The idea here is to enable enforce isolation legacy
behavior for gfx 10+ and as such, we can adjust the
behavior to better suite that. This aligns with how
windows handles resets and seems to work reliably
in my testing on GFX 10+. For older chips, or if
enforce isolation is disabled, the soft recovery (wave
kill) tends to work better. Additionally, only
force completion if enforce isolation is not enabled.
Signed-off-by: Alex Deucher <alexander.deuc...@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index e57401ef85140..b1618a33d5f77 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -112,6 +112,7 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct
drm_sched_job *s_job)
amdgpu_job_core_dump(adev, job);
if (amdgpu_gpu_recovery &&
+ !job->enforce_isolation &&
amdgpu_ring_soft_recovery(ring, job->vmid, s_job->s_fence->parent))
{
dev_err(adev->dev, "ring %s timeout, but soft recovered\n",
s_job->sched->name);
@@ -161,7 +162,7 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct
drm_sched_job *s_job)
if (!r) {
if (is_guilty) {
atomic_inc(&ring->adev->gpu_reset_counter);
- if (ring->funcs->type != AMDGPU_RING_TYPE_GFX)
+ if (!job->enforce_isolation)
amdgpu_fence_driver_force_completion(ring);
}
drm_sched_wqueue_start(&ring->sched);
--
2.49.0
