1 files changed, 89 insertions, 60 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
index 38f739fb727b..bd67f4cb8e6c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
@@ -50,8 +50,10 @@ struct amdgpu_mn {
 	struct hlist_node	node;
 
 	/* objects protected by lock */
-	struct mutex		lock;
-	struct rb_root		objects;
+	struct rw_semaphore	lock;
+	struct rb_root_cached	objects;
+	struct mutex		read_lock;
+	atomic_t		recursion;
 };
 
 struct amdgpu_mn_node {
@@ -74,17 +76,17 @@ static void amdgpu_mn_destroy(struct work_struct *work)
 	struct amdgpu_bo *bo, *next_bo;
 
 	mutex_lock(&adev->mn_lock);
-	mutex_lock(&rmn->lock);
+	down_write(&rmn->lock);
 	hash_del(&rmn->node);
-	rbtree_postorder_for_each_entry_safe(node, next_node, &rmn->objects,
-					     it.rb) {
+	rbtree_postorder_for_each_entry_safe(node, next_node,
+					     &rmn->objects.rb_root, it.rb) {
 		list_for_each_entry_safe(bo, next_bo, &node->bos, mn_list) {
 			bo->mn = NULL;
 			list_del_init(&bo->mn_list);
 		}
 		kfree(node);
 	}
-	mutex_unlock(&rmn->lock);
+	up_write(&rmn->lock);
 	mutex_unlock(&adev->mn_lock);
 	mmu_notifier_unregister_no_release(&rmn->mn, rmn->mm);
 	kfree(rmn);
@@ -106,6 +108,53 @@ static void amdgpu_mn_release(struct mmu_notifier *mn,
 	schedule_work(&rmn->work);
 }
 
+
+/**
+ * amdgpu_mn_lock - take the write side lock for this mn
+ */
+void amdgpu_mn_lock(struct amdgpu_mn *mn)
+{
+	if (mn)
+		down_write(&mn->lock);
+}
+
+/**
+ * amdgpu_mn_unlock - drop the write side lock for this mn
+ */
+void amdgpu_mn_unlock(struct amdgpu_mn *mn)
+{
+	if (mn)
+		up_write(&mn->lock);
+}
+
+/**
+ * amdgpu_mn_read_lock - take the rmn read lock
+ *
+ * @rmn: our notifier
+ *
+ * Take the rmn read side lock.
+ */
+static void amdgpu_mn_read_lock(struct amdgpu_mn *rmn)
+{
+	mutex_lock(&rmn->read_lock);
+	if (atomic_inc_return(&rmn->recursion) == 1)
+		down_read_non_owner(&rmn->lock);
+	mutex_unlock(&rmn->read_lock);
+}
+
+/**
+ * amdgpu_mn_read_unlock - drop the rmn read lock
+ *
+ * @rmn: our notifier
+ *
+ * Drop the rmn read side lock.
+ */
+static void amdgpu_mn_read_unlock(struct amdgpu_mn *rmn)
+{
+	if (atomic_dec_return(&rmn->recursion) == 0)
+		up_read_non_owner(&rmn->lock);
+}
+
 /**
  * amdgpu_mn_invalidate_node - unmap all BOs of a node
  *
@@ -126,54 +175,13 @@ static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node,
 		if (!amdgpu_ttm_tt_affect_userptr(bo->tbo.ttm, start, end))
 			continue;
 
-		r = amdgpu_bo_reserve(bo, true);
-		if (r) {
-			DRM_ERROR("(%ld) failed to reserve user bo\n", r);
-			continue;
-		}
-
 		r = reservation_object_wait_timeout_rcu(bo->tbo.resv,
 			true, false, MAX_SCHEDULE_TIMEOUT);
 		if (r <= 0)
 			DRM_ERROR("(%ld) failed to wait for user bo\n", r);
 
-		amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
-		r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-		if (r)
-			DRM_ERROR("(%ld) failed to validate user bo\n", r);
-
-		amdgpu_bo_unreserve(bo);
-	}
-}
-
-/**
- * amdgpu_mn_invalidate_page - callback to notify about mm change
- *
- * @mn: our notifier
- * @mn: the mm this callback is about
- * @address: address of invalidate page
- *
- * Invalidation of a single page. Blocks for all BOs mapping it
- * and unmap them by move them into system domain again.
- */
-static void amdgpu_mn_invalidate_page(struct mmu_notifier *mn,
-				      struct mm_struct *mm,
-				      unsigned long address)
-{
-	struct amdgpu_mn *rmn = container_of(mn, struct amdgpu_mn, mn);
-	struct interval_tree_node *it;
-
-	mutex_lock(&rmn->lock);
-
-	it = interval_tree_iter_first(&rmn->objects, address, address);
-	if (it) {
-		struct amdgpu_mn_node *node;
-
-		node = container_of(it, struct amdgpu_mn_node, it);
-		amdgpu_mn_invalidate_node(node, address, address);
+		amdgpu_ttm_tt_mark_user_pages(bo->tbo.ttm);
 	}
-
-	mutex_unlock(&rmn->lock);
 }
 
 /**
@@ -198,7 +206,7 @@ static void amdgpu_mn_invalidate_range_start(struct mmu_notifier *mn,
 	/* notification is exclusive, but interval is inclusive */
 	end -= 1;
 
-	mutex_lock(&rmn->lock);
+	amdgpu_mn_read_lock(rmn);
 
 	it = interval_tree_iter_first(&rmn->objects, start, end);
 	while (it) {
@@ -209,14 +217,32 @@ static void amdgpu_mn_invalidate_range_start(struct mmu_notifier *mn,
 
 		amdgpu_mn_invalidate_node(node, start, end);
 	}
+}
 
-	mutex_unlock(&rmn->lock);
+/**
+ * amdgpu_mn_invalidate_range_end - callback to notify about mm change
+ *
+ * @mn: our notifier
+ * @mn: the mm this callback is about
+ * @start: start of updated range
+ * @end: end of updated range
+ *
+ * Release the lock again to allow new command submissions.
+ */
+static void amdgpu_mn_invalidate_range_end(struct mmu_notifier *mn,
+					   struct mm_struct *mm,
+					   unsigned long start,
+					   unsigned long end)
+{
+	struct amdgpu_mn *rmn = container_of(mn, struct amdgpu_mn, mn);
+
+	amdgpu_mn_read_unlock(rmn);
 }
 
 static const struct mmu_notifier_ops amdgpu_mn_ops = {
 	.release = amdgpu_mn_release,
-	.invalidate_page = amdgpu_mn_invalidate_page,
 	.invalidate_range_start = amdgpu_mn_invalidate_range_start,
+	.invalidate_range_end = amdgpu_mn_invalidate_range_end,
 };
 
 /**
@@ -226,7 +252,7 @@ static const struct mmu_notifier_ops amdgpu_mn_ops = {
  *
  * Creates a notifier context for current->mm.
  */
-static struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
+struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
 {
 	struct mm_struct *mm = current->mm;
 	struct amdgpu_mn *rmn;
@@ -251,8 +277,10 @@ static struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
 	rmn->adev = adev;
 	rmn->mm = mm;
 	rmn->mn.ops = &amdgpu_mn_ops;
-	mutex_init(&rmn->lock);
-	rmn->objects = RB_ROOT;
+	init_rwsem(&rmn->lock);
+	rmn->objects = RB_ROOT_CACHED;
+	mutex_init(&rmn->read_lock);
+	atomic_set(&rmn->recursion, 0);
 
 	r = __mmu_notifier_register(&rmn->mn, mm);
 	if (r)
@@ -298,7 +326,7 @@ int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr)
 
 	INIT_LIST_HEAD(&bos);
 
-	mutex_lock(&rmn->lock);
+	down_write(&rmn->lock);
 
 	while ((it = interval_tree_iter_first(&rmn->objects, addr, end))) {
 		kfree(node);
@@ -312,7 +340,7 @@ int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr)
 	if (!node) {
 		node = kmalloc(sizeof(struct amdgpu_mn_node), GFP_KERNEL);
 		if (!node) {
-			mutex_unlock(&rmn->lock);
+			up_write(&rmn->lock);
 			return -ENOMEM;
 		}
 	}
@@ -327,7 +355,7 @@ int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr)
 
 	interval_tree_insert(&node->it, &rmn->objects);
 
-	mutex_unlock(&rmn->lock);
+	up_write(&rmn->lock);
 
 	return 0;
 }
@@ -353,13 +381,13 @@ void amdgpu_mn_unregister(struct amdgpu_bo *bo)
 		return;
 	}
 
-	mutex_lock(&rmn->lock);
+	down_write(&rmn->lock);
 
 	/* save the next list entry for later */
 	head = bo->mn_list.next;
 
 	bo->mn = NULL;
-	list_del(&bo->mn_list);
+	list_del_init(&bo->mn_list);
 
 	if (list_empty(head)) {
 		struct amdgpu_mn_node *node;
@@ -368,6 +396,7 @@ void amdgpu_mn_unregister(struct amdgpu_bo *bo)
 		kfree(node);
 	}
 
-	mutex_unlock(&rmn->lock);
+	up_write(&rmn->lock);
 	mutex_unlock(&adev->mn_lock);
 }
+