[Bug 95261] R5 M330 GPU lockup with DPM + high power states

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=95261

            Bug ID: 95261
           Summary: R5 M330 GPU lockup with DPM + high power states
           Product: DRI
           Version: DRI git
          Hardware: x86-64 (AMD64)
                OS: Linux (All)
            Status: NEW
          Severity: normal
          Priority: medium
         Component: DRM/Radeon
          Assignee: dri-devel at lists.freedesktop.org
          Reporter: andrzej.mendel at gmail.com

Created attachment 123454
  --> https://bugs.freedesktop.org/attachment.cgi?id=123454&action=edit
/var/log/syslog at the moment of hangup (photo)

System: Ubuntu 16.06 with Mesa git (padoka ppa) and kernel 4.6-rc6
GPU: Intel HD 5500 + AMD R5 M330 - all command below run with DRI_PRIME=1

I get GPU hangups, which result in freeze after a soft reset, whenever the load
is high enough to push the GPU into higher power states.

If I run, for example, glmark2, only the first frame gets rendered and then the
GPU lockups. After ~30s system freezes with information about GPU soft reset as
the last message in syslog (see attachment)

If I run a non-GPU-intensive commands (say, glxgears with vsync), then the GPU
stays in low power states and I do not get this hangup. If I force lower power
states (echo battery > /sys/class/drm/card1/device/power_dpm_state), I don't
get this hangup even with glmark2.

If I force high power states (echo performance >
/sys/class/drm/card1/device/power_dpm_state; echo high >
/sys/class/drm/card1/device/power_dpm_force_performance_level; echo on >
/sys/class/drm/card1/device/power/control) then I do not get the hangup as long
as there is no activity at all on the GPU. A simple glxgears is enough to
trigger a hangup in this situation.

This bug is present since at least kernel 4.2.

I would appreciate any info on how to debug this further and will provide more
info if requested.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20160504/5e2b6804/attachment.html>