Re: [PATCH 2/5] md: the latest try for improve io stats accounting

[Guoqing Jiang <jgq516@gmail.com>]

On 5/18/21 6:12 PM, Artur Paszkiewicz wrote:
> On 18.05.2021 07:32, Guoqing Jiang wrote:
>> +     /*
>> +      * We don't clone bio for multipath, raid1 and raid10 since we can reuse
>> +      * their clone infrastructure.
>> +      */
>> +     if (blk_queue_io_stat(bio->bi_bdev->bd_disk->queue) &&
>> +         (bio->bi_pool != &mddev->md_io_bs) &&
>> +         (mddev->level != 1) && (mddev->level != 10) &&
>> +         (mddev->level != LEVEL_MULTIPATH)) {
> Maybe add a flag to struct md_personality and check it here? Something
> that will be set only for the personalities which clone the bio
> themselves.

Good point.

> Doesn't this need to check the bio->bi_pool also against mddev->bio_set
> to skip the bios split by md? Similarly to the check against
> bio_chain_endio which you did before.

Hmm, raid0 allocates split bio from mddev->bio_set, but raid5 is
different, it splits bio from r5conf->bio_split. So either let raid5 also
splits bio from mddev->bio_set, or add an additional checking for
raid5. Thoughts?

Thanks,
Guoqing