Merge branch 'drm-next-4.13' of git://people.freedesktop.org/~agd5f/linux into drm-next

A few more things for 4.13:
- Semaphore support using sync objects
- Drop fb location programming
- Optimize bo list ioctl

* 'drm-next-4.13' of git://people.freedesktop.org/~agd5f/linux:
  drm/amdgpu: Optimize mutex usage (v4)
  drm/amdgpu: Optimization of AMDGPU_BO_LIST_OP_CREATE (v2)
  amdgpu: use drm sync objects for shared semaphores (v6)
  amdgpu/cs: split out fence dependency checking (v2)
  drm/amdgpu: don't check the default value for vm size

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

@@ -869,6 +869,8 @@ struct amdgpu_fpriv {
 
 struct amdgpu_bo_list {
 	struct mutex lock;
+	struct rcu_head rhead;
+	struct kref refcount;
 	struct amdgpu_bo *gds_obj;
 	struct amdgpu_bo *gws_obj;
 	struct amdgpu_bo *oa_obj;
@@ -1159,6 +1161,9 @@ struct amdgpu_cs_parser {
 
 	/* user fence */
 	struct amdgpu_bo_list_entry	uf_entry;
+
+	unsigned num_post_dep_syncobjs;
+	struct drm_syncobj **post_dep_syncobjs;
 };
 
 #define AMDGPU_PREAMBLE_IB_PRESENT          (1 << 0) /* bit set means command submit involves a preamble IB */

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

@@ -35,33 +35,59 @@
 #define AMDGPU_BO_LIST_MAX_PRIORITY	32u
 #define AMDGPU_BO_LIST_NUM_BUCKETS	(AMDGPU_BO_LIST_MAX_PRIORITY + 1)
 
-static int amdgpu_bo_list_create(struct amdgpu_fpriv *fpriv,
+static int amdgpu_bo_list_set(struct amdgpu_device *adev,
-				 struct amdgpu_bo_list **result,
+				     struct drm_file *filp,
+				     struct amdgpu_bo_list *list,
+				     struct drm_amdgpu_bo_list_entry *info,
+				     unsigned num_entries);
+
+static void amdgpu_bo_list_release_rcu(struct kref *ref)
+{
+	unsigned i;
+	struct amdgpu_bo_list *list = container_of(ref, struct amdgpu_bo_list,
+						   refcount);
+
+	for (i = 0; i < list->num_entries; ++i)
+		amdgpu_bo_unref(&list->array[i].robj);
+
+	mutex_destroy(&list->lock);
+	kvfree(list->array);
+	kfree_rcu(list, rhead);
+}
+
+static int amdgpu_bo_list_create(struct amdgpu_device *adev,
+				 struct drm_file *filp,
+				 struct drm_amdgpu_bo_list_entry *info,
+				 unsigned num_entries,
 				 int *id)
 {
 	int r;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
+	struct amdgpu_bo_list *list;
 
-	*result = kzalloc(sizeof(struct amdgpu_bo_list), GFP_KERNEL);
+	list = kzalloc(sizeof(struct amdgpu_bo_list), GFP_KERNEL);
-	if (!*result)
+	if (!list)
 		return -ENOMEM;
 
+	/* initialize bo list*/
+	mutex_init(&list->lock);
+	kref_init(&list->refcount);
+	r = amdgpu_bo_list_set(adev, filp, list, info, num_entries);
+	if (r) {
+		kfree(list);
+		return r;
+	}
+
+	/* idr alloc should be called only after initialization of bo list. */
 	mutex_lock(&fpriv->bo_list_lock);
-	r = idr_alloc(&fpriv->bo_list_handles, *result,
+	r = idr_alloc(&fpriv->bo_list_handles, list, 1, 0, GFP_KERNEL);
-		      1, 0, GFP_KERNEL);
+	mutex_unlock(&fpriv->bo_list_lock);
 	if (r < 0) {
-		mutex_unlock(&fpriv->bo_list_lock);
+		kfree(list);
-		kfree(*result);
 		return r;
 	}
 	*id = r;
 
-	mutex_init(&(*result)->lock);
-	(*result)->num_entries = 0;
-	(*result)->array = NULL;
-
-	mutex_lock(&(*result)->lock);
-	mutex_unlock(&fpriv->bo_list_lock);
-
 	return 0;
 }
 
@@ -71,13 +97,9 @@ static void amdgpu_bo_list_destroy(struct amdgpu_fpriv *fpriv, int id)
 
 	mutex_lock(&fpriv->bo_list_lock);
 	list = idr_remove(&fpriv->bo_list_handles, id);
-	if (list) {
-		/* Another user may have a reference to this list still */
-		mutex_lock(&list->lock);
-		mutex_unlock(&list->lock);
-		amdgpu_bo_list_free(list);
-	}
 	mutex_unlock(&fpriv->bo_list_lock);
+	if (list)
+		kref_put(&list->refcount, amdgpu_bo_list_release_rcu);
 }
 
 static int amdgpu_bo_list_set(struct amdgpu_device *adev,
@@ -172,11 +194,17 @@ amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id)
 {
 	struct amdgpu_bo_list *result;
 
-	mutex_lock(&fpriv->bo_list_lock);
+	rcu_read_lock();
 	result = idr_find(&fpriv->bo_list_handles, id);
-	if (result)
+
-		mutex_lock(&result->lock);
+	if (result) {
-	mutex_unlock(&fpriv->bo_list_lock);
+		if (kref_get_unless_zero(&result->refcount))
+			mutex_lock(&result->lock);
+		else
+			result = NULL;
+	}
+	rcu_read_unlock();
+
 	return result;
 }
 
@@ -214,6 +242,7 @@ void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
 void amdgpu_bo_list_put(struct amdgpu_bo_list *list)
 {
 	mutex_unlock(&list->lock);
+	kref_put(&list->refcount, amdgpu_bo_list_release_rcu);
 }
 
 void amdgpu_bo_list_free(struct amdgpu_bo_list *list)
@@ -273,16 +302,10 @@ int amdgpu_bo_list_ioctl(struct drm_device *dev, void *data,
 
 	switch (args->in.operation) {
 	case AMDGPU_BO_LIST_OP_CREATE:
-		r = amdgpu_bo_list_create(fpriv, &list, &handle);
+		r = amdgpu_bo_list_create(adev, filp, info, args->in.bo_number,
+					  &handle);
 		if (r)
 			goto error_free;
-
-		r = amdgpu_bo_list_set(adev, filp, list, info,
-					      args->in.bo_number);
-		amdgpu_bo_list_put(list);
-		if (r)
-			goto error_free;
-
 		break;
 
 	case AMDGPU_BO_LIST_OP_DESTROY:

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

@@ -27,6 +27,7 @@
 #include <linux/pagemap.h>
 #include <drm/drmP.h>
 #include <drm/amdgpu_drm.h>
+#include <drm/drm_syncobj.h>
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
 
@@ -154,6 +155,8 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 			break;
 
 		case AMDGPU_CHUNK_ID_DEPENDENCIES:
+		case AMDGPU_CHUNK_ID_SYNCOBJ_IN:
+		case AMDGPU_CHUNK_ID_SYNCOBJ_OUT:
 			break;
 
 		default:
@@ -682,6 +685,11 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
 		ttm_eu_backoff_reservation(&parser->ticket,
 					   &parser->validated);
 	}
+
+	for (i = 0; i < parser->num_post_dep_syncobjs; i++)
+		drm_syncobj_put(parser->post_dep_syncobjs[i]);
+	kfree(parser->post_dep_syncobjs);
+
 	dma_fence_put(parser->fence);
 
 	if (parser->ctx)
@@ -923,65 +931,150 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 	return 0;
 }
 
+static int amdgpu_cs_process_fence_dep(struct amdgpu_cs_parser *p,
+				       struct amdgpu_cs_chunk *chunk)
+{
+	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	unsigned num_deps;
+	int i, r;
+	struct drm_amdgpu_cs_chunk_dep *deps;
+
+	deps = (struct drm_amdgpu_cs_chunk_dep *)chunk->kdata;
+	num_deps = chunk->length_dw * 4 /
+		sizeof(struct drm_amdgpu_cs_chunk_dep);
+
+	for (i = 0; i < num_deps; ++i) {
+		struct amdgpu_ring *ring;
+		struct amdgpu_ctx *ctx;
+		struct dma_fence *fence;
+
+		ctx = amdgpu_ctx_get(fpriv, deps[i].ctx_id);
+		if (ctx == NULL)
+			return -EINVAL;
+
+		r = amdgpu_queue_mgr_map(p->adev, &ctx->queue_mgr,
+					 deps[i].ip_type,
+					 deps[i].ip_instance,
+					 deps[i].ring, &ring);
+		if (r) {
+			amdgpu_ctx_put(ctx);
+			return r;
+		}
+
+		fence = amdgpu_ctx_get_fence(ctx, ring,
+					     deps[i].handle);
+		if (IS_ERR(fence)) {
+			r = PTR_ERR(fence);
+			amdgpu_ctx_put(ctx);
+			return r;
+		} else if (fence) {
+			r = amdgpu_sync_fence(p->adev, &p->job->sync,
+					      fence);
+			dma_fence_put(fence);
+			amdgpu_ctx_put(ctx);
+			if (r)
+				return r;
+		}
+	}
+	return 0;
+}
+
+static int amdgpu_syncobj_lookup_and_add_to_sync(struct amdgpu_cs_parser *p,
+						 uint32_t handle)
+{
+	int r;
+	struct dma_fence *fence;
+	r = drm_syncobj_fence_get(p->filp, handle, &fence);
+	if (r)
+		return r;
+
+	r = amdgpu_sync_fence(p->adev, &p->job->sync, fence);
+	dma_fence_put(fence);
+
+	return r;
+}
+
+static int amdgpu_cs_process_syncobj_in_dep(struct amdgpu_cs_parser *p,
+					    struct amdgpu_cs_chunk *chunk)
+{
+	unsigned num_deps;
+	int i, r;
+	struct drm_amdgpu_cs_chunk_sem *deps;
+
+	deps = (struct drm_amdgpu_cs_chunk_sem *)chunk->kdata;
+	num_deps = chunk->length_dw * 4 /
+		sizeof(struct drm_amdgpu_cs_chunk_sem);
+
+	for (i = 0; i < num_deps; ++i) {
+		r = amdgpu_syncobj_lookup_and_add_to_sync(p, deps[i].handle);
+		if (r)
+			return r;
+	}
+	return 0;
+}
+
+static int amdgpu_cs_process_syncobj_out_dep(struct amdgpu_cs_parser *p,
+					     struct amdgpu_cs_chunk *chunk)
+{
+	unsigned num_deps;
+	int i;
+	struct drm_amdgpu_cs_chunk_sem *deps;
+	deps = (struct drm_amdgpu_cs_chunk_sem *)chunk->kdata;
+	num_deps = chunk->length_dw * 4 /
+		sizeof(struct drm_amdgpu_cs_chunk_sem);
+
+	p->post_dep_syncobjs = kmalloc_array(num_deps,
+					     sizeof(struct drm_syncobj *),
+					     GFP_KERNEL);
+	p->num_post_dep_syncobjs = 0;
+
+	for (i = 0; i < num_deps; ++i) {
+		p->post_dep_syncobjs[i] = drm_syncobj_find(p->filp, deps[i].handle);
+		if (!p->post_dep_syncobjs[i])
+			return -EINVAL;
+		p->num_post_dep_syncobjs++;
+	}
+	return 0;
+}
+
 static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 				  struct amdgpu_cs_parser *p)
 {
-	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	int i, r;
-	int i, j, r;
 
 	for (i = 0; i < p->nchunks; ++i) {
-		struct drm_amdgpu_cs_chunk_dep *deps;
 		struct amdgpu_cs_chunk *chunk;
-		unsigned num_deps;
 
 		chunk = &p->chunks[i];
 
-		if (chunk->chunk_id != AMDGPU_CHUNK_ID_DEPENDENCIES)
+		if (chunk->chunk_id == AMDGPU_CHUNK_ID_DEPENDENCIES) {
-			continue;
+			r = amdgpu_cs_process_fence_dep(p, chunk);
-
+			if (r)
-		deps = (struct drm_amdgpu_cs_chunk_dep *)chunk->kdata;
-		num_deps = chunk->length_dw * 4 /
-			sizeof(struct drm_amdgpu_cs_chunk_dep);
-
-		for (j = 0; j < num_deps; ++j) {
-			struct amdgpu_ring *ring;
-			struct amdgpu_ctx *ctx;
-			struct dma_fence *fence;
-
-			ctx = amdgpu_ctx_get(fpriv, deps[j].ctx_id);
-			if (ctx == NULL)
-				return -EINVAL;
-
-			r = amdgpu_queue_mgr_map(adev, &ctx->queue_mgr,
-						 deps[j].ip_type,
-						 deps[j].ip_instance,
-						 deps[j].ring, &ring);
-			if (r) {
-				amdgpu_ctx_put(ctx);
 				return r;
-			}
+		} else if (chunk->chunk_id == AMDGPU_CHUNK_ID_SYNCOBJ_IN) {
-
+			r = amdgpu_cs_process_syncobj_in_dep(p, chunk);
-			fence = amdgpu_ctx_get_fence(ctx, ring,
+			if (r)
-						     deps[j].handle);
+				return r;
-			if (IS_ERR(fence)) {
+		} else if (chunk->chunk_id == AMDGPU_CHUNK_ID_SYNCOBJ_OUT) {
-				r = PTR_ERR(fence);
+			r = amdgpu_cs_process_syncobj_out_dep(p, chunk);
-				amdgpu_ctx_put(ctx);
+			if (r)
 				return r;
-
-			} else if (fence) {
-				r = amdgpu_sync_fence(adev, &p->job->sync,
-						      fence);
-				dma_fence_put(fence);
-				amdgpu_ctx_put(ctx);
-				if (r)
-					return r;
-			}
 		}
 	}
 
 	return 0;
 }
 
+static void amdgpu_cs_post_dependencies(struct amdgpu_cs_parser *p)
+{
+	int i;
+
+	for (i = 0; i < p->num_post_dep_syncobjs; ++i) {
+		drm_syncobj_replace_fence(p->filp, p->post_dep_syncobjs[i],
+					  p->fence);
+	}
+}
+
 static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 			    union drm_amdgpu_cs *cs)
 {
@@ -1002,6 +1095,9 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	job->owner = p->filp;
 	job->fence_ctx = entity->fence_context;
 	p->fence = dma_fence_get(&job->base.s_fence->finished);
+
+	amdgpu_cs_post_dependencies(p);
+
 	cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, p->fence);
 	job->uf_sequence = cs->out.handle;
 	amdgpu_job_free_resources(job);
@@ -1009,7 +1105,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 
 	trace_amdgpu_cs_ioctl(job);
 	amd_sched_entity_push_job(&job->base);
-
 	return 0;
 }
 

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

@@ -1073,6 +1073,10 @@ def_value:
 
 static void amdgpu_check_vm_size(struct amdgpu_device *adev)
 {
+	/* no need to check the default value */
+	if (amdgpu_vm_size == -1)
+		return;
+
 	if (!amdgpu_check_pot_argument(amdgpu_vm_size)) {
 		dev_warn(adev->dev, "VM size (%d) must be a power of 2\n",
 			 amdgpu_vm_size);

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

@@ -782,7 +782,7 @@ static struct drm_driver kms_driver = {
 	.driver_features =
 	    DRIVER_USE_AGP |
 	    DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | DRIVER_GEM |
-	    DRIVER_PRIME | DRIVER_RENDER | DRIVER_MODESET,
+	    DRIVER_PRIME | DRIVER_RENDER | DRIVER_MODESET | DRIVER_SYNCOBJ,
 	.load = amdgpu_driver_load_kms,
 	.open = amdgpu_driver_open_kms,
 	.postclose = amdgpu_driver_postclose_kms,

include/uapi/drm/amdgpu_drm.h

@@ -440,6 +440,8 @@ struct drm_amdgpu_gem_va {
 #define AMDGPU_CHUNK_ID_IB		0x01
 #define AMDGPU_CHUNK_ID_FENCE		0x02
 #define AMDGPU_CHUNK_ID_DEPENDENCIES	0x03
+#define AMDGPU_CHUNK_ID_SYNCOBJ_IN      0x04
+#define AMDGPU_CHUNK_ID_SYNCOBJ_OUT     0x05
 
 struct drm_amdgpu_cs_chunk {
 	__u32		chunk_id;
@@ -507,6 +509,10 @@ struct drm_amdgpu_cs_chunk_fence {
 	__u32 offset;
 };
 
+struct drm_amdgpu_cs_chunk_sem {
+	__u32 handle;
+};
+
 struct drm_amdgpu_cs_chunk_data {
 	union {
 		struct drm_amdgpu_cs_chunk_ib		ib_data;