Re: [PATCH v2 1/4] block, bfq: add support to track if root_group have any pending requests

On 2021/08/27 1:00, Paolo Valente wrote:

quoted

Il giorno 6 ago 2021, alle ore 04:08, Yu Kuai [off-list ref] ha scritto:

Add a new member in bfq_data to track number of queues that are in
root_group with any pending requests.

maybe modify the last part of the sentence as: ... and that have some pending request

quoted

This will be used in next patch
to optmize queue idle judgment when root_group doesn't have any
pending requests.

Signed-off-by: Yu Kuai <redacted>
---
block/bfq-iosched.c |  8 +++++++-
block/bfq-iosched.h | 13 +++++++++++--
block/bfq-wf2q.c    | 37 ++++++++++++++++++++++++-------------
3 files changed, 42 insertions(+), 16 deletions(-)

diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 727955918563..7c6b412f9a9c 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c

@@ -859,8 +859,14 @@ void __bfq_weights_tree_remove(struct bfq_data *bfqd,

void bfq_weights_tree_remove(struct bfq_data *bfqd,
			     struct bfq_queue *bfqq)
{
-	struct bfq_entity *entity = bfqq->entity.parent;
+	struct bfq_entity *entity = &bfqq->entity;
+
+	if (entity->in_groups_with_pending_reqs) {
+		entity->in_groups_with_pending_reqs = false;
+		bfqd->num_queues_with_pending_reqs_in_root--;

Here you cross the comment "The decrement of
num_groups_with_pending_reqs is not performed immediately upon ...".

Find a way to
- move that comment up, and to make it correct for this slightly
   different decrement
- leave a correct comment (probably shorter) in the original position

quoted

+	}

+	entity = entity->parent;
	for_each_entity(entity) {
		struct bfq_sched_data *sd = entity->my_sched_data;

diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h
index 99c2a3cb081e..610769214f72 100644
--- a/block/bfq-iosched.h
+++ b/block/bfq-iosched.h

@@ -195,7 +195,12 @@ struct bfq_entity {

	/* flag, set to request a weight, ioprio or ioprio_class change  */
	int prio_changed;

-	/* flag, set if the entity is counted in groups_with_pending_reqs */
+	/*
+	 * If entity represents bfq_group, this flag will set

will be set?

quoted

if the group is
+	 * not root_group and have

has

quoted

any pending requests; If entity represents

one more nit: use lowercase after semicolon

quoted

+	 * bfq_queue, this flag will set

will be set

quoted

if the queue is in root_group and have
+	 * any pending requests.
+	 */
	bool in_groups_with_pending_reqs;

The name of the above field follows from the fact that entity is
counted in groups_with_pending_reqs.  You change this fact, because,
in your patch, a queue is not counted in groups_with_pending_reqs.
But you leave the same name. This creates confusion.

quoted

	/* last child queue of entity created (for non-leaf entities) */

@@ -539,7 +544,11 @@ struct bfq_data {

	 * with no request waiting for completion.
	 */
	unsigned int num_groups_with_pending_reqs;
-
+	/*
+	 * number of queues that are in root_group with at least one request
+	 * waiting for completion.

please link somehow this comment to the long comment that comes before it

quoted

+	 */
+	unsigned int num_queues_with_pending_reqs_in_root;

Why using two counters?  I mean, couldn't you simply count also the
root group in num_groups_with_pending_reqs?

Hi, Paolo

Thanks for taking time reviewing these patches

I was doing this too complicated, while counting root group into
num_groups_with_pending_reqs is much easier. I'll do this in next
iteration.

Thanks
Yu Kuai

quoted

	/*
	 * Per-class (RT, BE, IDLE) number of bfq_queues containing
	 * requests (including the queue in service, even if it is

diff --git a/block/bfq-wf2q.c b/block/bfq-wf2q.c
index 7a462df71f68..188c8f907219 100644
--- a/block/bfq-wf2q.c
+++ b/block/bfq-wf2q.c

@@ -946,6 +946,29 @@ static void bfq_update_fin_time_enqueue(struct bfq_entity *entity,

	bfq_active_insert(st, entity);
}

+static void bfq_update_groups_with_pending_reqs(struct bfq_entity *entity)
+{
+#ifdef CONFIG_BFQ_GROUP_IOSCHED
+	struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
+

why do you introduce an extra variable bfqq, instead of doing as in
the original version of the code?  In addition, you remove the comment
/* bfq_group */

Thanks,
Paolo

quoted

+	if (bfqq) {
+		if (!entity->parent && !entity->in_groups_with_pending_reqs) {
+			entity->in_groups_with_pending_reqs = true;
+			bfqq->bfqd->num_queues_with_pending_reqs_in_root++;
+		}
+	} else {
+		if (!entity->in_groups_with_pending_reqs) {
+			struct bfq_group *bfqg =
+				container_of(entity, struct bfq_group, entity);
+			struct bfq_data *bfqd = bfqg->bfqd;
+
+			entity->in_groups_with_pending_reqs = true;
+			bfqd->num_groups_with_pending_reqs++;
+		}
+	}
+#endif
+}
+
/**
  * __bfq_activate_entity - handle activation of entity.
  * @entity: the entity being activated.

@@ -999,19 +1022,7 @@ static void __bfq_activate_entity(struct bfq_entity *entity,

		entity->on_st_or_in_serv = true;
	}

-#ifdef CONFIG_BFQ_GROUP_IOSCHED
-	if (!bfq_entity_to_bfqq(entity)) { /* bfq_group */
-		struct bfq_group *bfqg =
-			container_of(entity, struct bfq_group, entity);
-		struct bfq_data *bfqd = bfqg->bfqd;
-
-		if (!entity->in_groups_with_pending_reqs) {
-			entity->in_groups_with_pending_reqs = true;
-			bfqd->num_groups_with_pending_reqs++;
-		}
-	}
-#endif
-
+	bfq_update_groups_with_pending_reqs(entity);
	bfq_update_fin_time_enqueue(entity, st, backshifted);
}

-- 
2.31.1

.