On 4/26/21 10:00 AM, Ming Lei wrote:
> On Mon, Apr 26, 2021 at 09:19:20AM +0200, Hannes Reinecke wrote:
>> On 4/22/21 2:20 PM, Ming Lei wrote:
>>> Limit at most 8 queues are polled in each blk_pull(), avoid to
>>> add extra latency when queue depth is high.
>>>
>>> Reviewed-by: Jeffle Xu <jefflexu@xxxxxxxxxxxxxxxxx>
>>> Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx>
>>> ---
>>> block/blk-poll.c | 78 ++++++++++++++++++++++++++++++++++--------------
>>> 1 file changed, 55 insertions(+), 23 deletions(-)
>>>
>>> diff --git a/block/blk-poll.c b/block/blk-poll.c
>>> index 249d73ff6f81..20e7c47cc984 100644
>>> --- a/block/blk-poll.c
>>> +++ b/block/blk-poll.c
>>> @@ -288,36 +288,32 @@ static void bio_grp_list_move(struct bio_grp_list *dst,
>>> src->nr_grps -= cnt;
>>> }
>>>
>>> -static int blk_mq_poll_io(struct bio *bio)
>>> +#define POLL_HCTX_MAX_CNT 8
>>> +
>>> +static bool blk_add_unique_hctx(struct blk_mq_hw_ctx **data, int *cnt,
>>> + struct blk_mq_hw_ctx *hctx)
>>> {
>>> - struct request_queue *q = bio->bi_bdev->bd_disk->queue;
>>> - blk_qc_t cookie = bio_get_poll_data(bio);
>>> - int ret = 0;
>>> + int i;
>>>
>>> - /* wait until the bio is submitted really */
>>> - if (!blk_qc_t_ready(cookie))
>>> - return 0;
>>>
>>> - if (!bio_flagged(bio, BIO_DONE) && blk_qc_t_valid(cookie)) {
>>> - struct blk_mq_hw_ctx *hctx =
>>> - q->queue_hw_ctx[blk_qc_t_to_queue_num(cookie)];
>>> + for (i = 0; i < *cnt; i++) {
>>> + if (data[i] == hctx)
>>> + goto exit;
>>> + }
>>>
>>> - ret += blk_mq_poll_hctx(q, hctx);
>>> + if (i < POLL_HCTX_MAX_CNT) {
>>> + data[i] = hctx;
>>> + (*cnt)++;
>>> }
>>> - return ret;
>>> + exit:
>>> + return *cnt == POLL_HCTX_MAX_CNT;
>>> }
>>>
>>> -static int blk_bio_poll_and_end_io(struct bio_grp_list *grps)
>>> +static void blk_build_poll_queues(struct bio_grp_list *grps,
>>> + struct blk_mq_hw_ctx **data, int *cnt)
>>> {
>>> - int ret = 0;
>>> int i;
>>>
>>> - /*
>>> - * Poll hw queue first.
>>> - *
>>> - * TODO: limit max poll times and make sure to not poll same
>>> - * hw queue one more time.
>>> - */
>>> for (i = 0; i < grps->nr_grps; i++) {
>>> struct bio_grp_list_data *grp = &grps->head[i];
>>> struct bio *bio;
>>> @@ -325,11 +321,31 @@ static int blk_bio_poll_and_end_io(struct bio_grp_list *grps)
>>> if (bio_grp_list_grp_empty(grp))
>>> continue;
>>>
>>> - for (bio = grp->list.head; bio; bio = bio->bi_poll)
>>> - ret += blk_mq_poll_io(bio);
>>> + for (bio = grp->list.head; bio; bio = bio->bi_poll) {
>>> + blk_qc_t cookie;
>>> + struct blk_mq_hw_ctx *hctx;
>>> + struct request_queue *q;
>>> +
>>> + if (bio_flagged(bio, BIO_DONE))
>>> + continue;
>>> +
>>> + /* wait until the bio is submitted really */
>>> + cookie = bio_get_poll_data(bio);
>>> + if (!blk_qc_t_ready(cookie) || !blk_qc_t_valid(cookie))
>>> + continue;
>>> +
>>> + q = bio->bi_bdev->bd_disk->queue;
>>> + hctx = q->queue_hw_ctx[blk_qc_t_to_queue_num(cookie)];
>>> + if (blk_add_unique_hctx(data, cnt, hctx))
>>> + return;
>>> + }
>>> }
>>> +}
>>> +
>>> +static void blk_bio_poll_reap_ios(struct bio_grp_list *grps)
>>> +{
>>> + int i;
>>>
>>> - /* reap bios */
>>> for (i = 0; i < grps->nr_grps; i++) {
>>> struct bio_grp_list_data *grp = &grps->head[i];
>>> struct bio *bio;
>>> @@ -354,6 +370,22 @@ static int blk_bio_poll_and_end_io(struct bio_grp_list *grps)
>>> }
>>> __bio_grp_list_merge(&grp->list, &bl);
>>> }
>>> +}
>>> +
>>> +static int blk_bio_poll_and_end_io(struct bio_grp_list *grps)
>>> +{
>>> + int ret = 0;
>>> + int i;
>>> + struct blk_mq_hw_ctx *hctx[POLL_HCTX_MAX_CNT];
>>> + int cnt = 0;
>>> +
>>> + blk_build_poll_queues(grps, hctx, &cnt);
>>> +
>>> + for (i = 0; i < cnt; i++)
>>> + ret += blk_mq_poll_hctx(hctx[i]->queue, hctx[i]);
>>> +
>>> + blk_bio_poll_reap_ios(grps);
>>> +
>>> return ret;
>>> }
>>>
>>>
>> Can't we make it a sysfs attribute instead of hard-coding it?
>> '8' seems a bit arbitrary to me, I'd rather have the ability to modify it...
>
> I'd rather not add such code in the feature 'enablement' stage since I doesn't
> observe the number plays a big role yet. It is added for holding hw queues to
> be polled on stack variables, also avoid to add too much latency if there is
> too many bios from too many hw queues to be reaped.
>
> Also the actual polled hw queues can be observed easily via bpftrace, so debug
> purpose from sysfs isn't necessary too.
>
Okay. You can add my
Reviewed-by: Hannes Reinecke <hare@xxxxxxx>
Cheers,
Hannes
--
Dr. Hannes Reinecke Kernel Storage Architect
hare@xxxxxxx +49 911 74053 688
SUSE Software Solutions Germany GmbH, 90409 Nürnberg
GF: F. Imendörffer, HRB 36809 (AG Nürnberg)
