MPT2SAS deadlocking on failed disk

MPT2SAS deadlocking on failed disk

[Hubert Tonneau]

Hi,

In stock Linux 2.6.35.7 kernel,
with mpt2sas driver connected to two disks (no raid at controler level),
and Linux MD software RAID1,

if I hot unplug one of the disks,
instead of the failed disk beeing removed from the MD raid set,
I get locked disk subsystem with all applications trying to write to
the disks beeing locked. No kernel message related to disks, only
to applications beeing locked.

This bug is probably not related specificaly to kernel 2.6.35.7
because I had not used LSI controlers for years.

Regards,
Hubert Tonneau

Re: MPT2SAS deadlocking on failed disk

[Louis-David Mitterrand]

On Sun, Oct 24, 2010 at 11:36:18AM +0000, Hubert Tonneau wrote:
> Hi,
> 
> In stock Linux 2.6.35.7 kernel,
> with mpt2sas driver connected to two disks (no raid at controler level),
> and Linux MD software RAID1,
> 
> if I hot unplug one of the disks,
> instead of the failed disk beeing removed from the MD raid set,
> I get locked disk subsystem with all applications trying to write to
> the disks beeing locked. No kernel message related to disks, only
> to applications beeing locked.
> 
> This bug is probably not related specificaly to kernel 2.6.35.7
> because I had not used LSI controlers for years.

FWIW I had horrible problems on 2.6.35.x (grml) with the mpt2sas driver
(like running lilo would eject most disks in a running raid6).

All these problems went away with 2.6.36.

Re: MPT2SAS deadlocking on failed disk

[Hubert Tonneau]

Hi,

> Louis-David Mitterrand wrote:
>
> FWIW I had horrible problems on 2.6.35.x (grml) with the mpt2sas driver
> (like running lilo would eject most disks in a running raid6).
>
> All these problems went away with 2.6.36.

I did the same experiment on 2.6.36: the result is the same.
What is still not clear is: is the problem in MD software RAID, or in mpt2sas ?

Regards,
Hubert Tonneau


Hubert Tonneau wrote:
>
> Hi,
> 
> In stock Linux 2.6.35.7 kernel,
> with mpt2sas driver connected to two disks (no raid at controler level),
> and Linux MD software RAID1,
> 
> if I hot unplug one of the disks,
> instead of the failed disk beeing removed from the MD raid set,
> I get locked disk subsystem with all applications trying to write to
> the disks beeing locked. No kernel message related to disks, only
> to applications beeing locked.
> 
> This bug is probably not related specificaly to kernel 2.6.35.7
> because I had not used LSI controlers for years.
> 
> Regards,
> Hubert Tonneau