drm/amdgpu: fix and cleanup user fence handling v2

We leaked the BO in the error pass, additional to that we only have
one user fence for all IBs in a job.

v2: remove white space changes

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 25 insertions, 30 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 9ab2f0886a14..2bbeeb07c187 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -87,33 +87,30 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
 }
 
 static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
-				      struct amdgpu_user_fence *uf,
-				      struct drm_amdgpu_cs_chunk_fence *fence_data)
+				      struct drm_amdgpu_cs_chunk_fence *data,
+				      uint32_t *offset)
 {
 	struct drm_gem_object *gobj;
-	uint32_t handle;
 
-	handle = fence_data->handle;
 	gobj = drm_gem_object_lookup(p->adev->ddev, p->filp,
-				     fence_data->handle);
+				     data->handle);
 	if (gobj == NULL)
 		return -EINVAL;
 
-	uf->bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
-	uf->offset = fence_data->offset;
-
-	if (amdgpu_ttm_tt_get_usermm(uf->bo->tbo.ttm)) {
-		drm_gem_object_unreference_unlocked(gobj);
-		return -EINVAL;
-	}
-
-	p->uf_entry.robj = amdgpu_bo_ref(uf->bo);
+	p->uf_entry.robj = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
 	p->uf_entry.priority = 0;
 	p->uf_entry.tv.bo = &p->uf_entry.robj->tbo;
 	p->uf_entry.tv.shared = true;
 	p->uf_entry.user_pages = NULL;
+	*offset = data->offset;
 
 	drm_gem_object_unreference_unlocked(gobj);
+
+	if (amdgpu_ttm_tt_get_usermm(p->uf_entry.robj->tbo.ttm)) {
+		amdgpu_bo_unref(&p->uf_entry.robj);
+		return -EINVAL;
+	}
+
 	return 0;
 }
 
@@ -124,8 +121,8 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 	union drm_amdgpu_cs *cs = data;
 	uint64_t *chunk_array_user;
 	uint64_t *chunk_array;
-	struct amdgpu_user_fence uf = {};
 	unsigned size, num_ibs = 0;
+	uint32_t uf_offset = 0;
 	int i;
 	int ret;
 
@@ -200,7 +197,8 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 				goto free_partial_kdata;
 			}
 
-			ret = amdgpu_cs_user_fence_chunk(p, &uf, (void *)p->chunks[i].kdata);
+			ret = amdgpu_cs_user_fence_chunk(p, p->chunks[i].kdata,
+							 &uf_offset);
 			if (ret)
 				goto free_partial_kdata;
 
@@ -219,7 +217,10 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 	if (ret)
 		goto free_all_kdata;
 
-	p->job->uf = uf;
+	if (p->uf_entry.robj) {
+		p->job->uf_bo = amdgpu_bo_ref(p->uf_entry.robj);
+		p->job->uf_offset = uf_offset;
+	}
 
 	kfree(chunk_array);
 	return 0;
@@ -377,7 +378,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 	INIT_LIST_HEAD(&duplicates);
 	amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);
 
-	if (p->job->uf.bo)
+	if (p->uf_entry.robj)
 		list_add(&p->uf_entry.tv.head, &p->validated);
 
 	if (need_mmap_lock)
@@ -760,17 +761,11 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 		j++;
 	}
 
-	/* wrap the last IB with user fence */
-	if (parser->job->uf.bo) {
-		struct amdgpu_ib *ib = &parser->job->ibs[parser->job->num_ibs - 1];
-
-		/* UVD & VCE fw doesn't support user fences */
-		if (parser->job->ring->type == AMDGPU_RING_TYPE_UVD ||
-		    parser->job->ring->type == AMDGPU_RING_TYPE_VCE)
-			return -EINVAL;
-
-		ib->user = &parser->job->uf;
-	}
+	/* UVD & VCE fw doesn't support user fences */
+	if (parser->job->uf_bo && (
+	    parser->job->ring->type == AMDGPU_RING_TYPE_UVD ||
+	    parser->job->ring->type == AMDGPU_RING_TYPE_VCE))
+		return -EINVAL;
 
 	return 0;
 }
@@ -856,7 +851,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	job->ctx = entity->fence_context;
 	p->fence = fence_get(fence);
 	cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, fence);
-	job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
+	job->uf_sequence = cs->out.handle;
 
 	trace_amdgpu_cs_ioctl(job);
 	amd_sched_entity_push_job(&job->base);