[Intel-gfx] [PATCH 0/4] Wrap up ILK-style atomic watermarks, try two

[Jani Nikula]

On Tue, 03 Nov 2015, Matt Roper <matthew.d.roper at intel.com> wrote:
> Patches #3 and #4 here are the final two patches from the atomic watermark
> series that was posted here:
>
>    http://lists.freedesktop.org/archives/intel-gfx/2015-September/076634.html
>
> We had to pull those out when Jani reported a BDW boot regression (divide by
> zero during watermark calculation).  Although we never found a smoking gun for
> that divide by zero, I haven't been able to reproduce the issue on a similar
> system.  There's been a lot of code churn since that time, so I'm hoping that
> we've either already fixed the issue without realizing it, or that the extra
> paranoia added in patch #2 here will avoid the crash and highlight the culprit.
>
> The first patch here solves a legitimate bug that could cause a divide-by-zero
> (just not the one Jani was seeing).  The second patch adds extra guards on
> divide operations to verify our invariants and ensure that bugs elsewhere in
> the driver can't lead to a fatal divide-by-zero (at least on the ILK
> codepaths).
>
> Please don't merge #3 or #4 here until we at least get a positive test result
> from Jani.

Still gives me warnings. I didn't try one patch at a time, but here are
the dmesgs before http://pastebin.com/GVG9UC1U and after
http://pastebin.com/1HaXTJ3Z applying the series on top of today's
nightly.

BR,
Jani.


-- 
Jani Nikula, Intel Open Source Technology Center