tmt@brunix.UUCP (Tom Talpey) (03/14/85)
This caused trouble for me when I wrote a daemon which used a mailbox
IPC structure (under V7) and which was awakened by SIGTRAPs from requesting
processes. It hung occasionally from races before a pause() and had to
be manually goosed when a single requester was active. My solution was
to set a 1-second alarm at every signal catch and to reset the tick
AFTER a completed pause(). This doesn't eliminate the problem, but the
alarm(1) call most always wakes it up one second later and even if THAT
sequence races, the next one might win, etc etc.
sig_catch(sig) {
signal(sig, sig_catch);
alarm(1);
}
sig_wait() {
signal(SIGTRAP, sig_catch);
signal(SIGALRM, sig_catch);
pause();
alarm(0);
}
This worked quite well for me, modulo the occasional extra pauses, but
the daemon wasn't very time-critical and the disk driver lost more
signals than the race ever did. The code sprouted about three or four
such hooks by the time it worked reliably.
Tom Talpey, Brown University Network Operations; ...brunix!tmt