Thread (5 messages) 5 messages, 2 authors, 2009-10-14

Re: MVSAS 1669:mvs_abort_task:rc= 5

From: Thomas Fjellstrom <hidden>
Date: 2009-10-11 22:59:01
Also in: linux-scsi, lkml

On Sun October 11 2009, Christian Vilhelm wrote:
Thomas Fjellstrom wrote:
quoted
Hi,

I've been trying to get an AOC-SASLP-MV8 card (pcie x4 2 port SAS card)
to work with linux for the past month or so. I've recently just RMAed my
first card, and tested the new one under linux, and I see the same
problems.

The very first time I made a new array off the controller, formated (with
xfs) and mounted the volume, it seemed to work. ioozone even seemed to
run for a while. Sadly after a few minutes I got a stream of
mvs_abort_task messages in dmesg, and any accesses to the volume, or any
disks connected to the controller lock up.

After that I updated my 2.6.31 kernel to 2.6.32-rc3-git2 off of
kernel.org, and the volume fails to mount with the same mvs_abort_task
messages.
I have the exact same problem with another Marvell 88SE64xx based card,
namely an Areca ARC-1300ix-16 and the mvsas driver.
If the disks are just used alone, with a filesystem on them, all seems
to work fine. dd and badblocks run fine on them. Mounting them,
reading/writing work fine. The error seem to popup but rarely when
several disks are used simultaneously.
But, an absolute sure way to trigger the error is to assemble (or
create) a md raid array with the disks. I join a syslog extract from the
error. You can see it happens seconds after the array creation.
I tried :
1) disabling the write cache on the disks => same error
2) disabling NCQ : in mv_sas.h :
#define MV_DISABLE_NCQ  1
same error.
Afer a while, the devices handled by the card are just dropped from the
system and the card stops working at all, a reboot is necessary.
I have found that a proper reboot is impossible once the card/driver starts 
misbehaving. Anything that tries to do anything with the md device, or any of 
the component drives will hang. Even kernel threads it seems. A reboot or a 
shutdown hangs when it tries to sync the md device, and ALT+SYSRQ+S/U both 
hang. After the first Alt+sysrq+s it will register more of them, but it won't 
print the "Emergency Sync Complete" message.
Does anyone have a working config based on a Marvell 64xx card ?

I'm willing to explore solutions, patches or anything, just tell me what
to do to help.

Christian Vilhelm.

-- 
Thomas Fjellstrom
tfjellstrom@shaw.ca
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help