drm/i915: "Flush Me Harder" required on gen6+
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Thu, 28 Jun 2012 07:48:42 +0000 (09:48 +0200)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Thu, 28 Jun 2012 19:06:25 +0000 (21:06 +0200)
commit97f209bcfc0c5db08d9badf8cbafd489f22a6e44
tree941bafd5b138df165195359973ad24b878469ef2
parenta8b0bbabf756bfb45a712b823ba41f5c95f85589
drm/i915: "Flush Me Harder" required on gen6+

The prep to remove the flushing list in

commit cc889e0f6ce6a63c62db17d702ecfed86d58083f
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Jun 13 20:45:19 2012 +0200

    drm/i915: disable flushing_list/gpu_write_list

causes quite some decent regressions. We can fix this by setting the
CS_STALL bit to ensure that the following seqno write happens only
after the cache flush has completed. But only do that when the caller
actually wants the flush (and not also when we invalidate caches
before starting the next batch).

I've looked through all our ancient scrolls about gen6+ pipe control
workarounds, and this seems to be indeed a legal combination: We're
allowed to set the CS_STALL bit when we flush the render cache (which
we do).

While yelling at this code, also pass back the return value from
intel_emit_post_sync_nonzero_flush properly.

v2: Instead of emitting more pipe controls, set the CS_STALL bit on
the write flush as suggested by Chris Wilson. It seems to work, too.

Cc: Eric Anholt <eric@anholt.net>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=51436
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=51429
Tested-by: Lu Hua <huax.lu@intel.com>
Tested-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/intel_ringbuffer.c
This page took 0.030462 seconds and 5 git commands to generate.