ATA problems again ... general problem of ICH7 or ATA?

[Greg Byshenk]

On Sun, Aug 20, 2006 at 01:38:55PM +0100, Matt Dawson wrote:
> On Sunday 20 August 2006 13:00, freebsd-stable-request at freebsd.org wrote:

> > Do you mean different type of cables, or just another piece? I can't
> > change cables by myself, servers are dedicated from provider, but as I
> > can saw, they picked whole new machine from their HW storage and put new
> > Samsung disk drives in. So these two last machines are brand new with
> > new cables. (Probably with a same type of cables - all machines are ASUS
> > RS120)
 
> I can confirm the same behaviour with a ULi M1689/Newcastle Athlon64 based 
> system running 6.1-RELEASE-p3 (i386). ad6 just detaches without warning and 
> it takes a reboot to bring it back. atacontrol reinit has no effect. Tried 
> the following to resolve the problems:
 
> - Changed cables (both ad4 and ad6)
> - Changed SATA power to legacy
> - Moved the NIC and anything else from the shared PCI INT (thought I'd cracked 
> it at this point as it was stable for a month, then it lost ad6 on a nightly 
> dump)
> - Remade my gmirror array as an ar. Put it straight back to gmirror again when 
> I found out what a pain it is to rebuild after ad6 disappears.

I am not sure if it is related, but...  I experienced a similar sort of
problem, although the details in my case are quite different.

What was similar was that I would "lose" two ATA drives from an array,
inexplicably.  Reconfiguring the same drives and rebuilding would cause
them to work perfectly again -- for some number of days, after which 
the same failure would occur.

What is different is that this was with a 3Ware RAID controller -- 
which made removing/raconfiguring/rebuilding much easier -- but I was
seeing the exact same errors.

This happened four times (with the same errors that have been discussed
here), running 6.1 STABLE as of June 22.  Before attempting to RMA the
drives, I tried an updated kernel, 6.1 STABLE as of July 19.  Strangely
enough, the problems disappeared.

So, while I have not checked everything that has changed, it _might_ be
worth trying 6.1 STABLE...
 

-- 
greg byshenk  -  gbyshenk at byshenk.net  -  Leiden, NL