Subject: Re: 1.6.2_RC4 kernel panic
To: ali (Anders Lindgren) <dat94ali@ludat.lth.se>
From: Andrey Petrov <petrov@NetBSD.org>
List: port-sparc64
Date: 03/03/2004 11:38:38
On Wed, Mar 03, 2004 at 09:22:42AM +0100, ali (Anders Lindgren) wrote:
> On Tue, 2 Mar 2004, Andrey Petrov wrote:
> 
> > On Wed, Feb 25, 2004 at 02:23:06PM +0100, ali (Anders Lindgren) wrote:
> > > Tried to build ncurses from pkgsrc on 1.6.2_RC4/sparc64,
> > > with pkgsrc gcc 3.3.2. It crashes pretty bad, and "sync"
> > > from "db" crashes again so something really fucked the kernel up.
> > >
> > > This is on:
> > > 	NetBSD 1.6.2 (IDA) #0: Thu Feb 12 22:15:42 CET 2004
> > > and:
> > > 	cpu0 at mainbus0: SUNW,UltraSPARC @ 143 MHz, version 0 FPU
> > >
> > > I hope this can be of some help to someone. Kernel config attached.
> > >
> >
> > How reproducable is that? I'd try -current.
> 
> Probably not. I rebuilt ncurses, and it just worked.
> 
> > There are several important improvements in it, integrated gcc33,
> > threads, and lot of kernel fixes. Unless your problem is ultra1-related
> > I wouldn't expect -current to be worth.
> 
> *cough* I just _downgraded_ to 1.6.2_RC4 because the -current I was
> running (1.6ZH and later 1.6ZI, Jan -> early Feb) was kernel panicking
> at such a rate as to render the machine unusable, and this machine needs
> to be up and running -- it's running my named, mail and http servers.
> I really really want to run -current instead if I can.
> 

I tried to set up a machine running 1.6 branch but every time something
else came up. For me it would be easier if you had -current. And it's
been stable on my load.

>   It would kernel panic in VFS, and other places, during pkgsrc builds
> etc., and when I ^C'd a foreground named9 during testing it crashed so
> hard I had to drive to the server room and power cycle the machine (you
> know you've crashed a sun hard when the serial console doesn't work! :-) )
> which is a major nuisance since said server room is 35km from my home.
> Unfortunately I didn't get a chance to save those ddb trace outputs. :-/
> 

Yeah, that's not friendly environment to experiment in it.

>   So far 1.6.2 has been otherwise stable, but I will surely switch back to
> -current for the integrated gcc3 and working pthreads as soon as someone
> can confirm it's stable enough again (my previous Nov 15 -current seemed
> rather stable!) to run my named, mail and http server on.
> 

I'll try them. More details about program versions, build environment
and load would help.

	Andrey