Reporting -EIO from i915_wait_request() has proven very troublematic
over the years, with numerous hard-to-reproduce bugs cropping up in the
corner case of where a reset occurs and the code wasn't expecting such
an error.
If the we reset the GPU or have detected a hang and wish to reset the
GPU, the request is forcibly complete and the wait broken. Currently, we
report either -EAGAIN or -EIO in order for the caller to retreat and
restart the wait (if appropriate) after dropping and then reacquiring
the struct_mutex (essential to allow the GPU reset to proceed). However,
if we take the view that the request is complete (no further work will
be done on it by the GPU because it is dead and soon to be reset), then
we can proceed with the task at hand and then drop the struct_mutex
allowing the reset to occur. This transfers the burden of checking
whether it is safe to proceed to the caller, which in all but one
instance it is safe - completely eliminating the source of all spurious
-EIO.
Of note, we only have two API entry points where we expect that
userspace can observe an EIO. First is when submitting an execbuf, if
the GPU is terminally wedged, then the operation cannot succeed and an
-EIO is reported. Secondly, existing userspace uses the throttle ioctl
to detect an already wedged GPU before starting using HW acceleration
(or to confirm that the GPU is wedged after an error condition). So if
the GPU is wedged when the user calls throttle, also report -EIO.
v2: Split more carefully the change to i915_wait_request() and assorted
ABI from the reset handling.
v3: Add a couple of WARN_ON(EIO) to the interruptible modesetting code
so that we don't start to leak EIO there in future (and break our hang
resistant modesetting).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1460565315-7748-9-git-send-email-chris@chris-wilson.co.uk
Link: http://patchwork.freedesktop.org/patch/msgid/1460565315-7748-1-git-send-email-chris@chris-wilson.co.uk
As the request is only valid during the same global reset epoch, we can
record the current reset_counter when constructing the request and reuse
it when waiting upon that request in future. This removes a very hairy
atomic check serialised by the struct_mutex at the time of waiting and
allows us to transfer those waits to a central dispatcher for all
waiters and all requests.
PS: With per-engine resets, we obviously cannot assume a global reset
epoch for the requests - a per-engine epoch makes the most sense. The
challenge then is how to handle checking in the waiter for when to break
the wait, as the fine-grained reset may also want to requeue the
request (i.e. the assumption that just because the epoch changes the
request is completed may be broken - or we just avoid breaking that
assumption with the fine-grained resets).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1460565315-7748-7-git-send-email-chris@chris-wilson.co.uk
Since we only ever use the drm_i915_private from the stored
i915_mm_struct->dev, save some electrons by storing the right
backpointer.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1459864801-28606-3-git-send-email-chris@chris-wilson.co.uk
Holding a reference to the containing task_struct is not sufficient to
prevent the mm_struct from being reaped under memory pressure. If this
happens whilst we are calling get_user_pages(), explosions erupt -
sometimes an immediate GPF, sometimes page flag corruption. To prevent
the target mm from being reaped as we are reading from it, acquire a
reference before we begin.
Testcase: igt/gem_shrink/*userptr
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: stable@vger.kernel.org
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1459864801-28606-2-git-send-email-chris@chris-wilson.co.uk
In order to ensure that all invalidations are completed before the
operation returns to userspace (i.e. before the munmap() syscall returns)
we need to wait upon the outstanding operations.
We are allowed to block inside the invalidate_range_start callback, and
as struct_mutex is the inner lock with mmap_sem we can wait upon the
struct_mutex without provoking lockdep into warning about a deadlock.
However, we don't actually want to wait upon outstanding rendering
whilst holding the struct_mutex if we can help it otherwise we also
block other processes from submitting work to the GPU. So first we do a
wait without the lock and then when we reacquire the lock, we double
check that everything is ready for removing the invalidated pages.
Finally to wait upon the outstanding unpinning tasks, we create a
private workqueue as a means to conveniently wait upon all at once. The
drawback is that this workqueue is per-mm, so any threads concurrently
invalidating objects will wait upon each other. The advantage of using
the workqueue is that we can wait in parallel for completion of
rendering and unpinning of several objects (of particular importance if
the process terminates with a whole mm full of objects).
v2: Apply a cup of tea to the changelog.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94699
Testcase: igt/gem_userptr_blits/sync-unmap-cycles
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1459864801-28606-1-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJXCva8AAoJEHm+PkMAQRiGXBoIAIkrjxdbuT2nS9A3tHwkiFXa
6/Th1UjbNaoLuZ+MckQHayAD9NcWY9lVjOUmFsSiSWMCQK/rTWDl8x5ITputrY2V
VuhrJCwI7huEtu6GpRaJaUgwtdOjhIHz1Ue2MCdNIbKX3l+LjVyyJ9Vo8rruvZcR
fC7kiivH04fYX58oQ+SHymCg54ny3qJEPT8i4+g26686m11hvZLI3UAs2PAn6ut+
atCjxdQ4yLN3DWsbjuA7wYGWhTgFloxL4TIoisuOUc3FXnSi/ivIbXZvu4lUfisz
LA2JBhfII3AEMBWG9xfGbXPijJTT4q7yNlTD0oYcnMtAt/Roh2F04asqB1LetEY=
=bri6
-----END PGP SIGNATURE-----
Merge tag 'v4.6-rc3' into drm-intel-next-queued
Linux 4.6-rc3
Backmerge requested by Chris Wilson to make his patches apply cleanly.
Tiny conflict in vmalloc.c with the (properly acked and all) patch in
drm-intel-next:
commit 4da56b99d9
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Mon Apr 4 14:46:42 2016 +0100
mm/vmap: Add a notifier for when we run out of vmap address space
and Linus' tree.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
I have instances where I want to use drm_malloc_ab() but with a custom
gfp mask. And with those, where I want a temporary allocation, I want to
try a high-order kmalloc() before using a vmalloc().
So refactor my usage into drm_malloc_gfp().
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: dri-devel@lists.freedesktop.org
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: Dave Airlie <airlied@redhat.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1460113874-17366-6-git-send-email-chris@chris-wilson.co.uk
PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time
ago with promise that one day it will be possible to implement page
cache with bigger chunks than PAGE_SIZE.
This promise never materialized. And unlikely will.
We have many places where PAGE_CACHE_SIZE assumed to be equal to
PAGE_SIZE. And it's constant source of confusion on whether
PAGE_CACHE_* or PAGE_* constant should be used in a particular case,
especially on the border between fs and mm.
Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much
breakage to be doable.
Let's stop pretending that pages in page cache are special. They are
not.
The changes are pretty straight-forward:
- <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>;
- <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>;
- PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN};
- page_cache_get() -> get_page();
- page_cache_release() -> put_page();
This patch contains automated changes generated with coccinelle using
script below. For some reason, coccinelle doesn't patch header files.
I've called spatch for them manually.
The only adjustment after coccinelle is revert of changes to
PAGE_CAHCE_ALIGN definition: we are going to drop it later.
There are few places in the code where coccinelle didn't reach. I'll
fix them manually in a separate patch. Comments and documentation also
will be addressed with the separate patch.
virtual patch
@@
expression E;
@@
- E << (PAGE_CACHE_SHIFT - PAGE_SHIFT)
+ E
@@
expression E;
@@
- E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT)
+ E
@@
@@
- PAGE_CACHE_SHIFT
+ PAGE_SHIFT
@@
@@
- PAGE_CACHE_SIZE
+ PAGE_SIZE
@@
@@
- PAGE_CACHE_MASK
+ PAGE_MASK
@@
expression E;
@@
- PAGE_CACHE_ALIGN(E)
+ PAGE_ALIGN(E)
@@
expression E;
@@
- page_cache_get(E)
+ get_page(E)
@@
expression E;
@@
- page_cache_release(E)
+ put_page(E)
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull drm updates from Dave Airlie:
"This is the main drm pull request for 4.6 kernel.
Overall the coolest thing here for me is the nouveau maxwell signed
firmware support from NVidia, it's taken a long while to extract this
from them.
I also wish the ARM vendors just designed one set of display IP, ARM
display block proliferation is definitely increasing.
Core:
- drm_event cleanups
- Internal API cleanup making mode_fixup optional.
- Apple GMUX vga switcheroo support.
- DP AUX testing interface
Panel:
- Refactoring of DSI core for use over more transports.
New driver:
- ARM hdlcd driver
i915:
- FBC/PSR (framebuffer compression, panel self refresh) enabled by default.
- Ongoing atomic display support work
- Ongoing runtime PM work
- Pixel clock limit checks
- VBT DSI description support
- GEM fixes
- GuC firmware scheduler enhancements
amdkfd:
- Deferred probing fixes to avoid make file or link ordering.
amdgpu/radeon:
- ACP support for i2s audio support.
- Command Submission/GPU scheduler/GPUVM optimisations
- Initial GPU reset support for amdgpu
vmwgfx:
- Support for DX10 gen mipmaps
- Pageflipping and other fixes.
exynos:
- Exynos5420 SoC support for FIMD
- Exynos5422 SoC support for MIPI-DSI
nouveau:
- GM20x secure boot support - adds acceleration for Maxwell GPUs.
- GM200 support
- GM20B clock driver support
- Power sensors work
etnaviv:
- Correctness fixes for GPU cache flushing
- Better support for i.MX6 systems.
imx-drm:
- VBlank IRQ support
- Fence support
- OF endpoint support
msm:
- HDMI support for 8996 (snapdragon 820)
- Adreno 430 support
- Timestamp queries support
virtio-gpu:
- Fixes for Android support.
rockchip:
- Add support for Innosilicion HDMI
rcar-du:
- Support for 4 crtcs
- R8A7795 support
- RCar Gen 3 support
omapdrm:
- HDMI interlace output support
- dma-buf import support
- Refactoring to remove a lot of legacy code.
tilcdc:
- Rewrite of pageflipping code
- dma-buf support
- pinctrl support
vc4:
- HDMI modesetting bug fixes
- Significant 3D performance improvement.
fsl-dcu (FreeScale):
- Lots of fixes
tegra:
- Two small fixes
sti:
- Atomic support for planes
- Improved HDMI support"
* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (1063 commits)
drm/amdgpu: release_pages requires linux/pagemap.h
drm/sti: restore mode_fixup callback
drm/amdgpu/gfx7: add MTYPE definition
drm/amdgpu: removing BO_VAs shouldn't be interruptible
drm/amd/powerplay: show uvd/vce power gate enablement for tonga.
drm/amd/powerplay: show uvd/vce power gate info for fiji
drm/amdgpu: use sched fence if possible
drm/amdgpu: move ib.fence to job.fence
drm/amdgpu: give a fence param to ib_free
drm/amdgpu: include the right version of gmc header files for iceland
drm/radeon: fix indentation.
drm/amd/powerplay: add uvd/vce dpm enabling flag to fix the performance issue for CZ
drm/amdgpu: switch back to 32bit hw fences v2
drm/amdgpu: remove amdgpu_fence_is_signaled
drm/amdgpu: drop the extra fence range check v2
drm/amdgpu: signal fences directly in amdgpu_fence_process
drm/amdgpu: cleanup amdgpu_fence_wait_empty v2
drm/amdgpu: keep all fences in an RCU protected array v2
drm/amdgpu: add number of hardware submissions to amdgpu_fence_driver_init_ring
drm/amdgpu: RCU protected amd_sched_fence_release
...
commit e5756c10d8
Author: Imre Deak <imre.deak@intel.com>
Date: Fri Aug 14 18:43:30 2015 +0300
drm/i915/bxt: don't allow cached GEM mappings on A stepping
Added an exception of disallowing snooping for Broxton A
stepping hardware but userptr was still enabling it regardless.
Move the check to HAS_SNOOP now that it is used from multiple
call sites and use it.
v2: Userptr cannot be supported when it cannot be coherent and
generalize the code better. (Chris Wilson)
v3: Make has_snoop true only when !has_llc. (Chris Wilson)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1456920631-34302-1-git-send-email-tvrtko.ursulin@linux.intel.com
Elsewhere we have adopted the convention of using '_link' to denote
elements in the list (and '_list' for the actual list_head itself), and
that the name should indicate which list the link belongs to (and
preferrably not just where the link is being stored).
s/vma_link/obj_link/ (we iterate over obj->vma_list)
s/mm_list/vm_link/ (we iterate over vm->[in]active_list)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
For protection keys, we need to understand whether protections
should be enforced in software or not. In general, we enforce
protections when working on our own task, but not when on others.
We call these "current" and "remote" operations.
This patch introduces a new get_user_pages() variant:
get_user_pages_remote()
Which is a replacement for when get_user_pages() is called on
non-current tsk/mm.
We also introduce a new gup flag: FOLL_REMOTE which can be used
for the "__" gup variants to get this new behavior.
The uprobes is_trap_at_addr() location holds mmap_sem and
calls get_user_pages(current->mm) on an instruction address. This
makes it a pretty unique gup caller. Being an instruction access
and also really originating from the kernel (vs. the app), I opted
to consider this a 'remote' access where protection keys will not
be enforced.
Without protection keys, this patch should not change any behavior.
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Hansen <dave@sr71.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: jack@suse.cz
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/20160212210154.3F0E51EA@viggo.jf.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
commit 033908aed5
Author: Dave Gordon <david.s.gordon@intel.com>
Date: Thu Dec 10 18:51:23 2015 +0000
drm/i915: mark GEM object pages dirty when mapped & written by the CPU
introduced a check into i915_gem_object_get_dirty_pages() that returned
a NULL pointer when called with a bad object, one that was not backed by
shmemfs. This WARN was too strict as we can work on all struct page
backed objects, and resulted in a WARN + GPF for existing userspace. In
order to differentiate the various types of objects, add a new flags field
to the i915_gem_object_ops struct to describe their capabilities, with
the first flag being whether the object has struct pages.
v2: Drop silly const before an integer in the structure declaration.
Testcase: igt/gem_userptr_blits/relocations
Reported-and-tested-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Dave Gordon <david.s.gordon@intel.com>
Cc: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Tested-by: Michal Winiarski <michal.winiarski@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Fixes: 033908aed5 ("drm/i915: mark GEM object pages dirty when mapped & written by the CPU")
Link: http://patchwork.freedesktop.org/patch/msgid/1453487551-16799-1-git-send-email-chris@chris-wilson.co.uk
(cherry picked from commit de4726649b)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
commit 033908aed5
Author: Dave Gordon <david.s.gordon@intel.com>
Date: Thu Dec 10 18:51:23 2015 +0000
drm/i915: mark GEM object pages dirty when mapped & written by the CPU
introduced a check into i915_gem_object_get_dirty_pages() that returned
a NULL pointer when called with a bad object, one that was not backed by
shmemfs. This WARN was too strict as we can work on all struct page
backed objects, and resulted in a WARN + GPF for existing userspace. In
order to differentiate the various types of objects, add a new flags field
to the i915_gem_object_ops struct to describe their capabilities, with
the first flag being whether the object has struct pages.
v2: Drop silly const before an integer in the structure declaration.
Testcase: igt/gem_userptr_blits/relocations
Reported-and-tested-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Dave Gordon <david.s.gordon@intel.com>
Cc: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Dave Gordon <david.s.gordon@intel.com>
Reviewed-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>
Tested-by: Michal Winiarski <michal.winiarski@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1453487551-16799-1-git-send-email-chris@chris-wilson.co.uk
The generic interval tree we use to speed up range invalidation is an
augmented rbtree that can report all overlapping intervals for a given
range. Therefore we do not need to degrade to a linear list if we find
overlapping objects. Oops.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1453397563-2848-1-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Pull drm updates from Dave Airlie:
"I Was Almost Tempted To Capitalise Every Word, but then I decided I
couldn't read it myself!
I've also got one pull request for the sti driver outstanding. It
relied on a commit in Greg's tree and I didn't find out in time, that
commit is in your tree now so I might send that along once this is
merged.
I also had the accidental misfortune to have access to a Skylake on my
desk for a few days, and I've had to encourage Intel to try harder,
which seems to be happening now.
Here is the main drm-next pull request for 4.4.
Highlights:
New driver:
vc4 driver for the Rasberry Pi VPU.
(From Eric Anholt at Broadcom.)
Core:
Atomic fbdev support
Atomic helpers for runtime pm
dp/aux i2c STATUS_UPDATE handling
struct_mutex usage cleanups.
Generic of probing support.
Documentation:
Kerneldoc for VGA switcheroo code.
Rename to gpu instead of drm to reflect scope.
i915:
Skylake GuC firmware fixes
HPD A support
VBT backlight fallbacks
Fastboot by default for some systems
FBC work
BXT/SKL workarounds
Skylake deeper sleep state fixes
amdgpu:
Enable GPU scheduler by default
New atombios opcodes
GPUVM debugging options
Stoney support.
Fencing cleanups.
radeon:
More efficient CS checking
nouveau:
gk20a instance memory handling improvements.
Improved PGOB detection and GK107 support
Kepler GDDR5 PLL statbility improvement
G8x/GT2xx reclock improvements
new userspace API compatiblity fixes.
virtio-gpu:
Add 3D support - qemu 2.5 has it merged for it's gtk backend.
msm:
Initial msm88896 (snapdragon 8200)
exynos:
HDMI cleanups
Enable mixer driver byt default
Add DECON-TV support
vmwgfx:
Move to using memremap + fixes.
rcar-du:
Add support for R8A7793/4 DU
armada:
Remove support for non-component mode
Improved plane handling
Power savings while in DPMS off.
tda998x:
Remove unused slave encoder support
Use more HDMI helpers
Fix EDID read handling
dwhdmi:
Interlace video mode support for ipu-v3/dw_hdmi
Hotplug state fixes
Audio driver integration
imx:
More color formats support.
tegra:
Minor fixes/improvements"
[ Merge fixup: remove unused variable 'dev' that had all uses removed in
commit 4e270f0880: "drm/gem: Drop struct_mutex requirement from
drm_gem_mmap_obj" ]
* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (764 commits)
drm/vmwgfx: Relax irq locking somewhat
drm/vmwgfx: Properly flush cursor updates and page-flips
drm/i915/skl: disable display side power well support for now
drm/i915: Extend DSL readout fix to BDW and SKL.
drm/i915: Do graphics device reset under forcewake
drm/i915: Skip fence installation for objects with rotated views (v4)
vga_switcheroo: Drop client power state VGA_SWITCHEROO_INIT
drm/amdgpu: group together common fence implementation
drm/amdgpu: remove AMDGPU_FENCE_OWNER_MOVE
drm/amdgpu: remove now unused fence functions
drm/amdgpu: fix fence fallback check
drm/amdgpu: fix stoping the scheduler timeout
drm/amdgpu: cleanup on error in amdgpu_cs_ioctl()
drm/i915: Fix locking around GuC firmware load
drm/amdgpu: update Fiji's Golden setting
drm/amdgpu: update Fiji's rev id
drm/amdgpu: extract common code in vi_common_early_init
drm/amd/scheduler: don't oops on failure to load
drm/amdgpu: don't oops on failure to load (v2)
drm/amdgpu: don't VT switch on suspend
...
Pinning a userptr onto the hardware raises interesting questions about
the lifetime of such a surface as the framebuffer extends that life
beyond the client's address space. That is the hardware will need to
keep scanning out from the backing storage even after the client wants
to remap its address space. As the hardware pins the backing storage,
the userptr becomes invalid and this raises a WARN when the clients
tries to unmap its address space. The situation can be even more
complicated when the buffer is passed between processes, between a
client and display server, where the lifetime and hardware access is
even more confusing. Deny it.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: stable@vger.kernel.org
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Whilst discussing possible ways to trigger an invalidate_range on a
userptr with an aliased GGTT mmapping (and so cause a struct_mutex
deadlock), the conclusion is that we can, and we must, prevent any
possible deadlock by avoiding taking the mutex at all during
invalidate_range. This has numerous advantages all of which stem from
avoid the sleeping function from inside the unknown context. In
particular, it simplifies the invalidate_range because we no longer
have to juggle the spinlock/mutex and can just hold the spinlock
for the entire walk. To compensate, we have to make get_pages a bit more
complicated in order to serialise with a pending cancel_userptr worker.
As we hold the struct_mutex, we have no choice but to return EAGAIN and
hope that the worker is then flushed before we retry after reacquiring
the struct_mutex.
The important caveat is that the invalidate_range itself is no longer
synchronous. There exists a small but definite period in time in which
the old PTE's page remain accessible via the GPU. Note however that the
physical pages themselves are not invalidated by the mmu_notifier, just
the CPU view of the address space. The impact should be limited to a
delay in pages being flushed, rather than a possibility of writing to
the wrong pages. The only race condition that this worsens is remapping
an userptr active on the GPU where fresh work may still reference the
old pages due to struct_mutex contention. Given that userspace is racing
with the GPU, it is fair to say that the results are undefined.
v2: Only queue (and importantly only take one refcnt) the worker once.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Michał Winiarski found a really evil way to trigger a struct_mutex
deadlock with userptr. He found that if he allocated a userptr bo and
then GTT mmaped another bo, or even itself, at the same address as the
userptr using MAP_FIXED, he could then cause a deadlock any time we then
had to invalidate the GTT mmappings (so at will). Tvrtko then found by
repeatedly allocating GTT mmappings he could alias with an old userptr
mmap and also trigger the deadlock.
To counter act the deadlock, we make the observation that we only need
to take the struct_mutex if the object has any pages to revoke, and that
before userspace can alias with the userptr address space, it must have
invalidated the userptr->pages. Thus if we can check for those pages
outside of the struct_mutex, we can avoid the deadlock. To do so we
introduce a separate flag for userptr objects that we can inspect from
the mmu-notifier underneath its spinlock.
The patch makes one eye-catching change. That is the removal serial=0
after detecting a to-be-freed object inside the invalidate walker. I
felt setting serial=0 was a questionable pessimisation: it denies us the
chance to reuse the current iterator for the next loop (before it is
freed) and being explicit makes the reader question the validity of the
locking (since the object-free race could occur elsewhere). The
serialisation of the iterator is through the spinlock, if the object is
freed before the next loop then the notifier.serial will be incremented
and we start the walk from the beginning as we detect the invalid cache.
To try and tame the error paths and interactions with the userptr->active
flag, we have to do a fair amount of rearranging of get_pages_userptr().
v2: Grammar fixes
v3: Reorder set-active so that it is only set when obj->pages is set
(and so needs cancellation). Only the order of setting obj->pages and
the active-flag is crucial. Calling gup after invalidate-range begin
means the userptr sees the new set of backing storage (and so will not
need to invalidate its new pages), but we have to be careful not to set
the active-flag prior to successfully establishing obj->pages.
v4: Take the active->flag early so we know in the mmu-notifier when we
have to cancel a pending gup-worker.
v5: Rearrange the error path so that is not so convoluted
v6: Set pinned to 0 when negative before calling release_pages()
Reported-by: Michał Winiarski <michal.winiarski@intel.com>
Testcase: igt/gem_userptr_blits/map-fixed*
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: stable@vger.kernel.org
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The userptr worker allows for a slight race condition where upon there
may two or more threads calling get_user_pages for the same object. When
we have the array of pages, then we serialise the update of the object.
However, the worker should only overwrite the obj->userptr.work pointer
if and only if it is the active one. Currently we clear it for a
secondary worker with the effect that we may rarely force a second
lookup.
v2: Rebase and rename a variable to avoid 80cols
v3: Mention v2
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
GTT was only 32b and its max value is 4GB. In order to allow objects
bigger than 4GB in 48b PPGTT, i915_gem_userptr_ioctl we could check
against max 48b range (1ULL << 48).
But since the check no longer applies, just kill the limit.
v2: Use the default ctx to infer the ppgtt max size (Akash).
v3: Just kill the limit, it was only there for early detection of an
error when used for execbuffer (Chris).
Cc: Akash Goel <akash.goel@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Akash Goel <akash.goel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We have 3 types of DMA mappings for GEM objects:
1. physically contiguous for stolen and for objects needing contiguous
memory
2. DMA-buf mappings imported via a DMA-buf attach operation
3. SG DMA mappings for shmem backed and userptr objects
For 1. and 2. the lifetime of the DMA mapping matches the lifetime of the
corresponding backing pages and so in practice we create/release the
mapping in the object's get_pages/put_pages callback.
For 3. the lifetime of the mapping matches that of any existing GPU binding
of the object, so we'll create the mapping when the object is bound to
the first vma and release the mapping when the object is unbound from its
last vma.
Since the object can be bound to multiple vmas, we can end up creating a
new DMA mapping in the 3. case even if the object already had one. This
is not allowed by the DMA API and can lead to leaked mapping data and
IOMMU memory space starvation in certain cases. For example HW IOMMU
drivers (intel_iommu) allocate a new range from their memory space
whenever a mapping is created, silently overriding a pre-existing
mapping.
Fix this by moving the creation/removal of DMA mappings to the object's
get_pages/put_pages callbacks. These callbacks already check for and do
an early return in case of any nested calls. This way objects of the 3.
case also become more like the other object types.
I noticed this issue by enabling DMA debugging, which got disabled after
a while due to its internal mapping tables getting full. It also reported
errors in connection to random other drivers that did a DMA mapping for
an address that was previously mapped by i915 but was never released.
Besides these diagnostic messages and the memory space starvation
problem for IOMMUs, I'm not aware of this causing a real issue.
The fix is based on a patch from Chris.
v2:
- move the DMA mapping create/remove calls to the get_pages/put_pages
callbacks instead of adding new callbacks for these (Chris)
v3:
- also fix the get_page cache logic on the userptr async path (Chris)
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Mika encountered one pathological scenario under X where acquiring all
the mm locks (required to insert a mmu notifier) was very slow, so slow
that by the time we tried to lock the struct_mutex with the usual call
to i915_mutex_lock_interruptible(), X's signal timer had fired causing
us to restart the ioctl (and so looped indefinitely).
While I suspect this is the result of another bug (something leaking mm
perhaps?) we can forgo the error checking and interuptible nature of the
lock here so we only have to pay the expense once and get on with it.
This does expose the userptr creation routine to a driver livelock
though by not being interruptible.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
[danvet: Init ret to avoid issues reported by PRTS.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This results in a warning when building out of tree:
"cc1: warning: include/drm: No such file or directory [enabled by default]"
Most code already uses #include <drm/foo.h> correctly, so fix the
instances that don't.
Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
It's possible for invalidate_range_start mmu notifier callback to race
against userptr object release. If the gem object was released prior to
obtaining the spinlock in invalidate_range_start we're hitting null
pointer dereference.
Testcase: igt/gem_userptr_blits/stress-mm-invalidate-close
Testcase: igt/gem_userptr_blits/stress-mm-invalidate-close-overlap
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
[Jani: added code comment suggested by Chris]
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
sg_alloc_table_from_pages() can build us a table with coalesced ranges which
means we need to iterate over pages and not sg table entries when releasing
page references.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: "Barbalho, Rafael" <rafael.barbalho@intel.com>
Tested-by: Rafael Barbalho <rafael.barbalho@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
[danvet: Remove unused local variable sg.]
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
During release of the GEM object we hold the struct_mutex. As the
object may be holding onto the last reference for the task->mm,
calling mmput() may trigger exit_mmap() which close the vma
which will call drm_gem_vm_close() and attempt to reacquire
the struct_mutex. In order to avoid that recursion, we have
to defer the mmput() until after we drop the struct_mutex,
i.e. we need to schedule a worker to do the clean up. A further issue
spotted by Tvrtko was caused when we took a GTT mmapping of a userptr
buffer object. In that case, we would never call mmput as the object
would be cyclically referenced by the GTT mmapping and not freed upon
process exit - keeping the entire process mm alive after the process
task was reaped. The fix employed is to replace the mm_users/mmput()
reference handling to mm_count/mmdrop() for the shared i915_mm_struct.
INFO: task test_surfaces:1632 blocked for more than 120 seconds.
Tainted: GF O 3.14.5+ #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
test_surfaces D 0000000000000000 0 1632 1590 0x00000082
ffff88014914baa8 0000000000000046 0000000000000000 ffff88014914a010
0000000000012c40 0000000000012c40 ffff8800a0058210 ffff88014784b010
ffff88014914a010 ffff880037b1c820 ffff8800a0058210 ffff880037b1c824
Call Trace:
[<ffffffff81582499>] schedule+0x29/0x70
[<ffffffff815825fe>] schedule_preempt_disabled+0xe/0x10
[<ffffffff81583b93>] __mutex_lock_slowpath+0x183/0x220
[<ffffffff81583c53>] mutex_lock+0x23/0x40
[<ffffffffa005c2a3>] drm_gem_vm_close+0x33/0x70 [drm]
[<ffffffff8115a483>] remove_vma+0x33/0x70
[<ffffffff8115a5dc>] exit_mmap+0x11c/0x170
[<ffffffff8104d6eb>] mmput+0x6b/0x100
[<ffffffffa00f44b9>] i915_gem_userptr_release+0x89/0xc0 [i915]
[<ffffffffa00e6706>] i915_gem_free_object+0x126/0x250 [i915]
[<ffffffffa005c06a>] drm_gem_object_free+0x2a/0x40 [drm]
[<ffffffffa005cc32>] drm_gem_object_handle_unreference_unlocked+0xe2/0x120 [drm]
[<ffffffffa005ccd4>] drm_gem_object_release_handle+0x64/0x90 [drm]
[<ffffffff8127ffeb>] idr_for_each+0xab/0x100
[<ffffffffa005cc70>] ? drm_gem_object_handle_unreference_unlocked+0x120/0x120 [drm]
[<ffffffff81583c46>] ? mutex_lock+0x16/0x40
[<ffffffffa005c354>] drm_gem_release+0x24/0x40 [drm]
[<ffffffffa005b82b>] drm_release+0x3fb/0x480 [drm]
[<ffffffff8118d482>] __fput+0xb2/0x260
[<ffffffff8118d6de>] ____fput+0xe/0x10
[<ffffffff8106f27f>] task_work_run+0x8f/0xf0
[<ffffffff81052228>] do_exit+0x1a8/0x480
[<ffffffff81052551>] do_group_exit+0x51/0xc0
[<ffffffff810525d7>] SyS_exit_group+0x17/0x20
[<ffffffff8158e092>] system_call_fastpath+0x16/0x1b
v2: Incorporate feedback from Tvrtko and remove the unnessary mm
referencing when creating the i915_mm_struct and improve some of the
function names and comments.
Reported-by: Jacek Danecki <jacek.danecki@intel.com>
Test-case: igt/gem_userptr_blits/process-exit*
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Tested-by: "Gong, Zhipeng" <zhipeng.gong@intel.com>
Cc: Jacek Danecki <jacek.danecki@intel.com>
Cc: "Ursulin, Tvrtko" <tvrtko.ursulin@intel.com>
Reviewed-by: "Ursulin, Tvrtko" <tvrtko.ursulin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: stable@vger.kernel.org # hold off until 3.17 ships for additional testing
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Move the code around in order to acquire and release the spinlock in the
same function and in the same block. This keeps static analysers happy
and the reader sane.
Suggested-by: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Whilst I strongly advise against doing so for the implicit coherency
issues between the multiple buffer objects accessing the same backing
store, it nevertheless is a valid use case, akin to mmaping the same
file multiple times.
The reason why we forbade it earlier was that our use of the interval
tree for fast invalidation upon vma changes excluded overlapping
objects. So in the case where the user wishes to create such pairs of
overlapping objects, we degrade the range invalidation to walkin the
linear list of objects associated with the mm.
A situation where overlapping objects could arise is the lax implementation
of MIT-SHM Pixmaps in the xserver. A second situation is where the user
wishes to have different access modes to a region of memory (e.g. access
through a read-only userptr buffer and through a normal userptr buffer).
v2: Compile for mmu-notifiers after tweaking
v3: Rename is_linear/has_linear
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: "Li, Victor Y" <victor.y.li@intel.com>
Cc: "Kelley, Sean V" <sean.v.kelley@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: "Gong, Zhipeng" <zhipeng.gong@intel.com>
Cc: Akash Goel <akash.goel@intel.com>
Cc: "Volkin, Bradley D" <bradley.d.volkin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
During the range invalidate, we walk the list of buffers associated with
the mmu_notifer and find the ones that overlap the range. An
optimisation is made to speed up the iteration by assuming the previous
iter is still valid whilst the tree is unmodified. This exposes a bug
when a range invalidate is triggered after we have just created the
mmu_notifier, but before attaching any buffers. In that case, we presume
we have an unmodified list and start walking from the last iter which is
NULL. Oops.
The easiest fix is then to initialise the serial of the tree to 1.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Testecase: igt/gem_userptr_blts/stress-mm
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
By exporting the ability to map user address and inserting PTEs
representing their backing pages into the GTT, we can exploit UMA in order
to utilize normal application data as a texture source or even as a
render target (depending upon the capabilities of the chipset). This has
a number of uses, with zero-copy downloads to the GPU and efficient
readback making the intermixed streaming of CPU and GPU operations
fairly efficient. This ability has many widespread implications from
faster rendering of client-side software rasterisers (chromium),
mitigation of stalls due to read back (firefox) and to faster pipelining
of texture data (such as pixel buffer objects in GL or data blobs in CL).
v2: Compile with CONFIG_MMU_NOTIFIER
v3: We can sleep while performing invalidate-range, which we can utilise
to drop our page references prior to the kernel manipulating the vma
(for either discard or cloning) and so protect normal users.
v4: Only run the invalidate notifier if the range intercepts the bo.
v5: Prevent userspace from attempting to GTT mmap non-page aligned buffers
v6: Recheck after reacquire mutex for lost mmu.
v7: Fix implicit padding of ioctl struct by rounding to next 64bit boundary.
v8: Fix rebasing error after forwarding porting the back port.
v9: Limit the userptr to page aligned entries. We now expect userspace
to handle all the offset-in-page adjustments itself.
v10: Prevent vma from being copied across fork to avoid issues with cow.
v11: Drop vma behaviour changes -- locking is nigh on impossible.
Use a worker to load user pages to avoid lock inversions.
v12: Use get_task_mm()/mmput() for correct refcounting of mm.
v13: Use a worker to release the mmu_notifier to avoid lock inversion
v14: Decouple mmu_notifier from struct_mutex using a custom mmu_notifer
with its own locking and tree of objects for each mm/mmu_notifier.
v15: Prevent overlapping userptr objects, and invalidate all objects
within the mmu_notifier range
v16: Fix a typo for iterating over multiple objects in the range and
rearrange error path to destroy the mmu_notifier locklessly.
Also close a race between invalidate_range and the get_pages_worker.
v17: Close a race between get_pages_worker/invalidate_range and fresh
allocations of the same userptr range - and notice that
struct_mutex was presumed to be held when during creation it wasn't.
v18: Sigh. Fix the refactor of st_set_pages() to allocate enough memory
for the struct sg_table and to clear it before reporting an error.
v19: Always error out on read-only userptr requests as we don't have the
hardware infrastructure to support them at the moment.
v20: Refuse to implement read-only support until we have the required
infrastructure - but reserve the bit in flags for future use.
v21: use_mm() is not required for get_user_pages(). It is only meant to
be used to fix up the kernel thread's current->mm for use with
copy_user().
v22: Use sg_alloc_table_from_pages for that chunky feeling
v23: Export a function for sanity checking dma-buf rather than encode
userptr details elsewhere, and clean up comments based on
suggestions by Bradley.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: "Gong, Zhipeng" <zhipeng.gong@intel.com>
Cc: Akash Goel <akash.goel@intel.com>
Cc: "Volkin, Bradley D" <bradley.d.volkin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Reviewed-by: Brad Volkin <bradley.d.volkin@intel.com>
[danvet: Frob ioctl allocation to pick the next one - will cause a bit
of fuss with create2 apparently, but such are the rules.]
[danvet2: oops, forgot to git add after manual patch application]
[danvet3: Appease sparse.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
