Bug 104930 - [ivb] GPU hang in taisei
Summary: [ivb] GPU hang in taisei
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/DRI/i965 (show other bugs)
Version: unspecified
Hardware: x86-64 (AMD64) Linux (All)
: medium normal
Assignee: Intel 3D Bugs Mailing List
QA Contact: Intel 3D Bugs Mailing List
Depends on:
Reported: 2018-02-03 13:46 UTC by Vitaly Shukela
Modified: 2019-09-25 19:08 UTC (History)
3 users (show)

See Also:
i915 platform:
i915 features:

/sys/class/drm/card0/error (38.07 KB, application/x-7z-compressed)
2018-02-03 13:46 UTC, Vitaly Shukela
glxinfo output (25.00 KB, text/plain)
2018-02-05 13:08 UTC, Vitaly Shukela
hw used in tests (15.92 KB, text/plain)
2018-02-19 08:44 UTC, Vladimir Los
mesa and Co used in tests (31.44 KB, text/plain)
2018-02-19 08:44 UTC, Vladimir Los

Description Vitaly Shukela 2018-02-03 13:46:56 UTC
Created attachment 137153 [details]

[4519177.538749] [drm] GPU HANG: ecode 7:0:0x85fffdfc, in taisei [7309], reason: Hang on render ring, action: reset
[4519177.538754] [drm] GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.
[4519177.538755] [drm] Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel
[4519177.538756] [drm] drm/i915 developers can then reassign to the right component if it's not a kernel issue.
[4519177.538757] [drm] The gpu crash dump is required to analyze gpu hangs, so please always attach it.
[4519177.538759] [drm] GPU crash dump saved to /sys/class/drm/card0/error
[4519177.538806] drm/i915: Resetting chip after gpu hang
[4519187.537419] drm/i915: Resetting chip after gpu hang
[4519195.537271] drm/i915: Resetting chip after gpu hang
[4519205.697549] drm/i915: Resetting chip after gpu hang
[4519321.535217] drm/i915: Resetting chip after gpu hang
[4519331.615043] drm/i915: Resetting chip after gpu hang

Permanent graphical glitch (horizontal dashed white lines) is visible after it recovers from the hang, until the game exits (intel_do_flush_locked failed: Input/output error) after the second glitch.

Comment 1 Vitaly Shukela 2018-02-03 13:52:54 UTC
The glitch is still visible in multiple programs, if they are fullscreened.
Comment 2 Tapani Pälli 2018-02-05 12:12:13 UTC
Which version of Mesa are you running on?

I tried on Mesa master branch on my system and unfortunately did not get taisei to launch.

--- 8< ---
taisei: ../src/mesa/drivers/dri/i965/intel_fbo.c:455: intel_create_winsys_renderbuffer: Assertion `screen->mesa_format_supports_render[rb->Format]' failed.

Program received signal SIGABRT, Aborted.
0x00007ffff7a2966b in raise () from /lib64/libc.so.6
(gdb) bt
#0  0x00007ffff7a2966b in raise () from /lib64/libc.so.6
#1  0x00007ffff7a2b381 in abort () from /lib64/libc.so.6
#2  0x00007ffff7a218fa in __assert_fail_base () from /lib64/libc.so.6
#3  0x00007ffff7a21972 in __assert_fail () from /lib64/libc.so.6
#4  0x00007ffff17933c3 in intel_create_winsys_renderbuffer (screen=0x55555589baf0, format=MESA_FORMAT_Z_UNORM16, num_samples=0) at ../src/mesa/drivers/dri/i965/intel_fbo.c:455
#5  0x00007ffff179340d in intel_create_private_renderbuffer (screen=0x55555589baf0, format=MESA_FORMAT_Z_UNORM16, num_samples=0) at ../src/mesa/drivers/dri/i965/intel_fbo.c:479
#6  0x00007ffff17a42a5 in intelCreateBuffer (dri_screen=0x55555589aff0, driDrawPriv=0x555555a03e30, mesaVis=0x5555559b7900, isPixmap=0 '\000')
    at ../src/mesa/drivers/dri/i965/intel_screen.c:1681
#7  0x00007ffff17ad68a in driCreateNewDrawable (screen=0x55555589aff0, config=0x5555559b7900, data=0x555555a18230) at ../src/mesa/drivers/dri/common/dri_util.c:697
#8  0x00007ffff2ac3d87 in dri2_wl_create_window_surface (drv=0x55555586fdb0, disp=0x55555586f7b0, conf=0x5555559b9af0, native_window=0x555555a035c0, attrib_list=0x0)
    at ../src/egl/drivers/dri2/platform_wayland.c:223
#9  0x00007ffff2ab93fc in dri2_create_window_surface (drv=0x55555586fdb0, dpy=0x55555586f7b0, conf=0x5555559b9af0, native_window=0x555555a035c0, attrib_list=0x0)
    at ../src/egl/drivers/dri2/egl_dri2.c:1575
#10 0x00007ffff2aaaaf3 in _eglCreateWindowSurfaceCommon (disp=0x55555586f7b0, config=0x5555559b9af0, native_window=0x555555a035c0, attrib_list=0x0) at ../src/egl/main/eglapi.c:905
#11 0x00007ffff2aaabd1 in eglCreateWindowSurface (dpy=0x55555586f7b0, config=0x5555559b9af0, window=93824997144000, attrib_list=0x0) at ../src/egl/main/eglapi.c:921
#12 0x00007ffff7778d0c in SDL_EGL_CreateSurface () from /lib64/libSDL2-2.0.so.0
Comment 3 Tapani Pälli 2018-02-05 12:20:51 UTC
ok well now got it going by simply ignoring that assert .. I don't see GPU hang, my system is HSW so this might be something ivb specific indeed
Comment 4 Vitaly Shukela 2018-02-05 13:08:03 UTC
How do I look for Mesa version?


Attached glxinfo output
Comment 5 Vitaly Shukela 2018-02-05 13:08:06 UTC
Created attachment 137169 [details]
glxinfo output
Comment 6 Kenneth Graunke 2018-02-05 17:49:48 UTC
Your drivers are about a year out of date - Mesa 13.0.x came out in November 2016.  Please try upgrading to Mesa 17.3.3 or 18.0-rc3.
Comment 7 Vladimir Los 2018-02-19 08:43:12 UTC
No any described effects appeared.
Checked and tested in several different versions of Mesa (13.06.6 & 17.x.x) and corresponded noted drivers and SDL2 libs.
Built and installed in different combinations.
system: Intel® Core™ i5-3470 CPU @ 3.20GHz × 4 : 8Gb RAM 
uname: 4.13.0-32-generic #35~16.04.1-Ubuntu SMP x86_64 x86_64 x86_64 GNU/Linux
lshw and glxinfo (latest used in tests) output could be seen in the attachments.
Comment 8 Vladimir Los 2018-02-19 08:44:01 UTC
Created attachment 137432 [details]
hw used in tests
Comment 9 Vladimir Los 2018-02-19 08:44:32 UTC
Created attachment 137433 [details]
mesa and Co used in tests
Comment 10 Elizabeth 2018-03-14 23:17:21 UTC
Hello everybody,
Could you try latest release 17.3.6, it has important fixes for GPU hangs that could help your issue. If the problem is still reproducible, how can it be reliably triggered?
Thank you.
Comment 11 GitLab Migration User 2019-09-25 19:08:23 UTC
-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/1684.

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.