[OpenAFS-devel] Re: segfault in process.c:savecontext() during lwp init -- how to debug?

[Russ Allbery]

Adam Megacz <megacz@hcoop.net> writes:
> Marc Dionne <marc.c.dionne@gmail.com> writes:

>>>    (gdb) bt
>>>    #0  0xb7df69eb in savecontext () from /lib/libnss_afs.so.2
>>>    #1  0xb7df6687 in LWP_CreateProcess () from /lib/libnss_afs.so.2

>> What architecture is this on, and which glibc?

> Linux 2.6.18, Debian's libc6 2.7-18.  It's a recent etch->lenny upgrade.

What architecture?

>> In the past this has generally been a symptom of not dealing properly
>> with setjmp/longjmp pointer mangling in glibc.  The solution has been
>> to move to the ucontext interface (defining USE_UCONTEXT), or to
>> implement a pointer demangling function (ptr_mangle()).

> Hrm.  This problem is occurring in a Debian package; does that mean
> that we have to talk the maintainers of the openafs package into
> changing the options with which that package it is built?

The Debian packages are built with USE_UCONTEXT on every platform that
supports it and have been for quite a while.  I suspect that you have a
different problem.

-- 
Russ Allbery (rra@stanford.edu)             <http://www.eyrie.org/~eagle/>