Re: [PATCH] block: Make blk_drain_queue() work for stopped queues

[Bart Van Assche <bvanassche@acm.org>]

On 03/21/12 03:37, Dan Williams wrote:
> On Tue, Mar 20, 2012 at 2:01 PM, Dan Williams <dan.j.williams@intel.com> wrote:
>> On Tue, Mar 20, 2012 at 1:06 PM, Bart Van Assche <bvanassche@acm.org> wrote:
>> [..]
>>> - Fix a null pointer dereference triggered by sd during device removal.
>> Hi Bart,
>>
>> Do you have a log of the backtrace in this case?  I'm going to put
>> this patch into our libsas/isci test environment.
> We beat on this patch pretty severely in our environment and appeared
> to only trigger a hung_task timeout when our driver / libata took too
> long to recovery for a 15 device unplug.

Thanks for testing - that's appreciated.

The null pointer dereference triggered during device removal was
originally reported by Jun'ichi Nomura. A call stack can be found here:
http://www.spinics.net/lists/linux-scsi/msg56254.html.

Regarding invoking blk_cleanup_queue() on a stopped queue: some code I
was testing could trigger this. But as far as I can see both the fc and
iSCSI transport layer code take care to unblock a queue before
destroying it, so these transports are not affected. There are other
(non-SCSI) block drivers though that can stop and restart the queue. I
haven't analyzed all of them. So I'm not sure there is currently any
upstream code that invokes blk_cleanup_queue() on a stopped queue.

Bart.