NFS or ATA driver causes FS corruption?
Bjoern A. Zeeb
bzeeb-lists at lists.zabbadoz.net
Sun May 9 09:30:12 PDT 2004
On Sun, 9 May 2004, Adriaan de Groot wrote:
> On Saturday 08 May 2004 20:25, Bjoern A. Zeeb wrote:
> > On Mon, 1 Mar 2004, it was written:
> > > Kris Kennaway wrote:
> > > > ad0: WARNING - WRITE_DMA interrupt was seen but timeout fired LBA=9440
> > > > ad0: WARNING - WRITE_DMA interrupt was seen but timeout fired LBA=20904
> > >
> > > The above means that *something* is stomping on the taskqueue that
> > > should take care of returning finished requests to the system (they are
> I see this (timeouts fired) as well on my Asus K8V with a single S-ATA disk;
> it happens only when the machine is ridiculously loaded (like feeding a
> 5500-message mbox file into sa-learn while doing two different make -j6
> compiles and also cvsupping the FBSD tree along with another 2G source repo).
> Haven't noticed any averse effects, though - the machine chokes for 30
> seconds or so and then carries on.
I have to hard reset the machine here :(
I can reproduce it with not too much IO. copying sources; make
I tried to do everything mentioned in the errata for 5.2(.1) but
nothing helped. It also happened with DMA disbaled (an d I thing it
logged PIO errors at that time).
I am now going to build a world from NFS src to NFS obj to not
have ATA traffic and to slow down things so this will not happen
[ obj dir could be oon the same machine I suspect but the installworld
will then most likely make the machine go berserk again ].
Are you running Release or HEAD ?
If this isn't fix in HEAD yet I am very interested in every
patches/things to try to fix this. The machine in question should
become my new in house server and I do not like another IO problem
once I night when backup is running...
Bjoern A. Zeeb bzeeb at Zabbadoz dot NeT
56 69 73 69 74 http://www.zabbadoz.net/
More information about the freebsd-amd64