]> git.karo-electronics.de Git - karo-tx-linux.git/commit
drm/i915: clear fencing tracking state when retiring requests
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Wed, 11 Apr 2012 23:27:57 +0000 (01:27 +0200)
committerBen Hutchings <ben@decadent.org.uk>
Wed, 17 Oct 2012 02:50:13 +0000 (03:50 +0100)
commit309c9c15f367d6466b24c1affb69c72145fc006d
tree766e69f2f7160f3d11545fbdbde039b0fdc169bd
parente25aa0dd616effbeedc4acee4a9b8fe2150c2afd
drm/i915: clear fencing tracking state when retiring requests

commit 15a13bbdffb0d6288a5dd04aee9736267da1335f upstream.

This fixes a resume regression introduced in

commit 7dd4906586274f3945f2aeaaa5a33b451c3b4bba
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Mar 21 10:48:18 2012 +0000

    drm/i915: Mark untiled BLT commands as fenced on gen2/3

which fixed fencing tracking for untiled blt commands.

A side effect of that patch was that now also untiled objects have a
non-zero obj->last_fenced_seqno to track when a fence can be set up
after a pipelined tiling change. Unfortunately this was only cleared
by the fence setup and teardown code, resulting in tons of untiled but
inactive objects with non-zero last_fenced_seqno.

Now after resume we completely reset the seqno tracking, both on the
driver side (by setting dev_priv->next_seqno = 1) and on the hw side
(by allocating a new hws page, which contains the seqnos). Hilarity
and indefinite waits ensued from the stale seqnos in
obj->last_fenced_seqno from before the suspend.

The fix is to properly clear the fencing tracking state like we
already do for the normal gpu rendering while moving objects off the
active list.

Reported-and-tested-by: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Jiri Slaby <jslaby@suse.cz>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
drivers/gpu/drm/i915/i915_gem.c