drm/i915: Adjust hangcheck EIO semantics

Owain Ainsworth reported an issue between the interaction of the
hangcheck and userspace immediately (and permanently) falling back to
s/w rasterisation. In order to break the mutex and begin resetting the
GPU, we must abort the current operation (usually within the wait) and
climb sufficiently far back up the call chain to drop the mutex. In his
implementation, Owain has a loop within the ioctl handler to detect the
hang and then sleep until the error handler has run. I've chosen to
return to userspace and report an EAGAIN which should trigger the
userspace ioctl handler to repeat the call (simply because it felt less
invasive...). Before hitting a wedged GPU, we then wait upon completion
of the error handler.

Reported-by: Owain G. Ainsworth <zerooa@googlemail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

4 files changed, 68 insertions, 8 deletions

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 2611e85bdd3d..f2ff258cdfd5 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -363,6 +363,7 @@ typedef struct drm_i915_private {
 	spinlock_t error_lock;
 	struct drm_i915_error_state *first_error;
 	struct work_struct error_work;
+	struct completion error_completion;
 	struct workqueue_struct *wq;
 
 	/* Display functions */
@@ -957,6 +958,7 @@ 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,
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 9185f098822d..a7283092c233 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -61,6 +61,37 @@ static void i915_gem_free_object_tail(struct drm_gem_object *obj);
 static LIST_HEAD(shrink_list);
 static DEFINE_SPINLOCK(shrink_list_lock);
 
+int
+i915_gem_check_is_wedged(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct completion *x = &dev_priv->error_completion;
+	unsigned long flags;
+	int ret;
+
+	if (!atomic_read(&dev_priv->mm.wedged))
+		return 0;
+
+	ret = wait_for_completion_interruptible(x);
+	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;
+}
+
+
 static inline bool
 i915_gem_object_is_inactive(struct drm_i915_gem_object *obj_priv)
 {
@@ -1848,15 +1879,15 @@ i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
 
 	BUG_ON(seqno == 0);
 
+	if (atomic_read(&dev_priv->mm.wedged))
+		return -EAGAIN;
+
 	if (seqno == dev_priv->next_seqno) {
 		seqno = i915_add_request(dev, NULL, NULL, ring);
 		if (seqno == 0)
 			return -ENOMEM;
 	}
 
-	if (atomic_read(&dev_priv->mm.wedged))
-		return -EIO;
-
 	if (!i915_seqno_passed(ring->get_seqno(dev, ring), seqno)) {
 		if (HAS_PCH_SPLIT(dev))
 			ier = I915_READ(DEIER) | I915_READ(GTIER);
@@ -1890,7 +1921,7 @@ i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
 		trace_i915_gem_request_wait_end(dev, seqno);
 	}
 	if (atomic_read(&dev_priv->mm.wedged))
-		ret = -EIO;
+		ret = -EAGAIN;
 
 	if (ret && ret != -ERESTARTSYS)
 		DRM_ERROR("%s returns %d (awaiting %d at %d, next %d)\n",
@@ -3569,13 +3600,17 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
 	struct drm_clip_rect *cliprects = NULL;
 	struct drm_i915_gem_relocation_entry *relocs = NULL;
 	struct drm_i915_gem_request *request = NULL;
-	int ret = 0, ret2, i, pinned = 0;
+	int ret, ret2, i, pinned = 0;
 	uint64_t exec_offset;
 	uint32_t reloc_index;
 	int pin_tries, flips;
 
 	struct intel_ring_buffer *ring = NULL;
 
+	ret = i915_gem_check_is_wedged(dev);
+	if (ret)
+		return ret;
+
 #if WATCH_EXEC
 	DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n",
 		  (int) args->buffers_ptr, args->buffer_count, args->batch_len);
@@ -3639,7 +3674,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
 
 	if (atomic_read(&dev_priv->mm.wedged)) {
 		mutex_unlock(&dev->struct_mutex);
-		ret = -EIO;
+		ret = -EAGAIN;
 		goto pre_mutex_err;
 	}
 
@@ -4126,6 +4161,10 @@ i915_gem_pin_ioctl(struct drm_device *dev, void *data,
 	struct drm_i915_gem_object *obj_priv;
 	int ret;
 
+	ret = i915_gem_check_is_wedged(dev);
+	if (ret)
+		return ret;
+
 	mutex_lock(&dev->struct_mutex);
 
 	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
@@ -4215,9 +4254,15 @@ int
 i915_gem_busy_ioctl(struct drm_device *dev, void *data,
 		    struct drm_file *file_priv)
 {
+	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct drm_i915_gem_busy *args = data;
 	struct drm_gem_object *obj;
 	struct drm_i915_gem_object *obj_priv;
+	int ret;
+
+	ret = i915_gem_check_is_wedged(dev);
+	if (ret)
+		return ret;
 
 	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
 	if (obj == NULL) {
@@ -4228,6 +4273,11 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data,
 
 	mutex_lock(&dev->struct_mutex);
 
+	if (atomic_read(&dev_priv->mm.wedged)) {
+		ret = -EAGAIN;
+		goto unlock;
+	}
+
 	/* Count all active objects as busy, even if they are currently not used
 	 * by the gpu. Users of this interface expect objects to eventually
 	 * become non-busy without any further actions, therefore emit any
@@ -4256,9 +4306,10 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data,
 		args->busy = obj_priv->active;
 	}
 
+unlock:
 	drm_gem_object_unreference(obj);
 	mutex_unlock(&dev->struct_mutex);
-	return 0;
+	return ret;
 }
 
 int
@@ -4643,6 +4694,7 @@ i915_gem_load(struct drm_device *dev)
 		INIT_LIST_HEAD(&dev_priv->fence_regs[i].lru_list);
 	INIT_DELAYED_WORK(&dev_priv->mm.retire_work,
 			  i915_gem_retire_work_handler);
+	init_completion(&dev_priv->error_completion);
 	spin_lock(&shrink_list_lock);
 	list_add(&dev_priv->mm.shrink_list, &shrink_list);
 	spin_unlock(&shrink_list_lock);
diff --git a/drivers/gpu/drm/i915/i915_gem_tiling.c b/drivers/gpu/drm/i915/i915_gem_tiling.c
index b09b157f6ada..8c9ffc4768ee 100644
--- a/drivers/gpu/drm/i915/i915_gem_tiling.c
+++ b/drivers/gpu/drm/i915/i915_gem_tiling.c
@@ -273,7 +273,11 @@ i915_gem_set_tiling(struct drm_device *dev, void *data,
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_gem_object *obj;
 	struct drm_i915_gem_object *obj_priv;
-	int ret = 0;
+	int ret;
+
+	ret = i915_gem_check_is_wedged(dev);
+	if (ret)
+		return ret;
 
 	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
 	if (obj == NULL)
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 245a07e6f1a4..aaa0f1b9d6e1 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -406,6 +406,7 @@ static void i915_error_work_func(struct work_struct *work)
 			atomic_set(&dev_priv->mm.wedged, 0);
 			kobject_uevent_env(&dev->primary->kdev.kobj, KOBJ_CHANGE, reset_done_event);
 		}
+		complete_all(&dev_priv->error_completion);
 	}
 }
 
@@ -869,6 +870,7 @@ static void i915_handle_error(struct drm_device *dev, bool wedged)
 	i915_report_and_clear_eir(dev);
 
 	if (wedged) {
+		INIT_COMPLETION(dev_priv->error_completion);
 		atomic_set(&dev_priv->mm.wedged, 1);
 
 		/*