[Bug 94814] New: Purging GPU memory, out of memory kernel error, plenty of memory available

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=94814

            Bug ID: 94814
           Summary: Purging GPU memory, out of memory kernel error, plenty
                    of memory available
           Product: DRI
           Version: DRI git
          Hardware: x86-64 (AMD64)
                OS: Linux (All)
            Status: NEW
          Severity: normal
          Priority: medium
         Component: DRM/Intel
          Assignee: intel-gfx-bugs at lists.freedesktop.org
          Reporter: a.nielsen at shikadi.net
        QA Contact: intel-gfx-bugs at lists.freedesktop.org
                CC: intel-gfx-bugs at lists.freedesktop.org

Created attachment 122700
  --> https://bugs.freedesktop.org/attachment.cgi?id=122700&action=edit
dmesg output after OOM event

In response to this request:
https://lists.freedesktop.org/archives/intel-gfx/2016-February/088489.html

I have run the drm-intel-nightly kernel for 18 days, two hours and experienced
this problem again.

It usually manifests when using mplayer fullscreen (windowed is fine) at 2K or
4K resolution.

The video framerate grinds to a crawl, mplayer spits out messages about the
computer being too slow for playback, then applications start getting killed by
the kernel OOM killer.

The system has around 14GB of free memory at this point, so it's obviously not
main memory it has run out of, although the kernel still tries to kill
processes to free up main memory anyway.

The mailing list post has copies of dmesg from a normal kernel, and I will
attach copies of dmesg output for drm-intel-nightly.  To produce this dmesg
output, I ran mplayer in fullscreen mode at 4K (720p video), after a few
minutes it ran slowly so I exited mplayer and observed one process was killed. 
I immediately launched mplayer again and it was slow from the start, and then
when I exited I observed another process was killed.

I can now no longer run mplayer as it will kill all my processes one by one, so
to run mplayer again I will need to restart my machine.  It almost seems like
there is a video memory leak, video memory is becoming fragmented, or something
along those lines.  It always seems to take at least two weeks for the problem
to surface.

System architecture: x86_64
Kernel version: 4.5.0-1-drm-intel-nightly
Distribution: Arch Linux
Machine: Intel DH87MC motherboard, latest BIOS as of drm-intel-nightly
compilation date (reflashed before booting this kernel)
Display connector: DP at 4K, DVI at 1600x1200 portrait, HDMI at 1600x1200 portrait

I only just saw the request to use drm.debug=0x1e so I will add that to my
kernel command line now and reboot in a day or two if I don't hear back with
any more tests to run, and in another 18 days or so when the issue happens
again I'll attach the more detailed logs.

-- 
You are receiving this mail because:
You are the assignee for the bug.
You are on the CC list for the bug.
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-gfx-bugs/attachments/20160404/f901c041/attachment.html>