linux/drivers/gpu/drm/i915/i915_gem_gtt.c

3523 lines
91 KiB
C
Raw Normal View History

/*
* Copyright © 2010 Daniel Vetter
* Copyright © 2011-2014 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*
*/
#include <linux/seq_file.h>
#include <drm/drmP.h>
#include <drm/i915_drm.h>
#include "i915_drv.h"
#include "i915_vgpu.h"
#include "i915_trace.h"
#include "intel_drv.h"
/**
* DOC: Global GTT views
*
* Background and previous state
*
* Historically objects could exists (be bound) in global GTT space only as
* singular instances with a view representing all of the object's backing pages
* in a linear fashion. This view will be called a normal view.
*
* To support multiple views of the same object, where the number of mapped
* pages is not equal to the backing store, or where the layout of the pages
* is not linear, concept of a GGTT view was added.
*
* One example of an alternative view is a stereo display driven by a single
* image. In this case we would have a framebuffer looking like this
* (2x2 pages):
*
* 12
* 34
*
* Above would represent a normal GGTT view as normally mapped for GPU or CPU
* rendering. In contrast, fed to the display engine would be an alternative
* view which could look something like this:
*
* 1212
* 3434
*
* In this example both the size and layout of pages in the alternative view is
* different from the normal view.
*
* Implementation and usage
*
* GGTT views are implemented using VMAs and are distinguished via enum
* i915_ggtt_view_type and struct i915_ggtt_view.
*
* A new flavour of core GEM functions which work with GGTT bound objects were
* added with the _ggtt_ infix, and sometimes with _view postfix to avoid
* renaming in large amounts of code. They take the struct i915_ggtt_view
* parameter encapsulating all metadata required to implement a view.
*
* As a helper for callers which are only interested in the normal view,
* globally const i915_ggtt_view_normal singleton instance exists. All old core
* GEM API functions, the ones not taking the view parameter, are operating on,
* or with the normal GGTT view.
*
* Code wanting to add or use a new GGTT view needs to:
*
* 1. Add a new enum with a suitable name.
* 2. Extend the metadata in the i915_ggtt_view structure if required.
* 3. Add support to i915_get_vma_pages().
*
* New views are required to build a scatter-gather table from within the
* i915_get_vma_pages function. This table is stored in the vma.ggtt_view and
* exists for the lifetime of an VMA.
*
* Core API is designed to have copy semantics which means that passed in
* struct i915_ggtt_view does not need to be persistent (left around after
* calling the core API functions).
*
*/
static int
i915_get_ggtt_vma_pages(struct i915_vma *vma);
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
const struct i915_ggtt_view i915_ggtt_view_normal;
const struct i915_ggtt_view i915_ggtt_view_rotated = {
.type = I915_GGTT_VIEW_ROTATED
};
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
static int sanitize_enable_ppgtt(struct drm_device *dev, int enable_ppgtt)
{
bool has_aliasing_ppgtt;
bool has_full_ppgtt;
has_aliasing_ppgtt = INTEL_INFO(dev)->gen >= 6;
has_full_ppgtt = INTEL_INFO(dev)->gen >= 7;
if (intel_vgpu_active(dev))
has_full_ppgtt = false; /* emulation is too hard */
/*
* We don't allow disabling PPGTT for gen9+ as it's a requirement for
* execlists, the sole mechanism available to submit work.
*/
if (INTEL_INFO(dev)->gen < 9 &&
(enable_ppgtt == 0 || !has_aliasing_ppgtt))
return 0;
if (enable_ppgtt == 1)
return 1;
if (enable_ppgtt == 2 && has_full_ppgtt)
return 2;
drm/i915: Disable full ppgtt by default There are too many oustanding issues: - Fence handling in the current code is broken. There's a patch series from me, but it's blocked on and extended review (which includes writing the testcases). - IOMMU mapping handling is broken, we need to properly refcount it - currently it gets destroyed when the first vma is unbound, so way too early. - There's a pending reset issue on snb. Since Mika's reset work and full ppgtt have been pulled in in separate branches and ended up intermittingly breaking each another it's unclear who's the exact culprit here. - We still have persistent evidince of crazy recursion bugs through vma_unbind and ppgtt_relase, e.g. https://bugs.freedesktop.org/show_bug.cgi?id=73383 This issue (and a few others meanwhile resolved) have blocked our performance measuring/tuning group since 3 months. - Secure batch dispatching is broken. This is blocking Brad Volkin's command checker work since 3 months. All these issues are confirmed to only happen when full ppgtt is enabled, falling back to aliasing ppgtt resolves them. But even aliasing ppgtt itself still has a regression: - We currently unconditionally bind objects into the aliasing ppgtt, which means all priviledged objects like ringbuffers are visible to unpriviledged access again. On top of that this also breaks the command checker for aliasing ppgtt, since it can't hide the validated batch any more. Furthermore topic/full-ppgtt has never been reviewed: - Lifetime rules around vma unbinding/release are unclear, resulting into this awesome hack called ppgtt_release. Which seems to take the blame for most of the recursion fallout. - Context/ring init works different on gpu reset than anywhere else. Such differeneces have in the past always lead to really hard to track down bugs. - Aliasing ppgtt is treated in a bunch of places as a real address space, but it isn't - the real address space is always the global gtt in that case. This results in a bit a mess between contexts and ppgtt object, further complication the context/ppgtt/vma lifetime rules. - We don't have any docs describing the overall concepts introduced with full ppgtt. A short, concise overview describing vmas and some of the strange bits around them (like the unbound vmas used by execbuf, or the new binding rules) really is needed. Note that a lot of the post topic/full-ppgtt merge fallout has already been addressed, this entire list here of 10 issues really only contains the still outstanding issues. Finally the 3.15 merge window is approaching and I think we need to use the remaining time to ensure that our fallback option of using aliasing ppgtt is in solid shape. Hence I think it's time to throw the switch. While at it demote the helper from static inline status because really. Cc: Ben Widawsky <ben@bwidawsk.net> Cc: Dave Airlie <airlied@gmail.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-03-06 08:40:43 +00:00
#ifdef CONFIG_INTEL_IOMMU
/* Disable ppgtt on SNB if VT-d is on. */
if (INTEL_INFO(dev)->gen == 6 && intel_iommu_gfx_mapped) {
DRM_INFO("Disabling PPGTT because VT-d is on\n");
return 0;
drm/i915: Disable full ppgtt by default There are too many oustanding issues: - Fence handling in the current code is broken. There's a patch series from me, but it's blocked on and extended review (which includes writing the testcases). - IOMMU mapping handling is broken, we need to properly refcount it - currently it gets destroyed when the first vma is unbound, so way too early. - There's a pending reset issue on snb. Since Mika's reset work and full ppgtt have been pulled in in separate branches and ended up intermittingly breaking each another it's unclear who's the exact culprit here. - We still have persistent evidince of crazy recursion bugs through vma_unbind and ppgtt_relase, e.g. https://bugs.freedesktop.org/show_bug.cgi?id=73383 This issue (and a few others meanwhile resolved) have blocked our performance measuring/tuning group since 3 months. - Secure batch dispatching is broken. This is blocking Brad Volkin's command checker work since 3 months. All these issues are confirmed to only happen when full ppgtt is enabled, falling back to aliasing ppgtt resolves them. But even aliasing ppgtt itself still has a regression: - We currently unconditionally bind objects into the aliasing ppgtt, which means all priviledged objects like ringbuffers are visible to unpriviledged access again. On top of that this also breaks the command checker for aliasing ppgtt, since it can't hide the validated batch any more. Furthermore topic/full-ppgtt has never been reviewed: - Lifetime rules around vma unbinding/release are unclear, resulting into this awesome hack called ppgtt_release. Which seems to take the blame for most of the recursion fallout. - Context/ring init works different on gpu reset than anywhere else. Such differeneces have in the past always lead to really hard to track down bugs. - Aliasing ppgtt is treated in a bunch of places as a real address space, but it isn't - the real address space is always the global gtt in that case. This results in a bit a mess between contexts and ppgtt object, further complication the context/ppgtt/vma lifetime rules. - We don't have any docs describing the overall concepts introduced with full ppgtt. A short, concise overview describing vmas and some of the strange bits around them (like the unbound vmas used by execbuf, or the new binding rules) really is needed. Note that a lot of the post topic/full-ppgtt merge fallout has already been addressed, this entire list here of 10 issues really only contains the still outstanding issues. Finally the 3.15 merge window is approaching and I think we need to use the remaining time to ensure that our fallback option of using aliasing ppgtt is in solid shape. Hence I think it's time to throw the switch. While at it demote the helper from static inline status because really. Cc: Ben Widawsky <ben@bwidawsk.net> Cc: Dave Airlie <airlied@gmail.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-03-06 08:40:43 +00:00
}
#endif
/* Early VLV doesn't have this */
if (IS_VALLEYVIEW(dev) && !IS_CHERRYVIEW(dev) &&
dev->pdev->revision < 0xb) {
DRM_DEBUG_DRIVER("disabling PPGTT on pre-B3 step VLV\n");
return 0;
}
if (INTEL_INFO(dev)->gen >= 8 && i915.enable_execlists)
return 2;
else
return has_aliasing_ppgtt ? 1 : 0;
drm/i915: Disable full ppgtt by default There are too many oustanding issues: - Fence handling in the current code is broken. There's a patch series from me, but it's blocked on and extended review (which includes writing the testcases). - IOMMU mapping handling is broken, we need to properly refcount it - currently it gets destroyed when the first vma is unbound, so way too early. - There's a pending reset issue on snb. Since Mika's reset work and full ppgtt have been pulled in in separate branches and ended up intermittingly breaking each another it's unclear who's the exact culprit here. - We still have persistent evidince of crazy recursion bugs through vma_unbind and ppgtt_relase, e.g. https://bugs.freedesktop.org/show_bug.cgi?id=73383 This issue (and a few others meanwhile resolved) have blocked our performance measuring/tuning group since 3 months. - Secure batch dispatching is broken. This is blocking Brad Volkin's command checker work since 3 months. All these issues are confirmed to only happen when full ppgtt is enabled, falling back to aliasing ppgtt resolves them. But even aliasing ppgtt itself still has a regression: - We currently unconditionally bind objects into the aliasing ppgtt, which means all priviledged objects like ringbuffers are visible to unpriviledged access again. On top of that this also breaks the command checker for aliasing ppgtt, since it can't hide the validated batch any more. Furthermore topic/full-ppgtt has never been reviewed: - Lifetime rules around vma unbinding/release are unclear, resulting into this awesome hack called ppgtt_release. Which seems to take the blame for most of the recursion fallout. - Context/ring init works different on gpu reset than anywhere else. Such differeneces have in the past always lead to really hard to track down bugs. - Aliasing ppgtt is treated in a bunch of places as a real address space, but it isn't - the real address space is always the global gtt in that case. This results in a bit a mess between contexts and ppgtt object, further complication the context/ppgtt/vma lifetime rules. - We don't have any docs describing the overall concepts introduced with full ppgtt. A short, concise overview describing vmas and some of the strange bits around them (like the unbound vmas used by execbuf, or the new binding rules) really is needed. Note that a lot of the post topic/full-ppgtt merge fallout has already been addressed, this entire list here of 10 issues really only contains the still outstanding issues. Finally the 3.15 merge window is approaching and I think we need to use the remaining time to ensure that our fallback option of using aliasing ppgtt is in solid shape. Hence I think it's time to throw the switch. While at it demote the helper from static inline status because really. Cc: Ben Widawsky <ben@bwidawsk.net> Cc: Dave Airlie <airlied@gmail.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-03-06 08:40:43 +00:00
}
static int ppgtt_bind_vma(struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 unused)
{
u32 pte_flags = 0;
/* Currently applicable only to VLV */
if (vma->obj->gt_ro)
pte_flags |= PTE_READ_ONLY;
vma->vm->insert_entries(vma->vm, vma->obj->pages, vma->node.start,
cache_level, pte_flags);
return 0;
}
static void ppgtt_unbind_vma(struct i915_vma *vma)
{
vma->vm->clear_range(vma->vm,
vma->node.start,
vma->obj->base.size,
true);
}
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
static gen8_pte_t gen8_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
bool valid)
{
gen8_pte_t pte = valid ? _PAGE_PRESENT | _PAGE_RW : 0;
pte |= addr;
switch (level) {
case I915_CACHE_NONE:
pte |= PPAT_UNCACHED_INDEX;
break;
case I915_CACHE_WT:
pte |= PPAT_DISPLAY_ELLC_INDEX;
break;
default:
pte |= PPAT_CACHED_INDEX;
break;
}
return pte;
}
static gen8_pde_t gen8_pde_encode(const dma_addr_t addr,
const enum i915_cache_level level)
{
gen8_pde_t pde = _PAGE_PRESENT | _PAGE_RW;
pde |= addr;
if (level != I915_CACHE_NONE)
pde |= PPAT_CACHED_PDE_INDEX;
else
pde |= PPAT_UNCACHED_INDEX;
return pde;
}
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
#define gen8_pdpe_encode gen8_pde_encode
#define gen8_pml4e_encode gen8_pde_encode
static gen6_pte_t snb_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
bool valid, u32 unused)
{
gen6_pte_t pte = valid ? GEN6_PTE_VALID : 0;
pte |= GEN6_PTE_ADDR_ENCODE(addr);
switch (level) {
case I915_CACHE_L3_LLC:
case I915_CACHE_LLC:
pte |= GEN6_PTE_CACHE_LLC;
break;
case I915_CACHE_NONE:
pte |= GEN6_PTE_UNCACHED;
break;
default:
drm/i915: Use BUILD_BUG if possible in the i915 WARN_ON Faster feedback to errors is always better. This is inspired by the addition to WARN_ONs to mask/enable helpers for registers to make sure callers have the arguments ordered correctly: Pretty much always the arguments are static. We use WARN_ON(1) a lot in default switch statements though where we should always handle all cases. So add a new macro specifically for that. The idea to use __builtin_constant_p is from Chris Wilson. v2: Use the ({}) gcc-ism to avoid the static inline, suggested by Dave. My first attempt used __cond as the temp var, which is the same used by BUILD_BUG_ON, but with inverted sense. Hilarity ensued, so sprinkle i915 into the name. Also use a temporary variable to only evaluate the condition once, suggested by Damien. v3: It's crazy but apparently 32bit gcc can't compile out the BUILD_BUG_ON in a lot of cases and just falls over. I have no idea why, but until clue grows just disable this nifty idea on 32bit builds. Reported by 0-day builder. v4: Got it all wrong, apparently its the gcc version. We need 4.9+. Now reported by Imre. v5: Chris suggested to add the case to MISSING_CASE for speedier debug. v6: Even some gcc 4.9 versions don't see through the maze, so give up for now. Keep the skeleton and MISSING_CASE stuff though. Cc: Imre Deak <imre.deak@intel.com> Cc: Damien Lespiau <damien.lespiau@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Dave Gordon <david.s.gordon@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
2014-12-08 15:40:10 +00:00
MISSING_CASE(level);
}
return pte;
}
static gen6_pte_t ivb_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
bool valid, u32 unused)
{
gen6_pte_t pte = valid ? GEN6_PTE_VALID : 0;
pte |= GEN6_PTE_ADDR_ENCODE(addr);
switch (level) {
case I915_CACHE_L3_LLC:
pte |= GEN7_PTE_CACHE_L3_LLC;
break;
case I915_CACHE_LLC:
pte |= GEN6_PTE_CACHE_LLC;
break;
case I915_CACHE_NONE:
pte |= GEN6_PTE_UNCACHED;
break;
default:
drm/i915: Use BUILD_BUG if possible in the i915 WARN_ON Faster feedback to errors is always better. This is inspired by the addition to WARN_ONs to mask/enable helpers for registers to make sure callers have the arguments ordered correctly: Pretty much always the arguments are static. We use WARN_ON(1) a lot in default switch statements though where we should always handle all cases. So add a new macro specifically for that. The idea to use __builtin_constant_p is from Chris Wilson. v2: Use the ({}) gcc-ism to avoid the static inline, suggested by Dave. My first attempt used __cond as the temp var, which is the same used by BUILD_BUG_ON, but with inverted sense. Hilarity ensued, so sprinkle i915 into the name. Also use a temporary variable to only evaluate the condition once, suggested by Damien. v3: It's crazy but apparently 32bit gcc can't compile out the BUILD_BUG_ON in a lot of cases and just falls over. I have no idea why, but until clue grows just disable this nifty idea on 32bit builds. Reported by 0-day builder. v4: Got it all wrong, apparently its the gcc version. We need 4.9+. Now reported by Imre. v5: Chris suggested to add the case to MISSING_CASE for speedier debug. v6: Even some gcc 4.9 versions don't see through the maze, so give up for now. Keep the skeleton and MISSING_CASE stuff though. Cc: Imre Deak <imre.deak@intel.com> Cc: Damien Lespiau <damien.lespiau@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Dave Gordon <david.s.gordon@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
2014-12-08 15:40:10 +00:00
MISSING_CASE(level);
}
return pte;
}
static gen6_pte_t byt_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
bool valid, u32 flags)
{
gen6_pte_t pte = valid ? GEN6_PTE_VALID : 0;
pte |= GEN6_PTE_ADDR_ENCODE(addr);
if (!(flags & PTE_READ_ONLY))
pte |= BYT_PTE_WRITEABLE;
if (level != I915_CACHE_NONE)
pte |= BYT_PTE_SNOOPED_BY_CPU_CACHES;
return pte;
}
static gen6_pte_t hsw_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
bool valid, u32 unused)
{
gen6_pte_t pte = valid ? GEN6_PTE_VALID : 0;
pte |= HSW_PTE_ADDR_ENCODE(addr);
if (level != I915_CACHE_NONE)
drm/i915/hsw: Change default LLC age to 3 The default LLC age was changed: commit 0d8ff15e9a15f2b393e53337a107b7a1e5919b6d Author: Ben Widawsky <benjamin.widawsky@intel.com> Date: Thu Jul 4 11:02:03 2013 -0700 drm/i915/hsw: Set correct Haswell PTE encodings. On the surface it would seem setting a default age wouldn't matter because all GEM BOs are aged similarly, so the order in which objects are evicted would not be subject to aging. The current working theory as to why this caused a regression though is that LLC is a bit special in that it is shared with the CPU. Presumably (not verified) the CPU fetches cachelines with age 3, and therefore recently cached GPU objects would be evicted before similar CPU object first when the LLC is full. It stands to reason therefore that this would negatively impact CPU bound benchmarks - but those seem to be low on the priority list. eLLC OTOH does not have this same property as LLC. It should be used entirely for the GPU, and so the age really shouldn't matter. Furthermore, we have no evidence to suggest one is better than another on eLLC. Since we've never properly supported eLLC before no, there should be no regression. If the GPU client really wants "younger" objects, they should use MOCS. v2: Drop the extra #define (Chad) v3: Actually git add v4: Pimped commit message Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67062 Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by: Chad Versace <chad.versace@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-08-05 06:47:29 +00:00
pte |= HSW_WB_LLC_AGE3;
return pte;
}
static gen6_pte_t iris_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
bool valid, u32 unused)
{
gen6_pte_t pte = valid ? GEN6_PTE_VALID : 0;
pte |= HSW_PTE_ADDR_ENCODE(addr);
switch (level) {
case I915_CACHE_NONE:
break;
case I915_CACHE_WT:
pte |= HSW_WT_ELLC_LLC_AGE3;
break;
default:
pte |= HSW_WB_ELLC_LLC_AGE3;
break;
}
return pte;
}
static int __setup_page_dma(struct drm_device *dev,
struct i915_page_dma *p, gfp_t flags)
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
{
struct device *device = &dev->pdev->dev;
p->page = alloc_page(flags);
if (!p->page)
return -ENOMEM;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
p->daddr = dma_map_page(device,
p->page, 0, 4096, PCI_DMA_BIDIRECTIONAL);
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
if (dma_mapping_error(device, p->daddr)) {
__free_page(p->page);
return -EINVAL;
}
return 0;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
}
static int setup_page_dma(struct drm_device *dev, struct i915_page_dma *p)
{
return __setup_page_dma(dev, p, GFP_KERNEL);
}
static void cleanup_page_dma(struct drm_device *dev, struct i915_page_dma *p)
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
{
if (WARN_ON(!p->page))
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
return;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
dma_unmap_page(&dev->pdev->dev, p->daddr, 4096, PCI_DMA_BIDIRECTIONAL);
__free_page(p->page);
memset(p, 0, sizeof(*p));
}
static void *kmap_page_dma(struct i915_page_dma *p)
{
return kmap_atomic(p->page);
}
/* We use the flushing unmap only with ppgtt structures:
* page directories, page tables and scratch pages.
*/
static void kunmap_page_dma(struct drm_device *dev, void *vaddr)
{
/* There are only few exceptions for gen >=6. chv and bxt.
* And we are not sure about the latter so play safe for now.
*/
if (IS_CHERRYVIEW(dev) || IS_BROXTON(dev))
drm_clflush_virt_range(vaddr, PAGE_SIZE);
kunmap_atomic(vaddr);
}
#define kmap_px(px) kmap_page_dma(px_base(px))
#define kunmap_px(ppgtt, vaddr) kunmap_page_dma((ppgtt)->base.dev, (vaddr))
#define setup_px(dev, px) setup_page_dma((dev), px_base(px))
#define cleanup_px(dev, px) cleanup_page_dma((dev), px_base(px))
#define fill_px(dev, px, v) fill_page_dma((dev), px_base(px), (v))
#define fill32_px(dev, px, v) fill_page_dma_32((dev), px_base(px), (v))
static void fill_page_dma(struct drm_device *dev, struct i915_page_dma *p,
const uint64_t val)
{
int i;
uint64_t * const vaddr = kmap_page_dma(p);
for (i = 0; i < 512; i++)
vaddr[i] = val;
kunmap_page_dma(dev, vaddr);
}
static void fill_page_dma_32(struct drm_device *dev, struct i915_page_dma *p,
const uint32_t val32)
{
uint64_t v = val32;
v = v << 32 | val32;
fill_page_dma(dev, p, v);
}
static struct i915_page_scratch *alloc_scratch_page(struct drm_device *dev)
{
struct i915_page_scratch *sp;
int ret;
sp = kzalloc(sizeof(*sp), GFP_KERNEL);
if (sp == NULL)
return ERR_PTR(-ENOMEM);
ret = __setup_page_dma(dev, px_base(sp), GFP_DMA32 | __GFP_ZERO);
if (ret) {
kfree(sp);
return ERR_PTR(ret);
}
set_pages_uc(px_page(sp), 1);
return sp;
}
static void free_scratch_page(struct drm_device *dev,
struct i915_page_scratch *sp)
{
set_pages_wb(px_page(sp), 1);
cleanup_px(dev, sp);
kfree(sp);
}
static struct i915_page_table *alloc_pt(struct drm_device *dev)
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
{
struct i915_page_table *pt;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
const size_t count = INTEL_INFO(dev)->gen >= 8 ?
GEN8_PTES : GEN6_PTES;
int ret = -ENOMEM;
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
pt = kzalloc(sizeof(*pt), GFP_KERNEL);
if (!pt)
return ERR_PTR(-ENOMEM);
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
pt->used_ptes = kcalloc(BITS_TO_LONGS(count), sizeof(*pt->used_ptes),
GFP_KERNEL);
if (!pt->used_ptes)
goto fail_bitmap;
ret = setup_px(dev, pt);
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
if (ret)
goto fail_page_m;
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
return pt;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
fail_page_m:
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
kfree(pt->used_ptes);
fail_bitmap:
kfree(pt);
return ERR_PTR(ret);
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
}
static void free_pt(struct drm_device *dev, struct i915_page_table *pt)
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
{
cleanup_px(dev, pt);
kfree(pt->used_ptes);
kfree(pt);
}
static void gen8_initialize_pt(struct i915_address_space *vm,
struct i915_page_table *pt)
{
gen8_pte_t scratch_pte;
scratch_pte = gen8_pte_encode(px_dma(vm->scratch_page),
I915_CACHE_LLC, true);
fill_px(vm->dev, pt, scratch_pte);
}
static void gen6_initialize_pt(struct i915_address_space *vm,
struct i915_page_table *pt)
{
gen6_pte_t scratch_pte;
WARN_ON(px_dma(vm->scratch_page) == 0);
scratch_pte = vm->pte_encode(px_dma(vm->scratch_page),
I915_CACHE_LLC, true, 0);
fill32_px(vm->dev, pt, scratch_pte);
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
}
static struct i915_page_directory *alloc_pd(struct drm_device *dev)
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
{
struct i915_page_directory *pd;
int ret = -ENOMEM;
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
pd = kzalloc(sizeof(*pd), GFP_KERNEL);
if (!pd)
return ERR_PTR(-ENOMEM);
pd->used_pdes = kcalloc(BITS_TO_LONGS(I915_PDES),
sizeof(*pd->used_pdes), GFP_KERNEL);
if (!pd->used_pdes)
goto fail_bitmap;
ret = setup_px(dev, pd);
if (ret)
goto fail_page_m;
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
return pd;
fail_page_m:
kfree(pd->used_pdes);
fail_bitmap:
kfree(pd);
return ERR_PTR(ret);
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
}
static void free_pd(struct drm_device *dev, struct i915_page_directory *pd)
{
if (px_page(pd)) {
cleanup_px(dev, pd);
kfree(pd->used_pdes);
kfree(pd);
}
}
static void gen8_initialize_pd(struct i915_address_space *vm,
struct i915_page_directory *pd)
{
gen8_pde_t scratch_pde;
scratch_pde = gen8_pde_encode(px_dma(vm->scratch_pt), I915_CACHE_LLC);
fill_px(vm->dev, pd, scratch_pde);
}
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
static int __pdp_init(struct drm_device *dev,
struct i915_page_directory_pointer *pdp)
{
size_t pdpes = I915_PDPES_PER_PDP(dev);
pdp->used_pdpes = kcalloc(BITS_TO_LONGS(pdpes),
sizeof(unsigned long),
GFP_KERNEL);
if (!pdp->used_pdpes)
return -ENOMEM;
pdp->page_directory = kcalloc(pdpes, sizeof(*pdp->page_directory),
GFP_KERNEL);
if (!pdp->page_directory) {
kfree(pdp->used_pdpes);
/* the PDP might be the statically allocated top level. Keep it
* as clean as possible */
pdp->used_pdpes = NULL;
return -ENOMEM;
}
return 0;
}
static void __pdp_fini(struct i915_page_directory_pointer *pdp)
{
kfree(pdp->used_pdpes);
kfree(pdp->page_directory);
pdp->page_directory = NULL;
}
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
static struct
i915_page_directory_pointer *alloc_pdp(struct drm_device *dev)
{
struct i915_page_directory_pointer *pdp;
int ret = -ENOMEM;
WARN_ON(!USES_FULL_48BIT_PPGTT(dev));
pdp = kzalloc(sizeof(*pdp), GFP_KERNEL);
if (!pdp)
return ERR_PTR(-ENOMEM);
ret = __pdp_init(dev, pdp);
if (ret)
goto fail_bitmap;
ret = setup_px(dev, pdp);
if (ret)
goto fail_page_m;
return pdp;
fail_page_m:
__pdp_fini(pdp);
fail_bitmap:
kfree(pdp);
return ERR_PTR(ret);
}
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
static void free_pdp(struct drm_device *dev,
struct i915_page_directory_pointer *pdp)
{
__pdp_fini(pdp);
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
if (USES_FULL_48BIT_PPGTT(dev)) {
cleanup_px(dev, pdp);
kfree(pdp);
}
}
static void gen8_initialize_pdp(struct i915_address_space *vm,
struct i915_page_directory_pointer *pdp)
{
gen8_ppgtt_pdpe_t scratch_pdpe;
scratch_pdpe = gen8_pdpe_encode(px_dma(vm->scratch_pd), I915_CACHE_LLC);
fill_px(vm->dev, pdp, scratch_pdpe);
}
static void gen8_initialize_pml4(struct i915_address_space *vm,
struct i915_pml4 *pml4)
{
gen8_ppgtt_pml4e_t scratch_pml4e;
scratch_pml4e = gen8_pml4e_encode(px_dma(vm->scratch_pdp),
I915_CACHE_LLC);
fill_px(vm->dev, pml4, scratch_pml4e);
}
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
static void
gen8_setup_page_directory(struct i915_hw_ppgtt *ppgtt,
struct i915_page_directory_pointer *pdp,
struct i915_page_directory *pd,
int index)
{
gen8_ppgtt_pdpe_t *page_directorypo;
if (!USES_FULL_48BIT_PPGTT(ppgtt->base.dev))
return;
page_directorypo = kmap_px(pdp);
page_directorypo[index] = gen8_pdpe_encode(px_dma(pd), I915_CACHE_LLC);
kunmap_px(ppgtt, page_directorypo);
}
static void
gen8_setup_page_directory_pointer(struct i915_hw_ppgtt *ppgtt,
struct i915_pml4 *pml4,
struct i915_page_directory_pointer *pdp,
int index)
{
gen8_ppgtt_pml4e_t *pagemap = kmap_px(pml4);
WARN_ON(!USES_FULL_48BIT_PPGTT(ppgtt->base.dev));
pagemap[index] = gen8_pml4e_encode(px_dma(pdp), I915_CACHE_LLC);
kunmap_px(ppgtt, pagemap);
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
}
/* Broadwell Page Directory Pointer Descriptors */
static int gen8_write_pdp(struct drm_i915_gem_request *req,
unsigned entry,
dma_addr_t addr)
{
struct intel_engine_cs *ring = req->ring;
int ret;
BUG_ON(entry >= 4);
ret = intel_ring_begin(req, 6);
if (ret)
return ret;
intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1));
intel_ring_emit(ring, GEN8_RING_PDP_UDW(ring, entry));
intel_ring_emit(ring, upper_32_bits(addr));
intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1));
intel_ring_emit(ring, GEN8_RING_PDP_LDW(ring, entry));
intel_ring_emit(ring, lower_32_bits(addr));
intel_ring_advance(ring);
return 0;
}
2015-07-30 10:06:23 +00:00
static int gen8_legacy_mm_switch(struct i915_hw_ppgtt *ppgtt,
struct drm_i915_gem_request *req)
{
int i, ret;
for (i = GEN8_LEGACY_PDPES - 1; i >= 0; i--) {
const dma_addr_t pd_daddr = i915_page_dir_dma_addr(ppgtt, i);
ret = gen8_write_pdp(req, i, pd_daddr);
if (ret)
return ret;
}
return 0;
}
2015-07-30 10:06:23 +00:00
static int gen8_48b_mm_switch(struct i915_hw_ppgtt *ppgtt,
struct drm_i915_gem_request *req)
{
return gen8_write_pdp(req, 0, px_dma(&ppgtt->pml4));
}
static void gen8_ppgtt_clear_pte_range(struct i915_address_space *vm,
struct i915_page_directory_pointer *pdp,
uint64_t start,
uint64_t length,
gen8_pte_t scratch_pte)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
gen8_pte_t *pt_vaddr;
drm/i915/gen8: Add 4 level support in insert_entries and clear_range When 48b is enabled, gen8_ppgtt_insert_entries needs to read the Page Map Level 4 (PML4), before it selects which Page Directory Pointer (PDP) it will write to. Similarly, gen8_ppgtt_clear_range needs to get the correct PDP/PD range. This patch was inspired by Ben's "Depend exclusively on map and unmap_vma". v2: Rebase after s/page_tables/page_table/. v3: Remove unnecessary pdpe loop in gen8_ppgtt_clear_range_4lvl and use clamp_pdp in gen8_ppgtt_insert_entries (Akash). v4: Merge gen8_ppgtt_clear_range_4lvl into gen8_ppgtt_clear_range to maintain symmetry with gen8_ppgtt_insert_entries (Akash). v5: Do not mix pages and bytes in insert_entries (Akash). v6: Prevent overflow in sg_nents << PAGE_SHIFT, when inserting 4GB at once. v7: Rebase after Mika's ppgtt cleanup / scratch merge patch series. Use gen8_px_index functions, and remove unnecessary number of pages parameter in insert_pte_entries. v8: Change gen8_ppgtt_clear_pte_range to stop at PDP boundary, instead of adding and extra clamp function; remove unnecessary pdp_start/pdp_len variables (Akash). v9: pages->orig_nents instead of sg_nents(pages->sgl) to get the length (Akash). v10: Remove pdp warning check ingen8_ppgtt_insert_pte_entries until this commit (Akash). Reviewed-by: Akash Goel <akash.goel@intel.com> (v9) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-08-03 08:53:27 +00:00
unsigned pdpe = gen8_pdpe_index(start);
unsigned pde = gen8_pde_index(start);
unsigned pte = gen8_pte_index(start);
unsigned num_entries = length >> PAGE_SHIFT;
unsigned last_pte, i;
if (WARN_ON(!pdp))
return;
while (num_entries) {
struct i915_page_directory *pd;
struct i915_page_table *pt;
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
2015-07-30 10:02:03 +00:00
if (WARN_ON(!pdp->page_directory[pdpe]))
break;
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
2015-07-30 10:02:03 +00:00
pd = pdp->page_directory[pdpe];
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
if (WARN_ON(!pd->page_table[pde]))
break;
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
pt = pd->page_table[pde];
if (WARN_ON(!px_page(pt)))
break;
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
last_pte = pte + num_entries;
if (last_pte > GEN8_PTES)
last_pte = GEN8_PTES;
pt_vaddr = kmap_px(pt);
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
for (i = pte; i < last_pte; i++) {
pt_vaddr[i] = scratch_pte;
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
num_entries--;
}
kunmap_px(ppgtt, pt);
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
pte = 0;
if (++pde == I915_PDES) {
drm/i915/gen8: Add 4 level support in insert_entries and clear_range When 48b is enabled, gen8_ppgtt_insert_entries needs to read the Page Map Level 4 (PML4), before it selects which Page Directory Pointer (PDP) it will write to. Similarly, gen8_ppgtt_clear_range needs to get the correct PDP/PD range. This patch was inspired by Ben's "Depend exclusively on map and unmap_vma". v2: Rebase after s/page_tables/page_table/. v3: Remove unnecessary pdpe loop in gen8_ppgtt_clear_range_4lvl and use clamp_pdp in gen8_ppgtt_insert_entries (Akash). v4: Merge gen8_ppgtt_clear_range_4lvl into gen8_ppgtt_clear_range to maintain symmetry with gen8_ppgtt_insert_entries (Akash). v5: Do not mix pages and bytes in insert_entries (Akash). v6: Prevent overflow in sg_nents << PAGE_SHIFT, when inserting 4GB at once. v7: Rebase after Mika's ppgtt cleanup / scratch merge patch series. Use gen8_px_index functions, and remove unnecessary number of pages parameter in insert_pte_entries. v8: Change gen8_ppgtt_clear_pte_range to stop at PDP boundary, instead of adding and extra clamp function; remove unnecessary pdp_start/pdp_len variables (Akash). v9: pages->orig_nents instead of sg_nents(pages->sgl) to get the length (Akash). v10: Remove pdp warning check ingen8_ppgtt_insert_pte_entries until this commit (Akash). Reviewed-by: Akash Goel <akash.goel@intel.com> (v9) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-08-03 08:53:27 +00:00
if (++pdpe == I915_PDPES_PER_PDP(vm->dev))
break;
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
pde = 0;
}
}
}
static void gen8_ppgtt_clear_range(struct i915_address_space *vm,
uint64_t start,
uint64_t length,
bool use_scratch)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
gen8_pte_t scratch_pte = gen8_pte_encode(px_dma(vm->scratch_page),
I915_CACHE_LLC, use_scratch);
drm/i915/gen8: Add 4 level support in insert_entries and clear_range When 48b is enabled, gen8_ppgtt_insert_entries needs to read the Page Map Level 4 (PML4), before it selects which Page Directory Pointer (PDP) it will write to. Similarly, gen8_ppgtt_clear_range needs to get the correct PDP/PD range. This patch was inspired by Ben's "Depend exclusively on map and unmap_vma". v2: Rebase after s/page_tables/page_table/. v3: Remove unnecessary pdpe loop in gen8_ppgtt_clear_range_4lvl and use clamp_pdp in gen8_ppgtt_insert_entries (Akash). v4: Merge gen8_ppgtt_clear_range_4lvl into gen8_ppgtt_clear_range to maintain symmetry with gen8_ppgtt_insert_entries (Akash). v5: Do not mix pages and bytes in insert_entries (Akash). v6: Prevent overflow in sg_nents << PAGE_SHIFT, when inserting 4GB at once. v7: Rebase after Mika's ppgtt cleanup / scratch merge patch series. Use gen8_px_index functions, and remove unnecessary number of pages parameter in insert_pte_entries. v8: Change gen8_ppgtt_clear_pte_range to stop at PDP boundary, instead of adding and extra clamp function; remove unnecessary pdp_start/pdp_len variables (Akash). v9: pages->orig_nents instead of sg_nents(pages->sgl) to get the length (Akash). v10: Remove pdp warning check ingen8_ppgtt_insert_pte_entries until this commit (Akash). Reviewed-by: Akash Goel <akash.goel@intel.com> (v9) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-08-03 08:53:27 +00:00
if (!USES_FULL_48BIT_PPGTT(vm->dev)) {
gen8_ppgtt_clear_pte_range(vm, &ppgtt->pdp, start, length,
scratch_pte);
} else {
uint64_t templ4, pml4e;
struct i915_page_directory_pointer *pdp;
gen8_for_each_pml4e(pdp, &ppgtt->pml4, start, length, templ4, pml4e) {
gen8_ppgtt_clear_pte_range(vm, pdp, start, length,
scratch_pte);
}
}
}
static void
gen8_ppgtt_insert_pte_entries(struct i915_address_space *vm,
struct i915_page_directory_pointer *pdp,
struct sg_page_iter *sg_iter,
uint64_t start,
enum i915_cache_level cache_level)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
gen8_pte_t *pt_vaddr;
drm/i915/gen8: Add 4 level support in insert_entries and clear_range When 48b is enabled, gen8_ppgtt_insert_entries needs to read the Page Map Level 4 (PML4), before it selects which Page Directory Pointer (PDP) it will write to. Similarly, gen8_ppgtt_clear_range needs to get the correct PDP/PD range. This patch was inspired by Ben's "Depend exclusively on map and unmap_vma". v2: Rebase after s/page_tables/page_table/. v3: Remove unnecessary pdpe loop in gen8_ppgtt_clear_range_4lvl and use clamp_pdp in gen8_ppgtt_insert_entries (Akash). v4: Merge gen8_ppgtt_clear_range_4lvl into gen8_ppgtt_clear_range to maintain symmetry with gen8_ppgtt_insert_entries (Akash). v5: Do not mix pages and bytes in insert_entries (Akash). v6: Prevent overflow in sg_nents << PAGE_SHIFT, when inserting 4GB at once. v7: Rebase after Mika's ppgtt cleanup / scratch merge patch series. Use gen8_px_index functions, and remove unnecessary number of pages parameter in insert_pte_entries. v8: Change gen8_ppgtt_clear_pte_range to stop at PDP boundary, instead of adding and extra clamp function; remove unnecessary pdp_start/pdp_len variables (Akash). v9: pages->orig_nents instead of sg_nents(pages->sgl) to get the length (Akash). v10: Remove pdp warning check ingen8_ppgtt_insert_pte_entries until this commit (Akash). Reviewed-by: Akash Goel <akash.goel@intel.com> (v9) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-08-03 08:53:27 +00:00
unsigned pdpe = gen8_pdpe_index(start);
unsigned pde = gen8_pde_index(start);
unsigned pte = gen8_pte_index(start);
pt_vaddr = NULL;
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
while (__sg_page_iter_next(sg_iter)) {
if (pt_vaddr == NULL) {
2015-07-30 10:02:03 +00:00
struct i915_page_directory *pd = pdp->page_directory[pdpe];
struct i915_page_table *pt = pd->page_table[pde];
pt_vaddr = kmap_px(pt);
}
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
pt_vaddr[pte] =
gen8_pte_encode(sg_page_iter_dma_address(sg_iter),
cache_level, true);
if (++pte == GEN8_PTES) {
kunmap_px(ppgtt, pt_vaddr);
pt_vaddr = NULL;
if (++pde == I915_PDES) {
drm/i915/gen8: Add 4 level support in insert_entries and clear_range When 48b is enabled, gen8_ppgtt_insert_entries needs to read the Page Map Level 4 (PML4), before it selects which Page Directory Pointer (PDP) it will write to. Similarly, gen8_ppgtt_clear_range needs to get the correct PDP/PD range. This patch was inspired by Ben's "Depend exclusively on map and unmap_vma". v2: Rebase after s/page_tables/page_table/. v3: Remove unnecessary pdpe loop in gen8_ppgtt_clear_range_4lvl and use clamp_pdp in gen8_ppgtt_insert_entries (Akash). v4: Merge gen8_ppgtt_clear_range_4lvl into gen8_ppgtt_clear_range to maintain symmetry with gen8_ppgtt_insert_entries (Akash). v5: Do not mix pages and bytes in insert_entries (Akash). v6: Prevent overflow in sg_nents << PAGE_SHIFT, when inserting 4GB at once. v7: Rebase after Mika's ppgtt cleanup / scratch merge patch series. Use gen8_px_index functions, and remove unnecessary number of pages parameter in insert_pte_entries. v8: Change gen8_ppgtt_clear_pte_range to stop at PDP boundary, instead of adding and extra clamp function; remove unnecessary pdp_start/pdp_len variables (Akash). v9: pages->orig_nents instead of sg_nents(pages->sgl) to get the length (Akash). v10: Remove pdp warning check ingen8_ppgtt_insert_pte_entries until this commit (Akash). Reviewed-by: Akash Goel <akash.goel@intel.com> (v9) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-08-03 08:53:27 +00:00
if (++pdpe == I915_PDPES_PER_PDP(vm->dev))
break;
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
pde = 0;
}
pte = 0;
}
}
if (pt_vaddr)
kunmap_px(ppgtt, pt_vaddr);
}
static void gen8_ppgtt_insert_entries(struct i915_address_space *vm,
struct sg_table *pages,
uint64_t start,
enum i915_cache_level cache_level,
u32 unused)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
struct sg_page_iter sg_iter;
__sg_page_iter_start(&sg_iter, pages->sgl, sg_nents(pages->sgl), 0);
drm/i915/gen8: Add 4 level support in insert_entries and clear_range When 48b is enabled, gen8_ppgtt_insert_entries needs to read the Page Map Level 4 (PML4), before it selects which Page Directory Pointer (PDP) it will write to. Similarly, gen8_ppgtt_clear_range needs to get the correct PDP/PD range. This patch was inspired by Ben's "Depend exclusively on map and unmap_vma". v2: Rebase after s/page_tables/page_table/. v3: Remove unnecessary pdpe loop in gen8_ppgtt_clear_range_4lvl and use clamp_pdp in gen8_ppgtt_insert_entries (Akash). v4: Merge gen8_ppgtt_clear_range_4lvl into gen8_ppgtt_clear_range to maintain symmetry with gen8_ppgtt_insert_entries (Akash). v5: Do not mix pages and bytes in insert_entries (Akash). v6: Prevent overflow in sg_nents << PAGE_SHIFT, when inserting 4GB at once. v7: Rebase after Mika's ppgtt cleanup / scratch merge patch series. Use gen8_px_index functions, and remove unnecessary number of pages parameter in insert_pte_entries. v8: Change gen8_ppgtt_clear_pte_range to stop at PDP boundary, instead of adding and extra clamp function; remove unnecessary pdp_start/pdp_len variables (Akash). v9: pages->orig_nents instead of sg_nents(pages->sgl) to get the length (Akash). v10: Remove pdp warning check ingen8_ppgtt_insert_pte_entries until this commit (Akash). Reviewed-by: Akash Goel <akash.goel@intel.com> (v9) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-08-03 08:53:27 +00:00
if (!USES_FULL_48BIT_PPGTT(vm->dev)) {
gen8_ppgtt_insert_pte_entries(vm, &ppgtt->pdp, &sg_iter, start,
cache_level);
} else {
struct i915_page_directory_pointer *pdp;
uint64_t templ4, pml4e;
uint64_t length = (uint64_t)pages->orig_nents << PAGE_SHIFT;
gen8_for_each_pml4e(pdp, &ppgtt->pml4, start, length, templ4, pml4e) {
gen8_ppgtt_insert_pte_entries(vm, pdp, &sg_iter,
start, cache_level);
}
}
}
static void gen8_free_page_tables(struct drm_device *dev,
struct i915_page_directory *pd)
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
{
int i;
if (!px_page(pd))
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
return;
for_each_set_bit(i, pd->used_pdes, I915_PDES) {
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
if (WARN_ON(!pd->page_table[i]))
continue;
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
free_pt(dev, pd->page_table[i]);
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
pd->page_table[i] = NULL;
}
}
static int gen8_init_scratch(struct i915_address_space *vm)
{
struct drm_device *dev = vm->dev;
vm->scratch_page = alloc_scratch_page(dev);
if (IS_ERR(vm->scratch_page))
return PTR_ERR(vm->scratch_page);
vm->scratch_pt = alloc_pt(dev);
if (IS_ERR(vm->scratch_pt)) {
free_scratch_page(dev, vm->scratch_page);
return PTR_ERR(vm->scratch_pt);
}
vm->scratch_pd = alloc_pd(dev);
if (IS_ERR(vm->scratch_pd)) {
free_pt(dev, vm->scratch_pt);
free_scratch_page(dev, vm->scratch_page);
return PTR_ERR(vm->scratch_pd);
}
if (USES_FULL_48BIT_PPGTT(dev)) {
vm->scratch_pdp = alloc_pdp(dev);
if (IS_ERR(vm->scratch_pdp)) {
free_pd(dev, vm->scratch_pd);
free_pt(dev, vm->scratch_pt);
free_scratch_page(dev, vm->scratch_page);
return PTR_ERR(vm->scratch_pdp);
}
}
gen8_initialize_pt(vm, vm->scratch_pt);
gen8_initialize_pd(vm, vm->scratch_pd);
if (USES_FULL_48BIT_PPGTT(dev))
gen8_initialize_pdp(vm, vm->scratch_pdp);
return 0;
}
static int gen8_ppgtt_notify_vgt(struct i915_hw_ppgtt *ppgtt, bool create)
{
enum vgt_g2v_type msg;
struct drm_device *dev = ppgtt->base.dev;
struct drm_i915_private *dev_priv = dev->dev_private;
unsigned int offset = vgtif_reg(pdp0_lo);
int i;
if (USES_FULL_48BIT_PPGTT(dev)) {
u64 daddr = px_dma(&ppgtt->pml4);
I915_WRITE(offset, lower_32_bits(daddr));
I915_WRITE(offset + 4, upper_32_bits(daddr));
msg = (create ? VGT_G2V_PPGTT_L4_PAGE_TABLE_CREATE :
VGT_G2V_PPGTT_L4_PAGE_TABLE_DESTROY);
} else {
for (i = 0; i < GEN8_LEGACY_PDPES; i++) {
u64 daddr = i915_page_dir_dma_addr(ppgtt, i);
I915_WRITE(offset, lower_32_bits(daddr));
I915_WRITE(offset + 4, upper_32_bits(daddr));
offset += 8;
}
msg = (create ? VGT_G2V_PPGTT_L3_PAGE_TABLE_CREATE :
VGT_G2V_PPGTT_L3_PAGE_TABLE_DESTROY);
}
I915_WRITE(vgtif_reg(g2v_notify), msg);
return 0;
}
static void gen8_free_scratch(struct i915_address_space *vm)
{
struct drm_device *dev = vm->dev;
if (USES_FULL_48BIT_PPGTT(dev))
free_pdp(dev, vm->scratch_pdp);
free_pd(dev, vm->scratch_pd);
free_pt(dev, vm->scratch_pt);
free_scratch_page(dev, vm->scratch_page);
}
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
static void gen8_ppgtt_cleanup_3lvl(struct drm_device *dev,
struct i915_page_directory_pointer *pdp)
{
int i;
2015-07-30 10:02:03 +00:00
for_each_set_bit(i, pdp->used_pdpes, I915_PDPES_PER_PDP(dev)) {
if (WARN_ON(!pdp->page_directory[i]))
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
continue;
2015-07-30 10:02:03 +00:00
gen8_free_page_tables(dev, pdp->page_directory[i]);
free_pd(dev, pdp->page_directory[i]);
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
}
2015-07-30 10:02:03 +00:00
free_pdp(dev, pdp);
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
}
static void gen8_ppgtt_cleanup_4lvl(struct i915_hw_ppgtt *ppgtt)
{
int i;
for_each_set_bit(i, ppgtt->pml4.used_pml4es, GEN8_PML4ES_PER_PML4) {
if (WARN_ON(!ppgtt->pml4.pdps[i]))
continue;
gen8_ppgtt_cleanup_3lvl(ppgtt->base.dev, ppgtt->pml4.pdps[i]);
}
cleanup_px(ppgtt->base.dev, &ppgtt->pml4);
}
static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
if (intel_vgpu_active(vm->dev))
gen8_ppgtt_notify_vgt(ppgtt, false);
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
if (!USES_FULL_48BIT_PPGTT(ppgtt->base.dev))
gen8_ppgtt_cleanup_3lvl(ppgtt->base.dev, &ppgtt->pdp);
else
gen8_ppgtt_cleanup_4lvl(ppgtt);
2015-07-30 10:02:03 +00:00
gen8_free_scratch(vm);
}
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/**
* gen8_ppgtt_alloc_pagetabs() - Allocate page tables for VA range.
2015-07-30 10:02:03 +00:00
* @vm: Master vm structure.
* @pd: Page directory for this address range.
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
* @start: Starting virtual address to begin allocations.
2015-07-30 10:02:03 +00:00
* @length: Size of the allocations.
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
* @new_pts: Bitmap set by function with new allocations. Likely used by the
* caller to free on error.
*
* Allocate the required number of page tables. Extremely similar to
* gen8_ppgtt_alloc_page_directories(). The main difference is here we are limited by
* the page directory boundary (instead of the page directory pointer). That
* boundary is 1GB virtual. Therefore, unlike gen8_ppgtt_alloc_page_directories(), it is
* possible, and likely that the caller will need to use multiple calls of this
* function to achieve the appropriate allocation.
*
* Return: 0 if success; negative error code otherwise.
*/
2015-07-30 10:02:03 +00:00
static int gen8_ppgtt_alloc_pagetabs(struct i915_address_space *vm,
struct i915_page_directory *pd,
uint64_t start,
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
uint64_t length,
unsigned long *new_pts)
{
2015-07-30 10:02:03 +00:00
struct drm_device *dev = vm->dev;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
struct i915_page_table *pt;
uint64_t temp;
uint32_t pde;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
gen8_for_each_pde(pt, pd, start, length, temp, pde) {
/* Don't reallocate page tables */
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
if (test_bit(pde, pd->used_pdes)) {
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/* Scratch is never allocated this way */
2015-07-30 10:02:03 +00:00
WARN_ON(pt == vm->scratch_pt);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
continue;
}
pt = alloc_pt(dev);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
if (IS_ERR(pt))
goto unwind_out;
2015-07-30 10:02:03 +00:00
gen8_initialize_pt(vm, pt);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
pd->page_table[pde] = pt;
__set_bit(pde, new_pts);
trace_i915_page_table_entry_alloc(vm, pde, start, GEN8_PDE_SHIFT);
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
}
return 0;
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
unwind_out:
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
for_each_set_bit(pde, new_pts, I915_PDES)
free_pt(dev, pd->page_table[pde]);
drm/i915/bdw: Reorganize PT allocations The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 2MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed. In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages. To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages. NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff. v2/NOTE2: This patch predated commit: 6f1cc993518462ccf039e195fabd47e7aa5bfd13 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Dec 31 15:50:31 2013 +0000 drm/i915: Avoid dereference past end of page arr It fixed the same issue as that patch, but because of the limbo state of PPGTT, Chris patch was merged instead. The excess churn is a result of my using my original patch, which has my preferred naming. Primarily act_* is changed to which_*, but it's mostly the same otherwise. I've kept the convention Chris used for the pte wrap (I had something slightly different, and broken - but fixable) v3: Rename which_p[..]e to drop which_ (Chris) Remove BUG_ON in inner loop (Chris) Redo the pde/pdpe wrap logic (Chris) v4: s/1MB/2MB in commit message (Imre) Plug leaking gen8_pt_pages in both the error path, as well as general free case (Imre) v5: Rename leftover "which_" variables (Imre) Add the pde = 0 wrap that was missed from v3 (Imre) Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Squash in fixup from Ben.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-02-20 19:51:21 +00:00
return -ENOMEM;
}
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/**
* gen8_ppgtt_alloc_page_directories() - Allocate page directories for VA range.
2015-07-30 10:02:03 +00:00
* @vm: Master vm structure.
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
* @pdp: Page directory pointer for this address range.
* @start: Starting virtual address to begin allocations.
2015-07-30 10:02:03 +00:00
* @length: Size of the allocations.
* @new_pds: Bitmap set by function with new allocations. Likely used by the
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
* caller to free on error.
*
* Allocate the required number of page directories starting at the pde index of
* @start, and ending at the pde index @start + @length. This function will skip
* over already allocated page directories within the range, and only allocate
* new ones, setting the appropriate pointer within the pdp as well as the
* correct position in the bitmap @new_pds.
*
* The function will only allocate the pages within the range for a give page
* directory pointer. In other words, if @start + @length straddles a virtually
* addressed PDP boundary (512GB for 4k pages), there will be more allocations
* required by the caller, This is not currently possible, and the BUG in the
* code will prevent it.
*
* Return: 0 if success; negative error code otherwise.
*/
2015-07-30 10:02:03 +00:00
static int
gen8_ppgtt_alloc_page_directories(struct i915_address_space *vm,
struct i915_page_directory_pointer *pdp,
uint64_t start,
uint64_t length,
unsigned long *new_pds)
{
2015-07-30 10:02:03 +00:00
struct drm_device *dev = vm->dev;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
struct i915_page_directory *pd;
uint64_t temp;
uint32_t pdpe;
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
uint32_t pdpes = I915_PDPES_PER_PDP(dev);
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
WARN_ON(!bitmap_empty(new_pds, pdpes));
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
gen8_for_each_pdpe(pd, pdp, start, length, temp, pdpe) {
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
if (test_bit(pdpe, pdp->used_pdpes))
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
continue;
pd = alloc_pd(dev);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
if (IS_ERR(pd))
goto unwind_out;
2015-07-30 10:02:03 +00:00
gen8_initialize_pd(vm, pd);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
pdp->page_directory[pdpe] = pd;
__set_bit(pdpe, new_pds);
trace_i915_page_directory_entry_alloc(vm, pdpe, start, GEN8_PDPE_SHIFT);
}
return 0;
unwind_out:
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
for_each_set_bit(pdpe, new_pds, pdpes)
free_pd(dev, pdp->page_directory[pdpe]);
return -ENOMEM;
}
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
/**
* gen8_ppgtt_alloc_page_dirpointers() - Allocate pdps for VA range.
* @vm: Master vm structure.
* @pml4: Page map level 4 for this address range.
* @start: Starting virtual address to begin allocations.
* @length: Size of the allocations.
* @new_pdps: Bitmap set by function with new allocations. Likely used by the
* caller to free on error.
*
* Allocate the required number of page directory pointers. Extremely similar to
* gen8_ppgtt_alloc_page_directories() and gen8_ppgtt_alloc_pagetabs().
* The main difference is here we are limited by the pml4 boundary (instead of
* the page directory pointer).
*
* Return: 0 if success; negative error code otherwise.
*/
static int
gen8_ppgtt_alloc_page_dirpointers(struct i915_address_space *vm,
struct i915_pml4 *pml4,
uint64_t start,
uint64_t length,
unsigned long *new_pdps)
{
struct drm_device *dev = vm->dev;
struct i915_page_directory_pointer *pdp;
uint64_t temp;
uint32_t pml4e;
WARN_ON(!bitmap_empty(new_pdps, GEN8_PML4ES_PER_PML4));
gen8_for_each_pml4e(pdp, pml4, start, length, temp, pml4e) {
if (!test_bit(pml4e, pml4->used_pml4es)) {
pdp = alloc_pdp(dev);
if (IS_ERR(pdp))
goto unwind_out;
gen8_initialize_pdp(vm, pdp);
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
pml4->pdps[pml4e] = pdp;
__set_bit(pml4e, new_pdps);
trace_i915_page_directory_pointer_entry_alloc(vm,
pml4e,
start,
GEN8_PML4E_SHIFT);
}
}
return 0;
unwind_out:
for_each_set_bit(pml4e, new_pdps, GEN8_PML4ES_PER_PML4)
free_pdp(dev, pml4->pdps[pml4e]);
return -ENOMEM;
}
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
static void
free_gen8_temp_bitmaps(unsigned long *new_pds, unsigned long *new_pts)
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
{
kfree(new_pts);
kfree(new_pds);
}
/* Fills in the page directory bitmap, and the array of page tables bitmap. Both
* of these are based on the number of PDPEs in the system.
*/
static
int __must_check alloc_gen8_temp_bitmaps(unsigned long **new_pds,
unsigned long **new_pts,
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
uint32_t pdpes)
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
{
unsigned long *pds;
unsigned long *pts;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
pds = kcalloc(BITS_TO_LONGS(pdpes), sizeof(unsigned long), GFP_TEMPORARY);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
if (!pds)
return -ENOMEM;
pts = kcalloc(pdpes, BITS_TO_LONGS(I915_PDES) * sizeof(unsigned long),
GFP_TEMPORARY);
if (!pts)
goto err_out;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
*new_pds = pds;
*new_pts = pts;
return 0;
err_out:
free_gen8_temp_bitmaps(pds, pts);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
return -ENOMEM;
}
/* PDE TLBs are a pain to invalidate on GEN8+. When we modify
* the page table structures, we mark them dirty so that
* context switching/execlist queuing code takes extra steps
* to ensure that tlbs are flushed.
*/
static void mark_tlbs_dirty(struct i915_hw_ppgtt *ppgtt)
{
ppgtt->pd_dirty_rings = INTEL_INFO(ppgtt->base.dev)->ring_mask;
}
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
static int gen8_alloc_va_range_3lvl(struct i915_address_space *vm,
struct i915_page_directory_pointer *pdp,
uint64_t start,
uint64_t length)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
unsigned long *new_page_dirs, *new_page_tables;
2015-07-30 10:02:03 +00:00
struct drm_device *dev = vm->dev;
struct i915_page_directory *pd;
const uint64_t orig_start = start;
const uint64_t orig_length = length;
uint64_t temp;
uint32_t pdpe;
2015-07-30 10:02:03 +00:00
uint32_t pdpes = I915_PDPES_PER_PDP(dev);
int ret;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/* Wrap is never okay since we can only represent 48b, and we don't
* actually use the other side of the canonical address space.
*/
if (WARN_ON(start + length < start))
return -ENODEV;
2015-07-30 10:02:03 +00:00
if (WARN_ON(start + length > vm->total))
return -ENODEV;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
ret = alloc_gen8_temp_bitmaps(&new_page_dirs, &new_page_tables, pdpes);
if (ret)
return ret;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/* Do the allocations first so we can easily bail out */
2015-07-30 10:02:03 +00:00
ret = gen8_ppgtt_alloc_page_directories(vm, pdp, start, length,
new_page_dirs);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
if (ret) {
free_gen8_temp_bitmaps(new_page_dirs, new_page_tables);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
return ret;
}
/* For every page directory referenced, allocate page tables */
2015-07-30 10:02:03 +00:00
gen8_for_each_pdpe(pd, pdp, start, length, temp, pdpe) {
ret = gen8_ppgtt_alloc_pagetabs(vm, pd, start, length,
new_page_tables + pdpe * BITS_TO_LONGS(I915_PDES));
if (ret)
goto err_out;
}
start = orig_start;
length = orig_length;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/* Allocations have completed successfully, so set the bitmaps, and do
* the mappings. */
2015-07-30 10:02:03 +00:00
gen8_for_each_pdpe(pd, pdp, start, length, temp, pdpe) {
gen8_pde_t *const page_directory = kmap_px(pd);
struct i915_page_table *pt;
uint64_t pd_len = length;
uint64_t pd_start = start;
uint32_t pde;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/* Every pd should be allocated, we just did that above. */
WARN_ON(!pd);
gen8_for_each_pde(pt, pd, pd_start, pd_len, temp, pde) {
/* Same reasoning as pd */
WARN_ON(!pt);
WARN_ON(!pd_len);
WARN_ON(!gen8_pte_count(pd_start, pd_len));
/* Set our used ptes within the page table */
bitmap_set(pt->used_ptes,
gen8_pte_index(pd_start),
gen8_pte_count(pd_start, pd_len));
/* Our pde is now pointing to the pagetable, pt */
__set_bit(pde, pd->used_pdes);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/* Map the PDE to the page table */
page_directory[pde] = gen8_pde_encode(px_dma(pt),
I915_CACHE_LLC);
trace_i915_page_table_entry_map(&ppgtt->base, pde, pt,
gen8_pte_index(start),
gen8_pte_count(start, length),
GEN8_PTES);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
/* NB: We haven't yet mapped ptes to pages. At this
* point we're still relying on insert_entries() */
}
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
kunmap_px(ppgtt, page_directory);
2015-07-30 10:02:03 +00:00
__set_bit(pdpe, pdp->used_pdpes);
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
gen8_setup_page_directory(ppgtt, pdp, pd, pdpe);
}
free_gen8_temp_bitmaps(new_page_dirs, new_page_tables);
mark_tlbs_dirty(ppgtt);
return 0;
err_out:
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
while (pdpe--) {
for_each_set_bit(temp, new_page_tables + pdpe *
BITS_TO_LONGS(I915_PDES), I915_PDES)
2015-07-30 10:02:03 +00:00
free_pt(dev, pdp->page_directory[pdpe]->page_table[temp]);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
}
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
for_each_set_bit(pdpe, new_page_dirs, pdpes)
2015-07-30 10:02:03 +00:00
free_pd(dev, pdp->page_directory[pdpe]);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
free_gen8_temp_bitmaps(new_page_dirs, new_page_tables);
mark_tlbs_dirty(ppgtt);
return ret;
}
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
static int gen8_alloc_va_range_4lvl(struct i915_address_space *vm,
struct i915_pml4 *pml4,
uint64_t start,
uint64_t length)
{
DECLARE_BITMAP(new_pdps, GEN8_PML4ES_PER_PML4);
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
struct i915_page_directory_pointer *pdp;
uint64_t temp, pml4e;
int ret = 0;
/* Do the pml4 allocations first, so we don't need to track the newly
* allocated tables below the pdp */
bitmap_zero(new_pdps, GEN8_PML4ES_PER_PML4);
/* The pagedirectory and pagetable allocations are done in the shared 3
* and 4 level code. Just allocate the pdps.
*/
ret = gen8_ppgtt_alloc_page_dirpointers(vm, pml4, start, length,
new_pdps);
if (ret)
return ret;
WARN(bitmap_weight(new_pdps, GEN8_PML4ES_PER_PML4) > 2,
"The allocation has spanned more than 512GB. "
"It is highly likely this is incorrect.");
gen8_for_each_pml4e(pdp, pml4, start, length, temp, pml4e) {
WARN_ON(!pdp);
ret = gen8_alloc_va_range_3lvl(vm, pdp, start, length);
if (ret)
goto err_out;
gen8_setup_page_directory_pointer(ppgtt, pml4, pdp, pml4e);
}
bitmap_or(pml4->used_pml4es, new_pdps, pml4->used_pml4es,
GEN8_PML4ES_PER_PML4);
return 0;
err_out:
for_each_set_bit(pml4e, new_pdps, GEN8_PML4ES_PER_PML4)
gen8_ppgtt_cleanup_3lvl(vm->dev, pml4->pdps[pml4e]);
return ret;
}
static int gen8_alloc_va_range(struct i915_address_space *vm,
uint64_t start, uint64_t length)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
if (USES_FULL_48BIT_PPGTT(vm->dev))
return gen8_alloc_va_range_4lvl(vm, &ppgtt->pml4, start, length);
else
return gen8_alloc_va_range_3lvl(vm, &ppgtt->pdp, start, length);
}
static void gen8_dump_pdp(struct i915_page_directory_pointer *pdp,
uint64_t start, uint64_t length,
gen8_pte_t scratch_pte,
struct seq_file *m)
{
struct i915_page_directory *pd;
uint64_t temp;
uint32_t pdpe;
gen8_for_each_pdpe(pd, pdp, start, length, temp, pdpe) {
struct i915_page_table *pt;
uint64_t pd_len = length;
uint64_t pd_start = start;
uint32_t pde;
if (!test_bit(pdpe, pdp->used_pdpes))
continue;
seq_printf(m, "\tPDPE #%d\n", pdpe);
gen8_for_each_pde(pt, pd, pd_start, pd_len, temp, pde) {
uint32_t pte;
gen8_pte_t *pt_vaddr;
if (!test_bit(pde, pd->used_pdes))
continue;
pt_vaddr = kmap_px(pt);
for (pte = 0; pte < GEN8_PTES; pte += 4) {
uint64_t va =
(pdpe << GEN8_PDPE_SHIFT) |
(pde << GEN8_PDE_SHIFT) |
(pte << GEN8_PTE_SHIFT);
int i;
bool found = false;
for (i = 0; i < 4; i++)
if (pt_vaddr[pte + i] != scratch_pte)
found = true;
if (!found)
continue;
seq_printf(m, "\t\t0x%llx [%03d,%03d,%04d]: =", va, pdpe, pde, pte);
for (i = 0; i < 4; i++) {
if (pt_vaddr[pte + i] != scratch_pte)
seq_printf(m, " %llx", pt_vaddr[pte + i]);
else
seq_puts(m, " SCRATCH ");
}
seq_puts(m, "\n");
}
/* don't use kunmap_px, it could trigger
* an unnecessary flush.
*/
kunmap_atomic(pt_vaddr);
}
}
}
static void gen8_dump_ppgtt(struct i915_hw_ppgtt *ppgtt, struct seq_file *m)
{
struct i915_address_space *vm = &ppgtt->base;
uint64_t start = ppgtt->base.start;
uint64_t length = ppgtt->base.total;
gen8_pte_t scratch_pte = gen8_pte_encode(px_dma(vm->scratch_page),
I915_CACHE_LLC, true);
if (!USES_FULL_48BIT_PPGTT(vm->dev)) {
gen8_dump_pdp(&ppgtt->pdp, start, length, scratch_pte, m);
} else {
uint64_t templ4, pml4e;
struct i915_pml4 *pml4 = &ppgtt->pml4;
struct i915_page_directory_pointer *pdp;
gen8_for_each_pml4e(pdp, pml4, start, length, templ4, pml4e) {
if (!test_bit(pml4e, pml4->used_pml4es))
continue;
seq_printf(m, " PML4E #%llu\n", pml4e);
gen8_dump_pdp(pdp, start, length, scratch_pte, m);
}
}
}
static int gen8_preallocate_top_level_pdps(struct i915_hw_ppgtt *ppgtt)
{
unsigned long *new_page_dirs, *new_page_tables;
uint32_t pdpes = I915_PDPES_PER_PDP(dev);
int ret;
/* We allocate temp bitmap for page tables for no gain
* but as this is for init only, lets keep the things simple
*/
ret = alloc_gen8_temp_bitmaps(&new_page_dirs, &new_page_tables, pdpes);
if (ret)
return ret;
/* Allocate for all pdps regardless of how the ppgtt
* was defined.
*/
ret = gen8_ppgtt_alloc_page_directories(&ppgtt->base, &ppgtt->pdp,
0, 1ULL << 32,
new_page_dirs);
if (!ret)
*ppgtt->pdp.used_pdpes = *new_page_dirs;
free_gen8_temp_bitmaps(new_page_dirs, new_page_tables);
return ret;
}
/*
* GEN8 legacy ppgtt programming is accomplished through a max 4 PDP registers
* with a net effect resembling a 2-level page table in normal x86 terms. Each
* PDP represents 1GB of memory 4 * 512 * 512 * 4096 = 4GB legacy 32b address
* space.
*
*/
static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt)
{
int ret;
ret = gen8_init_scratch(&ppgtt->base);
if (ret)
return ret;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
ppgtt->base.start = 0;
ppgtt->base.cleanup = gen8_ppgtt_cleanup;
ppgtt->base.allocate_va_range = gen8_alloc_va_range;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
ppgtt->base.insert_entries = gen8_ppgtt_insert_entries;
ppgtt->base.clear_range = gen8_ppgtt_clear_range;
ppgtt->base.unbind_vma = ppgtt_unbind_vma;
ppgtt->base.bind_vma = ppgtt_bind_vma;
ppgtt->debug_dump = gen8_dump_ppgtt;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
if (USES_FULL_48BIT_PPGTT(ppgtt->base.dev)) {
ret = setup_px(ppgtt->base.dev, &ppgtt->pml4);
if (ret)
goto free_scratch;
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
gen8_initialize_pml4(&ppgtt->base, &ppgtt->pml4);
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
ppgtt->base.total = 1ULL << 48;
2015-07-30 10:06:23 +00:00
ppgtt->switch_mm = gen8_48b_mm_switch;
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
} else {
ret = __pdp_init(ppgtt->base.dev, &ppgtt->pdp);
if (ret)
goto free_scratch;
ppgtt->base.total = 1ULL << 32;
2015-07-30 10:06:23 +00:00
ppgtt->switch_mm = gen8_legacy_mm_switch;
drm/i915/gen8: implement alloc/free for 4lvl PML4 has no special attributes, and there will always be a PML4. So simply initialize it at creation, and destroy it at the end. The code for 4lvl is able to call into the existing 3lvl page table code to handle all of the lower levels. v2: Return something at the end of gen8_alloc_va_range_4lvl to keep the compiler happy. And define ret only in one place. Updated gen8_ppgtt_unmap_pages and gen8_ppgtt_free to handle 4lvl. v3: Use i915_dma_unmap_single instead of pci API. Fix a couple of incorrect checks when unmapping pdp and pd pages (Akash). v4: Call __pdp_fini also for 32b PPGTT. Clean up alloc_pdp param list. v5: Prevent (harmless) out of range access in gen8_for_each_pml4e. v6: Simplify alloc_vma_range_4lvl and gen8_ppgtt_init_common error paths. (Akash) v7: Rebase, s/gen8_ppgtt_free_*/gen8_ppgtt_cleanup_*/. v8: Change location of pml4_init/fini. It will make next patches cleaner. v9: Rebase after Mika's ppgtt cleanup / scratch merge patch series, while trying to reuse as much as possible for pdp alloc. pml4_init/fini replaced by setup/cleanup_px macros. v10: Rebase after Mika's merged ppgtt cleanup patch series. v11: Rebase after final merged version of Mika's ppgtt/scratch patches. v12: Fix pdpe start value in trace (Akash) v13: Define all 4lvl functions in this patch directly, instead of previous patches, add i915_page_directory_pointer_entry_alloc here, use test_bit to detect when pdp is already allocated (Akash). v14: Move pdp allocation into a new gen8_ppgtt_alloc_page_dirpointers funtion, as we do for pds and pts; move pd and pdp setup functions to this patch (Akash). v15: Added kfree(pdp) from previous patch to this (Akash). Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-30 10:05:29 +00:00
trace_i915_page_directory_pointer_entry_alloc(&ppgtt->base,
0, 0,
GEN8_PML4E_SHIFT);
if (intel_vgpu_active(ppgtt->base.dev)) {
ret = gen8_preallocate_top_level_pdps(ppgtt);
if (ret)
goto free_scratch;
}
}
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
if (intel_vgpu_active(ppgtt->base.dev))
gen8_ppgtt_notify_vgt(ppgtt, true);
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
return 0;
drm/i915/gen8: Make pdp allocation more dynamic This transitional patch doesn't do much for the existing code. However, it should make upcoming patches to use the full 48b address space a bit easier. 32-bit ppgtt uses just 4 PDPs, while 48-bit ppgtt will have up-to 512; this patch prepares the existing functions to query the right number of pdps at run-time. This also means that used_pdpes should also be allocated during ppgtt_init, as the bitmap size will depend on the ppgtt address range selected. v2: Renamed pdp_free to be similar to pd/pt (unmap_and_free_pdp). v3: To facilitate testing, 48b mode will be available on Broadwell and GEN9+, when i915.enable_ppgtt = 3. v4: Rebase after s/page_tables/page_table/, added extra information about 4-level page table formats and use IS_ENABLED macro. v5: Check CONFIG_X86_64 instead of CONFIG_64BIT. v6: Rebase after Mika's ppgtt cleanup / scratch merge patch series, and follow his nomenclature in pdp functions (there is no alloc_pdp yet). v7: Rebase after merged version of Mika's ppgtt cleanup patch series. v8: Rebase after final merged version of Mika's ppgtt/scratch patches. v9: Introduce PML4 (and 48-bit checks) until next patch (Akash). v10: Also use test_bit to detect when pd/pt are already allocated (Akash) Cc: Akash Goel <akash.goel@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Akash Goel <akash.goel@intel.com> [danvet: Amend commit message as suggested by Michel.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-07-29 16:23:46 +00:00
free_scratch:
gen8_free_scratch(&ppgtt->base);
return ret;
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
}
static void gen6_dump_ppgtt(struct i915_hw_ppgtt *ppgtt, struct seq_file *m)
{
struct i915_address_space *vm = &ppgtt->base;
struct i915_page_table *unused;
gen6_pte_t scratch_pte;
uint32_t pd_entry;
uint32_t pte, pde, temp;
uint32_t start = ppgtt->base.start, length = ppgtt->base.total;
scratch_pte = vm->pte_encode(px_dma(vm->scratch_page),
I915_CACHE_LLC, true, 0);
gen6_for_each_pde(unused, &ppgtt->pd, start, length, temp, pde) {
u32 expected;
gen6_pte_t *pt_vaddr;
const dma_addr_t pt_addr = px_dma(ppgtt->pd.page_table[pde]);
pd_entry = readl(ppgtt->pd_addr + pde);
expected = (GEN6_PDE_ADDR_ENCODE(pt_addr) | GEN6_PDE_VALID);
if (pd_entry != expected)
seq_printf(m, "\tPDE #%d mismatch: Actual PDE: %x Expected PDE: %x\n",
pde,
pd_entry,
expected);
seq_printf(m, "\tPDE: %x\n", pd_entry);
pt_vaddr = kmap_px(ppgtt->pd.page_table[pde]);
for (pte = 0; pte < GEN6_PTES; pte+=4) {
unsigned long va =
(pde * PAGE_SIZE * GEN6_PTES) +
(pte * PAGE_SIZE);
int i;
bool found = false;
for (i = 0; i < 4; i++)
if (pt_vaddr[pte + i] != scratch_pte)
found = true;
if (!found)
continue;
seq_printf(m, "\t\t0x%lx [%03d,%04d]: =", va, pde, pte);
for (i = 0; i < 4; i++) {
if (pt_vaddr[pte + i] != scratch_pte)
seq_printf(m, " %08x", pt_vaddr[pte + i]);
else
seq_puts(m, " SCRATCH ");
}
seq_puts(m, "\n");
}
kunmap_px(ppgtt, pt_vaddr);
}
}
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
/* Write pde (index) from the page directory @pd to the page table @pt */
static void gen6_write_pde(struct i915_page_directory *pd,
const int pde, struct i915_page_table *pt)
{
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
/* Caller needs to make sure the write completes if necessary */
struct i915_hw_ppgtt *ppgtt =
container_of(pd, struct i915_hw_ppgtt, pd);
u32 pd_entry;
pd_entry = GEN6_PDE_ADDR_ENCODE(px_dma(pt));
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
pd_entry |= GEN6_PDE_VALID;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
writel(pd_entry, ppgtt->pd_addr + pde);
}
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
/* Write all the page tables found in the ppgtt structure to incrementing page
* directories. */
static void gen6_write_page_range(struct drm_i915_private *dev_priv,
struct i915_page_directory *pd,
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
uint32_t start, uint32_t length)
{
struct i915_page_table *pt;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
uint32_t pde, temp;
gen6_for_each_pde(pt, pd, start, length, temp, pde)
gen6_write_pde(pd, pde, pt);
/* Make sure write is complete before other code can use this page
* table. Also require for WC mapped PTEs */
readl(dev_priv->gtt.gsm);
}
static uint32_t get_pd_offset(struct i915_hw_ppgtt *ppgtt)
{
BUG_ON(ppgtt->pd.base.ggtt_offset & 0x3f);
return (ppgtt->pd.base.ggtt_offset / 64) << 16;
}
static int hsw_mm_switch(struct i915_hw_ppgtt *ppgtt,
struct drm_i915_gem_request *req)
{
struct intel_engine_cs *ring = req->ring;
int ret;
/* NB: TLBs must be flushed and invalidated before a switch */
ret = ring->flush(req, I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
if (ret)
return ret;
ret = intel_ring_begin(req, 6);
if (ret)
return ret;
intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(2));
intel_ring_emit(ring, RING_PP_DIR_DCLV(ring));
intel_ring_emit(ring, PP_DIR_DCLV_2G);
intel_ring_emit(ring, RING_PP_DIR_BASE(ring));
intel_ring_emit(ring, get_pd_offset(ppgtt));
intel_ring_emit(ring, MI_NOOP);
intel_ring_advance(ring);
return 0;
}
static int vgpu_mm_switch(struct i915_hw_ppgtt *ppgtt,
struct drm_i915_gem_request *req)
{
struct intel_engine_cs *ring = req->ring;
struct drm_i915_private *dev_priv = to_i915(ppgtt->base.dev);
I915_WRITE(RING_PP_DIR_DCLV(ring), PP_DIR_DCLV_2G);
I915_WRITE(RING_PP_DIR_BASE(ring), get_pd_offset(ppgtt));
return 0;
}
static int gen7_mm_switch(struct i915_hw_ppgtt *ppgtt,
struct drm_i915_gem_request *req)
{
struct intel_engine_cs *ring = req->ring;
int ret;
/* NB: TLBs must be flushed and invalidated before a switch */
ret = ring->flush(req, I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
if (ret)
return ret;
ret = intel_ring_begin(req, 6);
if (ret)
return ret;
intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(2));
intel_ring_emit(ring, RING_PP_DIR_DCLV(ring));
intel_ring_emit(ring, PP_DIR_DCLV_2G);
intel_ring_emit(ring, RING_PP_DIR_BASE(ring));
intel_ring_emit(ring, get_pd_offset(ppgtt));
intel_ring_emit(ring, MI_NOOP);
intel_ring_advance(ring);
/* XXX: RCS is the only one to auto invalidate the TLBs? */
if (ring->id != RCS) {
ret = ring->flush(req, I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
if (ret)
return ret;
}
return 0;
}
static int gen6_mm_switch(struct i915_hw_ppgtt *ppgtt,
struct drm_i915_gem_request *req)
{
struct intel_engine_cs *ring = req->ring;
struct drm_device *dev = ppgtt->base.dev;
struct drm_i915_private *dev_priv = dev->dev_private;
I915_WRITE(RING_PP_DIR_DCLV(ring), PP_DIR_DCLV_2G);
I915_WRITE(RING_PP_DIR_BASE(ring), get_pd_offset(ppgtt));
POSTING_READ(RING_PP_DIR_DCLV(ring));
return 0;
}
static void gen8_ppgtt_enable(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct intel_engine_cs *ring;
int j;
for_each_ring(ring, dev_priv, j) {
2015-07-30 10:06:23 +00:00
u32 four_level = USES_FULL_48BIT_PPGTT(dev) ? GEN8_GFX_PPGTT_48B : 0;
I915_WRITE(RING_MODE_GEN7(ring),
2015-07-30 10:06:23 +00:00
_MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE | four_level));
}
}
static void gen7_ppgtt_enable(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct intel_engine_cs *ring;
uint32_t ecochk, ecobits;
int i;
ecobits = I915_READ(GAC_ECO_BITS);
I915_WRITE(GAC_ECO_BITS, ecobits | ECOBITS_PPGTT_CACHE64B);
ecochk = I915_READ(GAM_ECOCHK);
if (IS_HASWELL(dev)) {
ecochk |= ECOCHK_PPGTT_WB_HSW;
} else {
ecochk |= ECOCHK_PPGTT_LLC_IVB;
ecochk &= ~ECOCHK_PPGTT_GFDT_IVB;
}
I915_WRITE(GAM_ECOCHK, ecochk);
for_each_ring(ring, dev_priv, i) {
/* GFX_MODE is per-ring on gen7+ */
I915_WRITE(RING_MODE_GEN7(ring),
_MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE));
}
}
static void gen6_ppgtt_enable(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
uint32_t ecochk, gab_ctl, ecobits;
ecobits = I915_READ(GAC_ECO_BITS);
I915_WRITE(GAC_ECO_BITS, ecobits | ECOBITS_SNB_BIT |
ECOBITS_PPGTT_CACHE64B);
gab_ctl = I915_READ(GAB_CTL);
I915_WRITE(GAB_CTL, gab_ctl | GAB_CTL_CONT_AFTER_PAGEFAULT);
ecochk = I915_READ(GAM_ECOCHK);
I915_WRITE(GAM_ECOCHK, ecochk | ECOCHK_SNB_BIT | ECOCHK_PPGTT_CACHE64B);
I915_WRITE(GFX_MODE, _MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE));
}
/* PPGTT support for Sandybdrige/Gen6 and later */
static void gen6_ppgtt_clear_range(struct i915_address_space *vm,
uint64_t start,
uint64_t length,
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
bool use_scratch)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
gen6_pte_t *pt_vaddr, scratch_pte;
unsigned first_entry = start >> PAGE_SHIFT;
unsigned num_entries = length >> PAGE_SHIFT;
unsigned act_pt = first_entry / GEN6_PTES;
unsigned first_pte = first_entry % GEN6_PTES;
unsigned last_pte, i;
scratch_pte = vm->pte_encode(px_dma(vm->scratch_page),
I915_CACHE_LLC, true, 0);
while (num_entries) {
last_pte = first_pte + num_entries;
if (last_pte > GEN6_PTES)
last_pte = GEN6_PTES;
pt_vaddr = kmap_px(ppgtt->pd.page_table[act_pt]);
for (i = first_pte; i < last_pte; i++)
pt_vaddr[i] = scratch_pte;
kunmap_px(ppgtt, pt_vaddr);
num_entries -= last_pte - first_pte;
first_pte = 0;
act_pt++;
}
}
static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,
struct sg_table *pages,
uint64_t start,
enum i915_cache_level cache_level, u32 flags)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
gen6_pte_t *pt_vaddr;
unsigned first_entry = start >> PAGE_SHIFT;
unsigned act_pt = first_entry / GEN6_PTES;
unsigned act_pte = first_entry % GEN6_PTES;
struct sg_page_iter sg_iter;
drm/i915: Avoid dereference past end of page array in gen6_ppgtt_insert_entries() [ 89.237347] BUG: unable to handle kernel paging request at ffff880096326000 [ 89.237369] IP: [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237382] PGD 2272067 PUD 25df0e067 PMD 25de5c067 PTE 8000000096326060 [ 89.237394] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC [ 89.237404] CPU: 1 PID: 1981 Comm: gem_concurrent_ Not tainted 3.13.0-rc4+ #639 [ 89.237411] Hardware name: Intel Corporation 2012 Client Platform/Emerald Lake 2, BIOS ACRVMBY1.86C.0078.P00.1201161002 01/16/2012 [ 89.237420] task: ffff88024c038030 ti: ffff88024b130000 task.ti: ffff88024b130000 [ 89.237425] RIP: 0010:[<ffffffff81347227>] [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237435] RSP: 0018:ffff88024b131ae0 EFLAGS: 00010286 [ 89.237440] RAX: ffff880096325000 RBX: 0000000000000400 RCX: 0000000000001000 [ 89.237445] RDX: 0000000000000200 RSI: 0000000000000001 RDI: 0000000000000010 [ 89.237451] RBP: ffff88024b131b30 R08: ffff88024cc3aef0 R09: 0000000000000000 [ 89.237456] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88024cc3ae00 [ 89.237462] R13: ffff88024a578000 R14: 0000000000000001 R15: ffff88024a578ffc [ 89.237469] FS: 00007ff5475d8900(0000) GS:ffff88025d020000(0000) knlGS:0000000000000000 [ 89.237475] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 89.237480] CR2: ffff880096326000 CR3: 000000024d531000 CR4: 00000000001407e0 [ 89.237485] Stack: [ 89.237488] ffff880000000000 0000020000000000 ffff88024b23f2c0 0000000100000000 [ 89.237499] 0000000000000001 000000000007ffff ffff8801e7bf5ac0 ffff8801e7bf5ac0 [ 89.237510] ffff88024cc3ae00 ffff880248a2ee40 ffff88024b131b58 ffffffff813455ed [ 89.237521] Call Trace: [ 89.237528] [<ffffffff813455ed>] ppgtt_bind_vma+0x3d/0x60 [ 89.237534] [<ffffffff8133d8dc>] i915_gem_object_pin+0x55c/0x6a0 [ 89.237541] [<ffffffff8134275b>] i915_gem_execbuffer_reserve_vma.isra.14+0x5b/0x110 [ 89.237548] [<ffffffff81342a88>] i915_gem_execbuffer_reserve+0x278/0x2c0 [ 89.237555] [<ffffffff81343d29>] i915_gem_do_execbuffer.isra.22+0x699/0x1250 [ 89.237562] [<ffffffff81344d91>] ? i915_gem_execbuffer2+0x51/0x290 [ 89.237569] [<ffffffff81344de6>] i915_gem_execbuffer2+0xa6/0x290 [ 89.237575] [<ffffffff813014f2>] drm_ioctl+0x4d2/0x610 [ 89.237582] [<ffffffff81080bf1>] ? cpuacct_account_field+0xa1/0xc0 [ 89.237588] [<ffffffff81080b55>] ? cpuacct_account_field+0x5/0xc0 [ 89.237597] [<ffffffff811371c0>] do_vfs_ioctl+0x300/0x520 [ 89.237603] [<ffffffff810757a1>] ? vtime_account_user+0x91/0xa0 [ 89.237610] [<ffffffff810e40eb>] ? context_tracking_user_exit+0x9b/0xe0 [ 89.237617] [<ffffffff81083d7d>] ? trace_hardirqs_on+0xd/0x10 [ 89.237623] [<ffffffff81137425>] SyS_ioctl+0x45/0x80 [ 89.237630] [<ffffffff815afffa>] tracesys+0xd4/0xd9 [ 89.237634] Code: 5b 41 5c 41 5d 41 5e 41 5f 5d c3 66 0f 1f 84 00 00 00 00 00 83 45 bc 01 49 8b 84 24 78 01 00 00 65 ff 0c 25 e0 b8 00 00 8b 55 bc <4c> 8b 2c d0 65 ff 04 25 e0 b8 00 00 49 8b 45 00 48 c1 e8 2d 48 [ 89.237741] RIP [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237749] RSP <ffff88024b131ae0> [ 89.237753] CR2: ffff880096326000 [ 89.237758] ---[ end trace 27416ba8b18d496c ]--- This bug dates back to the original introduction of the gen6_ppgtt_insert_entries() Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Ben Widawsky <benjamin.widawsky@intel.com> Reviewed-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Dropped cc: stable since without full ppgtt there's no way we'll access the last page directory with this function since that range is occupied (only in the allocator) with the ppgtt pdes. Without aliasing we can start to use that range and blow up.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-31 15:50:30 +00:00
pt_vaddr = NULL;
for_each_sg_page(pages->sgl, &sg_iter, pages->nents, 0) {
drm/i915: Avoid dereference past end of page array in gen6_ppgtt_insert_entries() [ 89.237347] BUG: unable to handle kernel paging request at ffff880096326000 [ 89.237369] IP: [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237382] PGD 2272067 PUD 25df0e067 PMD 25de5c067 PTE 8000000096326060 [ 89.237394] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC [ 89.237404] CPU: 1 PID: 1981 Comm: gem_concurrent_ Not tainted 3.13.0-rc4+ #639 [ 89.237411] Hardware name: Intel Corporation 2012 Client Platform/Emerald Lake 2, BIOS ACRVMBY1.86C.0078.P00.1201161002 01/16/2012 [ 89.237420] task: ffff88024c038030 ti: ffff88024b130000 task.ti: ffff88024b130000 [ 89.237425] RIP: 0010:[<ffffffff81347227>] [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237435] RSP: 0018:ffff88024b131ae0 EFLAGS: 00010286 [ 89.237440] RAX: ffff880096325000 RBX: 0000000000000400 RCX: 0000000000001000 [ 89.237445] RDX: 0000000000000200 RSI: 0000000000000001 RDI: 0000000000000010 [ 89.237451] RBP: ffff88024b131b30 R08: ffff88024cc3aef0 R09: 0000000000000000 [ 89.237456] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88024cc3ae00 [ 89.237462] R13: ffff88024a578000 R14: 0000000000000001 R15: ffff88024a578ffc [ 89.237469] FS: 00007ff5475d8900(0000) GS:ffff88025d020000(0000) knlGS:0000000000000000 [ 89.237475] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 89.237480] CR2: ffff880096326000 CR3: 000000024d531000 CR4: 00000000001407e0 [ 89.237485] Stack: [ 89.237488] ffff880000000000 0000020000000000 ffff88024b23f2c0 0000000100000000 [ 89.237499] 0000000000000001 000000000007ffff ffff8801e7bf5ac0 ffff8801e7bf5ac0 [ 89.237510] ffff88024cc3ae00 ffff880248a2ee40 ffff88024b131b58 ffffffff813455ed [ 89.237521] Call Trace: [ 89.237528] [<ffffffff813455ed>] ppgtt_bind_vma+0x3d/0x60 [ 89.237534] [<ffffffff8133d8dc>] i915_gem_object_pin+0x55c/0x6a0 [ 89.237541] [<ffffffff8134275b>] i915_gem_execbuffer_reserve_vma.isra.14+0x5b/0x110 [ 89.237548] [<ffffffff81342a88>] i915_gem_execbuffer_reserve+0x278/0x2c0 [ 89.237555] [<ffffffff81343d29>] i915_gem_do_execbuffer.isra.22+0x699/0x1250 [ 89.237562] [<ffffffff81344d91>] ? i915_gem_execbuffer2+0x51/0x290 [ 89.237569] [<ffffffff81344de6>] i915_gem_execbuffer2+0xa6/0x290 [ 89.237575] [<ffffffff813014f2>] drm_ioctl+0x4d2/0x610 [ 89.237582] [<ffffffff81080bf1>] ? cpuacct_account_field+0xa1/0xc0 [ 89.237588] [<ffffffff81080b55>] ? cpuacct_account_field+0x5/0xc0 [ 89.237597] [<ffffffff811371c0>] do_vfs_ioctl+0x300/0x520 [ 89.237603] [<ffffffff810757a1>] ? vtime_account_user+0x91/0xa0 [ 89.237610] [<ffffffff810e40eb>] ? context_tracking_user_exit+0x9b/0xe0 [ 89.237617] [<ffffffff81083d7d>] ? trace_hardirqs_on+0xd/0x10 [ 89.237623] [<ffffffff81137425>] SyS_ioctl+0x45/0x80 [ 89.237630] [<ffffffff815afffa>] tracesys+0xd4/0xd9 [ 89.237634] Code: 5b 41 5c 41 5d 41 5e 41 5f 5d c3 66 0f 1f 84 00 00 00 00 00 83 45 bc 01 49 8b 84 24 78 01 00 00 65 ff 0c 25 e0 b8 00 00 8b 55 bc <4c> 8b 2c d0 65 ff 04 25 e0 b8 00 00 49 8b 45 00 48 c1 e8 2d 48 [ 89.237741] RIP [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237749] RSP <ffff88024b131ae0> [ 89.237753] CR2: ffff880096326000 [ 89.237758] ---[ end trace 27416ba8b18d496c ]--- This bug dates back to the original introduction of the gen6_ppgtt_insert_entries() Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Ben Widawsky <benjamin.widawsky@intel.com> Reviewed-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Dropped cc: stable since without full ppgtt there's no way we'll access the last page directory with this function since that range is occupied (only in the allocator) with the ppgtt pdes. Without aliasing we can start to use that range and blow up.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-31 15:50:30 +00:00
if (pt_vaddr == NULL)
pt_vaddr = kmap_px(ppgtt->pd.page_table[act_pt]);
drm/i915: Avoid dereference past end of page array in gen6_ppgtt_insert_entries() [ 89.237347] BUG: unable to handle kernel paging request at ffff880096326000 [ 89.237369] IP: [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237382] PGD 2272067 PUD 25df0e067 PMD 25de5c067 PTE 8000000096326060 [ 89.237394] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC [ 89.237404] CPU: 1 PID: 1981 Comm: gem_concurrent_ Not tainted 3.13.0-rc4+ #639 [ 89.237411] Hardware name: Intel Corporation 2012 Client Platform/Emerald Lake 2, BIOS ACRVMBY1.86C.0078.P00.1201161002 01/16/2012 [ 89.237420] task: ffff88024c038030 ti: ffff88024b130000 task.ti: ffff88024b130000 [ 89.237425] RIP: 0010:[<ffffffff81347227>] [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237435] RSP: 0018:ffff88024b131ae0 EFLAGS: 00010286 [ 89.237440] RAX: ffff880096325000 RBX: 0000000000000400 RCX: 0000000000001000 [ 89.237445] RDX: 0000000000000200 RSI: 0000000000000001 RDI: 0000000000000010 [ 89.237451] RBP: ffff88024b131b30 R08: ffff88024cc3aef0 R09: 0000000000000000 [ 89.237456] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88024cc3ae00 [ 89.237462] R13: ffff88024a578000 R14: 0000000000000001 R15: ffff88024a578ffc [ 89.237469] FS: 00007ff5475d8900(0000) GS:ffff88025d020000(0000) knlGS:0000000000000000 [ 89.237475] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 89.237480] CR2: ffff880096326000 CR3: 000000024d531000 CR4: 00000000001407e0 [ 89.237485] Stack: [ 89.237488] ffff880000000000 0000020000000000 ffff88024b23f2c0 0000000100000000 [ 89.237499] 0000000000000001 000000000007ffff ffff8801e7bf5ac0 ffff8801e7bf5ac0 [ 89.237510] ffff88024cc3ae00 ffff880248a2ee40 ffff88024b131b58 ffffffff813455ed [ 89.237521] Call Trace: [ 89.237528] [<ffffffff813455ed>] ppgtt_bind_vma+0x3d/0x60 [ 89.237534] [<ffffffff8133d8dc>] i915_gem_object_pin+0x55c/0x6a0 [ 89.237541] [<ffffffff8134275b>] i915_gem_execbuffer_reserve_vma.isra.14+0x5b/0x110 [ 89.237548] [<ffffffff81342a88>] i915_gem_execbuffer_reserve+0x278/0x2c0 [ 89.237555] [<ffffffff81343d29>] i915_gem_do_execbuffer.isra.22+0x699/0x1250 [ 89.237562] [<ffffffff81344d91>] ? i915_gem_execbuffer2+0x51/0x290 [ 89.237569] [<ffffffff81344de6>] i915_gem_execbuffer2+0xa6/0x290 [ 89.237575] [<ffffffff813014f2>] drm_ioctl+0x4d2/0x610 [ 89.237582] [<ffffffff81080bf1>] ? cpuacct_account_field+0xa1/0xc0 [ 89.237588] [<ffffffff81080b55>] ? cpuacct_account_field+0x5/0xc0 [ 89.237597] [<ffffffff811371c0>] do_vfs_ioctl+0x300/0x520 [ 89.237603] [<ffffffff810757a1>] ? vtime_account_user+0x91/0xa0 [ 89.237610] [<ffffffff810e40eb>] ? context_tracking_user_exit+0x9b/0xe0 [ 89.237617] [<ffffffff81083d7d>] ? trace_hardirqs_on+0xd/0x10 [ 89.237623] [<ffffffff81137425>] SyS_ioctl+0x45/0x80 [ 89.237630] [<ffffffff815afffa>] tracesys+0xd4/0xd9 [ 89.237634] Code: 5b 41 5c 41 5d 41 5e 41 5f 5d c3 66 0f 1f 84 00 00 00 00 00 83 45 bc 01 49 8b 84 24 78 01 00 00 65 ff 0c 25 e0 b8 00 00 8b 55 bc <4c> 8b 2c d0 65 ff 04 25 e0 b8 00 00 49 8b 45 00 48 c1 e8 2d 48 [ 89.237741] RIP [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237749] RSP <ffff88024b131ae0> [ 89.237753] CR2: ffff880096326000 [ 89.237758] ---[ end trace 27416ba8b18d496c ]--- This bug dates back to the original introduction of the gen6_ppgtt_insert_entries() Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Ben Widawsky <benjamin.widawsky@intel.com> Reviewed-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Dropped cc: stable since without full ppgtt there's no way we'll access the last page directory with this function since that range is occupied (only in the allocator) with the ppgtt pdes. Without aliasing we can start to use that range and blow up.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-31 15:50:30 +00:00
pt_vaddr[act_pte] =
vm->pte_encode(sg_page_iter_dma_address(&sg_iter),
cache_level, true, flags);
if (++act_pte == GEN6_PTES) {
kunmap_px(ppgtt, pt_vaddr);
drm/i915: Avoid dereference past end of page array in gen6_ppgtt_insert_entries() [ 89.237347] BUG: unable to handle kernel paging request at ffff880096326000 [ 89.237369] IP: [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237382] PGD 2272067 PUD 25df0e067 PMD 25de5c067 PTE 8000000096326060 [ 89.237394] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC [ 89.237404] CPU: 1 PID: 1981 Comm: gem_concurrent_ Not tainted 3.13.0-rc4+ #639 [ 89.237411] Hardware name: Intel Corporation 2012 Client Platform/Emerald Lake 2, BIOS ACRVMBY1.86C.0078.P00.1201161002 01/16/2012 [ 89.237420] task: ffff88024c038030 ti: ffff88024b130000 task.ti: ffff88024b130000 [ 89.237425] RIP: 0010:[<ffffffff81347227>] [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237435] RSP: 0018:ffff88024b131ae0 EFLAGS: 00010286 [ 89.237440] RAX: ffff880096325000 RBX: 0000000000000400 RCX: 0000000000001000 [ 89.237445] RDX: 0000000000000200 RSI: 0000000000000001 RDI: 0000000000000010 [ 89.237451] RBP: ffff88024b131b30 R08: ffff88024cc3aef0 R09: 0000000000000000 [ 89.237456] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88024cc3ae00 [ 89.237462] R13: ffff88024a578000 R14: 0000000000000001 R15: ffff88024a578ffc [ 89.237469] FS: 00007ff5475d8900(0000) GS:ffff88025d020000(0000) knlGS:0000000000000000 [ 89.237475] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 89.237480] CR2: ffff880096326000 CR3: 000000024d531000 CR4: 00000000001407e0 [ 89.237485] Stack: [ 89.237488] ffff880000000000 0000020000000000 ffff88024b23f2c0 0000000100000000 [ 89.237499] 0000000000000001 000000000007ffff ffff8801e7bf5ac0 ffff8801e7bf5ac0 [ 89.237510] ffff88024cc3ae00 ffff880248a2ee40 ffff88024b131b58 ffffffff813455ed [ 89.237521] Call Trace: [ 89.237528] [<ffffffff813455ed>] ppgtt_bind_vma+0x3d/0x60 [ 89.237534] [<ffffffff8133d8dc>] i915_gem_object_pin+0x55c/0x6a0 [ 89.237541] [<ffffffff8134275b>] i915_gem_execbuffer_reserve_vma.isra.14+0x5b/0x110 [ 89.237548] [<ffffffff81342a88>] i915_gem_execbuffer_reserve+0x278/0x2c0 [ 89.237555] [<ffffffff81343d29>] i915_gem_do_execbuffer.isra.22+0x699/0x1250 [ 89.237562] [<ffffffff81344d91>] ? i915_gem_execbuffer2+0x51/0x290 [ 89.237569] [<ffffffff81344de6>] i915_gem_execbuffer2+0xa6/0x290 [ 89.237575] [<ffffffff813014f2>] drm_ioctl+0x4d2/0x610 [ 89.237582] [<ffffffff81080bf1>] ? cpuacct_account_field+0xa1/0xc0 [ 89.237588] [<ffffffff81080b55>] ? cpuacct_account_field+0x5/0xc0 [ 89.237597] [<ffffffff811371c0>] do_vfs_ioctl+0x300/0x520 [ 89.237603] [<ffffffff810757a1>] ? vtime_account_user+0x91/0xa0 [ 89.237610] [<ffffffff810e40eb>] ? context_tracking_user_exit+0x9b/0xe0 [ 89.237617] [<ffffffff81083d7d>] ? trace_hardirqs_on+0xd/0x10 [ 89.237623] [<ffffffff81137425>] SyS_ioctl+0x45/0x80 [ 89.237630] [<ffffffff815afffa>] tracesys+0xd4/0xd9 [ 89.237634] Code: 5b 41 5c 41 5d 41 5e 41 5f 5d c3 66 0f 1f 84 00 00 00 00 00 83 45 bc 01 49 8b 84 24 78 01 00 00 65 ff 0c 25 e0 b8 00 00 8b 55 bc <4c> 8b 2c d0 65 ff 04 25 e0 b8 00 00 49 8b 45 00 48 c1 e8 2d 48 [ 89.237741] RIP [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237749] RSP <ffff88024b131ae0> [ 89.237753] CR2: ffff880096326000 [ 89.237758] ---[ end trace 27416ba8b18d496c ]--- This bug dates back to the original introduction of the gen6_ppgtt_insert_entries() Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Ben Widawsky <benjamin.widawsky@intel.com> Reviewed-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Dropped cc: stable since without full ppgtt there's no way we'll access the last page directory with this function since that range is occupied (only in the allocator) with the ppgtt pdes. Without aliasing we can start to use that range and blow up.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-31 15:50:30 +00:00
pt_vaddr = NULL;
act_pt++;
act_pte = 0;
}
}
drm/i915: Avoid dereference past end of page array in gen6_ppgtt_insert_entries() [ 89.237347] BUG: unable to handle kernel paging request at ffff880096326000 [ 89.237369] IP: [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237382] PGD 2272067 PUD 25df0e067 PMD 25de5c067 PTE 8000000096326060 [ 89.237394] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC [ 89.237404] CPU: 1 PID: 1981 Comm: gem_concurrent_ Not tainted 3.13.0-rc4+ #639 [ 89.237411] Hardware name: Intel Corporation 2012 Client Platform/Emerald Lake 2, BIOS ACRVMBY1.86C.0078.P00.1201161002 01/16/2012 [ 89.237420] task: ffff88024c038030 ti: ffff88024b130000 task.ti: ffff88024b130000 [ 89.237425] RIP: 0010:[<ffffffff81347227>] [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237435] RSP: 0018:ffff88024b131ae0 EFLAGS: 00010286 [ 89.237440] RAX: ffff880096325000 RBX: 0000000000000400 RCX: 0000000000001000 [ 89.237445] RDX: 0000000000000200 RSI: 0000000000000001 RDI: 0000000000000010 [ 89.237451] RBP: ffff88024b131b30 R08: ffff88024cc3aef0 R09: 0000000000000000 [ 89.237456] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88024cc3ae00 [ 89.237462] R13: ffff88024a578000 R14: 0000000000000001 R15: ffff88024a578ffc [ 89.237469] FS: 00007ff5475d8900(0000) GS:ffff88025d020000(0000) knlGS:0000000000000000 [ 89.237475] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 89.237480] CR2: ffff880096326000 CR3: 000000024d531000 CR4: 00000000001407e0 [ 89.237485] Stack: [ 89.237488] ffff880000000000 0000020000000000 ffff88024b23f2c0 0000000100000000 [ 89.237499] 0000000000000001 000000000007ffff ffff8801e7bf5ac0 ffff8801e7bf5ac0 [ 89.237510] ffff88024cc3ae00 ffff880248a2ee40 ffff88024b131b58 ffffffff813455ed [ 89.237521] Call Trace: [ 89.237528] [<ffffffff813455ed>] ppgtt_bind_vma+0x3d/0x60 [ 89.237534] [<ffffffff8133d8dc>] i915_gem_object_pin+0x55c/0x6a0 [ 89.237541] [<ffffffff8134275b>] i915_gem_execbuffer_reserve_vma.isra.14+0x5b/0x110 [ 89.237548] [<ffffffff81342a88>] i915_gem_execbuffer_reserve+0x278/0x2c0 [ 89.237555] [<ffffffff81343d29>] i915_gem_do_execbuffer.isra.22+0x699/0x1250 [ 89.237562] [<ffffffff81344d91>] ? i915_gem_execbuffer2+0x51/0x290 [ 89.237569] [<ffffffff81344de6>] i915_gem_execbuffer2+0xa6/0x290 [ 89.237575] [<ffffffff813014f2>] drm_ioctl+0x4d2/0x610 [ 89.237582] [<ffffffff81080bf1>] ? cpuacct_account_field+0xa1/0xc0 [ 89.237588] [<ffffffff81080b55>] ? cpuacct_account_field+0x5/0xc0 [ 89.237597] [<ffffffff811371c0>] do_vfs_ioctl+0x300/0x520 [ 89.237603] [<ffffffff810757a1>] ? vtime_account_user+0x91/0xa0 [ 89.237610] [<ffffffff810e40eb>] ? context_tracking_user_exit+0x9b/0xe0 [ 89.237617] [<ffffffff81083d7d>] ? trace_hardirqs_on+0xd/0x10 [ 89.237623] [<ffffffff81137425>] SyS_ioctl+0x45/0x80 [ 89.237630] [<ffffffff815afffa>] tracesys+0xd4/0xd9 [ 89.237634] Code: 5b 41 5c 41 5d 41 5e 41 5f 5d c3 66 0f 1f 84 00 00 00 00 00 83 45 bc 01 49 8b 84 24 78 01 00 00 65 ff 0c 25 e0 b8 00 00 8b 55 bc <4c> 8b 2c d0 65 ff 04 25 e0 b8 00 00 49 8b 45 00 48 c1 e8 2d 48 [ 89.237741] RIP [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170 [ 89.237749] RSP <ffff88024b131ae0> [ 89.237753] CR2: ffff880096326000 [ 89.237758] ---[ end trace 27416ba8b18d496c ]--- This bug dates back to the original introduction of the gen6_ppgtt_insert_entries() Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Ben Widawsky <benjamin.widawsky@intel.com> Reviewed-by: Ben Widawsky <ben@bwidawsk.net> [danvet: Dropped cc: stable since without full ppgtt there's no way we'll access the last page directory with this function since that range is occupied (only in the allocator) with the ppgtt pdes. Without aliasing we can start to use that range and blow up.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-31 15:50:30 +00:00
if (pt_vaddr)
kunmap_px(ppgtt, pt_vaddr);
}
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
static int gen6_alloc_va_range(struct i915_address_space *vm,
uint64_t start_in, uint64_t length_in)
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
{
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
DECLARE_BITMAP(new_page_tables, I915_PDES);
struct drm_device *dev = vm->dev;
struct drm_i915_private *dev_priv = dev->dev_private;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
struct i915_page_table *pt;
uint32_t start, length, start_save, length_save;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
uint32_t pde, temp;
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
int ret;
if (WARN_ON(start_in + length_in > ppgtt->base.total))
return -ENODEV;
start = start_save = start_in;
length = length_save = length_in;
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
bitmap_zero(new_page_tables, I915_PDES);
/* The allocation is done in two stages so that we can bail out with
* minimal amount of pain. The first stage finds new page tables that
* need allocation. The second stage marks use ptes within the page
* tables.
*/
gen6_for_each_pde(pt, &ppgtt->pd, start, length, temp, pde) {
if (pt != vm->scratch_pt) {
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
WARN_ON(bitmap_empty(pt->used_ptes, GEN6_PTES));
continue;
}
/* We've already allocated a page table */
WARN_ON(!bitmap_empty(pt->used_ptes, GEN6_PTES));
pt = alloc_pt(dev);
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
if (IS_ERR(pt)) {
ret = PTR_ERR(pt);
goto unwind_out;
}
gen6_initialize_pt(vm, pt);
ppgtt->pd.page_table[pde] = pt;
__set_bit(pde, new_page_tables);
trace_i915_page_table_entry_alloc(vm, pde, start, GEN6_PDE_SHIFT);
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
}
start = start_save;
length = length_save;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
gen6_for_each_pde(pt, &ppgtt->pd, start, length, temp, pde) {
DECLARE_BITMAP(tmp_bitmap, GEN6_PTES);
bitmap_zero(tmp_bitmap, GEN6_PTES);
bitmap_set(tmp_bitmap, gen6_pte_index(start),
gen6_pte_count(start, length));
if (__test_and_clear_bit(pde, new_page_tables))
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
gen6_write_pde(&ppgtt->pd, pde, pt);
trace_i915_page_table_entry_map(vm, pde, pt,
gen6_pte_index(start),
gen6_pte_count(start, length),
GEN6_PTES);
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
bitmap_or(pt->used_ptes, tmp_bitmap, pt->used_ptes,
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
GEN6_PTES);
}
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
WARN_ON(!bitmap_empty(new_page_tables, I915_PDES));
/* Make sure write is complete before other code can use this page
* table. Also require for WC mapped PTEs */
readl(dev_priv->gtt.gsm);
drm/i915: Track page table reload need This patch was formerly known as, "Force pd restore when PDEs change, gen6-7." I had to change the name because it is needed for GEN8 too. The real issue this is trying to solve is when a new object is mapped into the current address space. The GPU does not snoop the new mapping so we must do the gen specific action to reload the page tables. GEN8 and GEN7 do differ in the way they load page tables for the RCS. GEN8 does so with the context restore, while GEN7 requires the proper load commands in the command streamer. Non-render is similar for both. Caveat for GEN7 The docs say you cannot change the PDEs of a currently running context. We never map new PDEs of a running context, and expect them to be present - so I think this is okay. (We can unmap, but this should also be okay since we only unmap unreferenced objects that the GPU shouldn't be tryingto va->pa xlate.) The MI_SET_CONTEXT command does have a flag to signal that even if the context is the same, force a reload. It's unclear exactly what this does, but I have a hunch it's the right thing to do. The logic assumes that we always emit a context switch after mapping new PDEs, and before we submit a batch. This is the case today, and has been the case since the inception of hardware contexts. A note in the comment let's the user know. It's not just for gen8. If the current context has mappings change, we need a context reload to switch v2: Rebased after ppgtt clean up patches. Split the warning for aliasing and true ppgtt options. And do not break aliasing ppgtt, where to->ppgtt is always null. v3: Invalidate PPGTT TLBs inside alloc_va_range. v4: Rename ppgtt_invalidate_tlbs to mark_tlbs_dirty and move pd_dirty_rings from i915_address_space to i915_hw_ppgtt. Fixes when neither ctx->ppgtt and aliasing_ppgtt exist. v5: Removed references to teardown_va_range. v6: Updated needs_pd_load_pre/post. v7: Fix pd_dirty_rings check in needs_pd_load_post, and update/move comment about updated PDEs to object_pin/bind (Mika). Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-19 12:53:28 +00:00
mark_tlbs_dirty(ppgtt);
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
return 0;
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
unwind_out:
for_each_set_bit(pde, new_page_tables, I915_PDES) {
struct i915_page_table *pt = ppgtt->pd.page_table[pde];
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
ppgtt->pd.page_table[pde] = vm->scratch_pt;
free_pt(vm->dev, pt);
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
}
mark_tlbs_dirty(ppgtt);
return ret;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
}
static int gen6_init_scratch(struct i915_address_space *vm)
{
struct drm_device *dev = vm->dev;
vm->scratch_page = alloc_scratch_page(dev);
if (IS_ERR(vm->scratch_page))
return PTR_ERR(vm->scratch_page);
vm->scratch_pt = alloc_pt(dev);
if (IS_ERR(vm->scratch_pt)) {
free_scratch_page(dev, vm->scratch_page);
return PTR_ERR(vm->scratch_pt);
}
gen6_initialize_pt(vm, vm->scratch_pt);
return 0;
}
static void gen6_free_scratch(struct i915_address_space *vm)
{
struct drm_device *dev = vm->dev;
free_pt(dev, vm->scratch_pt);
free_scratch_page(dev, vm->scratch_page);
}
static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
{
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt, base);
struct i915_page_table *pt;
uint32_t pde;
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
drm_mm_remove_node(&ppgtt->node);
gen6_for_all_pdes(pt, ppgtt, pde) {
if (pt != vm->scratch_pt)
free_pt(ppgtt->base.dev, pt);
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
}
drm/i915: Create page table allocators As we move toward dynamic page table allocation, it becomes much easier to manage our data structures if break do things less coarsely by breaking up all of our actions into individual tasks. This makes the code easier to write, read, and verify. Aside from the dissection of the allocation functions, the patch statically allocates the page table structures without a page directory. This remains the same for all platforms, The patch itself should not have much functional difference. The primary noticeable difference is the fact that page tables are no longer allocated, but rather statically declared as part of the page directory. This has non-zero overhead, but things gain additional complexity as a result. This patch exists for a few reasons: 1. Splitting out the functions allows easily combining GEN6 and GEN8 code. Page tables have no difference based on GEN8. As we'll see in a future patch when we add the DMA mappings to the allocations, it requires only one small change to make work, and error handling should just fall into place. 2. Unless we always want to allocate all page tables under a given PDE, we'll have to eventually break this up into an array of pointers (or pointer to pointer). 3. Having the discrete functions is easier to review, and understand. All allocations and frees now take place in just a couple of locations. Reviewing, and catching leaks should be easy. 4. Less important: the GFP flags are confined to one location, which makes playing around with such things trivial. v2: Updated commit message to explain why this patch exists v3: For lrc, s/pdp.page_directory[i].daddr/pdp.page_directory[i]->daddr/ v4: Renamed free_pt/pd_single functions to unmap_and_free_pt/pd (Daniel) v5: Added additional safety checks in gen8 clear/free/unmap. v6: Use WARN_ON and return -EINVAL in alloc_pt_range (Mika). v7: Make err_out loop symmetrical to the way we allocate in alloc_pt_range. Also s/page_tables/page_table and correct commit message (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-24 16:22:36 +00:00
gen6_free_scratch(vm);
}
static int gen6_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt)
{
struct i915_address_space *vm = &ppgtt->base;
struct drm_device *dev = ppgtt->base.dev;
struct drm_i915_private *dev_priv = dev->dev_private;
bool retried = false;
int ret;
drm/i915: Use drm_mm for PPGTT PDEs When PPGTT support was originally enabled, it was only designed to support 1 PPGTT. It therefore made sense to simply hide the GGTT space required to enable this from the drm_mm allocator. Since we intend to support full PPGTT, which means more than 1, and they can be created and destroyed ad hoc it will be required to use the proper allocation techniques we already have. The first step here is to make the existing single PPGTT use the allocator. The astute observer will notice that we are reserving space in the GGTT for the PDEs for the lifetime of the address space, and would be right to question whether or not this is a good idea. It does not make a difference with this current patch only the aliasing PPGTT (indeed the PDEs should still be hidden from the shrinker). For the future, we are allocating from top to bottom to avoid using the precious "gtt space" The GGTT space at that point should only be used for scanout, HW contexts, ringbuffers, HWSP, PDEs, and a couple of other small buffers (potentially) used by the kernel. Everything else should be mapped into a PPGTT. To put the consumption in more tangible terms, it takes approximately 4 sets of PDEs to equal one 19x10 framebuffer (with no fancy stride or alignment constraints). 3/4 of the total [average] GGTT can be used for PDEs, and hopefully never touch the 1/4 that the framebuffer needs. The astute, and persistent observer might ask about the page tables which are also pinned for the address space. This waste is unfortunate. We use 2MB of memory per address space. We leave wrapping the PDEs as a real GEM object as a TODO. v2: Align PDEs to 64b in GTT Allocate the node dynamically so we can use drm_mm_put_block Now tested on IGT Allocate node at the top to avoid fragmentation (Chris) v3: Use Chris' top down allocator v4: Embed drm_mm_node into ppgtt struct (Jesse) Remove hunks which didn't belong (Jesse) v5: Don't subtract guard page since we now killed the guard page prior to this patch. (Ben) v6: Rebased and removed guard page stuff. Added a chunk to the commit message Allow adding a context to mappable region v7: Undo v3, so we can make the drm patch last in the series Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v4) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> squash: drm/i915: allow PPGTT to use mappable Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:11:07 +00:00
/* PPGTT PDEs reside in the GGTT and consists of 512 entries. The
* allocator works in address space sizes, so it's multiplied by page
* size. We allocate at the top of the GTT to avoid fragmentation.
*/
BUG_ON(!drm_mm_initialized(&dev_priv->gtt.base.mm));
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
ret = gen6_init_scratch(vm);
if (ret)
return ret;
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
alloc:
drm/i915: Use drm_mm for PPGTT PDEs When PPGTT support was originally enabled, it was only designed to support 1 PPGTT. It therefore made sense to simply hide the GGTT space required to enable this from the drm_mm allocator. Since we intend to support full PPGTT, which means more than 1, and they can be created and destroyed ad hoc it will be required to use the proper allocation techniques we already have. The first step here is to make the existing single PPGTT use the allocator. The astute observer will notice that we are reserving space in the GGTT for the PDEs for the lifetime of the address space, and would be right to question whether or not this is a good idea. It does not make a difference with this current patch only the aliasing PPGTT (indeed the PDEs should still be hidden from the shrinker). For the future, we are allocating from top to bottom to avoid using the precious "gtt space" The GGTT space at that point should only be used for scanout, HW contexts, ringbuffers, HWSP, PDEs, and a couple of other small buffers (potentially) used by the kernel. Everything else should be mapped into a PPGTT. To put the consumption in more tangible terms, it takes approximately 4 sets of PDEs to equal one 19x10 framebuffer (with no fancy stride or alignment constraints). 3/4 of the total [average] GGTT can be used for PDEs, and hopefully never touch the 1/4 that the framebuffer needs. The astute, and persistent observer might ask about the page tables which are also pinned for the address space. This waste is unfortunate. We use 2MB of memory per address space. We leave wrapping the PDEs as a real GEM object as a TODO. v2: Align PDEs to 64b in GTT Allocate the node dynamically so we can use drm_mm_put_block Now tested on IGT Allocate node at the top to avoid fragmentation (Chris) v3: Use Chris' top down allocator v4: Embed drm_mm_node into ppgtt struct (Jesse) Remove hunks which didn't belong (Jesse) v5: Don't subtract guard page since we now killed the guard page prior to this patch. (Ben) v6: Rebased and removed guard page stuff. Added a chunk to the commit message Allow adding a context to mappable region v7: Undo v3, so we can make the drm patch last in the series Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v4) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> squash: drm/i915: allow PPGTT to use mappable Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:11:07 +00:00
ret = drm_mm_insert_node_in_range_generic(&dev_priv->gtt.base.mm,
&ppgtt->node, GEN6_PD_SIZE,
GEN6_PD_ALIGN, 0,
0, dev_priv->gtt.base.total,
DRM_MM_TOPDOWN);
if (ret == -ENOSPC && !retried) {
ret = i915_gem_evict_something(dev, &dev_priv->gtt.base,
GEN6_PD_SIZE, GEN6_PD_ALIGN,
drm/i915: Prevent negative relocation deltas from wrapping This is pure evil. Userspace, I'm looking at you SNA, repacks batch buffers on the fly after generation as they are being passed to the kernel for execution. These batches also contain self-referenced relocations as a single buffer encompasses the state commands, kernels, vertices and sampler. During generation the buffers are placed at known offsets within the full batch, and then the relocation deltas (as passed to the kernel) are tweaked as the batch is repacked into a smaller buffer. This means that userspace is passing negative relocations deltas, which subsequently wrap to large values if the batch is at a low address. The GPU hangs when it then tries to use the large value as a base for its address offsets, rather than wrapping back to the real value (as one would hope). As the GPU uses positive offsets from the base, we can treat the relocation address as the minimum address read by the GPU. For the upper bound, we trust that userspace will not read beyond the end of the buffer. So, how do we fix negative relocations from wrapping? We can either check that every relocation looks valid when we write it, and then position each object such that we prevent the offset wraparound, or we just special-case the self-referential behaviour of SNA and force all batches to be above 256k. Daniel prefers the latter approach. This fixes a GPU hang when it tries to use an address (relocation + offset) greater than the GTT size. The issue would occur quite easily with full-ppgtt as each fd gets its own VM space, so low offsets would often be handed out. However, with the rearrangement of the low GTT due to capturing the BIOS framebuffer, it is already affecting kernels 3.15 onwards. I think only IVB+ is susceptible to this bug, but the workaround should only kick in rarely, so it seems sensible to always apply it. v3: Use a bias for batch buffers to prevent small negative delta relocations from wrapping. v4 from Daniel: - s/BIAS/BATCH_OFFSET_BIAS/ - Extract eb_vma_misplaced/i915_vma_misplaced since the conditions were growing rather cumbersome. - Add a comment to eb_get_batch explaining why we do this. - Apply the batch offset bias everywhere but mention that we've only observed it on gen7 gpus. - Drop PIN_OFFSET_FIX for now, that slipped in from a feature patch. v5: Add static to eb_get_batch, spotted by 0-day tester. Testcase: igt/gem_bad_reloc Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78533 Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> (v3) Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-05-23 06:48:08 +00:00
I915_CACHE_NONE,
0, dev_priv->gtt.base.total,
0);
if (ret)
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
goto err_out;
retried = true;
goto alloc;
}
drm/i915: Use drm_mm for PPGTT PDEs When PPGTT support was originally enabled, it was only designed to support 1 PPGTT. It therefore made sense to simply hide the GGTT space required to enable this from the drm_mm allocator. Since we intend to support full PPGTT, which means more than 1, and they can be created and destroyed ad hoc it will be required to use the proper allocation techniques we already have. The first step here is to make the existing single PPGTT use the allocator. The astute observer will notice that we are reserving space in the GGTT for the PDEs for the lifetime of the address space, and would be right to question whether or not this is a good idea. It does not make a difference with this current patch only the aliasing PPGTT (indeed the PDEs should still be hidden from the shrinker). For the future, we are allocating from top to bottom to avoid using the precious "gtt space" The GGTT space at that point should only be used for scanout, HW contexts, ringbuffers, HWSP, PDEs, and a couple of other small buffers (potentially) used by the kernel. Everything else should be mapped into a PPGTT. To put the consumption in more tangible terms, it takes approximately 4 sets of PDEs to equal one 19x10 framebuffer (with no fancy stride or alignment constraints). 3/4 of the total [average] GGTT can be used for PDEs, and hopefully never touch the 1/4 that the framebuffer needs. The astute, and persistent observer might ask about the page tables which are also pinned for the address space. This waste is unfortunate. We use 2MB of memory per address space. We leave wrapping the PDEs as a real GEM object as a TODO. v2: Align PDEs to 64b in GTT Allocate the node dynamically so we can use drm_mm_put_block Now tested on IGT Allocate node at the top to avoid fragmentation (Chris) v3: Use Chris' top down allocator v4: Embed drm_mm_node into ppgtt struct (Jesse) Remove hunks which didn't belong (Jesse) v5: Don't subtract guard page since we now killed the guard page prior to this patch. (Ben) v6: Rebased and removed guard page stuff. Added a chunk to the commit message Allow adding a context to mappable region v7: Undo v3, so we can make the drm patch last in the series Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v4) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> squash: drm/i915: allow PPGTT to use mappable Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:11:07 +00:00
if (ret)
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
goto err_out;
drm/i915: Use drm_mm for PPGTT PDEs When PPGTT support was originally enabled, it was only designed to support 1 PPGTT. It therefore made sense to simply hide the GGTT space required to enable this from the drm_mm allocator. Since we intend to support full PPGTT, which means more than 1, and they can be created and destroyed ad hoc it will be required to use the proper allocation techniques we already have. The first step here is to make the existing single PPGTT use the allocator. The astute observer will notice that we are reserving space in the GGTT for the PDEs for the lifetime of the address space, and would be right to question whether or not this is a good idea. It does not make a difference with this current patch only the aliasing PPGTT (indeed the PDEs should still be hidden from the shrinker). For the future, we are allocating from top to bottom to avoid using the precious "gtt space" The GGTT space at that point should only be used for scanout, HW contexts, ringbuffers, HWSP, PDEs, and a couple of other small buffers (potentially) used by the kernel. Everything else should be mapped into a PPGTT. To put the consumption in more tangible terms, it takes approximately 4 sets of PDEs to equal one 19x10 framebuffer (with no fancy stride or alignment constraints). 3/4 of the total [average] GGTT can be used for PDEs, and hopefully never touch the 1/4 that the framebuffer needs. The astute, and persistent observer might ask about the page tables which are also pinned for the address space. This waste is unfortunate. We use 2MB of memory per address space. We leave wrapping the PDEs as a real GEM object as a TODO. v2: Align PDEs to 64b in GTT Allocate the node dynamically so we can use drm_mm_put_block Now tested on IGT Allocate node at the top to avoid fragmentation (Chris) v3: Use Chris' top down allocator v4: Embed drm_mm_node into ppgtt struct (Jesse) Remove hunks which didn't belong (Jesse) v5: Don't subtract guard page since we now killed the guard page prior to this patch. (Ben) v6: Rebased and removed guard page stuff. Added a chunk to the commit message Allow adding a context to mappable region v7: Undo v3, so we can make the drm patch last in the series Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v4) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> squash: drm/i915: allow PPGTT to use mappable Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:11:07 +00:00
if (ppgtt->node.start < dev_priv->gtt.mappable_end)
DRM_DEBUG("Forced to use aperture for PDEs\n");
return 0;
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
err_out:
gen6_free_scratch(vm);
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
return ret;
}
static int gen6_ppgtt_alloc(struct i915_hw_ppgtt *ppgtt)
{
return gen6_ppgtt_allocate_page_directories(ppgtt);
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
}
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
static void gen6_scratch_va_range(struct i915_hw_ppgtt *ppgtt,
uint64_t start, uint64_t length)
{
struct i915_page_table *unused;
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
uint32_t pde, temp;
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
gen6_for_each_pde(unused, &ppgtt->pd, start, length, temp, pde)
ppgtt->pd.page_table[pde] = ppgtt->base.scratch_pt;
}
static int gen6_ppgtt_init(struct i915_hw_ppgtt *ppgtt)
{
struct drm_device *dev = ppgtt->base.dev;
struct drm_i915_private *dev_priv = dev->dev_private;
int ret;
ppgtt->base.pte_encode = dev_priv->gtt.base.pte_encode;
if (IS_GEN6(dev)) {
ppgtt->switch_mm = gen6_mm_switch;
} else if (IS_HASWELL(dev)) {
ppgtt->switch_mm = hsw_mm_switch;
} else if (IS_GEN7(dev)) {
ppgtt->switch_mm = gen7_mm_switch;
} else
BUG();
if (intel_vgpu_active(dev))
ppgtt->switch_mm = vgpu_mm_switch;
ret = gen6_ppgtt_alloc(ppgtt);
if (ret)
return ret;
ppgtt->base.allocate_va_range = gen6_alloc_va_range;
ppgtt->base.clear_range = gen6_ppgtt_clear_range;
ppgtt->base.insert_entries = gen6_ppgtt_insert_entries;
ppgtt->base.unbind_vma = ppgtt_unbind_vma;
ppgtt->base.bind_vma = ppgtt_bind_vma;
ppgtt->base.cleanup = gen6_ppgtt_cleanup;
ppgtt->base.start = 0;
ppgtt->base.total = I915_PDES * GEN6_PTES * PAGE_SIZE;
ppgtt->debug_dump = gen6_dump_ppgtt;
ppgtt->pd.base.ggtt_offset =
ppgtt->node.start / PAGE_SIZE * sizeof(gen6_pte_t);
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
ppgtt->pd_addr = (gen6_pte_t __iomem *)dev_priv->gtt.gsm +
ppgtt->pd.base.ggtt_offset / sizeof(gen6_pte_t);
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
gen6_scratch_va_range(ppgtt, 0, ppgtt->base.total);
drm/i915: Track GEN6 page table usage Instead of implementing the full tracking + dynamic allocation, this patch does a bit less than half of the work, by tracking and warning on unexpected conditions. The tracking itself follows which PTEs within a page table are currently being used for objects. The next patch will modify this to actually allocate the page tables only when necessary. With the current patch there isn't much in the way of making a gen agnostic range allocation function. However, in the next patch we'll add more specificity which makes having separate functions a bit easier to manage. One important change introduced here is that DMA mappings are created/destroyed at the same page directories/tables are allocated/deallocated. Notice that aliasing PPGTT is not managed here. The patch which actually begins dynamic allocation/teardown explains the reasoning for this. v2: s/pdp.page_directory/pdp.page_directories Make a scratch page allocation helper v3: Rebase and expand commit message. v4: Allocate required pagetables only when it is needed, _bind_to_vm instead of bind_vma (Daniel). v5: Rebased to remove the unnecessary noise in the diff, also: - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK. - Removed unnecessary checks in gen6_alloc_va_range. - Changed map/unmap_px_single macros to use dma functions directly and be part of a static inline function instead. - Moved drm_device plumbing through page tables operation to its own patch. - Moved allocate/teardown_va_range calls until they are fully implemented (in subsequent patch). - Merged pt and scratch_pt unmap_and_free path. - Moved scratch page allocator helper to the patch that will use it. v6: Reduce complexity by not tearing down pagetables dynamically, the same can be achieved while freeing empty vms. (Daniel) v7: s/i915_dma_map_px_single/i915_dma_map_single s/gen6_write_pdes/gen6_write_pde Prevent a NULL case when only GGTT is available. (Mika) v8: Rebased after s/page_tables/page_table/. v9: Reworked i915_pte_index and i915_pte_count. Also exercise bitmap allocation here (gen6_alloc_va_range) and fix incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika). Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-16 16:00:56 +00:00
gen6_write_page_range(dev_priv, &ppgtt->pd, 0, ppgtt->base.total);
DRM_DEBUG_DRIVER("Allocated pde space (%lldM) at GTT entry: %llx\n",
ppgtt->node.size >> 20,
ppgtt->node.start / PAGE_SIZE);
DRM_DEBUG("Adding PPGTT at offset %x\n",
ppgtt->pd.base.ggtt_offset << 10);
return 0;
}
static int __hw_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt)
{
ppgtt->base.dev = dev;
if (INTEL_INFO(dev)->gen < 8)
return gen6_ppgtt_init(ppgtt);
else
drm/i915/gen8: Dynamic page table allocations This finishes off the dynamic page tables allocations, in the legacy 3 level style that already exists. Most everything has already been setup to this point, the patch finishes off the enabling by setting the appropriate function pointers. In LRC mode, contexts need to know the PDPs when they are populated. With dynamic page table allocations, these PDPs may not exist yet. Check if PDPs have been allocated and use the scratch page if they do not exist yet. Before submission, update the PDPs in the logic ring context as PDPs have been allocated. v2: Update aliasing/true ppgtt allocate/teardown/clear functions for gen 6 & 7. v3: Rebase. v4: Remove BUG() from ppgtt_unbind_vma, but keep checking that either teardown_va_range or clear_range functions exist (Daniel). v5: Similar to gen6, in init, gen8_ppgtt_clear_range call is only needed for aliasing ppgtt. Zombie tracking was originally added for teardown function and is no longer required. v6: Update err_out case in gen8_alloc_va_range (missed from lastest rebase). v7: Rebase after s/page_tables/page_table/. v8: Updated scratch_pt check after scratch flag was removed in previous patch. v9: Note that lrc mode needs to be updated to support init state without any PDP. v10: Unmap correct page_table in gen8_alloc_va_range's error case, clean-up gen8_aliasing_ppgtt_init (remove duplicated map), and initialize PTs during page table allocation. v11: Squashed LRC enabling commit, otherwise LRC mode would be left broken until it was updated to handle the init case without any PDP. v12: Do not overallocate new_pts bitmap, make alloc_gen8_temp_bitmaps static and don't abuse of inline functions. (Mika) Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-04-08 11:13:34 +00:00
return gen8_ppgtt_init(ppgtt);
}
static void i915_address_space_init(struct i915_address_space *vm,
struct drm_i915_private *dev_priv)
{
drm_mm_init(&vm->mm, vm->start, vm->total);
vm->dev = dev_priv->dev;
INIT_LIST_HEAD(&vm->active_list);
INIT_LIST_HEAD(&vm->inactive_list);
list_add_tail(&vm->global_link, &dev_priv->vm_list);
}
int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt)
{
struct drm_i915_private *dev_priv = dev->dev_private;
int ret = 0;
ret = __hw_ppgtt_init(dev, ppgtt);
if (ret == 0) {
kref_init(&ppgtt->ref);
i915_address_space_init(&ppgtt->base, dev_priv);
}
return ret;
}
int i915_ppgtt_init_hw(struct drm_device *dev)
{
/* In the case of execlists, PPGTT is enabled by the context descriptor
* and the PDPs are contained within the context itself. We don't
* need to do anything here. */
if (i915.enable_execlists)
return 0;
if (!USES_PPGTT(dev))
return 0;
if (IS_GEN6(dev))
gen6_ppgtt_enable(dev);
else if (IS_GEN7(dev))
gen7_ppgtt_enable(dev);
else if (INTEL_INFO(dev)->gen >= 8)
gen8_ppgtt_enable(dev);
else
drm/i915: Use BUILD_BUG if possible in the i915 WARN_ON Faster feedback to errors is always better. This is inspired by the addition to WARN_ONs to mask/enable helpers for registers to make sure callers have the arguments ordered correctly: Pretty much always the arguments are static. We use WARN_ON(1) a lot in default switch statements though where we should always handle all cases. So add a new macro specifically for that. The idea to use __builtin_constant_p is from Chris Wilson. v2: Use the ({}) gcc-ism to avoid the static inline, suggested by Dave. My first attempt used __cond as the temp var, which is the same used by BUILD_BUG_ON, but with inverted sense. Hilarity ensued, so sprinkle i915 into the name. Also use a temporary variable to only evaluate the condition once, suggested by Damien. v3: It's crazy but apparently 32bit gcc can't compile out the BUILD_BUG_ON in a lot of cases and just falls over. I have no idea why, but until clue grows just disable this nifty idea on 32bit builds. Reported by 0-day builder. v4: Got it all wrong, apparently its the gcc version. We need 4.9+. Now reported by Imre. v5: Chris suggested to add the case to MISSING_CASE for speedier debug. v6: Even some gcc 4.9 versions don't see through the maze, so give up for now. Keep the skeleton and MISSING_CASE stuff though. Cc: Imre Deak <imre.deak@intel.com> Cc: Damien Lespiau <damien.lespiau@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Dave Gordon <david.s.gordon@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
2014-12-08 15:40:10 +00:00
MISSING_CASE(INTEL_INFO(dev)->gen);
return 0;
}
int i915_ppgtt_init_ring(struct drm_i915_gem_request *req)
{
struct drm_i915_private *dev_priv = req->ring->dev->dev_private;
struct i915_hw_ppgtt *ppgtt = dev_priv->mm.aliasing_ppgtt;
if (i915.enable_execlists)
return 0;
if (!ppgtt)
return 0;
return ppgtt->switch_mm(ppgtt, req);
}
struct i915_hw_ppgtt *
i915_ppgtt_create(struct drm_device *dev, struct drm_i915_file_private *fpriv)
{
struct i915_hw_ppgtt *ppgtt;
int ret;
ppgtt = kzalloc(sizeof(*ppgtt), GFP_KERNEL);
if (!ppgtt)
return ERR_PTR(-ENOMEM);
ret = i915_ppgtt_init(dev, ppgtt);
if (ret) {
kfree(ppgtt);
return ERR_PTR(ret);
}
ppgtt->file_priv = fpriv;
trace_i915_ppgtt_create(&ppgtt->base);
return ppgtt;
}
void i915_ppgtt_release(struct kref *kref)
{
struct i915_hw_ppgtt *ppgtt =
container_of(kref, struct i915_hw_ppgtt, ref);
trace_i915_ppgtt_release(&ppgtt->base);
/* vmas should already be unbound */
WARN_ON(!list_empty(&ppgtt->base.active_list));
WARN_ON(!list_empty(&ppgtt->base.inactive_list));
list_del(&ppgtt->base.global_link);
drm_mm_takedown(&ppgtt->base.mm);
ppgtt->base.cleanup(&ppgtt->base);
kfree(ppgtt);
}
extern int intel_iommu_gfx_mapped;
/* Certain Gen5 chipsets require require idling the GPU before
* unmapping anything from the GTT when VT-d is enabled.
*/
static bool needs_idle_maps(struct drm_device *dev)
{
#ifdef CONFIG_INTEL_IOMMU
/* Query intel_iommu to see if we need the workaround. Presumably that
* was loaded first.
*/
if (IS_GEN5(dev) && IS_MOBILE(dev) && intel_iommu_gfx_mapped)
return true;
#endif
return false;
}
static bool do_idling(struct drm_i915_private *dev_priv)
{
bool ret = dev_priv->mm.interruptible;
if (unlikely(dev_priv->gtt.do_idle_maps)) {
dev_priv->mm.interruptible = false;
if (i915_gpu_idle(dev_priv->dev)) {
DRM_ERROR("Couldn't idle GPU\n");
/* Wait a bit, in hopes it avoids the hang */
udelay(10);
}
}
return ret;
}
static void undo_idling(struct drm_i915_private *dev_priv, bool interruptible)
{
if (unlikely(dev_priv->gtt.do_idle_maps))
dev_priv->mm.interruptible = interruptible;
}
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
void i915_check_and_clear_faults(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct intel_engine_cs *ring;
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
int i;
if (INTEL_INFO(dev)->gen < 6)
return;
for_each_ring(ring, dev_priv, i) {
u32 fault_reg;
fault_reg = I915_READ(RING_FAULT_REG(ring));
if (fault_reg & RING_FAULT_VALID) {
DRM_DEBUG_DRIVER("Unexpected fault\n"
"\tAddr: 0x%08lx\n"
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
"\tAddress space: %s\n"
"\tSource ID: %d\n"
"\tType: %d\n",
fault_reg & PAGE_MASK,
fault_reg & RING_FAULT_GTTSEL_MASK ? "GGTT" : "PPGTT",
RING_FAULT_SRCID(fault_reg),
RING_FAULT_FAULT_TYPE(fault_reg));
I915_WRITE(RING_FAULT_REG(ring),
fault_reg & ~RING_FAULT_VALID);
}
}
POSTING_READ(RING_FAULT_REG(&dev_priv->ring[RCS]));
}
static void i915_ggtt_flush(struct drm_i915_private *dev_priv)
{
if (INTEL_INFO(dev_priv->dev)->gen < 6) {
intel_gtt_chipset_flush();
} else {
I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN);
POSTING_READ(GFX_FLSH_CNTL_GEN6);
}
}
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
void i915_gem_suspend_gtt_mappings(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
/* Don't bother messing with faults pre GEN6 as we have little
* documentation supporting that it's a good idea.
*/
if (INTEL_INFO(dev)->gen < 6)
return;
i915_check_and_clear_faults(dev);
dev_priv->gtt.base.clear_range(&dev_priv->gtt.base,
dev_priv->gtt.base.start,
dev_priv->gtt.base.total,
true);
i915_ggtt_flush(dev_priv);
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
}
int i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj)
{
if (!dma_map_sg(&obj->base.dev->pdev->dev,
obj->pages->sgl, obj->pages->nents,
PCI_DMA_BIDIRECTIONAL))
return -ENOSPC;
return 0;
}
static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)
{
#ifdef writeq
writeq(pte, addr);
#else
iowrite32((u32)pte, addr);
iowrite32(pte >> 32, addr + 4);
#endif
}
static void gen8_ggtt_insert_entries(struct i915_address_space *vm,
struct sg_table *st,
uint64_t start,
enum i915_cache_level level, u32 unused)
{
struct drm_i915_private *dev_priv = vm->dev->dev_private;
unsigned first_entry = start >> PAGE_SHIFT;
gen8_pte_t __iomem *gtt_entries =
(gen8_pte_t __iomem *)dev_priv->gtt.gsm + first_entry;
int i = 0;
struct sg_page_iter sg_iter;
dma_addr_t addr = 0; /* shut up gcc */
for_each_sg_page(st->sgl, &sg_iter, st->nents, 0) {
addr = sg_dma_address(sg_iter.sg) +
(sg_iter.sg_pgoffset << PAGE_SHIFT);
gen8_set_pte(&gtt_entries[i],
gen8_pte_encode(addr, level, true));
i++;
}
/*
* XXX: This serves as a posting read to make sure that the PTE has
* actually been updated. There is some concern that even though
* registers and PTEs are within the same BAR that they are potentially
* of NUMA access patterns. Therefore, even with the way we assume
* hardware should work, we must keep this posting read for paranoia.
*/
if (i != 0)
WARN_ON(readq(&gtt_entries[i-1])
!= gen8_pte_encode(addr, level, true));
/* This next bit makes the above posting read even more important. We
* want to flush the TLBs only after we're certain all the PTE updates
* have finished.
*/
I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN);
POSTING_READ(GFX_FLSH_CNTL_GEN6);
}
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
/*
* Binds an object into the global gtt with the specified cache level. The object
* will be accessible to the GPU via commands whose operands reference offsets
* within the global GTT as well as accessible by the GPU through the GMADR
* mapped BAR (dev_priv->mm.gtt->gtt).
*/
static void gen6_ggtt_insert_entries(struct i915_address_space *vm,
struct sg_table *st,
uint64_t start,
enum i915_cache_level level, u32 flags)
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
{
struct drm_i915_private *dev_priv = vm->dev->dev_private;
unsigned first_entry = start >> PAGE_SHIFT;
gen6_pte_t __iomem *gtt_entries =
(gen6_pte_t __iomem *)dev_priv->gtt.gsm + first_entry;
int i = 0;
struct sg_page_iter sg_iter;
dma_addr_t addr = 0;
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
for_each_sg_page(st->sgl, &sg_iter, st->nents, 0) {
addr = sg_page_iter_dma_address(&sg_iter);
iowrite32(vm->pte_encode(addr, level, true, flags), &gtt_entries[i]);
i++;
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
}
/* XXX: This serves as a posting read to make sure that the PTE has
* actually been updated. There is some concern that even though
* registers and PTEs are within the same BAR that they are potentially
* of NUMA access patterns. Therefore, even with the way we assume
* hardware should work, we must keep this posting read for paranoia.
*/
if (i != 0) {
unsigned long gtt = readl(&gtt_entries[i-1]);
WARN_ON(gtt != vm->pte_encode(addr, level, true, flags));
}
/* This next bit makes the above posting read even more important. We
* want to flush the TLBs only after we're certain all the PTE updates
* have finished.
*/
I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN);
POSTING_READ(GFX_FLSH_CNTL_GEN6);
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
}
static void gen8_ggtt_clear_range(struct i915_address_space *vm,
uint64_t start,
uint64_t length,
bool use_scratch)
{
struct drm_i915_private *dev_priv = vm->dev->dev_private;
unsigned first_entry = start >> PAGE_SHIFT;
unsigned num_entries = length >> PAGE_SHIFT;
gen8_pte_t scratch_pte, __iomem *gtt_base =
(gen8_pte_t __iomem *) dev_priv->gtt.gsm + first_entry;
const int max_entries = gtt_total_entries(dev_priv->gtt) - first_entry;
int i;
if (WARN(num_entries > max_entries,
"First entry = %d; Num entries = %d (max=%d)\n",
first_entry, num_entries, max_entries))
num_entries = max_entries;
scratch_pte = gen8_pte_encode(px_dma(vm->scratch_page),
I915_CACHE_LLC,
use_scratch);
for (i = 0; i < num_entries; i++)
gen8_set_pte(&gtt_base[i], scratch_pte);
readl(gtt_base);
}
static void gen6_ggtt_clear_range(struct i915_address_space *vm,
uint64_t start,
uint64_t length,
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
bool use_scratch)
{
struct drm_i915_private *dev_priv = vm->dev->dev_private;
unsigned first_entry = start >> PAGE_SHIFT;
unsigned num_entries = length >> PAGE_SHIFT;
gen6_pte_t scratch_pte, __iomem *gtt_base =
(gen6_pte_t __iomem *) dev_priv->gtt.gsm + first_entry;
const int max_entries = gtt_total_entries(dev_priv->gtt) - first_entry;
int i;
if (WARN(num_entries > max_entries,
"First entry = %d; Num entries = %d (max=%d)\n",
first_entry, num_entries, max_entries))
num_entries = max_entries;
scratch_pte = vm->pte_encode(px_dma(vm->scratch_page),
I915_CACHE_LLC, use_scratch, 0);
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
for (i = 0; i < num_entries; i++)
iowrite32(scratch_pte, &gtt_base[i]);
readl(gtt_base);
}
static void i915_ggtt_insert_entries(struct i915_address_space *vm,
struct sg_table *pages,
uint64_t start,
enum i915_cache_level cache_level, u32 unused)
{
unsigned int flags = (cache_level == I915_CACHE_NONE) ?
AGP_USER_MEMORY : AGP_USER_CACHED_MEMORY;
intel_gtt_insert_sg_entries(pages, start >> PAGE_SHIFT, flags);
}
static void i915_ggtt_clear_range(struct i915_address_space *vm,
uint64_t start,
uint64_t length,
drm/i915: Disable GGTT PTEs on GEN6+ suspend Once the machine gets to a certain point in the suspend process, we expect the GPU to be idle. If it is not, we might corrupt memory. Empirically (with an early version of this patch) we have seen this is not the case. We cannot currently explain why the latent GPU writes occur. In the technical sense, this patch is a workaround in that we have an issue we can't explain, and the patch indirectly solves the issue. However, it's really better than a workaround because we understand why it works, and it really should be a safe thing to do in all cases. The noticeable effect other than the debug messages would be an increase in the suspend time. I have not measure how expensive it actually is. I think it would be good to spend further time to root cause why we're seeing these latent writes, but it shouldn't preclude preventing the fallout. NOTE: It should be safe (and makes some sense IMO) to also keep the VALID bit unset on resume when we clear_range(). I've opted not to do this as properly clearing those bits at some later point would be extra work. v2: Fix bugzilla link Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321 Tested-by: Takashi Iwai <tiwai@suse.de> Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Tested-By: Todd Previte <tprevite@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-10-16 16:21:30 +00:00
bool unused)
{
unsigned first_entry = start >> PAGE_SHIFT;
unsigned num_entries = length >> PAGE_SHIFT;
intel_gtt_clear_range(first_entry, num_entries);
}
static int ggtt_bind_vma(struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 flags)
{
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
struct drm_device *dev = vma->vm->dev;
struct drm_i915_private *dev_priv = dev->dev_private;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
struct drm_i915_gem_object *obj = vma->obj;
struct sg_table *pages = obj->pages;
u32 pte_flags = 0;
int ret;
ret = i915_get_ggtt_vma_pages(vma);
if (ret)
return ret;
pages = vma->ggtt_view.pages;
/* Currently applicable only to VLV */
if (obj->gt_ro)
pte_flags |= PTE_READ_ONLY;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
if (!dev_priv->mm.aliasing_ppgtt || flags & GLOBAL_BIND) {
vma->vm->insert_entries(vma->vm, pages,
vma->node.start,
cache_level, pte_flags);
/* Note the inconsistency here is due to absence of the
* aliasing ppgtt on gen4 and earlier. Though we always
* request PIN_USER for execbuffer (translated to LOCAL_BIND),
* without the appgtt, we cannot honour that request and so
* must substitute it with a global binding. Since we do this
* behind the upper layers back, we need to explicitly set
* the bound flag ourselves.
*/
vma->bound |= GLOBAL_BIND;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
}
if (dev_priv->mm.aliasing_ppgtt && flags & LOCAL_BIND) {
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
struct i915_hw_ppgtt *appgtt = dev_priv->mm.aliasing_ppgtt;
appgtt->base.insert_entries(&appgtt->base, pages,
vma->node.start,
cache_level, pte_flags);
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
}
return 0;
}
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
static void ggtt_unbind_vma(struct i915_vma *vma)
{
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
struct drm_device *dev = vma->vm->dev;
struct drm_i915_private *dev_priv = dev->dev_private;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
struct drm_i915_gem_object *obj = vma->obj;
const uint64_t size = min_t(uint64_t,
obj->base.size,
vma->node.size);
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
if (vma->bound & GLOBAL_BIND) {
vma->vm->clear_range(vma->vm,
vma->node.start,
size,
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
true);
}
if (dev_priv->mm.aliasing_ppgtt && vma->bound & LOCAL_BIND) {
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
struct i915_hw_ppgtt *appgtt = dev_priv->mm.aliasing_ppgtt;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
appgtt->base.clear_range(&appgtt->base,
vma->node.start,
size,
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
true);
}
}
void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj)
{
struct drm_device *dev = obj->base.dev;
struct drm_i915_private *dev_priv = dev->dev_private;
bool interruptible;
interruptible = do_idling(dev_priv);
dma_unmap_sg(&dev->pdev->dev, obj->pages->sgl, obj->pages->nents,
PCI_DMA_BIDIRECTIONAL);
undo_idling(dev_priv, interruptible);
}
static void i915_gtt_color_adjust(struct drm_mm_node *node,
unsigned long color,
u64 *start,
u64 *end)
{
if (node->color != color)
*start += 4096;
if (!list_empty(&node->node_list)) {
node = list_entry(node->node_list.next,
struct drm_mm_node,
node_list);
if (node->allocated && node->color != color)
*end -= 4096;
}
}
static int i915_gem_setup_global_gtt(struct drm_device *dev,
u64 start,
u64 mappable_end,
u64 end)
{
/* Let GEM Manage all of the aperture.
*
* However, leave one page at the end still bound to the scratch page.
* There are a number of places where the hardware apparently prefetches
* past the end of the object, and we've seen multiple hangs with the
* GPU head pointer stuck in a batchbuffer bound at the last page of the
* aperture. One page should be enough to keep any prefetching inside
* of the aperture.
*/
struct drm_i915_private *dev_priv = dev->dev_private;
struct i915_address_space *ggtt_vm = &dev_priv->gtt.base;
struct drm_mm_node *entry;
struct drm_i915_gem_object *obj;
unsigned long hole_start, hole_end;
int ret;
BUG_ON(mappable_end > end);
ggtt_vm->start = start;
/* Subtract the guard page before address space initialization to
* shrink the range used by drm_mm */
ggtt_vm->total = end - start - PAGE_SIZE;
i915_address_space_init(ggtt_vm, dev_priv);
ggtt_vm->total += PAGE_SIZE;
if (intel_vgpu_active(dev)) {
ret = intel_vgt_balloon(dev);
if (ret)
return ret;
}
if (!HAS_LLC(dev))
ggtt_vm->mm.color_adjust = i915_gtt_color_adjust;
/* Mark any preallocated objects as occupied */
list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
struct i915_vma *vma = i915_gem_obj_to_vma(obj, ggtt_vm);
DRM_DEBUG_KMS("reserving preallocated space: %llx + %zx\n",
i915_gem_obj_ggtt_offset(obj), obj->base.size);
WARN_ON(i915_gem_obj_ggtt_bound(obj));
ret = drm_mm_reserve_node(&ggtt_vm->mm, &vma->node);
if (ret) {
DRM_DEBUG_KMS("Reservation failed: %i\n", ret);
return ret;
}
vma->bound |= GLOBAL_BIND;
list_add_tail(&vma->mm_list, &ggtt_vm->inactive_list);
}
/* Clear any non-preallocated blocks */
drm_mm_for_each_hole(entry, &ggtt_vm->mm, hole_start, hole_end) {
DRM_DEBUG_KMS("clearing unused GTT space: [%lx, %lx]\n",
hole_start, hole_end);
ggtt_vm->clear_range(ggtt_vm, hole_start,
hole_end - hole_start, true);
}
/* And finally clear the reserved guard page */
ggtt_vm->clear_range(ggtt_vm, end - PAGE_SIZE, PAGE_SIZE, true);
if (USES_PPGTT(dev) && !USES_FULL_PPGTT(dev)) {
struct i915_hw_ppgtt *ppgtt;
ppgtt = kzalloc(sizeof(*ppgtt), GFP_KERNEL);
if (!ppgtt)
return -ENOMEM;
ret = __hw_ppgtt_init(dev, ppgtt);
if (ret) {
ppgtt->base.cleanup(&ppgtt->base);
kfree(ppgtt);
return ret;
}
if (ppgtt->base.allocate_va_range)
ret = ppgtt->base.allocate_va_range(&ppgtt->base, 0,
ppgtt->base.total);
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
if (ret) {
ppgtt->base.cleanup(&ppgtt->base);
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
kfree(ppgtt);
return ret;
drm/i915: Finish gen6/7 dynamic page table allocation This patch continues on the idea from "Track GEN6 page table usage". From here on, in the steady state, PDEs are all pointing to the scratch page table (as recommended in the spec). When an object is allocated in the VA range, the code will determine if we need to allocate a page for the page table. Similarly when the object is destroyed, we will remove, and free the page table pointing the PDE back to the scratch page. Following patches will work to unify the code a bit as we bring in GEN8 support. GEN6 and GEN8 are different enough that I had a hard time to get to this point with as much common code as I do. The aliasing PPGTT must pre-allocate all of the page tables. There are a few reasons for this. Two trivial ones: aliasing ppgtt goes through the ggtt paths, so it's hard to maintain, we currently do not restore the default context (assuming the previous force reload is indeed necessary). Most importantly though, the only way (it seems from empirical evidence) to invalidate the CS TLBs on non-render ring is to either use ring sync (which requires actually stopping the rings in order to synchronize when the sync completes vs. where you are in execution), or to reload DCLV. Since without full PPGTT we do not ever reload the DCLV register, there is no good way to achieve this. The simplest solution is just to not support dynamic page table creation/destruction in the aliasing PPGTT. We could always reload DCLV, but this seems like quite a bit of excess overhead only to save at most 2MB-4k of memory for the aliasing PPGTT page tables. v2: Make the page table bitmap declared inside the function (Chris) Simplify the way scratching address space works. Move the alloc/teardown tracepoints up a level in the call stack so that both all implementations get the trace. v3: Updated trace event to spit out a name v4: Aliasing ppgtt is now initialized differently (in setup global gtt) v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check for trace, as it is no longer possible after the PPGTT cleanup patch series of a couple of months ago (Daniel). v6: Implement changes from code review (Daniel): - allocate/teardown_va_range calls added. - Add a scratch page allocation helper (only need the address). - Move trace events to a new patch. - Use updated mark_tlbs_dirty. - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init. v7: teardown_va_range removed (Daniel). In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt. v8: Rebase after s/page_tables/page_table/. v9: Remove unnecessary scratch flag in page_table struct, future patches can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes redundant. Initialize scratch_pt and pt. (Mika) v10: Clean up aliasing ppgtt init error path and prevent leaking the ppgtt obj when init fails. (Mika) Updated commit author. (Daniel) Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+) Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-24 15:46:22 +00:00
}
ppgtt->base.clear_range(&ppgtt->base,
ppgtt->base.start,
ppgtt->base.total,
true);
dev_priv->mm.aliasing_ppgtt = ppgtt;
}
return 0;
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
}
void i915_gem_init_global_gtt(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
u64 gtt_size, mappable_size;
gtt_size = dev_priv->gtt.base.total;
mappable_size = dev_priv->gtt.mappable_end;
i915_gem_setup_global_gtt(dev, 0, mappable_size, gtt_size);
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
}
void i915_global_gtt_cleanup(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct i915_address_space *vm = &dev_priv->gtt.base;
if (dev_priv->mm.aliasing_ppgtt) {
struct i915_hw_ppgtt *ppgtt = dev_priv->mm.aliasing_ppgtt;
ppgtt->base.cleanup(&ppgtt->base);
}
if (drm_mm_initialized(&vm->mm)) {
if (intel_vgpu_active(dev))
intel_vgt_deballoon();
drm_mm_takedown(&vm->mm);
list_del(&vm->global_link);
}
vm->cleanup(vm);
}
static unsigned int gen6_get_total_gtt_size(u16 snb_gmch_ctl)
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
{
snb_gmch_ctl >>= SNB_GMCH_GGMS_SHIFT;
snb_gmch_ctl &= SNB_GMCH_GGMS_MASK;
return snb_gmch_ctl << 20;
}
static unsigned int gen8_get_total_gtt_size(u16 bdw_gmch_ctl)
{
bdw_gmch_ctl >>= BDW_GMCH_GGMS_SHIFT;
bdw_gmch_ctl &= BDW_GMCH_GGMS_MASK;
if (bdw_gmch_ctl)
bdw_gmch_ctl = 1 << bdw_gmch_ctl;
#ifdef CONFIG_X86_32
/* Limit 32b platforms to a 2GB GGTT: 4 << 20 / pte size * PAGE_SIZE */
if (bdw_gmch_ctl > 4)
bdw_gmch_ctl = 4;
#endif
return bdw_gmch_ctl << 20;
}
static unsigned int chv_get_total_gtt_size(u16 gmch_ctrl)
{
gmch_ctrl >>= SNB_GMCH_GGMS_SHIFT;
gmch_ctrl &= SNB_GMCH_GGMS_MASK;
if (gmch_ctrl)
return 1 << (20 + gmch_ctrl);
return 0;
}
static size_t gen6_get_stolen_size(u16 snb_gmch_ctl)
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
{
snb_gmch_ctl >>= SNB_GMCH_GMS_SHIFT;
snb_gmch_ctl &= SNB_GMCH_GMS_MASK;
return snb_gmch_ctl << 25; /* 32 MB units */
}
static size_t gen8_get_stolen_size(u16 bdw_gmch_ctl)
{
bdw_gmch_ctl >>= BDW_GMCH_GMS_SHIFT;
bdw_gmch_ctl &= BDW_GMCH_GMS_MASK;
return bdw_gmch_ctl << 25; /* 32 MB units */
}
static size_t chv_get_stolen_size(u16 gmch_ctrl)
{
gmch_ctrl >>= SNB_GMCH_GMS_SHIFT;
gmch_ctrl &= SNB_GMCH_GMS_MASK;
/*
* 0x0 to 0x10: 32MB increments starting at 0MB
* 0x11 to 0x16: 4MB increments starting at 8MB
* 0x17 to 0x1d: 4MB increments start at 36MB
*/
if (gmch_ctrl < 0x11)
return gmch_ctrl << 25;
else if (gmch_ctrl < 0x17)
return (gmch_ctrl - 0x11 + 2) << 22;
else
return (gmch_ctrl - 0x17 + 9) << 22;
}
static size_t gen9_get_stolen_size(u16 gen9_gmch_ctl)
{
gen9_gmch_ctl >>= BDW_GMCH_GMS_SHIFT;
gen9_gmch_ctl &= BDW_GMCH_GMS_MASK;
if (gen9_gmch_ctl < 0xf0)
return gen9_gmch_ctl << 25; /* 32 MB units */
else
/* 4MB increments starting at 0xf0 for 4MB */
return (gen9_gmch_ctl - 0xf0 + 1) << 22;
}
static int ggtt_probe_common(struct drm_device *dev,
size_t gtt_size)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct i915_page_scratch *scratch_page;
phys_addr_t gtt_phys_addr;
/* For Modern GENs the PTEs and register space are split in the BAR */
gtt_phys_addr = pci_resource_start(dev->pdev, 0) +
(pci_resource_len(dev->pdev, 0) / 2);
/*
* On BXT writes larger than 64 bit to the GTT pagetable range will be
* dropped. For WC mappings in general we have 64 byte burst writes
* when the WC buffer is flushed, so we can't use it, but have to
* resort to an uncached mapping. The WC issue is easily caught by the
* readback check when writing GTT PTE entries.
*/
if (IS_BROXTON(dev))
dev_priv->gtt.gsm = ioremap_nocache(gtt_phys_addr, gtt_size);
else
dev_priv->gtt.gsm = ioremap_wc(gtt_phys_addr, gtt_size);
if (!dev_priv->gtt.gsm) {
DRM_ERROR("Failed to map the gtt page table\n");
return -ENOMEM;
}
scratch_page = alloc_scratch_page(dev);
if (IS_ERR(scratch_page)) {
DRM_ERROR("Scratch setup failed\n");
/* iounmap will also get called at remove, but meh */
iounmap(dev_priv->gtt.gsm);
return PTR_ERR(scratch_page);
}
dev_priv->gtt.base.scratch_page = scratch_page;
return 0;
}
/* The GGTT and PPGTT need a private PPAT setup in order to handle cacheability
* bits. When using advanced contexts each context stores its own PAT, but
* writing this data shouldn't be harmful even in those cases. */
static void bdw_setup_private_ppat(struct drm_i915_private *dev_priv)
{
uint64_t pat;
pat = GEN8_PPAT(0, GEN8_PPAT_WB | GEN8_PPAT_LLC) | /* for normal objects, no eLLC */
GEN8_PPAT(1, GEN8_PPAT_WC | GEN8_PPAT_LLCELLC) | /* for something pointing to ptes? */
GEN8_PPAT(2, GEN8_PPAT_WT | GEN8_PPAT_LLCELLC) | /* for scanout with eLLC */
GEN8_PPAT(3, GEN8_PPAT_UC) | /* Uncached objects, mostly for scanout */
GEN8_PPAT(4, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(0)) |
GEN8_PPAT(5, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(1)) |
GEN8_PPAT(6, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(2)) |
GEN8_PPAT(7, GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(3));
if (!USES_PPGTT(dev_priv->dev))
/* Spec: "For GGTT, there is NO pat_sel[2:0] from the entry,
* so RTL will always use the value corresponding to
* pat_sel = 000".
* So let's disable cache for GGTT to avoid screen corruptions.
* MOCS still can be used though.
* - System agent ggtt writes (i.e. cpu gtt mmaps) already work
* before this patch, i.e. the same uncached + snooping access
* like on gen6/7 seems to be in effect.
* - So this just fixes blitter/render access. Again it looks
* like it's not just uncached access, but uncached + snooping.
* So we can still hold onto all our assumptions wrt cpu
* clflushing on LLC machines.
*/
pat = GEN8_PPAT(0, GEN8_PPAT_UC);
/* XXX: spec defines this as 2 distinct registers. It's unclear if a 64b
* write would work. */
I915_WRITE(GEN8_PRIVATE_PAT_LO, pat);
I915_WRITE(GEN8_PRIVATE_PAT_HI, pat >> 32);
}
static void chv_setup_private_ppat(struct drm_i915_private *dev_priv)
{
uint64_t pat;
/*
* Map WB on BDW to snooped on CHV.
*
* Only the snoop bit has meaning for CHV, the rest is
* ignored.
*
* The hardware will never snoop for certain types of accesses:
* - CPU GTT (GMADR->GGTT->no snoop->memory)
* - PPGTT page tables
* - some other special cycles
*
* As with BDW, we also need to consider the following for GT accesses:
* "For GGTT, there is NO pat_sel[2:0] from the entry,
* so RTL will always use the value corresponding to
* pat_sel = 000".
* Which means we must set the snoop bit in PAT entry 0
* in order to keep the global status page working.
*/
pat = GEN8_PPAT(0, CHV_PPAT_SNOOP) |
GEN8_PPAT(1, 0) |
GEN8_PPAT(2, 0) |
GEN8_PPAT(3, 0) |
GEN8_PPAT(4, CHV_PPAT_SNOOP) |
GEN8_PPAT(5, CHV_PPAT_SNOOP) |
GEN8_PPAT(6, CHV_PPAT_SNOOP) |
GEN8_PPAT(7, CHV_PPAT_SNOOP);
I915_WRITE(GEN8_PRIVATE_PAT_LO, pat);
I915_WRITE(GEN8_PRIVATE_PAT_HI, pat >> 32);
}
static int gen8_gmch_probe(struct drm_device *dev,
u64 *gtt_total,
size_t *stolen,
phys_addr_t *mappable_base,
u64 *mappable_end)
{
struct drm_i915_private *dev_priv = dev->dev_private;
u64 gtt_size;
u16 snb_gmch_ctl;
int ret;
/* TODO: We're not aware of mappable constraints on gen8 yet */
*mappable_base = pci_resource_start(dev->pdev, 2);
*mappable_end = pci_resource_len(dev->pdev, 2);
if (!pci_set_dma_mask(dev->pdev, DMA_BIT_MASK(39)))
pci_set_consistent_dma_mask(dev->pdev, DMA_BIT_MASK(39));
pci_read_config_word(dev->pdev, SNB_GMCH_CTRL, &snb_gmch_ctl);
if (INTEL_INFO(dev)->gen >= 9) {
*stolen = gen9_get_stolen_size(snb_gmch_ctl);
gtt_size = gen8_get_total_gtt_size(snb_gmch_ctl);
} else if (IS_CHERRYVIEW(dev)) {
*stolen = chv_get_stolen_size(snb_gmch_ctl);
gtt_size = chv_get_total_gtt_size(snb_gmch_ctl);
} else {
*stolen = gen8_get_stolen_size(snb_gmch_ctl);
gtt_size = gen8_get_total_gtt_size(snb_gmch_ctl);
}
*gtt_total = (gtt_size / sizeof(gen8_pte_t)) << PAGE_SHIFT;
if (IS_CHERRYVIEW(dev) || IS_BROXTON(dev))
chv_setup_private_ppat(dev_priv);
else
bdw_setup_private_ppat(dev_priv);
ret = ggtt_probe_common(dev, gtt_size);
dev_priv->gtt.base.clear_range = gen8_ggtt_clear_range;
dev_priv->gtt.base.insert_entries = gen8_ggtt_insert_entries;
dev_priv->gtt.base.bind_vma = ggtt_bind_vma;
dev_priv->gtt.base.unbind_vma = ggtt_unbind_vma;
return ret;
}
static int gen6_gmch_probe(struct drm_device *dev,
u64 *gtt_total,
size_t *stolen,
phys_addr_t *mappable_base,
u64 *mappable_end)
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
{
struct drm_i915_private *dev_priv = dev->dev_private;
unsigned int gtt_size;
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
u16 snb_gmch_ctl;
int ret;
*mappable_base = pci_resource_start(dev->pdev, 2);
*mappable_end = pci_resource_len(dev->pdev, 2);
/* 64/512MB is the current min/max we actually know of, but this is just
* a coarse sanity check.
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
*/
if ((*mappable_end < (64<<20) || (*mappable_end > (512<<20)))) {
DRM_ERROR("Unknown GMADR size (%llx)\n",
dev_priv->gtt.mappable_end);
return -ENXIO;
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
}
if (!pci_set_dma_mask(dev->pdev, DMA_BIT_MASK(40)))
pci_set_consistent_dma_mask(dev->pdev, DMA_BIT_MASK(40));
pci_read_config_word(dev->pdev, SNB_GMCH_CTRL, &snb_gmch_ctl);
Revert "drm/i915: Calculate correct stolen size for GEN7+" This reverts commit 03752f5b7b77b95d83479885040950fba1250850. This revert requires a bit of explanation on how I understand things work. Internally the architects/designers decide how the stolen encoding works. We put it in a doc. BIOS writers take these docs and implement it. Driver writers read the doc too, and read the value left by the BIOS writers, and then we make magic. The failing here is that in the docs we had[1] contained two different definitions for this register for Gen7. (We have both a PCI register, and an MMIO, and each of these were different). At the time [2] of 03752f5, we asked the architects what the correct value should be; but that doesn't match the reality (BIOS) unfortunately. So on all machines I can get my hands on, this revert is the right thing to do. I've also worked with the product group to confirm that they agree this revert is what we should do. People using HW made my "people" who both write their own BIOS, and have access to our docs (Apple?). Investigations are still ongoing about whether we need to add a list of machines needing special handling, but this patch should be the right thing for pretty much everyone. [1] The docs are still wrong on this one. Now instead of two registers with two definitions, we have one register with BOTH definitions, progress? [2] The open source PRMs have the "wrong" definitions in chapter Volume 1 part6, section 1.1.12. This digging was inspired by Paulo. Cc: Paulo Zanoni <przanoni@gmail.com> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org> [danvet: Augment the patch saying that it's still a bit unclear whether there are any machines out there with "wrong" firmware and whether we need to add a list to handle them specially.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-05-01 18:00:34 +00:00
*stolen = gen6_get_stolen_size(snb_gmch_ctl);
gtt_size = gen6_get_total_gtt_size(snb_gmch_ctl);
*gtt_total = (gtt_size / sizeof(gen6_pte_t)) << PAGE_SHIFT;
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
ret = ggtt_probe_common(dev, gtt_size);
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
dev_priv->gtt.base.clear_range = gen6_ggtt_clear_range;
dev_priv->gtt.base.insert_entries = gen6_ggtt_insert_entries;
dev_priv->gtt.base.bind_vma = ggtt_bind_vma;
dev_priv->gtt.base.unbind_vma = ggtt_unbind_vma;
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
return ret;
}
static void gen6_gmch_remove(struct i915_address_space *vm)
drm/i915: Stop using AGP layer for GEN6+ As a quick hack we make the old intel_gtt structure mutable so we can fool a bunch of the existing code which depends on elements in that data structure. We can/should try to remove this in a subsequent patch. This should preserve the old gtt init behavior which upon writing these patches seems incorrect. The next patch will fix these things. The one exception is VLV which doesn't have the preserved flush control write behavior. Since we want to do that for all GEN6+ stuff, we'll handle that in a later patch. Mainstream VLV support doesn't actually exist yet anyway. v2: Update the comment to remove the "voodoo" Check that the last pte written matches what we readback v3: actually kill cache_level_to_agp_type since most of the flags will disappear in an upcoming patch v4: v3 was actually not what we wanted (Daniel) Make the ggtt bind assertions better and stricter (Chris) Fix some uncaught errors at gtt init (Chris) Some other random stuff that Chris wanted v5: check for i==0 in gen6_ggtt_bind_object to shut up gcc (Ben) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Reviewed-by [v4]: Chris Wilson <chris@chris-wilson.co.uk> [danvet: Make the cache_level -> agp_flags conversion for pre-gen6 a tad more robust by mapping everything != CACHE_NONE to the cached agp flag - we have a 1:1 uncached mapping, but different modes of cacheable (at least on later generations). Suggested by Chris Wilson.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2012-11-04 17:21:27 +00:00
{
struct i915_gtt *gtt = container_of(vm, struct i915_gtt, base);
iounmap(gtt->gsm);
free_scratch_page(vm->dev, vm->scratch_page);
}
static int i915_gmch_probe(struct drm_device *dev,
u64 *gtt_total,
size_t *stolen,
phys_addr_t *mappable_base,
u64 *mappable_end)
{
struct drm_i915_private *dev_priv = dev->dev_private;
int ret;
ret = intel_gmch_probe(dev_priv->bridge_dev, dev_priv->dev->pdev, NULL);
if (!ret) {
DRM_ERROR("failed to set up gmch\n");
return -EIO;
}
intel_gtt_get(gtt_total, stolen, mappable_base, mappable_end);
dev_priv->gtt.do_idle_maps = needs_idle_maps(dev_priv->dev);
dev_priv->gtt.base.insert_entries = i915_ggtt_insert_entries;
dev_priv->gtt.base.clear_range = i915_ggtt_clear_range;
dev_priv->gtt.base.bind_vma = ggtt_bind_vma;
dev_priv->gtt.base.unbind_vma = ggtt_unbind_vma;
if (unlikely(dev_priv->gtt.do_idle_maps))
DRM_INFO("applying Ironlake quirks for intel_iommu\n");
return 0;
}
static void i915_gmch_remove(struct i915_address_space *vm)
{
intel_gmch_remove();
}
int i915_gem_gtt_init(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct i915_gtt *gtt = &dev_priv->gtt;
int ret;
if (INTEL_INFO(dev)->gen <= 5) {
gtt->gtt_probe = i915_gmch_probe;
gtt->base.cleanup = i915_gmch_remove;
} else if (INTEL_INFO(dev)->gen < 8) {
gtt->gtt_probe = gen6_gmch_probe;
gtt->base.cleanup = gen6_gmch_remove;
if (IS_HASWELL(dev) && dev_priv->ellc_size)
gtt->base.pte_encode = iris_pte_encode;
else if (IS_HASWELL(dev))
gtt->base.pte_encode = hsw_pte_encode;
else if (IS_VALLEYVIEW(dev))
gtt->base.pte_encode = byt_pte_encode;
else if (INTEL_INFO(dev)->gen >= 7)
gtt->base.pte_encode = ivb_pte_encode;
else
gtt->base.pte_encode = snb_pte_encode;
} else {
dev_priv->gtt.gtt_probe = gen8_gmch_probe;
dev_priv->gtt.base.cleanup = gen6_gmch_remove;
}
gtt->base.dev = dev;
ret = gtt->gtt_probe(dev, &gtt->base.total, &gtt->stolen_size,
&gtt->mappable_base, &gtt->mappable_end);
if (ret)
return ret;
/* GMADR is the PCI mmio aperture into the global GTT. */
DRM_INFO("Memory usable by graphics device = %lluM\n",
gtt->base.total >> 20);
DRM_DEBUG_DRIVER("GMADR size = %lldM\n", gtt->mappable_end >> 20);
DRM_DEBUG_DRIVER("GTT stolen size = %zdM\n", gtt->stolen_size >> 20);
#ifdef CONFIG_INTEL_IOMMU
if (intel_iommu_gfx_mapped)
DRM_INFO("VT-d active for gfx access\n");
#endif
/*
* i915.enable_ppgtt is read-only, so do an early pass to validate the
* user's requested state against the hardware/driver capabilities. We
* do this now so that we can print out any log messages once rather
* than every time we check intel_enable_ppgtt().
*/
i915.enable_ppgtt = sanitize_enable_ppgtt(dev, i915.enable_ppgtt);
DRM_DEBUG_DRIVER("ppgtt mode: %i\n", i915.enable_ppgtt);
return 0;
}
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
void i915_gem_restore_gtt_mappings(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct drm_i915_gem_object *obj;
struct i915_address_space *vm;
struct i915_vma *vma;
bool flush;
i915_check_and_clear_faults(dev);
/* First fill our portion of the GTT with scratch pages */
dev_priv->gtt.base.clear_range(&dev_priv->gtt.base,
dev_priv->gtt.base.start,
dev_priv->gtt.base.total,
true);
/* Cache flush objects bound into GGTT and rebind them. */
vm = &dev_priv->gtt.base;
list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) {
flush = false;
list_for_each_entry(vma, &obj->vma_list, vma_link) {
if (vma->vm != vm)
continue;
WARN_ON(i915_vma_bind(vma, obj->cache_level,
PIN_UPDATE));
flush = true;
}
if (flush)
i915_gem_clflush_object(obj, obj->pin_display);
}
if (INTEL_INFO(dev)->gen >= 8) {
if (IS_CHERRYVIEW(dev) || IS_BROXTON(dev))
chv_setup_private_ppat(dev_priv);
else
bdw_setup_private_ppat(dev_priv);
return;
}
if (USES_PPGTT(dev)) {
list_for_each_entry(vm, &dev_priv->vm_list, global_link) {
/* TODO: Perhaps it shouldn't be gen6 specific */
struct i915_hw_ppgtt *ppgtt =
container_of(vm, struct i915_hw_ppgtt,
base);
if (i915_is_ggtt(vm))
ppgtt = dev_priv->mm.aliasing_ppgtt;
gen6_write_page_range(dev_priv, &ppgtt->pd,
0, ppgtt->base.total);
}
}
i915_ggtt_flush(dev_priv);
}
static struct i915_vma *
__i915_gem_vma_create(struct drm_i915_gem_object *obj,
struct i915_address_space *vm,
const struct i915_ggtt_view *ggtt_view)
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
{
struct i915_vma *vma;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
if (WARN_ON(i915_is_ggtt(vm) != !!ggtt_view))
return ERR_PTR(-EINVAL);
vma = kmem_cache_zalloc(to_i915(obj->base.dev)->vmas, GFP_KERNEL);
if (vma == NULL)
return ERR_PTR(-ENOMEM);
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
INIT_LIST_HEAD(&vma->vma_link);
INIT_LIST_HEAD(&vma->mm_list);
INIT_LIST_HEAD(&vma->exec_list);
vma->vm = vm;
vma->obj = obj;
if (i915_is_ggtt(vm))
vma->ggtt_view = *ggtt_view;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
list_add_tail(&vma->vma_link, &obj->vma_list);
if (!i915_is_ggtt(vm))
i915_ppgtt_get(i915_vm_to_ppgtt(vm));
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
return vma;
}
struct i915_vma *
i915_gem_obj_lookup_or_create_vma(struct drm_i915_gem_object *obj,
struct i915_address_space *vm)
{
struct i915_vma *vma;
vma = i915_gem_obj_to_vma(obj, vm);
if (!vma)
vma = __i915_gem_vma_create(obj, vm,
i915_is_ggtt(vm) ? &i915_ggtt_view_normal : NULL);
return vma;
}
struct i915_vma *
i915_gem_obj_lookup_or_create_ggtt_vma(struct drm_i915_gem_object *obj,
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
const struct i915_ggtt_view *view)
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
{
struct i915_address_space *ggtt = i915_obj_to_ggtt(obj);
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
struct i915_vma *vma;
if (WARN_ON(!view))
return ERR_PTR(-EINVAL);
vma = i915_gem_obj_to_ggtt_view(obj, view);
if (IS_ERR(vma))
return vma;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
if (!vma)
vma = __i915_gem_vma_create(obj, ggtt, view);
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
return vma;
drm/i915: Create bind/unbind abstraction for VMAs To sum up what goes on here, we abstract the vma binding, similarly to the previous object binding. This helps for distinguishing legacy binding, versus modern binding. To keep the code churn as minimal as possible, I am leaving in insert_entries(). It serves as the per platform pte writing basically. bind_vma and insert_entries do share a lot of similarities, and I did have designs to combine the two, but as mentioned already... too much churn in an already massive patchset. What follows are the 3 commits which existed discretely in the original submissions. Upon rebasing on Broadwell support, it became clear that separation was not good, and only made for more error prone code. Below are the 3 commit messages with all their history. drm/i915: Add bind/unbind object functions to VMA drm/i915: Use the new vm [un]bind functions drm/i915: reduce vm->insert_entries() usage drm/i915: Add bind/unbind object functions to VMA As we plumb the code with more VM information, it has become more obvious that the easiest way to deal with bind and unbind is to simply put the function pointers in the vm, and let those choose the correct way to handle the page table updates. This change allows many places in the code to simply be vm->bind, and not have to worry about distinguishing PPGTT vs GGTT. Notice that this patch has no impact on functionality. I've decided to save the actual change until the next patch because I think it's easier to review that way. I'm happy to squash the two, or let Daniel do it on merge. v2: Make ggtt handle the quirky aliasing ppgtt Add flags to bind object to support above Don't ever call bind/unbind directly for PPGTT until we have real, full PPGTT (use NULLs to assert this) Make sure we rebind the ggtt if there already is a ggtt binding. This happens on set cache levels. Use VMA for bind/unbind (Daniel, Ben) v3: Reorganize ggtt_vma_bind to be more concise and easier to read (Ville). Change logic in unbind to only unbind ggtt when there is a global mapping, and to remove a redundant check if the aliasing ppgtt exists. v4: Make the bind function a bit smarter about the cache levels to avoid unnecessary multiple remaps. "I accept it is a wart, I think unifying the pin_vma / bind_vma could be unified later" (Chris) Removed the git notes, and put version info here. (Daniel) v5: Update the comment to not suck (Chris) v6: Move bind/unbind to the VMA. It makes more sense in the VMA structure (always has, but I was previously lazy). With this change, it will allow us to keep a distinct insert_entries. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: Use the new vm [un]bind functions Building on the last patch which created the new function pointers in the VM for bind/unbind, here we actually put those new function pointers to use. Split out as a separate patch to aid in review. I'm fine with squashing into the previous patch if people request it. v2: Updated to address the smart ggtt which can do aliasing as needed Make sure we bind to global gtt when mappable and fenceable. I thought we could get away without this initialy, but we cannot. v3: Make the global GTT binding explicitly use the ggtt VM for bind_vma(). While at it, use the new ggtt_vma helper (Chris) At this point the original mailing list thread diverges. ie. v4^: use target_obj instead of obj for gen6 relocate_entry vma->bind_vma() can be called safely during pin. So simply do that instead of the complicated conditionals. Don't restore PPGTT bound objects on resume path Bug fix in resume path for globally bound Bos Properly handle secure dispatch Rebased on vma bind/unbind conversion Signed-off-by: Ben Widawsky <ben@bwidawsk.net> drm/i915: reduce vm->insert_entries() usage FKA: drm/i915: eliminate vm->insert_entries() With bind/unbind function pointers in place, we no longer need insert_entries. We could, and want, to remove clear_range, however it's not totally easy at this point. Since it's used in a couple of place still that don't only deal in objects: setup, ppgtt init, and restore gtt mappings. v2: Don't actually remove insert_entries, just limit its usage. It will be useful when we introduce gen8. It will always be called from the vma bind/unbind. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1) Signed-off-by: Ben Widawsky <ben@bwidawsk.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-06 22:10:56 +00:00
}
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
static struct scatterlist *
rotate_pages(dma_addr_t *in, unsigned int offset,
unsigned int width, unsigned int height,
struct sg_table *st, struct scatterlist *sg)
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
{
unsigned int column, row;
unsigned int src_idx;
if (!sg) {
st->nents = 0;
sg = st->sgl;
}
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
for (column = 0; column < width; column++) {
src_idx = width * (height - 1) + column;
for (row = 0; row < height; row++) {
st->nents++;
/* We don't need the pages, but need to initialize
* the entries so the sg list can be happily traversed.
* The only thing we need are DMA addresses.
*/
sg_set_page(sg, NULL, PAGE_SIZE, 0);
sg_dma_address(sg) = in[offset + src_idx];
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
sg_dma_len(sg) = PAGE_SIZE;
sg = sg_next(sg);
src_idx -= width;
}
}
return sg;
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
}
static struct sg_table *
intel_rotate_fb_obj_pages(struct i915_ggtt_view *ggtt_view,
struct drm_i915_gem_object *obj)
{
struct intel_rotation_info *rot_info = &ggtt_view->rotation_info;
unsigned int size_pages = rot_info->size >> PAGE_SHIFT;
unsigned int size_pages_uv;
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
struct sg_page_iter sg_iter;
unsigned long i;
dma_addr_t *page_addr_list;
struct sg_table *st;
unsigned int uv_start_page;
struct scatterlist *sg;
int ret = -ENOMEM;
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
/* Allocate a temporary list of source pages for random access. */
page_addr_list = drm_malloc_ab(obj->base.size / PAGE_SIZE,
sizeof(dma_addr_t));
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
if (!page_addr_list)
return ERR_PTR(ret);
/* Account for UV plane with NV12. */
if (rot_info->pixel_format == DRM_FORMAT_NV12)
size_pages_uv = rot_info->size_uv >> PAGE_SHIFT;
else
size_pages_uv = 0;
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
/* Allocate target SG list. */
st = kmalloc(sizeof(*st), GFP_KERNEL);
if (!st)
goto err_st_alloc;
ret = sg_alloc_table(st, size_pages + size_pages_uv, GFP_KERNEL);
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
if (ret)
goto err_sg_alloc;
/* Populate source page list from the object. */
i = 0;
for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, 0) {
page_addr_list[i] = sg_page_iter_dma_address(&sg_iter);
i++;
}
/* Rotate the pages. */
sg = rotate_pages(page_addr_list, 0,
rot_info->width_pages, rot_info->height_pages,
st, NULL);
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
/* Append the UV plane if NV12. */
if (rot_info->pixel_format == DRM_FORMAT_NV12) {
uv_start_page = size_pages;
/* Check for tile-row un-alignment. */
if (offset_in_page(rot_info->uv_offset))
uv_start_page--;
rot_info->uv_start_page = uv_start_page;
rotate_pages(page_addr_list, uv_start_page,
rot_info->width_pages_uv,
rot_info->height_pages_uv,
st, sg);
}
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
DRM_DEBUG_KMS(
"Created rotated page mapping for object size %zu (pitch=%u, height=%u, pixel_format=0x%x, %ux%u tiles, %u pages (%u plane 0)).\n",
obj->base.size, rot_info->pitch, rot_info->height,
rot_info->pixel_format, rot_info->width_pages,
rot_info->height_pages, size_pages + size_pages_uv,
size_pages);
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
drm_free_large(page_addr_list);
return st;
err_sg_alloc:
kfree(st);
err_st_alloc:
drm_free_large(page_addr_list);
DRM_DEBUG_KMS(
"Failed to create rotated mapping for object size %zu! (%d) (pitch=%u, height=%u, pixel_format=0x%x, %ux%u tiles, %u pages (%u plane 0))\n",
obj->base.size, ret, rot_info->pitch, rot_info->height,
rot_info->pixel_format, rot_info->width_pages,
rot_info->height_pages, size_pages + size_pages_uv,
size_pages);
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
return ERR_PTR(ret);
}
static struct sg_table *
intel_partial_pages(const struct i915_ggtt_view *view,
struct drm_i915_gem_object *obj)
{
struct sg_table *st;
struct scatterlist *sg;
struct sg_page_iter obj_sg_iter;
int ret = -ENOMEM;
st = kmalloc(sizeof(*st), GFP_KERNEL);
if (!st)
goto err_st_alloc;
ret = sg_alloc_table(st, view->params.partial.size, GFP_KERNEL);
if (ret)
goto err_sg_alloc;
sg = st->sgl;
st->nents = 0;
for_each_sg_page(obj->pages->sgl, &obj_sg_iter, obj->pages->nents,
view->params.partial.offset)
{
if (st->nents >= view->params.partial.size)
break;
sg_set_page(sg, NULL, PAGE_SIZE, 0);
sg_dma_address(sg) = sg_page_iter_dma_address(&obj_sg_iter);
sg_dma_len(sg) = PAGE_SIZE;
sg = sg_next(sg);
st->nents++;
}
return st;
err_sg_alloc:
kfree(st);
err_st_alloc:
return ERR_PTR(ret);
}
static int
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
i915_get_ggtt_vma_pages(struct i915_vma *vma)
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
{
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
int ret = 0;
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
if (vma->ggtt_view.pages)
return 0;
if (vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL)
vma->ggtt_view.pages = vma->obj->pages;
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
else if (vma->ggtt_view.type == I915_GGTT_VIEW_ROTATED)
vma->ggtt_view.pages =
intel_rotate_fb_obj_pages(&vma->ggtt_view, vma->obj);
else if (vma->ggtt_view.type == I915_GGTT_VIEW_PARTIAL)
vma->ggtt_view.pages =
intel_partial_pages(&vma->ggtt_view, vma->obj);
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
else
WARN_ONCE(1, "GGTT view %u not implemented!\n",
vma->ggtt_view.type);
if (!vma->ggtt_view.pages) {
DRM_ERROR("Failed to get pages for GGTT view type %u!\n",
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
vma->ggtt_view.type);
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
ret = -EINVAL;
} else if (IS_ERR(vma->ggtt_view.pages)) {
ret = PTR_ERR(vma->ggtt_view.pages);
vma->ggtt_view.pages = NULL;
DRM_ERROR("Failed to get pages for VMA view type %u (%d)!\n",
vma->ggtt_view.type, ret);
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
}
drm/i915/skl: Support secondary (rotated) frame buffer mapping 90/270 rotated scanout needs a rotated GTT view of the framebuffer. This is put in a separate VMA with a dedicated ggtt view and wired such that it is created when a framebuffer is pinned to a 90/270 rotated plane. Rotation is only possible with Yb/Yf buffers and error is propagated to user space in case of a mismatch. Special rotated page view is constructed at the VMA creation time by borrowing the DMA addresses from obj->pages. v2: * Do not bother with pages for rotated sg list, just populate the DMA addresses. (Daniel Vetter) * Checkpatch cleanup. v3: * Rebased on top of new plane handling (create rotated mapping when setting the rotation property). * Unpin rotated VMA on unpinning from display plane. * Simplify rotation check using bitwise AND. (Chris Wilson) v4: * Fix unpinning of optional rotated mapping so it is really considered to be optional. v5: * Rebased for fb modifier changes. * Rebased for atomic commit. * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter) v6: * Rebased after preparatory work has been extracted out. (Daniel Vetter) v7: * Slightly simplified tiling geometry calculation. * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter) v8: * Do not use i915_gem_obj_size to get object size since that actually returns the size of an VMA which may not exist. * Rebased for ggtt view changes. v9: * Rebased after code review changes on the preceding patches. * Tidy function definitions. (Joonas Lahtinen) For: VIZ-4726 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4) Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-03-23 11:10:36 +00:00
return ret;
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
}
/**
* i915_vma_bind - Sets up PTEs for an VMA in it's corresponding address space.
* @vma: VMA to map
* @cache_level: mapping cache level
* @flags: flags like global or local mapping
*
* DMA addresses are taken from the scatter-gather table of this object (or of
* this VMA in case of non-default GGTT views) and PTE entries set up.
* Note that DMA addresses are also the only part of the SG table we care about.
*/
int i915_vma_bind(struct i915_vma *vma, enum i915_cache_level cache_level,
u32 flags)
{
int ret;
u32 bind_flags;
if (WARN_ON(flags == 0))
return -EINVAL;
bind_flags = 0;
if (flags & PIN_GLOBAL)
bind_flags |= GLOBAL_BIND;
if (flags & PIN_USER)
bind_flags |= LOCAL_BIND;
if (flags & PIN_UPDATE)
bind_flags |= vma->bound;
else
bind_flags &= ~vma->bound;
if (bind_flags == 0)
return 0;
if (vma->bound == 0 && vma->vm->allocate_va_range) {
trace_i915_va_alloc(vma->vm,
vma->node.start,
vma->node.size,
VM_TO_TRACE_NAME(vma->vm));
/* XXX: i915_vma_pin() will fix this +- hack */
vma->pin_count++;
ret = vma->vm->allocate_va_range(vma->vm,
vma->node.start,
vma->node.size);
vma->pin_count--;
if (ret)
return ret;
}
ret = vma->vm->bind_vma(vma, cache_level, bind_flags);
if (ret)
return ret;
vma->bound |= bind_flags;
drm/i915: Infrastructure for supporting different GGTT views per object Things like reliable GGTT mappings and mirrored 2d-on-3d display will need to map objects into the same address space multiple times. Added a GGTT view concept and linked it with the VMA to distinguish between multiple instances per address space. New objects and GEM functions which do not take this new view as a parameter assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the previous behaviour. This now means that objects can have multiple VMA entries so the code which assumed there will only be one also had to be modified. Alternative GGTT views are supposed to borrow DMA addresses from obj->pages which is DMA mapped on first VMA instantiation and unmapped on the last one going away. v2: * Removed per view special casing in i915_gem_ggtt_prepare / finish_object in favour of creating and destroying DMA mappings on first VMA instantiation and last VMA destruction. (Daniel Vetter) * Simplified i915_vma_unbind which does not need to count the GGTT views. (Daniel Vetter) * Also moved obj->map_and_fenceable reset under the same check. * Checkpatch cleanups. v3: * Only retire objects once the last VMA is unbound. v4: * Keep scatter-gather table for alternative views persistent for the lifetime of the VMA. * Propagate binding errors to callers and handle appropriately. v5: * Explicitly look for normal GGTT view in i915_gem_obj_bound to align usage in i915_gem_object_ggtt_unpin. (Michel Thierry) * Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry) * Removed stray semi-colon in i915_gem_object_set_cache_level. For: VIZ-4544 Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Michel Thierry <michel.thierry@intel.com> [danvet: Drop hunk from i915_gem_shrink since it's just prettification but upsets a __must_check warning.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-12-10 17:27:58 +00:00
return 0;
}
/**
* i915_ggtt_view_size - Get the size of a GGTT view.
* @obj: Object the view is of.
* @view: The view in question.
*
* @return The size of the GGTT view in bytes.
*/
size_t
i915_ggtt_view_size(struct drm_i915_gem_object *obj,
const struct i915_ggtt_view *view)
{
if (view->type == I915_GGTT_VIEW_NORMAL) {
return obj->base.size;
} else if (view->type == I915_GGTT_VIEW_ROTATED) {
return view->rotation_info.size;
} else if (view->type == I915_GGTT_VIEW_PARTIAL) {
return view->params.partial.size << PAGE_SHIFT;
} else {
WARN_ONCE(1, "GGTT view %u not implemented!\n", view->type);
return obj->base.size;
}
}