Subject: Re: SS10 unstable under high load (SMP)
To: Michael-John Turner <mj@turner.org.za>
From: Atte =?iso-8859-1?Q?Peltom=E4ki?= <atte.peltomaki@iki.fi>
List: port-sparc
Date: 09/13/2007 19:52:10
On Thu, Jul 19, 2007 at 10:46:28AM +0200, Michael-John Turner wrote:
> On Thu, Jul 19, 2007 at 11:17:49AM +0300, Atte Peltom?ki wrote:
> > Tried, and failed miserably. After having hubertf read
> > pthread.buildlink3.mk for me enough to locate the undocumented
> > PREFER_NATIVE_PTHREADS environment variable and setting it to "no" in
> > python24 Makefile, I get this:
> 
> [snip]
> 
> Ah yes, I've seen this on sparc64 systems when trying to build certain
> packages and not using native pthreads (eg clamav). A quick 'n dirty hack
> is to do the following:
> - in /usr/include/signal.h, comment out the definitions of
>   pthread_sigmask() and pthread_kill()
> - in /usr/include/sys/types.h, comment out the include of <pthread_types.h>
> 
> That normally works for me, YYMV of course. I've been meaning to see if
> pkgsrc could handle this tranparently, but the round tuits have been
> lacking.

I have now done the suggested changes and succesfully compiled python
against GNU pthreads. If you're wondering why it took so long, well - my
summer vacation did not include any computers whatsoever ;-)

Anyway. It seems the process is deadlocking. I'll keep on poking at it,
but here's a ktrace dump of it:

http://worm.hole.fi/atte/python-dump.ktrace
(It's 3.3MB on a 2Mbit line, so it can take a small while to download)

p.s. I'm at EuroBSDCon right now, if you're here too come say hi!
(Koston@IRC)

-- 
Atte Peltomäki