Re: ProcessManager on Solaris

[David Daney <ddaney at avtrex dot com>]

Eric Botcazou wrote:
> > It should be pretty trival to fix boehm-gc to block SIGCHLD. Basically
> > you should just need to copy the pthread_sigmask() code from
> > block_sigchld in posix-threads.cc into GC_pthread_create in
> > solaris_pthreads.c. I think it would be fine to make that change in
> > order to fix this problem, so please submit a patch.
>
>
> Indeed, if I'm allowed to directly patch Boehm GC, the fix is trivial.  
> Attached, tested on SPARC64/Solaris 9.
>
>
> 2005-02-16  Eric Botcazou  <ebotcazou@libertysurf.fr>
>
> 	* solaris_threads.c (GC_thr_daemon): Block SIGCHLD.
>
>
>
>
> ------------------------------------------------------------------------
>
> Index: solaris_threads.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/boehm-gc/solaris_threads.c,v
> retrieving revision 1.12
> diff -u -p -r1.12 solaris_threads.c
> --- solaris_threads.c	16 Aug 2004 16:32:26 -0000	1.12
> +++ solaris_threads.c	15 Feb 2005 23:09:05 -0000
> @@ -743,7 +743,13 @@ void * GC_thr_daemon(void * dummy)
>      register GC_thread t;
>      register int i;
>      register int result;
> -    
> +    sigset_t mask;
> +
> +    /* Libgcj's ProcessManager requires that SIGCHLD be delivered to it. */
> +    sigemptyset (&mask);
> +    sigaddset (&mask, SIGCHLD);
> +    thr_sigsetmask (SIG_BLOCK, &mask, NULL);
> +
>      for(;;) {
>        start:
>          result = thr_join((thread_t)0, &departed, &status);

What happens if the SIGCHLD is received between the beginning of this 
function and the thr_sigsetmask?

I am not a Solaris threads expert, but it looks like a race condition.

New threads are documented to inherit their signal mask from the parent. 
Where is the Solaris equivalent of libjava/posix-threads.cc ?  The 
signal mask is set in there for posix-threads (look for calls to 
block_sigchld() in _Jv_ThreadRegister and _Jv_ThreadStart.

David Daney.