agp/intel: Reinforce the barrier after GTT updates
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 10 Apr 2020 08:35:35 +0000 (09:35 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 17 Jun 2020 14:40:36 +0000 (16:40 +0200)
commitbfc12ff9050b5d0705b867c1fcd26e7f2af4c2da
tree5f4905837a9b2a844cc90c78871d5fb004cd3b77
parente81b05e535e8f4f636b81bdb8a4533cd42d869c8
agp/intel: Reinforce the barrier after GTT updates

commit f30d3ced9fafa03e4855508929b5b6334907f45e upstream.

After changing the timing between GTT updates and execution on the GPU,
we started seeing sporadic failures on Ironlake. These were narrowed
down to being an insufficiently strong enough barrier/delay after
updating the GTT and scheduling execution on the GPU. By forcing the
uncached read, and adding the missing barrier for the singular
insert_page (relocation paths), the sporadic failures go away.

Fixes: 983d308cb8f6 ("agp/intel: Serialise after GTT updates")
Fixes: 3497971a71d8 ("agp/intel: Flush chipset writes after updating a single PTE")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Andi Shyti <andi.shyti@intel.com>
Cc: stable@vger.kernel.org # v4.0+
Link: https://patchwork.freedesktop.org/patch/msgid/20200410083535.25464-1-chris@chris-wilson.co.uk
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/char/agp/intel-gtt.c