Skip to contentSkip to conversation

drm/i915: don't clflush gem objects in stolen memory

PATCH In review Imre Deak 3 participants 5 messages Feb 13, 2013

Subsystems (2)
lore
Share & tools
lore

Imre Deak

author
#0 lore raw

PATCH 1/1 drm/i915: don't clflush gem objects in stolen memory

As explained by Chris Wilson gem objects in stolen memory are always coherent with the GPU so we don't need to ever flush the CPU caches for these.

This fixes a breakage - at least with the compact sg patches applied - during the resume/restore gtt mappings path, when we tried to clflush an FB object in stolen memory, but since stolen objects don't have backing pages we passed an invalid page pointer to drm_clflush_page().

---
 drivers/gpu/drm/i915/i915_gem.c |    7 +++++++
 1 file changed, 7 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +7 -0
index 8ea9df9..9cf6e84 100644
@@ -3019,6 +3019,13 @@ i915_gem_clflush_object(struct drm_i915_gem_object *obj)
 	if (obj->pages == NULL)
 		return;
 
+	/*
+	 * Stolen memory is always coherent with the GPU as it is explicitly
+	 * marked as wc by the system, or the system is cache-coherent.
+	 */
+	if (obj->stolen)
+		return;
+
 	/* If the GPU is snooping the contents of the CPU cache,
 	 * we do not need to manually clear the CPU cache lines.  However,
 	 * the caches are only snooped when the render cache is
Signature · 1 line
1.7.10.4
Reviewed-by on v1 1/1
#1 lore raw
Quoted 10 lines · Imre Deak

As explained by Chris Wilson gem objects in stolen memory are always coherent with the GPU so we don't need to ever flush the CPU caches for these.

This fixes a breakage - at least with the compact sg patches applied - during the resume/restore gtt mappings path, when we tried to clflush an FB object in stolen memory, but since stolen objects don't have backing pages we passed an invalid page pointer to drm_clflush_page().

To the best of my knowledge, this is correct:

Though the stolen framework landed in 3.8, tough call as to whether this should be in 3.8 as well given the backported clflush fix... I guess we are simply too late, so drm-intel-next +
Cc: [email protected]

Imre do you mind digging up the sha of both the introduction of stolen and the clflush of unbounded upon resume?
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
author
#2 lore raw
Quoted 15 lines · Chris Wilson
Quoted 10 lines · Imre Deak

As explained by Chris Wilson gem objects in stolen memory are always coherent with the GPU so we don't need to ever flush the CPU caches for these.

This fixes a breakage - at least with the compact sg patches applied - during the resume/restore gtt mappings path, when we tried to clflush an FB object in stolen memory, but since stolen objects don't have backing pages we passed an invalid page pointer to drm_clflush_page().

To the best of my knowledge, this is correct:

Quoted 7 lines

Though the stolen framework landed in 3.8, tough call as to whether this should be in 3.8 as well given the backported clflush fix... I guess we are simply too late, so drm-intel-next +
Cc: [email protected]

Imre do you mind digging up the sha of both the introduction of stolen and the clflush of unbounded upon resume?

Since afaics this problem relates only to _FBs_ in stolen memory the first one is:

commit 0ffb0ff283cca16f72caf29c44496d83b0c291fb

Author: Chris Wilson <[email protected]>
Date:   Thu Nov 15 11:32:27 2012 +0000

drm/i915: Allocate fbcon from stolen memory

and the second:

commit a8e93126a6f10d0a14ba8407ec112b1b3a5e2e97

Author: Chris Wilson <[email protected]>
Date:   Wed Dec 8 14:28:54 2010 +0000

drm/i915/gtt: Clear the cachelines upon resume

Required for my pineview system to not barf after resuming.

--Imre

#3 lore raw
Quoted 8 lines · Imre Deak

Since afaics this problem relates only to _FBs_ in stolen memory the first one is:

commit 0ffb0ff283cca16f72caf29c44496d83b0c291fb

Author: Chris Wilson <[email protected]>
Date:   Thu Nov 15 11:32:27 2012 +0000

drm/i915: Allocate fbcon from stolen memory

I would argue the oversight is in the general stolen enabling, not usage. Speaking as the guilty party.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#4 lore raw
In reply to Chris Wilson
Quoted 15 lines · Chris Wilson
Quoted 10 lines · Imre Deak

As explained by Chris Wilson gem objects in stolen memory are always coherent with the GPU so we don't need to ever flush the CPU caches for these.

This fixes a breakage - at least with the compact sg patches applied - during the resume/restore gtt mappings path, when we tried to clflush an FB object in stolen memory, but since stolen objects don't have backing pages we passed an invalid page pointer to drm_clflush_page().

To the best of my knowledge, this is correct:

Quoted 7 lines

Though the stolen framework landed in 3.8, tough call as to whether this should be in 3.8 as well given the backported clflush fix... I guess we are simply too late, so drm-intel-next +
Cc: [email protected]

Imre do you mind digging up the sha of both the introduction of stolen and the clflush of unbounded upon resume?

Stolen allocations seem to be only in 3.9-next, so no cc: stable. Queued up, thanks for the patch an review.
-Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch