Hi,
here is the V12, it differs from V11 in that it applies the
recommendation by Damien in [2].
Here is the whole description of this patch series again. This
extension addresses the following issue. Single-LUN multi-actuator
SCSI drives, as well as all multi-actuator SATA drives appear as a
single device to the I/O subsystem [1]. Yet they address commands to
different actuators internally, as a function of Logical Block
Addressing (LBAs). A given sector is reachable by only one of the
actuators. For example, Seagate’s Serial Advanced Technology
Attachment (SATA) version contains two actuators and maps the lower
half of the SATA LBA space to the lower actuator and the upper half to
the upper actuator.
Evidently, to fully utilize actuators, no actuator must be left idle
or underutilized while there is pending I/O for it. To reach this
goal, the block layer must somehow control the load of each actuator
individually. This series enriches BFQ with such a per-actuator
control, as a first step. Then it also adds a simple mechanism for
guaranteeing that actuators with pending I/O are never left idle.
See [1] for a more detailed overview of the problem and of the
solutions implemented in this patch series. There you will also find
some preliminary performance results.
Thanks,
Paolo
[1] https://www.linaro.org/blog/budget-fair-queueing-bfq-linux-io-scheduler-optimizations-for-multi-actuator-sata-hard-drives/
[2] https://lore.kernel.org/lkml/[email protected]/T/#m28f40c3b060087a230998a3b55a021020c396cab
Davide Zini (3):
block, bfq: split also async bfq_queues on a per-actuator basis
block, bfq: inject I/O to underutilized actuators
block, bfq: balance I/O injection among underutilized actuators
Federico Gavioli (1):
block, bfq: retrieve independent access ranges from request queue
Paolo Valente (4):
block, bfq: split sync bfq_queues on a per-actuator basis
block, bfq: forbid stable merging of queues associated with different
actuators
block, bfq: move io_cq-persistent bfqq data into a dedicated struct
block, bfq: turn bfqq_data into an array in bfq_io_cq
block/bfq-cgroup.c | 93 +++----
block/bfq-iosched.c | 591 ++++++++++++++++++++++++++++++--------------
block/bfq-iosched.h | 142 ++++++++---
block/bfq-wf2q.c | 2 +-
4 files changed, 569 insertions(+), 259 deletions(-)
--
2.20.1
Single-LUN multi-actuator SCSI drives, as well as all multi-actuator
SATA drives appear as a single device to the I/O subsystem [1]. Yet
they address commands to different actuators internally, as a function
of Logical Block Addressing (LBAs). A given sector is reachable by
only one of the actuators. For example, Seagate’s Serial Advanced
Technology Attachment (SATA) version contains two actuators and maps
the lower half of the SATA LBA space to the lower actuator and the
upper half to the upper actuator.
Evidently, to fully utilize actuators, no actuator must be left idle
or underutilized while there is pending I/O for it. The block layer
must somehow control the load of each actuator individually. This
commit lays the ground for allowing BFQ to provide such a per-actuator
control.
BFQ associates an I/O-request sync bfq_queue with each process doing
synchronous I/O, or with a group of processes, in case of queue
merging. Then BFQ serves one bfq_queue at a time. While in service, a
bfq_queue is emptied in request-position order. Yet the same process,
or group of processes, may generate I/O for different actuators. In
this case, different streams of I/O (each for a different actuator)
get all inserted into the same sync bfq_queue. So there is basically
no individual control on when each stream is served, i.e., on when the
I/O requests of the stream are picked from the bfq_queue and
dispatched to the drive.
This commit enables BFQ to control the service of each actuator
individually for synchronous I/O, by simply splitting each sync
bfq_queue into N queues, one for each actuator. In other words, a sync
bfq_queue is now associated to a pair (process, actuator). As a
consequence of this split, the per-queue proportional-share policy
implemented by BFQ will guarantee that the sync I/O generated for each
actuator, by each process, receives its fair share of service.
This is just a preparatory patch. If the I/O of the same process
happens to be sent to different queues, then each of these queues may
undergo queue merging. To handle this event, the bfq_io_cq data
structure must be properly extended. In addition, stable merging must
be disabled to avoid loss of control on individual actuators. Finally,
also async queues must be split. These issues are described in detail
and addressed in next commits. As for this commit, although multiple
per-process bfq_queues are provided, the I/O of each process or group
of processes is still sent to only one queue, regardless of the
actuator the I/O is for. The forwarding to distinct bfq_queues will be
enabled after addressing the above issues.
[1] https://www.linaro.org/blog/budget-fair-queueing-bfq-linux-io-scheduler-optimizations-for-multi-actuator-sata-hard-drives/
Signed-off-by: Gabriele Felici <[email protected]>
Signed-off-by: Carmine Zaccagnino <[email protected]>
Signed-off-by: Paolo Valente <[email protected]>
---
block/bfq-cgroup.c | 91 ++++++++++++------------
block/bfq-iosched.c | 164 +++++++++++++++++++++++++++++---------------
block/bfq-iosched.h | 51 +++++++++++---
3 files changed, 199 insertions(+), 107 deletions(-)
diff --git a/block/bfq-cgroup.c b/block/bfq-cgroup.c
index 1b2829e99dad..bd7bf0f8121d 100644
--- a/block/bfq-cgroup.c
+++ b/block/bfq-cgroup.c
@@ -714,6 +714,46 @@ void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
bfq_put_queue(bfqq);
}
+static void bfq_sync_bfqq_move(struct bfq_data *bfqd,
+ struct bfq_queue *sync_bfqq,
+ struct bfq_io_cq *bic,
+ struct bfq_group *bfqg,
+ unsigned int act_idx)
+{
+ struct bfq_queue *bfqq;
+
+ if (!sync_bfqq->new_bfqq && !bfq_bfqq_coop(sync_bfqq)) {
+ /* We are the only user of this bfqq, just move it */
+ if (sync_bfqq->entity.sched_data != &bfqg->sched_data)
+ bfq_bfqq_move(bfqd, sync_bfqq, bfqg);
+ return;
+ }
+
+ /*
+ * The queue was merged to a different queue. Check
+ * that the merge chain still belongs to the same
+ * cgroup.
+ */
+ for (bfqq = sync_bfqq; bfqq; bfqq = bfqq->new_bfqq)
+ if (bfqq->entity.sched_data != &bfqg->sched_data)
+ break;
+ if (bfqq) {
+ /*
+ * Some queue changed cgroup so the merge is not valid
+ * anymore. We cannot easily just cancel the merge (by
+ * clearing new_bfqq) as there may be other processes
+ * using this queue and holding refs to all queues
+ * below sync_bfqq->new_bfqq. Similarly if the merge
+ * already happened, we need to detach from bfqq now
+ * so that we cannot merge bio to a request from the
+ * old cgroup.
+ */
+ bfq_put_cooperator(sync_bfqq);
+ bfq_release_process_ref(bfqd, sync_bfqq);
+ bic_set_bfqq(bic, NULL, true, act_idx);
+ }
+}
+
/**
* __bfq_bic_change_cgroup - move @bic to @bfqg.
* @bfqd: the queue descriptor.
@@ -728,53 +768,20 @@ static void __bfq_bic_change_cgroup(struct bfq_data *bfqd,
struct bfq_io_cq *bic,
struct bfq_group *bfqg)
{
- struct bfq_queue *async_bfqq = bic_to_bfqq(bic, false);
- struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, true);
- struct bfq_entity *entity;
+ unsigned int act_idx;
- if (async_bfqq) {
- entity = &async_bfqq->entity;
+ for (act_idx = 0; act_idx < bfqd->num_actuators; act_idx++) {
+ struct bfq_queue *async_bfqq = bic_to_bfqq(bic, false, act_idx);
+ struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, true, act_idx);
- if (entity->sched_data != &bfqg->sched_data) {
- bic_set_bfqq(bic, NULL, false);
+ if (async_bfqq &&
+ async_bfqq->entity.sched_data != &bfqg->sched_data) {
+ bic_set_bfqq(bic, NULL, false, act_idx);
bfq_release_process_ref(bfqd, async_bfqq);
}
- }
- if (sync_bfqq) {
- if (!sync_bfqq->new_bfqq && !bfq_bfqq_coop(sync_bfqq)) {
- /* We are the only user of this bfqq, just move it */
- if (sync_bfqq->entity.sched_data != &bfqg->sched_data)
- bfq_bfqq_move(bfqd, sync_bfqq, bfqg);
- } else {
- struct bfq_queue *bfqq;
-
- /*
- * The queue was merged to a different queue. Check
- * that the merge chain still belongs to the same
- * cgroup.
- */
- for (bfqq = sync_bfqq; bfqq; bfqq = bfqq->new_bfqq)
- if (bfqq->entity.sched_data !=
- &bfqg->sched_data)
- break;
- if (bfqq) {
- /*
- * Some queue changed cgroup so the merge is
- * not valid anymore. We cannot easily just
- * cancel the merge (by clearing new_bfqq) as
- * there may be other processes using this
- * queue and holding refs to all queues below
- * sync_bfqq->new_bfqq. Similarly if the merge
- * already happened, we need to detach from
- * bfqq now so that we cannot merge bio to a
- * request from the old cgroup.
- */
- bfq_put_cooperator(sync_bfqq);
- bfq_release_process_ref(bfqd, sync_bfqq);
- bic_set_bfqq(bic, NULL, true);
- }
- }
+ if (sync_bfqq)
+ bfq_sync_bfqq_move(bfqd, sync_bfqq, bic, bfqg, act_idx);
}
}
diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 16f43bbc575a..e889d50147e8 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -377,16 +377,23 @@ static const unsigned long bfq_late_stable_merging = 600;
#define RQ_BIC(rq) ((struct bfq_io_cq *)((rq)->elv.priv[0]))
#define RQ_BFQQ(rq) ((rq)->elv.priv[1])
-struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync)
+struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync,
+ unsigned int actuator_idx)
{
- return bic->bfqq[is_sync];
+ if (is_sync)
+ return bic->bfqq[1][actuator_idx];
+
+ return bic->bfqq[0][actuator_idx];
}
static void bfq_put_stable_ref(struct bfq_queue *bfqq);
-void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync)
+void bic_set_bfqq(struct bfq_io_cq *bic,
+ struct bfq_queue *bfqq,
+ bool is_sync,
+ unsigned int actuator_idx)
{
- struct bfq_queue *old_bfqq = bic->bfqq[is_sync];
+ struct bfq_queue *old_bfqq = bic->bfqq[is_sync][actuator_idx];
/* Clear bic pointer if bfqq is detached from this bic */
if (old_bfqq && old_bfqq->bic == bic)
@@ -405,7 +412,10 @@ void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync)
* we cancel the stable merge if
* bic->stable_merge_bfqq == bfqq.
*/
- bic->bfqq[is_sync] = bfqq;
+ if (is_sync)
+ bic->bfqq[1][actuator_idx] = bfqq;
+ else
+ bic->bfqq[0][actuator_idx] = bfqq;
if (bfqq && bic->stable_merge_bfqq == bfqq) {
/*
@@ -678,9 +688,9 @@ static void bfq_limit_depth(blk_opf_t opf, struct blk_mq_alloc_data *data)
{
struct bfq_data *bfqd = data->q->elevator->elevator_data;
struct bfq_io_cq *bic = bfq_bic_lookup(data->q);
- struct bfq_queue *bfqq = bic ? bic_to_bfqq(bic, op_is_sync(opf)) : NULL;
int depth;
unsigned limit = data->q->nr_requests;
+ unsigned int act_idx;
/* Sync reads have full depth available */
if (op_is_sync(opf) && !op_is_write(opf)) {
@@ -690,14 +700,25 @@ static void bfq_limit_depth(blk_opf_t opf, struct blk_mq_alloc_data *data)
limit = (limit * depth) >> bfqd->full_depth_shift;
}
- /*
- * Does queue (or any parent entity) exceed number of requests that
- * should be available to it? Heavily limit depth so that it cannot
- * consume more available requests and thus starve other entities.
- */
- if (bfqq && bfqq_request_over_limit(bfqq, limit))
- depth = 1;
+ for (act_idx = 0; act_idx < bfqd->num_actuators; act_idx++) {
+ struct bfq_queue *bfqq;
+
+ if (bic)
+ bfqq = bic_to_bfqq(bic, op_is_sync(opf), act_idx);
+ else
+ break;
+ /*
+ * Does queue (or any parent entity) exceed number of
+ * requests that should be available to it? Heavily
+ * limit depth so that it cannot consume more
+ * available requests and thus starve other entities.
+ */
+ if (bfqq && bfqq_request_over_limit(bfqq, limit)) {
+ depth = 1;
+ break;
+ }
+ }
bfq_log(bfqd, "[%s] wr_busy %d sync %d depth %u",
__func__, bfqd->wr_busy_queues, op_is_sync(opf), depth);
if (depth)
@@ -1766,6 +1787,18 @@ static bool bfq_bfqq_higher_class_or_weight(struct bfq_queue *bfqq,
return bfqq_weight > in_serv_weight;
}
+/*
+ * Get the index of the actuator that will serve bio.
+ */
+static unsigned int bfq_actuator_index(struct bfq_data *bfqd, struct bio *bio)
+{
+ /*
+ * Multi-actuator support not complete yet, so always return 0
+ * for the moment (to keep incomplete mechanisms off).
+ */
+ return 0;
+}
+
static bool bfq_better_to_idle(struct bfq_queue *bfqq);
static void bfq_bfqq_handle_idle_busy_switch(struct bfq_data *bfqd,
@@ -2098,7 +2131,7 @@ static void bfq_check_waker(struct bfq_data *bfqd, struct bfq_queue *bfqq,
* We reset waker detection logic also if too much time has passed
* since the first detection. If wakeups are rare, pointless idling
* doesn't hurt throughput that much. The condition below makes sure
- * we do not uselessly idle blocking waker in more than 1/64 cases.
+ * we do not uselessly idle blocking waker in more than 1/64 cases.
*/
if (bfqd->last_completed_rq_bfqq !=
bfqq->tentative_waker_bfqq ||
@@ -2418,7 +2451,8 @@ static bool bfq_bio_merge(struct request_queue *q, struct bio *bio,
*/
bfq_bic_update_cgroup(bic, bio);
- bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf));
+ bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf),
+ bfq_actuator_index(bfqd, bio));
} else {
bfqd->bio_bfqq = NULL;
}
@@ -3114,7 +3148,7 @@ bfq_merge_bfqqs(struct bfq_data *bfqd, struct bfq_io_cq *bic,
/*
* Merge queues (that is, let bic redirect its requests to new_bfqq)
*/
- bic_set_bfqq(bic, new_bfqq, true);
+ bic_set_bfqq(bic, new_bfqq, true, bfqq->actuator_idx);
bfq_mark_bfqq_coop(new_bfqq);
/*
* new_bfqq now belongs to at least two bics (it is a shared queue):
@@ -4748,11 +4782,8 @@ static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd)
*/
if (bfq_bfqq_wait_request(bfqq) ||
(bfqq->dispatched != 0 && bfq_better_to_idle(bfqq))) {
- struct bfq_queue *async_bfqq =
- bfqq->bic && bfqq->bic->bfqq[0] &&
- bfq_bfqq_busy(bfqq->bic->bfqq[0]) &&
- bfqq->bic->bfqq[0]->next_rq ?
- bfqq->bic->bfqq[0] : NULL;
+ unsigned int act_idx = bfqq->actuator_idx;
+ struct bfq_queue *async_bfqq = NULL;
struct bfq_queue *blocked_bfqq =
!hlist_empty(&bfqq->woken_list) ?
container_of(bfqq->woken_list.first,
@@ -4760,6 +4791,10 @@ static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd)
woken_list_node)
: NULL;
+ if (bfqq->bic && bfqq->bic->bfqq[0][act_idx] &&
+ bfq_bfqq_busy(bfqq->bic->bfqq[0][act_idx]) &&
+ bfqq->bic->bfqq[0][act_idx]->next_rq)
+ async_bfqq = bfqq->bic->bfqq[0][act_idx];
/*
* The next four mutually-exclusive ifs decide
* whether to try injection, and choose the queue to
@@ -4844,7 +4879,7 @@ static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd)
icq_to_bic(async_bfqq->next_rq->elv.icq) == bfqq->bic &&
bfq_serv_to_charge(async_bfqq->next_rq, async_bfqq) <=
bfq_bfqq_budget_left(async_bfqq))
- bfqq = bfqq->bic->bfqq[0];
+ bfqq = bfqq->bic->bfqq[0][act_idx];
else if (bfqq->waker_bfqq &&
bfq_bfqq_busy(bfqq->waker_bfqq) &&
bfqq->waker_bfqq->next_rq &&
@@ -5305,48 +5340,54 @@ static void bfq_exit_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq)
bfq_release_process_ref(bfqd, bfqq);
}
-static void bfq_exit_icq_bfqq(struct bfq_io_cq *bic, bool is_sync)
+static void bfq_exit_icq_bfqq(struct bfq_io_cq *bic, bool is_sync,
+ unsigned int actuator_idx)
{
- struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync);
+ struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync, actuator_idx);
struct bfq_data *bfqd;
if (bfqq)
bfqd = bfqq->bfqd; /* NULL if scheduler already exited */
if (bfqq && bfqd) {
- unsigned long flags;
-
- spin_lock_irqsave(&bfqd->lock, flags);
bfq_exit_bfqq(bfqd, bfqq);
- bic_set_bfqq(bic, NULL, is_sync);
- spin_unlock_irqrestore(&bfqd->lock, flags);
+ bic_set_bfqq(bic, NULL, is_sync, actuator_idx);
}
}
static void bfq_exit_icq(struct io_cq *icq)
{
struct bfq_io_cq *bic = icq_to_bic(icq);
+ struct bfq_data *bfqd = bic_to_bfqd(bic);
+ unsigned long flags;
+ unsigned int act_idx;
+ /*
+ * If bfqd and thus bfqd->num_actuators is not available any
+ * longer, then cycle over all possible per-actuator bfqqs in
+ * next loop. We rely on bic being zeroed on creation, and
+ * therefore on its unused per-actuator fields being NULL.
+ */
+ unsigned int num_actuators = BFQ_MAX_ACTUATORS;
- if (bic->stable_merge_bfqq) {
- struct bfq_data *bfqd = bic->stable_merge_bfqq->bfqd;
+ /*
+ * bfqd is NULL if scheduler already exited, and in that case
+ * this is the last time these queues are accessed.
+ */
+ if (bfqd) {
+ spin_lock_irqsave(&bfqd->lock, flags);
+ num_actuators = bfqd->num_actuators;
+ }
- /*
- * bfqd is NULL if scheduler already exited, and in
- * that case this is the last time bfqq is accessed.
- */
- if (bfqd) {
- unsigned long flags;
+ if (bic->stable_merge_bfqq)
+ bfq_put_stable_ref(bic->stable_merge_bfqq);
- spin_lock_irqsave(&bfqd->lock, flags);
- bfq_put_stable_ref(bic->stable_merge_bfqq);
- spin_unlock_irqrestore(&bfqd->lock, flags);
- } else {
- bfq_put_stable_ref(bic->stable_merge_bfqq);
- }
+ for (act_idx = 0; act_idx < num_actuators; act_idx++) {
+ bfq_exit_icq_bfqq(bic, true, act_idx);
+ bfq_exit_icq_bfqq(bic, false, act_idx);
}
- bfq_exit_icq_bfqq(bic, true);
- bfq_exit_icq_bfqq(bic, false);
+ if (bfqd)
+ spin_unlock_irqrestore(&bfqd->lock, flags);
}
/*
@@ -5423,23 +5464,25 @@ static void bfq_check_ioprio_change(struct bfq_io_cq *bic, struct bio *bio)
bic->ioprio = ioprio;
- bfqq = bic_to_bfqq(bic, false);
+ bfqq = bic_to_bfqq(bic, false, bfq_actuator_index(bfqd, bio));
if (bfqq) {
bfq_release_process_ref(bfqd, bfqq);
bfqq = bfq_get_queue(bfqd, bio, false, bic, true);
- bic_set_bfqq(bic, bfqq, false);
+ bic_set_bfqq(bic, bfqq, false, bfq_actuator_index(bfqd, bio));
}
- bfqq = bic_to_bfqq(bic, true);
+ bfqq = bic_to_bfqq(bic, true, bfq_actuator_index(bfqd, bio));
if (bfqq)
bfq_set_next_ioprio_data(bfqq, bic);
}
static void bfq_init_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
- struct bfq_io_cq *bic, pid_t pid, int is_sync)
+ struct bfq_io_cq *bic, pid_t pid, int is_sync,
+ unsigned int act_idx)
{
u64 now_ns = ktime_get_ns();
+ bfqq->actuator_idx = act_idx;
RB_CLEAR_NODE(&bfqq->entity.rb_node);
INIT_LIST_HEAD(&bfqq->fifo);
INIT_HLIST_NODE(&bfqq->burst_list_node);
@@ -5692,7 +5735,7 @@ static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd,
if (bfqq) {
bfq_init_bfqq(bfqd, bfqq, bic, current->pid,
- is_sync);
+ is_sync, bfq_actuator_index(bfqd, bio));
bfq_init_entity(&bfqq->entity, bfqg);
bfq_log_bfqq(bfqd, bfqq, "allocated");
} else {
@@ -6007,7 +6050,8 @@ static bool __bfq_insert_request(struct bfq_data *bfqd, struct request *rq)
* then complete the merge and redirect it to
* new_bfqq.
*/
- if (bic_to_bfqq(RQ_BIC(rq), 1) == bfqq)
+ if (bic_to_bfqq(RQ_BIC(rq), true,
+ bfq_actuator_index(bfqd, rq->bio)) == bfqq)
bfq_merge_bfqqs(bfqd, RQ_BIC(rq),
bfqq, new_bfqq);
@@ -6562,7 +6606,7 @@ bfq_split_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq)
return bfqq;
}
- bic_set_bfqq(bic, NULL, true);
+ bic_set_bfqq(bic, NULL, true, bfqq->actuator_idx);
bfq_put_cooperator(bfqq);
@@ -6576,7 +6620,8 @@ static struct bfq_queue *bfq_get_bfqq_handle_split(struct bfq_data *bfqd,
bool split, bool is_sync,
bool *new_queue)
{
- struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync);
+ unsigned int act_idx = bfq_actuator_index(bfqd, bio);
+ struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync, act_idx);
if (likely(bfqq && bfqq != &bfqd->oom_bfqq))
return bfqq;
@@ -6588,7 +6633,7 @@ static struct bfq_queue *bfq_get_bfqq_handle_split(struct bfq_data *bfqd,
bfq_put_queue(bfqq);
bfqq = bfq_get_queue(bfqd, bio, is_sync, bic, split);
- bic_set_bfqq(bic, bfqq, is_sync);
+ bic_set_bfqq(bic, bfqq, is_sync, act_idx);
if (split && is_sync) {
if ((bic->was_in_burst_list && bfqd->large_burst) ||
bic->saved_in_large_burst)
@@ -7036,8 +7081,10 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
* Our fallback bfqq if bfq_find_alloc_queue() runs into OOM issues.
* Grab a permanent reference to it, so that the normal code flow
* will not attempt to free it.
+ * Set zero as actuator index: we will pretend that
+ * all I/O requests are for the same actuator.
*/
- bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, NULL, 1, 0);
+ bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, NULL, 1, 0, 0);
bfqd->oom_bfqq.ref++;
bfqd->oom_bfqq.new_ioprio = BFQ_DEFAULT_QUEUE_IOPRIO;
bfqd->oom_bfqq.new_ioprio_class = IOPRIO_CLASS_BE;
@@ -7056,6 +7103,13 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
bfqd->queue = q;
+ /*
+ * Multi-actuator support not complete yet, unconditionally
+ * set to only one actuator for the moment (to keep incomplete
+ * mechanisms off).
+ */
+ bfqd->num_actuators = 1;
+
INIT_LIST_HEAD(&bfqd->dispatch);
hrtimer_init(&bfqd->idle_slice_timer, CLOCK_MONOTONIC,
diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h
index 41aa151ccc22..3c6b21b2cd95 100644
--- a/block/bfq-iosched.h
+++ b/block/bfq-iosched.h
@@ -33,6 +33,14 @@
*/
#define BFQ_SOFTRT_WEIGHT_FACTOR 100
+/*
+ * Maximum number of actuators supported. This constant is used simply
+ * to define the size of the static array that will contain
+ * per-actuator data. The current value is hopefully a good upper
+ * bound to the possible number of actuators of any actual drive.
+ */
+#define BFQ_MAX_ACTUATORS 8
+
struct bfq_entity;
/**
@@ -227,12 +235,14 @@ struct bfq_ttime {
* struct bfq_queue - leaf schedulable entity.
*
* A bfq_queue is a leaf request queue; it can be associated with an
- * io_context or more, if it is async or shared between cooperating
- * processes. @cgroup holds a reference to the cgroup, to be sure that it
- * does not disappear while a bfqq still references it (mostly to avoid
- * races between request issuing and task migration followed by cgroup
- * destruction).
- * All the fields are protected by the queue lock of the containing bfqd.
+ * io_context or more, if it is async or shared between cooperating
+ * processes. Besides, it contains I/O requests for only one actuator
+ * (an io_context is associated with a different bfq_queue for each
+ * actuator it generates I/O for). @cgroup holds a reference to the
+ * cgroup, to be sure that it does not disappear while a bfqq still
+ * references it (mostly to avoid races between request issuing and
+ * task migration followed by cgroup destruction). All the fields are
+ * protected by the queue lock of the containing bfqd.
*/
struct bfq_queue {
/* reference counter */
@@ -397,6 +407,9 @@ struct bfq_queue {
* the woken queues when this queue exits.
*/
struct hlist_head woken_list;
+
+ /* index of the actuator this queue is associated with */
+ unsigned int actuator_idx;
};
/**
@@ -405,8 +418,17 @@ struct bfq_queue {
struct bfq_io_cq {
/* associated io_cq structure */
struct io_cq icq; /* must be the first member */
- /* array of two process queues, the sync and the async */
- struct bfq_queue *bfqq[2];
+ /*
+ * Matrix of associated process queues: first row for async
+ * queues, second row sync queues. Each row contains one
+ * column for each actuator. An I/O request generated by the
+ * process is inserted into the queue pointed by bfqq[i][j] if
+ * the request is to be served by the j-th actuator of the
+ * drive, where i==0 or i==1, depending on whether the request
+ * is async or sync. So there is a distinct queue for each
+ * actuator.
+ */
+ struct bfq_queue *bfqq[2][BFQ_MAX_ACTUATORS];
/* per (request_queue, blkcg) ioprio */
int ioprio;
#ifdef CONFIG_BFQ_GROUP_IOSCHED
@@ -772,6 +794,13 @@ struct bfq_data {
*/
unsigned int word_depths[2][2];
unsigned int full_depth_shift;
+
+ /*
+ * Number of independent actuators. This is equal to 1 in
+ * case of single-actuator drives.
+ */
+ unsigned int num_actuators;
+
};
enum bfqq_state_flags {
@@ -969,8 +998,10 @@ struct bfq_group {
extern const int bfq_timeout;
-struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync);
-void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync);
+struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync,
+ unsigned int actuator_idx);
+void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync,
+ unsigned int actuator_idx);
struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic);
void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq);
void bfq_weights_tree_add(struct bfq_queue *bfqq);
--
2.20.1
When a bfq_queue Q is merged with another queue, several pieces of
information are saved about Q. These pieces are stored in the
bfqq_data field in the bfq_io_cq data structure of the process
associated with Q.
Yet, with a multi-actuator drive, a process may get associated with
multiple bfq_queues: one queue for each of the N actuators. Each of
these queues may undergo a merge. So, the bfq_io_cq data structure
must be able to accommodate the above information for N queues.
This commit solves this problem by turning the bfqq_data scalar field
into an array of N elements (and by changing code so as to handle
this array).
This solution is written under the assumption that bfq_queues
associated with different actuators cannot be cross-merged. This
assumption holds naturally with basic queue merging: the latter is
triggered by spatial locality, and sectors for different actuators are
not close to each other (apart from the corner case of the last
sectors served by a given actuator and the first sectors served by the
next actuator). As for stable cross-merging, the assumption here is
that it is disabled.
Reviewed-by: Damien Le Moal <[email protected]>
Signed-off-by: Gabriele Felici <[email protected]>
Signed-off-by: Gianmarco Lusvardi <[email protected]>
Signed-off-by: Giulio Barabino <[email protected]>
Signed-off-by: Emiliano Maccaferri <[email protected]>
Signed-off-by: Paolo Valente <[email protected]>
---
block/bfq-iosched.c | 100 ++++++++++++++++++++++++++------------------
block/bfq-iosched.h | 12 ++++--
2 files changed, 67 insertions(+), 45 deletions(-)
diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 92ad0ec4efcc..a9ac7b6f3b81 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -408,7 +408,7 @@ void bic_set_bfqq(struct bfq_io_cq *bic,
* we cancel the stable merge if
* bic->stable_merge_bfqq == bfqq.
*/
- struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data;
+ struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data[actuator_idx];
/* Clear bic pointer if bfqq is detached from this bic */
if (old_bfqq && old_bfqq->bic == bic)
@@ -1139,9 +1139,10 @@ static void
bfq_bfqq_resume_state(struct bfq_queue *bfqq, struct bfq_data *bfqd,
struct bfq_io_cq *bic, bool bfq_already_existing)
{
- struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data;
unsigned int old_wr_coeff = 1;
bool busy = bfq_already_existing && bfq_bfqq_busy(bfqq);
+ unsigned int a_idx = bfqq->actuator_idx;
+ struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data[a_idx];
if (bfqq_data->saved_has_short_ttime)
bfq_mark_bfqq_has_short_ttime(bfqq);
@@ -1822,7 +1823,9 @@ static void bfq_bfqq_handle_idle_busy_switch(struct bfq_data *bfqd,
arrived_in_time = ktime_get_ns() <=
bfqq->ttime.last_end_request +
bfqd->bfq_slice_idle * 3;
-
+ unsigned int act_idx = bfq_actuator_index(bfqd, rq->bio);
+ bool bfqq_non_merged_or_stably_merged =
+ bfqq->bic || RQ_BIC(rq)->bfqq_data[act_idx].stably_merged;
/*
* bfqq deserves to be weight-raised if:
@@ -1856,9 +1859,8 @@ static void bfq_bfqq_handle_idle_busy_switch(struct bfq_data *bfqd,
*/
wr_or_deserves_wr = bfqd->low_latency &&
(bfqq->wr_coeff > 1 ||
- (bfq_bfqq_sync(bfqq) &&
- (bfqq->bic || RQ_BIC(rq)->bfqq_data.stably_merged) &&
- (*interactive || soft_rt)));
+ (bfq_bfqq_sync(bfqq) && bfqq_non_merged_or_stably_merged &&
+ (*interactive || soft_rt)));
/*
* Using the last flag, update budget and check whether bfqq
@@ -2832,6 +2834,35 @@ static bool bfq_may_be_close_cooperator(struct bfq_queue *bfqq,
static bool idling_boosts_thr_without_issues(struct bfq_data *bfqd,
struct bfq_queue *bfqq);
+static struct bfq_queue *
+bfq_setup_stable_merge(struct bfq_data *bfqd, struct bfq_queue *bfqq,
+ struct bfq_queue *stable_merge_bfqq,
+ struct bfq_iocq_bfqq_data *bfqq_data)
+{
+ int proc_ref = min(bfqq_process_refs(bfqq),
+ bfqq_process_refs(stable_merge_bfqq));
+ struct bfq_queue *new_bfqq;
+
+ if (idling_boosts_thr_without_issues(bfqd, bfqq) ||
+ proc_ref == 0)
+ return NULL;
+
+ /* next function will take at least one ref */
+ new_bfqq = bfq_setup_merge(bfqq, stable_merge_bfqq);
+
+ if (new_bfqq) {
+ bfqq_data->stably_merged = true;
+ if (new_bfqq->bic) {
+ unsigned int new_a_idx = new_bfqq->actuator_idx;
+ struct bfq_iocq_bfqq_data *new_bfqq_data =
+ &new_bfqq->bic->bfqq_data[new_a_idx];
+
+ new_bfqq_data->stably_merged = true;
+ }
+ }
+ return new_bfqq;
+}
+
/*
* Attempt to schedule a merge of bfqq with the currently in-service
* queue or with a close queue among the scheduled queues. Return
@@ -2857,7 +2888,8 @@ bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq,
void *io_struct, bool request, struct bfq_io_cq *bic)
{
struct bfq_queue *in_service_bfqq, *new_bfqq;
- struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data;
+ unsigned int a_idx = bfqq->actuator_idx;
+ struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data[a_idx];
/* if a merge has already been setup, then proceed with that first */
if (bfqq->new_bfqq)
@@ -2887,29 +2919,15 @@ bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq,
msecs_to_jiffies(bfq_late_stable_merging))) {
struct bfq_queue *stable_merge_bfqq =
bfqq_data->stable_merge_bfqq;
- int proc_ref = min(bfqq_process_refs(bfqq),
- bfqq_process_refs(stable_merge_bfqq));
/* deschedule stable merge, because done or aborted here */
bfq_put_stable_ref(stable_merge_bfqq);
bfqq_data->stable_merge_bfqq = NULL;
- if (!idling_boosts_thr_without_issues(bfqd, bfqq) &&
- proc_ref > 0) {
- /* next function will take at least one ref */
- struct bfq_queue *new_bfqq =
- bfq_setup_merge(bfqq, stable_merge_bfqq);
-
- if (new_bfqq) {
- bfqq_data->stably_merged = true;
- if (new_bfqq->bic)
- new_bfqq->bic->bfqq_data.stably_merged =
- true;
- }
- return new_bfqq;
- } else
- return NULL;
+ return bfq_setup_stable_merge(bfqd, bfqq,
+ stable_merge_bfqq,
+ bfqq_data);
}
}
@@ -3004,7 +3022,8 @@ bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq,
static void bfq_bfqq_save_state(struct bfq_queue *bfqq)
{
struct bfq_io_cq *bic = bfqq->bic;
- struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data;
+ unsigned int a_idx = bfqq->actuator_idx;
+ struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data[a_idx];
/*
* If !bfqq->bic, the queue is already shared or its requests
@@ -3015,7 +3034,7 @@ static void bfq_bfqq_save_state(struct bfq_queue *bfqq)
return;
bfqq_data->saved_last_serv_time_ns = bfqq->last_serv_time_ns;
- bfqq_data->saved_inject_limit = bfqq->inject_limit;
+ bfqq_data->saved_inject_limit = bfqq->inject_limit;
bfqq_data->saved_decrease_time_jif = bfqq->decrease_time_jif;
bfqq_data->saved_weight = bfqq->entity.orig_weight;
@@ -5381,7 +5400,7 @@ static void bfq_exit_icq(struct io_cq *icq)
* therefore on its unused per-actuator fields being NULL.
*/
unsigned int num_actuators = BFQ_MAX_ACTUATORS;
- struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data;
+ struct bfq_iocq_bfqq_data *bfqq_data = bic->bfqq_data;
/*
* bfqd is NULL if scheduler already exited, and in that case
@@ -5392,10 +5411,10 @@ static void bfq_exit_icq(struct io_cq *icq)
num_actuators = bfqd->num_actuators;
}
- if (bfqq_data->stable_merge_bfqq)
- bfq_put_stable_ref(bfqq_data->stable_merge_bfqq);
-
for (act_idx = 0; act_idx < num_actuators; act_idx++) {
+ if (bfqq_data[act_idx].stable_merge_bfqq)
+ bfq_put_stable_ref(bfqq_data[act_idx].stable_merge_bfqq);
+
bfq_exit_icq_bfqq(bic, true, act_idx);
bfq_exit_icq_bfqq(bic, false, act_idx);
}
@@ -5582,16 +5601,16 @@ bfq_do_early_stable_merge(struct bfq_data *bfqd, struct bfq_queue *bfqq,
struct bfq_io_cq *bic,
struct bfq_queue *last_bfqq_created)
{
+ unsigned int a_idx = last_bfqq_created->actuator_idx;
struct bfq_queue *new_bfqq =
bfq_setup_merge(bfqq, last_bfqq_created);
- struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data;
if (!new_bfqq)
return bfqq;
if (new_bfqq->bic)
- new_bfqq->bic->bfqq_data.stably_merged = true;
- bfqq_data->stably_merged = true;
+ new_bfqq->bic->bfqq_data[a_idx].stably_merged = true;
+ bic->bfqq_data[a_idx].stably_merged = true;
/*
* Reusing merge functions. This implies that
@@ -5660,7 +5679,6 @@ static struct bfq_queue *bfq_do_or_sched_stable_merge(struct bfq_data *bfqd,
&bfqd->last_bfqq_created;
struct bfq_queue *last_bfqq_created = *source_bfqq;
- struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data;
/*
* If last_bfqq_created has not been set yet, then init it. If
@@ -5722,7 +5740,8 @@ static struct bfq_queue *bfq_do_or_sched_stable_merge(struct bfq_data *bfqd,
/*
* Record the bfqq to merge to.
*/
- bfqq_data->stable_merge_bfqq = last_bfqq_created;
+ bic->bfqq_data[last_bfqq_created->actuator_idx].stable_merge_bfqq =
+ last_bfqq_created;
}
}
@@ -6643,7 +6662,7 @@ static struct bfq_queue *bfq_get_bfqq_handle_split(struct bfq_data *bfqd,
{
unsigned int act_idx = bfq_actuator_index(bfqd, bio);
struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync, act_idx);
- struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data;
+ struct bfq_iocq_bfqq_data *bfqq_data = &bic->bfqq_data[act_idx];
if (likely(bfqq && bfqq != &bfqd->oom_bfqq))
return bfqq;
@@ -6751,7 +6770,7 @@ static struct bfq_queue *bfq_init_rq(struct request *rq)
struct bfq_queue *bfqq;
bool new_queue = false;
bool bfqq_already_existing = false, split = false;
- struct bfq_iocq_bfqq_data *bfqq_data;
+ unsigned int a_idx = bfq_actuator_index(bfqd, bio);
if (unlikely(!rq->elv.icq))
return NULL;
@@ -6775,17 +6794,16 @@ static struct bfq_queue *bfq_init_rq(struct request *rq)
bfqq = bfq_get_bfqq_handle_split(bfqd, bic, bio, false, is_sync,
&new_queue);
- bfqq_data = &bic->bfqq_data;
-
if (likely(!new_queue)) {
/* If the queue was seeky for too long, break it apart. */
if (bfq_bfqq_coop(bfqq) && bfq_bfqq_split_coop(bfqq) &&
- !bfqq_data->stably_merged) {
+ !bic->bfqq_data[a_idx].stably_merged) {
struct bfq_queue *old_bfqq = bfqq;
/* Update bic before losing reference to bfqq */
if (bfq_bfqq_in_large_burst(bfqq))
- bfqq_data->saved_in_large_burst = true;
+ bic->bfqq_data[a_idx].saved_in_large_burst =
+ true;
bfqq = bfq_split_bfqq(bic, bfqq);
split = true;
diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h
index 8e0215b8cc30..5a6f888930ad 100644
--- a/block/bfq-iosched.h
+++ b/block/bfq-iosched.h
@@ -418,7 +418,7 @@ struct bfq_queue {
struct bfq_iocq_bfqq_data {
/*
* Snapshot of the has_short_time flag before merging; taken
- * to remember its value while the queue is merged, so as to
+ * to remember its values while the queue is merged, so as to
* be able to restore it in case of split.
*/
bool saved_has_short_ttime;
@@ -432,7 +432,7 @@ struct bfq_iocq_bfqq_data {
u64 saved_tot_idle_time;
/*
- * Same purpose as the previous fields for the value of the
+ * Same purpose as the previous fields for the values of the
* field keeping the queue's belonging to a large burst
*/
bool saved_in_large_burst;
@@ -495,8 +495,12 @@ struct bfq_io_cq {
uint64_t blkcg_serial_nr; /* the current blkcg serial */
#endif
- /* persistent data for associated synchronous process queue */
- struct bfq_iocq_bfqq_data bfqq_data;
+ /*
+ * Persistent data for associated synchronous process queues
+ * (one queue per actuator, see field bfqq above). In
+ * particular, each of these queues may undergo a merge.
+ */
+ struct bfq_iocq_bfqq_data bfqq_data[BFQ_MAX_ACTUATORS];
unsigned int requests; /* Number of requests this process has in flight */
};
--
2.20.1
From: Davide Zini <[email protected]>
Similarly to sync bfq_queues, also async bfq_queues need to be split
on a per-actuator basis.
Reviewed-by: Damien Le Moal <[email protected]>
Signed-off-by: Paolo Valente <[email protected]>
Signed-off-by: Davide Zini <[email protected]>
---
block/bfq-iosched.c | 41 +++++++++++++++++++++++------------------
block/bfq-iosched.h | 8 ++++----
2 files changed, 27 insertions(+), 22 deletions(-)
diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index a9ac7b6f3b81..f29444d53e6a 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -2624,14 +2624,16 @@ static void bfq_bfqq_end_wr(struct bfq_queue *bfqq)
void bfq_end_wr_async_queues(struct bfq_data *bfqd,
struct bfq_group *bfqg)
{
- int i, j;
-
- for (i = 0; i < 2; i++)
- for (j = 0; j < IOPRIO_NR_LEVELS; j++)
- if (bfqg->async_bfqq[i][j])
- bfq_bfqq_end_wr(bfqg->async_bfqq[i][j]);
- if (bfqg->async_idle_bfqq)
- bfq_bfqq_end_wr(bfqg->async_idle_bfqq);
+ int i, j, k;
+
+ for (k = 0; k < bfqd->num_actuators; k++) {
+ for (i = 0; i < 2; i++)
+ for (j = 0; j < IOPRIO_NR_LEVELS; j++)
+ if (bfqg->async_bfqq[i][j][k])
+ bfq_bfqq_end_wr(bfqg->async_bfqq[i][j][k]);
+ if (bfqg->async_idle_bfqq[k])
+ bfq_bfqq_end_wr(bfqg->async_idle_bfqq[k]);
+ }
}
static void bfq_end_wr(struct bfq_data *bfqd)
@@ -5579,18 +5581,18 @@ static void bfq_init_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
static struct bfq_queue **bfq_async_queue_prio(struct bfq_data *bfqd,
struct bfq_group *bfqg,
- int ioprio_class, int ioprio)
+ int ioprio_class, int ioprio, int act_idx)
{
switch (ioprio_class) {
case IOPRIO_CLASS_RT:
- return &bfqg->async_bfqq[0][ioprio];
+ return &bfqg->async_bfqq[0][ioprio][act_idx];
case IOPRIO_CLASS_NONE:
ioprio = IOPRIO_BE_NORM;
fallthrough;
case IOPRIO_CLASS_BE:
- return &bfqg->async_bfqq[1][ioprio];
+ return &bfqg->async_bfqq[1][ioprio][act_idx];
case IOPRIO_CLASS_IDLE:
- return &bfqg->async_idle_bfqq;
+ return &bfqg->async_idle_bfqq[act_idx];
default:
return NULL;
}
@@ -5763,7 +5765,8 @@ static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd,
bfqg = bfq_bio_bfqg(bfqd, bio);
if (!is_sync) {
async_bfqq = bfq_async_queue_prio(bfqd, bfqg, ioprio_class,
- ioprio);
+ ioprio,
+ bfq_actuator_index(bfqd, bio));
bfqq = *async_bfqq;
if (bfqq)
goto out;
@@ -6986,13 +6989,15 @@ static void __bfq_put_async_bfqq(struct bfq_data *bfqd,
*/
void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg)
{
- int i, j;
+ int i, j, k;
- for (i = 0; i < 2; i++)
- for (j = 0; j < IOPRIO_NR_LEVELS; j++)
- __bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j]);
+ for (k = 0; k < bfqd->num_actuators; k++) {
+ for (i = 0; i < 2; i++)
+ for (j = 0; j < IOPRIO_NR_LEVELS; j++)
+ __bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j][k]);
- __bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq);
+ __bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq[k]);
+ }
}
/*
diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h
index 5a6f888930ad..de2b2af643e5 100644
--- a/block/bfq-iosched.h
+++ b/block/bfq-iosched.h
@@ -980,8 +980,8 @@ struct bfq_group {
struct bfq_data *bfqd;
- struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS];
- struct bfq_queue *async_idle_bfqq;
+ struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS][BFQ_MAX_ACTUATORS];
+ struct bfq_queue *async_idle_bfqq[BFQ_MAX_ACTUATORS];
struct bfq_entity *my_entity;
@@ -998,8 +998,8 @@ struct bfq_group {
struct bfq_entity entity;
struct bfq_sched_data sched_data;
- struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS];
- struct bfq_queue *async_idle_bfqq;
+ struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS][BFQ_MAX_ACTUATORS];
+ struct bfq_queue *async_idle_bfqq[BFQ_MAX_ACTUATORS];
struct rb_root rq_pos_tree;
};
--
2.20.1
From: Davide Zini <[email protected]>
Upon the invocation of its dispatch function, BFQ returns the next I/O
request of the in-service bfq_queue, unless some exception holds. One
such exception is that there is some underutilized actuator, different
from the actuator for which the in-service queue contains I/O, and
that some other bfq_queue happens to contain I/O for such an
actuator. In this case, the next I/O request of the latter bfq_queue,
and not of the in-service bfq_queue, is returned (I/O is injected from
that bfq_queue). To find such an actuator, a linear scan, in
increasing index order, is performed among actuators.
Performing a linear scan entails a prioritization among actuators: an
underutilized actuator may be considered for injection only if all
actuators with a lower index are currently fully utilized, or if there
is no pending I/O for any lower-index actuator that happens to be
underutilized.
This commits breaks this prioritization and tends to distribute
injection uniformly across actuators. This is obtained by adding the
following condition to the linear scan: even if an actuator A is
underutilized, A is however skipped if its load is higher than that of
the next actuator.
Reviewed-by: Damien Le Moal <[email protected]>
Signed-off-by: Paolo Valente <[email protected]>
Signed-off-by: Davide Zini <[email protected]>
---
block/bfq-iosched.c | 18 +++++++++++++-----
1 file changed, 13 insertions(+), 5 deletions(-)
diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 132c5874eefc..f69dfa659e34 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -4771,10 +4771,16 @@ bfq_find_active_bfqq_for_actuator(struct bfq_data *bfqd, int idx)
/*
* Perform a linear scan of each actuator, until an actuator is found
- * for which the following two conditions hold: the load of the
- * actuator is below the threshold (see comments on actuator_load_threshold
- * for details), and there is a queue that contains I/O for that
- * actuator. On success, return that queue.
+ * for which the following three conditions hold: the load of the
+ * actuator is below the threshold (see comments on
+ * actuator_load_threshold for details) and lower than that of the
+ * next actuator (comments on this extra condition below), and there
+ * is a queue that contains I/O for that actuator. On success, return
+ * that queue.
+ *
+ * Performing a plain linear scan entails a prioritization among
+ * actuators. The extra condition above breaks this prioritization and
+ * tends to distribute injection uniformly across actuators.
*/
static struct bfq_queue *
bfq_find_bfqq_for_underused_actuator(struct bfq_data *bfqd)
@@ -4782,7 +4788,9 @@ bfq_find_bfqq_for_underused_actuator(struct bfq_data *bfqd)
int i;
for (i = 0 ; i < bfqd->num_actuators; i++) {
- if (bfqd->rq_in_driver[i] < bfqd->actuator_load_threshold) {
+ if (bfqd->rq_in_driver[i] < bfqd->actuator_load_threshold &&
+ (i == bfqd->num_actuators - 1 ||
+ bfqd->rq_in_driver[i] < bfqd->rq_in_driver[i+1])) {
struct bfq_queue *bfqq =
bfq_find_active_bfqq_for_actuator(bfqd, i);
--
2.20.1
From: Davide Zini <[email protected]>
The main service scheme of BFQ for sync I/O is serving one sync
bfq_queue at a time, for a while. In particular, BFQ enforces this
scheme when it deems the latter necessary to boost throughput or
to preserve service guarantees. Unfortunately, when BFQ enforces
this policy, only one actuator at a time gets served for a while,
because each bfq_queue contains I/O only for one actuator. The
other actuators may remain underutilized.
Actually, BFQ may serve (inject) extra I/O, taken from other
bfq_queues, in parallel with that of the in-service queue. This
injection mechanism may provide the ground for dealing also with
the above actuator-underutilization problem. Yet BFQ does not take
the actuator load into account when choosing which queue to pick
extra I/O from. In addition, BFQ may happen to inject extra I/O
only when the in-service queue is temporarily empty.
In view of these facts, this commit extends the
injection mechanism in such a way that the latter:
(1) takes into account also the actuator load;
(2) checks such a load on each dispatch, and injects I/O for an
underutilized actuator, if there is one and there is I/O for it.
To perform the check in (2), this commit introduces a load
threshold, currently set to 4. A linear scan of each actuator is
performed, until an actuator is found for which the following two
conditions hold: the load of the actuator is below the threshold,
and there is at least one non-in-service queue that contains I/O
for that actuator. If such a pair (actuator, queue) is found, then
the head request of that queue is returned for dispatch, instead
of the head request of the in-service queue.
We have set the threshold, empirically, to the minimum possible
value for which an actuator is fully utilized, or close to be
fully utilized. By doing so, injected I/O 'steals' as few
drive-queue slots as possibile to the in-service queue. This
reduces as much as possible the probability that the service of
I/O from the in-service bfq_queue gets delayed because of slot
exhaustion, i.e., because all the slots of the drive queue are
filled with I/O injected from other queues (NCQ provides for 32
slots).
This new mechanism also counters actuator underutilization in the
case of asymmetric configurations of bfq_queues. Namely if there
are few bfq_queues containing I/O for some actuators and many
bfq_queues containing I/O for other actuators. Or if the
bfq_queues containing I/O for some actuators have lower weights
than the other bfq_queues.
Reviewed-by: Damien Le Moal <[email protected]>
Signed-off-by: Paolo Valente <[email protected]>
Signed-off-by: Davide Zini <[email protected]>
---
block/bfq-cgroup.c | 2 +-
block/bfq-iosched.c | 136 ++++++++++++++++++++++++++++++++------------
block/bfq-iosched.h | 39 ++++++++++++-
block/bfq-wf2q.c | 2 +-
4 files changed, 139 insertions(+), 40 deletions(-)
diff --git a/block/bfq-cgroup.c b/block/bfq-cgroup.c
index bd7bf0f8121d..a6e8da5f5cfd 100644
--- a/block/bfq-cgroup.c
+++ b/block/bfq-cgroup.c
@@ -708,7 +708,7 @@ void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
bfq_activate_bfqq(bfqd, bfqq);
}
- if (!bfqd->in_service_queue && !bfqd->rq_in_driver)
+ if (!bfqd->in_service_queue && !bfqd->tot_rq_in_driver)
bfq_schedule_dispatch(bfqd);
/* release extra ref taken above, bfqq may happen to be freed now */
bfq_put_queue(bfqq);
diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 1db7fb18f99e..132c5874eefc 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -2263,9 +2263,9 @@ static void bfq_add_request(struct request *rq)
* elapsed.
*/
if (bfqq == bfqd->in_service_queue &&
- (bfqd->rq_in_driver == 0 ||
+ (bfqd->tot_rq_in_driver == 0 ||
(bfqq->last_serv_time_ns > 0 &&
- bfqd->rqs_injected && bfqd->rq_in_driver > 0)) &&
+ bfqd->rqs_injected && bfqd->tot_rq_in_driver > 0)) &&
time_is_before_eq_jiffies(bfqq->decrease_time_jif +
msecs_to_jiffies(10))) {
bfqd->last_empty_occupied_ns = ktime_get_ns();
@@ -2289,7 +2289,7 @@ static void bfq_add_request(struct request *rq)
* will be set in case injection is performed
* on bfqq before rq is completed).
*/
- if (bfqd->rq_in_driver == 0)
+ if (bfqd->tot_rq_in_driver == 0)
bfqd->rqs_injected = false;
}
}
@@ -2654,11 +2654,14 @@ void bfq_end_wr_async_queues(struct bfq_data *bfqd,
static void bfq_end_wr(struct bfq_data *bfqd)
{
struct bfq_queue *bfqq;
+ int i;
spin_lock_irq(&bfqd->lock);
- list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
- bfq_bfqq_end_wr(bfqq);
+ for (i = 0; i < bfqd->num_actuators; i++) {
+ list_for_each_entry(bfqq, &bfqd->active_list[i], bfqq_list)
+ bfq_bfqq_end_wr(bfqq);
+ }
list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list)
bfq_bfqq_end_wr(bfqq);
bfq_end_wr_async(bfqd);
@@ -3615,13 +3618,13 @@ static void bfq_update_peak_rate(struct bfq_data *bfqd, struct request *rq)
* - start a new observation interval with this dispatch
*/
if (now_ns - bfqd->last_dispatch > 100*NSEC_PER_MSEC &&
- bfqd->rq_in_driver == 0)
+ bfqd->tot_rq_in_driver == 0)
goto update_rate_and_reset;
/* Update sampling information */
bfqd->peak_rate_samples++;
- if ((bfqd->rq_in_driver > 0 ||
+ if ((bfqd->tot_rq_in_driver > 0 ||
now_ns - bfqd->last_completion < BFQ_MIN_TT)
&& !BFQ_RQ_SEEKY(bfqd, bfqd->last_position, rq))
bfqd->sequential_samples++;
@@ -3886,10 +3889,8 @@ static bool idling_needed_for_service_guarantees(struct bfq_data *bfqd,
return false;
return (bfqq->wr_coeff > 1 &&
- (bfqd->wr_busy_queues <
- tot_busy_queues ||
- bfqd->rq_in_driver >=
- bfqq->dispatched + 4)) ||
+ (bfqd->wr_busy_queues < tot_busy_queues ||
+ bfqd->tot_rq_in_driver >= bfqq->dispatched + 4)) ||
bfq_asymmetric_scenario(bfqd, bfqq) ||
tot_busy_queues == 1;
}
@@ -4660,6 +4661,8 @@ bfq_choose_bfqq_for_injection(struct bfq_data *bfqd)
{
struct bfq_queue *bfqq, *in_serv_bfqq = bfqd->in_service_queue;
unsigned int limit = in_serv_bfqq->inject_limit;
+ int i;
+
/*
* If
* - bfqq is not weight-raised and therefore does not carry
@@ -4691,7 +4694,7 @@ bfq_choose_bfqq_for_injection(struct bfq_data *bfqd)
)
limit = 1;
- if (bfqd->rq_in_driver >= limit)
+ if (bfqd->tot_rq_in_driver >= limit)
return NULL;
/*
@@ -4706,11 +4709,12 @@ bfq_choose_bfqq_for_injection(struct bfq_data *bfqd)
* (and re-added only if it gets new requests, but then it
* is assigned again enough budget for its new backlog).
*/
- list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
- if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&
- (in_serv_always_inject || bfqq->wr_coeff > 1) &&
- bfq_serv_to_charge(bfqq->next_rq, bfqq) <=
- bfq_bfqq_budget_left(bfqq)) {
+ for (i = 0; i < bfqd->num_actuators; i++) {
+ list_for_each_entry(bfqq, &bfqd->active_list[i], bfqq_list)
+ if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&
+ (in_serv_always_inject || bfqq->wr_coeff > 1) &&
+ bfq_serv_to_charge(bfqq->next_rq, bfqq) <=
+ bfq_bfqq_budget_left(bfqq)) {
/*
* Allow for only one large in-flight request
* on non-rotational devices, for the
@@ -4735,22 +4739,69 @@ bfq_choose_bfqq_for_injection(struct bfq_data *bfqd)
else
limit = in_serv_bfqq->inject_limit;
- if (bfqd->rq_in_driver < limit) {
+ if (bfqd->tot_rq_in_driver < limit) {
bfqd->rqs_injected = true;
return bfqq;
}
}
+ }
+
+ return NULL;
+}
+
+static struct bfq_queue *
+bfq_find_active_bfqq_for_actuator(struct bfq_data *bfqd, int idx)
+{
+ struct bfq_queue *bfqq;
+
+ if (bfqd->in_service_queue &&
+ bfqd->in_service_queue->actuator_idx == idx)
+ return bfqd->in_service_queue;
+
+ list_for_each_entry(bfqq, &bfqd->active_list[idx], bfqq_list) {
+ if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&
+ bfq_serv_to_charge(bfqq->next_rq, bfqq) <=
+ bfq_bfqq_budget_left(bfqq)) {
+ return bfqq;
+ }
+ }
return NULL;
}
+/*
+ * Perform a linear scan of each actuator, until an actuator is found
+ * for which the following two conditions hold: the load of the
+ * actuator is below the threshold (see comments on actuator_load_threshold
+ * for details), and there is a queue that contains I/O for that
+ * actuator. On success, return that queue.
+ */
+static struct bfq_queue *
+bfq_find_bfqq_for_underused_actuator(struct bfq_data *bfqd)
+{
+ int i;
+
+ for (i = 0 ; i < bfqd->num_actuators; i++) {
+ if (bfqd->rq_in_driver[i] < bfqd->actuator_load_threshold) {
+ struct bfq_queue *bfqq =
+ bfq_find_active_bfqq_for_actuator(bfqd, i);
+
+ if (bfqq)
+ return bfqq;
+ }
+ }
+
+ return NULL;
+}
+
+
/*
* Select a queue for service. If we have a current queue in service,
* check whether to continue servicing it, or retrieve and set a new one.
*/
static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd)
{
- struct bfq_queue *bfqq;
+ struct bfq_queue *bfqq, *inject_bfqq;
struct request *next_rq;
enum bfqq_expiration reason = BFQQE_BUDGET_TIMEOUT;
@@ -4772,6 +4823,15 @@ static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd)
goto expire;
check_queue:
+ /*
+ * If some actuator is underutilized, but the in-service
+ * queue does not contain I/O for that actuator, then try to
+ * inject I/O for that actuator.
+ */
+ inject_bfqq = bfq_find_bfqq_for_underused_actuator(bfqd);
+ if (inject_bfqq && inject_bfqq != bfqq)
+ return inject_bfqq;
+
/*
* This loop is rarely executed more than once. Even when it
* happens, it is much more convenient to re-execute this loop
@@ -5127,11 +5187,11 @@ static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
/*
* We exploit the bfq_finish_requeue_request hook to
- * decrement rq_in_driver, but
+ * decrement tot_rq_in_driver, but
* bfq_finish_requeue_request will not be invoked on
* this request. So, to avoid unbalance, just start
- * this request, without incrementing rq_in_driver. As
- * a negative consequence, rq_in_driver is deceptively
+ * this request, without incrementing tot_rq_in_driver. As
+ * a negative consequence, tot_rq_in_driver is deceptively
* lower than it should be while this request is in
* service. This may cause bfq_schedule_dispatch to be
* invoked uselessly.
@@ -5140,7 +5200,7 @@ static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
* bfq_finish_requeue_request hook, if defined, is
* probably invoked also on this request. So, by
* exploiting this hook, we could 1) increment
- * rq_in_driver here, and 2) decrement it in
+ * tot_rq_in_driver here, and 2) decrement it in
* bfq_finish_requeue_request. Such a solution would
* let the value of the counter be always accurate,
* but it would entail using an extra interface
@@ -5169,7 +5229,7 @@ static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
* Of course, serving one request at a time may cause loss of
* throughput.
*/
- if (bfqd->strict_guarantees && bfqd->rq_in_driver > 0)
+ if (bfqd->strict_guarantees && bfqd->tot_rq_in_driver > 0)
goto exit;
bfqq = bfq_select_queue(bfqd);
@@ -5180,7 +5240,8 @@ static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
if (rq) {
inc_in_driver_start_rq:
- bfqd->rq_in_driver++;
+ bfqd->rq_in_driver[bfqq->actuator_idx]++;
+ bfqd->tot_rq_in_driver++;
start_rq:
rq->rq_flags |= RQF_STARTED;
}
@@ -6247,7 +6308,7 @@ static void bfq_update_hw_tag(struct bfq_data *bfqd)
struct bfq_queue *bfqq = bfqd->in_service_queue;
bfqd->max_rq_in_driver = max_t(int, bfqd->max_rq_in_driver,
- bfqd->rq_in_driver);
+ bfqd->tot_rq_in_driver);
if (bfqd->hw_tag == 1)
return;
@@ -6258,7 +6319,7 @@ static void bfq_update_hw_tag(struct bfq_data *bfqd)
* sum is not exact, as it's not taking into account deactivated
* requests.
*/
- if (bfqd->rq_in_driver + bfqd->queued <= BFQ_HW_QUEUE_THRESHOLD)
+ if (bfqd->tot_rq_in_driver + bfqd->queued <= BFQ_HW_QUEUE_THRESHOLD)
return;
/*
@@ -6269,7 +6330,7 @@ static void bfq_update_hw_tag(struct bfq_data *bfqd)
if (bfqq && bfq_bfqq_has_short_ttime(bfqq) &&
bfqq->dispatched + bfqq->queued[0] + bfqq->queued[1] <
BFQ_HW_QUEUE_THRESHOLD &&
- bfqd->rq_in_driver < BFQ_HW_QUEUE_THRESHOLD)
+ bfqd->tot_rq_in_driver < BFQ_HW_QUEUE_THRESHOLD)
return;
if (bfqd->hw_tag_samples++ < BFQ_HW_QUEUE_SAMPLES)
@@ -6290,7 +6351,8 @@ static void bfq_completed_request(struct bfq_queue *bfqq, struct bfq_data *bfqd)
bfq_update_hw_tag(bfqd);
- bfqd->rq_in_driver--;
+ bfqd->rq_in_driver[bfqq->actuator_idx]--;
+ bfqd->tot_rq_in_driver--;
bfqq->dispatched--;
if (!bfqq->dispatched && !bfq_bfqq_busy(bfqq)) {
@@ -6410,7 +6472,7 @@ static void bfq_completed_request(struct bfq_queue *bfqq, struct bfq_data *bfqd)
BFQQE_NO_MORE_REQUESTS);
}
- if (!bfqd->rq_in_driver)
+ if (!bfqd->tot_rq_in_driver)
bfq_schedule_dispatch(bfqd);
}
@@ -6541,13 +6603,13 @@ static void bfq_update_inject_limit(struct bfq_data *bfqd,
* conditions to do it, or we can lower the last base value
* computed.
*
- * NOTE: (bfqd->rq_in_driver == 1) means that there is no I/O
+ * NOTE: (bfqd->tot_rq_in_driver == 1) means that there is no I/O
* request in flight, because this function is in the code
* path that handles the completion of a request of bfqq, and,
* in particular, this function is executed before
- * bfqd->rq_in_driver is decremented in such a code path.
+ * bfqd->tot_rq_in_driver is decremented in such a code path.
*/
- if ((bfqq->last_serv_time_ns == 0 && bfqd->rq_in_driver == 1) ||
+ if ((bfqq->last_serv_time_ns == 0 && bfqd->tot_rq_in_driver == 1) ||
tot_time_ns < bfqq->last_serv_time_ns) {
if (bfqq->last_serv_time_ns == 0) {
/*
@@ -6557,7 +6619,7 @@ static void bfq_update_inject_limit(struct bfq_data *bfqd,
bfqq->inject_limit = max_t(unsigned int, 1, old_limit);
}
bfqq->last_serv_time_ns = tot_time_ns;
- } else if (!bfqd->rqs_injected && bfqd->rq_in_driver == 1)
+ } else if (!bfqd->rqs_injected && bfqd->tot_rq_in_driver == 1)
/*
* No I/O injected and no request still in service in
* the drive: these are the exact conditions for
@@ -7212,7 +7274,8 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
bfqd->num_groups_with_pending_reqs = 0;
#endif
- INIT_LIST_HEAD(&bfqd->active_list);
+ INIT_LIST_HEAD(&bfqd->active_list[0]);
+ INIT_LIST_HEAD(&bfqd->active_list[1]);
INIT_LIST_HEAD(&bfqd->idle_list);
INIT_HLIST_HEAD(&bfqd->burst_list);
@@ -7257,6 +7320,9 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
ref_wr_duration[blk_queue_nonrot(bfqd->queue)];
bfqd->peak_rate = ref_rate[blk_queue_nonrot(bfqd->queue)] * 2 / 3;
+ /* see comments on the definition of next field inside bfq_data */
+ bfqd->actuator_load_threshold = 4;
+
spin_lock_init(&bfqd->lock);
/*
diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h
index 830dda1f9322..058af701bbbe 100644
--- a/block/bfq-iosched.h
+++ b/block/bfq-iosched.h
@@ -590,7 +590,12 @@ struct bfq_data {
/* number of queued requests */
int queued;
/* number of requests dispatched and waiting for completion */
- int rq_in_driver;
+ int tot_rq_in_driver;
+ /*
+ * number of requests dispatched and waiting for completion
+ * for each actuator
+ */
+ int rq_in_driver[BFQ_MAX_ACTUATORS];
/* true if the device is non rotational and performs queueing */
bool nonrot_with_queueing;
@@ -684,8 +689,13 @@ struct bfq_data {
/* maximum budget allotted to a bfq_queue before rescheduling */
int bfq_max_budget;
- /* list of all the bfq_queues active on the device */
- struct list_head active_list;
+ /*
+ * List of all the bfq_queues active for a specific actuator
+ * on the device. Keeping active queues separate on a
+ * per-actuator basis helps implementing per-actuator
+ * injection more efficiently.
+ */
+ struct list_head active_list[BFQ_MAX_ACTUATORS];
/* list of all the bfq_queues idle on the device */
struct list_head idle_list;
@@ -821,6 +831,29 @@ struct bfq_data {
sector_t sector[BFQ_MAX_ACTUATORS];
sector_t nr_sectors[BFQ_MAX_ACTUATORS];
struct blk_independent_access_range ia_ranges[BFQ_MAX_ACTUATORS];
+
+ /*
+ * If the number of I/O requests queued in the device for a
+ * given actuator is below next threshold, then the actuator
+ * is deemed as underutilized. If this condition is found to
+ * hold for some actuator upon a dispatch, but (i) the
+ * in-service queue does not contain I/O for that actuator,
+ * while (ii) some other queue does contain I/O for that
+ * actuator, then the head I/O request of the latter queue is
+ * returned (injected), instead of the head request of the
+ * currently in-service queue.
+ *
+ * We set the threshold, empirically, to the minimum possible
+ * value for which an actuator is fully utilized, or close to
+ * be fully utilized. By doing so, injected I/O 'steals' as
+ * few drive-queue slots as possibile to the in-service
+ * queue. This reduces as much as possible the probability
+ * that the service of I/O from the in-service bfq_queue gets
+ * delayed because of slot exhaustion, i.e., because all the
+ * slots of the drive queue are filled with I/O injected from
+ * other queues (NCQ provides for 32 slots).
+ */
+ unsigned int actuator_load_threshold;
};
enum bfqq_state_flags {
diff --git a/block/bfq-wf2q.c b/block/bfq-wf2q.c
index ea4c3d757fdd..7941b6f07391 100644
--- a/block/bfq-wf2q.c
+++ b/block/bfq-wf2q.c
@@ -493,7 +493,7 @@ static void bfq_active_insert(struct bfq_service_tree *st,
bfq_update_active_tree(node);
if (bfqq)
- list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list);
+ list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list[bfqq->actuator_idx]);
bfq_inc_active_entities(entity);
}
--
2.20.1
From: Federico Gavioli <[email protected]>
This patch implements the code to gather the content of the
independent_access_ranges structure from the request_queue and copy
it into the queue's bfq_data. This copy is done at queue initialization.
We copy the access ranges into the bfq_data to avoid taking the queue
lock each time we access the ranges.
This implementation, however, puts a limit to the maximum independent
ranges supported by the scheduler. Such a limit is equal to the constant
BFQ_MAX_ACTUATORS. This limit was placed to avoid the allocation of
dynamic memory.
Reviewed-by: Damien Le Moal <[email protected]>
Co-developed-by: Rory Chen <[email protected]>
Signed-off-by: Rory Chen <[email protected]>
Signed-off-by: Federico Gavioli <[email protected]>
Signed-off-by: Paolo Valente <[email protected]>
---
block/bfq-iosched.c | 59 +++++++++++++++++++++++++++++++++++++++------
block/bfq-iosched.h | 8 +++++-
2 files changed, 58 insertions(+), 9 deletions(-)
diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index f29444d53e6a..1db7fb18f99e 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -1797,10 +1797,25 @@ static bool bfq_bfqq_higher_class_or_weight(struct bfq_queue *bfqq,
*/
static unsigned int bfq_actuator_index(struct bfq_data *bfqd, struct bio *bio)
{
- /*
- * Multi-actuator support not complete yet, so always return 0
- * for the moment (to keep incomplete mechanisms off).
- */
+ unsigned int i;
+ sector_t end;
+
+ /* no search needed if one or zero ranges present */
+ if (bfqd->num_actuators == 1)
+ return 0;
+
+ /* bio_end_sector(bio) gives the sector after the last one */
+ end = bio_end_sector(bio) - 1;
+
+ for (i = 0; i < bfqd->num_actuators; i++) {
+ if (end >= bfqd->sector[i] &&
+ end < bfqd->sector[i] + bfqd->nr_sectors[i])
+ return i;
+ }
+
+ WARN_ONCE(true,
+ "bfq_actuator_index: bio sector out of ranges: end=%llu\n",
+ end);
return 0;
}
@@ -7109,6 +7124,8 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
{
struct bfq_data *bfqd;
struct elevator_queue *eq;
+ unsigned int i;
+ struct blk_independent_access_ranges *ia_ranges = q->disk->ia_ranges;
eq = elevator_alloc(q, e);
if (!eq)
@@ -7151,12 +7168,38 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
bfqd->queue = q;
+ bfqd->num_actuators = 1;
/*
- * Multi-actuator support not complete yet, unconditionally
- * set to only one actuator for the moment (to keep incomplete
- * mechanisms off).
+ * If the disk supports multiple actuators, copy independent
+ * access ranges from the request queue structure.
*/
- bfqd->num_actuators = 1;
+ spin_lock_irq(&q->queue_lock);
+ if (ia_ranges) {
+ /*
+ * Check if the disk ia_ranges size exceeds the current bfq
+ * actuator limit.
+ */
+ if (ia_ranges->nr_ia_ranges > BFQ_MAX_ACTUATORS) {
+ pr_crit("nr_ia_ranges higher than act limit: iars=%d, max=%d.\n",
+ ia_ranges->nr_ia_ranges, BFQ_MAX_ACTUATORS);
+ pr_crit("Falling back to single actuator mode.\n");
+ } else {
+ bfqd->num_actuators = ia_ranges->nr_ia_ranges;
+
+ for (i = 0; i < bfqd->num_actuators; i++) {
+ bfqd->sector[i] = ia_ranges->ia_range[i].sector;
+ bfqd->nr_sectors[i] =
+ ia_ranges->ia_range[i].nr_sectors;
+ }
+ }
+ }
+
+ /* Otherwise use single-actuator dev info */
+ if (bfqd->num_actuators == 1) {
+ bfqd->sector[0] = 0;
+ bfqd->nr_sectors[0] = get_capacity(q->disk);
+ }
+ spin_unlock_irq(&q->queue_lock);
INIT_LIST_HEAD(&bfqd->dispatch);
diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h
index de2b2af643e5..830dda1f9322 100644
--- a/block/bfq-iosched.h
+++ b/block/bfq-iosched.h
@@ -814,7 +814,13 @@ struct bfq_data {
* case of single-actuator drives.
*/
unsigned int num_actuators;
-
+ /*
+ * Disk independent access ranges for each actuator
+ * in this device.
+ */
+ sector_t sector[BFQ_MAX_ACTUATORS];
+ sector_t nr_sectors[BFQ_MAX_ACTUATORS];
+ struct blk_independent_access_range ia_ranges[BFQ_MAX_ACTUATORS];
};
enum bfqq_state_flags {
--
2.20.1
On 12/23/22 00:21, Paolo Valente wrote:
> Single-LUN multi-actuator SCSI drives, as well as all multi-actuator
> SATA drives appear as a single device to the I/O subsystem [1]. Yet
> they address commands to different actuators internally, as a function
> of Logical Block Addressing (LBAs). A given sector is reachable by
> only one of the actuators. For example, Seagate’s Serial Advanced
> Technology Attachment (SATA) version contains two actuators and maps
> the lower half of the SATA LBA space to the lower actuator and the
> upper half to the upper actuator.
>
> Evidently, to fully utilize actuators, no actuator must be left idle
> or underutilized while there is pending I/O for it. The block layer
> must somehow control the load of each actuator individually. This
> commit lays the ground for allowing BFQ to provide such a per-actuator
> control.
>
> BFQ associates an I/O-request sync bfq_queue with each process doing
> synchronous I/O, or with a group of processes, in case of queue
> merging. Then BFQ serves one bfq_queue at a time. While in service, a
> bfq_queue is emptied in request-position order. Yet the same process,
> or group of processes, may generate I/O for different actuators. In
> this case, different streams of I/O (each for a different actuator)
> get all inserted into the same sync bfq_queue. So there is basically
> no individual control on when each stream is served, i.e., on when the
> I/O requests of the stream are picked from the bfq_queue and
> dispatched to the drive.
>
> This commit enables BFQ to control the service of each actuator
> individually for synchronous I/O, by simply splitting each sync
> bfq_queue into N queues, one for each actuator. In other words, a sync
> bfq_queue is now associated to a pair (process, actuator). As a
> consequence of this split, the per-queue proportional-share policy
> implemented by BFQ will guarantee that the sync I/O generated for each
> actuator, by each process, receives its fair share of service.
>
> This is just a preparatory patch. If the I/O of the same process
> happens to be sent to different queues, then each of these queues may
> undergo queue merging. To handle this event, the bfq_io_cq data
> structure must be properly extended. In addition, stable merging must
> be disabled to avoid loss of control on individual actuators. Finally,
> also async queues must be split. These issues are described in detail
> and addressed in next commits. As for this commit, although multiple
> per-process bfq_queues are provided, the I/O of each process or group
> of processes is still sent to only one queue, regardless of the
> actuator the I/O is for. The forwarding to distinct bfq_queues will be
> enabled after addressing the above issues.
>
> [1] https://www.linaro.org/blog/budget-fair-queueing-bfq-linux-io-scheduler-optimizations-for-multi-actuator-sata-hard-drives/
>
> Signed-off-by: Gabriele Felici <[email protected]>
> Signed-off-by: Carmine Zaccagnino <[email protected]>
> Signed-off-by: Paolo Valente <[email protected]>
One styles nit below.
Reviewed-by: Damien Le Moal <[email protected]>
> @@ -690,14 +700,25 @@ static void bfq_limit_depth(blk_opf_t opf, struct blk_mq_alloc_data *data)
> limit = (limit * depth) >> bfqd->full_depth_shift;
> }
>
> - /*
> - * Does queue (or any parent entity) exceed number of requests that
> - * should be available to it? Heavily limit depth so that it cannot
> - * consume more available requests and thus starve other entities.
> - */
> - if (bfqq && bfqq_request_over_limit(bfqq, limit))
> - depth = 1;
> + for (act_idx = 0; act_idx < bfqd->num_actuators; act_idx++) {
> + struct bfq_queue *bfqq;
> +
> + if (bic)
> + bfqq = bic_to_bfqq(bic, op_is_sync(opf), act_idx);
> + else
> + break;
>
> + /*
> + * Does queue (or any parent entity) exceed number of
> + * requests that should be available to it? Heavily
> + * limit depth so that it cannot consume more
> + * available requests and thus starve other entities.
> + */
> + if (bfqq && bfqq_request_over_limit(bfqq, limit)) {
> + depth = 1;
> + break;
> + }
You could reverse the if condition to make this cleaner, or even better,
include the bic test in the for loop:
for (act_idx = 0; bic && act_idx < bfqd->num_actuators; act_idx++) {
struct bfq_queue *bfqq;
/*
* Does queue (or any parent entity) exceed number of
* requests that should be available to it? Heavily
* limit depth so that it cannot consume more
* available requests and thus starve other entities.
*/
bfqq = bic_to_bfqq(bic, op_is_sync(opf), act_idx);
if (bfqq && bfqq_request_over_limit(bfqq, limit)) {
depth = 1;
break;
}
}
--
Damien Le Moal
Western Digital Research
> Il giorno 27 dic 2022, alle ore 02:37, Damien Le Moal <[email protected]> ha scritto:
>
> On 12/23/22 00:21, Paolo Valente wrote:
>> Single-LUN multi-actuator SCSI drives, as well as all multi-actuator
>> SATA drives appear as a single device to the I/O subsystem [1]. Yet
>> they address commands to different actuators internally, as a function
>> of Logical Block Addressing (LBAs). A given sector is reachable by
>> only one of the actuators. For example, Seagate’s Serial Advanced
>> Technology Attachment (SATA) version contains two actuators and maps
>> the lower half of the SATA LBA space to the lower actuator and the
>> upper half to the upper actuator.
>>
>> Evidently, to fully utilize actuators, no actuator must be left idle
>> or underutilized while there is pending I/O for it. The block layer
>> must somehow control the load of each actuator individually. This
>> commit lays the ground for allowing BFQ to provide such a per-actuator
>> control.
>>
>> BFQ associates an I/O-request sync bfq_queue with each process doing
>> synchronous I/O, or with a group of processes, in case of queue
>> merging. Then BFQ serves one bfq_queue at a time. While in service, a
>> bfq_queue is emptied in request-position order. Yet the same process,
>> or group of processes, may generate I/O for different actuators. In
>> this case, different streams of I/O (each for a different actuator)
>> get all inserted into the same sync bfq_queue. So there is basically
>> no individual control on when each stream is served, i.e., on when the
>> I/O requests of the stream are picked from the bfq_queue and
>> dispatched to the drive.
>>
>> This commit enables BFQ to control the service of each actuator
>> individually for synchronous I/O, by simply splitting each sync
>> bfq_queue into N queues, one for each actuator. In other words, a sync
>> bfq_queue is now associated to a pair (process, actuator). As a
>> consequence of this split, the per-queue proportional-share policy
>> implemented by BFQ will guarantee that the sync I/O generated for each
>> actuator, by each process, receives its fair share of service.
>>
>> This is just a preparatory patch. If the I/O of the same process
>> happens to be sent to different queues, then each of these queues may
>> undergo queue merging. To handle this event, the bfq_io_cq data
>> structure must be properly extended. In addition, stable merging must
>> be disabled to avoid loss of control on individual actuators. Finally,
>> also async queues must be split. These issues are described in detail
>> and addressed in next commits. As for this commit, although multiple
>> per-process bfq_queues are provided, the I/O of each process or group
>> of processes is still sent to only one queue, regardless of the
>> actuator the I/O is for. The forwarding to distinct bfq_queues will be
>> enabled after addressing the above issues.
>>
>> [1] https://www.linaro.org/blog/budget-fair-queueing-bfq-linux-io-scheduler-optimizations-for-multi-actuator-sata-hard-drives/
>>
>> Signed-off-by: Gabriele Felici <[email protected]>
>> Signed-off-by: Carmine Zaccagnino <[email protected]>
>> Signed-off-by: Paolo Valente <[email protected]>
>
> One styles nit below.
>
> Reviewed-by: Damien Le Moal <[email protected]>
>
>> @@ -690,14 +700,25 @@ static void bfq_limit_depth(blk_opf_t opf, struct blk_mq_alloc_data *data)
>> limit = (limit * depth) >> bfqd->full_depth_shift;
>> }
>>
>> - /*
>> - * Does queue (or any parent entity) exceed number of requests that
>> - * should be available to it? Heavily limit depth so that it cannot
>> - * consume more available requests and thus starve other entities.
>> - */
>> - if (bfqq && bfqq_request_over_limit(bfqq, limit))
>> - depth = 1;
>> + for (act_idx = 0; act_idx < bfqd->num_actuators; act_idx++) {
>> + struct bfq_queue *bfqq;
>> +
>> + if (bic)
>> + bfqq = bic_to_bfqq(bic, op_is_sync(opf), act_idx);
>> + else
>> + break;
>>
>> + /*
>> + * Does queue (or any parent entity) exceed number of
>> + * requests that should be available to it? Heavily
>> + * limit depth so that it cannot consume more
>> + * available requests and thus starve other entities.
>> + */
>> + if (bfqq && bfqq_request_over_limit(bfqq, limit)) {
>> + depth = 1;
>> + break;
>> + }
>
> You could reverse the if condition to make this cleaner, or even better,
> include the bic test in the for loop:
>
> for (act_idx = 0; bic && act_idx < bfqd->num_actuators; act_idx++) {
> struct bfq_queue *bfqq;
>
> /*
> * Does queue (or any parent entity) exceed number of
> * requests that should be available to it? Heavily
> * limit depth so that it cannot consume more
> * available requests and thus starve other entities.
> */
> bfqq = bic_to_bfqq(bic, op_is_sync(opf), act_idx);
> if (bfqq && bfqq_request_over_limit(bfqq, limit)) {
> depth = 1;
> break;
> }
> }
>
Done, thanks for this improvement.
Sending a V13, with all patches tagged as reviewed.
Thanks,
Paolo
>
> --
> Damien Le Moal
> Western Digital Research