[PATCH] [v2] drm/i915/bdw: Always issue a force restore

[PATCH] [v2] drm/i915/bdw: Always issue a force restore

[Ben Widawsky]

The PDPs seem to get screwed up otherwise, specifically PDP0. I am not
really clear why this is required, it just works with full PPGTT.

v2: Only do it for gen8, to limit regression potential

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78891
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78934
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78935
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78936
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78937

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
---
 drivers/gpu/drm/i915/i915_gem_context.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index 3ffe308..c5330ab 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -568,6 +568,9 @@ mi_set_context(struct intel_engine_cs *ring,
 	else
 		intel_ring_emit(ring, MI_NOOP);
 
+	if (INTEL_INFO(ring->dev)->gen == 8)
+		hw_flags |= MI_FORCE_RESTORE;
+
 	intel_ring_emit(ring, MI_NOOP);
 	intel_ring_emit(ring, MI_SET_CONTEXT);
 	intel_ring_emit(ring, i915_gem_obj_ggtt_offset(new_context->obj) |
-- 
1.9.3

[PATCH] [v3] drm/i915/bdw: Always issue a force restore

[Ben Widawsky]

The PDPs seem to get screwed up otherwise, specifically PDP0. I am not
really clear why this is required, it just works with full PPGTT.

v2: Only do it for gen8, to limit regression potential

v3: Fix the bugzilla links

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78891
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78935
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78936
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78937
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78938

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
---
 drivers/gpu/drm/i915/i915_gem_context.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index 3ffe308..c5330ab 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -568,6 +568,9 @@ mi_set_context(struct intel_engine_cs *ring,
 	else
 		intel_ring_emit(ring, MI_NOOP);
 
+	if (INTEL_INFO(ring->dev)->gen == 8)
+		hw_flags |= MI_FORCE_RESTORE;
+
 	intel_ring_emit(ring, MI_NOOP);
 	intel_ring_emit(ring, MI_SET_CONTEXT);
 	intel_ring_emit(ring, i915_gem_obj_ggtt_offset(new_context->obj) |
-- 
1.9.3

Re: [PATCH] [v3] drm/i915/bdw: Always issue a force restore

[Ben Widawsky]

On Wed, May 28, 2014 at 04:06:58PM -0700, Ben Widawsky wrote:
> The PDPs seem to get screwed up otherwise, specifically PDP0. I am not
> really clear why this is required, it just works with full PPGTT.
> 
> v2: Only do it for gen8, to limit regression potential
> 
> v3: Fix the bugzilla links

I should note there is a problem with this patch. Without full PPGTT, we
also do a restore inhibit on the default context, and therefore any test
which is running and not using contexts (like everything in IGT) is
programming restore inhibit as well as force restore, which as has been
pointed out in the semaphore series is weird at best.

All I know is the patch fixed at least a few of the bugs mentioned
below. I guess we'll figure out what to do when the QA results come in.

[snip]

-- 
Ben Widawsky, Intel Open Source Technology Center

Re: [PATCH] [v3] drm/i915/bdw: Always issue a force restore

[Ben Widawsky]

On Wed, May 28, 2014 at 10:54:21PM -0700, Ben Widawsky wrote:
> On Wed, May 28, 2014 at 04:06:58PM -0700, Ben Widawsky wrote:
> > The PDPs seem to get screwed up otherwise, specifically PDP0. I am not
> > really clear why this is required, it just works with full PPGTT.
> > 
> > v2: Only do it for gen8, to limit regression potential
> > 
> > v3: Fix the bugzilla links
> 
> I should note there is a problem with this patch. Without full PPGTT, we
> also do a restore inhibit on the default context, and therefore any test
> which is running and not using contexts (like everything in IGT) is
> programming restore inhibit as well as force restore, which as has been
> pointed out in the semaphore series is weird at best.
> 
> All I know is the patch fixed at least a few of the bugs mentioned
> below. I guess we'll figure out what to do when the QA results come in.
> 
> [snip]
> 

Okay, now I am really confused. I've changed machines and none of these
tests are passing for me now. Until I figured out what happened, please
ignore this thread. Sorry for the noise.


-- 
Ben Widawsky, Intel Open Source Technology Center