Re: core dumps from multi-threaded (kernel cloned) processes

Philip Gladstone (philip@raptor.com)
Fri, 14 Nov 1997 14:21:10 -0500


kwrohrer@enteract.com wrote:
> A refinement: assuming that you dump the register contents, stack
> pointer included, for all the threads in the core created by the dying
> thread, there's no reason to mark the others non-dumpable, assuming they
> survive the experience. If one were to dump to "core.%d",pid_of_dead_thread,
> instead of just to "core", that would be another reason not to mark the
> other threads non-dumpable. Or, kill the whole heavyweight process
> after dumping core from the perspective of the crashed thread.

I'm a little unclear on the semantics of posix threads and seg faults --
namely if the signal is not caught, does the whole process die, or just
that
thread? If the former case, then I'd just terminate the whole
heavyweight
process, but if the latter case, then I think that the linux pthreads is
behaving incorrectly.

by the way, does nayone know if where the ELF core file spec is -- which
defines the notes of type "CORE"? [I have the ELF spec itself...]

Philip

-- 
Philip Gladstone                           +1 617 487 7700
Raptor Systems, Waltham, MA         http://www.raptor.com/
Our new daughter: http://www.mwmc.com/Extweb/Cybernursery/17423662.htm