amdgpu_ring_set_fence_errors_and_reemit() and amdgpu_ring_backup_unprocessed_commands() masked last_seq and sync_seq to fence-array indices before the loop, then walked a do/while until the two indices met. When hardware completed every command before the driver signalled the fences, last_seq == sync_seq, but the masked-index do/while could not tell that empty interval from a full table and walked the whole fence array, backing up and replaying already-completed commands.
Read the raw sequence numbers, return early through amdgpu_fence_process() when they are equal, and mask only at the array index. Guard the force-completion against a NULL guilty_fence so a non-guilty backup does not dereference it. Signed-off-by: Jesse Zhang <[email protected]> --- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 24 +++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 6537e8845fb6..3017ddcefbe2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -724,8 +724,14 @@ void amdgpu_ring_set_fence_errors_and_reemit(struct amdgpu_ring *ring, bool is_guilty_fence; bool is_guilty_context; - last_seq = amdgpu_fence_read(ring) & ring->fence_drv.num_fences_mask; - seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask; + last_seq = amdgpu_fence_read(ring); + seq = ring->fence_drv.sync_seq; + + /* Hardware completion can precede the interrupt which signals fences. */ + if (last_seq == seq) { + amdgpu_fence_process(ring); + return; + } /* If there is nothing to reemit, return early and set an error on the fence * if applicable. If all of the fences are siganlled, this will be a nop. @@ -743,9 +749,8 @@ void amdgpu_ring_set_fence_errors_and_reemit(struct amdgpu_ring *ring, spin_lock_irqsave(&ring->fence_drv.lock, flags); do { last_seq++; - last_seq &= ring->fence_drv.num_fences_mask; - ptr = &ring->fence_drv.fences[last_seq]; + ptr = &ring->fence_drv.fences[last_seq & ring->fence_drv.num_fences_mask]; rcu_read_lock(); unprocessed = rcu_dereference(*ptr); @@ -802,8 +807,8 @@ void amdgpu_ring_backup_unprocessed_commands(struct amdgpu_ring *ring, struct amdgpu_fence *fence; u32 seq, last_seq; - last_seq = amdgpu_fence_read(ring) & ring->fence_drv.num_fences_mask; - seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask; + last_seq = amdgpu_fence_read(ring); + seq = ring->fence_drv.sync_seq; ring->ring_backup_entries_to_copy = 0; /* @@ -816,11 +821,10 @@ void amdgpu_ring_backup_unprocessed_commands(struct amdgpu_ring *ring, } ring->guilty_fence = guilty_fence; - do { + while (last_seq != seq) { last_seq++; - last_seq &= ring->fence_drv.num_fences_mask; - ptr = &ring->fence_drv.fences[last_seq]; + ptr = &ring->fence_drv.fences[last_seq & ring->fence_drv.num_fences_mask]; rcu_read_lock(); unprocessed = rcu_dereference(*ptr); @@ -830,7 +834,7 @@ void amdgpu_ring_backup_unprocessed_commands(struct amdgpu_ring *ring, amdgpu_ring_backup_unprocessed_command(ring, fence); } rcu_read_unlock(); - } while (last_seq != seq); + } } struct amdgpu_fence * -- 2.49.0
