drm/amdgpu: use entity instead of ring for CS

Further demangle ring from entity handling.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -523,7 +523,7 @@ struct amdgpu_cs_parser {
 
 	/* scheduler job object */
 	struct amdgpu_job	*job;
-	struct amdgpu_ring	*ring;
+	struct drm_sched_entity	*entity;
 
 	/* buffer objects */
 	struct ww_acquire_ctx		ticket;

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

@@ -893,13 +893,13 @@ 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 *p)
 {
+	struct amdgpu_ring *ring = to_amdgpu_ring(p->entity->rq->sched);
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
 	struct amdgpu_vm *vm = &fpriv->vm;
-	struct amdgpu_ring *ring = p->ring;
 	int r;
 
 	/* Only for UVD/VCE VM emulation */
-	if (p->ring->funcs->parse_cs || p->ring->funcs->patch_cs_in_place) {
+	if (ring->funcs->parse_cs || ring->funcs->patch_cs_in_place) {
 		unsigned i, j;
 
 		for (i = 0, j = 0; i < p->nchunks && j < p->job->num_ibs; i++) {
@@ -940,7 +940,7 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
 			offset = m->start * AMDGPU_GPU_PAGE_SIZE;
 			kptr += va_start - offset;
 
-			if (p->ring->funcs->parse_cs) {
+			if (ring->funcs->parse_cs) {
 				memcpy(ib->ptr, kptr, chunk_ib->ib_bytes);
 				amdgpu_bo_kunmap(aobj);
 
@@ -979,14 +979,15 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 {
 	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
 	struct amdgpu_vm *vm = &fpriv->vm;
-	int i, j;
 	int r, ce_preempt = 0, de_preempt = 0;
+	struct amdgpu_ring *ring;
+	int i, j;
 
 	for (i = 0, j = 0; i < parser->nchunks && j < parser->job->num_ibs; i++) {
 		struct amdgpu_cs_chunk *chunk;
 		struct amdgpu_ib *ib;
 		struct drm_amdgpu_cs_chunk_ib *chunk_ib;
-		struct amdgpu_ring *ring;
+		struct drm_sched_entity *entity;
 
 		chunk = &parser->chunks[i];
 		ib = &parser->job->ibs[j];
@@ -1008,9 +1009,9 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 				return -EINVAL;
 		}
 
-		r = amdgpu_ctx_get_ring(parser->ctx, chunk_ib->ip_type,
+		r = amdgpu_ctx_get_entity(parser->ctx, chunk_ib->ip_type,
 					  chunk_ib->ip_instance, chunk_ib->ring,
-					&ring);
+					  &entity);
 		if (r)
 			return r;
 
@@ -1018,14 +1019,14 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 			parser->job->preamble_status |=
 				AMDGPU_PREAMBLE_IB_PRESENT;
 
-		if (parser->ring && parser->ring != ring)
+		if (parser->entity && parser->entity != entity)
 			return -EINVAL;
 
-		parser->ring = ring;
+		parser->entity = entity;
 
-		r =  amdgpu_ib_get(adev, vm,
-					ring->funcs->parse_cs ? chunk_ib->ib_bytes : 0,
-					ib);
+		ring = to_amdgpu_ring(entity->rq->sched);
+		r =  amdgpu_ib_get(adev, vm, ring->funcs->parse_cs ?
+				   chunk_ib->ib_bytes : 0, ib);
 		if (r) {
 			DRM_ERROR("Failed to get ib !\n");
 			return r;
@@ -1039,12 +1040,13 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 	}
 
 	/* UVD & VCE fw doesn't support user fences */
+	ring = to_amdgpu_ring(parser->entity->rq->sched);
 	if (parser->job->uf_addr && (
-	    parser->ring->funcs->type == AMDGPU_RING_TYPE_UVD ||
-	    parser->ring->funcs->type == AMDGPU_RING_TYPE_VCE))
+	    ring->funcs->type == AMDGPU_RING_TYPE_UVD ||
+	    ring->funcs->type == AMDGPU_RING_TYPE_VCE))
 		return -EINVAL;
 
-	return amdgpu_ctx_wait_prev_fence(parser->ctx, parser->ring->idx);
+	return amdgpu_ctx_wait_prev_fence(parser->ctx, parser->entity);
 }
 
 static int amdgpu_cs_process_fence_dep(struct amdgpu_cs_parser *p,
@@ -1060,23 +1062,23 @@ static int amdgpu_cs_process_fence_dep(struct amdgpu_cs_parser *p,
 		sizeof(struct drm_amdgpu_cs_chunk_dep);
 
 	for (i = 0; i < num_deps; ++i) {
-		struct amdgpu_ring *ring;
 		struct amdgpu_ctx *ctx;
+		struct drm_sched_entity *entity;
 		struct dma_fence *fence;
 
 		ctx = amdgpu_ctx_get(fpriv, deps[i].ctx_id);
 		if (ctx == NULL)
 			return -EINVAL;
 
-		r = amdgpu_ctx_get_ring(ctx, deps[i].ip_type,
+		r = amdgpu_ctx_get_entity(ctx, deps[i].ip_type,
 					  deps[i].ip_instance,
-					deps[i].ring, &ring);
+					  deps[i].ring, &entity);
 		if (r) {
 			amdgpu_ctx_put(ctx);
 			return r;
 		}
 
-		fence = amdgpu_ctx_get_fence(ctx, ring,
+		fence = amdgpu_ctx_get_fence(ctx, entity,
 					     deps[i].handle);
 		if (IS_ERR(fence)) {
 			r = PTR_ERR(fence);
@@ -1195,9 +1197,9 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 			    union drm_amdgpu_cs *cs)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	struct amdgpu_ring *ring = p->ring;
-	struct drm_sched_entity *entity = &p->ctx->rings[ring->idx].entity;
+	struct drm_sched_entity *entity = p->entity;
 	enum drm_sched_priority priority;
+	struct amdgpu_ring *ring;
 	struct amdgpu_bo_list_entry *e;
 	struct amdgpu_job *job;
 	uint64_t seq;
@@ -1227,7 +1229,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	job->owner = p->filp;
 	p->fence = dma_fence_get(&job->base.s_fence->finished);
 
-	r = amdgpu_ctx_add_fence(p->ctx, ring, p->fence, &seq);
+	r = amdgpu_ctx_add_fence(p->ctx, entity, p->fence, &seq);
 	if (r) {
 		dma_fence_put(p->fence);
 		dma_fence_put(&job->base.s_fence->finished);
@@ -1332,7 +1334,7 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
 {
 	union drm_amdgpu_wait_cs *wait = data;
 	unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
-	struct amdgpu_ring *ring = NULL;
+	struct drm_sched_entity *entity;
 	struct amdgpu_ctx *ctx;
 	struct dma_fence *fence;
 	long r;
@@ -1341,14 +1343,14 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
 	if (ctx == NULL)
 		return -EINVAL;
 
-	r = amdgpu_ctx_get_ring(ctx, wait->in.ip_type, wait->in.ip_instance,
-				wait->in.ring, &ring);
+	r = amdgpu_ctx_get_entity(ctx, wait->in.ip_type, wait->in.ip_instance,
+				  wait->in.ring, &entity);
 	if (r) {
 		amdgpu_ctx_put(ctx);
 		return r;
 	}
 
-	fence = amdgpu_ctx_get_fence(ctx, ring, wait->in.handle);
+	fence = amdgpu_ctx_get_fence(ctx, entity, wait->in.handle);
 	if (IS_ERR(fence))
 		r = PTR_ERR(fence);
 	else if (fence) {
@@ -1380,7 +1382,7 @@ static struct dma_fence *amdgpu_cs_get_fence(struct amdgpu_device *adev,
 					     struct drm_file *filp,
 					     struct drm_amdgpu_fence *user)
 {
-	struct amdgpu_ring *ring;
+	struct drm_sched_entity *entity;
 	struct amdgpu_ctx *ctx;
 	struct dma_fence *fence;
 	int r;
@@ -1389,14 +1391,14 @@ static struct dma_fence *amdgpu_cs_get_fence(struct amdgpu_device *adev,
 	if (ctx == NULL)
 		return ERR_PTR(-EINVAL);
 
-	r = amdgpu_ctx_get_ring(ctx, user->ip_type, user->ip_instance,
-				user->ring, &ring);
+	r = amdgpu_ctx_get_entity(ctx, user->ip_type, user->ip_instance,
+				  user->ring, &entity);
 	if (r) {
 		amdgpu_ctx_put(ctx);
 		return ERR_PTR(r);
 	}
 
-	fence = amdgpu_ctx_get_fence(ctx, ring, user->seq_no);
+	fence = amdgpu_ctx_get_fence(ctx, entity, user->seq_no);
 	amdgpu_ctx_put(ctx);
 
 	return fence;

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

@@ -27,6 +27,9 @@
 #include "amdgpu.h"
 #include "amdgpu_sched.h"
 
+#define to_amdgpu_ctx_ring(e)	\
+	container_of((e), struct amdgpu_ctx_ring, entity)
+
 static int amdgpu_ctx_priority_permit(struct drm_file *filp,
 				      enum drm_sched_priority priority)
 {
@@ -151,12 +154,12 @@ static void amdgpu_ctx_fini(struct kref *ref)
 	kfree(ctx);
 }
 
-int amdgpu_ctx_get_ring(struct amdgpu_ctx *ctx,
-			u32 hw_ip, u32 instance, u32 ring,
-			struct amdgpu_ring **out_ring)
+int amdgpu_ctx_get_entity(struct amdgpu_ctx *ctx, u32 hw_ip, u32 instance,
+			  u32 ring, struct drm_sched_entity **entity)
 {
 	struct amdgpu_device *adev = ctx->adev;
 	unsigned num_rings = 0;
+	struct amdgpu_ring *out_ring;
 
 	/* Right now all IPs have only one instance - multiple rings. */
 	if (instance != 0) {
@@ -166,39 +169,39 @@ int amdgpu_ctx_get_ring(struct amdgpu_ctx *ctx,
 
 	switch (hw_ip) {
 	case AMDGPU_HW_IP_GFX:
-		*out_ring = &adev->gfx.gfx_ring[ring];
+		out_ring = &adev->gfx.gfx_ring[ring];
 		num_rings = adev->gfx.num_gfx_rings;
 		break;
 	case AMDGPU_HW_IP_COMPUTE:
-		*out_ring = &adev->gfx.compute_ring[ring];
+		out_ring = &adev->gfx.compute_ring[ring];
 		num_rings = adev->gfx.num_compute_rings;
 		break;
 	case AMDGPU_HW_IP_DMA:
-		*out_ring = &adev->sdma.instance[ring].ring;
+		out_ring = &adev->sdma.instance[ring].ring;
 		num_rings = adev->sdma.num_instances;
 		break;
 	case AMDGPU_HW_IP_UVD:
-		*out_ring = &adev->uvd.inst[0].ring;
+		out_ring = &adev->uvd.inst[0].ring;
 		num_rings = adev->uvd.num_uvd_inst;
 		break;
 	case AMDGPU_HW_IP_VCE:
-		*out_ring = &adev->vce.ring[ring];
+		out_ring = &adev->vce.ring[ring];
 		num_rings = adev->vce.num_rings;
 		break;
 	case AMDGPU_HW_IP_UVD_ENC:
-		*out_ring = &adev->uvd.inst[0].ring_enc[ring];
+		out_ring = &adev->uvd.inst[0].ring_enc[ring];
 		num_rings = adev->uvd.num_enc_rings;
 		break;
 	case AMDGPU_HW_IP_VCN_DEC:
-		*out_ring = &adev->vcn.ring_dec;
+		out_ring = &adev->vcn.ring_dec;
 		num_rings = 1;
 		break;
 	case AMDGPU_HW_IP_VCN_ENC:
-		*out_ring = &adev->vcn.ring_enc[ring];
+		out_ring = &adev->vcn.ring_enc[ring];
 		num_rings = adev->vcn.num_enc_rings;
 		break;
 	case AMDGPU_HW_IP_VCN_JPEG:
-		*out_ring = &adev->vcn.ring_jpeg;
+		out_ring = &adev->vcn.ring_jpeg;
 		num_rings = 1;
 		break;
 	default:
@@ -209,6 +212,7 @@ int amdgpu_ctx_get_ring(struct amdgpu_ctx *ctx,
 	if (ring > num_rings)
 		return -EINVAL;
 
+	*entity = &ctx->rings[out_ring->idx].entity;
 	return 0;
 }
 
@@ -414,13 +418,14 @@ int amdgpu_ctx_put(struct amdgpu_ctx *ctx)
 	return 0;
 }
 
-int amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
-			      struct dma_fence *fence, uint64_t* handler)
+int amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx,
+			 struct drm_sched_entity *entity,
+			 struct dma_fence *fence, uint64_t* handle)
 {
-	struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
+	struct amdgpu_ctx_ring *cring = to_amdgpu_ctx_ring(entity);
 	uint64_t seq = cring->sequence;
-	unsigned idx = 0;
 	struct dma_fence *other = NULL;
+	unsigned idx = 0;
 
 	idx = seq & (amdgpu_sched_jobs - 1);
 	other = cring->fences[idx];
@@ -435,22 +440,23 @@ int amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
 	spin_unlock(&ctx->ring_lock);
 
 	dma_fence_put(other);
-	if (handler)
-		*handler = seq;
+	if (handle)
+		*handle = seq;
 
 	return 0;
 }
 
 struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
-				       struct amdgpu_ring *ring, uint64_t seq)
+				       struct drm_sched_entity *entity,
+				       uint64_t seq)
 {
-	struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
+	struct amdgpu_ctx_ring *cring = to_amdgpu_ctx_ring(entity);
 	struct dma_fence *fence;
 
 	spin_lock(&ctx->ring_lock);
 
 	if (seq == ~0ull)
-		seq = ctx->rings[ring->idx].sequence - 1;
+		seq = cring->sequence - 1;
 
 	if (seq >= cring->sequence) {
 		spin_unlock(&ctx->ring_lock);
@@ -494,9 +500,10 @@ void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
 	}
 }
 
-int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx, unsigned ring_id)
+int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx,
+			       struct drm_sched_entity *entity)
 {
-	struct amdgpu_ctx_ring *cring = &ctx->rings[ring_id];
+	struct amdgpu_ctx_ring *cring = to_amdgpu_ctx_ring(entity);
 	unsigned idx = cring->sequence & (amdgpu_sched_jobs - 1);
 	struct dma_fence *other = cring->fences[idx];
 

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

@@ -61,20 +61,22 @@ struct amdgpu_ctx_mgr {
 struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
 int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
 
-int amdgpu_ctx_get_ring(struct amdgpu_ctx *ctx,
-			u32 hw_ip, u32 instance, u32 ring,
-			struct amdgpu_ring **out_ring);
-int amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
+int amdgpu_ctx_get_entity(struct amdgpu_ctx *ctx, u32 hw_ip, u32 instance,
+			  u32 ring, struct drm_sched_entity **entity);
+int amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx,
+			 struct drm_sched_entity *entity,
 			 struct dma_fence *fence, uint64_t *seq);
 struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
-				   struct amdgpu_ring *ring, uint64_t seq);
+				       struct drm_sched_entity *entity,
+				       uint64_t seq);
 void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
 				  enum drm_sched_priority priority);
 
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp);
 
-int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx, unsigned ring_id);
+int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx,
+			       struct drm_sched_entity *entity);
 
 void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr);
 void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr);

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

@@ -150,10 +150,10 @@ TRACE_EVENT(amdgpu_cs,
 
 	    TP_fast_assign(
 			   __entry->bo_list = p->bo_list;
-			   __entry->ring = p->ring->idx;
+			   __entry->ring = to_amdgpu_ring(p->entity->rq->sched)->idx;
 			   __entry->dw = p->job->ibs[i].length_dw;
 			   __entry->fences = amdgpu_fence_count_emitted(
-				p->ring);
+				to_amdgpu_ring(p->entity->rq->sched));
 			   ),
 	    TP_printk("bo_list=%p, ring=%u, dw=%u, fences=%u",
 		      __entry->bo_list, __entry->ring, __entry->dw,

drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c

@@ -1264,11 +1264,12 @@ static int uvd_v7_0_ring_test_ring(struct amdgpu_ring *ring)
 static int uvd_v7_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
 					   uint32_t ib_idx)
 {
+	struct amdgpu_ring *ring = to_amdgpu_ring(p->entity->rq->sched);
 	struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
 	unsigned i;
 
 	/* No patching necessary for the first instance */
-	if (!p->ring->me)
+	if (!ring->me)
 		return 0;
 
 	for (i = 0; i < ib->length_dw; i += 2) {