MFI Driver Behavior

[Erich Weiler]

Hi All - I've posted on the forums but no one seems to have any ideas... 
  I have an odd lock up issue with a Perc H800 controller under 
8.2-PRERELEASE.

We have a FreeBSD server running:

Code:

FreeBSD 8.2-PRERELEASE (GENERIC) #0: Thu Dec 16 14:59:46 PST 2010

It's a Dell R610. It has two MD1200 disk arrays on it, SAS chained 
together. The controller that manages them is a Perc H800, with the 
latest firmware available.

I have the disks exported JBOD from the controller. And, the disks are 
roped into a ZFS filesystem, which is exported via NFS to the local net.

Everything works well most of the time, but every once in a while (like 
once every few days), the filesystem completely hangs and we see these 
errors on the console:

mfi0: COMMAND 0xffffff80009b5870 TIMEOUT AFTER 61793 SECONDS
mfi0: COMMAND 0xffffff80009b5870 TIMEOUT AFTER 61823 SECONDS
mfi0: COMMAND 0xffffff80009b5870 TIMEOUT AFTER 61853 SECONDS
mfi0: COMMAND 0xffffff80009b5870 TIMEOUT AFTER 61923 SECONDS

(this is after the filesystem has been hung for a day)

etc... When I start poking around with mfiutil, it shows everything is 
OK, the disks are all OK, the volumes are good, the event logs show no 
errors. The "Patrol" feature is disabled. The battery is fine.

After running "mfiutil show volumes", the lockup magically frees itself. 
But, I don't want them to happen in the first place, and I certainly 
don't want to have to manually run a "mfiutil show volumes" or whatever 
to unlock it every time. Has anyone seen this before?

I've actually tried another H800 controller we had on the shelf as well, 
just to rule out a hardware problem with the first one, but we see the 
same behavior on both controllers.

"zpool status" also shows the disks as all OK, and a "zpool scrub" turns 
up no problems.

Any insight much appreciated!!  Since multiple controllers exhibit the 
same behavior, I was thinking it's falling more into a driver issue at 
this point.  I hope I'm right!  I emailed the author of the MFI driver 
for FreeBSD, but have not heard anything back, so I was hoping someone 
here would have an idea of where I could turn next.

If even there was a way I could determine what the "0xffffff80009b5870" 
MFI command is, that would be a big help, so I would have a better idea 
of where to continue my investigations.

-erich