drm/amdgpu: move ring from IBs into job

We can't submit to multiple rings at the same time anyway.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -771,7 +771,6 @@ struct amdgpu_ib {
 	uint32_t			length_dw;
 	uint64_t			gpu_addr;
 	uint32_t			*ptr;
-	struct amdgpu_ring		*ring;
 	struct amdgpu_fence		*fence;
 	struct amdgpu_user_fence        *user;
 	bool				grabbed_vmid;
@@ -1178,10 +1177,10 @@ struct amdgpu_gfx {
 	unsigned ce_ram_size;
 };
 
-int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm,
+int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		  unsigned size, struct amdgpu_ib *ib);
 void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib);
-int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
+int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 		       struct amdgpu_ib *ib, void *owner);
 int amdgpu_ib_pool_init(struct amdgpu_device *adev);
 void amdgpu_ib_pool_fini(struct amdgpu_device *adev);
@@ -1239,6 +1238,7 @@ struct amdgpu_cs_parser {
 struct amdgpu_job {
 	struct amd_sched_job    base;
 	struct amdgpu_device	*adev;
+	struct amdgpu_ring	*ring;
 	struct amdgpu_ib	*ibs;
 	uint32_t		num_ibs;
 	void			*owner;

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

@@ -542,26 +542,25 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 }
 
 static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
-				 struct amdgpu_cs_parser *parser)
+				 struct amdgpu_cs_parser *p)
 {
-	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
+	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
 	struct amdgpu_vm *vm = &fpriv->vm;
-	struct amdgpu_ring *ring;
+	struct amdgpu_ring *ring = p->job->ring;
 	int i, r;
 
 	/* Only for UVD/VCE VM emulation */
-	for (i = 0; i < parser->job->num_ibs; i++) {
-		ring = parser->job->ibs[i].ring;
 	if (ring->funcs->parse_cs) {
-			r = amdgpu_ring_parse_cs(ring, parser, i);
+		for (i = 0; i < p->job->num_ibs; i++) {
+			r = amdgpu_ring_parse_cs(ring, p, i);
 			if (r)
 				return r;
 		}
 	}
 
-	r = amdgpu_bo_vm_update_pte(parser, vm);
+	r = amdgpu_bo_vm_update_pte(p, vm);
 	if (!r)
-		amdgpu_cs_sync_rings(parser);
+		amdgpu_cs_sync_rings(p);
 
 	return r;
 }
@@ -603,6 +602,11 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 		if (r)
 			return r;
 
+		if (parser->job->ring && parser->job->ring != ring)
+			return -EINVAL;
+
+		parser->job->ring = ring;
+
 		if (ring->funcs->parse_cs) {
 			struct amdgpu_bo_va_mapping *m;
 			struct amdgpu_bo *aobj = NULL;
@@ -631,7 +635,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 			offset = ((uint64_t)m->it.start) * AMDGPU_GPU_PAGE_SIZE;
 			kptr += chunk_ib->va_start - offset;
 
-			r =  amdgpu_ib_get(ring, NULL, chunk_ib->ib_bytes, ib);
+			r =  amdgpu_ib_get(adev, NULL, chunk_ib->ib_bytes, ib);
 			if (r) {
 				DRM_ERROR("Failed to get ib !\n");
 				return r;
@@ -640,7 +644,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 			memcpy(ib->ptr, kptr, chunk_ib->ib_bytes);
 			amdgpu_bo_kunmap(aobj);
 		} else {
-			r =  amdgpu_ib_get(ring, vm, 0, ib);
+			r =  amdgpu_ib_get(adev, vm, 0, ib);
 			if (r) {
 				DRM_ERROR("Failed to get ib !\n");
 				return r;
@@ -680,8 +684,8 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 		struct amdgpu_ib *ib = &parser->job->ibs[parser->job->num_ibs - 1];
 
 		/* UVD & VCE fw doesn't support user fences */
-		if (ib->ring->type == AMDGPU_RING_TYPE_UVD ||
-		    ib->ring->type == AMDGPU_RING_TYPE_VCE)
+		if (parser->job->ring->type == AMDGPU_RING_TYPE_UVD ||
+		    parser->job->ring->type == AMDGPU_RING_TYPE_VCE)
 			return -EINVAL;
 
 		ib->user = &parser->job->uf;
@@ -757,7 +761,7 @@ static int amdgpu_cs_free_job(struct amdgpu_job *job)
 static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 			    union drm_amdgpu_cs *cs)
 {
-	struct amdgpu_ring * ring = p->job->ibs->ring;
+	struct amdgpu_ring *ring = p->job->ring;
 	struct amd_sched_fence *fence;
 	struct amdgpu_job *job;
 
@@ -766,7 +770,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 
 	job->base.sched = &ring->sched;
 	job->base.s_entity = &p->ctx->rings[ring->idx].entity;
-	job->adev = p->adev;
 	job->owner = p->filp;
 	job->free_job = amdgpu_cs_free_job;
 

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

@@ -55,10 +55,9 @@ static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev);
  * suballocator.
  * Returns 0 on success, error on failure.
  */
-int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm,
+int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		  unsigned size, struct amdgpu_ib *ib)
 {
-	struct amdgpu_device *adev = ring->adev;
 	int r;
 
 	if (size) {
@@ -77,7 +76,6 @@ int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm,
 
 	amdgpu_sync_create(&ib->sync);
 
-	ib->ring = ring;
 	ib->vm = vm;
 
 	return 0;
@@ -120,11 +118,11 @@ void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib)
  * a CONST_IB), it will be put on the ring prior to the DE IB.  Prior
  * to SI there was just a DE IB.
  */
-int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
+int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 		       struct amdgpu_ib *ibs, void *owner)
 {
+	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_ib *ib = &ibs[0];
-	struct amdgpu_ring *ring;
 	struct amdgpu_ctx *ctx, *old_ctx;
 	struct amdgpu_vm *vm;
 	unsigned i;
@@ -133,7 +131,6 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 	if (num_ibs == 0)
 		return -EINVAL;
 
-	ring = ibs->ring;
 	ctx = ibs->ctx;
 	vm = ibs->vm;
 
@@ -178,7 +175,7 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 	for (i = 0; i < num_ibs; ++i) {
 		ib = &ibs[i];
 
-		if (ib->ring != ring || ib->ctx != ctx || ib->vm != vm) {
+		if (ib->ctx != ctx || ib->vm != vm) {
 			ring->current_ctx = old_ctx;
 			amdgpu_ring_undo(ring);
 			return -EINVAL;

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

@@ -70,7 +70,7 @@ static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job)
 	struct fence *fence = amdgpu_sync_get_fence(sync);
 
 	if (fence == NULL && vm && !job->ibs->grabbed_vmid) {
-		struct amdgpu_ring *ring = job->ibs->ring;
+		struct amdgpu_ring *ring = job->ring;
 		int r;
 
 		r = amdgpu_vm_grab_id(vm, ring, sync,
@@ -98,7 +98,7 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job)
 	}
 	job = to_amdgpu_job(sched_job);
 	trace_amdgpu_sched_run_job(job);
-	r = amdgpu_ib_schedule(job->adev, job->num_ibs, job->ibs, job->owner);
+	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job->owner);
 	if (r) {
 		DRM_ERROR("Error scheduling IBs (%d)\n", r);
 		goto err;
@@ -142,6 +142,7 @@ int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
 	*f = fence_get(&job->base.s_fence->base);
 
 	job->adev = adev;
+	job->ring = ring;
 	job->ibs = ibs;
 	job->num_ibs = num_ibs;
 	job->owner = owner;

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

@@ -38,10 +38,10 @@ TRACE_EVENT(amdgpu_cs,
 
 	    TP_fast_assign(
 			   __entry->bo_list = p->bo_list;
-			   __entry->ring = p->job->ibs[i].ring->idx;
+			   __entry->ring = p->job->ring->idx;
 			   __entry->dw = p->job->ibs[i].length_dw;
 			   __entry->fences = amdgpu_fence_count_emitted(
-				p->job->ibs[i].ring);
+				p->job->ring);
 			   ),
 	    TP_printk("bo_list=%p, ring=%u, dw=%u, fences=%u",
 		      __entry->bo_list, __entry->ring, __entry->dw,
@@ -65,7 +65,7 @@ TRACE_EVENT(amdgpu_cs_ioctl,
 			   __entry->sched_job = &job->base;
 			   __entry->ib = job->ibs;
 			   __entry->fence = &job->base.s_fence->base;
-			   __entry->ring_name = job->ibs[0].ring->name;
+			   __entry->ring_name = job->ring->name;
 			   __entry->num_ibs = job->num_ibs;
 			   ),
 	    TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u",
@@ -90,7 +90,7 @@ TRACE_EVENT(amdgpu_sched_run_job,
 			   __entry->sched_job = &job->base;
 			   __entry->ib = job->ibs;
 			   __entry->fence = &job->base.s_fence->base;
-			   __entry->ring_name = job->ibs[0].ring->name;
+			   __entry->ring_name = job->ring->name;
 			   __entry->num_ibs = job->num_ibs;
 			   ),
 	    TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u",

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

@@ -1030,7 +1030,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 	if (!ib)
 		return -ENOMEM;
 
-	r = amdgpu_ib_get(ring, NULL, num_dw * 4, ib);
+	r = amdgpu_ib_get(adev, NULL, num_dw * 4, ib);
 	if (r) {
 		kfree(ib);
 		return r;

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

@@ -867,7 +867,7 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 		r = -ENOMEM;
 		goto err;
 	}
-	r = amdgpu_ib_get(ring, NULL, 64, ib);
+	r = amdgpu_ib_get(adev, NULL, 64, ib);
 	if (r)
 		goto err1;
 

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

@@ -377,7 +377,7 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
 	if (!ib)
 		return -ENOMEM;
-	r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib);
+	r = amdgpu_ib_get(adev, NULL, ib_size_dw * 4, ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 		kfree(ib);
@@ -463,7 +463,7 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 	if (!ib)
 		return -ENOMEM;
 
-	r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib);
+	r = amdgpu_ib_get(adev, NULL, ib_size_dw * 4, ib);
 	if (r) {
 		kfree(ib);
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);

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

@@ -355,7 +355,7 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 	if (!ib)
 		goto error;
 
-	r = amdgpu_ib_get(ring, NULL, 64, ib);
+	r = amdgpu_ib_get(adev, NULL, 64, ib);
 	if (r)
 		goto error_free;
 
@@ -448,7 +448,7 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 	if (!ib)
 		return -ENOMEM;
 
-	r = amdgpu_ib_get(ring, NULL, ndw * 4, ib);
+	r = amdgpu_ib_get(adev, NULL, ndw * 4, ib);
 	if (r) {
 		kfree(ib);
 		return r;
@@ -737,7 +737,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	if (!ib)
 		return -ENOMEM;
 
-	r = amdgpu_ib_get(ring, NULL, ndw * 4, ib);
+	r = amdgpu_ib_get(adev, NULL, ndw * 4, ib);
 	if (r) {
 		kfree(ib);
 		return r;

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

@@ -621,7 +621,7 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
 	tmp = 0xCAFEDEAD;
 	adev->wb.wb[index] = cpu_to_le32(tmp);
 	memset(&ib, 0, sizeof(ib));
-	r = amdgpu_ib_get(ring, NULL, 256, &ib);
+	r = amdgpu_ib_get(adev, NULL, 256, &ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 		goto err0;

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -2631,7 +2631,7 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring)
 	}
 	WREG32(scratch, 0xCAFEDEAD);
 	memset(&ib, 0, sizeof(ib));
-	r = amdgpu_ib_get(ring, NULL, 256, &ib);
+	r = amdgpu_ib_get(adev, NULL, 256, &ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 		goto err1;

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -699,7 +699,7 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring)
 	}
 	WREG32(scratch, 0xCAFEDEAD);
 	memset(&ib, 0, sizeof(ib));
-	r = amdgpu_ib_get(ring, NULL, 256, &ib);
+	r = amdgpu_ib_get(adev, NULL, 256, &ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 		goto err1;
@@ -1171,7 +1171,7 @@ static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
 
 	/* allocate an indirect buffer to put the commands in */
 	memset(&ib, 0, sizeof(ib));
-	r = amdgpu_ib_get(ring, NULL, total_size, &ib);
+	r = amdgpu_ib_get(adev, NULL, total_size, &ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 		return r;

drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c

@@ -674,7 +674,7 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
 	tmp = 0xCAFEDEAD;
 	adev->wb.wb[index] = cpu_to_le32(tmp);
 	memset(&ib, 0, sizeof(ib));
-	r = amdgpu_ib_get(ring, NULL, 256, &ib);
+	r = amdgpu_ib_get(adev, NULL, 256, &ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 		goto err0;

drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c

@@ -825,7 +825,7 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
 	tmp = 0xCAFEDEAD;
 	adev->wb.wb[index] = cpu_to_le32(tmp);
 	memset(&ib, 0, sizeof(ib));
-	r = amdgpu_ib_get(ring, NULL, 256, &ib);
+	r = amdgpu_ib_get(adev, NULL, 256, &ib);
 	if (r) {
 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
 		goto err0;