@@ -566,6 +566,8 @@ Description:
device to that new IO scheduler. Note that writing an IO
scheduler name to this file will attempt to load that IO
scheduler module, if it isn't already present in the system.
+ Support setting scheduler & nr_requests together,
+ e.g. echo mq-deadline:256 > scheduler
What: /sys/block/<disk>/queue/stable_writes
@@ -438,7 +438,8 @@ static int blk_mq_init_sched_shared_tags(struct request_queue *queue)
}
/* caller must have a reference to @e, will grab another one if successful */
-int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)
+int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e,
+ unsigned long nr_rqs)
{
unsigned int flags = q->tag_set->flags;
struct blk_mq_hw_ctx *hctx;
@@ -451,8 +452,11 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)
* since we don't split into sync/async like the old code did.
* Additionally, this is a per-hw queue depth.
*/
- q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth,
- BLKDEV_DEFAULT_RQ);
+ if (nr_rqs)
+ q->nr_requests = nr_rqs;
+ else
+ q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth,
+ BLKDEV_DEFAULT_RQ);
if (blk_mq_is_shared_tags(flags)) {
ret = blk_mq_init_sched_shared_tags(q);
@@ -18,7 +18,8 @@ void __blk_mq_sched_restart(struct blk_mq_hw_ctx *hctx);
void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx);
-int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e);
+int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e,
+ unsigned long nr_rqs);
void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e);
void blk_mq_sched_free_rqs(struct request_queue *q);
@@ -4707,7 +4707,7 @@ static void blk_mq_elv_switch_back(struct list_head *head,
kfree(qe);
mutex_lock(&q->sysfs_lock);
- elevator_switch(q, t);
+ elevator_switch(q, t, 0);
/* drop the reference acquired in blk_mq_elv_switch_none */
elevator_put(t);
mutex_unlock(&q->sysfs_lock);
@@ -287,7 +287,8 @@ bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
bool blk_insert_flush(struct request *rq);
-int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
+int elevator_switch(struct request_queue *q, struct elevator_type *new_e,
+ unsigned long nr_rqs);
void elevator_disable(struct request_queue *q);
void elevator_exit(struct request_queue *q);
int elv_register_queue(struct request_queue *q, bool uevent);
@@ -639,7 +639,7 @@ void elevator_init_mq(struct request_queue *q)
blk_mq_freeze_queue(q);
blk_mq_cancel_work_sync(q);
- err = blk_mq_init_sched(q, e);
+ err = blk_mq_init_sched(q, e, 0);
blk_mq_unfreeze_queue(q);
@@ -657,7 +657,8 @@ void elevator_init_mq(struct request_queue *q)
* If switching fails, we are most likely running out of memory and not able
* to restore the old io scheduler, so leaving the io scheduler being none.
*/
-int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
+int elevator_switch(struct request_queue *q, struct elevator_type *new_e,
+ unsigned long nr_rqs)
{
int ret;
@@ -671,7 +672,7 @@ int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
elevator_exit(q);
}
- ret = blk_mq_init_sched(q, new_e);
+ ret = blk_mq_init_sched(q, new_e, nr_rqs);
if (ret)
goto out_unfreeze;
@@ -715,7 +716,8 @@ void elevator_disable(struct request_queue *q)
/*
* Switch this queue to the given IO scheduler.
*/
-static int elevator_change(struct request_queue *q, const char *elevator_name)
+static int elevator_change(struct request_queue *q, const char *elevator_name,
+ unsigned long nr_rqs)
{
struct elevator_type *e;
int ret;
@@ -740,7 +742,7 @@ static int elevator_change(struct request_queue *q, const char *elevator_name)
if (!e)
return -EINVAL;
}
- ret = elevator_switch(q, e);
+ ret = elevator_switch(q, e, nr_rqs);
elevator_put(e);
return ret;
}
@@ -749,13 +751,33 @@ ssize_t elv_iosched_store(struct request_queue *q, const char *buf,
size_t count)
{
char elevator_name[ELV_NAME_MAX];
+ unsigned long nr_rqs;
+ const char *num;
+ char *tmp;
int ret;
if (!elv_support_iosched(q))
return count;
+ /*
+ * Support setting scheduler & nr_requests together
+ * e.g. echo mq-deadline:256 > scheduler
+ */
+ tmp = strchr(buf, ':');
+ if (tmp) {
+ *tmp = '\0';
+ num = tmp + 1;
+
+ ret = kstrtoul(num, 10, &nr_rqs);
+ if (ret || nr_rqs > MAX_SCHED_RQ)
+ return -EINVAL;
+
+ if (nr_rqs < BLKDEV_MIN_RQ)
+ nr_rqs = BLKDEV_MIN_RQ;
+ }
+
strscpy(elevator_name, buf, sizeof(elevator_name));
- ret = elevator_change(q, strstrip(elevator_name));
+ ret = elevator_change(q, strstrip(elevator_name), nr_rqs);
if (!ret)
return count;
return ret;
When switching the scheduler, the nr_requests defaults to an initial value. If we want to change the nr_requests after switching the scheduler, we need to perform the setting twice. It means the update process for nr_requests needs to be run twice. And updating nr_requests may cost dozens of milliseconds in our test environment. [2.091825] init: Command 'write /sys/block/sda/queue/nr_requests 96' took 24ms [2.123248] init: Command 'write /sys/block/sdb/queue/nr_requests 96' took 31ms [2.146620] init: Command 'write /sys/block/sdc/queue/nr_requests 96' took 22ms [2.171127] init: Command 'write /sys/block/sdd/queue/nr_requests 96' took 24ms [2.207286] init: Command 'write /sys/block/sde/queue/nr_requests 96' took 36ms [2.222515] init: Command 'write /sys/block/sdf/queue/nr_requests 96' took 15ms [2.246704] init: Command 'write /sys/block/sdg/queue/nr_requests 96' took 24ms So provide a way to configure scheduler and nr_requests together. Signed-off-by: Yang Yang <yang.yang@vivo.com> --- Documentation/ABI/stable/sysfs-block | 2 ++ block/blk-mq-sched.c | 10 +++++--- block/blk-mq-sched.h | 3 ++- block/blk-mq.c | 2 +- block/blk.h | 3 ++- block/elevator.c | 34 +++++++++++++++++++++++----- 6 files changed, 42 insertions(+), 12 deletions(-)