GDB backtrace does not work correctly for Thumb-2 on armel

Ulrich, I saw "gdb" and I thought of you! ;-) Would you be tempted to try reproducing this issue with gdb tip, and if it's still an issue, to work on a fix? :-)

I was unable to investigate much in the original reported scenario, because the libc.so binary is missing from the tarball:

uweigand@beagle:~$ tar tfz t2-bt-testcase.tar.gz
usr/sbin/rsyslogd-
lib/libz.so.1
lib/libpthread.so.0
lib/libdl.so.2
lib/librt.so.1
lib/libgcc_s.so.1
lib/libc.so.6
usr/lib/debug/usr/sbin/rsyslogd
usr/lib/debug/lib/libpthread-2.11.1.so
usr/lib/debug/lib/libdl-2.11.1.so
usr/lib/debug/lib/librt-2.11.1.so
usr/lib/debug/lib/libc-2.11.1.so

Note that lib/libc.so.6, which is just a symbolic link, is there, but lib/libc-2.11.1.so, which is where the link goes to, is missing.

Nevertheless, I was able to reproduce similar problems when attempting to backtrace out of interrupted system calls. For example, I've been using the following example:

#include <unistd.h>

int main (void)
{
  sleep (1000);
}

Running this in GDB 7.2 and interrupting results in:

uweigand@beagle:~$ ./gdb-7_2-build/gdb/gdb ./a.out
GNU gdb (GDB) 7.1.90.20100709-cvs
Copyright (C) 2010 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law. Type "show copying"
and "show warranty" for details.
This GDB was configured as "armv7l-unknown-linux-gnueabi".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>...
Reading symbols from /home/uweigand/a.out...done.
(gdb) run
Starting program: /home/uweigand/a.out
^C
Program received signal SIGINT, Interrupt.
0x4008e02c in nanosleep () from /lib/libc.so.6
(gdb) bt
#0 0x4008e02c in nanosleep () from /lib/libc.so.6
#1 0x4008de70 in sleep () from /lib/libc.so.6
#2 0x4008de70 in sleep () from /lib/libc.so.6
Backtrace stopped: previous frame identical to this frame (corrupt stack?)
(gdb)

However, when I install the libc6-dbg debuginfo package, backtraces work:

uweigand@beagle:~$ ./gdb-7_2-build/gdb/gdb ./a.out
GNU gdb (GDB) 7.1.90.20100709-cvs
Copyright (C) 2010 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law. Type "show copying"
and "show warranty" for details.
This GDB was configured as "armv7l-unknown-linux-gnueabi".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>...
Reading symbols from /home/uweigand/a.out...done.
(gdb) run
Starting program: /home/uweigand/a.out
^C
Program received signal SIGINT, Interrupt.
0x4008e02c in nanosleep () at ../sysdeps/unix/syscall-template.S:82
82 ../sysdeps/unix/syscall-template.S: No such file or directory.
        in ../sysdeps/unix/syscall-template.S
(gdb) bt
#0 0x4008e02c in nanosleep () at ../sysdeps/unix/syscall-template.S:82
#1 0x4008de70 in __sleep (seconds=0) at ../sysdeps/unix/sysv/linux/sleep.c:138
#2 0x000083c0 in main () at t.c:6
(gdb)

The reason why backtraces do not work without debug info is that GDB will fall back to analyzing function prologue code instead. This requires GDB to understand and decode all instructions that may be used in such ...

I've now committed a patch to mainline GDB that significantly improves backtraces for Thumb-2 code even in the absence of DWARF CFI debug info. This will still not handle absolutely all cases (e.g. hand-written assembler routines that don't follow usual conventions), but should still help quite a bit:
http://sourceware.org/ml/gdb-patches/2010-10/msg00132.html

Sorry, this one seems to have fallen through the cracks on the package side. If this was fixed upstream in 2010-10, can we assume that the fix is present in the 7.4-0ubuntu1 package in precise?

Yes, the fix is present in GDB 7.4 (and in fact already GDB 7.3) release.