drm/i915: Defer reporting EIO until we try to use the GPU

Instead of reporting EIO upfront in the entrance of an ioctl that may or
may not attempt to use the GPU, defer the actual detection of an invalid
ioctl to when we issue a GPU instruction. This allows us to continue to
use bo in video memory (via pread/pwrite and mmap) after the GPU has hung.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

drivers/gpu/drm/i915/i915_drv.h

@@ -1052,7 +1052,6 @@ extern void i915_mem_takedown(struct mem_block **heap);
 extern void i915_mem_release(struct drm_device * dev,
 			     struct drm_file *file_priv, struct mem_block *heap);
 /* i915_gem.c */
-int i915_gem_check_is_wedged(struct drm_device *dev);
 int i915_gem_init_ioctl(struct drm_device *dev, void *data,
 			struct drm_file *file_priv);
 int i915_gem_create_ioctl(struct drm_device *dev, void *data,

drivers/gpu/drm/i915/i915_gem.c

@@ -75,8 +75,8 @@ static void i915_gem_info_remove_obj(struct drm_i915_private *dev_priv,
 	dev_priv->mm.object_memory -= size;
 }
 
-int
-i915_gem_check_is_wedged(struct drm_device *dev)
+static int
+i915_gem_wait_for_error(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct completion *x = &dev_priv->error_completion;
@@ -90,27 +90,24 @@ i915_gem_check_is_wedged(struct drm_device *dev)
 	if (ret)
 		return ret;
 
-	/* Success, we reset the GPU! */
-	if (!atomic_read(&dev_priv->mm.wedged))
-		return 0;
-
-	/* GPU is hung, bump the completion count to account for
-	 * the token we just consumed so that we never hit zero and
-	 * end up waiting upon a subsequent completion event that
-	 * will never happen.
-	 */
-	spin_lock_irqsave(&x->wait.lock, flags);
-	x->done++;
-	spin_unlock_irqrestore(&x->wait.lock, flags);
-	return -EIO;
+	if (atomic_read(&dev_priv->mm.wedged)) {
+		/* GPU is hung, bump the completion count to account for
+		 * the token we just consumed so that we never hit zero and
+		 * end up waiting upon a subsequent completion event that
+		 * will never happen.
+		 */
+		spin_lock_irqsave(&x->wait.lock, flags);
+		x->done++;
+		spin_unlock_irqrestore(&x->wait.lock, flags);
+	}
+	return 0;
 }
 
 int i915_mutex_lock_interruptible(struct drm_device *dev)
 {
-	struct drm_i915_private *dev_priv = dev->dev_private;
 	int ret;
 
-	ret = i915_gem_check_is_wedged(dev);
+	ret = i915_gem_wait_for_error(dev);
 	if (ret)
 		return ret;
 
@@ -118,11 +115,6 @@ int i915_mutex_lock_interruptible(struct drm_device *dev)
 	if (ret)
 		return ret;
 
-	if (atomic_read(&dev_priv->mm.wedged)) {
-		mutex_unlock(&dev->struct_mutex);
-		return -EAGAIN;
-	}
-
 	WARN_ON(i915_verify_lists(dev));
 	return 0;
 }

drivers/gpu/drm/i915/i915_gem_tiling.c

@@ -284,11 +284,6 @@ i915_gem_set_tiling(struct drm_device *dev, void *data,
 	struct drm_i915_gem_set_tiling *args = data;
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_i915_gem_object *obj;
-	int ret;
-
-	ret = i915_gem_check_is_wedged(dev);
-	if (ret)
-		return ret;
 
 	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
 	if (obj == NULL)

drivers/gpu/drm/i915/intel_ringbuffer.c

@@ -980,9 +980,13 @@ int intel_wait_ring_buffer(struct intel_ring_buffer *ring, int n)
 int intel_ring_begin(struct intel_ring_buffer *ring,
 		     int num_dwords)
 {
+	struct drm_i915_private *dev_priv = ring->dev->dev_private;
 	int n = 4*num_dwords;
 	int ret;
 
+	if (unlikely(atomic_read(&dev_priv->mm.wedged)))
+		return -EIO;
+
 	if (unlikely(ring->tail + n > ring->effective_size)) {
 		ret = intel_wrap_ring_buffer(ring);
 		if (unlikely(ret))