[dm-devel] [PATCH] dm-mpath: requeue I/O during pg_init

[Junichi Nomura]

On 11/05/13 22:45, Hannes Reinecke wrote:
> On 11/05/2013 02:31 PM, Alasdair G Kergon wrote:
>> On Tue, Nov 05, 2013 at 02:10:55PM +0100, Hannes Reinecke wrote:
>>> If this is just a left-over from the original port to request-based
>>> (for bio-based we _have_ to queue internally as there's no request
>>> queue to be had), fine, we should be removing it.
>>
>> I think that is the case.

Yes. That's the case.
Kiyoshi and I was removing it but couldn't take time to audit the isolation
of pg_init state machine from the process_queued_ios.

>>> But there _might_ be some corner cases which require us to do internal
>>> queueing.
>>  
>> We *only* add I/O to the internal queue in map_io() - which can always
>> be replaced with REQUEUE,  As long as we still 'wake up' the queue
>> immediately when we are ready to receive the I/O, I can't think of any
>> other reason.  And it would let us remove quite a bit of tricky code!
>>
> Precisely what I was thinking.
> 
> I'll be cobbling together a patch.

-- 
Jun'ichi Nomura, NEC Corporation