ZFS + replacing failing hard-drive.

[Rick C. Petty]

On Thu, Apr 19, 2007 at 09:10:48PM +0200, Dag-Erling Sm?rgrav wrote:
> "Rick C. Petty" <rick-freebsd at kiwi-computer.com> writes:
> > What do you mean by recently?  I've seen this problem which started around
> > 5.4-RELEASE (perhaps earlier) and on, including 6.0-R thru 6.2-stable as of
> > a few weeks ago.  Would the interrupt filtering be present on these
> > systems?
> 
> No, and in fact it's not used by the ata driver in -CURRENT, so it's
> probably not relevant.
> 
> I said recently because I didn't start having these problems until
> around early march.  However, this could be due to changes in usage
> patterns.

Oh.  :(  Okay.  I had hoped otherwise...

I'm in the process of looking at the commands sent to the PDC chips from
linux's sata_promise driver and comparing what's different about ours.

Linux properly reinitialized the card (tested with rmmod & modprobe), but
the command which is causing the drives to be reprobed may affect the whole
card vs. a single channel.  I'll have time to play around with the FreeBSD
kernel this weekend.  Unfortunately, I couldn't figure out a way in linux
to drop just a channel and reattach it, and I didn't see anything in their
code which handles this case.  However, I'd rather have the whole Promise
card pause while reinitializing all channels than not be able to bring up a
single channel.

Still, it would be useful to figure out the cause of the DMA failures and
avoid this altogether.  Anyone volunteering to help?  sos@ ?

-- Rick C. Petty