drm/msm: Don't overwrite hw fence in hw_init

Prior to the last commit, this could result in setting the GPU written fence value back to an older value, if we had missed updating completed_fence prior to suspend. This was mostly harmless as the GPU would eventually overwrite it again with the correct value. But we should just not do this. Instead just leave a sanity check that the fence looks plausible (in case the GPU scribbled on memory).

Rob Clark committed 4 years ago

Commit c8af2191 parent 3c7a5222 files · +9 −4Browse files

Unified · Split

2 files

drivers/gpu/drm/msm/adreno/adreno_gpu.c +8 −3
drivers/gpu/drm/msm/msm_gpu.c +1 −1

drivers/gpu/drm/msm/adreno/adreno_gpu.c+8−3

@@ -498,10 +498,15 @@ int adreno_hw_init(struct msm_gpu *gpu) · 497 unmodified lines
498	498
499	499	ring->cur = ring->start;
500	500	ring->next = ring->start;
501		−
502		− /* reset completed fence seqno: */
503		− ring->memptrs->fence = ring->fctx->completed_fence;
504	501	ring->memptrs->rptr = 0;
	502	+
	503	+ /* Detect and clean up an impossible fence, ie. if GPU managed
	504	+ * to scribble something invalid, we don't want that to confuse
	505	+ * us into mistakingly believing that submits have completed.
	506	+ */
	507	+ if (fence_before(ring->fctx->last_fence, ring->memptrs->fence)) {
	508	+ ring->memptrs->fence = ring->fctx->last_fence;
	509	+ }
505	510	}
506	511
507	512	return 0;

drivers/gpu/drm/msm/msm_gpu.c+1−1

@@ -418,7 +418,7 @@ static void recover_worker(struct kthread_work *work) · 417 unmodified lines
418	418	* one more to clear the faulting submit
419	419	*/
420	420	if (ring == cur_ring)
421		− fence++;
	421	+ ring->memptrs->fence = ++fence;
422	422
423	423	msm_update_fence(ring->fctx, fence);
424	424	}