handle error completions the same way as FreeBSD

4 files changed, 66 insertions, 8 deletions

diff --git a/sys/dev/pci/drm/i915_drv.c b/sys/dev/pci/drm/i915_drv.c
index 89a6b71776e..d3681b07293 100644
--- a/sys/dev/pci/drm/i915_drv.c
+++ b/sys/dev/pci/drm/i915_drv.c
@@ -1000,6 +1000,7 @@ inteldrm_attach(struct device *parent, struct device *self, void *aux)
 	timeout_set(&dev_priv->hangcheck_timer, i915_hangcheck_elapsed, dev_priv);
 	dev_priv->next_seqno = 1;
 	dev_priv->mm.suspended = 1;
+	dev_priv->error_completion = 0;
 
 	/* On GEN3 we really need to make sure the ARB C3 LP bit is set */
 	if (IS_GEN3(dev)) {
@@ -1068,6 +1069,7 @@ inteldrm_attach(struct device *parent, struct device *self, void *aux)
 	mtx_init(&dev_priv->rps.lock, IPL_NONE);
 	mtx_init(&dev_priv->dpio_lock, IPL_NONE);
 	mtx_init(&mchdev_lock, IPL_NONE);
+	mtx_init(&dev_priv->error_completion_lock, IPL_NONE);
 
 	rw_init(&dev_priv->rps.hw_lock, "rpshw");
 
diff --git a/sys/dev/pci/drm/i915_drv.h b/sys/dev/pci/drm/i915_drv.h
index 2eaf075a67b..c10f5e59c81 100644
--- a/sys/dev/pci/drm/i915_drv.h
+++ b/sys/dev/pci/drm/i915_drv.h
@@ -776,6 +776,9 @@ struct inteldrm_softc {
 
 	unsigned int fsb_freq, mem_freq, is_ddr3;
 
+	int error_completion;
+	struct mutex error_completion_lock;
+
 	time_t last_gpu_reset;
 
 	struct intel_fbdev *fbdev;
diff --git a/sys/dev/pci/drm/i915_gem.c b/sys/dev/pci/drm/i915_gem.c
index 1dc1f4c2e2d..ef34c19a734 100644
--- a/sys/dev/pci/drm/i915_gem.c
+++ b/sys/dev/pci/drm/i915_gem.c
@@ -81,6 +81,7 @@ int i915_gem_check_olr(struct intel_ring_buffer *, u32);
 void i915_gem_object_truncate(struct drm_i915_gem_object *obj);
 int i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,
 				unsigned alignment, bool map_and_fenceable);
+int i915_gem_wait_for_error(struct drm_device *);
 
 extern int ticks;
 
@@ -99,18 +100,48 @@ i915_gem_object_fence_lost(struct drm_i915_gem_object *obj)
 
 // i915_gem_info_add_obj
 // i915_gem_info_remove_obj
-// i915_gem_wait_for_error
+
+int
+i915_gem_wait_for_error(struct drm_device *dev)
+{
+	drm_i915_private_t *dev_priv= dev->dev_private;
+	int ret;
+
+	if (!atomic_read(&dev_priv->mm.wedged))
+		return (0);
+
+	/*
+	 * Only wait 10 seconds for the gpu reset to complete to avoid hanging
+	 * userspace. If it takes that long something really bad is going on and
+	 * we should simply try to bail out and fail as gracefully as possible.
+	 */
+	mtx_enter(&dev_priv->error_completion_lock);
+	while (dev_priv->error_completion == 0) {
+		ret = -msleep(&dev_priv->error_completion,
+		    &dev_priv->error_completion_lock, PCATCH, "915wco", 10*hz);
+		if (ret != 0) {
+			mtx_leave(&dev_priv->error_completion_lock);
+			return (ret);
+		}
+	}
+	mtx_leave(&dev_priv->error_completion_lock);
+
+	if (atomic_read(&dev_priv->mm.wedged)) {
+		mtx_enter(&dev_priv->error_completion_lock);
+		dev_priv->error_completion++;
+		mtx_leave(&dev_priv->error_completion_lock);
+	}
+	return (0);
+}
 
 int
 i915_mutex_lock_interruptible(struct drm_device *dev)
 {
 	int ret;
 
-#ifdef notyet
 	ret = i915_gem_wait_for_error(dev);
 	if (ret)
 		return ret;
-#endif
 
 	ret = rw_enter(&dev->dev_lock, RW_WRITE | RW_INTR);
 	if (ret)
@@ -426,8 +457,26 @@ int
 i915_gem_check_wedge(struct inteldrm_softc *dev_priv,
 		     bool interruptible)
 {
-	if (dev_priv->mm.wedged)
-		return (EIO);
+	if (atomic_read(&dev_priv->mm.wedged) != 0) {
+		bool recovery_complete;
+
+		/* Give the error handler a chance to run. */
+		mtx_enter(&dev_priv->error_completion_lock);
+		recovery_complete = (&dev_priv->error_completion) > 0;
+		mtx_leave(&dev_priv->error_completion_lock);
+		
+		/* Non-interruptible callers can't handle -EAGAIN, hence return
+		 * -EIO unconditionally for these. */
+		if (!interruptible)
+			return -EIO;
+
+		/* Recovery complete, but still wedged means reset failure. */
+		if (recovery_complete)
+			return -EIO;
+
+		return -EAGAIN;
+	}
+
 	return 0;
 }
 
@@ -465,8 +514,9 @@ i915_wait_seqno(struct intel_ring_buffer *ring, uint32_t seqno)
 	int ret = 0;
 
 	/* Check first because poking a wedged chip is bad. */
-	if (dev_priv->mm.wedged)
-		return (EIO);
+	ret = i915_gem_check_wedge(dev_priv, dev_priv->mm.interruptible);
+	if (ret)
+		return (ret);
 
 	ret = i915_gem_check_olr(ring, seqno);
 	if (ret)
diff --git a/sys/dev/pci/drm/i915_irq.c b/sys/dev/pci/drm/i915_irq.c
index 8f4a86438fb..b670483851f 100644
--- a/sys/dev/pci/drm/i915_irq.c
+++ b/sys/dev/pci/drm/i915_irq.c
@@ -943,7 +943,10 @@ i915_error_work_func(void *arg1, void *arg2)
 			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);
+		mtx_enter(&dev_priv->error_completion_lock);
+		dev_priv->error_completion++;
+		wakeup(&dev_priv->error_completion);
+		mtx_leave(&dev_priv->error_completion_lock);
 	}
 }