Re: mdraid causing mvsas to lockup? (was: Re: recommended 4port SATA controller ?)

[Thomas Fjellstrom <tfjellstrom@shaw.ca>]

On Fri September 18 2009, Thomas Fjellstrom wrote:
> On Thu September 17 2009, Thomas Fjellstrom wrote:
> > On Thu September 17 2009, Kristleifur Daðason wrote:
> > > On Thu, Sep 17, 2009 at 11:02 PM, Thomas Fjellstrom
> > > <tfjellstrom@shaw.ca>
> >
> > wrote:
> > > > On Thu September 17 2009, John Bridges wrote:
> > > >> I'm a fan of the SuperMicro AOC-SAT2-MV8, great card.
> > > >> http://www.supermicro.com/products/accessories/addon/AOC-SAT2-MV8.cf
> > > >>m
> > > >>
> > > >> It's an 8 port PCI-X card, works in both PCI and PCI-X slots.
> > > >>
> > > >> SATA2
> > > >>
> > > >> Drivers for Linux are stable, built in.
> > > >
> > > > Have you had any experience with the AOC-SASLP-MV8? I've got one and
> > > > have been having no end of issues with it under linux.
> > > >
> > > > --
> > > > Thomas Fjellstrom
> > > > tfjellstrom@shaw.ca
> > > > --
> > >
> > > I have,
> > >
> > > or rather, I've tried to get an AOC-SASLP-MV8 card going. I think I
> > > can safely say that at least Linux kernel 2.6.31 is a requirement. The
> > > card was basically useless with everything up to 2.6.30, then I tried
> > > 2.6.31-rc5 on a whim and it kicked in. Built-in driver support, that
> > > is. However it wasn't stable, it dropped disks when syncing a large
> > > array. I've been meaning to test on 2.6.31 final, and am pretty
> > > optimistic.
> >
> > Yeah, the driver didn't appear till .30. I have 2.6.31-git4 installed
> > right now, and no matter what I do, the controller starts spewing errors:
> >
> > [ 1455.698186] drivers/scsi/mvsas/mv_sas.c 1669:mvs_abort_task:rc= 5
> > [ 1455.698196] drivers/scsi/mvsas/mv_sas.c 1608:mvs_query_task:rc= 5
> > ...
> > [ 1424.708085] end_request: I/O error, dev sdh, sector 3072
> > [ 1424.708106] sd 0:0:3:0: [sdh] Unhandled error code
> > [ 1424.708111] sd 0:0:3:0: [sdh] Result: hostbyte=DID_OK
> > driverbyte=DRIVER_TIMEOUT
> > [ 1424.708118] sd 0:0:3:0: [sdh] CDB: Read(10): 28 00 00 00 08 00 00 04
> > 00 00
> >
> > And thats with perfectly good disks, and with smartd/hddtemp disabled
> > (they were causing one of my disks to barf).
> >
> > All I have to do is start a read from any disk, and after a few minutes,
> >  the card starts erroring out, and then dies.
> >
> > It actually seems like it got more unstable from .30 to .31.
> >
> > I've been trying to get some help with it on the lkml/ide/scsi lists for
> > a while now, one person has tried to help, but thats about it.
> 
> Very strange. I've found that reading from all 4 drives currently connected
>  to the controller at once, works. I have 4 dd commands, one reading off
>  each drive, and so far no errors, the dd commands aren't locking up, and
>  they are going full speed (120MB/s per drive).
> 
> If however I attempt to bring up the md raid0 array ontop of these disks,
>  the controller locks up, and all of the disks become inaccessible.
> 
> Maybe it has something to do with it, but just as the system is booting, I
>  get the following, maybe related, maybe not:
> 
> ata_id[5183]: HDIO_GET_IDENTITY failed for '/dev/block/8:96'
> ata_id[5188]: HDIO_GET_IDENTITY failed for '/dev/block/8:112'
> ata_id[5184]: HDIO_GET_IDENTITY failed for '/dev/block/8:80'
> 
> (those map to sdg, sdh, and sdf in that order, no report for sde, the first
> disk in the controller)
> 

So I've let the controller and disks sit all day after finishing a full read 
test (dd if=/dev/sd[efgh] of=/dev/null bs=8M) with all four 1TB drives going 
at the same time, and I've had no errors at all. All four dd commands finished 
without error, and went at full speed.

If I attempt to activate an md raid0 array ontop of any disks on this 
controller the controller starts having a fit, and all disks are inaccessible 
till a hard reset (the machine won't fully reboot, or turn off, as the 
"flushing scsi cache" or "shutting down LVM" steps will hang waiting on drives 
on the wedged controller.

I would really like to get this fixed, if there's anything more I can do to 
help narrow down the problem further, I'll do my best.

-- 
Thomas Fjellstrom
tfjellstrom@shaw.ca
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html