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>
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 9185f09..a728309 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -61,6 +61,37 @@
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 @@
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 @@
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 @@
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 @@
if (atomic_read(&dev_priv->mm.wedged)) {
mutex_unlock(&dev->struct_mutex);
- ret = -EIO;
+ ret = -EAGAIN;
goto pre_mutex_err;
}
@@ -4126,6 +4161,10 @@
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 @@
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 @@
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 @@
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 @@
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);