| Message ID | 20230407235822.1672286-2-bvanassche@acm.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Submit zoned writes in order | expand |
On 4/8/23 08:58, Bart Van Assche wrote: > Send zoned writes inserted by the device mapper to the I/O scheduler. > This prevents that zoned writes get reordered if a device mapper driver > has been stacked on top of a driver for a zoned block device. > > Cc: Christoph Hellwig <hch@lst.de> > Cc: Damien Le Moal <damien.lemoal@opensource.wdc.com> > Cc: Ming Lei <ming.lei@redhat.com> > Cc: Mike Snitzer <snitzer@kernel.org> > Signed-off-by: Bart Van Assche <bvanassche@acm.org> > --- > block/blk-mq.c | 16 +++++++++++++--- > block/blk.h | 19 +++++++++++++++++++ > 2 files changed, 32 insertions(+), 3 deletions(-) > > diff --git a/block/blk-mq.c b/block/blk-mq.c > index db93b1a71157..fefc9a728e0e 100644 > --- a/block/blk-mq.c > +++ b/block/blk-mq.c > @@ -3008,9 +3008,19 @@ blk_status_t blk_insert_cloned_request(struct request *rq) > blk_account_io_start(rq); > > /* > - * Since we have a scheduler attached on the top device, > - * bypass a potential scheduler on the bottom device for > - * insert. > + * Send zoned writes to the I/O scheduler if an I/O scheduler has been > + * attached. > + */ > + if (q->elevator && blk_rq_is_seq_zoned_write(rq)) { > + blk_mq_sched_insert_request(rq, /*at_head=*/false, > + /*run_queue=*/true, > + /*async=*/false); > + return BLK_STS_OK; > + } Looks OK to me, but as I understand it, this function is used only for request based device mapper, none of which currently support zoned block devices if I a not mistaken. So is this change really necessary ? > + > + /* > + * If no I/O scheduler has been attached or if the request is not a > + * zoned write bypass the I/O scheduler attached to the bottom device. > */ > blk_mq_run_dispatch_ops(q, > ret = blk_mq_request_issue_directly(rq, true)); > diff --git a/block/blk.h b/block/blk.h > index d65d96994a94..4b6f8d7a6b84 100644 > --- a/block/blk.h > +++ b/block/blk.h > @@ -118,6 +118,25 @@ static inline bool bvec_gap_to_prev(const struct queue_limits *lim, > return __bvec_gap_to_prev(lim, bprv, offset); > } > > +/** > + * blk_rq_is_seq_zoned_write() - Whether @rq is a write request for a sequential zone. > + * @rq: Request to examine. > + * > + * In this context sequential zone means either a sequential write required or > + * to a sequential write preferred zone. > + */ > +static inline bool blk_rq_is_seq_zoned_write(struct request *rq) > +{ > + switch (req_op(rq)) { > + case REQ_OP_WRITE: > + case REQ_OP_WRITE_ZEROES: > + return disk_zone_is_seq(rq->q->disk, blk_rq_pos(rq)); > + case REQ_OP_ZONE_APPEND: > + default: > + return false; > + } > +} > + > static inline bool rq_mergeable(struct request *rq) > { > if (blk_rq_is_passthrough(rq))
On 4/10/23 00:42, Damien Le Moal wrote: > On 4/8/23 08:58, Bart Van Assche wrote: >> Send zoned writes inserted by the device mapper to the I/O scheduler. >> This prevents that zoned writes get reordered if a device mapper driver >> has been stacked on top of a driver for a zoned block device. >> >> Cc: Christoph Hellwig <hch@lst.de> >> Cc: Damien Le Moal <damien.lemoal@opensource.wdc.com> >> Cc: Ming Lei <ming.lei@redhat.com> >> Cc: Mike Snitzer <snitzer@kernel.org> >> Signed-off-by: Bart Van Assche <bvanassche@acm.org> >> --- >> block/blk-mq.c | 16 +++++++++++++--- >> block/blk.h | 19 +++++++++++++++++++ >> 2 files changed, 32 insertions(+), 3 deletions(-) >> >> diff --git a/block/blk-mq.c b/block/blk-mq.c >> index db93b1a71157..fefc9a728e0e 100644 >> --- a/block/blk-mq.c >> +++ b/block/blk-mq.c >> @@ -3008,9 +3008,19 @@ blk_status_t blk_insert_cloned_request(struct request *rq) >> blk_account_io_start(rq); >> >> /* >> - * Since we have a scheduler attached on the top device, >> - * bypass a potential scheduler on the bottom device for >> - * insert. >> + * Send zoned writes to the I/O scheduler if an I/O scheduler has been >> + * attached. >> + */ >> + if (q->elevator && blk_rq_is_seq_zoned_write(rq)) { >> + blk_mq_sched_insert_request(rq, /*at_head=*/false, >> + /*run_queue=*/true, >> + /*async=*/false); >> + return BLK_STS_OK; >> + } > > Looks OK to me, but as I understand it, this function is used only for request > based device mapper, none of which currently support zoned block devices if I a > not mistaken. So is this change really necessary ? Hi Damien, Probably not. I will double check whether this patch can be dropped. Thanks, Bart.
diff --git a/block/blk-mq.c b/block/blk-mq.c index db93b1a71157..fefc9a728e0e 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -3008,9 +3008,19 @@ blk_status_t blk_insert_cloned_request(struct request *rq) blk_account_io_start(rq); /* - * Since we have a scheduler attached on the top device, - * bypass a potential scheduler on the bottom device for - * insert. + * Send zoned writes to the I/O scheduler if an I/O scheduler has been + * attached. + */ + if (q->elevator && blk_rq_is_seq_zoned_write(rq)) { + blk_mq_sched_insert_request(rq, /*at_head=*/false, + /*run_queue=*/true, + /*async=*/false); + return BLK_STS_OK; + } + + /* + * If no I/O scheduler has been attached or if the request is not a + * zoned write bypass the I/O scheduler attached to the bottom device. */ blk_mq_run_dispatch_ops(q, ret = blk_mq_request_issue_directly(rq, true)); diff --git a/block/blk.h b/block/blk.h index d65d96994a94..4b6f8d7a6b84 100644 --- a/block/blk.h +++ b/block/blk.h @@ -118,6 +118,25 @@ static inline bool bvec_gap_to_prev(const struct queue_limits *lim, return __bvec_gap_to_prev(lim, bprv, offset); } +/** + * blk_rq_is_seq_zoned_write() - Whether @rq is a write request for a sequential zone. + * @rq: Request to examine. + * + * In this context sequential zone means either a sequential write required or + * to a sequential write preferred zone. + */ +static inline bool blk_rq_is_seq_zoned_write(struct request *rq) +{ + switch (req_op(rq)) { + case REQ_OP_WRITE: + case REQ_OP_WRITE_ZEROES: + return disk_zone_is_seq(rq->q->disk, blk_rq_pos(rq)); + case REQ_OP_ZONE_APPEND: + default: + return false; + } +} + static inline bool rq_mergeable(struct request *rq) { if (blk_rq_is_passthrough(rq))
Send zoned writes inserted by the device mapper to the I/O scheduler. This prevents that zoned writes get reordered if a device mapper driver has been stacked on top of a driver for a zoned block device. Cc: Christoph Hellwig <hch@lst.de> Cc: Damien Le Moal <damien.lemoal@opensource.wdc.com> Cc: Ming Lei <ming.lei@redhat.com> Cc: Mike Snitzer <snitzer@kernel.org> Signed-off-by: Bart Van Assche <bvanassche@acm.org> --- block/blk-mq.c | 16 +++++++++++++--- block/blk.h | 19 +++++++++++++++++++ 2 files changed, 32 insertions(+), 3 deletions(-)