> Not sure it's related, but it sounds like the hotplug failures being
> seen with libsas hotplug.  Have not had a chance yet to track it down
> further with isci, but mvsas developer Xiangliang Yu is reporting:

It looks somewhat different: he has a bad pointer, not a NULL pointer,
and it's in elv_completed_request() (but that may be related to 
the different request types)

> 
> http://marc.info/?l=linux-scsi&m=130707166512002&w=2
> 
> He tracked it down to a suspected regression between .39-rc4 and
> .39-rc5 but did not finalize the bisect.

I can check if it disappears with .39-rc4.

> The isci driver sees this and other signatures all seemingly related
> to early device tear down and only when pulling a drive with in-flight
> i/o.

In my case it's not really in flight IO, but umount after pull with an
idle drive.

-Andi