On Wed, Jan 04, 2017 at 03:36:16PM +, Tvrtko Ursulin wrote:
> You beat me to solving this interesting bug. :) With the
> comments/commit slightly improved:
CI chose to ignore this, so sent the improved comments to trybot:
https://patchwork.freedesktop.org/series/17506/
Thanks,
-Chris
--
On 04/01/2017 15:45, Chris Wilson wrote:
On Wed, Jan 04, 2017 at 03:36:16PM +, Tvrtko Ursulin wrote:
On 04/01/2017 14:51, Chris Wilson wrote:
During the fence registers are clobbered by a GPU reset. Hence if
During the reset I suppose, although it would sound still a bit
redundant. So
On Wed, Jan 04, 2017 at 03:36:16PM +, Tvrtko Ursulin wrote:
>
> On 04/01/2017 14:51, Chris Wilson wrote:
> >During the fence registers are clobbered by a GPU reset. Hence if
>
> During the reset I suppose, although it would sound still a bit
> redundant. So maybe "Since GPU reset clobbers
On 04/01/2017 14:51, Chris Wilson wrote:
During the fence registers are clobbered by a GPU reset. Hence if
During the reset I suppose, although it would sound still a bit
redundant. So maybe "Since GPU reset clobbers the fence registers, if
there is concurrent..."
there is concurrent
During the fence registers are clobbered by a GPU reset. Hence if
there is concurrent user access to a fenced region via a GTT mmaping,
the access will not be fenced during the reset (until we restore the
fences afterwards). In order to prevent invalid access during the reset,
before we clobber
