drm/amdgpu: move the context from the IBs into the job

We only have one context for all IBs.
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>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -743,7 +743,6 @@ struct amdgpu_ib {
 	struct amdgpu_user_fence        *user;
 	unsigned			vm_id;
 	uint64_t			vm_pd_addr;
-	uint64_t			ctx;
 	uint32_t			gds_base, gds_size;
 	uint32_t			gws_base, gws_size;
 	uint32_t			oa_base, oa_size;
@@ -1262,6 +1261,7 @@ struct amdgpu_job {
 	struct fence		*fence; /* the hw fence */
 	uint32_t		num_ibs;
 	void			*owner;
+	uint64_t		ctx;
 	struct amdgpu_user_fence uf;
 };
 #define to_amdgpu_job(sched_job)		\

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -741,7 +741,6 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 
 		ib->length_dw = chunk_ib->ib_bytes / 4;
 		ib->flags = chunk_ib->flags;
-		ib->ctx = parser->ctx->rings[ring->idx].entity.fence_context;
 		j++;
 	}
 
@@ -840,6 +839,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 			    union drm_amdgpu_cs *cs)
 {
 	struct amdgpu_ring *ring = p->job->ring;
+	struct amd_sched_entity *entity = &p->ctx->rings[ring->idx].entity;
 	struct fence *fence;
 	struct amdgpu_job *job;
 	int r;
@@ -848,16 +848,16 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	p->job = NULL;
 
 	r = amd_sched_job_init(&job->base, &ring->sched,
-						&p->ctx->rings[ring->idx].entity,
-						amdgpu_job_timeout_func,
-						amdgpu_job_free_func,
-						p->filp, &fence);
+			       entity, amdgpu_job_timeout_func,
+			       amdgpu_job_free_func,
+			       p->filp, &fence);
 	if (r) {
 		amdgpu_job_free(job);
 		return r;
 	}
 
 	job->owner = p->filp;
+	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;

drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c

@@ -121,18 +121,26 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_ib *ib = &ibs[0];
-	struct fence *hwf;
-	struct amdgpu_vm *vm = NULL;
-	unsigned i, patch_offset = ~0;
 	bool skip_preamble, need_ctx_switch;
+	unsigned patch_offset = ~0;
+	struct amdgpu_vm *vm;
+	struct fence *hwf;
+	uint64_t ctx;
 
+	unsigned i;
 	int r = 0;
 
 	if (num_ibs == 0)
 		return -EINVAL;
 
-	if (job) /* for domain0 job like ring test, ibs->job is not assigned */
+	/* ring tests don't use a job */
+	if (job) {
 		vm = job->vm;
+		ctx = job->ctx;
+	} else {
+		vm = NULL;
+		ctx = 0;
+	}
 
 	if (!ring->ready) {
 		dev_err(adev->dev, "couldn't schedule ib\n");
@@ -170,8 +178,8 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 	/* always set cond_exec_polling to CONTINUE */
 	*ring->cond_exe_cpu_addr = 1;
 
-	skip_preamble = ring->current_ctx == ib->ctx;
-	need_ctx_switch = ring->current_ctx != ib->ctx;
+	skip_preamble = ring->current_ctx == ctx;
+	need_ctx_switch = ring->current_ctx != ctx;
 	for (i = 0; i < num_ibs; ++i) {
 		ib = &ibs[i];
 
@@ -209,7 +217,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 	if (patch_offset != ~0 && ring->funcs->patch_cond_exec)
 		amdgpu_ring_patch_cond_exec(ring, patch_offset);
 
-	ring->current_ctx = ibs->ctx;
+	ring->current_ctx = ctx;
 	amdgpu_ring_commit(ring);
 	return 0;
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_job.c

@@ -122,14 +122,13 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
 		return -EINVAL;
 
 	r = amd_sched_job_init(&job->base, &ring->sched,
-							entity,
-							amdgpu_job_timeout_func,
-							amdgpu_job_free_func,
-							owner, &fence);
+			       entity, amdgpu_job_timeout_func,
+			       amdgpu_job_free_func, owner, &fence);
 	if (r)
 		return r;
 
 	job->owner = owner;
+	job->ctx = entity->fence_context;
 	*f = fence_get(fence);
 	amd_sched_entity_push_job(&job->base);