Re: kern/38060: blocking TSTP is broken

[yamt%mwd.biglobe.ne.jp@localhost (YAMAMOTO Takashi)]

> The following reply was made to PR kern/38060; it has been noted by GNATS.
> 
> From: Andrew Doran <ad%netbsd.org@localhost>
> To: gnats-bugs%NetBSD.org@localhost
> Cc: 
> Subject: Re: kern/38060: blocking TSTP is broken
> Date: Tue, 4 Mar 2008 15:55:21 +0000
> 
>  On Mon, Mar 03, 2008 at 12:20:03AM +0000, YAMAMOTO Takashi wrote:
>  
>  >  > how about this?
>  >  > 
>  >  > YAMAMOTO Takashi
>  >  > 
>  >  > 
>  >  > Index: kern_sig.c
>  >  > ===================================================================
>  >  > RCS file: /cvsroot/src/sys/kern/kern_sig.c,v
>  >  > retrieving revision 1.272
>  >  > diff -u -p -r1.272 kern_sig.c
>  >  > --- kern_sig.c  20 Feb 2008 11:48:46 -0000      1.272
>  >  > +++ kern_sig.c  2 Mar 2008 12:21:27 -0000
>  >  > @@ -1352,23 +1352,7 @@ kpsignal2(struct proc *p, ksiginfo_t *ks
>  >  >                 if ((prop & SA_CONT) != 0 && action == SIG_DFL)
>  >  >                         goto out;
>  >  >  
>  >  > -               if ((prop & SA_STOP) != 0 && action == SIG_DFL) {
>  >  > -                       /*
>  >  > -                        * If a child holding parent blocked, stopping 
> could
>  >  > -                        * cause deadlock: discard the signal.
>  >  > -                        */
>  >  > -                       if ((p->p_sflag & PS_PPWAIT) == 0) {
>  >  > -                               p->p_xstat = signo;
>  >  > -                               proc_stop(p, 1, signo);
>  >  > -                       }
>  >  > -                       goto out;
>  >  > -               } else {
>  >  > -                       /*
>  >  > -                        * Stop signals with the default action are 
> handled
>  >  > -                        * specially in issignal(), and so are not 
> enqueued.
>  >  > -                        */
>  >  > -                       sigput(&p->p_sigpend, p, kp);
>  >  > -               }
>  >  > +               sigput(&p->p_sigpend, p, kp);
>  >  >         } else {
>  >  >                 /*
>  >  >                  * Process is stopped or stopping.  If traced, then no
>  >  
>  
>  I think this should work. I can't remember why it stops there instead of
>  deferring to issignal().

because otherwise it makes nanosleep wake up earlier?
(it was fixed by kern_time.c rev.1.141.)

>  There is a question over signal masking. TSTP is job control and so operates
>  on the whole process, although masks are per-thread. I found nothing in the
>  standards about this. I think we can be lazy and assume that if any thread
>  has it unmasked, it can be taken.
>  
>  Andrew

i couldn't find anything in standards either.

YAMAMOTO Takashi