when loading bad page in the same nps mode, need to set the other fields
fields in eeprom records manually besides retired_page
Signed-off-by: ganglxie <ganglxie@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When the driver is unloaded, the interrupt source of
the rma device is not released, resulting in the failure
of hw_init when loading again using bad_page_threshold.
Signed-off-by: Ce Sun <cesun102@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Among the scheduler's statuses, the only one that indicates an error is
DRM_GPU_SCHED_STAT_ENODEV. Any status other than DRM_GPU_SCHED_STAT_ENODEV
signifies that the operation succeeded and the GPU is in a nominal state.
However, to provide more information about the GPU's status, it is needed
to convey more information than just "OK".
Therefore, rename DRM_GPU_SCHED_STAT_NOMINAL to
DRM_GPU_SCHED_STAT_RESET, which better communicates the meaning of this
status. The status DRM_GPU_SCHED_STAT_RESET indicates that the GPU has
hung, but it has been successfully reset and is now in a nominal state
again.
Reviewed-by: Philipp Stanner <phasta@kernel.org>
Link: https://lore.kernel.org/r/20250714-sched-skip-reset-v6-1-5c5ba4f55039@igalia.com
Signed-off-by: Maíra Canal <mcanal@igalia.com>
When a ring reset happens, amdgpu calls drm_dev_wedged_event() using
struct amdgpu_task_info *ti as one of the arguments. After using *ti, a
call to amdgpu_vm_put_task_info(ti) is required to correctly track its
lifetime.
However, it's called from a place that the ring reset path never reaches
due to a goto after drm_dev_wedged_event() is called. Move
amdgpu_vm_put_task_info() bellow the exit label to make sure that it's
called regardless of the code path.
amdgpu_vm_put_task_info() can only accept a valid address or NULL as
argument, so initialise *ti to make sure we can call this function if
*ti isn't used.
Fixes: a72002cb18 ("drm/amdgpu: Make use of drm_wedge_task_info")
Reported-by: Dave Airlie <airlied@gmail.com>
Closes: https://lore.kernel.org/dri-devel/CAPM=9tz0rQP8VZWKWyuF8kUMqRScxqoa6aVdwWw9=5yYxyYQ2Q@mail.gmail.com/
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://lore.kernel.org/r/20250704030629.1064397-1-andrealmeid@igalia.com
Signed-off-by: André Almeida <andrealmeid@igalia.com>
When hibernate with data center dGPUs, huge number of VRAM BOs evicted
to GTT and takes too much system memory. This will cause hibernation
fail due to insufficient memory for creating the hibernation image.
Move GTT BOs to shmem in KMD, then shmem to swap disk in kernel
hibernation code to make room for hibernation image.
Signed-off-by: Samuel Zhang <guoqing.zhang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://lore.kernel.org/r/20250710062313.3226149-3-guoqing.zhang@amd.com
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
This reverts commit 5fb90421fa.
The original patch moved `amdgpu_userq_mgr_fini()` to the driver's
`postclose` callback, which is called after `drm_gem_release()` in
the DRM file cleanup sequence.If a user application crashes or aborts
without cleaning up its user queues, 'drm_gem_release()` may free
GEM objects that are still referenced by active user queues, leading
to use-after-free. By reverting, we ensure that user queues are
disabled and cleaned up before any GEM objects are released,
preventing this class of bug. However, this reintroduces a race
during PCI hot-unplug, where device removal can race with per-file
cleanup, leading to use-after-free in suspend/unplug paths.
This will be fixed in the next patch.
Fixes: 5fb90421fa ("drm/amdgpu: fix slab-use-after-free in amdgpu_userq_mgr_fini+0x70c")
Signed-off-by: Vitaly Prosyak <vitaly.prosyak@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add a parameter to amdgpu_sdma_reset_engine() to let the
caller handle the kernel rings. This allows the kernel
rings to back up their unprocessed state if the reset comes in
via the drm scheduler rather than KFD.
Reviewed-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Move the force completion handling into the common
engine reset function. No need to duplicate it for
every IP version.
Reviewed-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For extended wait with retries on a PSP register value, add a noverbose
flag to avoid excessive error messages on each timeout.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Need to reinit the ring before remapping it and all of
the KIQ handling needs to be within the kiq lock.
Fixes: 1741281a15 ("drm/amdgpu/gfx10: add ring reset callbacks")
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Pass amdgpu device context instead of drm device context to some
amdgpu_device_* functions. DRM device context is not required in those
functions. No functional change.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
patch dd64956685 ("drm/amdgpu: Remove duplicated "context still
alive" check") removed ctx put, which will cause amdgpu_ctx_fini()
cannot be called and then cause some finished fence that added by
amdgpu_ctx_add_fence() cannot be released and cause memleak.
Fixes: dd64956685 ("drm/amdgpu: Remove duplicated "context still alive" check")
Signed-off-by: Lin.Cao <lincao12@amd.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 8cf66089e2)
Cc: stable@vger.kernel.org
SDMA 5.x only supports engine soft reset which resets
all queues on the engine. As such, we need to suspend
KFD queues around resets like we do for SDMA 4.x.
Reviewed-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 61feed0baa)
Fill in the SDMA ucode version checks for more SDMA 6.x parts.
v2: squash in fixes (Alex)
Reviewed-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
when eeprom initialization failed, we still support ras error injection,
and reserve bad pages, but do not save bad pages to eeprom
Signed-off-by: ganglxie <ganglxie@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The severity of CPER for BP threshold exceed event should be set as
CPER_SEV_FATAL to match the OOB implementation.
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Move the scheduler wqueue stopping and starting into
the ring reset callbacks. On some IPs we have to reset
an engine which may have multiple queues. Move the wqueue
handling into the backend so we can handle them as needed
based on the type of reset available.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Move guilty logic into the ring reset callbacks. This
allows each ring reset callback to better handle fence
errors and force completions in line with the reset
behavior for each IP. It also allows us to remove
the ring guilty callback since that logic now lives
in the reset callback.
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Move the force completion handling into each ring
reset function so that each engine can determine
whether or not it needs to force completion on the
jobs in the ring.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
What we actually care about is the amdgpu_fence object
so pass that in explicitly to avoid possible mistakes
in the future.
The job_run_counter handling can be safely removed at this
point as we no longer support job resubmission.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
patch dd64956685 ("drm/amdgpu: Remove duplicated "context still
alive" check") removed ctx put, which will cause amdgpu_ctx_fini()
cannot be called and then cause some finished fence that added by
amdgpu_ctx_add_fence() cannot be released and cause memleak.
Fixes: dd64956685 ("drm/amdgpu: Remove duplicated "context still alive" check")
Signed-off-by: Lin.Cao <lincao12@amd.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The "return true;" line wasn't indented. Also checkpatch likes when
we align the && conditions.
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Avoid dereferencing struct drm_gem_object.import_attach for the
imported dma-buf. The dma_buf field in the GEM object instance refers
to the same buffer. Prepares to make import_attach an implementation
detail of PRIME.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Instead of testing import_attach for imported GEM buffers, invoke
drm_gem_is_imported() to do the test.
v2:
- keep amdgpu_bo_print_info() as-is (Christian)
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
