Subject: Re: 2.0beta freezes
To: None <sigsegv@rambler.ru>
From: Sean Davis <dive@endersgame.net>
List: tech-kern
Date: 09/12/2004 13:06:20
On Sun, Sep 12, 2004 at 04:48:56PM +0100, sigsegv@rambler.ru wrote:
> Jukka Marin wrote:
> 
> >I have experienced a few freezes running 2.0beta kernels.  A moment ago,
> >I was about to save an image from gimp and the system froze.  The X display
> >remained visible, but the mouse and keyboard were all dead (no caps lock
> >operation even).  I waited for some 15 minutes and the system didn't reboot
> >or recover.  All the time, I was able to ping the system from a remote
> >machine.
> >
> >I have seen this before.  I thought it was a disk problem, replaced the
> >disk with a new one (different make as well), no help.  Sometimes the
> >system freezes for about 10 seconds and then comes back alive, sometimes
> >I have to push the reset button to get it going again.
> >
> >This has been going on for months.  The system is usually stable, but every
> >now and then (one or twice a month) it freezes like this.  I think it 
> >happens
> >during heavy disk / memory load, but I'm not completely sure.
> >
> > -jm
> >
> >
> > 
> >
> I had similar problems, which were attributed to mounting filesytems 
> with softdep option. I did a simple test, downloaded a 40MB tar file 
> (e.g. mozilla.tar.gz) and then unpacked it many times over and over 
> again, sooner or later, the whole system would just freeze. Mounting all 
> my filesystems without softdep option made the problem go away. I am not 
> a kernel programmer, but from what I can see, softdep code in NetBSD may 
> have some serious bugs.

It does. It was stable for years, but recently (I believe since UBC
happened) it's returned to dont-trust-it-further-than-you-can-throw-it
status. I had close to a dozen panics all in the softdep code. Now that I've
pulled it from all my kernels, and run everything with normall ffsv1 no
softdep, I can load the hell out of any of my machines and they won't crash.

Also, I've had 2.0_BETA crash on sparc64 for no known reason quite a few
times - switched that machine to -current (Sun Ultra 1E Model 170, aka Sun
Ultra 1 Creator), and no crashes since.

I'm of the opinion that 2.0_BETA is way more _BETA than -current is ;-)

-Sean

--
/~\ The ASCII
\ / Ribbon Campaign                   Sean Davis
 X  Against HTML                       aka dive
/ \ Email!