startx results in a segfault

[andornaut]

bspwm stopped working on my workstation as of 2c36390.

startx fails with the following in /var/log/syslog:

Sep 23 20:28:54 workstation acpid: client connected from 4092[0:0]
Sep 23 20:28:54 workstation acpid: 1 client rule loaded
Sep 23 20:28:55 workstation kernel: [  107.050820] NVRM: GPU at 0000:01:00: GPU-3736b905-88ce-b01a-e7f9-3e55a1fffb4c
Sep 23 20:28:55 workstation acpid: client connected from 4092[0:0]
Sep 23 20:28:55 workstation acpid: 1 client rule loaded
Sep 23 20:28:55 workstation kernel: [  107.387652] bspwm[4095]: segfault at 1f2000001f1 ip 000000000040d62a sp 00007fff4ab3dc40 error 4 in bspwm[400000+11000]

This is a dual-head workstation. I tested with an empty bspwmrc and a reduced .xinitrc:

[ -e "$PANEL_FIFO" ] && rm "$PANEL_FIFO"
mkfifo "$PANEL_FIFO"
exec bspwm -s "$PANEL_FIFO" -p W

The same version of bspwm works on my laptop.

[baskerville]

Grab this patch and apply it to the standard Git PKGBUILD.

Install the package generated from this patched PKGBUILD.

(Afterwards, check that file $(which bspwm) ends with not stripped.)

In a virtual terminal (without X running):

ulimit -c unlimited
startx
sudo systemd-coredumpctl gdb bspwm
set logging on
bt full
q

And post the content of gdb.txt.

[baskerville]

Might be fixed by 48d181f.

[andornaut]

I'm still experiencing the same issues with 48d181f.

I'm using Ubuntu which doesn't use PKGBUILD and also doesn't include systemd-coredumpctl. startx in the above instructions fails, though. Maybe I could try something else?

I did try installing via make debug && sudo make install. Now the /var/log/Xorg.0.log output contains:

import monitors
add monitor ...
add monitor ...
add desktop ...
xinit: connection to X server lost

I thought that perhaps this issue might be related so I checked the version of some dependencies:

When I install xcb-util-wm 0.3.9 from fedora 19, bspwm can start without segfaulting.

My advice to other Fedora users: Get the newest xcb-util-wm packages and 'make clean all' in bspwm.

On Ubuntu, xcb-util-wm is a source package which is used to build libxcb-ewmh-dev and libxcb-ewmh2.

It looks like my system already has version 0.3.9 of the above installed, as suggested in the above quotation:

Package: libxcb-ewmh2
Source: xcb-util-wm
Version: 0.3.9-2

Package: libxcb-ewmh-dev
Source: xcb-util-wm
Version: 0.3.9-2

Here are some other xcb package versions:

Package: libxcb-keysyms1-dev
Source: xcb-util-keysyms
Version: 0.3.9-1

Package: libxcb-keysyms1
Source: xcb-util-keysyms
Version: 0.3.9-1

Package: libx11-xcb-dev
Source: libx11
Version: 2:1.5.0-1ubuntu1

Package: libxcb1
Source: libxcb
Version: 1.8.1-2ubuntu2

Package: libxcb-util0-dev
Source: xcb-util
Version: 0.3.8-2build1

[andornaut]

This issue only occurs when I have a second monitor connected. If I disconnect my second monitor and run startx, then bspwm works as expected.

[andornaut]

This is the first commit with which I experience this issue: 80caab4

I've reverted to this commit, which works as expected: 1ef4d4f

[baskerville]

Can you make it crash by plugging your secondary monitor?

If so, you could run this in a virtual terminal (while bspwm is running):

su -c 'echo 0 > /proc/sys/kernel/yama/ptrace_scope'
gdb bspwm $(pgrep -x bspwm)
set logging on
c

Then go back to X and plug the second monitor.

Switch back to the virtual terminal holding gdb and issue:

bt full
q
y

[andornaut]

Can you make it crash by plugging your secondary monitor?

Nope. The second monitor isn't initialized though.

I can start Xorg with only exec xterm in .xinitrc then run gdb on bspwm, though:

Starting program: /usr/local/bin/bspwm 
warning: no loadable sections found in added symbol-file system-supplied DSO at 0x7ffff7ffa000
Program received signal SIGSEGV, Segmentation fault.
0x0000000000413f63 in ?? ()
#0  0x0000000000413f63 in ?? ()
No symbol table info available.
#1  0x000000090061a550 in ?? ()
No symbol table info available.
#2  0x00007fff00000000 in ?? ()
No symbol table info available.
#3  0x0000000000619280 in ?? ()
No symbol table info available.
#4  0x000001f2000001f1 in ?? ()
No symbol table info available.
#5  0x31706f746b736544 in ?? ()
No symbol table info available.
#6  0x00007fffffff9b00 in ?? ()
No symbol table info available.
#7  0x0000000000000005 in ?? ()
No symbol table info available.
#8  0x00007fffffff9b20 in ?? ()
No symbol table info available.
#9  0x00007fffffff9af0 in ?? ()
No symbol table info available.
#10 0x0000000000000005 in ?? ()
No symbol table info available.
#11 0x00007fffffff9b40 in ?? ()
No symbol table info available.
#12 0x0000000000000000 in ?? ()
No symbol table info available.
A debugging session is active.
    Inferior 1 [process 12735] will be killed.
Quit anyway? (y or n) 

Here's the process I followed:

git clone https://github.com/baskerville/bspwm.git
cd bspwm
make debug
sudo make install
# startx with only exec xterm in .xinitrc
# In the xterm
gdb
set logging on
target exec /usr/local/bin/bspwm
run
# Segfaults here
bt full
q
y

[baskerville]

This: target exec /usr/local/bin/bspwm doesn't load the symbols table, use file /usr/local/bin/bspwm instead.

[andornaut]

use file /usr/local/bin/bspwm instead.

Done. Here's the gdb.txt:

Starting program: /usr/local/bin/bspwm 
warning: no loadable sections found in added symbol-file system-supplied DSO at 0x7ffff7ffa000
Program received signal SIGSEGV, Segmentation fault.
0x0000000000413f63 in ewmh_update_desktop_names () at ewmh.c:80
80              for (j = 0; d->name[j] != '\0' && (i + j) < sizeof(names); j++)
#0  0x0000000000413f63 in ewmh_update_desktop_names () at ewmh.c:80
        d = 0x1f2000001f1
        m = 0x619280
        names = "Desktop1\000\235\377\377\377\177\000\000\005\000\000\000\000\000\000\000\340\235\377\377\377\177\000\000\260\235\377\377\377\177\000\000\005\000\000\000\000\000\000\000\000\236\377\377\377\177", '\000' , "/\\\214\367\377\177\000\000\022\000\a\000\003\002\000\000\227\001\000\000\006\000\000\000 \000\000\000\001", '\000' , "\260\235\377\377\377\177\000\000\225\214\336\367\377\177\000\000\001", '\000' , "\b\235\377\377\377\177\000\000\000\000\000\000\000\000\000\000\350A\255\367\377\177\000\000\000\000\000\000\000\000\000\000\240\235\377\377\377\177\000\000\000\000\000\000\000\000\000\000\350A\255\367\377\177\000\000\000\000\000\000\000\000\000\000"...
        i = 9
        j = 0
        names_len = 32767
#1  0x00000000004061f4 in add_desktop (m=0x621da0, d=0x621d00) at desktop.c:120
No locals.
#2  0x0000000000405bd8 in import_monitors () at monitor.c:300
        sres = 0x621910
        m = 0x621da0
        mm = 0x619280
        num = 2
        len = 6
        outputs = 0x621940
        cookies = 0x7fffffff9de0
        gpo = 0x620390
#3  0x00000000004042d3 in setup () at bspwm.c:224
        mask = 2048
        values = {64}
        net_atoms = {404, 395, 410, 407, 368, 405, 367, 372, 375, 378, 442, 
          387, 429, 428, 438, 434, 432, 430}
        qep = 0x6203f0
        ifo = 0x7ffff78b9b40
#4  0x000000000040396d in main (argc=1, argv=0x7fffffffe338) at bspwm.c:73
        descriptors = {__fds_bits = {0 }}
        socket_path = '\000' 
        fifo_path = 0x0
        sock_fd = 0
        ret_fd = 0
        dpy_fd = 0
        sel = 0
        n = 0
        sock_address = {sun_family = 0, sun_path = '\000' }
        rsp_len = 0
        msg = '\000' 
        rsp = '\000' 
        event = 0x0
        opt = -1 '\377'
        sp = 0x0
A debugging session is active.
    Inferior 1 [process 17165] will be killed.
Quit anyway? (y or n)

[baskerville]

Should be fixed by 6ae7601.

[andornaut]

Should be fixed by 6ae7601.

Confirmed. Thanks!