[Mesa-dev] [PATCH] i965: Fix Vertex URB Read Length calculation in 3DSTATE_SF on Gen6.

Daniel Vetter daniel at ffwll.ch
Wed Feb 6 01:35:17 PST 2013


On Tue, Feb 05, 2013 at 10:40:28PM +0100, Martin Steigerwald wrote:
> Am Dienstag, 5. Februar 2013 schrieb Daniel Vetter:
> > On Sat, Feb 02, 2013 at 09:22:55PM +0100, Martin Steigerwald wrote:
> > > About these messages: Uhm, bingo:
> > > 
> > > merkaba:~> zgrep -i "GPU hung" /var/log/kern.log*
> > > /var/log/kern.log:Jan 29 22:29:55 merkaba kernel: [135984.784399]
> > > [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung
> > > /var/log/kern.log:Jan 29 22:39:57 merkaba kernel: [136585.410291]
> > > [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung
> […]
> > > /var/log/kern.log.1:Jan 27 00:51:19 merkaba kernel: [28709.417383]
> > > [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung
> > > 
> > > 
> > > Nothing in older logs it seems:
> > > 
> > > merkaba:~> ls -l /var/log/kern.log*
> > > -rw-r----- 1 root adm  478589 Feb  2 20:55 /var/log/kern.log
> > > -rw-r----- 1 root adm 1318618 Jan 27 12:14 /var/log/kern.log.1
> > > -rw-r----- 1 root adm   52179 Jan 20 10:49 /var/log/kern.log.2.gz
> > > -rw-r----- 1 root adm   63021 Jan 13 11:18 /var/log/kern.log.3.gz
> > > -rw-r----- 1 root adm   51878 Jan  6 11:12 /var/log/kern.log.4.gz
> > > 
> > > 
> > > So all I know is that it didn´t happen with Mesa 8.0.5. I think 20th of
> > > January is about the time I started to test with Mesa 9.0.2 and then
> > > 9.2-devel + first your initial BLORB CopySubTexImage patch and then
> > > now all four of these and your Fix Vertex URB one from this thread.
> > 
> > Retesting with latest 3.8-rc kernels might be good - we've just merged a
> > few workaround patches for snb, which reportedly fix some hangs.
> > -Daniel
> 
> I have 3.8-rc6 running. Since then I mostly see radeon drm fixes.
> 
> Any specific fix you have in mind for me to look out for?

Nope, you have the ones I've thought off, nothing else comes to mind for
kernel stuff. I guess bisecting mesa would be a good next approach.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch


More information about the mesa-dev mailing list