On Mar 11, 2007, at 10:34 PM, Aaron J. Grier wrote:

> On Sun, Mar 11, 2007 at 03:16:11PM +0100, Johnny Billquist wrote:
>> I decided to sit down and look a bit more as well.
>> I haven't traced it down yet, but I think we can skip the MI scsi  
>> stuff.
>>  In fact, I think we can skip the scsi totally.
>>
>> From what I can see, it appears we have a problem with the callout
>> functions in the vax port. If I register a callout, it never gets
>> called. Either there is something that have broken in the timer
>> handling, or we're locking out the timer interrupts, I'd guess.
>> I'll see if I have some more time later tonight to look at this.
>>
>> Or does anyone know what might have happened recently?
>
> I ran into this on alpha.
>
> it ended up being the bool-ization (TRUE -> true) in the kernel  
> sources
> broke some machine-dependent handling of UVM_PAGE_IDLE_ZERO.
>
> http://mail-index.netbsd.org/port-alpha/2007/02/24/0000.html
>
> scsibus probing just happens to be the first place in bootup where the
> idle loop is called.
>
> I assume the vax fix will be similarly trivial.

Completely different problem.  This was due a latent bug in the softintr
dispatch that was revealed by the removal of spllowersoftclock.