1490994 - Crash in libpthread-2.28.so@0x126f3

Status: RESOLVED DUPLICATE of bug 1489735

(Core :: Graphics: CanvasWebGL, defect)

Description

[Marcia Knous [:marcia]]

This bug was filed from the Socorro interface and is
report bp-147b9241-5999-4c97-b395-9d4a80180910.
=============================================================

Seen while looking at nightly crash stats: https://bit.ly/2xanclP. Content crash which appears to have started using 20180904220134. 121 crashes/28 installations in 2 weeks. Some movie sites and netflix are among the URLs.

Possible regression range based on Build ID: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=2a4cf603095ae6c4dc01d9e2b4adc217cbfd2eaf&tochange=bade35bc346f3bc1f5e8b59d3d49c7a1ba28166c

Top 10 frames of crashing thread:

0 libpthread-2.28.so libpthread-2.28.so@0x126f3 
1 radeonsi_dri.so radeonsi_dri.so@0x51e1ca 
2 radeonsi_dri.so radeonsi_dri.so@0x50d2bd 
3 radeonsi_dri.so radeonsi_dri.so@0x29ef73 
4 radeonsi_dri.so radeonsi_dri.so@0x462f3d 
5 libGLX_mesa.so.0.0.0 libGLX_mesa.so.0.0.0@0x44924 
6 radeonsi_dri.so radeonsi_dri.so@0x461c25 
7 libGLX_mesa.so.0.0.0 libGLX_mesa.so.0.0.0@0x4b478 
8 libGLX_mesa.so.0.0.0 libGLX_mesa.so.0.0.0@0x7701f 
9 libGLX_mesa.so.0.0.0 libGLX_mesa.so.0.0.0@0x7701f 

=============================================================

Comment 1

[Andrew McCreight [:mccr8]]

There's another signature that looks similar: bp-33c734bc-3ce4-45d3-9de7-f72c90180913

Comment 2

[Julien Cristau [:jcristau] (back April 22)]

The content process is dying with SIGSYS.  Could this be seccomp sandbox related?

Comment 3

[Julien Cristau [:jcristau] (back April 22)]

Looking at https://crash-stats.mozilla.com/report/index/620f70eb-32b4-4308-8cc5-ca70d0180913 and libpthread debug symbols from http://archive.ubuntu.com/ubuntu/pool/main/g/glibc/libc6-dbg_2.28-0ubuntu1_amd64.deb:

$ addr2line -e usr/lib/debug/.build-id/d7/ce1c3b1a3169d65c31021947a9778ae91bf8a0.debug 0x130bf 
/build/glibc-B9XfQf/glibc-2.28/nptl/../sysdeps/unix/sysv/linux/pthread_setaffinity.c:34

Talking to a radeonsi developer he says this could indeed be fallout from mesa messing with thread affinity.

Comment 5

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

For future reference, the syscall number is reported as the crash “address”; there's no need to find debug symbols just for that.  (It's also available as rAX in the registers, but the address field can be used in searches.)  In this case it's 0xcb which is:

#define __NR_sched_setaffinity 203

The “signature” is always going to be a mess for these, because we don't have symbols for system libraries and every libc build will (in general) have a different offset for the same function.

Comment 6

[Julien Cristau [:jcristau] (back April 22)]

Good to know, thanks Jed!