friesen@psivax.UUCP (Stanley Friesen) (04/22/85)
I am having trouble in writing a routine to handle less than
one second sleeps on BSD 4.2 using setitimer and sigpause. The routine
I am using is similar to the code for sleep(). My problem, it keeps
hanging sfter a random number of calls. It seems that eventually it
does not return from sigpause. Why? What can I do about it?
The code is as follows:
#include <sys/time.h>
#include <signal.h>
int wakeup;
void nullfunc(){
wakeup = 1;
}
struct itimerval tm = {{0, 0}, {0, 1500}};
delay()
{
void nullfunc();
signal(SIGALRM, nullfunc);
wakeup = 0;
if(setitimer(ITIMER_REAL, &tm, NULL)) exit(1);
while(wakeup == 0) sigpause(0);
}
--
Sarima (Stanley Friesen)
{trwrb|allegra|cbosgd|hplabs|ihnp4|aero!uscvax!akgua}!sdcrdcf!psivax!friesen
or {ttdica|quad1|bellcore|scgvaxd}!psivax!friesengwyn@brl-tgr.ARPA (Doug Gwyn <gwyn>) (04/24/85)
> void nullfunc(){ > wakeup = 1; > } > ... > signal(SIGALRM, nullfunc); Sometimes the signal will fire HERE, ... > wakeup = 0; > if(setitimer(ITIMER_REAL, &tm, NULL)) exit(1); > while(wakeup == 0) sigpause(0); ... so you won't see it here. The simplest way to get sub-second naps on 4.2BSD is to do a select() specifying no file descriptors and a small timeout.
gwyn@brl-tgr.ARPA (Doug Gwyn <gwyn>) (04/24/85)
> Sometimes the signal will fire HERE, ... Oops, that was a bug on my part. The following is still true: > The simplest way to get sub-second naps on 4.2BSD is to > do a select() specifying no file descriptors and a small timeout.
chris@umcp-cs.UUCP (Chris Torek) (04/25/85)
Your code is one big race condition. If you aren't interested in
preserving the old interval timer, then use this (I've deleted the
unchanged code):
delay()
{
int omask, nmask;
#define mask(x) (1 << ((x) - 1))
signal(SIGALRM, nullfunc);
wakeup = 0;
omask = sigblock(mask(SIGALRM));
nmask = omask & ~mask(SIGALRM);
if (setitimer(ITIMER_REAL, &tm, (struct timeval *)NULL))
exit(1);
while (wakeup == 0)
sigpause(nmask);
sigsetmask(omask);
}
I'm assuming that sigpause() restores the signal mask that was in
effect before it's call; there is some very bogus-looking code in
the 4.2BSD kernel that is intended to do that, but I've never tested
it....
--
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 4251)
UUCP: {seismo,allegra,brl-bmd}!umcp-cs!chris
CSNet: chris@umcp-cs ARPA: chris@marylanddonn@utah-gr.UUCP (Donn Seeley) (04/29/85)
This is another perennial question... This time, though, I'll break down and post my 'all-singing, all-dancing' fsleep() routine to net.sources where everyone can beat on it. fsleep() is the same as sleep() but it takes a floating point argument instead of an integer one. I put some work into it trying to find all the race conditions and timing problems but I still can't claim perfection for it; corrections are welcomed... Donn Seeley University of Utah CS Dept donn@utah-cs.arpa 40 46' 6"N 111 50' 34"W (801) 581-5668 decvax!utah-cs!donn