[PATCH] drm/i915: Skip GPU wait for scanout pin while wedged

[PATCH] drm/i915: Skip GPU wait for scanout pin while wedged

[Keith Packard]

Failing to pin a scanout buffer will most likely lead to a black
screen, so if the GPU is wedged, then just let the pin happen and hope
that things work out OK.

v2: Just ignore any error from i915_gem_object_wait_rendering, as
suggested by Chris Wilson

Signed-off-by: Keith Packard <keithp@keithp.com>
---
 drivers/gpu/drm/i915/i915_gem.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index e9d1d5c..e46f273 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3113,7 +3113,7 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj,
 
 	if (pipelined != obj->ring) {
 		ret = i915_gem_object_wait_rendering(obj);
-		if (ret)
+		if (ret == -ERESTARTSYS)
 			return ret;
 	}
 
-- 
1.7.5.4

Re: [PATCH] drm/i915: Skip GPU wait for scanout pin while wedged

[Chris Wilson]

On Tue, 19 Jul 2011 17:28:53 -0700, Keith Packard <keithp@keithp.com> wrote:
> Failing to pin a scanout buffer will most likely lead to a black
> screen, so if the GPU is wedged, then just let the pin happen and hope
> that things work out OK.
> 
> v2: Just ignore any error from i915_gem_object_wait_rendering, as
> suggested by Chris Wilson
> 
> Signed-off-by: Keith Packard <keithp@keithp.com>

I couldn't find a way to make this any simpler, or spot why it couldn't
possible work, so
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

For the sake of our sanity, we should document which paths we think can
safely ignore GPU hangs. Can we detect lockups in any other functions?
Would any of the test signals help us diagnose which link in the chain
failed?
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

[PATCH] drm/i915: Skip GPU wait for scanout pin while wedged

[Keith Packard]

Failing to pin a scanout buffer will most likely lead to a black
screen, so if the GPU is wedged, then just let the pin happen and hope
that things work out OK.

Signed-off-by: Keith Packard <keithp@keithp.com>
---
 drivers/gpu/drm/i915/i915_gem.c |   12 +++++++++---
 1 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index e9d1d5c..d8d623e 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3112,9 +3112,15 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj,
 		return ret;
 
 	if (pipelined != obj->ring) {
-		ret = i915_gem_object_wait_rendering(obj);
-		if (ret)
-			return ret;
+		struct drm_device *dev = obj->base.dev;
+		drm_i915_private_t *dev_priv = dev->dev_private;
+		if (atomic_read(&dev_priv->mm.wedged)) {
+			DRM_ERROR("Skip GPU wait for scanout while wedged\n");
+		} else {
+			ret = i915_gem_object_wait_rendering(obj);
+			if (ret)
+				return ret;
+		}
 	}
 
 	/* The display engine is not coherent with the LLC cache on gen6.  As
-- 
1.7.5.4

Re: [PATCH] drm/i915: Skip GPU wait for scanout pin while wedged

[Chris Wilson]

On Tue, 19 Jul 2011 16:26:43 -0700, Keith Packard <keithp@keithp.com> wrote:
> Failing to pin a scanout buffer will most likely lead to a black
> screen, so if the GPU is wedged, then just let the pin happen and hope
> that things work out OK.

This doesn't prevent us returning an error should the wait-rendering abort
due to a GPU hang occurring in the middle of the wait.

i915_gem_object_pin_to_display_plane is used for more than just the
scanout, cursors and overlay are also included, otherwise we could just
ignore any errors. So,

/* Update the display engine regardless of any GPU hangs */
if (ret == -ERESTARTSYS)
	return ret;

-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: [PATCH] drm/i915: Skip GPU wait for scanout pin while wedged

[Keith Packard]

[-- Attachment #1.1: Type: text/plain, Size: 767 bytes --]

On Wed, 20 Jul 2011 01:03:17 +0100, Chris Wilson <chris@chris-wilson.co.uk> wrote:

> This doesn't prevent us returning an error should the wait-rendering abort
> due to a GPU hang occurring in the middle of the wait.

Yeah, should probably check the return value and ignore the error instead.

> i915_gem_object_pin_to_display_plane is used for more than just the
> scanout, cursors and overlay are also included, otherwise we could just
> ignore any errors. So,

Given that the GPU is wedged, I don't see any problem pinning any of
these instead of failing.

> /* Update the display engine regardless of any GPU hangs */
> if (ret == -ERESTARTSYS)
> 	return ret;

Yeah, that seems simpler in all ways. Much nicer.

-- 
keith.packard@intel.com

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 159 bytes --]

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx