HEADS UP [Re: Insta-panic with recent -current on Ultra10]

[Chris Knight]

> -----Original Message-----
> From: owner-freebsd-sparc64 at freebsd.org On Behalf Of Ken Smith
> Sent: Friday, 27 August 2004 11:01 PM
> To: Pyun YongHyeon
> Cc: Chris Knight; sparc64 at freebsd.org
> Subject: Re: HEADS UP [Re: Insta-panic with recent -current 
> on Ultra10]
> 
> On Fri, Aug 27, 2004 at 07:15:51PM +0900, Pyun YongHyeon wrote:
> > On Fri, Aug 27, 2004 at 06:03:02PM +1000, Chris Knight wrote:
> >  > Howdy,
> >  > 
> >  > I'm still getting this panic on sources updated just before
> >  > you tagged newvers.sh for BETA2.
> >  > 
> >  > ad0: 8693MB <ST39140A/861000> [17662/16/63] at ata2-master WDMA2
> >                                                               ^^^^^
> > Since you still use DMA on ad device, I'm afraid your kernel is not
> > updated with latest loader.
> > (src/sys/boot/sparc64/loader/main.c, rev 1.24 or 1.23.2.1)
> > Please re-cvsup and try again.
> > 
> 
> This is most likely the problem but it might not be a re-cvsup
> problem.
> If you are following The Right Way To Update (which virtually none of
> us do... :-) then you're installing the new kernel and rebooting
> before running installworld.  It's installworld that takes care of
> installing the new loader, which has hardcoded in it something that
> turns off ata DMA.
> 
Yeah, I remember seeing the commit for this. Simply forgot to engage
brain and either install the loader prior to rebooting on the new
kernel, or turning DMA off.

> You can get around this by stopping the loader before it boots the
> kernel (where it's typically counting down from 10), and do this:
> 
> 	set hw.ata.ata_dma=0
> 	boot
> 
> That will let the new kernel boot.  And then you can go ahead with the
> installworld.  The current version of the loader will be installed
> then, and this step shouldn't be necessary any more.
> 
This did the trick. Had I bothered to think things through, I would
have come to this conclusion. That's what I get for trying to wrap
things up early on a Friday :-)

> Scott was able to locate the problem that makes this necessary
> yesterday.
> It turned out to be something in the new ATA code after all and not
> what he had originally thought.  He sent the report of what's
> happening to Soeren, but it definitely won't be fixed by BETA2 which
> is building now...
> 
Cool. Thanks for the explanation.

Regards,
Chris Knight
Systems Administrator
E-Easy
Tel: +61 3 6334 9995  Fax: +61 3 6331 7032  Mob: +61 419 528 795
Web: http://www.e-easy.com.au