[Gta04-owner] dbus crashing with SIGBUS after suspend/resume
edorfaus at xepher.net
Thu Jun 27 00:59:02 CEST 2013
On 06/26/2013 09:22 PM, Radek Polak wrote:
> I have setup kernel and rootfs so that it produces coredumps. It appears that
> also udev and rsyslogd crashed. The crash happens in syscall (epoll_wait() for
> dbus and udev and in read() for rsyslogd). To me it looks that someone must
> have sent the SIGBUS to these daemons.
> Any ideas how to debug this further?
I haven't done much/any debugging of this kind, so take this with a
grain of salt, but a bit of searching found me someone else with a
SIGBUS in epoll_wait - but the main response he got to that was that the
signal was unlikely to come from epoll_wait, and that he should check
the stack traces from the other threads ("thread apply all where" in
gdb, apparently) to see if there's a "more interesting" stack
trace/thread to look at. I don't know if you've already done that, or if
those programs even use threading, but figured I'd mention it.
On Linux, the SIGBUS signal can apparently be triggered by several
things, including failure to page in a code page (due to memory pressure
(without the OOM killer) or a failed I/O request), errors during
accesses to memory-mapped devices, unaligned memory access (unlikely in
compiled code), bad pointer dereferencing, etc... So it's hard to tell
exactly why it might have been sent here, but I still think it's more
likely sent by the kernel than by some other program.
Personally, I'd initially suspect a problem with an mmapped device,
since udev is involved and we've had device problems before...
More information about the Gta04-owner