[9fans] silent limit in thread(2) Channels which can corrupt the stack?

[Fco.J.Ballesteros <nemo@plan9.escet.urjc.es>]

Hi,

while debugging a program which uses thread(2), I've come to a point
where an apparently correct program causes stack corruption on one of
the threads. Tracking down the problem, I've found that the program I
show below seems to exceed a silent limit in the implementation of
channels.  I think that if enough pressure is put on a channel it can
corrupt your stack without a warning. Please, correct me if I'm wrong,
I'd be happy to get any info regarding what's going on.

The point where I think the channel has problems is (libthread/channel.c):

	// enqueue on the channel
	for (i = 0; i < 32; i++)
		if ((c->qused & (1 << i)) == 0) {
			c->qentry[i] = &a;
			c->qused |= a.q = 1 << i;
			break;
		}

In case I'm right, I'd suggest calling
abort() when the limit is exceeded.


This is the stack I get:

  acid: At pc:0x00002eb2:sleep+0x7 /sys/src/libc/9syscall/sleep.s:5
  sleep() /sys/src/libc/9syscall/sleep.s:3
	called from main+0x197 /sys/src/libthread/thread.c:500
  main(argc=0x00000001,argv=0x7fffefec) /sys/src/libthread/thread.c:459
	called from _main+0x31 /sys/src/libc/386/main9.s:16
  acid:

and this is smallest the program I got that presents the problem:

Channel c;
int	t;

void s(void *p)
{
	for(;;){
		if (t)
			break;
		recv(&c, nil);
	}
}

void r(void *p)
{
	int i = 1;
	for(;;){
		send(&c, &i);
		if (t)
			break;
	}
}

void
threadmain()
{
	int i;
	chaninit(&c, sizeof(int), 0);
	for (i = 0; i < 50; i++)
		proccreate(r, (void*)i, 16*1024);
	for (i = 0; i < 50; i++)
		proccreate(s, (void*)i, 16*1024);
	sleep(5*1000);
	t = 1;
}