[Intel-gfx] [PATCH] drm/i915: Avoid promoting a simulated hang to 'wedged'

[Chris Wilson]

It appears that a beneficial side-effect of Mika's more accurate hangman
work is to speed up hang detection and execution. This exposes a bug in
the reset code that then treats repeated simulated hangs as an
indication that the machine is wedged. Jiggle the code around so that we
only do the simulation processing from the hangcheck and avoid confusing
it with a real hang.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=65060
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.c |   55 ++++++++++++++++-----------------------
 1 file changed, 23 insertions(+), 32 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index a1a936f..af22450 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -839,37 +839,14 @@ static int gen6_do_reset(struct drm_device *dev)
 
 int intel_gpu_reset(struct drm_device *dev)
 {
-	struct drm_i915_private *dev_priv = dev->dev_private;
-	int ret = -ENODEV;
-
 	switch (INTEL_INFO(dev)->gen) {
 	case 7:
-	case 6:
-		ret = gen6_do_reset(dev);
-		break;
-	case 5:
-		ret = ironlake_do_reset(dev);
-		break;
-	case 4:
-		ret = i965_do_reset(dev);
-		break;
-	case 2:
-		ret = i8xx_do_reset(dev);
-		break;
-	}
-
-	/* Also reset the gpu hangman. */
-	if (dev_priv->gpu_error.stop_rings) {
-		DRM_INFO("Simulated gpu hang, resetting stop_rings\n");
-		dev_priv->gpu_error.stop_rings = 0;
-		if (ret == -ENODEV) {
-			DRM_ERROR("Reset not implemented, but ignoring "
-				  "error for simulated gpu hangs\n");
-			ret = 0;
-		}
+	case 6: return gen6_do_reset(dev);
+	case 5: return ironlake_do_reset(dev);
+	case 4: return i965_do_reset(dev);
+	case 2: return i8xx_do_reset(dev);
+	default: return -ENODEV;
 	}
-
-	return ret;
 }
 
 /**
@@ -890,6 +867,7 @@ int intel_gpu_reset(struct drm_device *dev)
 int i915_reset(struct drm_device *dev)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
+	bool simulated;
 	int ret;
 
 	if (!i915_try_reset)
@@ -899,13 +877,26 @@ int i915_reset(struct drm_device *dev)
 
 	i915_gem_reset(dev);
 
-	ret = -ENODEV;
-	if (get_seconds() - dev_priv->gpu_error.last_reset < 5)
+	simulated = dev_priv->gpu_error.stop_rings != 0;
+
+	if (!simulated && get_seconds() - dev_priv->gpu_error.last_reset < 5) {
 		DRM_ERROR("GPU hanging too fast, declaring wedged!\n");
-	else
+		ret = -ENODEV;
+	} else {
 		ret = intel_gpu_reset(dev);
 
-	dev_priv->gpu_error.last_reset = get_seconds();
+		/* Also reset the gpu hangman. */
+		if (simulated) {
+			DRM_INFO("Simulated gpu hang, resetting stop_rings\n");
+			dev_priv->gpu_error.stop_rings = 0;
+			if (ret == -ENODEV) {
+				DRM_ERROR("Reset not implemented, but ignoring "
+					  "error for simulated gpu hangs\n");
+				ret = 0;
+			}
+		} else
+			dev_priv->gpu_error.last_reset = get_seconds();
+	}
 	if (ret) {
 		DRM_ERROR("Failed to reset chip.\n");
 		mutex_unlock(&dev->struct_mutex);
-- 
1.7.10.4