X.org crashes / Debian Bulleyes

[wasgurd]

Hi there,

After I've updated my system, I've got the trouble: when I use Fn + some functional key, X.org crashes.

Code:

[Sat Jan 25 02:01:57 2020] broken atomic modeset userspace detected, disabling atomic

And some error with Panfrost:

Code:

[Sat Jan 25 01:32:57 2020] panfrost ff9a0000.gpu: Unhandled Page fault in AS1 at VA 0x0000000002012680
                          Reason: TODO
                          raw fault status: 0x28002C3
                          decoded fault status: SLAVE FAULT
                          exception type 0xC3: TRANSLATION_FAULT_LEVEL3
                          access type 0x2: READ
                          source id 0x280
[Sat Jan 25 01:32:57 2020] panfrost ff9a0000.gpu: gpu sched timeout, js=0, config=0x3301, status=0x8, head=0x2329f80, tail=0x2329f80, sched_job=00000000d4418cb6
[Sat Jan 25 01:32:57 2020] ------------[ cut here ]------------
[Sat Jan 25 01:32:57 2020] WARNING: CPU: 3 PID: 366 at include/linux/dma-fence.h:533 drm_sched_resubmit_jobs+0x164/0x178 [gpu_sched]

It looks like a Debian issue, but maybe we can find a way to make Fn works.

[Antoine]

I am also experiencing a lot of Xorg crashes since my last apt upgrade on 24/01/2020. Looking at the apt history, there was a bunch of packages related to the video drivers in this update:

[...] libegl1:arm64 (1.1.0-1+b1, 1.3.0-7), libegl-mesa0:arm64 (19.2.6-1, 19.3.2-1), libgl1:arm64 (1.1.0-1+b1, 1.3.0-7), libglapi-mesa:arm64 (19.2.6-1, 19.3.2-1),  libgles2:arm64 (1.1.0-1+b1, 1.3.0-7), libgbm1:arm64 (19.2.6-1, 19.3.2-1), vainfo:arm64 (2.5.0+ds1-1, 2.6.0+ds1-1), libglx0:arm64 (1.1.0-1+b1, 1.3.0-7), libgl1-mesa-dri:arm64 (19.2.6-1, 19.3.2-1), mesa-vdpau-drivers:arm64 (19.2.6-1, 19.3.2-1), mesa-vulkan-drivers:arm64 (19.2.6-1, 19.3.2-1), mesa-va-drivers:arm64 (19.2.6-1, 19.3.2-1), libglx-mesa0:arm64 (19.2.6-1, 19.3.2-1), libglvnd0:arm64 (1.1.0-1+b1, 1.3.0-7) [...]

I am using the i3 window manager and I can reproduce easily the crash: opening a file dialog in firefox (Open/Save), launching some GTK3/Gnome applications (evince, gedit...).

The Xorg log is not very helpful with only the following entries:

Code:

[  3407.881] (EE)
[  3407.881] (EE) Backtrace:
[  3407.882] (EE) 0: /usr/lib/xorg/Xorg (OsLookupColor+0x188) [0xaaaae4b86880]
[  3407.882] (EE) unw_get_proc_info failed: no unwind info found [-10]
[  3407.882] (EE)
[  3407.882] (EE) Segmentation fault at address 0xaadb13f3ce38
[  3407.882] (EE)
Fatal server error:
[  3407.882] (EE) Caught signal 11 (Segmentation fault). Server aborting

This is very annoying as I cannot use some applications anymore. I have not been able to find the root cause of the crash yet...

[Antoine]

I have managed to find the bug is caused by the newly uploaded version libgl1-mesa-dri package (19.3.2-1), and more specifically to the code in rockchip_dri.so.
I have filled a bug in Debian bugtracker with a full backtrace of xorg crash: https://bugs.debian.org/cgi-bin/bugrepor...bug=949980
This bug is very annoying and I am surprised nobody is complaining...

[icecream95]

The 5.5 kernel has got a fix which helps a number of applications, mainly GTK ones.

There are also a pile of fixes and improvements for Mesa which are not in 19.3, so you'll want to install a recent Mesa git.

[xmixahlx]

i recommend using mesa-git instead of distro mesa. see here for details:
https://forum.pine64.org/showthread.php?tid=8953

[rillian]

I've seen this too. Thanks for tracking down a stack trace!

[ab1jx]

I am also experiencing a lot of Xorg crashes since my last apt upgrade on 24/01/2020. Looking at the apt history, there was a bunch of packages related to the video drivers in this update:

[...] libegl1:arm64 (1.1.0-1+b1, 1.3.0-7), libegl-mesa0:arm64 (19.2.6-1, 19.3.2-1), libgl1:arm64 (1.1.0-1+b1, 1.3.0-7), libglapi-mesa:arm64 (19.2.6-1, 19.3.2-1),  libgles2:arm64 (1.1.0-1+b1, 1.3.0-7), libgbm1:arm64 (19.2.6-1, 19.3.2-1), vainfo:arm64 (2.5.0+ds1-1, 2.6.0+ds1-1), libglx0:arm64 (1.1.0-1+b1, 1.3.0-7), libgl1-mesa-dri:arm64 (19.2.6-1, 19.3.2-1), mesa-vdpau-drivers:arm64 (19.2.6-1, 19.3.2-1), mesa-vulkan-drivers:arm64 (19.2.6-1, 19.3.2-1), mesa-va-drivers:arm64 (19.2.6-1, 19.3.2-1), libglx-mesa0:arm64 (19.2.6-1, 19.3.2-1), libglvnd0:arm64 (1.1.0-1+b1, 1.3.0-7) [...]

My deb uninstaller at https://sourceforge.net/projects/debuninstall/ can probably take out that set of debs.  Not sure of an easy way of putting the old versions back.  It works by caliing dpkg -r and detecting whether that returned an error or not.  It keeps working through the set of files as long as each pass removes at least one.

[xmixahlx]

you can either downgrade the packages with apt, or use mesa git (permanently or temporarily), or install a few day old git packages from experimental. so a lot of solutions.

[ab1jx]

I'm seeing that error with libgl1-mesa-dri 19.3.3-1 which was part of my install from the debootstrap yesterday.  I googled ""unw_get_proc_info" failed" and found this post.  This is testing after all, I'm not hugely surprised.  I just wonder if this deb is really the problem since it happens with 2 slightly different versions.

My crash happens starting X, I can see parts of my LXDE desktop then it crashes.  Right after the gpio-keys like the xorg log you posted on Debian bugs.  I was wondering if it was OsLookupColor and thinking about the color depth being apparently 24 instead of 32.

Code:

[   194.588] (II) event1  - gpio-keys: is tagged by udev as: Keyboard Switch
[   194.588] (II) event1  - gpio-keys: device is a keyboard
[   194.588] (II) event1  - gpio-keys: device is a switch device
[   197.152] (EE)
[   197.152] (EE) Backtrace:
[   197.155] (EE) 0: /usr/lib/xorg/Xorg (OsLookupColor+0x188) [0xaaaacf2208a0]
[   197.155] (EE) unw_get_proc_info failed: no unwind info found [-10]
[   197.155] (EE)
[   197.155] (EE) Segmentation fault at address 0xaadad3a895f8
[   197.155] (EE)
Fatal server error:
[   197.155] (EE) Caught signal 11 (Segmentation fault). Server aborting
[   197.155] (EE)
[   197.155] (EE)

Have you managed to work around the error?  I had this Bullseye installed and working once too, but I screwed it up so I've been trying several times to reinstall.

One sort of odd failure mode I saw a few times including this one was a blinking underline text cursor for part of a second, then a mouse cursor in the middle of the screen, then back to text, over and over for 15 minutes or  more.  A couple times I gave up at that, this time I flipped to another virtual terminal, logged in, set multiuser mode as the default.  But it comes from X not being able to start I think, for a week or so.  My first installation worked normally, that was January 15.  So sometime in that range of a month something broke.

I downloaded the arm64 deb from https://packages.debian.org/unstable/libgl1-mesa-dri and installed it but it was the same version (19.3.3-1) I already had so it made no difference.  I wish I had kept my old debs.  So I need to find debs in Bullseye updated in the last 30 days, intersected with debs I have.  Then try the deb it replaced.  This file is also in Buster as libgl1-mesa-dri_18.3.6-2+deb10u1_arm64.deb but it has different dependencies, I just tried it.  I think this file isn't the culprit, it's a different one.

I tried the real experimental from https://packages.debian.org/experimental...i/download but that runs into first wanting libgcc-s1, and that gets into wanting GCC 10, which probably wants a bunch more stuff.

By the stacktrace posted in the Debian bug report it dies in midgard_compile_shader_nir but I'm not doing anything with OpenGL at all.  LXDE does use a bit of GTK (2 I think) but other than that I'm doing nothing with Gnome.

[danielt]

I presume this goes away if you add modprobe.blacklist=panfrost to the kernel command line (albeit at the cost of graphics performance)?