[Intel-gfx] [PATCH] drm/i915: Repeat unbinding during free if interrupted (v3)

[Chris Wilson]

If during the freeing of an object the unbind is interrupted by a system
call, which is quite possible if we have outstanding GPU writes that
must be flushed, the unbind is silently aborted. This still leaves the
AGP region and backing pages allocated, and perhaps more importantly,
the object remains upon the various lists exposing us to memory
corruption.

I think this is the cause behind the use-after-free, such as

  Bug 15664 - Graphics hang and kernel backtrace when starting Azureus
              with Compiz enabled
  https://bugzilla.kernel.org/show_bug.cgi?id=15664

v2: Daniel Vetter reminded me that kernel space programming is never easy.
We cannot simply spin to clear the pending signal and so must deferred
the freeing of the object until later.
v3: Run from the top level retire requests.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: stable at kernel.org
---
 drivers/gpu/drm/i915/i915_drv.h |    8 +++++
 drivers/gpu/drm/i915/i915_gem.c |   56 ++++++++++++++++++++++++++++----------
 2 files changed, 49 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 0e7bf85..a66503c 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -551,6 +551,14 @@ typedef struct drm_i915_private {
 		struct list_head fence_list;
 
 		/**
+		 * List of objects currently pending being freed.
+		 *
+		 * These objects are no longer in use, but due to a signal
+		 * we were prevented from freeing them at the appointed time.
+		 */
+		struct list_head deferred_free_list;
+
+		/**
 		 * We leave the user IRQ off as much as possible,
 		 * but this means that requests will finish and never
 		 * be retired once the system goes idle. Set a timer to
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 3c7c0f7..1f2c5f5 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -52,6 +52,7 @@ static void i915_gem_clear_fence_reg(struct drm_gem_object *obj);
 static int i915_gem_phys_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
 				struct drm_i915_gem_pwrite *args,
 				struct drm_file *file_priv);
+static void i915_gem_free_object_tail(struct drm_gem_object *obj);
 
 static LIST_HEAD(shrink_list);
 static DEFINE_SPINLOCK(shrink_list_lock);
@@ -1746,6 +1747,15 @@ i915_gem_retire_requests(struct drm_device *dev)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
 
+	if (!list_empty(&dev_priv->mm.deferred_free_list)) {
+	    struct drm_i915_gem_object *obj_priv, *tmp;
+
+	    list_for_each_entry_safe(obj_priv, tmp,
+				     &dev_priv->mm.deferred_free_list,
+				     list)
+		i915_gem_free_object_tail(&obj_priv->base);
+	}
+
 	i915_gem_retire_requests_ring(dev, &dev_priv->render_ring);
 	if (HAS_BSD(dev))
 		i915_gem_retire_requests_ring(dev, &dev_priv->bsd_ring);
@@ -1929,11 +1939,12 @@ i915_gem_object_unbind(struct drm_gem_object *obj)
 	 * before we unbind.
 	 */
 	ret = i915_gem_object_set_to_cpu_domain(obj, 1);
-	if (ret) {
-		if (ret != -ERESTARTSYS)
-			DRM_ERROR("set_domain failed: %d\n", ret);
+	if (ret == -ERESTARTSYS)
 		return ret;
-	}
+	/* Continue on if we fail due to EIO, the GPU is hung so we
+	 * should be safe and we need to cleanup or else we might
+	 * cause memory corruption through use-after-free.
+	 */
 
 	BUG_ON(obj_priv->active);
 
@@ -1967,7 +1978,7 @@ i915_gem_object_unbind(struct drm_gem_object *obj)
 
 	trace_i915_gem_object_unbind(obj);
 
-	return 0;
+	return ret;
 }
 
 int
@@ -4288,20 +4299,19 @@ int i915_gem_init_object(struct drm_gem_object *obj)
 	return 0;
 }
 
-void i915_gem_free_object(struct drm_gem_object *obj)
+static void i915_gem_free_object_tail(struct drm_gem_object *obj)
 {
 	struct drm_device *dev = obj->dev;
+	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
+	int ret;
 
-	trace_i915_gem_object_destroy(obj);
-
-	while (obj_priv->pin_count > 0)
-		i915_gem_object_unpin(obj);
-
-	if (obj_priv->phys_obj)
-		i915_gem_detach_phys_object(dev, obj);
-
-	i915_gem_object_unbind(obj);
+	ret = i915_gem_object_unbind(obj);
+	if (ret == -ERESTARTSYS) {
+		list_move_tail(&obj_priv->list,
+			       &dev_priv->mm.deferred_free_list);
+		return;
+	}
 
 	if (obj_priv->mmap_offset)
 		i915_gem_free_mmap_offset(obj);
@@ -4313,6 +4323,22 @@ void i915_gem_free_object(struct drm_gem_object *obj)
 	kfree(obj_priv);
 }
 
+void i915_gem_free_object(struct drm_gem_object *obj)
+{
+	struct drm_device *dev = obj->dev;
+	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
+
+	trace_i915_gem_object_destroy(obj);
+
+	while (obj_priv->pin_count > 0)
+		i915_gem_object_unpin(obj);
+
+	if (obj_priv->phys_obj)
+		i915_gem_detach_phys_object(dev, obj);
+
+	i915_gem_free_object_tail(obj);
+}
+
 int
 i915_gem_idle(struct drm_device *dev)
 {
-- 
1.7.1