So far the kfd driver implemented same routines for runtime and system
wide suspend and resume (s2idle or mem). During system wide suspend the
kfd aquires an atomic lock that prevents any more user processes to
create queues and interact with kfd driver and amd gpu. This mechanism
created problem when amdgpu device is runtime suspended with BACO
enabled. Any application that relies on kfd driver fails to load because
the driver reports a locked kfd device since gpu is runtime suspended.
However, in an ideal case, when gpu is runtime suspended the kfd driver
should be able to:
- auto resume amdgpu driver whenever a client requests compute service
- prevent runtime suspend for amdgpu while kfd is in use
This change refactors the amdgpu and amdkfd drivers to support BACO and
runtime power management.
Reviewed-by: Oak Zeng <oak.zeng@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
During system suspend the kfd driver aquires a lock that prohibits
further kfd actions unless the gpu is resumed. This adds some info which
can be useful while debugging.
Reviewed-by: Oak Zeng <oak.zeng@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
At the moment, only DRM_MODE_ROTATE_180 is allowed when we try to apply
the rotation from the video mode parameters. It is also useful to allow
DRM_MODE_ROTATE_0 in case there is only a reflect option in the video mode
parameter (e.g. video=540x960,reflect_x).
DRM_MODE_ROTATE_0 means "no rotation" and should therefore not require
any special handling, so we can just add it to the if condition.
Signed-off-by: Stephan Gerhold <stephan@gerhold.net>
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Link: https://patchwork.freedesktop.org/patch/msgid/20200117153429.54700-3-stephan@gerhold.net
A rotation value should have exactly one rotation angle.
At the moment there is no validation for this when parsing video=
parameters from the command line. This causes problems later on
when we try to combine the command line rotation with the panel
orientation.
To make sure that we generate a valid rotation value:
- Set DRM_MODE_ROTATE_0 by default (if no rotate= option is set)
- Validate that there is exactly one rotation angle set
(i.e. specifying the rotate= option multiple times is invalid)
Signed-off-by: Stephan Gerhold <stephan@gerhold.net>
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Link: https://patchwork.freedesktop.org/patch/msgid/20200117153429.54700-2-stephan@gerhold.net
If we encounter a hang on a virtual engine, as we process the hang the
request may already have been moved back to the virtual engine (we are
processing the hang on the physical engine). We need to reclaim the
request from the virtual engine so that the locking is consistent and
local to the real engine on which we will hold the request for error
state capturing.
v2: Pull the reclamation into execlists_hold() and assert that cannot be
called from outside of the reset (i.e. with the tasklet disabled).
v3: Added selftest
v4: Drop the reference owned by the virtual engine
Fixes: ad18ba7b5e ("drm/i915/execlists: Offline error capture")
Testcase: igt/gem_exec_balancer/hang
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200122140243.495621-2-chris@chris-wilson.co.uk
(cherry picked from commit 989df3a7bd)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Thanks to preempt-to-busy, we leave the request on the HW as we submit
the preemption request. This means that the request may complete at any
moment as we process HW events, and in particular the request may be
retired as we are planning to capture it for a preemption timeout.
Be more careful while obtaining the request to capture after a
preemption timeout, and check to see if it completed before we were able
to put it on the on-hold list. If we do see it did complete just before
we capture the request, proclaim the preemption-timeout a false positive
and pardon the reset as we should hit an arbitration point momentarily
and so be able to process the preemption.
Note that even after we move the request to be on hold it may be retired
(as the reset to stop the HW comes after), so we do require to hold our
own reference as we work on the request for capture (and all of the
peeking at state within the request needs to be carefully protected).
Fixes: c3f1ed90e6 ("drm/i915/gt: Allow temporary suspension of inflight requests")
Closes: https://gitlab.freedesktop.org/drm/intel/issues/997
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200122140243.495621-1-chris@chris-wilson.co.uk
(cherry picked from commit 4ba5c086a1)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Currently, we skip error capture upon forced preemption. We apply forced
preemption when there is a higher priority request that should be
running but is being blocked, and we skip inline error capture so that
the preemption request is not further delayed by a user controlled
capture -- extending the denial of service.
However, preemption reset is also used for heartbeats and regular GPU
hangs. By skipping the error capture, we remove the ability to debug GPU
hangs.
In order to capture the error without delaying the preemption request
further, we can do an out-of-line capture by removing the guilty request
from the execution queue and scheduling a worker to dump that request.
When removing a request, we need to remove the entire context and all
descendants from the execution queue, so that they do not jump past.
Closes: https://gitlab.freedesktop.org/drm/intel/issues/738
Fixes: 3a7a92aba8 ("drm/i915/execlists: Force preemption")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200116184754.2860848-3-chris@chris-wilson.co.uk
(cherry picked from commit 748317386a)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
In order to support out-of-line error capture, we need to remove the
active request from HW and put it to one side while a worker compresses
and stores all the details associated with that request. (As that
compression may take an arbitrary user-controlled amount of time, we
want to let the engine continue running on other workloads while the
hanging request is dumped.) Not only do we need to remove the active
request, but we also have to remove its context and all requests that
were dependent on it (both in flight, queued and future submission).
Finally once the capture is complete, we need to be able to resubmit the
request and its dependents and allow them to execute.
v2: Replace stack recursion with a simple list.
v3: Check all the parents, not just the first, when searching for a
stuck ancestor!
References: https://gitlab.freedesktop.org/drm/intel/issues/738
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200116184754.2860848-2-chris@chris-wilson.co.uk
(cherry picked from commit 32ff621fd7)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
This patch reworks the whole delayed deletion of BOs which aren't idle.
Instead of having two counters for the BO structure we resurrect the BO
when we find that a deleted BO is not idle yet.
This has many advantages, especially that we don't need to
increment/decrement the BOs reference counter any more when it
moves on the LRUs.
v2: remove duplicate ttm_tt_destroy, fix holde lock for LRU move
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: xinhui pan <xinhui.pan@amd.com>
Link: https://patchwork.freedesktop.org/patch/352912/
As we use a mutex to serialise the first acquire (as it may be a lengthy
operation), but only an atomic decrement for the release, we have to
be careful in case a second thread races and completes both
acquire/release as the first finishes its acquire.
Thread A Thread B
i915_active_acquire i915_active_acquire
atomic_read() == 0 atomic_read() == 0
mutex_lock() mutex_lock()
atomic_read() == 0
ref->active();
atomic_inc()
mutex_unlock()
atomic_read() == 1
i915_active_release
atomic_dec_and_test() -> 0
ref->retire()
atomic_inc() -> 1
mutex_unlock()
So thread A has acquired the ref->active_count but since the ref was
still active at the time, it did not initialise it. By switching the
check inside the mutex to an atomic increment only if already active, we
close the race.
Fixes: c9ad602fea ("drm/i915: Split i915_active.mutex into an irq-safe spinlock for the rbtree")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200126102346.1877661-3-chris@chris-wilson.co.uk
(cherry picked from commit ac0e331a62)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
If we have a set of active engines marked as being non-persistent, we
lose track of those if the user replaces those engines with
I915_CONTEXT_PARAM_ENGINES. As part of our uABI contract is that
non-persistent requests are terminated if they are no longer being
tracked by the user's context (in order to prevent a lost request
causing an untracked and so unstoppable GPU hang), we need to apply the
same context cancellation upon changing engines.
v2: Track stale engines[] so we only reap at context closure.
v3: Tvrtko spotted races with closing contexts and set-engines, so add a
veneer of kill-everything paranoia to clean up after losing a race.
Fixes: a0e047156c ("drm/i915/gem: Make context persistence optional")
Testcase: igt/gem_ctx_peristence/replace
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200211144831.1011498-1-chris@chris-wilson.co.uk
[WHY]
Programming DPPCLK to the same value currently set may cause
underflow while playing video in certain conditions.
[HOW]
Only program DPPCLK if clock is not the same as the
previous value programmed.
Signed-off-by: Sung Lee <sung.lee@amd.com>
Reviewed-by: Yongqiang Sun <yongqiang.sun@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
In DCN hardware sequencer we do actually call ATOM_INIT correctly per
pipe. The workaround is not necessary for command table offloading.
[How]
Drop the workaround since it's not needed.
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Reviewed-by: Chris Park <Chris.Park@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix warning during switching to dpg pause mode for
VCN firmware Version ENC: 1.1 DEC: 1 VEP: 0 Revision: 16
Signed-off-by: James Zhu <James.Zhu@amd.com>
Acked-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
GDS clear workaround will cause gfx failure in suspend/resume case.
[ 98.679559] [drm:amdgpu_device_ip_late_init [amdgpu]] *ERROR* late_init of IP block <gfx_v9_0> failed -110
[ 98.679561] PM: dpm_run_callback(): pci_pm_resume+0x0/0xa0 returns -110
[ 98.679562] PM: Device 0000:03:00.0 failed to resume async: error -110
As this workaround is specific to the HW bug of GDS's ECC error
existing in cold boot up, so bypass this workaround in suspend/
resume case after booting up.
Signed-off-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
hwc->conf was designated specifically for AMD APU IOMMU purposes. This
could cause problems in performance and/or function since APU IOMMU
implementation is elsewhere. Also hwc->conf and hwc->config are
different members of an anonymous union so hwc->conf aliases as
hw->last_tag.
Signed-off-by: Jonathan Kim <Jonathan.Kim@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
On PowerPC, the compiler will tag object files with whether they
use hard or soft float FP ABI and whether they use 64 or 128-bit
long double ABI. On systems with 64-bit long double ABI, a tag
will get emitted whenever a double is used, as on those systems
a long double is the same as a double. This will prevent linkage
as other files are being compiled with hard-float.
On ppc64, this code will never actually get used for the time
being, as the only currently existing hardware using it are the
Renoir APUs. Therefore, until this is testable and can be fixed
properly, at least make sure the build will not fail.
Signed-off-by: Daniel Kolesa <daniel@octaforge.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Support pause_state for multiple instance, and it will fix vcn2.5 DPG mode
power off issue on instance 1.
Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Starting from 14nm, the PLL is built into the PHY and the PLL is mapped
to PHY on 1 to 1 basis. In the code, the DP port is mapped to a PLL that was not
initialized. This causes DP to HDMI dongle to not light up the display.
[How]
Initializations added for PLL2 when creating resources.
Signed-off-by: Isabel Zhang <isabel.zhang@amd.com>
Reviewed-by: Eric Yang <eric.yang2@amd.com>
Acked-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Currently on execlists, we use a local hwsp for the kernel_context,
rather than the engine's HWSP, as this is the default for execlists.
However, seqno wrap requires allocating a new HWSP cacheline, and may
require pinning a new HWSP page in the GGTT. This operation requiring
pinning in the GGTT is not allowed within the kernel_context timeline,
as doing so may require re-entering the kernel_context in order to evict
from the GGTT. As we want to avoid requiring a new HWSP for the
kernel_context, we can use the permanently pinned engine's HWSP instead.
However to do so we must prevent the use of semaphores reading the
kernel_context's HWSP, as the use of semaphores do not support rollover
onto the same cacheline. Fortunately, the kernel_context is mostly
isolated, so unlikely to give benefit to semaphores.
Reported-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200210205722.794180-5-chris@chris-wilson.co.uk
[WHY & HOW]
Previously drain clk was unconstrained and fill clk was constrained on fclk.
We want to change it to fill clk unconstrained and drain clock constrained
to dcfclk.
Signed-off-by: Sung Lee <sung.lee@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[WHY]
The optimized_require flag is needed to set watermarks and clocks lower
in certain conditions. This flag is set to true and then set to false
while programming front end in dcn20.
[HOW]
Do not set the flag to false while disabling plane.
Signed-off-by: Sung Lee <sung.lee@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Driver crash with psr feature enabled due to divide-by-zero error.
This is a regression after rework to calculate static screen frame
number entry time.
[How]
Correct order of operations to avoid divide-by-zero.
Signed-off-by: Roman Li <roman.li@amd.com>
Reviewed-by: Zhan Liu <Zhan.Liu@amd.com>
Acked-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
