If KFD_SUPPORT_IOMMU_V2 is not set, gcc warns:
drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_device.c:121:37: warning: ‘raven_device_info’ defined but not used [-Wunused-const-variable=]
static const struct kfd_device_info raven_device_info = {
^~~~~~~~~~~~~~~~~
As Huang Rui suggested, Raven already has the fallback path,
so it should be out of IOMMU v2 flag.
Suggested-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: Huang Rui <ray.huang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
XGMI support is more complicated than single device support as
questions of synchronization between the device recovering from
PCI error and other members of the hive are required.
Leaving this for next round.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cache the PCI state on boot and before each case where we might
loose it.
v2: Add pci_restore_state while caching the PCI state to avoid
breaking PCI core logic for stuff like suspend/resume.
v3: Extract pci_restore_state from amdgpu_device_cache_pci_state
to avoid superflous restores during GPU resets and suspend/resumes.
v4: Style fixes.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
At this point the ASIC is already post reset by the HW/PSP
so the HW not in proper state to be configured for suspension,
some blocks might be even gated and so best is to avoid touching it.
v2: Rename in_dpc to more meaningful name
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In function flr_work, we should do gpu recovery when no job
is running. Fix the logic by inverting it.
v2: modify the description
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Liu ChengZhe <ChengZhe.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In the resume stage of GPU recovery, start_cpsch will call pm_init
which set pm->allocated as false, cause the next pm_release_ib has
no chance to release ib memory.
Add pm_release_ib in stop_cpsch which will be called in the suspend
stage of GPU recovery.
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Paul needs 1a21e5b930 ("drm/ingenic: Fix leak of device_node
pointer") and 3b5b005ef7 ("drm/ingenic: Fix driver not probing when
IPU port is missing") from -fixes to be able to merge further ingenic
patches into -next.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This function should be an int, not a bool.
Presumably because we had the same 2 reverts in a slightly different
way, git got confused.
Thanks to Dan for reporting. :)
The conflict is between the 3 reverts in drm-fixes:
4993a8a378 ("Revert "drm/i915: Remove i915_gem_object_get_dirty_page()"")
ad5d95e4d5 ("Revert "drm/i915/gem: Async GPU relocations only"")
20561da3a2 ("Revert "drm/i915/gem: Delete unused code"")
And the slightly different combined revert in drm-intel-gt-next, but
with the same goal:
102a0a9051 ("Revert "drm/i915/gem: Async GPU relocations only"")
In the merge commit 1f4b2aca79 ("Merge tag
'drm-intel-gt-next-2020-09-07' of git://anongit.freedesktop.org/drm/drm-intel into drm-next") things
went wrong, but the merge commit view now doesn't show any conflict
anymore (as git tends to do when the resolution picks one or the other
branch).
The need to handle other than just true/false error codes in
__reloc_entry_gpu was added in the dma_resv locking changes in
c43ce12328 ("drm/i915: Use per object locking in execbuf, v12.")
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Dave Airlie <airlied@redhat.com>
[danvet: Explain this entire saga a lot better, adding tons of commit
references. Also note that this was merged before full intel-gfx-CI
results, only after BAT, since the breakage at the BAT run is already
severe enough to block all pre-merge testing.]
Fixes: 1f4b2aca79 ("Merge tag 'drm-intel-gt-next-2020-09-07' of git://anongit.freedesktop.org/drm/drm-intel into drm-next")
Acked-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20200910111225.2184193-1-maarten.lankhorst@linux.intel.com
(Same content as drm-intel-gt-next-2020-09-04-3, S-o-b's added)
UAPI Changes:
(- Potential implicit changes from WW locking refactoring)
Cross-subsystem Changes:
(- WW locking changes should align the i915 locking more with others)
Driver Changes:
- MAJOR: Apply WW locking across the driver (Maarten)
- Reverts for 5 commits to make applying WW locking faster (Maarten)
- Disable preparser around invalidations on Tigerlake for non-RCS engines (Chris)
- Add missing dma_fence_put() for error case of syncobj timeline (Chris)
- Parse command buffer earlier in eb_relocate(slow) to facilitate backoff (Maarten)
- Pin engine before pinning all objects (Maarten)
- Rework intel_context pinning to do everything outside of pin_mutex (Maarten)
- Avoid tracking GEM context until registered (Cc: stable, Chris)
- Provide a fastpath for waiting on vma bindings (Chris)
- Fixes to preempt-to-busy mechanism (Chris)
- Distinguish the virtual breadcrumbs from the irq breadcrumbs (Chris)
- Switch to object allocations for page directories (Chris)
- Hold context/request reference while breadcrumbs are active (Chris)
- Make sure execbuffer always passes ww state to i915_vma_pin (Maarten)
- Code refactoring to facilitate use of WW locking (Maarten)
- Locking refactoring to use more granular locking (Maarten, Chris)
- Support for multiple pinned timelines per engine (Chris)
- Move complication of I915_GEM_THROTTLE to the ioctl from general code (Chris)
- Make active tracking/vma page-directory stash work preallocated (Chris)
- Avoid flushing submission tasklet too often (Chris)
- Reduce context termination list iteration guard to RCU (Chris)
- Reductions to locking contention (Chris)
- Fixes for issues found by CI (Chris)
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Joonas Lahtinen <jlahtine@jlahtine-mobl.ger.corp.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200907130039.GA27766@jlahtine-mobl.ger.corp.intel.com
Commit '6f6a73c8b715d595977774d48450a734297ab21f' from Linus' tree
The fixes reverts cause a bit of a conflict pain with intel next,
start fixing it up here.
Signed-off-by: Dave Airlie <airlied@redhat.com>
In commit 4f0b4352bd ("drm/i915: Extract cdclk requirements checking
to separate function") the order of force_min_cdclk_changed check and
intel_modeset_checks(), was reversed. This broke the mechanism to
immediately force a new CDCLK minimum, and lead to driver probe
errors for display audio on GLK platform with 5.9-rc1 kernel. Fix
the issue by moving intel_modeset_checks() call later.
[vsyrjala: It also broke the ability of planes to bump up the cdclk
and thus could lead to underruns when eg. flipping from 32bpp to
64bpp framebuffer. To be clear, we still compute the new cdclk
correctly but fail to actually program it to the hardware due to
intel_set_cdclk_{pre,post}_plane_update() not getting called on
account of state->modeset==false.]
Fixes: 4f0b4352bd ("drm/i915: Extract cdclk requirements checking to separate function")
BugLink: https://github.com/thesofproject/linux/issues/2410
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200901151036.1312357-1-kai.vehmanen@linux.intel.com
(cherry picked from commit cf696856bc)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
These commits caused a regression on Lenovo t520 sandybridge
machine belonging to reporter. We are reverting them for 5.10
for other reasons, so just do it for 5.9 as well.
This reverts commit 7ac2d2536d.
Reported-by: Harald Arnesen <harald@skogtun.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
These commits caused a regression on Lenovo t520 sandybridge
machine belonging to reporter. We are reverting them for 5.10
for other reasons, so just do it for 5.9 as well.
This reverts commit 9e0f9464e2.
Reported-by: Harald Arnesen <harald@skogtun.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
These commits caused a regression on Lenovo t520 sandybridge
machine belonging to reporter. We are reverting them for 5.10
for other reasons, so just do it for 5.9 as well.
This reverts commit 763fedd6a2.
Reported-by: Harald Arnesen <harald@skogtun.org>
Signed-off-by: Dave Airlie <airied@redhat.com>
The hwsp_gtt object is used for sub-allocation and could therefore
be shared by many contexts causing unnecessary contention during
concurrent context pinning.
However since we're currently locking it only for pinning, it remains
resident until we unpin it, and therefore it's safe to drop the
lock early, allowing for concurrent thread access.
Signed-off-by: Thomas Hellström <thomas.hellstrom@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
To implement preempt-to-busy (and so efficient timeslicing and best utilization
of the hardware submission ports) we let the GPU run asynchronously in respect
to the ELSP submission queue. This created challenges in keeping and accessing
the driver state mirroring the asynchronous GPU execution.
Previous fix 1d9221e9d3 ("drm/i915: Skip signaling a signaled request")
however did not correctly serialize request retirement with the execution
callbacks.
We were using the i915_request.lock to serialise adding an execution callback
with __i915_request_submit. However, if we use an atomic llist_add to serialise
multiple waiters and then check to see if the request is already executing, we
can remove the irq-spinlock and fix serialization between retirement and
execution callbacks in one go.
v2: Avoid using the irq_work when outside of the irq-spinlocks, where we
can execute the callbacks immediately.
v3: Pay close attention to the order of setting ACTIVE on retirement, we
need to ensure the request is signaled and breadcrumbs detached before
we finish removing the request from the engine.
v4: Expanded commit message.
Fixes: 1d9221e9d3 ("drm/i915: Skip signaling a signaled request")
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/20200716142207.13003-2-chris@chris-wilson.co.uk
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
[Joonas: Rebased and reordered into drm-intel-gt-next branch]
[Joonas: Added expanded commit message from Tvrtko and Chris]
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
To implement preempt-to-busy (and so efficient timeslicing and best utilization
of the hardware submission ports) we let the GPU run asynchronously in respect
to the ELSP submission queue. This created challenges in keeping and accessing
the driver state mirroring the asynchronous GPU execution.
The latest occurence of this was spotted by KCSAN:
[ 1413.563200] BUG: KCSAN: data-race in __await_execution+0x217/0x370 [i915]
[ 1413.563221]
[ 1413.563236] race at unknown origin, with read to 0xffff88885bb6c478 of 8 bytes by task 9654 on cpu 1:
[ 1413.563548] __await_execution+0x217/0x370 [i915]
[ 1413.563891] i915_request_await_dma_fence+0x4eb/0x6a0 [i915]
[ 1413.564235] i915_request_await_object+0x421/0x490 [i915]
[ 1413.564577] i915_gem_do_execbuffer+0x29b7/0x3c40 [i915]
[ 1413.564967] i915_gem_execbuffer2_ioctl+0x22f/0x5c0 [i915]
[ 1413.564998] drm_ioctl_kernel+0x156/0x1b0
[ 1413.565022] drm_ioctl+0x2ff/0x480
[ 1413.565046] __x64_sys_ioctl+0x87/0xd0
[ 1413.565069] do_syscall_64+0x4d/0x80
[ 1413.565094] entry_SYSCALL_64_after_hwframe+0x44/0xa9
To complicate matters, we have to both avoid the read tearing of *active and
avoid any write tearing as perform the pending[] -> inflight[] promotion of the
execlists.
This is because we cannot rely on the memcpy doing u64 aligned copies on all
kernels/platforms and so we opt to open-code it with explicit WRITE_ONCE
annotations to satisfy KCSAN.
v2: When in doubt, write the same comment again.
v3: Expanded commit message.
Fixes: b55230e5e8 ("drm/i915: Check for awaits on still currently executing requests")
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/20200716142207.13003-1-chris@chris-wilson.co.uk
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
[Joonas: Rebased and reordered into drm-intel-gt-next branch]
[Joonas: Added expanded commit message from Tvrtko and Chris]
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
We want to lock all gem objects, including the engine context objects,
rework the throttling to ensure that we can do this. Now we only throttle
once, but can take eb_pin_engine while acquiring objects. This means we
will have to drop the lock to wait. If we don't have to throttle we can
still take the fastpath, if not we will take the slowpath and wait for
the throttle request while unlocked.
The engine has to be pinned as first step, otherwise gpu relocations
won't work.
Changes since v1:
- Only need to get a throttled request in the fastpath, no need for
a global flag any more.
- Always free the waited request correctly.
Changes since v2:
- Use intel_engine_pm_get()/put() to keeep engine pool alive during
EDEADLK handling.
Changes since v3:
- Fix small rq leak.
Changes since v4:
- Use a single reloc_context, for intel_context_pin_ww().
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200819140904.1708856-13-maarten.lankhorst@linux.intel.com
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Now that we changed execbuf submission slightly to allow us to do all
pinning in one place, we can now simply add ww versions on top of
struct_mutex. All we have to do is a separate path for -EDEADLK
handling, which needs to unpin all gem bo's before dropping the lock,
then starting over.
This finally allows us to do parallel submission, but because not
all of the pinning code uses the ww ctx yet, we cannot completely
drop struct_mutex yet.
Changes since v1:
- Keep struct_mutex for now. :(
Changes since v2:
- Make sure we always lock the ww context in slowpath.
Changes since v3:
- Don't call __eb_unreserve_vma in eb_move_to_gpu now; this can be
done on normal unlock path.
- Unconditionally release vmas and context.
Changes since v4:
- Rebased on top of struct_mutex reduction.
Changes since v5:
- Remove training wheels.
Changes since v6:
- Fix accidentally broken -ENOSPC handling.
Changes since v7:
- Handle gt buffer pool better.
Changes since v8:
- Properly clear variables, to make -EDEADLK handling not BUG.
Change since v9:
- Fix unpinning fence on pnv and below.
Changes since v10:
- Make relocation gpu chaining working again.
Changes since v11:
- Remove relocation chaining, pain to make it work.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200819140904.1708856-9-maarten.lankhorst@linux.intel.com
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
