Intel-GFX Archive on lore.kernel.org
 help / color / mirror / Atom feed
* A bunch of random execbuffer patches
@ 2012-12-03 11:48 Chris Wilson
  2012-12-03 11:48 ` [PATCH 01/14] drm/i915: Move the get_pages assertions up to the right layer Chris Wilson
                   ` (13 more replies)
  0 siblings, 14 replies; 22+ messages in thread
From: Chris Wilson @ 2012-12-03 11:48 UTC (permalink / raw)
  To: intel-gfx

So Daniel likes the "Tighten the checks for invalid relocation domains"
so much he wrote a i-g-t for it. Here is the patch to fix it - only I
had it inside a patch series to adjust the execbuffer API to reduce the
relocation overhead, which in turn requires a couple of other patches.
Then I threw in a couple of patches that have been baking nearby for
good measure...

It is the last patch to allow userspace to assume a known offset for a
particular bo that is both the most contentious and pulls in a bunch of
earlier patches - but those should be useful in their own right.

Please review, and look at the proposed changes to execbuffer carefully.
-Chris

^ permalink raw reply	[flat|nested] 22+ messages in thread
* [PATCH] drm/i915: Reduce memory pressure during shrinker by preallocating swizzle pages
@ 2012-12-03 16:30 Chris Wilson
  2012-12-03 17:29 ` Daniel Vetter
  0 siblings, 1 reply; 22+ messages in thread
From: Chris Wilson @ 2012-12-03 16:30 UTC (permalink / raw)
  To: intel-gfx

On a machine with bit17 swizzling, we need to store the bit17 of the
physical page address in put-pages. This requires a memory allocation,
on average less than a page, which may be difficult to satisfy is the
request to put-pages is on behalf of the shrinker. We could allow that
allocation to pull from the reserved memory pools, but it seems much
safer to preallocate the array for tiled objects on affected machines.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_tiling.c |   14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_gem_tiling.c b/drivers/gpu/drm/i915/i915_gem_tiling.c
index 8e59bb5..140ef6b 100644
--- a/drivers/gpu/drm/i915/i915_gem_tiling.c
+++ b/drivers/gpu/drm/i915/i915_gem_tiling.c
@@ -351,6 +351,20 @@ i915_gem_set_tiling(struct drm_device *dev, void *data,
 	/* we have to maintain this existing ABI... */
 	args->stride = obj->stride;
 	args->tiling_mode = obj->tiling_mode;
+
+	/* Try to preallocate memory required to save swizzling on put-pages */
+	if (dev_priv->mm.bit_6_swizzle_x == I915_BIT_6_SWIZZLE_9_10_17) {
+		if (obj->tiling_mode) {
+			if (obj->bit_17 == NULL) {
+				obj->bit_17 = kmalloc(BITS_TO_LONGS(obj->base.size >> PAGE_SHIFT) *
+						      sizeof(long), GFP_KERNEL);
+			}
+		} else {
+			kfree(obj->bit_17);
+			obj->bit_17 = NULL;
+		}
+	}
+
 	drm_gem_object_unreference(&obj->base);
 	mutex_unlock(&dev->struct_mutex);
 
-- 
1.7.10.4

^ permalink raw reply related	[flat|nested] 22+ messages in thread
end of thread, other threads:[~2012-12-07  0:15 UTC | newest]

Thread overview: 22+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-12-03 11:48 A bunch of random execbuffer patches Chris Wilson
2012-12-03 11:48 ` [PATCH 01/14] drm/i915: Move the get_pages assertions up to the right layer Chris Wilson
2012-12-03 11:49 ` [PATCH 02/14] drm/i915: Decouple the object from the unbound list before freeing pages Chris Wilson
2012-12-03 16:16   ` Daniel Vetter
2012-12-03 11:49 ` [PATCH 03/14] drm/i915: Bail if we attempt to allocate pages for a purged object Chris Wilson
2012-12-03 11:49 ` [PATCH 04/14] drm/i915: Defer the unbind for a fence change until the next get_fence() Chris Wilson
2012-12-03 11:49 ` [PATCH 05/14] drm/i915: Avoid forcing relocations through the mappable GTT or CPU Chris Wilson
2012-12-03 11:49 ` [PATCH 06/14] drm: Optionally create mm blocks from top-to-bottom Chris Wilson
2012-12-03 11:49 ` [PATCH 07/14] drm/i915: Preferentially allocate mappable GTT space to uncached bo Chris Wilson
2012-12-03 11:49 ` [PATCH 08/14] drm/i915: Tighten the checks for invalid relocation domains Chris Wilson
2012-12-03 11:49 ` [PATCH 09/14] drm/i915: Remove check for conflicting relocation write-domains Chris Wilson
2012-12-03 19:18   ` Daniel Vetter
2012-12-03 21:03     ` [PATCH] drm/i915: Reduce memory pressure during shrinker by preallocating swizzle pages Chris Wilson
2012-12-07  0:16       ` Daniel Vetter
2012-12-03 11:49 ` [PATCH 10/14] drm/i915: Take the handle idr spinlock once for looking up the exec objects Chris Wilson
2012-12-03 11:49 ` [PATCH 11/14] drm/i915: Move the execbuffer objects list from the stack into the tracker Chris Wilson
2012-12-03 11:49 ` [PATCH 12/14] drm/i915: Allow userspace to hint that the relocations were known Chris Wilson
2012-12-03 11:49 ` [PATCH 13/14] drm/i915: Use the reloc.handle as an index into the execbuffer array Chris Wilson
2012-12-03 11:49 ` [PATCH 14/14] drm/i915: Allow userspace to request an object at a specific offset Chris Wilson
  -- strict thread matches above, loose matches on Subject: below --
2012-12-03 16:30 [PATCH] drm/i915: Reduce memory pressure during shrinker by preallocating swizzle pages Chris Wilson
2012-12-03 17:29 ` Daniel Vetter
2012-12-03 17:54   ` Chris Wilson

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox