On Wed, Aug 04, 2004 at 01:03:08PM +0300, Jukka Marin wrote:
> On Wed, Aug 04, 2004 at 11:28:26AM +0200, Manuel Bouyer wrote:
> > > Hmm.  My desktop system appeared completely dead (I didn't try to
> > > ping it, had no other machine here at that time).  XFree didn't work,
> > > mouse pointer froze etc.  After a few minutes the system was back
> > 
> > Maybe it was trying to swap
> 
> With 1 GB of RAM, I doubt it .. but I guess it's _possible_.

With UBC, I found that when the file cache is large enouth (the free ram
is below the target) it will always swap (or remove and then page in read-only
pages, such a text of executables). It's only a few pages, so it's
hardly noticable unless you have a drive dedicated to system+swap, and
look at 'systat vm'

> 
> > > alive.  The weird thing is that there were no messages in syslog
> > > about the problem.  I didn't even know this was a disk problem until
> > > the disk locked up so bad that I had to power-cycle the system.
> > 
> > No message in dmesg either ?
> 
> Nope.  Is there any loop in the driver with no timeout and no printf()
> where the IDE bus/disk could block?

No, there shouldn't be any with no timeout. But in some case the timeout
can be several 10s of seconds, and there won't be any printf if it doens't
expire.

Also, in some case, I've seen the box hang hard on disk problem, I suspect
the drive had locked up the PCI bus (in which case the CPU is also probably
hung, as memory access are blocked). But if you run into such a case I
don't think it would recover by itself.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--