GNOME Bugzilla – Bug 651302
SIGILL raised in orc_process_int16_clamp
Last modified: 2012-10-17 10:55:01 UTC
While using Pidgin under Ubuntu Maverick on amd64, I am experiencing a random crash that sometimes occurs unattended. Details and stack trace are provided here: http://developer.pidgin.im/ticket/14185
Could you attach the output of orc-bugreport? (Or paste it if it's short.)
Hey David. Can you tell me a bit about what I need to do to get the output of orc-bugreport? I took a look in my package manager and couldn't find any such package containing orc-bugreport. I couldn't dig anything up on Google either.
On Ubuntu, liborc-0.4-dev.
Thanks Dave. I installed the package, but I'm not sure how to use orc-bugreport with Pidgin and GStreamer to get the information you requested?
Just run it from a terminal and attach the output to this bug
Thanks Sebastian. $ orc-bugreport Orc 0.4.14 - integrated testing tool Active backend: sse L1 cache: 32768 L2 cache: 3145728 L3 cache: 0 Family/Model/Stepping: 6/23/10 CPU name: Intel(R) Core(TM)2 Quad CPU Q9000 @ 2.00GHz Compiler options: sse2 sse3 ssse3 sse41 64bit Opcode test: No errors detected.
This is currently filed in my brain as "things that can't happen". What is probably happening is that this function is called successfully many times, and then one time (randomly) it fails with an illegal instruction. Although it is possible for Orc to generate illegal instructions, it would fail the first (and every) time, not randomly. Best I can guess is that the code memory is getting corrupted somehow. I shall think about it.
Thanks David. Let me know if I can reasonably be of assistance. Worst case is I just have to re-open Pidgin following a crash every couple of hours.
Hey David. Not to pester you, but a brief update: Nothing much new to report, other than the same habitual crash several times a day. I noticed it frequently happens when someone on my contact list signs on (e.g. a message alert is played).
Grrr, sorry about that. To work around the problem, you can set "ORC_CODE=backup" in the environment and run pidgin. That will use C backup code, which is unlikely to crash.
Thanks David, no problem. I'll give that a shot and that should keep it chugging along until we figure out what's making it segfault.
As a debug tool, maybe mprotecting write away from the pages with orc generated code could help determine if something (and what) is scribbling over them without having to use valgrind, which sounds unusable on a 'normal' use of Pidgin.
Does it make sense to keep this bug open, there doesn't seem to be much to go on, other than "something" must be writing into the memory where orc generated some code into, but we don't know if it's GStreamer or pidgin or some other library... Why is valgrind unusable for a normal use of pidgin ?
Let's close this then. It doesn't look like there's much to go on unfortunatly. If there's a bug in orc or GStreamer, it will pop up in other - hopefully easier to debug - contexts as well.