Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
drm/amdgpu: don't reemit ring contents more than once
If we cancel a bad job and reemit the ring contents, and
we get another timeout, cancel everything rather than reemitting.
The wptr markers are only relevant for the original emit. If
we reemit, the wptr markers are no longer correct.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
+19
-5
+17
-5
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+17
-5
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
···
709
709
struct amdgpu_ring *ring = af->ring;
710
710
unsigned long flags;
711
711
u32 seq, last_seq;
712
+
bool reemitted = false;
712
713
713
714
last_seq = amdgpu_fence_read(ring) & ring->fence_drv.num_fences_mask;
714
715
seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask;
···
727
726
if (unprocessed && !dma_fence_is_signaled_locked(unprocessed)) {
728
727
fence = container_of(unprocessed, struct amdgpu_fence, base);
729
728
730
-
if (fence == af)
729
+
if (fence->reemitted > 1)
730
+
reemitted = true;
731
+
else if (fence == af)
731
732
dma_fence_set_error(&fence->base, -ETIME);
732
733
else if (fence->context == af->context)
733
734
dma_fence_set_error(&fence->base, -ECANCELED);
···
737
734
rcu_read_unlock();
738
735
} while (last_seq != seq);
739
736
spin_unlock_irqrestore(&ring->fence_drv.lock, flags);
740
-
/* signal the guilty fence */
741
-
amdgpu_fence_write(ring, (u32)af->base.seqno);
742
-
amdgpu_fence_process(ring);
737
+
738
+
if (reemitted) {
739
+
/* if we've already reemitted once then just cancel everything */
740
+
amdgpu_fence_driver_force_completion(af->ring);
741
+
af->ring->ring_backup_entries_to_copy = 0;
742
+
} else {
743
+
/* signal the guilty fence */
744
+
amdgpu_fence_write(ring, (u32)af->base.seqno);
745
+
amdgpu_fence_process(ring);
746
+
}
743
747
}
744
748
745
749
void amdgpu_fence_save_wptr(struct amdgpu_fence *af)
···
794
784
/* save everything if the ring is not guilty, otherwise
795
785
* just save the content from other contexts.
796
786
*/
797
-
if (!guilty_fence || (fence->context != guilty_fence->context))
787
+
if (!fence->reemitted &&
788
+
(!guilty_fence || (fence->context != guilty_fence->context)))
798
789
amdgpu_ring_backup_unprocessed_command(ring, wptr,
799
790
fence->wptr);
800
791
wptr = fence->wptr;
792
+
fence->reemitted++;
801
793
}
802
794
rcu_read_unlock();
803
795
} while (last_seq != seq);