Merge tag 'drm-misc-fixes-2023-01-12' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

Several fixes for amdgpu (all addressing issues with fences), yet
another orientation quirk for a Lenovo device, a use-after-free fix for
virtio, a regression fix in TTM and a performance regression in drm
buddy.

Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Maxime Ripard <maxime@cerno.tech>
Link: https://patchwork.freedesktop.org/patch/msgid/20230112130954.pxt77g3a7rokha42@houat

2 files changed, 37 insertions, 18 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 8516c814bc9b..7b5ce00f0602 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -61,6 +61,8 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p,
 		amdgpu_ctx_put(p->ctx);
 		return -ECANCELED;
 	}
+
+	amdgpu_sync_create(&p->sync);
 	return 0;
 }
 
@@ -452,18 +454,6 @@ static int amdgpu_syncobj_lookup_and_add(struct amdgpu_cs_parser *p,
 	}
 
 	r = amdgpu_sync_fence(&p->sync, fence);
-	if (r)
-		goto error;
-
-	/*
-	 * When we have an explicit dependency it might be necessary to insert a
-	 * pipeline sync to make sure that all caches etc are flushed and the
-	 * next job actually sees the results from the previous one.
-	 */
-	if (fence->context == p->gang_leader->base.entity->fence_context)
-		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence);
-
-error:
 	dma_fence_put(fence);
 	return r;
 }
@@ -1188,10 +1178,19 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct drm_gpu_scheduler *sched;
 	struct amdgpu_bo_list_entry *e;
+	struct dma_fence *fence;
 	unsigned int i;
 	int r;
 
+	r = amdgpu_ctx_wait_prev_fence(p->ctx, p->entities[p->gang_leader_idx]);
+	if (r) {
+		if (r != -ERESTARTSYS)
+			DRM_ERROR("amdgpu_ctx_wait_prev_fence failed.\n");
+		return r;
+	}
+
 	list_for_each_entry(e, &p->validated, tv.head) {
 		struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
 		struct dma_resv *resv = bo->tbo.base.resv;
@@ -1211,10 +1210,24 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 			return r;
 	}
 
-	r = amdgpu_ctx_wait_prev_fence(p->ctx, p->entities[p->gang_leader_idx]);
-	if (r && r != -ERESTARTSYS)
-		DRM_ERROR("amdgpu_ctx_wait_prev_fence failed.\n");
-	return r;
+	sched = p->gang_leader->base.entity->rq->sched;
+	while ((fence = amdgpu_sync_get_fence(&p->sync))) {
+		struct drm_sched_fence *s_fence = to_drm_sched_fence(fence);
+
+		/*
+		 * When we have an dependency it might be necessary to insert a
+		 * pipeline sync to make sure that all caches etc are flushed and the
+		 * next job actually sees the results from the previous one
+		 * before we start executing on the same scheduler ring.
+		 */
+		if (!s_fence || s_fence->sched != sched)
+			continue;
+
+		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence);
+		if (r)
+			return r;
+	}
+	return 0;
 }
 
 static void amdgpu_cs_post_dependencies(struct amdgpu_cs_parser *p)
@@ -1254,9 +1267,12 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 			continue;
 
 		fence = &p->jobs[i]->base.s_fence->scheduled;
+		dma_fence_get(fence);
 		r = drm_sched_job_add_dependency(&leader->base, fence);
-		if (r)
+		if (r) {
+			dma_fence_put(fence);
 			goto error_cleanup;
+		}
 	}
 
 	if (p->gang_size > 1) {
@@ -1344,6 +1360,7 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser)
 {
 	unsigned i;
 
+	amdgpu_sync_free(&parser->sync);
 	for (i = 0; i < parser->num_post_deps; i++) {
 		drm_syncobj_put(parser->post_deps[i].syncobj);
 		kfree(parser->post_deps[i].chain);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
index bac7976975bd..dcd8c066bc1f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
@@ -391,8 +391,10 @@ int amdgpu_sync_push_to_job(struct amdgpu_sync *sync, struct amdgpu_job *job)
 
 		dma_fence_get(f);
 		r = drm_sched_job_add_dependency(&job->base, f);
-		if (r)
+		if (r) {
+			dma_fence_put(f);
 			return r;
+		}
 	}
 	return 0;
 }