glibc 2.3.4 on DEC Alpha segfaults

[Jakub Bogusz]

On Sun, Apr 10, 2005 at 11:50:41PM -0400, Andrew A. Gill wrote:
> glibc is segfaulting on DEC Alpha for me.
> 
> The last version that works is glibc-2.3.4-0.20040603.1.  The
> later CVS versions and glibc-2.3.4-[12] fail.  gdb won't let me
> do a backtrace (``no stack''), but I've got the following bits
> from ltrace (strace is attached):
> 
> xfce4-panel (4.2.0-1):
> 
> DEBUG: process_event.c:89: process_event(): event: sysret
> (SYS_osf_syscall [0])
> DEBUG: process_event.c:73: process_event(): event: signal
> (SIGSEGV [11])
> 
> dillo 0.8.4 (compiled against the new glibc):
> 
> DEBUG: process_event.c:89: process_event(): event: sysret
> (SYS_osf_syscall [0])
> DEBUG: process_event.c:85: process_event(): event: syscall
> (SYS_clone [312])
> DEBUG: breakpoints.c:103: disable_all_breakpoints(): Disabling
> breakpoints for
> pid 32336...
> DEBUG: process_event.c:89: process_event(): event: sysret
> (SYS_-1 [-1])
> DEBUG: process_event.c:81: process_event(): event: exit signal
> (SIGSEGV [11])
> 
> I'm not sure what's causing this--it may just be that my kernel
> (2.6.11) isn't configured for NPTL (glibc-2.3.4-[12].alpha uses
> NPTL, right?), or something easy.

glibc 2.3.4-[12] uses NPTL "where available" (i.e. on 2.6.x kernels) and
linuxthreads otherwise (when run on 2.4.x kernel).

There may be some problem with NPTL - I didn't test such configuration
as I have only 2.4.2x on alpha. dillo runs fine on glibc-2.3.4-1 with
linuxthreads. You can try it on 2.6.x by exporting
"LD_ASSUME_KERNEL=2.4.6" (and eventually use it as workaround until real
fix comes).


-- 
Jakub Bogusz    http://cyber.cs.net.pl/~qboosh/