Re: [PATCH] block: add timer on blkdev_dequeue_request() not elv_next_request()

[Tejun Heo <tj@kernel.org>]

Mike Anderson wrote:
> Jens Axboe <jens.axboe@oracle.com> wrote:
>> On Thu, Oct 30 2008, Tejun Heo wrote:
>>> Jens Axboe wrote:
>>>> That's actually a pretty dumb error, I'm surprised it hasn't reared its
>>>> ugly face in more ways. Presumably because the timeout is usually so
>>>> long, that we'll get to actually issuing and completing it within the
>>>> normal timeout anyway.
>>> Heh... it showed its ugly face in many different ways while I was
>>> playing with PMP connected via a very long eSATA cable.
>> Ah :-)
>>
>> If we had it wired up for eg the old IDE drivers, it would have shown up
>> quite quickly as well I think.
> 
> I am getting errors now and my system will not boot up. The system is
> connected to storage with active / passive paths. If we are doing a
> BLKPREP_KILL we will call elv_dequeue_request which will add the
> timer for the request we are killing. 
> 
> The attached patch is a quick patch to work around my issue, but we
> probably need something better. I would like to run some short timeout
> testing on it for a while (though that previously did not catch Tejun's
> issue). I will look at this more tomorrow unless someone beats me to it.

Aieee... I'm now trying to clean up the command fetch/completion
interface.  Please give me a day or two.  I'll clean it all up.

Thanks.

-- 
tejun