Re: Proglem doing CA calls after catching a signal

[Eric Norum <norume@aps.anl.gov>]

Chris Timossi wrote:
> To be honest, I dont know if it's safe for a signal handler to join a 
> thread context. There's at least one possible race condition in this 
> case: the signal handler could execute after ca_task_exit() has been 
> called. What other race conditions are you thinking about?

The POSIX documentation I've found includes neither pthread_setspecific nor 
pthread_getspecific in the list of routines safe to call from a signal 
handler.  For example, the Solaris documentation describes these routines as 
MT-Safe, but not as Async-Signal-Safe.  Given this, I would say that 
ca_current_context and ca_attach_context should not be called from a signal 
handler since they end up calling the pthread routines.

Can anybody with access to the POSIX standard check and confirm the unsafe 
nature of the pthread_[set/get]specific() routines?

> I'm sure Jeff will weigh in when he gets back.

That may be a while since he's currently caught up in the security situation 
at Los Alamos.

--
Eric Norum                                 norume@aps.anl.gov
Advanced Photon Source                     Phone: (630) 252-4793
Argonne National Laboratory