Re: kern/46402 (LWPs created after exit_lwp() is called can hang the process...)

[Greg Oster <oster%cs.usask.ca@localhost>]

On Sat, 22 Sep 2012 14:21:40 +0100
Mindaugas Rasiukevicius <rmind%netbsd.org@localhost> wrote:

> Greg Oster <oster%cs.usask.ca@localhost> wrote:
> >  > Synopsis: LWPs created after exit_lwp() is called can hang the
> >  > process...
> >  > 
> >  > State-Changed-From-To: analyzed->feedback
> >  > State-Changed-By: rmind%NetBSD.org@localhost
> >  > State-Changed-When: Wed, 19 Sep 2012 21:23:06 +0000
> >  > State-Changed-Why:
> >  > Can you please try the following patch?
> >  > 
> >  > http://www.netbsd.org/~rmind/lwp_wait_fix.diff
> >  > 
> >  
> >  After 376883 iterations on the test loop thing hung:
> >  
> >  USER      PID %CPU %MEM    VSZ   RSS TTY     STAT STARTED    TIME
> >  COMMAND UID   PID  PPID   CPU PRI NI    VSZ   RSS WCHAN    STAT
> >  TTY        TIME COMMAND
> >  root    27779  0.0  0.1 798304  1396 ttyp0   DEl+  2:19AM 0:00.01
> >  (t_cond   0 27779   410 93985  63  0 798304  1396 -        DEl+
> > ttyp0 0:00.01 (t_cond) root    27779  0.0  0.1 798304  1396 ttyp0
> > DEl+ 2:19AM 0:00.01 (t_cond   0 27779   410 93985  95  0 798304
> > 1396 lwpwait  DEl+ ttyp0 0:00.01 (t_cond)
> >  
> >  (sorry for the cut'n'paste mess... weekend plans have changed, and
> > I have to run.. I can attempt to test more over the next days...)
> 
> Thanks.  No hurry. :)  Perhaps you could get the output of ps from
> crash(8) or DDB?  Also, contents of struct proc of that process might
> be useful.

I can't explain why, but in another 3 million runs the above was the
*only* issue encountered.  When I test without the patch, it hangs
instantly (in the last case, on the first test case).

It's entirely possible that the t_cond hang was caused by something
unrelated... and I've been unable to replicate the issue since
rebooting the machine after the last hang.

I'd say "Please commit this patch".

Later...

Greg Oster