Subject: Re: About 'SIR Reset'
To: Pascal Cabaud <pc@eila.jussieu.fr>
From: Eduardo Horvath <eeh@netbsd.org>
List: port-sparc64
Date: 11/28/2005 17:27:16
On Fri, 25 Nov 2005, Pascal Cabaud wrote:

> Hi,
>
> I experimented 'SIR Reset' on an Ultra1e. Before searching on Google, I
> noticed that I/Os and 'SIR reset' freezes were correlated. On mail
> archives, I read "your cpu" or "your mb is gone"...
>
> *In my case*, I've all the reasons to think that it is the disk. On the
> same host with another disk, no 'SIR Reset' even after dumping the FS.
> On an Ultra10 with the same disk, 'SIR Reset' freezes appeared after
> hundreds MB of tar... On another Ultra1, same freezes with the disks
> after some big I/Os. With others disks, all these hosts are OK under the
> same conditions.
>
> The OS tested was NetBSD/sparc64 (2.0 and 2.1) but it doesn't care AFAIU.

SIR is a Software Initiated Reset.  I sprinkled some of those throughout 
locore.s primarily for diagnostic reasons in places where the machine 
state gets so stuffed up that it can't recover.

To find out what's really happening, after you get the SIR you need to 
type `.trap-registers' and write down the TPC values.  Then boot the 
kernel and see which TPC points to an SIR instruction.  Then correlate 
that with the code in locore.s and look at the comments around the SIR to 
determine what really went wrong.

Eduardo