Re: [PATCH V14 13/24] mmc: block: Add blk-mq support

[Ulf Hansson]

[...]

>>>>> +
>>>>> +enum mmc_issued mmc_blk_mq_issue_rq(struct mmc_queue *mq, struct request *req)
>>>>> +{
>>>>> +       struct mmc_blk_data *md = mq->blkdata;
>>>>> +       struct mmc_card *card = md->queue.card;
>>>>> +       struct mmc_host *host = card->host;
>>>>> +       int ret;
>>>>> +
>>>>> +       ret = mmc_blk_part_switch(card, md->part_type);
>>>>
>>>> What if there is an ongoing request, shouldn't you wait for that to
>>>> complete before switching partition?
>>>
>>> Two requests on the same queue cannot be on different partitions because we
>>> have a different queue (and block device) for each partition.
>>
>> That's not true for RPMB anymore I am afraid.
>>
>> RPMB shares the same queue as for the main eMMC partition, which is
>> because we strive towards fair I/O scheduling across the hole device.
>
> I hadn't thought of RPMB, but I think the logic is OK, which is good because
> it is the same as we presently have.  Here the md->part_type will be the
> main area even for RPMB.  So this switch won't do anything if we have a
> request in flight.  Then inside __mmc_blk_ioctl_cmd() the switch to RPMB is
> done, and afterwards mmc_blk_issue_drv_op() switches it back again.

Yes, you are right! No worries then!

[...]

>>>
>>> You are right, this logic does not support parallel dispatches.
>>>
>>
>> This do raises a question, don't you think it would be beneficial,
>> especially for CQE to allow parallel dispatches?
>>
>> I am not saying we should change this at this point, just that we may
>> consider changing this for future improvements.
>
> I think the benefit is limited because the time to dispatch a request is
> small compared with the time to complete a request. i.e. a number of
> requests can be queued before the first one has completed.  But yes, it is
> something to keep in mind.

Yeah, let's leave this for future considerations.

Kind regards
Uffe