[PATCH 43/43] drm/i915: Move object close under its own lock

[Chris Wilson <chris@chris-wilson.co.uk>]

Use i915_gem_object_lock() to guard the LUT and active reference to
allow us to break free of struct_mutex for handling GEM_CLOSE.

Testcase: igt/gem_close_race
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/gem/i915_gem_context.c   | 15 +++---
 .../gpu/drm/i915/gem/i915_gem_context_types.h |  6 ---
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 22 +++++---
 drivers/gpu/drm/i915/gem/i915_gem_object.c    | 51 +++++++++++++------
 .../gpu/drm/i915/gem/i915_gem_object_types.h  |  1 -
 .../gpu/drm/i915/gem/selftests/mock_context.c |  1 -
 drivers/gpu/drm/i915/i915_drv.h               |  4 +-
 drivers/gpu/drm/i915/i915_gem.c               |  1 +
 drivers/gpu/drm/i915/i915_vma.c               | 41 ++++++++++-----
 drivers/gpu/drm/i915/i915_vma.h               | 15 +++---
 .../gpu/drm/i915/selftests/mock_gem_device.c  |  1 +
 11 files changed, 97 insertions(+), 61 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
index d5bddfd3f42d..dac50cf67468 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
@@ -112,23 +112,17 @@ void i915_lut_handle_free(struct i915_lut_handle *lut)
 
 static void lut_close(struct i915_gem_context *ctx)
 {
-	struct i915_lut_handle *lut, *ln;
 	struct radix_tree_iter iter;
 	void __rcu **slot;
 
-	list_for_each_entry_safe(lut, ln, &ctx->handles_list, ctx_link) {
-		list_del(&lut->obj_link);
-		i915_lut_handle_free(lut);
-	}
+	lockdep_assert_held(&ctx->mutex);
 
 	rcu_read_lock();
 	radix_tree_for_each_slot(slot, &ctx->handles_vma, &iter, 0) {
 		struct i915_vma *vma = rcu_dereference_raw(*slot);
 
 		radix_tree_iter_delete(&ctx->handles_vma, &iter, slot);
-
 		vma->open_count--;
-		i915_vma_put(vma);
 	}
 	rcu_read_unlock();
 }
@@ -311,6 +305,8 @@ void i915_gem_context_release(struct kref *ref)
 
 static void context_close(struct i915_gem_context *ctx)
 {
+	mutex_lock(&ctx->mutex);
+
 	i915_gem_context_set_closed(ctx);
 
 	/*
@@ -329,6 +325,8 @@ static void context_close(struct i915_gem_context *ctx)
 		i915_ppgtt_close(ctx->ppgtt);
 
 	ctx->file_priv = ERR_PTR(-EBADF);
+
+	mutex_unlock(&ctx->mutex);
 	i915_gem_context_put(ctx);
 }
 
@@ -379,7 +377,6 @@ __create_hw_context(struct drm_i915_private *dev_priv,
 	spin_lock_init(&ctx->hw_contexts_lock);
 
 	INIT_RADIX_TREE(&ctx->handles_vma, GFP_KERNEL);
-	INIT_LIST_HEAD(&ctx->handles_list);
 	INIT_LIST_HEAD(&ctx->hw_id_link);
 
 	/* Default context will never have a file_priv */
@@ -1043,7 +1040,9 @@ static int set_ppgtt(struct i915_gem_context *ctx,
 		goto unlock;
 
 	/* Teardown the existing obj:vma cache, it will have to be rebuilt. */
+	mutex_lock(&ctx->mutex);
 	lut_close(ctx);
+	mutex_unlock(&ctx->mutex);
 
 	old = __set_ppgtt(ctx, ppgtt);
 
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context_types.h b/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
index 47ed1f65cc06..333eeacef5e3 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
@@ -177,12 +177,6 @@ struct i915_gem_context {
 	 * per vm, which may be one per context or shared with the global GTT)
 	 */
 	struct radix_tree_root handles_vma;
-
-	/** handles_list: reverse list of all the rbtree entries in use for
-	 * this context, which allows us to free all the allocations on
-	 * context close.
-	 */
-	struct list_head handles_list;
 };
 
 #endif /* __I915_GEM_CONTEXT_TYPES_H__ */
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 55817ee754b5..18f4a236540b 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -784,9 +784,6 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 	unsigned int i, batch;
 	int err;
 
-	if (unlikely(i915_gem_context_is_closed(eb->ctx)))
-		return -ENOENT;
-
 	if (unlikely(i915_gem_context_is_banned(eb->ctx)))
 		return -EIO;
 
@@ -795,6 +792,12 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 
 	batch = eb_batch_index(eb);
 
+	mutex_lock(&eb->ctx->mutex);
+	if (unlikely(i915_gem_context_is_closed(eb->ctx))) {
+		err = -ENOENT;
+		goto err_ctx;
+	}
+
 	for (i = 0; i < eb->buffer_count; i++) {
 		u32 handle = eb->exec[i].handle;
 		struct i915_lut_handle *lut;
@@ -831,10 +834,13 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 		/* transfer ref to ctx */
 		if (!vma->open_count++)
 			i915_vma_reopen(vma);
-		list_add(&lut->obj_link, &obj->lut_list);
-		list_add(&lut->ctx_link, &eb->ctx->handles_list);
-		lut->ctx = eb->ctx;
+
 		lut->handle = handle;
+		lut->ctx = i915_gem_context_get(eb->ctx);
+
+		i915_gem_object_lock(obj);
+		list_add(&lut->obj_link, &obj->lut_list);
+		i915_gem_object_unlock(obj);
 
 add_vma:
 		err = eb_add_vma(eb, i, batch, vma);
@@ -847,6 +853,8 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 			   eb_vma_misplaced(&eb->exec[i], vma, eb->flags[i]));
 	}
 
+	mutex_unlock(&eb->ctx->mutex);
+
 	eb->args->flags |= __EXEC_VALIDATED;
 	return eb_reserve(eb);
 
@@ -854,6 +862,8 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 	i915_gem_object_put(obj);
 err_vma:
 	eb->vma[i] = NULL;
+err_ctx:
+	mutex_unlock(&eb->ctx->mutex);
 	return err;
 }
 
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index d21a7095092c..7da9dbad55c9 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -125,41 +125,60 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,
 		!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE);
 }
 
+static void lut_handle_release(struct i915_lut_handle *lut)
+{
+	list_del(&lut->obj_link);
+	i915_gem_context_put(lut->ctx);
+	i915_lut_handle_free(lut);
+}
+
 void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file)
 {
-	struct drm_i915_private *i915 = to_i915(gem->dev);
 	struct drm_i915_gem_object *obj = to_intel_bo(gem);
 	struct drm_i915_file_private *fpriv = file->driver_priv;
 	struct i915_lut_handle *lut, *ln;
+	LIST_HEAD(close);
 
-	mutex_lock(&i915->drm.struct_mutex);
+	i915_gem_object_lock(obj);
 
 	list_for_each_entry_safe(lut, ln, &obj->lut_list, obj_link) {
 		struct i915_gem_context *ctx = lut->ctx;
-		struct i915_vma *vma;
 
-		GEM_BUG_ON(ctx->file_priv == ERR_PTR(-EBADF));
-		if (ctx->file_priv != fpriv)
+		/* prune any stale entries */
+		if (i915_gem_context_is_closed(ctx)) {
+			lut_handle_release(lut);
+			i915_gem_object_put(obj);
 			continue;
+		}
 
-		vma = radix_tree_delete(&ctx->handles_vma, lut->handle);
-		GEM_BUG_ON(vma->obj != obj);
+		if (ctx->file_priv == fpriv)
+			list_move(&lut->obj_link, &close);
+	}
 
-		/* We allow the process to have multiple handles to the same
+	list_for_each_entry_safe(lut, ln, &close, obj_link) {
+		struct i915_gem_context *ctx = lut->ctx;
+		struct i915_vma *vma;
+
+		i915_gem_object_unlock(obj);
+		mutex_lock(&ctx->mutex);
+		/*
+		 * We allow the process to have multiple handles to the same
 		 * vma, in the same fd namespace, by virtue of flink/open.
 		 */
-		GEM_BUG_ON(!vma->open_count);
-		if (!--vma->open_count && !i915_vma_is_ggtt(vma))
-			i915_vma_close(vma);
-
-		list_del(&lut->obj_link);
-		list_del(&lut->ctx_link);
+		vma = radix_tree_delete(&ctx->handles_vma, lut->handle);
+		if (vma) {
+			GEM_BUG_ON(!vma->open_count);
+			if (!--vma->open_count && !i915_vma_is_ggtt(vma))
+				i915_vma_close(vma);
+		}
+		mutex_unlock(&ctx->mutex);
+		i915_gem_object_lock(obj);
 
-		i915_lut_handle_free(lut);
+		lut_handle_release(lut);
 		i915_gem_object_put(obj);
 	}
 
-	mutex_unlock(&i915->drm.struct_mutex);
+	i915_gem_object_unlock(obj);
 }
 
 static bool discard_backing_storage(struct drm_i915_gem_object *obj)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
index 9f61220795a6..1d42b7681944 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
@@ -24,7 +24,6 @@ struct drm_i915_gem_object;
  */
 struct i915_lut_handle {
 	struct list_head obj_link;
-	struct list_head ctx_link;
 	struct i915_gem_context *ctx;
 	u32 handle;
 };
diff --git a/drivers/gpu/drm/i915/gem/selftests/mock_context.c b/drivers/gpu/drm/i915/gem/selftests/mock_context.c
index 7f88fd6b3ce8..c9d9969821e0 100644
--- a/drivers/gpu/drm/i915/gem/selftests/mock_context.c
+++ b/drivers/gpu/drm/i915/gem/selftests/mock_context.c
@@ -26,7 +26,6 @@ mock_context(struct drm_i915_private *i915,
 	spin_lock_init(&ctx->hw_contexts_lock);
 
 	INIT_RADIX_TREE(&ctx->handles_vma, GFP_KERNEL);
-	INIT_LIST_HEAD(&ctx->handles_list);
 	INIT_LIST_HEAD(&ctx->hw_id_link);
 	INIT_LIST_HEAD(&ctx->active_engines);
 	mutex_init(&ctx->mutex);
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index a15d7a81a330..a948fac346ae 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1983,10 +1983,12 @@ struct drm_i915_private {
 		} timelines;
 
 		struct list_head active_rings;
-		struct list_head closed_vma;
 		unsigned long active_engines;
 		u32 active_requests;
 
+		struct list_head closed_vma;
+		spinlock_t closed_lock; /* guards the list of closed_vma */
+
 		/**
 		 * Is the GPU currently considered idle, or busy executing
 		 * userspace requests? Whilst idle, we allow runtime power
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 8fcfd1795885..5512b4498142 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2078,6 +2078,7 @@ int i915_gem_init_early(struct drm_i915_private *dev_priv)
 
 	INIT_LIST_HEAD(&dev_priv->gt.active_rings);
 	INIT_LIST_HEAD(&dev_priv->gt.closed_vma);
+	spin_lock_init(&dev_priv->gt.closed_lock);
 
 	i915_gem_init__mm(dev_priv);
 
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 741ae66eeff8..9e95a572af72 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -142,6 +142,8 @@ vma_create(struct drm_i915_gem_object *obj,
 	vma->size = obj->base.size;
 	vma->display_alignment = I915_GTT_MIN_ALIGNMENT;
 
+	INIT_LIST_HEAD(&vma->closed_link);
+
 	if (view && view->type != I915_GGTT_VIEW_NORMAL) {
 		vma->ggtt_view = *view;
 		if (view->type == I915_GGTT_VIEW_PARTIAL) {
@@ -784,10 +786,9 @@ int __i915_vma_do_pin(struct i915_vma *vma,
 
 void i915_vma_close(struct i915_vma *vma)
 {
-	lockdep_assert_held(&vma->vm->i915->drm.struct_mutex);
+	struct drm_i915_private *i915 = vma->vm->i915;
 
 	GEM_BUG_ON(i915_vma_is_closed(vma));
-	vma->flags |= I915_VMA_CLOSED;
 
 	/*
 	 * We defer actually closing, unbinding and destroying the VMA until
@@ -801,17 +802,26 @@ void i915_vma_close(struct i915_vma *vma)
 	 * causing us to rebind the VMA once more. This ends up being a lot
 	 * of wasted work for the steady state.
 	 */
-	list_add_tail(&vma->closed_link, &vma->vm->i915->gt.closed_vma);
+	spin_lock(&i915->gt.closed_lock);
+	list_add(&vma->closed_link, &i915->gt.closed_vma);
+	spin_unlock(&i915->gt.closed_lock);
 }
 
-void i915_vma_reopen(struct i915_vma *vma)
+static void __i915_vma_remove_closed(struct i915_vma *vma)
 {
-	lockdep_assert_held(&vma->vm->i915->drm.struct_mutex);
+	struct drm_i915_private *i915 = vma->vm->i915;
 
-	if (vma->flags & I915_VMA_CLOSED) {
-		vma->flags &= ~I915_VMA_CLOSED;
-		list_del(&vma->closed_link);
-	}
+	if (!i915_vma_is_closed(vma))
+		return;
+
+	spin_lock(&i915->gt.closed_lock);
+	list_del_init(&vma->closed_link);
+	spin_unlock(&i915->gt.closed_lock);
+}
+
+void i915_vma_reopen(struct i915_vma *vma)
+{
+	__i915_vma_remove_closed(vma);
 }
 
 static void __i915_vma_destroy(struct i915_vma *vma)
@@ -845,8 +855,7 @@ void i915_vma_destroy(struct i915_vma *vma)
 
 	GEM_BUG_ON(i915_vma_is_pinned(vma));
 
-	if (i915_vma_is_closed(vma))
-		list_del(&vma->closed_link);
+	__i915_vma_remove_closed(vma);
 
 	WARN_ON(i915_vma_unbind(vma));
 	GEM_BUG_ON(i915_vma_is_active(vma));
@@ -858,12 +867,16 @@ void i915_vma_parked(struct drm_i915_private *i915)
 {
 	struct i915_vma *vma, *next;
 
+	spin_lock(&i915->gt.closed_lock);
 	list_for_each_entry_safe(vma, next, &i915->gt.closed_vma, closed_link) {
-		GEM_BUG_ON(!i915_vma_is_closed(vma));
+		list_del_init(&vma->closed_link);
+		spin_unlock(&i915->gt.closed_lock);
+
 		i915_vma_destroy(vma);
-	}
 
-	GEM_BUG_ON(!list_empty(&i915->gt.closed_vma));
+		spin_lock(&i915->gt.closed_lock);
+	}
+	spin_unlock(&i915->gt.closed_lock);
 }
 
 static void __i915_vma_iounmap(struct i915_vma *vma)
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 71ac7ee8620a..67bc67241c1b 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -104,10 +104,9 @@ struct i915_vma {
 
 #define I915_VMA_GGTT		BIT(11)
 #define I915_VMA_CAN_FENCE	BIT(12)
-#define I915_VMA_CLOSED		BIT(13)
-#define I915_VMA_USERFAULT_BIT	14
+#define I915_VMA_USERFAULT_BIT	13
 #define I915_VMA_USERFAULT	BIT(I915_VMA_USERFAULT_BIT)
-#define I915_VMA_GGTT_WRITE	BIT(15)
+#define I915_VMA_GGTT_WRITE	BIT(14)
 
 	struct i915_active active;
 	struct i915_active_request last_fence;
@@ -190,11 +189,6 @@ static inline bool i915_vma_is_map_and_fenceable(const struct i915_vma *vma)
 	return vma->flags & I915_VMA_CAN_FENCE;
 }
 
-static inline bool i915_vma_is_closed(const struct i915_vma *vma)
-{
-	return vma->flags & I915_VMA_CLOSED;
-}
-
 static inline bool i915_vma_set_userfault(struct i915_vma *vma)
 {
 	GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma));
@@ -211,6 +205,11 @@ static inline bool i915_vma_has_userfault(const struct i915_vma *vma)
 	return test_bit(I915_VMA_USERFAULT_BIT, &vma->flags);
 }
 
+static inline bool i915_vma_is_closed(const struct i915_vma *vma)
+{
+	return !list_empty(&vma->closed_link);
+}
+
 static inline u32 i915_ggtt_offset(const struct i915_vma *vma)
 {
 	GEM_BUG_ON(!i915_vma_is_ggtt(vma));
diff --git a/drivers/gpu/drm/i915/selftests/mock_gem_device.c b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
index 930f9cb1661f..37c9a5706f69 100644
--- a/drivers/gpu/drm/i915/selftests/mock_gem_device.c
+++ b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
@@ -206,6 +206,7 @@ struct drm_i915_private *mock_gem_device(void)
 
 	INIT_LIST_HEAD(&i915->gt.active_rings);
 	INIT_LIST_HEAD(&i915->gt.closed_vma);
+	spin_lock_init(&i915->gt.closed_lock);
 
 	mutex_lock(&i915->drm.struct_mutex);
 
-- 
2.20.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx