On Tue, Jan 26, 2021 at 06:26:02AM +0000, Damien Le Moal wrote:
> On 2021/01/26 15:07, Ming Lei wrote:
> > On Tue, Jan 26, 2021 at 04:06:06AM +0000, Damien Le Moal wrote:
> >> On 2021/01/26 12:58, Ming Lei wrote:
> >>> On Tue, Jan 26, 2021 at 10:32:34AM +0900, Changheun Lee wrote:
> >>>> bio size can grow up to 4GB when muli-page bvec is enabled.
> >>>> but sometimes it would lead to inefficient behaviors.
> >>>> in case of large chunk direct I/O, - 32MB chunk read in user space -
> >>>> all pages for 32MB would be merged to a bio structure if the pages
> >>>> physical addresses are contiguous. it makes some delay to submit
> >>>> until merge complete. bio max size should be limited to a proper size.
> >>>>
> >>>> When 32MB chunk read with direct I/O option is coming from userspace,
> >>>> kernel behavior is below now. it's timeline.
> >>>>
> >>>> | bio merge for 32MB. total 8,192 pages are merged.
> >>>> | total elapsed time is over 2ms.
> >>>> |------------------ ... ----------------------->|
> >>>> | 8,192 pages merged a bio.
> >>>> | at this time, first bio submit is done.
> >>>> | 1 bio is split to 32 read request and issue.
> >>>> |--------------->
> >>>> |--------------->
> >>>> |--------------->
> >>>> ......
> >>>> |--------------->
> >>>> |--------------->|
> >>>> total 19ms elapsed to complete 32MB read done from device. |
> >>>>
> >>>> If bio max size is limited with 1MB, behavior is changed below.
> >>>>
> >>>> | bio merge for 1MB. 256 pages are merged for each bio.
> >>>> | total 32 bio will be made.
> >>>> | total elapsed time is over 2ms. it's same.
> >>>> | but, first bio submit timing is fast. about 100us.
> >>>> |--->|--->|--->|---> ... -->|--->|--->|--->|--->|
> >>>> | 256 pages merged a bio.
> >>>> | at this time, first bio submit is done.
> >>>> | and 1 read request is issued for 1 bio.
> >>>> |--------------->
> >>>> |--------------->
> >>>> |--------------->
> >>>> ......
> >>>> |--------------->
> >>>> |--------------->|
> >>>> total 17ms elapsed to complete 32MB read done from device. |
> >>>>
> >>>> As a result, read request issue timing is faster if bio max size is limited.
> >>>> Current kernel behavior with multipage bvec, super large bio can be created.
> >>>> And it lead to delay first I/O request issue.
> >>>>
> >>>> Signed-off-by: Changheun Lee <nanich.lee@xxxxxxxxxxx>
> >>>> ---
> >>>> block/bio.c | 17 ++++++++++++++++-
> >>>> include/linux/bio.h | 4 +++-
> >>>> include/linux/blkdev.h | 3 +++
> >>>> 3 files changed, 22 insertions(+), 2 deletions(-)
> >>>>
> >>>> diff --git a/block/bio.c b/block/bio.c
> >>>> index 1f2cc1fbe283..ec0281889045 100644
> >>>> --- a/block/bio.c
> >>>> +++ b/block/bio.c
> >>>> @@ -287,6 +287,21 @@ void bio_init(struct bio *bio, struct bio_vec *table,
> >>>> }
> >>>> EXPORT_SYMBOL(bio_init);
> >>>>
> >>>> +unsigned int bio_max_size(struct bio *bio)
> >>>> +{
> >>>> + struct request_queue *q;
> >>>> +
> >>>> + if (!bio->bi_disk)
> >>>> + return UINT_MAX;
> >>>> +
> >>>> + q = bio->bi_disk->queue;
> >>>> + if (!blk_queue_limit_bio_size(q))
> >>>> + return UINT_MAX;
> >>>> +
> >>>> + return blk_queue_get_max_sectors(q, bio_op(bio)) << SECTOR_SHIFT;
> >>>> +}
> >>>> +EXPORT_SYMBOL(bio_max_size);
> >>>> +
> >>>> /**
> >>>> * bio_reset - reinitialize a bio
> >>>> * @bio: bio to reset
> >>>> @@ -877,7 +892,7 @@ bool __bio_try_merge_page(struct bio *bio, struct page *page,
> >>>> struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
> >>>>
> >>>> if (page_is_mergeable(bv, page, len, off, same_page)) {
> >>>> - if (bio->bi_iter.bi_size > UINT_MAX - len) {
> >>>> + if (bio->bi_iter.bi_size > bio_max_size(bio) - len) {
> >>>> *same_page = false;
> >>>> return false;
> >>>> }
> >>>
> >>> So far we don't need bio->bi_disk or bio->bi_bdev(will be changed in
> >>> Christoph's patch) during adding page to bio, so there is null ptr
> >>> refereance risk.
> >>>
> >>>> diff --git a/include/linux/bio.h b/include/linux/bio.h
> >>>> index 1edda614f7ce..cdb134ca7bf5 100644
> >>>> --- a/include/linux/bio.h
> >>>> +++ b/include/linux/bio.h
> >>>> @@ -100,6 +100,8 @@ static inline void *bio_data(struct bio *bio)
> >>>> return NULL;
> >>>> }
> >>>>
> >>>> +extern unsigned int bio_max_size(struct bio *);
> >>>> +
> >>>> /**
> >>>> * bio_full - check if the bio is full
> >>>> * @bio: bio to check
> >>>> @@ -113,7 +115,7 @@ static inline bool bio_full(struct bio *bio, unsigned len)
> >>>> if (bio->bi_vcnt >= bio->bi_max_vecs)
> >>>> return true;
> >>>>
> >>>> - if (bio->bi_iter.bi_size > UINT_MAX - len)
> >>>> + if (bio->bi_iter.bi_size > bio_max_size(bio) - len)
> >>>> return true;
> >>>>
> >>>> return false;
> >>>> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> >>>> index f94ee3089e01..3aeab9e7e97b 100644
> >>>> --- a/include/linux/blkdev.h
> >>>> +++ b/include/linux/blkdev.h
> >>>> @@ -621,6 +621,7 @@ struct request_queue {
> >>>> #define QUEUE_FLAG_RQ_ALLOC_TIME 27 /* record rq->alloc_time_ns */
> >>>> #define QUEUE_FLAG_HCTX_ACTIVE 28 /* at least one blk-mq hctx is active */
> >>>> #define QUEUE_FLAG_NOWAIT 29 /* device supports NOWAIT */
> >>>> +#define QUEUE_FLAG_LIMIT_BIO_SIZE 30 /* limit bio size */
> >>>>
> >>>> #define QUEUE_FLAG_MQ_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \
> >>>> (1 << QUEUE_FLAG_SAME_COMP) | \
> >>>> @@ -667,6 +668,8 @@ bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q);
> >>>> #define blk_queue_fua(q) test_bit(QUEUE_FLAG_FUA, &(q)->queue_flags)
> >>>> #define blk_queue_registered(q) test_bit(QUEUE_FLAG_REGISTERED, &(q)->queue_flags)
> >>>> #define blk_queue_nowait(q) test_bit(QUEUE_FLAG_NOWAIT, &(q)->queue_flags)
> >>>> +#define blk_queue_limit_bio_size(q) \
> >>>> + test_bit(QUEUE_FLAG_LIMIT_BIO_SIZE, &(q)->queue_flags)
> >>>
> >>> I don't think it is a good idea by adding queue flag for this purpose,
> >>> since this case just needs to limit bio size for not delay bio submission
> >>> too much, which is kind of logical thing, nothing to do with request queue.
> >>>
> >>> Just wondering why you don't take the following way:
> >>>
> >>>
> >>> diff --git a/block/bio.c b/block/bio.c
> >>> index 99040a7e6656..35852f7f47d4 100644
> >>> --- a/block/bio.c
> >>> +++ b/block/bio.c
> >>> @@ -1081,7 +1081,7 @@ static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter)
> >>> * It's intended for direct IO, so doesn't do PSI tracking, the caller is
> >>> * responsible for setting BIO_WORKINGSET if necessary.
> >>> */
> >>> -int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
> >>> +int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter, bool sync)
> >>> {
> >>> int ret = 0;
> >>>
> >>> @@ -1092,12 +1092,20 @@ int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
> >>> bio_set_flag(bio, BIO_NO_PAGE_REF);
> >>> return 0;
> >>> } else {
> >>> + /*
> >>> + * Don't add too many pages in case of sync dio for
> >>> + * avoiding delay bio submission too much especially
> >>> + * pinning user pages in memory isn't cheap.
> >>> + */
> >>> + const unsigned int max_size = sync ? (1U << 12) : UINT_MAX;
> >>
> >> 4KB max bio size ? That is a little small :)
> >
> > It should have been (1U << 20), :-(
>
> Sounds better !
>
> >
> >> In any case, I am not a fan of using an arbitrary value not related to the
> >> actual device characteristics. Wouldn't it be better to us the device
> >> max_sectors limit ? And that limit would need to be zone_append_max_sectors for
> >> zone append writes. So some helper like Changheun bio_max_size() may be useful.
> >
> > Firstly, bio->bi_disk may not be initialized when adding page to bio; secondly this
> > limit isn't really related with device, is it? Also if it is one queue limit, it has
> > to be stacked.
>
> 1MB can be used as a fallback default if the gendisk is not yet set. If it is,
IMO, only sync dio on slow machine needs such limit because pinning userspace
pages to memory may take a bit long, so far not see other workloads needs this limit.
Even today I get queries from client about why 4MB user space IO won't be converted to
4MB bio, some workload needs big size IO.
> using a queue limit that does not cause bio splitting after submit makes most
> sense as that avoid useless overhead. I agree, it is not critical, but it would
> be nice to have some number that causes less splitting than the arbitrary 1MB.
That is another story. Each fs bio needs two allocation(one fixed length
bio allocation, and variable length of bvec table), however bio splitting just
needs single fixed length bio allocation. So if the source bio(fs bio) for holding
data becomes smaller, splitting may become less, but more fs bio and bvec table
allocation may be involved, not sure this way always gets better performance.
Also in theory, bio splitting may not need to allocate one whole bio
allocation, what matters is just the actual position/size info of the
to-be-splitted bio.
> E,g, most HDDs will likely have that 1MB BIO split... And max_sectors is a
> stacked queue limit, no ? We could use max_hw_sectors too I think.
bio_add_page() is really fast path, and checking queue limit here may
hurt performance because queue_limits reference is added to the fast path.
Thanks,
Ming
