Re: [PATCH] umem: fix up unplugging.

[Tao Guo]

On Thu, Apr 5, 2012 at 9:19 PM, Jens Axboe <axboe@xxxxxxxxx> wrote:
> On 2012-04-05 22:09, Tao Guo wrote:
>> On Thu, Apr 5, 2012 at 4:51 PM, Jens Axboe <axboe@xxxxxxxxx> wrote:
>>> On 2012-04-04 17:58, Tao Guo wrote:
>>>> On Wed, Apr 4, 2012 at 4:18 PM, Jens Axboe <axboe@xxxxxxxxx> wrote:
>>>>> On 2012-04-04 16:20, Tao Guo wrote:
>>>>>> Hi Andrew,
>>>>>>
>>>>>> Thanks for your reply.
>>>>>>
>>>>>> Yes, without this patch the umem driver just doesn't work.
>>>>>> It is a bug introduced by commit 7eaceaccab5f40bbfda044629a6298616aeaed50.
>>>>>> In that patch, Jens removed the whole mm_unplug_device() function,
>>>>>> which used to be
>>>>>> the trigger to make umem start to work.
>>>>>
>>>>> Hmm indeed, that's isn't terribly useful. Why aren't we just calling
>>>>> activate_card() on addition of a bio?
>>>>>
>>>> It is the original design idea, the umem driver also wants to do batch
>>>> IO the same as
>>>> hard disk drive to fully utilize the IO bandwidth.
>>>
>>> Sure, just wondering whether the benefits have been tested. But yes, it
>>> certainly makes sense.
>>>
>>> Your patch looks buggy, though. What if the card is currently plugged on
>>> another tasks plug list? You seem to assume that it can only be plugged
>>> once.
>> No, it will register in its own plug callback list successfully, so
>> the unplug functions
>> will be called separately(twice the same).
>
> So the code loops through the current plug, checking if it's already on
> that list. If not, then it adds it. But what if it's on another plug
> list? The above loop could be done with just a
>
> Â Â Â Âif (!list_empty(&card->plug_cb.list))
> Â Â Â Â Â Â Â Â...
>
> instead. Right now you could have !list_empty(&card->plug_cb.list) but
> still add it to a different list.
>
mmm, I was stupid here: card->plug_cb can not be in two list the same
time. But I still need to make sure the final unplug will trigger the
activate() function.
I will sent another version to fix this problem.

Thanks,
-Tao
> --
> Jens Axboe
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/