re: install/46811: sysinst got "pid 0 is killed: exceeded RLIMIT_CPU" on netbsd-6/sparc

[matthew green <mrg%eterna.com.au@localhost>]

> >Description:
> I got the error message "pid 0 is killed: exceeded RLIMIT_CPU"
> repeatedly (1 message/sec?) on sysinst on NetBSD 6.0_BETA2/sparc (2012/08/16).
> But it does not seem to occur everytime.
> I use QEMU 0.13 .
> >How-To-Repeat:
> % qemu-system-sparc -boot d -nographic \
>  -hda sparc.img -cdrom NetBSD-6.0_BETA2-sparc.iso
> (reproducibility is not 100%...)

i think this is a bug in sched_pstats():

   1132                 runtm = p->p_rtime.sec;
[ ... ]
   1140                         runtm += l->l_rtime.sec;
   1179                 rlim = &p->p_rlimit[RLIMIT_CPU];
   1180                 sig = 0;
   1181                 if (__predict_false(runtm >= rlim->rlim_cur)) {
   1182                         if (runtm >= rlim->rlim_max) {
   1183                                 sig = SIGKILL;
   1184                                 log(LOG_NOTICE, "pid %d is killed: 
%s\n",
   1185                                         p->p_pid, "exceeded 
RLIMIT_CPU");
[ ... ]
   1196                 if (__predict_false(runtm < 0)) {
   1197                         if (!backwards) {
   1198                                 backwards = true;
   1199                                 printf("WARNING: negative runtime; "
   1200                                     "monotonic clock has gone 
backwards\n");
   1201                         }
   1202                 } else if (__predict_false(sig)) {
   1203                         KASSERT((p->p_flag & PK_SYSTEM) == 0);
   1204                         psignal(p, sig);
   1205                 }

ie, it prints the warning, but then due to runtm being negative doesn't
actually send the signal ever.  can you change line 1181 like so:

   1181                 if (__predict_false(runtm >= 0 && runtm >= 
rlim->rlim_cur)) {

and re-test?


.mrg.