[PATCH] Ignore realitme signals.

[PATCH] Ignore realitme signals.

[Kevin Buettner]

I've just committed the patch below.

The comment in the patch explains what it's about.  I've seen this
problem (in which rda terminates due to getting a signal intended for
one of the inferior threads and terminates) in at three different
environments now (all embedded linux, but different architectures). 
While I suspect that the real problem lies in the kernel, it certainly
doesn't hurt for RDA to ignore any real-time signals sent to it.

	* server.c (main): Ignore realtime signals.

Index: server.c
===================================================================
RCS file: /cvs/src/src/rda/unix/server.c,v
retrieving revision 1.4
diff -u -p -r1.4 server.c
--- server.c	19 Mar 2004 18:26:08 -0000	1.4
+++ server.c	20 Apr 2004 20:08:09 -0000
@@ -374,6 +374,17 @@ main (int argc, char **argv)
 
   signal (SIGCHLD, chld_handler);
 
+  {
+    int sig;
+    /* Ignore realtime signals.  We do this so as to not terminate
+       RDA if we inadvertently receive one of these signals.  The
+       realtime signals are used for thread support, and, for some
+       reason, some environments send these signals to RDA as well
+       as the application.  (And some don't.)  */
+    for (sig = SIGRTMIN; sig <= SIGRTMAX; sig++)
+      signal (sig, SIG_IGN);
+  }
+
   if (portno != 0)
     {
       gdbsocket_startup (portno, gdbserver.attach, process);

Re: [PATCH] Ignore realitme signals.

[Daniel Jacobowitz]

On Tue, Apr 20, 2004 at 01:13:08PM -0700, Kevin Buettner wrote:
> I've just committed the patch below.
> 
> The comment in the patch explains what it's about.  I've seen this
> problem (in which rda terminates due to getting a signal intended for
> one of the inferior threads and terminates) in at three different
> environments now (all embedded linux, but different architectures). 
> While I suspect that the real problem lies in the kernel, it certainly
> doesn't hurt for RDA to ignore any real-time signals sent to it.
> 
> 	* server.c (main): Ignore realtime signals.

At least some versions of the kernel will send a child's exit signal to
the debugger, instead of SIGCHLD - this should be fixed in 2.6 series
kernels, but I think it was broken through all of 2.2 and 2.4.  GDB had
the same problem for a while, IIRC.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer