Running torcs on drm-intel-next-2015-04-23 or earlier, with mesa 10.5.2 or current master, hangs the machine hard or can also segfault torcs. Can't get anything useful from the machine when it happens, since it's dead. Reproduced with SKL-S beta and SKL-Y alpha sdp's.
The segfault suggests an oops, so netconsole? Also try disabling fbdev to disable the kms panic handler.
netconsole shows nothing I don't know what the other sentence means.. build a special kernel?
(In reply to Timo Aaltonen from comment #2) > netconsole shows nothing > > I don't know what the other sentence means.. build a special kernel? Chris, did you mean to compile with CONFIG_DRM_I915_FBDEV disabled?
Yes, but even more so to compile without FBCON as we (tjaalton and I) discussed on irc. Basically it just prevents a machine hang in case we panic whilst holding the struct_mutex (because our atomic panic handler is not so atomic).
so I built a kernel without CONFIG_FRAMEBUFFER* and CONFIG_DRM_I915_FBDEV, but it still hangs the same way.. The kernel is 3.19 + backported i915 for SKL/CHV from 4.1-rc3 + WA fixes etc, but IIRC a recent-ish nightly hangs too.
Ok, that means you have a "it can't possibly happen but it does" GPU hang that kills the machine (as opposed easier to debug driver deadlock).
Does the hard hang still happens with latest -nightly, which contains the following commit? commit 7fd2d26921d1dd70732d8765d714ec3a023a3ca9 Author: Mika Kuoppala <mika.kuoppala@linux.intel.com> Date: Thu Jun 18 12:51:40 2015 +0300 drm/i915: Reset request handling for gen8+
I tried drm-intel-next-2015-06-19 which has it too, but it still hung the same way, also with i915.reset=0 like before.
ran that kernel plus mesa 10.6.0 with the patch from 90425 added, and now torcs just segfaults
I believe this got fixed by WaEnableGapsTsvCreditFix (245d96670d2655f7)
Thanks for the follow-up, Timo!
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.