From 28b8d66e0c1cc6c02b8159ae44aec359e12feefa Mon Sep 17 00:00:00 2001
From: Nicolai Hähnle <nicolai.haehnle@amd.com>
Date: Wed, 27 Jan 2016 11:04:19 -0500
Subject: drm/amdgpu: only move pt bos in LRU list on success
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This fixes a race condition in the error case: since the pt bos have not
necessarily been reserved in case of an error, we could move a pt bo that
is currently in the middle of being evicted/moved by another process,
which then resulted in a BUG_ON in ttm_bo_add_to_lru.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@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_cs.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 6f89f8e034d0..b882e8175615 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -478,9 +478,9 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
 	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
 	unsigned i;
 
-	amdgpu_vm_move_pt_bos_in_lru(parser->adev, &fpriv->vm);
-
 	if (!error) {
+		amdgpu_vm_move_pt_bos_in_lru(parser->adev, &fpriv->vm);
+
 		/* Sort the buffer list from the smallest to largest buffer,
 		 * which affects the order of buffers in the LRU list.
 		 * This assures that the smallest buffers are added first
-- 
cgit v1.2.3


From f69f90a113f283b33b7b750204d708711a42b396 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Mon, 21 Dec 2015 19:47:42 +0100
Subject: drm/amdgpu: fix amdgpu_cs_get_threshold_for_moves handling
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The threshold should only be computed once.

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    |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 21 ++++++++++++---------
 2 files changed, 14 insertions(+), 9 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index d0fee29ebeba..eaff96afe540 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1261,6 +1261,8 @@ struct amdgpu_cs_parser {
 	struct amdgpu_bo_list_entry	vm_pd;
 	struct list_head	validated;
 	struct fence		*fence;
+	uint64_t		bytes_moved_threshold;
+	uint64_t		bytes_moved;
 
 	struct amdgpu_ib	*ibs;
 	uint32_t		num_ibs;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index b882e8175615..2f80da016d57 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -336,14 +336,14 @@ static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
 	return max(bytes_moved_threshold, 1024*1024ull);
 }
 
-int amdgpu_cs_list_validate(struct amdgpu_device *adev,
-			    struct amdgpu_vm *vm,
+int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 			    struct list_head *validated)
 {
+	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_list_entry *lobj;
 	struct amdgpu_bo *bo;
-	u64 bytes_moved = 0, initial_bytes_moved;
-	u64 bytes_moved_threshold = amdgpu_cs_get_threshold_for_moves(adev);
+	u64 initial_bytes_moved;
 	int r;
 
 	list_for_each_entry(lobj, validated, tv.head) {
@@ -363,16 +363,16 @@ int amdgpu_cs_list_validate(struct amdgpu_device *adev,
 			 */
 			if ((lobj->allowed_domains & current_domain) != 0 &&
 			    (domain & current_domain) == 0 && /* will be moved */
-			    bytes_moved > bytes_moved_threshold) {
+			    p->bytes_moved > p->bytes_moved_threshold) {
 				/* don't move it */
 				domain = current_domain;
 			}
 
 		retry:
 			amdgpu_ttm_placement_from_domain(bo, domain);
-			initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
+			initial_bytes_moved = atomic64_read(&bo->adev->num_bytes_moved);
 			r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
-			bytes_moved += atomic64_read(&adev->num_bytes_moved) -
+			p->bytes_moved += atomic64_read(&bo->adev->num_bytes_moved) -
 				       initial_bytes_moved;
 
 			if (unlikely(r)) {
@@ -421,11 +421,14 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 
 	amdgpu_vm_get_pt_bos(&fpriv->vm, &duplicates);
 
-	r = amdgpu_cs_list_validate(p->adev, &fpriv->vm, &duplicates);
+	p->bytes_moved_threshold = amdgpu_cs_get_threshold_for_moves(p->adev);
+	p->bytes_moved = 0;
+
+	r = amdgpu_cs_list_validate(p, &duplicates);
 	if (r)
 		goto error_validate;
 
-	r = amdgpu_cs_list_validate(p->adev, &fpriv->vm, &p->validated);
+	r = amdgpu_cs_list_validate(p, &p->validated);
 
 error_validate:
 	if (r) {
-- 
cgit v1.2.3


From 36409d122cb84fa8f25a42b95a32c6090790e571 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Mon, 21 Dec 2015 20:31:35 +0100
Subject: drm/amdgpu: cleanup amdgpu_cs_list_validate
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

No need to actually check the current placement. Just use the allowed domains
when the threshold is reached.

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_cs.c | 66 ++++++++++++++++------------------
 1 file changed, 31 insertions(+), 35 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 2f80da016d57..c63efd7972d5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -342,48 +342,44 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
 	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_list_entry *lobj;
-	struct amdgpu_bo *bo;
 	u64 initial_bytes_moved;
 	int r;
 
 	list_for_each_entry(lobj, validated, tv.head) {
-		bo = lobj->robj;
-		if (!bo->pin_count) {
-			u32 domain = lobj->prefered_domains;
-			u32 current_domain =
-				amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
-
-			/* Check if this buffer will be moved and don't move it
-			 * if we have moved too many buffers for this IB already.
-			 *
-			 * Note that this allows moving at least one buffer of
-			 * any size, because it doesn't take the current "bo"
-			 * into account. We don't want to disallow buffer moves
-			 * completely.
-			 */
-			if ((lobj->allowed_domains & current_domain) != 0 &&
-			    (domain & current_domain) == 0 && /* will be moved */
-			    p->bytes_moved > p->bytes_moved_threshold) {
-				/* don't move it */
-				domain = current_domain;
-			}
+		struct amdgpu_bo *bo = lobj->robj;
+		uint32_t domain;
 
-		retry:
-			amdgpu_ttm_placement_from_domain(bo, domain);
-			initial_bytes_moved = atomic64_read(&bo->adev->num_bytes_moved);
-			r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
-			p->bytes_moved += atomic64_read(&bo->adev->num_bytes_moved) -
-				       initial_bytes_moved;
-
-			if (unlikely(r)) {
-				if (r != -ERESTARTSYS && domain != lobj->allowed_domains) {
-					domain = lobj->allowed_domains;
-					goto retry;
-				}
-				return r;
+		lobj->bo_va = amdgpu_vm_bo_find(vm, bo);
+		if (bo->pin_count)
+			continue;
+
+		/* Avoid moving this one if we have moved too many buffers
+		 * for this IB already.
+		 *
+		 * Note that this allows moving at least one buffer of
+		 * any size, because it doesn't take the current "bo"
+		 * into account. We don't want to disallow buffer moves
+		 * completely.
+		 */
+		if (p->bytes_moved <= p->bytes_moved_threshold)
+			domain = lobj->prefered_domains;
+		else
+			domain = lobj->allowed_domains;
+
+	retry:
+		amdgpu_ttm_placement_from_domain(bo, domain);
+		initial_bytes_moved = atomic64_read(&bo->adev->num_bytes_moved);
+		r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
+		p->bytes_moved += atomic64_read(&bo->adev->num_bytes_moved) -
+			       initial_bytes_moved;
+
+		if (unlikely(r)) {
+			if (r != -ERESTARTSYS && domain != lobj->allowed_domains) {
+				domain = lobj->allowed_domains;
+				goto retry;
 			}
+			return r;
 		}
-		lobj->bo_va = amdgpu_vm_bo_find(vm, bo);
 	}
 	return 0;
 }
-- 
cgit v1.2.3


From c3cca41e6249e9775adfc0ad8bc1af6fc434da9f Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Tue, 15 Dec 2015 14:41:33 +0100
Subject: drm/amdgpu: cleanup amdgpu_cs_parser structure
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Remove unused user_ptr field, group fields by usage.

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    | 21 +++++++++++----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c |  1 -
 2 files changed, 11 insertions(+), 11 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index af2981190ae0..9c1113a96ae0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1247,28 +1247,29 @@ struct amdgpu_cs_chunk {
 	uint32_t		chunk_id;
 	uint32_t		length_dw;
 	uint32_t		*kdata;
-	void __user		*user_ptr;
 };
 
 struct amdgpu_cs_parser {
 	struct amdgpu_device	*adev;
 	struct drm_file		*filp;
 	struct amdgpu_ctx	*ctx;
-	struct amdgpu_bo_list *bo_list;
+
 	/* chunks */
 	unsigned		nchunks;
 	struct amdgpu_cs_chunk	*chunks;
-	/* relocations */
-	struct amdgpu_bo_list_entry	vm_pd;
-	struct list_head	validated;
-	struct fence		*fence;
-	uint64_t		bytes_moved_threshold;
-	uint64_t		bytes_moved;
 
-	struct amdgpu_ib	*ibs;
+	/* indirect buffers */
 	uint32_t		num_ibs;
+	struct amdgpu_ib	*ibs;
 
-	struct ww_acquire_ctx	ticket;
+	/* buffer objects */
+	struct ww_acquire_ctx		ticket;
+	struct amdgpu_bo_list		*bo_list;
+	struct amdgpu_bo_list_entry	vm_pd;
+	struct list_head		validated;
+	struct fence			*fence;
+	uint64_t			bytes_moved_threshold;
+	uint64_t			bytes_moved;
 
 	/* user fence */
 	struct amdgpu_user_fence	uf;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index c63efd7972d5..03ec5e86d1a1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -217,7 +217,6 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 
 		size = p->chunks[i].length_dw;
 		cdata = (void __user *)(unsigned long)user_chunk.chunk_data;
-		p->chunks[i].user_ptr = cdata;
 
 		p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
 		if (p->chunks[i].kdata == NULL) {
-- 
cgit v1.2.3


From 2a7d9bdabec21825ef77f5705e463342a9d7fcea Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Fri, 18 Dec 2015 20:33:52 +0100
Subject: drm/amdgpu: cleanup amdgpu_cs_parser_relocs
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Rename it to amdgpu_cs_parser_bos and move validation and bo list init there.

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_cs.c | 21 ++++++++++-----------
 1 file changed, 10 insertions(+), 11 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 03ec5e86d1a1..5986da26a492 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -181,15 +181,12 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 		goto free_chunk;
 	}
 
-	p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
-
 	/* get chunks */
-	INIT_LIST_HEAD(&p->validated);
 	chunk_array_user = (uint64_t __user *)(unsigned long)(cs->in.chunks);
 	if (copy_from_user(chunk_array, chunk_array_user,
 			   sizeof(uint64_t)*cs->in.num_chunks)) {
 		ret = -EFAULT;
-		goto put_bo_list;
+		goto put_ctx;
 	}
 
 	p->nchunks = cs->in.num_chunks;
@@ -197,7 +194,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 			    GFP_KERNEL);
 	if (!p->chunks) {
 		ret = -ENOMEM;
-		goto put_bo_list;
+		goto put_ctx;
 	}
 
 	for (i = 0; i < p->nchunks; i++) {
@@ -273,9 +270,7 @@ free_partial_kdata:
 	for (; i >= 0; i--)
 		drm_free_large(p->chunks[i].kdata);
 	kfree(p->chunks);
-put_bo_list:
-	if (p->bo_list)
-		amdgpu_bo_list_put(p->bo_list);
+put_ctx:
 	amdgpu_ctx_put(p->ctx);
 free_chunk:
 	kfree(chunk_array);
@@ -383,7 +378,8 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 	return 0;
 }
 
-static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
+static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
+				union drm_amdgpu_cs *cs)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
 	struct amdgpu_cs_buckets buckets;
@@ -391,12 +387,15 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 	bool need_mmap_lock = false;
 	int i, r;
 
+	INIT_LIST_HEAD(&p->validated);
+
+	p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
 	if (p->bo_list) {
 		need_mmap_lock = p->bo_list->has_userptr;
 		amdgpu_cs_buckets_init(&buckets);
 		for (i = 0; i < p->bo_list->num_entries; i++)
 			amdgpu_cs_buckets_add(&buckets, &p->bo_list->array[i].tv.head,
-								  p->bo_list->array[i].priority);
+					      p->bo_list->array[i].priority);
 
 		amdgpu_cs_buckets_get_list(&buckets, &p->validated);
 	}
@@ -827,7 +826,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 		r = amdgpu_cs_handle_lockup(adev, r);
 		return r;
 	}
-	r = amdgpu_cs_parser_relocs(&parser);
+	r = amdgpu_cs_parser_bos(&parser, data);
 	if (r == -ENOMEM)
 		DRM_ERROR("Not enough memory for command submission!\n");
 	else if (r && r != -ERESTARTSYS)
-- 
cgit v1.2.3


From 636ce25c30010a8f393f5a1e67d5d4b7b66739e7 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Fri, 18 Dec 2015 21:26:47 +0100
Subject: drm/amdgpu: cleanup bo list bucket handling
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Move that into the BO list. No functional change.

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         |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 36 +++++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c      | 51 ++---------------------------
 3 files changed, 39 insertions(+), 50 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 9c1113a96ae0..b7056649062a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1097,6 +1097,8 @@ struct amdgpu_bo_list {
 
 struct amdgpu_bo_list *
 amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id);
+void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
+			     struct list_head *validated);
 void amdgpu_bo_list_put(struct amdgpu_bo_list *list);
 void amdgpu_bo_list_free(struct amdgpu_bo_list *list);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index f82a2dd83874..9da4bd02cec3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -32,6 +32,9 @@
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
 
+#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,
 				 struct amdgpu_bo_list **result,
 				 int *id)
@@ -106,7 +109,8 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 
 		entry->robj = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
 		drm_gem_object_unreference_unlocked(gobj);
-		entry->priority = info[i].bo_priority;
+		entry->priority = min(info[i].bo_priority,
+				      AMDGPU_BO_LIST_MAX_PRIORITY);
 		entry->prefered_domains = entry->robj->initial_domain;
 		entry->allowed_domains = entry->prefered_domains;
 		if (entry->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
@@ -161,6 +165,36 @@ amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id)
 	return result;
 }
 
+void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
+			     struct list_head *validated)
+{
+	/* This is based on the bucket sort with O(n) time complexity.
+	 * An item with priority "i" is added to bucket[i]. The lists are then
+	 * concatenated in descending order.
+	 */
+	struct list_head bucket[AMDGPU_BO_LIST_NUM_BUCKETS];
+	unsigned i;
+
+	for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++)
+		INIT_LIST_HEAD(&bucket[i]);
+
+	/* Since buffers which appear sooner in the relocation list are
+	 * likely to be used more often than buffers which appear later
+	 * in the list, the sort mustn't change the ordering of buffers
+	 * with the same priority, i.e. it must be stable.
+	 */
+	for (i = 0; i < list->num_entries; i++) {
+		unsigned priority = list->array[i].priority;
+
+		list_add_tail(&list->array[i].tv.head,
+			      &bucket[priority]);
+	}
+
+	/* Connect the sorted buckets in the output list. */
+	for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++)
+		list_splice(&bucket[i], validated);
+}
+
 void amdgpu_bo_list_put(struct amdgpu_bo_list *list)
 {
 	mutex_unlock(&list->lock);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 5986da26a492..d249e9e0a4ea 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -30,47 +30,6 @@
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
 
-#define AMDGPU_CS_MAX_PRIORITY		32u
-#define AMDGPU_CS_NUM_BUCKETS		(AMDGPU_CS_MAX_PRIORITY + 1)
-
-/* This is based on the bucket sort with O(n) time complexity.
- * An item with priority "i" is added to bucket[i]. The lists are then
- * concatenated in descending order.
- */
-struct amdgpu_cs_buckets {
-	struct list_head bucket[AMDGPU_CS_NUM_BUCKETS];
-};
-
-static void amdgpu_cs_buckets_init(struct amdgpu_cs_buckets *b)
-{
-	unsigned i;
-
-	for (i = 0; i < AMDGPU_CS_NUM_BUCKETS; i++)
-		INIT_LIST_HEAD(&b->bucket[i]);
-}
-
-static void amdgpu_cs_buckets_add(struct amdgpu_cs_buckets *b,
-				  struct list_head *item, unsigned priority)
-{
-	/* Since buffers which appear sooner in the relocation list are
-	 * likely to be used more often than buffers which appear later
-	 * in the list, the sort mustn't change the ordering of buffers
-	 * with the same priority, i.e. it must be stable.
-	 */
-	list_add_tail(item, &b->bucket[min(priority, AMDGPU_CS_MAX_PRIORITY)]);
-}
-
-static void amdgpu_cs_buckets_get_list(struct amdgpu_cs_buckets *b,
-				       struct list_head *out_list)
-{
-	unsigned i;
-
-	/* Connect the sorted buckets in the output list. */
-	for (i = 0; i < AMDGPU_CS_NUM_BUCKETS; i++) {
-		list_splice(&b->bucket[i], out_list);
-	}
-}
-
 int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
 		       u32 ip_instance, u32 ring,
 		       struct amdgpu_ring **out_ring)
@@ -382,22 +341,16 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 				union drm_amdgpu_cs *cs)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	struct amdgpu_cs_buckets buckets;
 	struct list_head duplicates;
 	bool need_mmap_lock = false;
-	int i, r;
+	int r;
 
 	INIT_LIST_HEAD(&p->validated);
 
 	p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
 	if (p->bo_list) {
 		need_mmap_lock = p->bo_list->has_userptr;
-		amdgpu_cs_buckets_init(&buckets);
-		for (i = 0; i < p->bo_list->num_entries; i++)
-			amdgpu_cs_buckets_add(&buckets, &p->bo_list->array[i].tv.head,
-					      p->bo_list->array[i].priority);
-
-		amdgpu_cs_buckets_get_list(&buckets, &p->validated);
+		amdgpu_bo_list_get_list(p->bo_list, &p->validated);
 	}
 
 	INIT_LIST_HEAD(&duplicates);
-- 
cgit v1.2.3


From 1ea863fd736eed8217a0b7bc5d0bd0de6b267b80 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Fri, 18 Dec 2015 22:13:12 +0100
Subject: drm/amdgpu: keep the prefered/allowed domains in the BO
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Stop copying that to the bo list entry, it doesn't change anyway.

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         |  5 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 15 ++++-----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c      | 10 ++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c     | 14 ++++++++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 15 +++++++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c      |  4 ----
 6 files changed, 29 insertions(+), 34 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index b7056649062a..7410d5bd37b9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -484,8 +484,6 @@ struct amdgpu_bo_list_entry {
 	struct amdgpu_bo		*robj;
 	struct ttm_validate_buffer	tv;
 	struct amdgpu_bo_va		*bo_va;
-	unsigned			prefered_domains;
-	unsigned			allowed_domains;
 	uint32_t			priority;
 };
 
@@ -522,7 +520,8 @@ struct amdgpu_bo {
 	/* Protected by gem.mutex */
 	struct list_head		list;
 	/* Protected by tbo.reserved */
-	u32				initial_domain;
+	u32				prefered_domains;
+	u32				allowed_domains;
 	struct ttm_place		placements[AMDGPU_GEM_DOMAIN_MAX + 1];
 	struct ttm_placement		placement;
 	struct ttm_buffer_object	tbo;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index 9da4bd02cec3..c4341dd4b6a4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -111,23 +111,16 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 		drm_gem_object_unreference_unlocked(gobj);
 		entry->priority = min(info[i].bo_priority,
 				      AMDGPU_BO_LIST_MAX_PRIORITY);
-		entry->prefered_domains = entry->robj->initial_domain;
-		entry->allowed_domains = entry->prefered_domains;
-		if (entry->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
-			entry->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
-		if (amdgpu_ttm_tt_has_userptr(entry->robj->tbo.ttm)) {
+		if (amdgpu_ttm_tt_has_userptr(entry->robj->tbo.ttm))
 			has_userptr = true;
-			entry->prefered_domains = AMDGPU_GEM_DOMAIN_GTT;
-			entry->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
-		}
 		entry->tv.bo = &entry->robj->tbo;
 		entry->tv.shared = true;
 
-		if (entry->prefered_domains == AMDGPU_GEM_DOMAIN_GDS)
+		if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GDS)
 			gds_obj = entry->robj;
-		if (entry->prefered_domains == AMDGPU_GEM_DOMAIN_GWS)
+		if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GWS)
 			gws_obj = entry->robj;
-		if (entry->prefered_domains == AMDGPU_GEM_DOMAIN_OA)
+		if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_OA)
 			oa_obj = entry->robj;
 
 		trace_amdgpu_bo_list_set(list, entry->robj);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index d249e9e0a4ea..a610465c5e0b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -107,8 +107,6 @@ static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
 	}
 
 	p->uf_entry.robj = amdgpu_bo_ref(p->uf.bo);
-	p->uf_entry.prefered_domains = AMDGPU_GEM_DOMAIN_GTT;
-	p->uf_entry.allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
 	p->uf_entry.priority = 0;
 	p->uf_entry.tv.bo = &p->uf_entry.robj->tbo;
 	p->uf_entry.tv.shared = true;
@@ -315,9 +313,9 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 		 * completely.
 		 */
 		if (p->bytes_moved <= p->bytes_moved_threshold)
-			domain = lobj->prefered_domains;
+			domain = bo->prefered_domains;
 		else
-			domain = lobj->allowed_domains;
+			domain = bo->allowed_domains;
 
 	retry:
 		amdgpu_ttm_placement_from_domain(bo, domain);
@@ -327,8 +325,8 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 			       initial_bytes_moved;
 
 		if (unlikely(r)) {
-			if (r != -ERESTARTSYS && domain != lobj->allowed_domains) {
-				domain = lobj->allowed_domains;
+			if (r != -ERESTARTSYS && domain != bo->allowed_domains) {
+				domain = bo->allowed_domains;
 				goto retry;
 			}
 			return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 7380f782cd14..84e850f81f39 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -252,6 +252,8 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
 		goto handle_lockup;
 
 	bo = gem_to_amdgpu_bo(gobj);
+	bo->prefered_domains = AMDGPU_GEM_DOMAIN_GTT;
+	bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
 	r = amdgpu_ttm_tt_set_userptr(bo->tbo.ttm, args->addr, args->flags);
 	if (r)
 		goto release_object;
@@ -628,7 +630,7 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 
 		info.bo_size = robj->gem_base.size;
 		info.alignment = robj->tbo.mem.page_alignment << PAGE_SHIFT;
-		info.domains = robj->initial_domain;
+		info.domains = robj->prefered_domains;
 		info.domain_flags = robj->flags;
 		amdgpu_bo_unreserve(robj);
 		if (copy_to_user(out, &info, sizeof(info)))
@@ -641,9 +643,13 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 			amdgpu_bo_unreserve(robj);
 			break;
 		}
-		robj->initial_domain = args->value & (AMDGPU_GEM_DOMAIN_VRAM |
-						      AMDGPU_GEM_DOMAIN_GTT |
-						      AMDGPU_GEM_DOMAIN_CPU);
+		robj->prefered_domains = args->value & (AMDGPU_GEM_DOMAIN_VRAM |
+							AMDGPU_GEM_DOMAIN_GTT |
+							AMDGPU_GEM_DOMAIN_CPU);
+		robj->allowed_domains = robj->prefered_domains;
+		if (robj->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
+			robj->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
+
 		amdgpu_bo_unreserve(robj);
 		break;
 	default:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index b8fbbd7699e4..8c1fc1f56342 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -254,12 +254,15 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 	bo->adev = adev;
 	INIT_LIST_HEAD(&bo->list);
 	INIT_LIST_HEAD(&bo->va);
-	bo->initial_domain = domain & (AMDGPU_GEM_DOMAIN_VRAM |
-				       AMDGPU_GEM_DOMAIN_GTT |
-				       AMDGPU_GEM_DOMAIN_CPU |
-				       AMDGPU_GEM_DOMAIN_GDS |
-				       AMDGPU_GEM_DOMAIN_GWS |
-				       AMDGPU_GEM_DOMAIN_OA);
+	bo->prefered_domains = domain & (AMDGPU_GEM_DOMAIN_VRAM |
+					 AMDGPU_GEM_DOMAIN_GTT |
+					 AMDGPU_GEM_DOMAIN_CPU |
+					 AMDGPU_GEM_DOMAIN_GDS |
+					 AMDGPU_GEM_DOMAIN_GWS |
+					 AMDGPU_GEM_DOMAIN_OA);
+	bo->allowed_domains = bo->prefered_domains;
+	if (!kernel && bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
+		bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
 
 	bo->flags = flags;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 9599f7559b3d..7e6414cffbef 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -89,8 +89,6 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 			 struct amdgpu_bo_list_entry *entry)
 {
 	entry->robj = vm->page_directory;
-	entry->prefered_domains = AMDGPU_GEM_DOMAIN_VRAM;
-	entry->allowed_domains = AMDGPU_GEM_DOMAIN_VRAM;
 	entry->priority = 0;
 	entry->tv.bo = &vm->page_directory->tbo;
 	entry->tv.shared = true;
@@ -1124,8 +1122,6 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 		}
 
 		entry->robj = pt;
-		entry->prefered_domains = AMDGPU_GEM_DOMAIN_VRAM;
-		entry->allowed_domains = AMDGPU_GEM_DOMAIN_VRAM;
 		entry->priority = 0;
 		entry->tv.bo = &entry->robj->tbo;
 		entry->tv.shared = true;
-- 
cgit v1.2.3


From 15486fd20cbe45cf56805ee5a67f7b25642efa9d Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Tue, 22 Dec 2015 16:06:12 +0100
Subject: drm/amdgpu: search only the BO list for VM mappings
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Make UVD/VCE VM emulation more efficient.

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_cs.c | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index a610465c5e0b..ac26e52d5923 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -928,30 +928,36 @@ struct amdgpu_bo_va_mapping *
 amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
 		       uint64_t addr, struct amdgpu_bo **bo)
 {
-	struct amdgpu_bo_list_entry *reloc;
 	struct amdgpu_bo_va_mapping *mapping;
+	unsigned i;
+
+	if (!parser->bo_list)
+		return NULL;
 
 	addr /= AMDGPU_GPU_PAGE_SIZE;
 
-	list_for_each_entry(reloc, &parser->validated, tv.head) {
-		if (!reloc->bo_va)
+	for (i = 0; i < parser->bo_list->num_entries; i++) {
+		struct amdgpu_bo_list_entry *lobj;
+
+		lobj = &parser->bo_list->array[i];
+		if (!lobj->bo_va)
 			continue;
 
-		list_for_each_entry(mapping, &reloc->bo_va->valids, list) {
+		list_for_each_entry(mapping, &lobj->bo_va->valids, list) {
 			if (mapping->it.start > addr ||
 			    addr > mapping->it.last)
 				continue;
 
-			*bo = reloc->bo_va->bo;
+			*bo = lobj->bo_va->bo;
 			return mapping;
 		}
 
-		list_for_each_entry(mapping, &reloc->bo_va->invalids, list) {
+		list_for_each_entry(mapping, &lobj->bo_va->invalids, list) {
 			if (mapping->it.start > addr ||
 			    addr > mapping->it.last)
 				continue;
 
-			*bo = reloc->bo_va->bo;
+			*bo = lobj->bo_va->bo;
 			return mapping;
 		}
 	}
-- 
cgit v1.2.3


From a8480309df2aa8b39e8f67b5202db8cebaff4b72 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Tue, 5 Jan 2016 16:03:39 +0100
Subject: drm/amdgpu: try to find BO VAs only for the BOs in the list
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The other ones don't have any VAs assigned anyway or are uninteresting to us.

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_cs.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index ac26e52d5923..0479ad5a66ed 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -290,8 +290,6 @@ static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
 int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 			    struct list_head *validated)
 {
-	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_list_entry *lobj;
 	u64 initial_bytes_moved;
 	int r;
@@ -300,7 +298,6 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 		struct amdgpu_bo *bo = lobj->robj;
 		uint32_t domain;
 
-		lobj->bo_va = amdgpu_vm_bo_find(vm, bo);
 		if (bo->pin_count)
 			continue;
 
@@ -374,6 +371,19 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 		goto error_validate;
 
 	r = amdgpu_cs_list_validate(p, &p->validated);
+	if (r)
+		goto error_validate;
+
+	if (p->bo_list) {
+		struct amdgpu_vm *vm = &fpriv->vm;
+		unsigned i;
+
+		for (i = 0; i < p->bo_list->num_entries; i++) {
+			struct amdgpu_bo *bo = p->bo_list->array[i].robj;
+
+			p->bo_list->array[i].bo_va = amdgpu_vm_bo_find(vm, bo);
+		}
+	}
 
 error_validate:
 	if (r) {
-- 
cgit v1.2.3


From cadf97b196a1e5b2db2606d53f77714e3e9cf4bb Mon Sep 17 00:00:00 2001
From: Chunming Zhou <David1.Zhou@amd.com>
Date: Fri, 15 Jan 2016 11:25:00 +0800
Subject: drm/amdgpu: clean up non-scheduler code path (v2)

Non-scheduler code is longer supported.

v2: agd: rebased on upstream

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Ken Wang  <Qingqing.Wang@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h       |  1 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c    | 11 +------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c   | 48 ++++++++++++++-----------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c   |  4 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 39 ++++++++++++-------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c    |  4 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 44 ++++++++++++----------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c   |  4 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c   |  4 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c   |  6 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c    |  9 ++----
 11 files changed, 64 insertions(+), 110 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 8c55c1d78ffe..9d3dff244217 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -82,7 +82,6 @@ extern int amdgpu_vm_size;
 extern int amdgpu_vm_block_size;
 extern int amdgpu_vm_fault_stop;
 extern int amdgpu_vm_debug;
-extern int amdgpu_enable_scheduler;
 extern int amdgpu_sched_jobs;
 extern int amdgpu_sched_hw_submission;
 extern int amdgpu_enable_semaphores;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 0479ad5a66ed..ddeba55c3b7d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -813,7 +813,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 	if (r)
 		goto out;
 
-	if (amdgpu_enable_scheduler && parser.num_ibs) {
+	if (parser.num_ibs) {
 		struct amdgpu_ring * ring = parser.ibs->ring;
 		struct amd_sched_fence *fence;
 		struct amdgpu_job *job;
@@ -858,15 +858,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 
 		trace_amdgpu_cs_ioctl(job);
 		amd_sched_entity_push_job(&job->base);
-
-	} else {
-		struct amdgpu_fence *fence;
-
-		r = amdgpu_ib_schedule(adev, parser.num_ibs, parser.ibs,
-				       parser.filp);
-		fence = parser.ibs[parser.num_ibs - 1].fence;
-		parser.fence = fence_get(&fence->base);
-		cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence;
 	}
 
 out:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 17d1fb12128a..f1f4b453ece1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -45,29 +45,27 @@ int amdgpu_ctx_init(struct amdgpu_device *adev, enum amd_sched_priority pri,
 		ctx->rings[i].fences = (void *)ctx->fences + sizeof(struct fence *) *
 			amdgpu_sched_jobs * i;
 	}
-	if (amdgpu_enable_scheduler) {
-		/* create context entity for each ring */
-		for (i = 0; i < adev->num_rings; i++) {
-			struct amd_sched_rq *rq;
-			if (pri >= AMD_SCHED_MAX_PRIORITY) {
-				kfree(ctx->fences);
-				return -EINVAL;
-			}
-			rq = &adev->rings[i]->sched.sched_rq[pri];
-			r = amd_sched_entity_init(&adev->rings[i]->sched,
-						  &ctx->rings[i].entity,
-						  rq, amdgpu_sched_jobs);
-			if (r)
-				break;
-		}
-
-		if (i < adev->num_rings) {
-			for (j = 0; j < i; j++)
-				amd_sched_entity_fini(&adev->rings[j]->sched,
-						      &ctx->rings[j].entity);
+	/* create context entity for each ring */
+	for (i = 0; i < adev->num_rings; i++) {
+		struct amd_sched_rq *rq;
+		if (pri >= AMD_SCHED_MAX_PRIORITY) {
 			kfree(ctx->fences);
-			return r;
+			return -EINVAL;
 		}
+		rq = &adev->rings[i]->sched.sched_rq[pri];
+		r = amd_sched_entity_init(&adev->rings[i]->sched,
+					  &ctx->rings[i].entity,
+					  rq, amdgpu_sched_jobs);
+		if (r)
+			break;
+	}
+
+	if (i < adev->num_rings) {
+		for (j = 0; j < i; j++)
+			amd_sched_entity_fini(&adev->rings[j]->sched,
+					      &ctx->rings[j].entity);
+		kfree(ctx->fences);
+		return r;
 	}
 	return 0;
 }
@@ -85,11 +83,9 @@ void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
 			fence_put(ctx->rings[i].fences[j]);
 	kfree(ctx->fences);
 
-	if (amdgpu_enable_scheduler) {
-		for (i = 0; i < adev->num_rings; i++)
-			amd_sched_entity_fini(&adev->rings[i]->sched,
-					      &ctx->rings[i].entity);
-	}
+	for (i = 0; i < adev->num_rings; i++)
+		amd_sched_entity_fini(&adev->rings[i]->sched,
+				      &ctx->rings[i].entity);
 }
 
 static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 11573fd1f053..8af888a2aa9f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -78,7 +78,6 @@ int amdgpu_vm_block_size = -1;
 int amdgpu_vm_fault_stop = 0;
 int amdgpu_vm_debug = 0;
 int amdgpu_exp_hw_support = 0;
-int amdgpu_enable_scheduler = 1;
 int amdgpu_sched_jobs = 32;
 int amdgpu_sched_hw_submission = 2;
 int amdgpu_powerplay = -1;
@@ -152,9 +151,6 @@ module_param_named(vm_debug, amdgpu_vm_debug, int, 0644);
 MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
 module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
 
-MODULE_PARM_DESC(enable_scheduler, "enable SW GPU scheduler (1 = enable (default), 0 = disable)");
-module_param_named(enable_scheduler, amdgpu_enable_scheduler, int, 0444);
-
 MODULE_PARM_DESC(sched_jobs, "the max number of jobs supported in the sw queue (default 32)");
 module_param_named(sched_jobs, amdgpu_sched_jobs, int, 0444);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 3671f9f220bd..cac03e743b58 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -472,6 +472,7 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
 int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
 {
 	int i, r;
+	long timeout;
 
 	ring->fence_drv.cpu_addr = NULL;
 	ring->fence_drv.gpu_addr = 0;
@@ -486,26 +487,24 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
 
 	init_waitqueue_head(&ring->fence_drv.fence_queue);
 
-	if (amdgpu_enable_scheduler) {
-		long timeout = msecs_to_jiffies(amdgpu_lockup_timeout);
-		if (timeout == 0) {
-			/*
-			 * FIXME:
-			 * Delayed workqueue cannot use it directly,
-			 * so the scheduler will not use delayed workqueue if
-			 * MAX_SCHEDULE_TIMEOUT is set.
-			 * Currently keep it simple and silly.
-			 */
-			timeout = MAX_SCHEDULE_TIMEOUT;
-		}
-		r = amd_sched_init(&ring->sched, &amdgpu_sched_ops,
-				   amdgpu_sched_hw_submission,
-				   timeout, ring->name);
-		if (r) {
-			DRM_ERROR("Failed to create scheduler on ring %s.\n",
-				  ring->name);
-			return r;
-		}
+	timeout = msecs_to_jiffies(amdgpu_lockup_timeout);
+	if (timeout == 0) {
+		/*
+		 * FIXME:
+		 * Delayed workqueue cannot use it directly,
+		 * so the scheduler will not use delayed workqueue if
+		 * MAX_SCHEDULE_TIMEOUT is set.
+		 * Currently keep it simple and silly.
+		 */
+		timeout = MAX_SCHEDULE_TIMEOUT;
+	}
+	r = amd_sched_init(&ring->sched, &amdgpu_sched_ops,
+			   amdgpu_sched_hw_submission,
+			   timeout, ring->name);
+	if (r) {
+		DRM_ERROR("Failed to create scheduler on ring %s.\n",
+			  ring->name);
+		return r;
 	}
 
 	return 0;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index 3b58d70b73cd..54cede30a69c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -199,10 +199,6 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
 		return r;
 	}
 
-	if (!amdgpu_enable_scheduler && ib->ctx)
-		ib->sequence = amdgpu_ctx_add_fence(ib->ctx, ring,
-						    &ib->fence->base);
-
 	/* wrap the last IB with fence */
 	if (ib->user) {
 		uint64_t addr = amdgpu_bo_gpu_offset(ib->user->bo);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index 438c05254695..dd9fac302e55 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -76,33 +76,25 @@ int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
 					 void *owner,
 					 struct fence **f)
 {
-	int r = 0;
-	if (amdgpu_enable_scheduler) {
-		struct amdgpu_job *job =
-			kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
-		if (!job)
-			return -ENOMEM;
-		job->base.sched = &ring->sched;
-		job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
-		job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner);
-		if (!job->base.s_fence) {
-			kfree(job);
-			return -ENOMEM;
-		}
-		*f = fence_get(&job->base.s_fence->base);
-
-		job->adev = adev;
-		job->ibs = ibs;
-		job->num_ibs = num_ibs;
-		job->owner = owner;
-		job->free_job = free_job;
-		amd_sched_entity_push_job(&job->base);
-	} else {
-		r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner);
-		if (r)
-			return r;
-		*f = fence_get(&ibs[num_ibs - 1].fence->base);
+	struct amdgpu_job *job =
+		kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
+	if (!job)
+		return -ENOMEM;
+	job->base.sched = &ring->sched;
+	job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
+	job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner);
+	if (!job->base.s_fence) {
+		kfree(job);
+		return -ENOMEM;
 	}
+	*f = fence_get(&job->base.s_fence->base);
+
+	job->adev = adev;
+	job->ibs = ibs;
+	job->num_ibs = num_ibs;
+	job->owner = owner;
+	job->free_job = free_job;
+	amd_sched_entity_push_job(&job->base);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 6442a06d6fdc..100bfd4a0707 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1070,10 +1070,6 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 	if (r)
 		goto error_free;
 
-	if (!amdgpu_enable_scheduler) {
-		amdgpu_ib_free(adev, ib);
-		kfree(ib);
-	}
 	return 0;
 error_free:
 	amdgpu_ib_free(adev, ib);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index 53f987aeeacf..72193f1c8e99 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -895,11 +895,7 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 		*fence = fence_get(f);
 	amdgpu_bo_unref(&bo);
 	fence_put(f);
-	if (amdgpu_enable_scheduler)
-		return 0;
 
-	amdgpu_ib_free(ring->adev, ib);
-	kfree(ib);
 	return 0;
 err2:
 	amdgpu_ib_free(ring->adev, ib);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index e882fbfacb12..16fbde9c5f56 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -432,8 +432,7 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 	if (fence)
 		*fence = fence_get(f);
 	fence_put(f);
-	if (amdgpu_enable_scheduler)
-		return 0;
+	return 0;
 err:
 	amdgpu_ib_free(adev, ib);
 	kfree(ib);
@@ -499,8 +498,7 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 	if (fence)
 		*fence = fence_get(f);
 	fence_put(f);
-	if (amdgpu_enable_scheduler)
-		return 0;
+	return 0;
 err:
 	amdgpu_ib_free(adev, ib);
 	kfree(ib);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 7e6414cffbef..cc28bdc02078 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -401,8 +401,7 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 	if (!r)
 		amdgpu_bo_fence(bo, fence, true);
 	fence_put(fence);
-	if (amdgpu_enable_scheduler)
-		return 0;
+	return 0;
 
 error_free:
 	amdgpu_ib_free(adev, ib);
@@ -536,7 +535,7 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 		fence_put(fence);
 	}
 
-	if (!amdgpu_enable_scheduler || ib->length_dw == 0) {
+	if (ib->length_dw == 0) {
 		amdgpu_ib_free(adev, ib);
 		kfree(ib);
 	}
@@ -819,10 +818,6 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 		*fence = fence_get(f);
 	}
 	fence_put(f);
-	if (!amdgpu_enable_scheduler) {
-		amdgpu_ib_free(adev, ib);
-		kfree(ib);
-	}
 	return 0;
 
 error_free:
-- 
cgit v1.2.3


From cd75dc6887f1e1fddc4ee8ef5ae4e40529607379 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Sun, 31 Jan 2016 11:30:55 +0100
Subject: drm/amdgpu: separate pushing CS to scheduler
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Move that out of the main IOCTL function.

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_cs.c | 96 ++++++++++++++++++----------------
 1 file changed, 50 insertions(+), 46 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index ddeba55c3b7d..a795af3ab0ad 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -766,6 +766,54 @@ static int amdgpu_cs_free_job(struct amdgpu_job *job)
 	return 0;
 }
 
+static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
+			    union drm_amdgpu_cs *cs)
+{
+	struct amdgpu_ring * ring = p->ibs->ring;
+	struct amd_sched_fence *fence;
+	struct amdgpu_job *job;
+
+	job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
+	if (!job)
+		return -ENOMEM;
+
+	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;
+
+	job->ibs = p->ibs;
+	job->num_ibs = p->num_ibs;
+	p->ibs = NULL;
+	p->num_ibs = 0;
+
+	if (job->ibs[job->num_ibs - 1].user) {
+		job->uf = p->uf;
+		job->ibs[job->num_ibs - 1].user = &job->uf;
+		p->uf.bo = NULL;
+	}
+
+	fence = amd_sched_fence_create(job->base.s_entity, p->filp);
+	if (!fence) {
+		amdgpu_cs_free_job(job);
+		kfree(job);
+		return -ENOMEM;
+	}
+
+	job->base.s_fence = fence;
+	p->fence = fence_get(&fence->base);
+
+	cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring,
+					      &fence->base);
+	job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
+
+	trace_amdgpu_cs_ioctl(job);
+	amd_sched_entity_push_job(&job->base);
+
+	return 0;
+}
+
 int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
 	struct amdgpu_device *adev = dev->dev_private;
@@ -813,52 +861,8 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 	if (r)
 		goto out;
 
-	if (parser.num_ibs) {
-		struct amdgpu_ring * ring = parser.ibs->ring;
-		struct amd_sched_fence *fence;
-		struct amdgpu_job *job;
-
-		job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
-		if (!job) {
-			r = -ENOMEM;
-			goto out;
-		}
-
-		job->base.sched = &ring->sched;
-		job->base.s_entity = &parser.ctx->rings[ring->idx].entity;
-		job->adev = parser.adev;
-		job->owner = parser.filp;
-		job->free_job = amdgpu_cs_free_job;
-
-		job->ibs = parser.ibs;
-		job->num_ibs = parser.num_ibs;
-		parser.ibs = NULL;
-		parser.num_ibs = 0;
-
-		if (job->ibs[job->num_ibs - 1].user) {
-			job->uf = parser.uf;
-			job->ibs[job->num_ibs - 1].user = &job->uf;
-			parser.uf.bo = NULL;
-		}
-
-		fence = amd_sched_fence_create(job->base.s_entity,
-					       parser.filp);
-		if (!fence) {
-			r = -ENOMEM;
-			amdgpu_cs_free_job(job);
-			kfree(job);
-			goto out;
-		}
-		job->base.s_fence = fence;
-		parser.fence = fence_get(&fence->base);
-
-		cs->out.handle = amdgpu_ctx_add_fence(parser.ctx, ring,
-						      &fence->base);
-		job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
-
-		trace_amdgpu_cs_ioctl(job);
-		amd_sched_entity_push_job(&job->base);
-	}
+	if (parser.num_ibs)
+		r = amdgpu_cs_submit(&parser, cs);
 
 out:
 	amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
-- 
cgit v1.2.3


From cc325d1913475655b81c0417ba04c84f168ac78c Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Mon, 8 Feb 2016 11:08:35 +0100
Subject: drm/amdgpu: check userptrs mm earlier
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Instead of when we try to bind it check the usermm when
we try to use it in the IOCTLs.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h         |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 16 +++++++++++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c      |  7 ++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c     |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c   |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     |  9 +++------
 7 files changed, 27 insertions(+), 15 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 4a6e87857875..a783c684b49b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -2364,7 +2364,7 @@ void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *rbo, u32 domain);
 bool amdgpu_ttm_bo_is_amdgpu_bo(struct ttm_buffer_object *bo);
 int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
 				     uint32_t flags);
-bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm);
+struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm);
 bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
 				  unsigned long end);
 bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index c4341dd4b6a4..90d6fc1618aa 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -93,6 +93,7 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 
 	bool has_userptr = false;
 	unsigned i;
+	int r;
 
 	array = drm_malloc_ab(num_entries, sizeof(struct amdgpu_bo_list_entry));
 	if (!array)
@@ -102,17 +103,26 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 	for (i = 0; i < num_entries; ++i) {
 		struct amdgpu_bo_list_entry *entry = &array[i];
 		struct drm_gem_object *gobj;
+		struct mm_struct *usermm;
 
 		gobj = drm_gem_object_lookup(adev->ddev, filp, info[i].bo_handle);
-		if (!gobj)
+		if (!gobj) {
+			r = -ENOENT;
 			goto error_free;
+		}
 
 		entry->robj = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
 		drm_gem_object_unreference_unlocked(gobj);
 		entry->priority = min(info[i].bo_priority,
 				      AMDGPU_BO_LIST_MAX_PRIORITY);
-		if (amdgpu_ttm_tt_has_userptr(entry->robj->tbo.ttm))
+		usermm = amdgpu_ttm_tt_get_usermm(entry->robj->tbo.ttm);
+		if (usermm) {
+			if (usermm != current->mm) {
+				r = -EPERM;
+				goto error_free;
+			}
 			has_userptr = true;
+		}
 		entry->tv.bo = &entry->robj->tbo;
 		entry->tv.shared = true;
 
@@ -142,7 +152,7 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 
 error_free:
 	drm_free_large(array);
-	return -ENOENT;
+	return r;
 }
 
 struct amdgpu_bo_list *
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index a795af3ab0ad..e7e384264202 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -101,7 +101,7 @@ static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
 	p->uf.bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
 	p->uf.offset = fence_data->offset;
 
-	if (amdgpu_ttm_tt_has_userptr(p->uf.bo->tbo.ttm)) {
+	if (amdgpu_ttm_tt_get_usermm(p->uf.bo->tbo.ttm)) {
 		drm_gem_object_unreference_unlocked(gobj);
 		return -EINVAL;
 	}
@@ -296,8 +296,13 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 
 	list_for_each_entry(lobj, validated, tv.head) {
 		struct amdgpu_bo *bo = lobj->robj;
+		struct mm_struct *usermm;
 		uint32_t domain;
 
+		usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm);
+		if (usermm && usermm != current->mm)
+			return -EPERM;
+
 		if (bo->pin_count)
 			continue;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 84e850f81f39..b1a611e79c94 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -310,7 +310,7 @@ int amdgpu_mode_dumb_mmap(struct drm_file *filp,
 		return -ENOENT;
 	}
 	robj = gem_to_amdgpu_bo(gobj);
-	if (amdgpu_ttm_tt_has_userptr(robj->tbo.ttm) ||
+	if (amdgpu_ttm_tt_get_usermm(robj->tbo.ttm) ||
 	    (robj->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)) {
 		drm_gem_object_unreference_unlocked(gobj);
 		return -EPERM;
@@ -638,7 +638,7 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 		break;
 	}
 	case AMDGPU_GEM_OP_SET_PLACEMENT:
-		if (amdgpu_ttm_tt_has_userptr(robj->tbo.ttm)) {
+		if (amdgpu_ttm_tt_get_usermm(robj->tbo.ttm)) {
 			r = -EPERM;
 			amdgpu_bo_unreserve(robj);
 			break;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 8c1fc1f56342..b74084ce734f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -370,7 +370,7 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
 	int r, i;
 	unsigned fpfn, lpfn;
 
-	if (amdgpu_ttm_tt_has_userptr(bo->tbo.ttm))
+	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm))
 		return -EPERM;
 
 	if (WARN_ON_ONCE(min_offset > max_offset))
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
index 59f735a933a9..a0bdb0e45c0f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
@@ -121,7 +121,7 @@ struct dma_buf *amdgpu_gem_prime_export(struct drm_device *dev,
 {
 	struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
 
-	if (amdgpu_ttm_tt_has_userptr(bo->tbo.ttm))
+	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm))
 		return ERR_PTR(-EPERM);
 
 	return drm_gem_prime_export(dev, gobj, flags);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 100bfd4a0707..cdb393db11a5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -499,9 +499,6 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
 	enum dma_data_direction direction = write ?
 		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
 
-	if (current->mm != gtt->usermm)
-		return -EPERM;
-
 	if (gtt->userflags & AMDGPU_GEM_USERPTR_ANONONLY) {
 		/* check that we only pin down anonymous memory
 		   to prevent problems with writeback */
@@ -773,14 +770,14 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
 	return 0;
 }
 
-bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm)
+struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
 
 	if (gtt == NULL)
-		return false;
+		return NULL;
 
-	return !!gtt->userptr;
+	return gtt->usermm;
 }
 
 bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
-- 
cgit v1.2.3


From 4acabfe3793eb9bf89f71cc0cef23dfb2a812916 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Sun, 31 Jan 2016 11:32:04 +0100
Subject: drm/amdgpu: fix num_ibs check
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Specifying no IBs on command submission is invalid, stop crashing
badly when somebody tries it.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 16 +++++-----------
 1 file changed, 5 insertions(+), 11 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index e7e384264202..55179efccfcf 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -211,6 +211,10 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 		}
 	}
 
+	if (p->num_ibs == 0) {
+		ret = -EINVAL;
+		goto free_all_kdata;
+	}
 
 	p->ibs = kcalloc(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL);
 	if (!p->ibs) {
@@ -551,9 +555,6 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
 	struct amdgpu_ring *ring;
 	int i, r;
 
-	if (parser->num_ibs == 0)
-		return 0;
-
 	/* Only for UVD/VCE VM emulation */
 	for (i = 0; i < parser->num_ibs; i++) {
 		ring = parser->ibs[i].ring;
@@ -660,9 +661,6 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 		j++;
 	}
 
-	if (!parser->num_ibs)
-		return 0;
-
 	/* add GDS resources to first IB */
 	if (parser->bo_list) {
 		struct amdgpu_bo *gds = parser->bo_list->gds_obj;
@@ -705,9 +703,6 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 	struct amdgpu_ib *ib;
 	int i, j, r;
 
-	if (!p->num_ibs)
-		return 0;
-
 	/* Add dependencies to first IB */
 	ib = &p->ibs[0];
 	for (i = 0; i < p->nchunks; ++i) {
@@ -866,8 +861,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 	if (r)
 		goto out;
 
-	if (parser.num_ibs)
-		r = amdgpu_cs_submit(&parser, cs);
+	r = amdgpu_cs_submit(&parser, cs);
 
 out:
 	amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
-- 
cgit v1.2.3


From 50838c8cc413de8da39c4c216ae05410845d5a44 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Wed, 3 Feb 2016 13:44:52 +0100
Subject: drm/amdgpu: add proper job alloc/free functions
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

And use them in the CS instead of allocating IBs and jobs separately.

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       | 12 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c    | 65 +++++++++++--------------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 33 ++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h |  6 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c   |  6 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c   |  2 +-
 6 files changed, 69 insertions(+), 55 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 935b9aa08474..5c55445f0260 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -796,6 +796,9 @@ enum amdgpu_ring_type {
 
 extern struct amd_sched_backend_ops amdgpu_sched_ops;
 
+int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
+		     struct amdgpu_job **job);
+void amdgpu_job_free(struct amdgpu_job *job);
 int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
 					 struct amdgpu_ring *ring,
 					 struct amdgpu_ib *ibs,
@@ -1216,9 +1219,8 @@ struct amdgpu_cs_parser {
 	unsigned		nchunks;
 	struct amdgpu_cs_chunk	*chunks;
 
-	/* indirect buffers */
-	uint32_t		num_ibs;
-	struct amdgpu_ib	*ibs;
+	/* scheduler job object */
+	struct amdgpu_job	*job;
 
 	/* buffer objects */
 	struct ww_acquire_ctx		ticket;
@@ -1249,14 +1251,14 @@ struct amdgpu_job {
 static inline u32 amdgpu_get_ib_value(struct amdgpu_cs_parser *p,
 				      uint32_t ib_idx, int idx)
 {
-	return p->ibs[ib_idx].ptr[idx];
+	return p->job->ibs[ib_idx].ptr[idx];
 }
 
 static inline void amdgpu_set_ib_value(struct amdgpu_cs_parser *p,
 				       uint32_t ib_idx, int idx,
 				       uint32_t value)
 {
-	p->ibs[ib_idx].ptr[idx] = value;
+	p->job->ibs[ib_idx].ptr[idx] = value;
 }
 
 /*
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 55179efccfcf..e9d88771783b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -121,7 +121,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 	uint64_t *chunk_array_user;
 	uint64_t *chunk_array;
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	unsigned size;
+	unsigned size, num_ibs = 0;
 	int i;
 	int ret;
 
@@ -186,7 +186,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 
 		switch (p->chunks[i].chunk_id) {
 		case AMDGPU_CHUNK_ID_IB:
-			p->num_ibs++;
+			++num_ibs;
 			break;
 
 		case AMDGPU_CHUNK_ID_FENCE:
@@ -211,16 +211,9 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 		}
 	}
 
-	if (p->num_ibs == 0) {
-		ret = -EINVAL;
+	ret = amdgpu_job_alloc(p->adev, num_ibs, &p->job);
+	if (ret)
 		goto free_all_kdata;
-	}
-
-	p->ibs = kcalloc(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!p->ibs) {
-		ret = -ENOMEM;
-		goto free_all_kdata;
-	}
 
 	kfree(chunk_array);
 	return 0;
@@ -414,7 +407,7 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 
 	list_for_each_entry(e, &p->validated, tv.head) {
 		struct reservation_object *resv = e->robj->tbo.resv;
-		r = amdgpu_sync_resv(p->adev, &p->ibs[0].sync, resv, p->filp);
+		r = amdgpu_sync_resv(p->adev, &p->job->ibs[0].sync, resv, p->filp);
 
 		if (r)
 			return r;
@@ -477,10 +470,8 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
 	for (i = 0; i < parser->nchunks; i++)
 		drm_free_large(parser->chunks[i].kdata);
 	kfree(parser->chunks);
-	if (parser->ibs)
-		for (i = 0; i < parser->num_ibs; i++)
-			amdgpu_ib_free(parser->adev, &parser->ibs[i]);
-	kfree(parser->ibs);
+	if (parser->job)
+		amdgpu_job_free(parser->job);
 	amdgpu_bo_unref(&parser->uf.bo);
 	amdgpu_bo_unref(&parser->uf_entry.robj);
 }
@@ -497,7 +488,7 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(adev, &p->ibs[0].sync, vm->page_directory_fence);
+	r = amdgpu_sync_fence(adev, &p->job->ibs[0].sync, vm->page_directory_fence);
 	if (r)
 		return r;
 
@@ -523,14 +514,14 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 				return r;
 
 			f = bo_va->last_pt_update;
-			r = amdgpu_sync_fence(adev, &p->ibs[0].sync, f);
+			r = amdgpu_sync_fence(adev, &p->job->ibs[0].sync, f);
 			if (r)
 				return r;
 		}
 
 	}
 
-	r = amdgpu_vm_clear_invalids(adev, vm, &p->ibs[0].sync);
+	r = amdgpu_vm_clear_invalids(adev, vm, &p->job->ibs[0].sync);
 
 	if (amdgpu_vm_debug && p->bo_list) {
 		/* Invalidate all BOs to test for userspace bugs */
@@ -556,8 +547,8 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
 	int i, r;
 
 	/* Only for UVD/VCE VM emulation */
-	for (i = 0; i < parser->num_ibs; i++) {
-		ring = parser->ibs[i].ring;
+	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);
 			if (r)
@@ -590,14 +581,14 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 	int i, j;
 	int r;
 
-	for (i = 0, j = 0; i < parser->nchunks && j < parser->num_ibs; i++) {
+	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;
 
 		chunk = &parser->chunks[i];
-		ib = &parser->ibs[j];
+		ib = &parser->job->ibs[j];
 		chunk_ib = (struct drm_amdgpu_cs_chunk_ib *)chunk->kdata;
 
 		if (chunk->chunk_id != AMDGPU_CHUNK_ID_IB)
@@ -666,7 +657,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 		struct amdgpu_bo *gds = parser->bo_list->gds_obj;
 		struct amdgpu_bo *gws = parser->bo_list->gws_obj;
 		struct amdgpu_bo *oa = parser->bo_list->oa_obj;
-		struct amdgpu_ib *ib = &parser->ibs[0];
+		struct amdgpu_ib *ib = &parser->job->ibs[0];
 
 		if (gds) {
 			ib->gds_base = amdgpu_bo_gpu_offset(gds);
@@ -683,7 +674,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 	}
 	/* wrap the last IB with user fence */
 	if (parser->uf.bo) {
-		struct amdgpu_ib *ib = &parser->ibs[parser->num_ibs - 1];
+		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 ||
@@ -704,7 +695,7 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 	int i, j, r;
 
 	/* Add dependencies to first IB */
-	ib = &p->ibs[0];
+	ib = &p->job->ibs[0];
 	for (i = 0; i < p->nchunks; ++i) {
 		struct drm_amdgpu_cs_chunk_dep *deps;
 		struct amdgpu_cs_chunk *chunk;
@@ -756,26 +747,19 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 
 static int amdgpu_cs_free_job(struct amdgpu_job *job)
 {
-	int i;
-	if (job->ibs)
-		for (i = 0; i < job->num_ibs; i++)
-			amdgpu_ib_free(job->adev, &job->ibs[i]);
-	kfree(job->ibs);
-	if (job->uf.bo)
-		amdgpu_bo_unref(&job->uf.bo);
+	amdgpu_job_free(job);
 	return 0;
 }
 
 static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 			    union drm_amdgpu_cs *cs)
 {
-	struct amdgpu_ring * ring = p->ibs->ring;
+	struct amdgpu_ring * ring = p->job->ibs->ring;
 	struct amd_sched_fence *fence;
 	struct amdgpu_job *job;
 
-	job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
-	if (!job)
-		return -ENOMEM;
+	job = p->job;
+	p->job = NULL;
 
 	job->base.sched = &ring->sched;
 	job->base.s_entity = &p->ctx->rings[ring->idx].entity;
@@ -783,11 +767,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	job->owner = p->filp;
 	job->free_job = amdgpu_cs_free_job;
 
-	job->ibs = p->ibs;
-	job->num_ibs = p->num_ibs;
-	p->ibs = NULL;
-	p->num_ibs = 0;
-
 	if (job->ibs[job->num_ibs - 1].user) {
 		job->uf = p->uf;
 		job->ibs[job->num_ibs - 1].user = &job->uf;
@@ -854,7 +833,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 	if (r)
 		goto out;
 
-	for (i = 0; i < parser.num_ibs; i++)
+	for (i = 0; i < parser.job->num_ibs; i++)
 		trace_amdgpu_cs(&parser, i);
 
 	r = amdgpu_cs_ib_vm_chunk(adev, &parser);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index 76a1f823d983..10d098e33707 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -28,6 +28,39 @@
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
 
+int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
+		     struct amdgpu_job **job)
+{
+	size_t size = sizeof(struct amdgpu_job);
+
+	if (num_ibs == 0)
+		return -EINVAL;
+
+	size += sizeof(struct amdgpu_ib) * num_ibs;
+
+	*job = kzalloc(size, GFP_KERNEL);
+	if (!*job)
+		return -ENOMEM;
+
+	(*job)->adev = adev;
+	(*job)->ibs = (void *)&(*job)[1];
+	(*job)->num_ibs = num_ibs;
+	(*job)->free_job = NULL;
+
+	return 0;
+}
+
+void amdgpu_job_free(struct amdgpu_job *job)
+{
+	unsigned i;
+
+	for (i = 0; i < job->num_ibs; ++i)
+		amdgpu_ib_free(job->adev, &job->ibs[i]);
+
+	amdgpu_bo_unref(&job->uf.bo);
+	/* TODO: Free the job structure here as well */
+}
+
 static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job)
 {
 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
index e7d2676839e1..f808b5a6e52b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
+++ b/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->ibs[i].ring->idx;
-			   __entry->dw = p->ibs[i].length_dw;
+			   __entry->ring = p->job->ibs[i].ring->idx;
+			   __entry->dw = p->job->ibs[i].length_dw;
 			   __entry->fences = amdgpu_fence_count_emitted(
-				p->ibs[i].ring);
+				p->job->ibs[i].ring);
 			   ),
 	    TP_printk("bo_list=%p, ring=%u, dw=%u, fences=%u",
 		      __entry->bo_list, __entry->ring, __entry->dw,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index 676ef699b6ad..f2bd2334bbe6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -702,7 +702,7 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx)
 static int amdgpu_uvd_cs_reg(struct amdgpu_uvd_cs_ctx *ctx,
 			     int (*cb)(struct amdgpu_uvd_cs_ctx *ctx))
 {
-	struct amdgpu_ib *ib = &ctx->parser->ibs[ctx->ib_idx];
+	struct amdgpu_ib *ib = &ctx->parser->job->ibs[ctx->ib_idx];
 	int i, r;
 
 	ctx->idx++;
@@ -748,7 +748,7 @@ static int amdgpu_uvd_cs_reg(struct amdgpu_uvd_cs_ctx *ctx,
 static int amdgpu_uvd_cs_packets(struct amdgpu_uvd_cs_ctx *ctx,
 				 int (*cb)(struct amdgpu_uvd_cs_ctx *ctx))
 {
-	struct amdgpu_ib *ib = &ctx->parser->ibs[ctx->ib_idx];
+	struct amdgpu_ib *ib = &ctx->parser->job->ibs[ctx->ib_idx];
 	int r;
 
 	for (ctx->idx = 0 ; ctx->idx < ib->length_dw; ) {
@@ -790,7 +790,7 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx)
 		[0x00000003]	=	2048,
 		[0x00000004]	=	0xFFFFFFFF,
 	};
-	struct amdgpu_ib *ib = &parser->ibs[ib_idx];
+	struct amdgpu_ib *ib = &parser->job->ibs[ib_idx];
 	int r;
 
 	if (ib->length_dw % 16) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 43aad45969a7..66b7bfafca24 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -603,7 +603,7 @@ static int amdgpu_vce_validate_handle(struct amdgpu_cs_parser *p,
  */
 int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx)
 {
-	struct amdgpu_ib *ib = &p->ibs[ib_idx];
+	struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
 	unsigned fb_idx = 0, bs_idx = 0;
 	int session_idx = -1;
 	bool destroyed = false;
-- 
cgit v1.2.3


From 4c0b242cf27094b666df6604420417e201260af9 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Mon, 1 Feb 2016 11:20:37 +0100
Subject: drm/amdgpu: cleanup user fence handling in the CS
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Don't keep that around twice.

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    |  1 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 29 +++++++++++++----------------
 2 files changed, 13 insertions(+), 17 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 5c55445f0260..364b0d1c4566 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1232,7 +1232,6 @@ struct amdgpu_cs_parser {
 	uint64_t			bytes_moved;
 
 	/* user fence */
-	struct amdgpu_user_fence	uf;
 	struct amdgpu_bo_list_entry	uf_entry;
 };
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index e9d88771783b..8f3b72f5c91c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -87,6 +87,7 @@ 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_gem_object *gobj;
@@ -98,15 +99,15 @@ static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
 	if (gobj == NULL)
 		return -EINVAL;
 
-	p->uf.bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
-	p->uf.offset = fence_data->offset;
+	uf->bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
+	uf->offset = fence_data->offset;
 
-	if (amdgpu_ttm_tt_get_usermm(p->uf.bo->tbo.ttm)) {
+	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(p->uf.bo);
+	p->uf_entry.robj = amdgpu_bo_ref(uf->bo);
 	p->uf_entry.priority = 0;
 	p->uf_entry.tv.bo = &p->uf_entry.robj->tbo;
 	p->uf_entry.tv.shared = true;
@@ -117,10 +118,11 @@ static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
 
 int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 {
+	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
 	union drm_amdgpu_cs *cs = data;
 	uint64_t *chunk_array_user;
 	uint64_t *chunk_array;
-	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct amdgpu_user_fence uf = {};
 	unsigned size, num_ibs = 0;
 	int i;
 	int ret;
@@ -196,7 +198,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 				goto free_partial_kdata;
 			}
 
-			ret = amdgpu_cs_user_fence_chunk(p, (void *)p->chunks[i].kdata);
+			ret = amdgpu_cs_user_fence_chunk(p, &uf, (void *)p->chunks[i].kdata);
 			if (ret)
 				goto free_partial_kdata;
 
@@ -215,6 +217,8 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 	if (ret)
 		goto free_all_kdata;
 
+	p->job->uf = uf;
+
 	kfree(chunk_array);
 	return 0;
 
@@ -353,7 +357,7 @@ 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->uf.bo)
+	if (p->job->uf.bo)
 		list_add(&p->uf_entry.tv.head, &p->validated);
 
 	if (need_mmap_lock)
@@ -472,7 +476,6 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo
 	kfree(parser->chunks);
 	if (parser->job)
 		amdgpu_job_free(parser->job);
-	amdgpu_bo_unref(&parser->uf.bo);
 	amdgpu_bo_unref(&parser->uf_entry.robj);
 }
 
@@ -673,7 +676,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 		}
 	}
 	/* wrap the last IB with user fence */
-	if (parser->uf.bo) {
+	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 */
@@ -681,7 +684,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
 		    ib->ring->type == AMDGPU_RING_TYPE_VCE)
 			return -EINVAL;
 
-		ib->user = &parser->uf;
+		ib->user = &parser->job->uf;
 	}
 
 	return 0;
@@ -767,12 +770,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	job->owner = p->filp;
 	job->free_job = amdgpu_cs_free_job;
 
-	if (job->ibs[job->num_ibs - 1].user) {
-		job->uf = p->uf;
-		job->ibs[job->num_ibs - 1].user = &job->uf;
-		p->uf.bo = NULL;
-	}
-
 	fence = amd_sched_fence_create(job->base.s_entity, p->filp);
 	if (!fence) {
 		amdgpu_cs_free_job(job);
-- 
cgit v1.2.3


From b07c60c0652c497af0c42c1278941f7c5a187fe9 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Sun, 31 Jan 2016 12:29:04 +0100
Subject: drm/amdgpu: move ring from IBs into job
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

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       |  6 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c    | 33 +++++++++++++++++--------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c    | 11 ++++-------
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c |  5 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h |  8 ++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c   |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c   |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c   |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c    |  6 +++---
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c     |  2 +-
 drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c     |  2 +-
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c     |  4 ++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c    |  2 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c    |  2 +-
 14 files changed, 45 insertions(+), 44 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index c23eea308991..dd6dd23fb3c7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 8f3b72f5c91c..d928165bfc33 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/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);
+	if (ring->funcs->parse_cs) {
+		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;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index b6737707eda2..47196ec593fc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index 10d098e33707..84453c1c4b07 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
index f808b5a6e52b..9ca3735c563c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
+++ b/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",
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index a7354798f087..6b63dcc3a773 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index f2bd2334bbe6..46f2bdb76d20 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/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;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 66b7bfafca24..bcbe4167d7b1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/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);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index d0acc9c0fd7f..b68642b47b7b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
index 1007a98f775b..f352d7559686 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index 40159d529e5b..5f57a820dacb 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 2ca36606c2ef..dbfa895c2ae5 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
index 5dfb13c56b90..49ec93cc7166 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
index 999a169299f0..8fe150f02634 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+++ b/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;
-- 
cgit v1.2.3


From d71518b5aa7c9c298ffbd12ddd23297e3373a37b Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Mon, 1 Feb 2016 12:20:25 +0100
Subject: drm/amdgpu: cleanup in kernel job submission
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Add a job_alloc_with_ib helper and proper job submission.

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       | 13 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c    | 10 +---
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 76 +++++++++++++--------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c   | 34 +++++-------
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c   | 37 +++++--------
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c   | 58 ++++++---------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c    | 87 ++++++++++---------------------
 7 files changed, 111 insertions(+), 204 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 387a26bba65a..6cefde4aab49 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -797,14 +797,11 @@ extern struct amd_sched_backend_ops amdgpu_sched_ops;
 
 int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
 		     struct amdgpu_job **job);
+int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
+			     struct amdgpu_job **job);
 void amdgpu_job_free(struct amdgpu_job *job);
-int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
-					 struct amdgpu_ring *ring,
-					 struct amdgpu_ib *ibs,
-					 unsigned num_ibs,
-					 int (*free_job)(struct amdgpu_job *),
-					 void *owner,
-					 struct fence **fence);
+int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
+		      void *owner, struct fence **f);
 
 struct amdgpu_ring {
 	struct amdgpu_device		*adev;
@@ -987,7 +984,6 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
 		       uint64_t addr);
 void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 		      struct amdgpu_bo_va *bo_va);
-int amdgpu_vm_free_job(struct amdgpu_job *job);
 
 /*
  * context related structures
@@ -1244,7 +1240,6 @@ struct amdgpu_job {
 	uint32_t		num_ibs;
 	void			*owner;
 	struct amdgpu_user_fence uf;
-	int (*free_job)(struct amdgpu_job *job);
 };
 #define to_amdgpu_job(sched_job)		\
 		container_of((sched_job), struct amdgpu_job, base)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index d928165bfc33..a5311623a489 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -752,12 +752,6 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 	return 0;
 }
 
-static int amdgpu_cs_free_job(struct amdgpu_job *job)
-{
-	amdgpu_job_free(job);
-	return 0;
-}
-
 static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 			    union drm_amdgpu_cs *cs)
 {
@@ -771,12 +765,10 @@ 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->owner = p->filp;
-	job->free_job = amdgpu_cs_free_job;
 
 	fence = amd_sched_fence_create(job->base.s_entity, p->filp);
 	if (!fence) {
-		amdgpu_cs_free_job(job);
-		kfree(job);
+		amdgpu_job_free(job);
 		return -ENOMEM;
 	}
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index cabb0fc28610..bbdda727f89a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -45,11 +45,26 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
 	(*job)->adev = adev;
 	(*job)->ibs = (void *)&(*job)[1];
 	(*job)->num_ibs = num_ibs;
-	(*job)->free_job = NULL;
 
 	return 0;
 }
 
+int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
+			     struct amdgpu_job **job)
+{
+	int r;
+
+	r = amdgpu_job_alloc(adev, 1, job);
+	if (r)
+		return r;
+
+	r = amdgpu_ib_get(adev, NULL, size, &(*job)->ibs[0]);
+	if (r)
+		kfree(*job);
+
+	return r;
+}
+
 void amdgpu_job_free(struct amdgpu_job *job)
 {
 	unsigned i;
@@ -58,7 +73,27 @@ void amdgpu_job_free(struct amdgpu_job *job)
 		amdgpu_ib_free(job->adev, &job->ibs[i]);
 
 	amdgpu_bo_unref(&job->uf.bo);
-	/* TODO: Free the job structure here as well */
+	kfree(job);
+}
+
+int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
+		      void *owner, struct fence **f)
+{
+	struct amdgpu_device *adev = job->adev;
+
+	job->ring = ring;
+	job->base.sched = &ring->sched;
+	job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
+	job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner);
+	if (!job->base.s_fence)
+		return -ENOMEM;
+
+	*f = fence_get(&job->base.s_fence->base);
+
+	job->owner = owner;
+	amd_sched_entity_push_job(&job->base);
+
+	return 0;
 }
 
 static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job)
@@ -106,10 +141,7 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job)
 	}
 
 err:
-	if (job->free_job)
-		job->free_job(job);
-
-	kfree(job);
+	amdgpu_job_free(job);
 	return fence;
 }
 
@@ -117,35 +149,3 @@ struct amd_sched_backend_ops amdgpu_sched_ops = {
 	.dependency = amdgpu_sched_dependency,
 	.run_job = amdgpu_sched_run_job,
 };
-
-int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
-					 struct amdgpu_ring *ring,
-					 struct amdgpu_ib *ibs,
-					 unsigned num_ibs,
-					 int (*free_job)(struct amdgpu_job *),
-					 void *owner,
-					 struct fence **f)
-{
-	struct amdgpu_job *job =
-		kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
-	if (!job)
-		return -ENOMEM;
-	job->base.sched = &ring->sched;
-	job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
-	job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner);
-	if (!job->base.s_fence) {
-		kfree(job);
-		return -ENOMEM;
-	}
-	*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;
-	job->free_job = free_job;
-	amd_sched_entity_push_job(&job->base);
-
-	return 0;
-}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 6b63dcc3a773..21c1a18c6d48 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1012,9 +1012,10 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 		       struct fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
+	struct amdgpu_job *job;
+
 	uint32_t max_bytes;
 	unsigned num_loops, num_dw;
-	struct amdgpu_ib *ib;
 	unsigned i;
 	int r;
 
@@ -1026,20 +1027,12 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 	while (num_dw & 0x7)
 		num_dw++;
 
-	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!ib)
-		return -ENOMEM;
-
-	r = amdgpu_ib_get(adev, NULL, num_dw * 4, ib);
-	if (r) {
-		kfree(ib);
+	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, &job);
+	if (r)
 		return r;
-	}
-
-	ib->length_dw = 0;
 
 	if (resv) {
-		r = amdgpu_sync_resv(adev, &ib->sync, resv,
+		r = amdgpu_sync_resv(adev, &job->ibs[0].sync, resv,
 				     AMDGPU_FENCE_OWNER_UNDEFINED);
 		if (r) {
 			DRM_ERROR("sync failed (%d).\n", r);
@@ -1050,27 +1043,24 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 	for (i = 0; i < num_loops; i++) {
 		uint32_t cur_size_in_bytes = min(byte_count, max_bytes);
 
-		amdgpu_emit_copy_buffer(adev, ib, src_offset, dst_offset,
-					cur_size_in_bytes);
+		amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_offset,
+					dst_offset, cur_size_in_bytes);
 
 		src_offset += cur_size_in_bytes;
 		dst_offset += cur_size_in_bytes;
 		byte_count -= cur_size_in_bytes;
 	}
 
-	amdgpu_ring_pad_ib(ring, ib);
-	WARN_ON(ib->length_dw > num_dw);
-	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
-						 &amdgpu_vm_free_job,
-						 AMDGPU_FENCE_OWNER_UNDEFINED,
-						 fence);
+	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
+	WARN_ON(job->ibs[0].length_dw > num_dw);
+	r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_UNDEFINED, fence);
 	if (r)
 		goto error_free;
 
 	return 0;
+
 error_free:
-	amdgpu_ib_free(adev, ib);
-	kfree(ib);
+	amdgpu_job_free(job);
 	return r;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index 46f2bdb76d20..4e98be8b29fb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -823,14 +823,6 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx)
 	return 0;
 }
 
-static int amdgpu_uvd_free_job(
-	struct amdgpu_job *job)
-{
-	amdgpu_ib_free(job->adev, job->ibs);
-	kfree(job->ibs);
-	return 0;
-}
-
 static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 			       struct amdgpu_bo *bo,
 			       struct fence **fence)
@@ -838,7 +830,8 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 	struct ttm_validate_buffer tv;
 	struct ww_acquire_ctx ticket;
 	struct list_head head;
-	struct amdgpu_ib *ib = NULL;
+	struct amdgpu_job *job;
+	struct amdgpu_ib *ib;
 	struct fence *f = NULL;
 	struct amdgpu_device *adev = ring->adev;
 	uint64_t addr;
@@ -862,15 +855,12 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 	r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
 	if (r)
 		goto err;
-	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!ib) {
-		r = -ENOMEM;
-		goto err;
-	}
-	r = amdgpu_ib_get(adev, NULL, 64, ib);
+
+	r = amdgpu_job_alloc_with_ib(adev, 64, &job);
 	if (r)
-		goto err1;
+		goto err;
 
+	ib = &job->ibs[0];
 	addr = amdgpu_bo_gpu_offset(bo);
 	ib->ptr[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0);
 	ib->ptr[1] = addr;
@@ -882,12 +872,9 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 		ib->ptr[i] = PACKET2(0);
 	ib->length_dw = 16;
 
-	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
-						 &amdgpu_uvd_free_job,
-						 AMDGPU_FENCE_OWNER_UNDEFINED,
-						 &f);
+	r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
 	if (r)
-		goto err2;
+		goto err_free;
 
 	ttm_eu_fence_buffer_objects(&ticket, &head, f);
 
@@ -897,10 +884,10 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring,
 	fence_put(f);
 
 	return 0;
-err2:
-	amdgpu_ib_free(ring->adev, ib);
-err1:
-	kfree(ib);
+
+err_free:
+	amdgpu_job_free(job);
+
 err:
 	ttm_eu_backoff_reservation(&ticket, &head);
 	return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index bcbe4167d7b1..66e97ea0e6b0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -346,14 +346,6 @@ void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
 	}
 }
 
-static int amdgpu_vce_free_job(
-	struct amdgpu_job *job)
-{
-	amdgpu_ib_free(job->adev, job->ibs);
-	kfree(job->ibs);
-	return 0;
-}
-
 /**
  * amdgpu_vce_get_create_msg - generate a VCE create msg
  *
@@ -368,21 +360,17 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 			      struct fence **fence)
 {
 	const unsigned ib_size_dw = 1024;
-	struct amdgpu_ib *ib = NULL;
+	struct amdgpu_job *job;
+	struct amdgpu_ib *ib;
 	struct fence *f = NULL;
-	struct amdgpu_device *adev = ring->adev;
 	uint64_t dummy;
 	int i, r;
 
-	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!ib)
-		return -ENOMEM;
-	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);
+	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
+	if (r)
 		return r;
-	}
+
+	ib = &job->ibs[0];
 
 	dummy = ib->gpu_addr + 1024;
 
@@ -423,19 +411,16 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 	for (i = ib->length_dw; i < ib_size_dw; ++i)
 		ib->ptr[i] = 0x0;
 
-	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
-						 &amdgpu_vce_free_job,
-						 AMDGPU_FENCE_OWNER_UNDEFINED,
-						 &f);
+	r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
 	if (r)
 		goto err;
 	if (fence)
 		*fence = fence_get(f);
 	fence_put(f);
 	return 0;
+
 err:
-	amdgpu_ib_free(adev, ib);
-	kfree(ib);
+	amdgpu_job_free(job);
 	return r;
 }
 
@@ -453,23 +438,17 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 			       struct fence **fence)
 {
 	const unsigned ib_size_dw = 1024;
-	struct amdgpu_ib *ib = NULL;
+	struct amdgpu_job *job;
+	struct amdgpu_ib *ib;
 	struct fence *f = NULL;
-	struct amdgpu_device *adev = ring->adev;
 	uint64_t dummy;
 	int i, r;
 
-	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!ib)
-		return -ENOMEM;
-
-	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);
+	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
+	if (r)
 		return r;
-	}
 
+	ib = &job->ibs[0];
 	dummy = ib->gpu_addr + 1024;
 
 	/* stitch together an VCE destroy msg */
@@ -489,19 +468,16 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 
 	for (i = ib->length_dw; i < ib_size_dw; ++i)
 		ib->ptr[i] = 0x0;
-	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
-						 &amdgpu_vce_free_job,
-						 AMDGPU_FENCE_OWNER_UNDEFINED,
-						 &f);
+	r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
 	if (r)
 		goto err;
 	if (fence)
 		*fence = fence_get(f);
 	fence_put(f);
 	return 0;
+
 err:
-	amdgpu_ib_free(adev, ib);
-	kfree(ib);
+	amdgpu_job_free(job);
 	return r;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index b68642b47b7b..fb003089f73c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -313,15 +313,6 @@ static void amdgpu_vm_update_pages(struct amdgpu_device *adev,
 	}
 }
 
-int amdgpu_vm_free_job(struct amdgpu_job *job)
-{
-	int i;
-	for (i = 0; i < job->num_ibs; i++)
-		amdgpu_ib_free(job->adev, &job->ibs[i]);
-	kfree(job->ibs);
-	return 0;
-}
-
 /**
  * amdgpu_vm_clear_bo - initially clear the page dir/table
  *
@@ -335,7 +326,7 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 {
 	struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring;
 	struct fence *fence = NULL;
-	struct amdgpu_ib *ib;
+	struct amdgpu_job *job;
 	unsigned entries;
 	uint64_t addr;
 	int r;
@@ -351,32 +342,25 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
 	addr = amdgpu_bo_gpu_offset(bo);
 	entries = amdgpu_bo_size(bo) / 8;
 
-	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!ib)
+	r = amdgpu_job_alloc_with_ib(adev, 64, &job);
+	if (r)
 		goto error;
 
-	r = amdgpu_ib_get(adev, NULL, 64, ib);
+	amdgpu_vm_update_pages(adev, NULL, 0, &job->ibs[0], addr, 0, entries,
+			       0, 0);
+	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
+
+	WARN_ON(job->ibs[0].length_dw > 64);
+	r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_VM, &fence);
 	if (r)
 		goto error_free;
 
-	ib->length_dw = 0;
-
-	amdgpu_vm_update_pages(adev, NULL, 0, ib, addr, 0, entries, 0, 0);
-	amdgpu_ring_pad_ib(ring, ib);
-
-	WARN_ON(ib->length_dw > 64);
-	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
-						 &amdgpu_vm_free_job,
-						 AMDGPU_FENCE_OWNER_VM,
-						 &fence);
-	if (!r)
-		amdgpu_bo_fence(bo, fence, true);
+	amdgpu_bo_fence(bo, fence, true);
 	fence_put(fence);
 	return 0;
 
 error_free:
-	amdgpu_ib_free(adev, ib);
-	kfree(ib);
+	amdgpu_job_free(job);
 
 error:
 	return r;
@@ -433,6 +417,7 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 	uint32_t incr = AMDGPU_VM_PTE_COUNT * 8;
 	uint64_t last_pde = ~0, last_pt = ~0;
 	unsigned count = 0, pt_idx, ndw;
+	struct amdgpu_job *job;
 	struct amdgpu_ib *ib;
 	struct fence *fence = NULL;
 
@@ -444,16 +429,11 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 	/* assume the worst case */
 	ndw += vm->max_pde_used * 6;
 
-	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!ib)
-		return -ENOMEM;
-
-	r = amdgpu_ib_get(adev, NULL, ndw * 4, ib);
-	if (r) {
-		kfree(ib);
+	r = amdgpu_job_alloc_with_ib(adev, ndw * 4, &job);
+	if (r)
 		return r;
-	}
-	ib->length_dw = 0;
+
+	ib = &job->ibs[0];
 
 	/* walk over the address space and update the page directory */
 	for (pt_idx = 0; pt_idx <= vm->max_pde_used; ++pt_idx) {
@@ -495,10 +475,7 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 		amdgpu_ring_pad_ib(ring, ib);
 		amdgpu_sync_resv(adev, &ib->sync, pd->tbo.resv, AMDGPU_FENCE_OWNER_VM);
 		WARN_ON(ib->length_dw > ndw);
-		r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
-							 &amdgpu_vm_free_job,
-							 AMDGPU_FENCE_OWNER_VM,
-							 &fence);
+		r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_VM, &fence);
 		if (r)
 			goto error_free;
 
@@ -506,18 +483,15 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 		fence_put(vm->page_directory_fence);
 		vm->page_directory_fence = fence_get(fence);
 		fence_put(fence);
-	}
 
-	if (ib->length_dw == 0) {
-		amdgpu_ib_free(adev, ib);
-		kfree(ib);
+	} else {
+		amdgpu_job_free(job);
 	}
 
 	return 0;
 
 error_free:
-	amdgpu_ib_free(adev, ib);
-	kfree(ib);
+	amdgpu_job_free(job);
 	return r;
 }
 
@@ -695,6 +669,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring;
 	void *owner = AMDGPU_FENCE_OWNER_VM;
 	unsigned nptes, ncmds, ndw;
+	struct amdgpu_job *job;
 	struct amdgpu_ib *ib;
 	struct fence *f = NULL;
 	int r;
@@ -733,15 +708,11 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 		ndw += 2 * 10;
 	}
 
-	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
-	if (!ib)
-		return -ENOMEM;
-
-	r = amdgpu_ib_get(adev, NULL, ndw * 4, ib);
-	if (r) {
-		kfree(ib);
+	r = amdgpu_job_alloc_with_ib(adev, ndw * 4, &job);
+	if (r)
 		return r;
-	}
+
+	ib = &job->ibs[0];
 
 	r = amdgpu_sync_resv(adev, &ib->sync, vm->page_directory->tbo.resv,
 			     owner);
@@ -757,10 +728,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 
 	amdgpu_ring_pad_ib(ring, ib);
 	WARN_ON(ib->length_dw > ndw);
-	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
-						 &amdgpu_vm_free_job,
-						 AMDGPU_FENCE_OWNER_VM,
-						 &f);
+	r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_VM, &f);
 	if (r)
 		goto error_free;
 
@@ -773,8 +741,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	return 0;
 
 error_free:
-	amdgpu_ib_free(adev, ib);
-	kfree(ib);
+	amdgpu_job_free(job);
 	return r;
 }
 
-- 
cgit v1.2.3


From e86f9ceee19ec028ae79a09fe1eaabf315d67969 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Mon, 8 Feb 2016 12:13:05 +0100
Subject: drm/amdgpu: move sync into job object
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

No need to keep that for every IB.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  3 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  | 14 ++++++--------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c  | 13 ++-----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 21 +++++++++++++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c |  6 ++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  |  5 +++--
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c   |  3 ++-
 drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c   |  3 ++-
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c   |  6 ++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c  |  3 ++-
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c  |  3 ++-
 13 files changed, 46 insertions(+), 38 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 6cefde4aab49..99e660fec190 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -776,7 +776,6 @@ struct amdgpu_ib {
 	bool				grabbed_vmid;
 	struct amdgpu_vm		*vm;
 	struct amdgpu_ctx		*ctx;
-	struct amdgpu_sync		sync;
 	uint32_t			gds_base, gds_size;
 	uint32_t			gws_base, gws_size;
 	uint32_t			oa_base, oa_size;
@@ -1178,6 +1177,7 @@ int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib);
 int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 		       struct amdgpu_ib *ib, void *owner,
+		       struct fence *last_vm_update,
 		       struct fence **f);
 int amdgpu_ib_pool_init(struct amdgpu_device *adev);
 void amdgpu_ib_pool_fini(struct amdgpu_device *adev);
@@ -1236,6 +1236,7 @@ struct amdgpu_job {
 	struct amd_sched_job    base;
 	struct amdgpu_device	*adev;
 	struct amdgpu_ring	*ring;
+	struct amdgpu_sync	sync;
 	struct amdgpu_ib	*ibs;
 	uint32_t		num_ibs;
 	void			*owner;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index a5311623a489..52c3eb96b199 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -411,7 +411,7 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 
 	list_for_each_entry(e, &p->validated, tv.head) {
 		struct reservation_object *resv = e->robj->tbo.resv;
-		r = amdgpu_sync_resv(p->adev, &p->job->ibs[0].sync, resv, p->filp);
+		r = amdgpu_sync_resv(p->adev, &p->job->sync, resv, p->filp);
 
 		if (r)
 			return r;
@@ -491,7 +491,7 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(adev, &p->job->ibs[0].sync, vm->page_directory_fence);
+	r = amdgpu_sync_fence(adev, &p->job->sync, vm->page_directory_fence);
 	if (r)
 		return r;
 
@@ -517,14 +517,14 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 				return r;
 
 			f = bo_va->last_pt_update;
-			r = amdgpu_sync_fence(adev, &p->job->ibs[0].sync, f);
+			r = amdgpu_sync_fence(adev, &p->job->sync, f);
 			if (r)
 				return r;
 		}
 
 	}
 
-	r = amdgpu_vm_clear_invalids(adev, vm, &p->job->ibs[0].sync);
+	r = amdgpu_vm_clear_invalids(adev, vm, &p->job->sync);
 
 	if (amdgpu_vm_debug && p->bo_list) {
 		/* Invalidate all BOs to test for userspace bugs */
@@ -698,11 +698,8 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 				  struct amdgpu_cs_parser *p)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	struct amdgpu_ib *ib;
 	int i, j, r;
 
-	/* Add dependencies to first IB */
-	ib = &p->job->ibs[0];
 	for (i = 0; i < p->nchunks; ++i) {
 		struct drm_amdgpu_cs_chunk_dep *deps;
 		struct amdgpu_cs_chunk *chunk;
@@ -740,7 +737,8 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
 				return r;
 
 			} else if (fence) {
-				r = amdgpu_sync_fence(adev, &ib->sync, fence);
+				r = amdgpu_sync_fence(adev, &p->job->sync,
+						      fence);
 				fence_put(fence);
 				amdgpu_ctx_put(ctx);
 				if (r)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index 4b2c176b83f2..b5bdd5d59b58 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -74,8 +74,6 @@ int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 			ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo);
 	}
 
-	amdgpu_sync_create(&ib->sync);
-
 	ib->vm = vm;
 
 	return 0;
@@ -91,7 +89,6 @@ int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
  */
 void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib)
 {
-	amdgpu_sync_free(&ib->sync);
 	amdgpu_sa_bo_free(adev, &ib->sa_bo, &ib->fence->base);
 	if (ib->fence)
 		fence_put(&ib->fence->base);
@@ -121,6 +118,7 @@ void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib)
  */
 int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 		       struct amdgpu_ib *ibs, void *owner,
+		       struct fence *last_vm_update,
 		       struct fence **f)
 {
 	struct amdgpu_device *adev = ring->adev;
@@ -152,16 +150,9 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 		return r;
 	}
 
-	r = amdgpu_sync_wait(&ibs->sync);
-	if (r) {
-		amdgpu_ring_undo(ring);
-		dev_err(adev->dev, "failed to sync wait (%d)\n", r);
-		return r;
-	}
-
 	if (vm) {
 		/* do context switch */
-		amdgpu_vm_flush(ring, vm, ib->sync.last_vm_update);
+		amdgpu_vm_flush(ring, vm, last_vm_update);
 
 		if (ring->funcs->emit_gds_switch)
 			amdgpu_ring_emit_gds_switch(ring, ib->vm->ids[ring->idx].id,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 6f3e757e056e..0f6719e0ace0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -46,6 +46,8 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
 	(*job)->ibs = (void *)&(*job)[1];
 	(*job)->num_ibs = num_ibs;
 
+	amdgpu_sync_create(&(*job)->sync);
+
 	return 0;
 }
 
@@ -73,6 +75,7 @@ void amdgpu_job_free(struct amdgpu_job *job)
 		amdgpu_ib_free(job->adev, &job->ibs[i]);
 
 	amdgpu_bo_unref(&job->uf.bo);
+	amdgpu_sync_free(&job->sync);
 	kfree(job);
 }
 
@@ -99,23 +102,22 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
 static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
 {
 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
-	struct amdgpu_sync *sync = &job->ibs->sync;
 	struct amdgpu_vm *vm = job->ibs->vm;
 
-	struct fence *fence = amdgpu_sync_get_fence(sync);
+	struct fence *fence = amdgpu_sync_get_fence(&job->sync);
 
 	if (fence == NULL && vm && !job->ibs->grabbed_vmid) {
 		struct amdgpu_ring *ring = job->ring;
 		int r;
 
-		r = amdgpu_vm_grab_id(vm, ring, sync,
+		r = amdgpu_vm_grab_id(vm, ring, &job->sync,
 				      &job->base.s_fence->base);
 		if (r)
 			DRM_ERROR("Error getting VM ID (%d)\n", r);
 		else
 			job->ibs->grabbed_vmid = true;
 
-		fence = amdgpu_sync_get_fence(sync);
+		fence = amdgpu_sync_get_fence(&job->sync);
 	}
 
 	return fence;
@@ -132,9 +134,16 @@ static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job)
 		return NULL;
 	}
 	job = to_amdgpu_job(sched_job);
+
+	r = amdgpu_sync_wait(&job->sync);
+	if (r) {
+		DRM_ERROR("failed to sync wait (%d)\n", r);
+		return NULL;
+	}
+
 	trace_amdgpu_sched_run_job(job);
-	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs,
-			       job->owner, &fence);
+	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job->owner,
+			       job->sync.last_vm_update, &fence);
 	if (r) {
 		DRM_ERROR("Error scheduling IBs (%d)\n", r);
 		goto err;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 21c1a18c6d48..e47d5188c886 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1032,7 +1032,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 		return r;
 
 	if (resv) {
-		r = amdgpu_sync_resv(adev, &job->ibs[0].sync, resv,
+		r = amdgpu_sync_resv(adev, &job->sync, resv,
 				     AMDGPU_FENCE_OWNER_UNDEFINED);
 		if (r) {
 			DRM_ERROR("sync failed (%d).\n", r);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index 00b608b6c8c4..c536630580f8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -874,7 +874,7 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo,
 
 	if (direct) {
 		r = amdgpu_ib_schedule(ring, 1, ib,
-				       AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+				       AMDGPU_FENCE_OWNER_UNDEFINED, NULL, &f);
 		if (r)
 			goto err_free;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 97c22212d048..fb2ce3ed9aab 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -411,7 +411,8 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 	for (i = ib->length_dw; i < ib_size_dw; ++i)
 		ib->ptr[i] = 0x0;
 
-	r = amdgpu_ib_schedule(ring, 1, ib, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+	r = amdgpu_ib_schedule(ring, 1, ib, AMDGPU_FENCE_OWNER_UNDEFINED,
+			       NULL, &f);
 	if (r)
 		goto err;
 
@@ -473,7 +474,8 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 
 	if (direct) {
 		r = amdgpu_ib_schedule(ring, 1, ib,
-				       AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+				       AMDGPU_FENCE_OWNER_UNDEFINED,
+				       NULL, &f);
 		if (r)
 			goto err;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index fb003089f73c..b291b1a4611a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -473,7 +473,8 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
 
 	if (ib->length_dw != 0) {
 		amdgpu_ring_pad_ib(ring, ib);
-		amdgpu_sync_resv(adev, &ib->sync, pd->tbo.resv, AMDGPU_FENCE_OWNER_VM);
+		amdgpu_sync_resv(adev, &job->sync, pd->tbo.resv,
+				 AMDGPU_FENCE_OWNER_VM);
 		WARN_ON(ib->length_dw > ndw);
 		r = amdgpu_job_submit(job, ring, AMDGPU_FENCE_OWNER_VM, &fence);
 		if (r)
@@ -714,7 +715,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 
 	ib = &job->ibs[0];
 
-	r = amdgpu_sync_resv(adev, &ib->sync, vm->page_directory->tbo.resv,
+	r = amdgpu_sync_resv(adev, &job->sync, vm->page_directory->tbo.resv,
 			     owner);
 	if (r)
 		goto error_free;
diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
index 6004dce23dc1..47fba3142a5a 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
@@ -633,7 +633,8 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
 	ib.ptr[3] = 1;
 	ib.ptr[4] = 0xDEADBEEF;
 	ib.length_dw = 5;
-	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED,
+			       NULL, &f);
 	if (r)
 		goto err1;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index 4dac79ac9d3d..fd6796657051 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -2641,7 +2641,8 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring)
 	ib.ptr[2] = 0xDEADBEEF;
 	ib.length_dw = 3;
 
-	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED,
+			       NULL, &f);
 	if (r)
 		goto err2;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 4a7708541723..415da6e100cd 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -709,7 +709,8 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring)
 	ib.ptr[2] = 0xDEADBEEF;
 	ib.length_dw = 3;
 
-	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED,
+			       NULL, &f);
 	if (r)
 		goto err2;
 
@@ -1264,7 +1265,8 @@ static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
 
 	/* shedule the ib on the ring */
-	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED,
+			       NULL, &f);
 	if (r) {
 		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
 		goto fail;
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
index 6a048b0b5fa7..423be6e10d01 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
@@ -691,7 +691,8 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring)
 	ib.ptr[7] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
 	ib.length_dw = 8;
 
-	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED,
+			       NULL, &f);
 	if (r)
 		goto err1;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
index 446aa016e96e..960462a9e941 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
@@ -842,7 +842,8 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring)
 	ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
 	ib.length_dw = 8;
 
-	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED, &f);
+	r = amdgpu_ib_schedule(ring, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED,
+			       NULL, &f);
 	if (r)
 		goto err1;
 
-- 
cgit v1.2.3


From 211dff5518d938d0c11205af3113d2c7a292dd98 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Mon, 22 Feb 2016 15:40:59 +0100
Subject: drm/amdgpu: group userptr in the BO list v2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

We need them together with the next patch.

v2: Don't take bo reference twice

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h         |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 24 +++++++++++++++---------
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c      |  3 ++-
 3 files changed, 18 insertions(+), 11 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 497ca587cc35..11a9f5899995 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1048,7 +1048,7 @@ struct amdgpu_bo_list {
 	struct amdgpu_bo *gds_obj;
 	struct amdgpu_bo *gws_obj;
 	struct amdgpu_bo *oa_obj;
-	bool has_userptr;
+	unsigned first_userptr;
 	unsigned num_entries;
 	struct amdgpu_bo_list_entry *array;
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index 4792f9d0b7d4..9763e52886fe 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -91,7 +91,7 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 	struct amdgpu_bo *gws_obj = adev->gds.gws_gfx_bo;
 	struct amdgpu_bo *oa_obj = adev->gds.oa_gfx_bo;
 
-	bool has_userptr = false;
+	unsigned last_entry = 0, first_userptr = num_entries;
 	unsigned i;
 	int r;
 
@@ -101,8 +101,9 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 	memset(array, 0, num_entries * sizeof(struct amdgpu_bo_list_entry));
 
 	for (i = 0; i < num_entries; ++i) {
-		struct amdgpu_bo_list_entry *entry = &array[i];
+		struct amdgpu_bo_list_entry *entry;
 		struct drm_gem_object *gobj;
+		struct amdgpu_bo *bo;
 		struct mm_struct *usermm;
 
 		gobj = drm_gem_object_lookup(adev->ddev, filp, info[i].bo_handle);
@@ -111,19 +112,24 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 			goto error_free;
 		}
 
-		entry->robj = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
+		bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
 		drm_gem_object_unreference_unlocked(gobj);
-		entry->priority = min(info[i].bo_priority,
-				      AMDGPU_BO_LIST_MAX_PRIORITY);
-		usermm = amdgpu_ttm_tt_get_usermm(entry->robj->tbo.ttm);
+
+		usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm);
 		if (usermm) {
 			if (usermm != current->mm) {
-				amdgpu_bo_unref(&entry->robj);
+				amdgpu_bo_unref(&bo);
 				r = -EPERM;
 				goto error_free;
 			}
-			has_userptr = true;
+			entry = &array[--first_userptr];
+		} else {
+			entry = &array[last_entry++];
 		}
+
+		entry->robj = bo;
+		entry->priority = min(info[i].bo_priority,
+				      AMDGPU_BO_LIST_MAX_PRIORITY);
 		entry->tv.bo = &entry->robj->tbo;
 		entry->tv.shared = true;
 
@@ -145,7 +151,7 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 	list->gds_obj = gds_obj;
 	list->gws_obj = gws_obj;
 	list->oa_obj = oa_obj;
-	list->has_userptr = has_userptr;
+	list->first_userptr = first_userptr;
 	list->array = array;
 	list->num_entries = num_entries;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 52c3eb96b199..7833dfb1ff6e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -350,7 +350,8 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 
 	p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
 	if (p->bo_list) {
-		need_mmap_lock = p->bo_list->has_userptr;
+		need_mmap_lock = p->bo_list->first_userptr !=
+			p->bo_list->num_entries;
 		amdgpu_bo_list_get_list(p->bo_list, &p->validated);
 	}
 
-- 
cgit v1.2.3


From 2f568dbd6b944c2e8c0c54b53c2211c23995e6a4 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Tue, 23 Feb 2016 12:36:59 +0100
Subject: drm/amdgpu: move get_user_pages out of amdgpu_ttm_tt_pin_userptr v6
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

That avoids lock inversion between the BO reservation lock
and the anon_vma lock.

v2:
* Changed amdgpu_bo_list_entry.user_pages to an array of pointers
* Lock mmap_sem only for get_user_pages
* Added invalidation of unbound userpointer BOs
* Fixed memory leak and page reference leak

v3 (chk):
* Revert locking mmap_sem only for_get user_pages
* Revert adding invalidation of unbound userpointer BOs
* Sanitize and fix error handling

v4 (chk):
* Init userpages pointer everywhere.
* Fix error handling when get_user_pages() fails.
* Add invalidation of unbound userpointer BOs again.

v5 (chk):
* Add maximum number of tries.

v6 (chk):
* Fix error handling when we run out of tries.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> (v4)
Acked-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h         |   5 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c |   1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c      | 114 +++++++++++++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c     |  23 ++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     |  53 +++++++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c      |   2 +
 6 files changed, 174 insertions(+), 24 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index e0727de9c2b2..e32ab13227d4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -471,6 +471,8 @@ struct amdgpu_bo_list_entry {
 	struct ttm_validate_buffer	tv;
 	struct amdgpu_bo_va		*bo_va;
 	uint32_t			priority;
+	struct page			**user_pages;
+	int				user_invalidated;
 };
 
 struct amdgpu_bo_va_mapping {
@@ -2365,11 +2367,14 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
 		       struct amdgpu_ring **out_ring);
 void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *rbo, u32 domain);
 bool amdgpu_ttm_bo_is_amdgpu_bo(struct ttm_buffer_object *bo);
+int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages);
 int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
 				     uint32_t flags);
 struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm);
 bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
 				  unsigned long end);
+bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
+				       int *last_invalidated);
 bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm);
 uint32_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
 				 struct ttm_mem_reg *mem);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index 9763e52886fe..eacd810fc09b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -200,6 +200,7 @@ void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
 
 		list_add_tail(&list->array[i].tv.head,
 			      &bucket[priority]);
+		list->array[i].user_pages = NULL;
 	}
 
 	/* Connect the sorted buckets in the output list. */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 7833dfb1ff6e..4f5ef4149e87 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -111,6 +111,7 @@ static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
 	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;
 
 	drm_gem_object_unreference_unlocked(gobj);
 	return 0;
@@ -297,6 +298,7 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 
 	list_for_each_entry(lobj, validated, tv.head) {
 		struct amdgpu_bo *bo = lobj->robj;
+		bool binding_userptr = false;
 		struct mm_struct *usermm;
 		uint32_t domain;
 
@@ -304,6 +306,15 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 		if (usermm && usermm != current->mm)
 			return -EPERM;
 
+		/* Check if we have user pages and nobody bound the BO already */
+		if (lobj->user_pages && bo->tbo.ttm->state != tt_bound) {
+			size_t size = sizeof(struct page *);
+
+			size *= bo->tbo.ttm->num_pages;
+			memcpy(bo->tbo.ttm->pages, lobj->user_pages, size);
+			binding_userptr = true;
+		}
+
 		if (bo->pin_count)
 			continue;
 
@@ -334,6 +345,11 @@ int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 			}
 			return r;
 		}
+
+		if (binding_userptr) {
+			drm_free_large(lobj->user_pages);
+			lobj->user_pages = NULL;
+		}
 	}
 	return 0;
 }
@@ -342,8 +358,10 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 				union drm_amdgpu_cs *cs)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct amdgpu_bo_list_entry *e;
 	struct list_head duplicates;
 	bool need_mmap_lock = false;
+	unsigned i, tries = 10;
 	int r;
 
 	INIT_LIST_HEAD(&p->validated);
@@ -364,9 +382,81 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 	if (need_mmap_lock)
 		down_read(&current->mm->mmap_sem);
 
-	r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true, &duplicates);
-	if (unlikely(r != 0))
-		goto error_reserve;
+	while (1) {
+		struct list_head need_pages;
+		unsigned i;
+
+		r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
+					   &duplicates);
+		if (unlikely(r != 0))
+			goto error_free_pages;
+
+		/* Without a BO list we don't have userptr BOs */
+		if (!p->bo_list)
+			break;
+
+		INIT_LIST_HEAD(&need_pages);
+		for (i = p->bo_list->first_userptr;
+		     i < p->bo_list->num_entries; ++i) {
+
+			e = &p->bo_list->array[i];
+
+			if (amdgpu_ttm_tt_userptr_invalidated(e->robj->tbo.ttm,
+				 &e->user_invalidated) && e->user_pages) {
+
+				/* We acquired a page array, but somebody
+				 * invalidated it. Free it an try again
+				 */
+				release_pages(e->user_pages,
+					      e->robj->tbo.ttm->num_pages,
+					      false);
+				drm_free_large(e->user_pages);
+				e->user_pages = NULL;
+			}
+
+			if (e->robj->tbo.ttm->state != tt_bound &&
+			    !e->user_pages) {
+				list_del(&e->tv.head);
+				list_add(&e->tv.head, &need_pages);
+
+				amdgpu_bo_unreserve(e->robj);
+			}
+		}
+
+		if (list_empty(&need_pages))
+			break;
+
+		/* Unreserve everything again. */
+		ttm_eu_backoff_reservation(&p->ticket, &p->validated);
+
+		/* We tried to often, just abort */
+		if (!--tries) {
+			r = -EDEADLK;
+			goto error_free_pages;
+		}
+
+		/* Fill the page arrays for all useptrs. */
+		list_for_each_entry(e, &need_pages, tv.head) {
+			struct ttm_tt *ttm = e->robj->tbo.ttm;
+
+			e->user_pages = drm_calloc_large(ttm->num_pages,
+							 sizeof(struct page*));
+			if (!e->user_pages) {
+				r = -ENOMEM;
+				goto error_free_pages;
+			}
+
+			r = amdgpu_ttm_tt_get_user_pages(ttm, e->user_pages);
+			if (r) {
+				drm_free_large(e->user_pages);
+				e->user_pages = NULL;
+				goto error_free_pages;
+			}
+		}
+
+		/* And try again. */
+		list_splice(&need_pages, &p->validated);
+	}
 
 	amdgpu_vm_get_pt_bos(&fpriv->vm, &duplicates);
 
@@ -398,10 +488,26 @@ error_validate:
 		ttm_eu_backoff_reservation(&p->ticket, &p->validated);
 	}
 
-error_reserve:
+error_free_pages:
+
 	if (need_mmap_lock)
 		up_read(&current->mm->mmap_sem);
 
+	if (p->bo_list) {
+		for (i = p->bo_list->first_userptr;
+		     i < p->bo_list->num_entries; ++i) {
+			e = &p->bo_list->array[i];
+
+			if (!e->user_pages)
+				continue;
+
+			release_pages(e->user_pages,
+				      e->robj->tbo.ttm->num_pages,
+				      false);
+			drm_free_large(e->user_pages);
+		}
+	}
+
 	return r;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 2e26a517f2d6..cb27754ec07f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -274,18 +274,23 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
 
 	if (args->flags & AMDGPU_GEM_USERPTR_VALIDATE) {
 		down_read(&current->mm->mmap_sem);
+
+		r = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm,
+						 bo->tbo.ttm->pages);
+		if (r)
+			goto unlock_mmap_sem;
+
 		r = amdgpu_bo_reserve(bo, true);
-		if (r) {
-			up_read(&current->mm->mmap_sem);
-			goto release_object;
-		}
+		if (r)
+			goto free_pages;
 
 		amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
 		r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
 		amdgpu_bo_unreserve(bo);
-		up_read(&current->mm->mmap_sem);
 		if (r)
-			goto release_object;
+			goto free_pages;
+
+		up_read(&current->mm->mmap_sem);
 	}
 
 	r = drm_gem_handle_create(filp, gobj, &handle);
@@ -297,6 +302,12 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
 	args->handle = handle;
 	return 0;
 
+free_pages:
+	release_pages(bo->tbo.ttm->pages, bo->tbo.ttm->num_pages, false);
+
+unlock_mmap_sem:
+	up_read(&current->mm->mmap_sem);
+
 release_object:
 	drm_gem_object_unreference_unlocked(gobj);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 53e627970798..ed0d2e7cc57d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -508,22 +508,18 @@ struct amdgpu_ttm_tt {
 	uint32_t		userflags;
 	spinlock_t              guptasklock;
 	struct list_head        guptasks;
+	atomic_t		mmu_invalidations;
 };
 
-/* prepare the sg table with the user pages */
-static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
+int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
 {
-	struct amdgpu_device *adev = amdgpu_get_adev(ttm->bdev);
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
-	unsigned pinned = 0, nents;
-	int r;
-
 	int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
-	enum dma_data_direction direction = write ?
-		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+	unsigned pinned = 0;
+	int r;
 
 	if (gtt->userflags & AMDGPU_GEM_USERPTR_ANONONLY) {
-		/* check that we only pin down anonymous memory
+		/* check that we only use anonymous memory
 		   to prevent problems with writeback */
 		unsigned long end = gtt->userptr + ttm->num_pages * PAGE_SIZE;
 		struct vm_area_struct *vma;
@@ -536,7 +532,7 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
 	do {
 		unsigned num_pages = ttm->num_pages - pinned;
 		uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
-		struct page **pages = ttm->pages + pinned;
+		struct page **p = pages + pinned;
 		struct amdgpu_ttm_gup_task_list guptask;
 
 		guptask.task = current;
@@ -545,7 +541,7 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
 		spin_unlock(&gtt->guptasklock);
 
 		r = get_user_pages(current, current->mm, userptr, num_pages,
-				   write, 0, pages, NULL);
+				   write, 0, p, NULL);
 
 		spin_lock(&gtt->guptasklock);
 		list_del(&guptask.list);
@@ -558,6 +554,25 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
 
 	} while (pinned < ttm->num_pages);
 
+	return 0;
+
+release_pages:
+	release_pages(pages, pinned, 0);
+	return r;
+}
+
+/* prepare the sg table with the user pages */
+static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
+{
+	struct amdgpu_device *adev = amdgpu_get_adev(ttm->bdev);
+	struct amdgpu_ttm_tt *gtt = (void *)ttm;
+	unsigned nents;
+	int r;
+
+	int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
+	enum dma_data_direction direction = write ?
+		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+
 	r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0,
 				      ttm->num_pages << PAGE_SHIFT,
 				      GFP_KERNEL);
@@ -576,9 +591,6 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
 
 release_sg:
 	kfree(ttm->sg);
-
-release_pages:
-	release_pages(ttm->pages, pinned, 0);
 	return r;
 }
 
@@ -803,6 +815,7 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
 	gtt->userflags = flags;
 	spin_lock_init(&gtt->guptasklock);
 	INIT_LIST_HEAD(&gtt->guptasks);
+	atomic_set(&gtt->mmu_invalidations, 0);
 
 	return 0;
 }
@@ -840,9 +853,21 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
 	}
 	spin_unlock(&gtt->guptasklock);
 
+	atomic_inc(&gtt->mmu_invalidations);
+
 	return true;
 }
 
+bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
+				       int *last_invalidated)
+{
+	struct amdgpu_ttm_tt *gtt = (void *)ttm;
+	int prev_invalidated = *last_invalidated;
+
+	*last_invalidated = atomic_read(&gtt->mmu_invalidations);
+	return prev_invalidated != *last_invalidated;
+}
+
 bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm)
 {
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index a0896c761108..cc967b44f104 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -95,6 +95,7 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 	entry->priority = 0;
 	entry->tv.bo = &vm->page_directory->tbo;
 	entry->tv.shared = true;
+	entry->user_pages = NULL;
 	list_add(&entry->tv.head, validated);
 }
 
@@ -1186,6 +1187,7 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 		entry->priority = 0;
 		entry->tv.bo = &entry->robj->tbo;
 		entry->tv.shared = true;
+		entry->user_pages = NULL;
 		vm->page_tables[pt_idx].addr = 0;
 	}
 
-- 
cgit v1.2.3


From 568d7c764ae01f3706085ac8f0d8a8ac7e826bd7 Mon Sep 17 00:00:00 2001
From: Stephen Rothwell <sfr@canb.auug.org.au>
Date: Thu, 17 Mar 2016 15:30:49 +1100
Subject: drm/amdgpu: release_pages requires linux/pagemap.h
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Reviewed-by: Christian König <christian.koenig@amd.com.
Signed-off-by: Dave Airlie <airlied@redhat.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c  | 1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 1 +
 2 files changed, 2 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 4f5ef4149e87..9392e50a7ba4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -25,6 +25,7 @@
  *    Jerome Glisse <glisse@freedesktop.org>
  */
 #include <linux/list_sort.h>
+#include <linux/pagemap.h>
 #include <drm/drmP.h>
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 3e128c52bb87..fa6a27bff298 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -26,6 +26,7 @@
  *          Jerome Glisse
  */
 #include <linux/ktime.h>
+#include <linux/pagemap.h>
 #include <drm/drmP.h>
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
-- 
cgit v1.2.3