blk-mq/scsi: Provide hostwide shared tags for SCSI HBAs
Subsystems (7)
v5 cover blk-mq/scsi: Provide hostwide shared tags for SCSI HBAs
Hi all,
here now is an updated version of the v2 patchset from John Garry, including the suggestions and reviews from the mailing list. John, apologies for hijacking your work :-)
For this version I've only added some slight modifications to Johns original patch (renaming variables etc); the contentious separate sbitmap allocation has been dropped in favour of Johns original version with pointers to the embedded sbitmap.
But more importantly I've reworked the scheduler tag allocation after discussions with Ming Lei.
Point is, hostwide shared tags can't really be resized; they surely cannot be increased (as it's a hardware limitation), and even decreasing is questionable as any modification here would affect all devices served by this HBA.
Scheduler tags, OTOH, can be considered as per-queue, as the I/O scheduler might want to look at all requests on all queues. As such the queue depth is distinct from the actual queue depth of the tagset.
Seeing that it is distinct the depth can now be changed independently of the underlying tagset, and there's no need ever to change the tagset itself.
I've also modified megaraid_sas, smartpqi and hpsa to take advantage of host_tags.
Performance for megaraid_sas is on par with the original implementation, with the added benefit that with this we should be able to handle cpu hotplug properly.
Differences to v4:
- Rework scheduler tag allocations
- Revert back to the original implementation from John
Differences to v3:
- Include reviews from Ming Lei
Differences to v2:
- Drop embedded tag bitmaps
- Do not share scheduling tags
- Add patches for hpsa and smartpqi
Differences to v1:
- Use a shared sbitmap, and not a separate shared tags (a big change!)
- Drop request.shared_tag
- Add RB tags
Hannes Reinecke (7):
blk-mq: rename blk_mq_update_tag_set_depth()
blk-mq: add WARN_ON in blk_mq_free_rqs()
blk-mq: move shared sbitmap into elevator queue
scsi: Add template flag 'host_tagset'
megaraid_sas: switch fusion adapters to MQ
smartpqi: enable host tagset
hpsa: enable host_tagset and switch to MQ
John Garry (3):
blk-mq: Remove some unused function arguments
blk-mq: Facilitate a shared sbitmap per tagset
scsi: hisi_sas: Switch v3 hw to MQ
Ming Lei (1):
blk-mq: rename BLK_MQ_F_TAG_SHARED as BLK_MQ_F_TAG_QUEUE_SHARED
block/bfq-iosched.c | 4 +- block/blk-mq-debugfs.c | 12 +-- block/blk-mq-sched.c | 22 +++++ block/blk-mq-tag.c | 140 +++++++++++++++++++++------- block/blk-mq-tag.h | 27 ++++-- block/blk-mq.c | 104 +++++++++++++++------ block/blk-mq.h | 7 +- block/blk-sysfs.c | 7 ++ block/kyber-iosched.c | 4 +- drivers/scsi/hisi_sas/hisi_sas.h | 3 +- drivers/scsi/hisi_sas/hisi_sas_main.c | 36 +++---- drivers/scsi/hisi_sas/hisi_sas_v3_hw.c | 86 +++++++---------- drivers/scsi/hpsa.c | 44 ++------- drivers/scsi/hpsa.h | 1 - drivers/scsi/megaraid/megaraid_sas.h | 1 - drivers/scsi/megaraid/megaraid_sas_base.c | 65 ++++--------- drivers/scsi/megaraid/megaraid_sas_fusion.c | 14 ++- drivers/scsi/scsi_lib.c | 2 + drivers/scsi/smartpqi/smartpqi_init.c | 38 ++++++-- include/linux/blk-mq.h | 7 +- include/linux/elevator.h | 3 + include/scsi/scsi_host.h | 3 + 22 files changed, 380 insertions(+), 250 deletions(-)
Signature · 1 line
2.16.4
PATCH v5 1/11 blk-mq: Remove some unused function arguments
From: John Garry <[email protected]>
The struct blk_mq_hw_ctx * argument in blk_mq_put_tag(), blk_mq_poll_nsecs(), and blk_mq_poll_hybrid_sleep() is unused, so remove it.
--- block/blk-mq-tag.c | 4 ++-- block/blk-mq-tag.h | 4 ++-- block/blk-mq.c | 10 ++++------ block/blk-mq.h | 2 +- 4 files changed, 9 insertions(+), 11 deletions(-)
block/blk-mq-tag.c +2 -2
@@ -191,8 +191,8 @@ unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data)return tag + tag_offset; } -void blk_mq_put_tag (struct blk_mq_hw_ctx *hctx,struct blk_mq_tags *tags, -struct blk_mq_ctx *ctx,unsigned int tag) +void blk_mq_put_tag (struct blk_mq_tags *tags,struct blk_mq_ctx *ctx, +unsigned int tag) {if (!blk_mq_tag_is_reserved (tags, tag)) {const int real_tag = tag - tags->nr_reserved_tags;
block/blk-mq-tag.h +2 -2
@@ -26,8 +26,8 @@ extern struct blk_mq_tags *blk_mq_init_tags(unsigned int nr_tags, unsigned int rextern void blk_mq_free_tags (struct blk_mq_tags *tags);extern unsigned int blk_mq_get_tag (struct blk_mq_alloc_data *data); -extern void blk_mq_put_tag (struct blk_mq_hw_ctx *hctx,struct blk_mq_tags *tags, -struct blk_mq_ctx *ctx,unsigned int tag); +extern void blk_mq_put_tag (struct blk_mq_tags *tags,struct blk_mq_ctx *ctx, +unsigned int tag);extern bool blk_mq_has_free_tags (struct blk_mq_tags *tags); extern int blk_mq_tag_update_depth(struct blk_mq_hw_ctx *hctx, struct blk_mq_tags **tags,
block/blk-mq.c +4 -6
@@ -499,9 +499,9 @@ static void __blk_mq_free_request(struct request *rq)blk_pm_mark_last_busy (rq); rq->mq_hctx =NULL ;if (rq->tag !=-1 ) -blk_mq_put_tag (hctx, hctx->tags, ctx, rq->tag); +blk_mq_put_tag (hctx->tags, ctx, rq->tag);if (sched_tag !=-1 ) -blk_mq_put_tag (hctx, hctx->sched_tags, ctx, sched_tag); +blk_mq_put_tag (hctx->sched_tags, ctx, sched_tag);blk_mq_sched_restart (hctx);blk_queue_exit (q); }
@@ -3354,7 +3354,6 @@ static void blk_mq_poll_stats_fn(struct blk_stat_callback *cb) }static unsigned long blk_mq_poll_nsecs (struct request_queue *q, -struct blk_mq_hw_ctx *hctx,struct request *rq) {unsigned long ret =0 ;
@@ -3387,7 +3386,6 @@ static unsigned long blk_mq_poll_nsecs(struct request_queue *q, }static bool blk_mq_poll_hybrid_sleep (struct request_queue *q, -struct blk_mq_hw_ctx *hctx,struct request *rq) {struct hrtimer_sleeper hs;
@@ -3407,7 +3405,7 @@ static bool blk_mq_poll_hybrid_sleep(struct request_queue *q,if (q->poll_nsec >0 ) nsecs = q->poll_nsec;else - nsecs =blk_mq_poll_nsecs (q, hctx, rq); + nsecs =blk_mq_poll_nsecs (q, rq);if (!nsecs)return false;
@@ -3462,7 +3460,7 @@ static bool blk_mq_poll_hybrid(struct request_queue *q,return false; } -return blk_mq_poll_hybrid_sleep (q, hctx, rq); +return blk_mq_poll_hybrid_sleep (q, rq); }/**
block/blk-mq.h +1 -1
@@ -208,7 +208,7 @@ static inline bool blk_mq_get_dispatch_budget(struct blk_mq_hw_ctx *hctx)static inline void __blk_mq_put_driver_tag (struct blk_mq_hw_ctx *hctx,struct request *rq) { -blk_mq_put_tag (hctx, hctx->tags, rq->mq_ctx, rq->tag); +blk_mq_put_tag (hctx->tags, rq->mq_ctx, rq->tag); rq->tag =-1 ;if (rq->rq_flags &RQF_MQ_INFLIGHT ) {
Signature · 1 line
2.16.4
PATCH v5 2/11 blk-mq: rename BLK_MQ_F_TAG_SHARED as BLK_MQ_F_TAG_QUEUE_SHARED
In reply to Hannes ReineckeFrom: Ming Lei <[email protected]>
BLK_MQ_F_TAG_SHARED actually means that tags is shared among request queues, all of which should belong to LUNs attached to same HBA.
So rename it to make the point explicitly.
--- block/blk-mq-debugfs.c | 2 +- block/blk-mq-tag.c | 2 +- block/blk-mq-tag.h | 4 ++-- block/blk-mq.c | 20 ++++++++++---------- include/linux/blk-mq.h | 2 +- 5 files changed, 15 insertions(+), 15 deletions(-)
block/blk-mq-debugfs.c +1 -1
@@ -236,7 +236,7 @@ static const char *const alloc_policy_name[] = {#define HCTX_FLAG_NAME (name) [ilog2(BLK_MQ_F_##name)] = #namestatic const char *const hctx_flag_name[] = {HCTX_FLAG_NAME (SHOULD_MERGE ), -HCTX_FLAG_NAME (TAG_SHARED ), +HCTX_FLAG_NAME (TAG_QUEUE_SHARED ),HCTX_FLAG_NAME (BLOCKING ),HCTX_FLAG_NAME (NO_SCHED ), };
block/blk-mq-tag.c +1 -1
@@ -73,7 +73,7 @@ static inline bool hctx_may_queue(struct blk_mq_hw_ctx *hctx, {unsigned int depth, users; -if (!hctx || !(hctx->flags &BLK_MQ_F_TAG_SHARED )) +if (!hctx || !(hctx->flags &BLK_MQ_F_TAG_QUEUE_SHARED ))return true;if (!test_bit (BLK_MQ_S_TAG_ACTIVE , &hctx->state))return true;
block/blk-mq-tag.h +2 -2
@@ -55,7 +55,7 @@ extern void __blk_mq_tag_idle(struct blk_mq_hw_ctx *);static inline bool blk_mq_tag_busy (struct blk_mq_hw_ctx *hctx) { -if (!(hctx->flags &BLK_MQ_F_TAG_SHARED )) +if (!(hctx->flags &BLK_MQ_F_TAG_QUEUE_SHARED ))return false;return __blk_mq_tag_busy (hctx);
@@ -63,7 +63,7 @@ static inline bool blk_mq_tag_busy(struct blk_mq_hw_ctx *hctx)static inline void blk_mq_tag_idle (struct blk_mq_hw_ctx *hctx) { -if (!(hctx->flags &BLK_MQ_F_TAG_SHARED )) +if (!(hctx->flags &BLK_MQ_F_TAG_QUEUE_SHARED ))return ;__blk_mq_tag_idle (hctx);
block/blk-mq.c +10 -10
@@ -302,7 +302,7 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data, rq->tag =-1 ; rq->internal_tag = tag; }else { -if (data->hctx->flags &BLK_MQ_F_TAG_SHARED ) { +if (data->hctx->flags &BLK_MQ_F_TAG_QUEUE_SHARED ) { rq_flags =RQF_MQ_INFLIGHT ;atomic_inc (&data->hctx->nr_active); }
@@ -1118,7 +1118,7 @@ static bool blk_mq_mark_tag_wait(struct blk_mq_hw_ctx *hctx,wait_queue_entry_t *wait;bool ret; -if (!(hctx->flags &BLK_MQ_F_TAG_SHARED )) { +if (!(hctx->flags &BLK_MQ_F_TAG_QUEUE_SHARED )) {blk_mq_sched_mark_restart_hctx (hctx);/*
@@ -1249,7 +1249,7 @@ bool blk_mq_dispatch_rq_list(struct request_queue *q, struct list_head *list,* For non-shared tags, the RESTART check * will suffice. */ -if (hctx->flags &BLK_MQ_F_TAG_SHARED ) +if (hctx->flags &BLK_MQ_F_TAG_QUEUE_SHARED ) no_tag = true;break ; }
@@ -2358,7 +2358,7 @@ blk_mq_alloc_hctx(struct request_queue *q, struct blk_mq_tag_set *set,spin_lock_init (&hctx->lock);INIT_LIST_HEAD (&hctx->dispatch); hctx->queue = q; - hctx->flags = set->flags & ~BLK_MQ_F_TAG_SHARED ; + hctx->flags = set->flags & ~BLK_MQ_F_TAG_QUEUE_SHARED ;INIT_LIST_HEAD (&hctx->hctx_list);
@@ -2575,9 +2575,9 @@ static void queue_set_hctx_shared(struct request_queue *q, bool shared)queue_for_each_hw_ctx (q, hctx, i) {if (shared) - hctx->flags |=BLK_MQ_F_TAG_SHARED ; + hctx->flags |=BLK_MQ_F_TAG_QUEUE_SHARED ;else - hctx->flags &= ~BLK_MQ_F_TAG_SHARED ; + hctx->flags &= ~BLK_MQ_F_TAG_QUEUE_SHARED ; } }
@@ -2603,7 +2603,7 @@ static void blk_mq_del_queue_tag_set(struct request_queue *q)list_del_rcu (&q->tag_set_list);if (list_is_singular (&set->tag_list)) {/* just transitioned to unshared */ - set->flags &= ~BLK_MQ_F_TAG_SHARED ; + set->flags &= ~BLK_MQ_F_TAG_QUEUE_SHARED ;/* update existing queue */ blk_mq_update_tag_set_depth (set, false); }
@@ -2620,12 +2620,12 @@ static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set,* Check to see if we're transitioning to shared (from 1 to 2 queues). */ if (!list_empty (&set->tag_list) && - !(set->flags &BLK_MQ_F_TAG_SHARED )) { - set->flags |=BLK_MQ_F_TAG_SHARED ; + !(set->flags &BLK_MQ_F_TAG_QUEUE_SHARED )) { + set->flags |=BLK_MQ_F_TAG_QUEUE_SHARED ;/* update existing queue */ blk_mq_update_tag_set_depth (set, true); } -if (set->flags &BLK_MQ_F_TAG_SHARED ) +if (set->flags &BLK_MQ_F_TAG_QUEUE_SHARED )queue_set_hctx_shared (q, true);list_add_tail_rcu (&q->tag_set_list, &set->tag_list);
include/linux/blk-mq.h +1 -1
@@ -225,7 +225,7 @@ struct blk_mq_ops {enum {BLK_MQ_F_SHOULD_MERGE =1 <<0 , -BLK_MQ_F_TAG_SHARED =1 <<1 , +BLK_MQ_F_TAG_QUEUE_SHARED =1 <<1 ,BLK_MQ_F_BLOCKING =1 <<5 ,BLK_MQ_F_NO_SCHED =1 <<6 ,BLK_MQ_F_ALLOC_POLICY_START_BIT =8 ,
Signature · 1 line
2.16.4
PATCH v5 3/11 blk-mq: rename blk_mq_update_tag_set_depth()
In reply to Hannes ReineckeThe function does not set the depth, but rather transitions from shared to non-shared queues and vice versa.
So rename it to blk_mq_update_tag_set_shared() to better reflect its purpose.
--- block/blk-mq-tag.c | 18 ++++++++++-------- block/blk-mq.c | 8 ++++---- 2 files changed, 14 insertions(+), 12 deletions(-)
block/blk-mq-tag.c +10 -8
@@ -440,24 +440,22 @@ static int bt_alloc(struct sbitmap_queue *bt, unsigned int depth, node); } -static struct blk_mq_tags *blk_mq_init_bitmap_tags (struct blk_mq_tags *tags, -int node,int alloc_policy) +static int blk_mq_init_bitmap_tags (struct blk_mq_tags *tags, +int node,int alloc_policy) {unsigned int depth = tags->nr_tags - tags->nr_reserved_tags;bool round_robin = alloc_policy ==BLK_TAG_ALLOC_RR ;if (bt_alloc (&tags->bitmap_tags, depth, round_robin, node)) - goto free_tags; +return -ENOMEM ;if (bt_alloc (&tags->breserved_tags, tags->nr_reserved_tags, round_robin, node)) goto free_bitmap_tags; -return tags; +return 0 ; free_bitmap_tags:sbitmap_queue_free (&tags->bitmap_tags); -free_tags: -kfree (tags); -return NULL ; +return -ENOMEM ; } struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags,
@@ -478,7 +476,11 @@ struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags, tags->nr_tags = total_tags; tags->nr_reserved_tags = reserved_tags; -return blk_mq_init_bitmap_tags (tags, node, alloc_policy); +if (blk_mq_init_bitmap_tags (tags, node, alloc_policy) <0 ) { +kfree (tags); + tags =NULL ; + } +return tags; }void blk_mq_free_tags (struct blk_mq_tags *tags)
block/blk-mq.c +4 -4
@@ -2581,8 +2581,8 @@ static void queue_set_hctx_shared(struct request_queue *q, bool shared) } } -static void blk_mq_update_tag_set_depth (struct blk_mq_tag_set *set, -bool shared) +static void blk_mq_update_tag_set_shared (struct blk_mq_tag_set *set, +bool shared) {struct request_queue *q;
@@ -2605,7 +2605,7 @@ static void blk_mq_del_queue_tag_set(struct request_queue *q)/* just transitioned to unshared */ set->flags &= ~BLK_MQ_F_TAG_QUEUE_SHARED ;/* update existing queue */ -blk_mq_update_tag_set_depth (set, false); +blk_mq_update_tag_set_shared (set, false); }mutex_unlock (&set->tag_list_lock);INIT_LIST_HEAD (&q->tag_set_list);
@@ -2623,7 +2623,7 @@ static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set, !(set->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) { set->flags |=BLK_MQ_F_TAG_QUEUE_SHARED ;/* update existing queue */ -blk_mq_update_tag_set_depth (set, true); +blk_mq_update_tag_set_shared (set, true); }if (set->flags &BLK_MQ_F_TAG_QUEUE_SHARED )queue_set_hctx_shared (q, true);
Signature · 1 line
2.16.4
Re: [PATCH 03/11] blk-mq: rename blk_mq_update_tag_set_depth()
Quoted 4 lines · Hannes Reinecke
The function does not set the depth, but rather transitions from shared to non-shared queues and vice versa.
So rename it toblk_mq_update_tag_set_shared()to better reflect its purpose.
We will probably need to split this patch into two, as we're doing more than just renaming.
The prep patches could be picked up earlier if we're lucky.
Thanks,
John
Quoted 88 lines
Signed-off-by Hannes Reinecke <[email protected]>--- block/blk-mq-tag.c | 18 ++++++++++-------- block/blk-mq.c | 8 ++++---- 2 files changed, 14 insertions(+), 12 deletions(-)block/blk-mq-tag.c +10
-8@@ -440,24 +440,22 @@ static int bt_alloc(struct sbitmap_queue *bt, unsigned int depth, node); } -static struct blk_mq_tags *blk_mq_init_bitmap_tags (struct blk_mq_tags *tags, -int node,int alloc_policy) +static int blk_mq_init_bitmap_tags (struct blk_mq_tags *tags, +int node,int alloc_policy) {unsigned int depth = tags->nr_tags - tags->nr_reserved_tags;bool round_robin = alloc_policy ==BLK_TAG_ALLOC_RR ;if (bt_alloc (&tags->bitmap_tags, depth, round_robin, node)) - goto free_tags; +return -ENOMEM ;if (bt_alloc (&tags->breserved_tags, tags->nr_reserved_tags, round_robin, node)) goto free_bitmap_tags; -return tags; +return 0 ; free_bitmap_tags:sbitmap_queue_free (&tags->bitmap_tags); -free_tags: -kfree (tags); -return NULL ; +return -ENOMEM ; } struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags,@@ -478,7 +476,11 @@ struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags, tags->nr_tags = total_tags; tags->nr_reserved_tags = reserved_tags; -return blk_mq_init_bitmap_tags (tags, node, alloc_policy); +if (blk_mq_init_bitmap_tags (tags, node, alloc_policy) <0 ) { +kfree (tags); + tags =NULL ; + } +return tags; }void blk_mq_free_tags (struct blk_mq_tags *tags)block/blk-mq.c +4
-4@@ -2581,8 +2581,8 @@ static void queue_set_hctx_shared(struct request_queue *q, bool shared) } } -static void blk_mq_update_tag_set_depth (struct blk_mq_tag_set *set, -bool shared) +static void blk_mq_update_tag_set_shared (struct blk_mq_tag_set *set, +bool shared) {struct request_queue *q;@@ -2605,7 +2605,7 @@ static void blk_mq_del_queue_tag_set(struct request_queue *q)/* just transitioned to unshared */ set->flags &= ~BLK_MQ_F_TAG_QUEUE_SHARED ;/* update existing queue */ -blk_mq_update_tag_set_depth (set, false); +blk_mq_update_tag_set_shared (set, false); }mutex_unlock (&set->tag_list_lock);INIT_LIST_HEAD (&q->tag_set_list);@@ -2623,7 +2623,7 @@ static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set, !(set->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) { set->flags |=BLK_MQ_F_TAG_QUEUE_SHARED ;/* update existing queue */ -blk_mq_update_tag_set_depth (set, true); +blk_mq_update_tag_set_shared (set, true); }if (set->flags &BLK_MQ_F_TAG_QUEUE_SHARED )queue_set_hctx_shared (q, true);
Re: [PATCH 03/11] blk-mq: rename blk_mq_update_tag_set_depth()
Quoted 93 lines · John Garry
Quoted 4 lines · Hannes Reinecke
The function does not set the depth, but rather transitions from shared to non-shared queues and vice versa.
So rename it toblk_mq_update_tag_set_shared()to better reflect its purpose.We will probably need to split this patch into two, as we're doing more than just renaming.
The prep patches could be picked up earlier if we're lucky.
Thanks,
JohnQuoted 79 lines
Signed-off-by Hannes Reinecke <[email protected]>---
block/blk-mq-tag.c| 18 ++++++++++--------
block/blk-mq.c | 8 ++++----2 files changed, 14 insertions(+), 12 deletions(-)block/blk-mq-tag.c +0
-0@@ -440,24 +440,22 @@ static int bt_alloc(struct sbitmap_queue *bt,unsigned int depth, node); }-static struct
blk_mq_tags*blk_mq_init_bitmap_tags(structblk_mq_tags*tags,
- int node, intalloc_policy) +static intblk_mq_init_bitmap_tags(struct blk_mq_tags *tags, + int node, int alloc_policy)
{
unsigned int depth =tags->nr_tags-tags->nr_reserved_tags; boolround_robin=alloc_policy==BLK_TAG_ALLOC_RR; if (bt_alloc(&tags->bitmap_tags, depth, round_robin, node)) - gotofree_tags;
+ return-ENOMEM;
if (bt_alloc(&tags->breserved_tags,tags->nr_reserved_tags,round_robin,
node))
gotofree_bitmap_tags;
- return tags;
+ return 0;
free_bitmap_tags:
sbitmap_queue_free(&tags->bitmap_tags); -free_tags:
- kfree(tags);
- return NULL;
+ return-ENOMEM;
}
structblk_mq_tags*blk_mq_init_tags(unsigned inttotal_tags,block/blk-mq-tag.c +5
-0@@ -478,7 +476,11 @@ struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags, tags->nr_tags = total_tags; tags->nr_reserved_tags = reserved_tags; - return blk_mq_init_bitmap_tags(tags, node, alloc_policy); + if (blk_mq_init_bitmap_tags(tags, node, alloc_policy) < 0) { + kfree(tags); + tags = NULL; + } + return tags; } void blk_mq_free_tags(struct blk_mq_tags *tags)block/blk-mq.c +2
-1@@ -2581,8 +2581,8 @@ static void queue_set_hctx_shared(struct request_queue *q, bool shared) } } -static void blk_mq_update_tag_set_depth(struct blk_mq_tag_set *set, - bool shared) +static void blk_mq_update_tag_set_shared (struct blk_mq_tag_set *set, + bool shared) { struct request_queue *q;@@ -2605,7 +2605,7 @@ static void
blk_mq_del_queue_tag_set(struct request_queue *q)
/* just transitioned to unshared */
set->flags&= ~BLK_MQ_F_TAG_QUEUE_SHARED;
/* update existing queue */
-blk_mq_update_tag_set_depth(set, false);
+blk_mq_update_tag_set_shared(set, false);
}
mutex_unlock(&set->tag_list_lock);INIT_LIST_HEAD(&q->tag_set_list);block/blk-mq.c +0
-0@@ -2623,7 +2623,7 @@ static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set, !(set->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) {
Quoted 9 lines
Quoted 8 lines
block/blk-mq.c +1
-1set->flags |= BLK_MQ_F_TAG_QUEUE_SHARED; /* update existing queue */ - blk_mq_update_tag_set_depth(set, true); + blk_mq_update_tag_set_shared(set, true); } if (set->flags & BLK_MQ_F_TAG_QUEUE_SHARED)
queue_set_hctx_shared(q, true);
Hmm. That shouldn't have happened; I'm sure I've had it in two patches originally. Oh well.
But I'd rather wait for feedback to my shared scheduler tag patch; guess that'll meet with some raise eyebrows ...
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Teamlead Storage & Networking [email protected] +49 911 74053 688 SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
PATCH v5 4/11 blk-mq: Facilitate a shared sbitmap per tagset
In reply to Hannes ReineckeFrom: John Garry <[email protected]>
Some SCSI HBAs (such as HPSA, megaraid, mpt3sas, hisi_sas_v3 ..) support multiple reply queues with single hostwide tags.
In addition, these drivers want to use interrupt assignment in pci_alloc_irq_vectors(PCI_IRQ_AFFINITY). However, as discussed in [0], CPU hotplug may cause in-flight IO completion to not be serviced when an interrupt is shutdown.
To solve that problem, Ming's patchset to drain hctx's should ensure no IOs are missed in-flight [1].
However, to take advantage of that patchset, we need to map the HBA HW queues to blk mq hctx's; to do that, we need to expose the HBA HW queues.
In making that transition, the per-SCSI command request tags are no longer unique per Scsi host - they are just unique per hctx. As such, the HBA LLDD would have to generate this tag internally, which has a certain performance overhead.
However another problem is that blk mq assumes the host may accept (Scsi_host.can_queue * #hw queue) commands. In [2], we removed the Scsi host busy counter, which would stop the LLDD being sent more than .can_queue commands; however, we should still ensure that the block layer does not issue more than .can_queue commands to the Scsi host.
To solve this problem, introduce a shared sbitmap per blk_mq_tag_set, which may be requested at init time.
New flag BLK_MQ_F_TAG_HCTX_SHARED should be set when requesting the tagset to indicate whether the shared sbitmap should be used.
Even when BLK_MQ_F_TAG_HCTX_SHARED is set, we still allocate a full set of tags and requests per hctx; the reason for this is that if we only allocate tags and requests for a single hctx - like hctx0 - we may break block drivers which expect a request be associated with a specific hctx, i.e. not hctx0.
This is based on work originally from Ming Lei in [3] and from Bart's suggestion in [4].
[0] https://lore.kernel.org/linux-block/[email protected]/
[1] https://lore.kernel.org/linux-block/[email protected]/
[2] https://lore.kernel.org/linux-scsi/[email protected]/
[3] https://lore.kernel.org/linux-block/[email protected]/
[4] https://lore.kernel.org/linux-block/[email protected]/T/#m3db0a602f095cbcbff27e9c884d6b4ae826144be
--- block/bfq-iosched.c | 4 +- block/blk-mq-debugfs.c | 10 ++--- block/blk-mq-sched.c | 14 ++++++ block/blk-mq-tag.c | 114 +++++++++++++++++++++++++++++++++++++++---------- block/blk-mq-tag.h | 17 ++++++-- block/blk-mq.c | 67 ++++++++++++++++++++++++++--- block/blk-mq.h | 5 +++ block/kyber-iosched.c | 4 +- include/linux/blk-mq.h | 9 ++++ 9 files changed, 204 insertions(+), 40 deletions(-)
block/bfq-iosched.c +2 -2
@@ -6327,8 +6327,8 @@ static void bfq_depth_updated(struct blk_mq_hw_ctx *hctx)struct blk_mq_tags *tags = hctx->sched_tags;unsigned int min_shallow; - min_shallow =bfq_update_depths (bfqd, &tags->bitmap_tags); -sbitmap_queue_min_shallow_depth (&tags->bitmap_tags, min_shallow); + min_shallow =bfq_update_depths (bfqd, tags->bitmap_tags); +sbitmap_queue_min_shallow_depth (tags->bitmap_tags, min_shallow); }static int bfq_init_hctx (struct blk_mq_hw_ctx *hctx,unsigned int index)
block/blk-mq-debugfs.c +5 -5
@@ -449,11 +449,11 @@ static void blk_mq_debugfs_tags_show(struct seq_file *m, atomic_read(&tags->active_queues));seq_puts (m,"\nbitmap_tags:\n" ); -sbitmap_queue_show (&tags->bitmap_tags, m); +sbitmap_queue_show (tags->bitmap_tags, m);if (tags->nr_reserved_tags) {seq_puts (m,"\nbreserved_tags:\n" ); -sbitmap_queue_show (&tags->breserved_tags, m); +sbitmap_queue_show (tags->breserved_tags, m); } }
@@ -483,8 +483,8 @@ static int hctx_tags_bitmap_show(void *data, struct seq_file *m) res =mutex_lock_interruptible (&q->sysfs_lock);if (res) goto out; -if (hctx->tags) -sbitmap_bitmap_show (&hctx->tags->bitmap_tags.sb, m); +if (hctx->tags)/* We should just iterate the relevant bits for this hctx FIXME */ +sbitmap_bitmap_show (&hctx->tags->bitmap_tags->sb, m);mutex_unlock (&q->sysfs_lock); out:
@@ -518,7 +518,7 @@ static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)if (res) goto out;if (hctx->sched_tags) -sbitmap_bitmap_show (&hctx->sched_tags->bitmap_tags.sb, m); +sbitmap_bitmap_show (&hctx->sched_tags->bitmap_tags->sb, m);mutex_unlock (&q->sysfs_lock); out:
block/blk-mq-sched.c +14 -0
@@ -492,6 +492,7 @@ static void blk_mq_sched_tags_teardown(struct request_queue *q)int blk_mq_init_sched (struct request_queue *q,struct elevator_type *e) { +struct blk_mq_tag_set *tag_set = q->tag_set;struct blk_mq_hw_ctx *hctx;struct elevator_queue *eq;unsigned int i;
@@ -537,6 +538,19 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)blk_mq_debugfs_register_sched_hctx (q, hctx); } +if (blk_mq_is_sbitmap_shared (tag_set)) { +if (!blk_mq_init_sched_shared_sbitmap (tag_set, q->nr_requests)) { + ret = -ENOMEM ; + goto err; + } +queue_for_each_hw_ctx (q, hctx, i) { +struct blk_mq_tags *tags = hctx->sched_tags; + + tags->bitmap_tags = &tag_set->__sched_bitmap_tags; + tags->breserved_tags = &tag_set->__sched_breserved_tags; + } + } +return 0 ; err:
block/blk-mq-tag.c +92 -22
200 lines not on this page · raw message
block/blk-mq-tag.h +14 -3
@@ -13,8 +13,11 @@ struct blk_mq_tags {atomic_t active_queues; -struct sbitmap_queue bitmap_tags; -struct sbitmap_queue breserved_tags; +struct sbitmap_queue *bitmap_tags; +struct sbitmap_queue *breserved_tags; + +struct sbitmap_queue __bitmap_tags; +struct sbitmap_queue __breserved_tags;struct request **rqs;struct request **static_rqs;
@@ -22,7 +25,15 @@ struct blk_mq_tags { }; -extern struct blk_mq_tags *blk_mq_init_tags (unsigned int nr_tags,unsigned int reserved_tags,int node,int alloc_policy); +extern bool blk_mq_init_shared_sbitmap (struct blk_mq_tag_set *tag_set); +extern void blk_mq_exit_shared_sbitmap (struct blk_mq_tag_set *tag_set); +extern bool blk_mq_init_sched_shared_sbitmap (struct blk_mq_tag_set *tag_set, +unsigned long nr_requests); +extern void blk_mq_exit_shared_sched_sbitmap (struct blk_mq_tag_set *tag_set); +extern struct blk_mq_tags *blk_mq_init_tags (struct blk_mq_tag_set *tag_set, +unsigned int nr_tags, +unsigned int reserved_tags, +int node,int alloc_policy);extern void blk_mq_free_tags (struct blk_mq_tags *tags);extern unsigned int blk_mq_get_tag (struct blk_mq_alloc_data *data);
block/blk-mq.c +61 -6
129 lines not on this page · raw message
block/blk-mq.h +5 -0
@@ -166,6 +166,11 @@ struct blk_mq_alloc_data {struct blk_mq_hw_ctx *hctx; }; +static inline bool blk_mq_is_sbitmap_shared (struct blk_mq_tag_set *tag_set) +{ +return !!(tag_set->flags &BLK_MQ_F_TAG_HCTX_SHARED ); +} +static inline struct blk_mq_tags *blk_mq_tags_from_data (struct blk_mq_alloc_data *data) {if (data->flags &BLK_MQ_REQ_INTERNAL )
block/kyber-iosched.c +2 -2
@@ -359,7 +359,7 @@ static unsigned int kyber_sched_tags_shift(struct request_queue *q)* All of the hardware queues have the same depth, so we can just grab * the shift of the first one. */ -return q->queue_hw_ctx[0 ]->sched_tags->bitmap_tags.sb.shift; +return q->queue_hw_ctx[0 ]->sched_tags->bitmap_tags->sb.shift; }static struct kyber_queue_data *kyber_queue_data_alloc (struct request_queue *q)
@@ -502,7 +502,7 @@ static int kyber_init_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) khd->batching =0 ; hctx->sched_data = khd; -sbitmap_queue_min_shallow_depth (&hctx->sched_tags->bitmap_tags, +sbitmap_queue_min_shallow_depth (hctx->sched_tags->bitmap_tags, kqd->async_depth);return 0 ;
include/linux/blk-mq.h +9 -0
@@ -109,6 +109,12 @@ struct blk_mq_tag_set {unsigned int flags;/* BLK_MQ_F_* */ void *driver_data; +struct sbitmap_queue __bitmap_tags; +struct sbitmap_queue __breserved_tags; + +struct sbitmap_queue __sched_bitmap_tags; +struct sbitmap_queue __sched_breserved_tags; +struct blk_mq_tags **tags;struct mutex tag_list_lock;
@@ -226,6 +232,9 @@ struct blk_mq_ops {enum {BLK_MQ_F_SHOULD_MERGE =1 <<0 ,BLK_MQ_F_TAG_QUEUE_SHARED =1 <<1 , +BLK_MQ_F_TAG_HCTX_SHARED =1 <<2 , +BLK_MQ_F_TAG_BITMAP_ALLOCATED =1 <<3 , +BLK_MQ_F_TAG_SCHED_BITMAP_ALLOCATED =1 <<4 ,BLK_MQ_F_BLOCKING =1 <<5 ,BLK_MQ_F_NO_SCHED =1 <<6 ,BLK_MQ_F_ALLOC_POLICY_START_BIT =8 ,
Signature · 1 line
2.16.4
Re: [PATCH 04/11] blk-mq: Facilitate a shared sbitmap per tagset
block/blk-mq-debugfs.c static int hctx_tags_bitmap_show(void *data, struct seq_file *m) · line 486
Quoted 8 lines
block/blk-mq-debugfs.c +1
-2@@ -483,8 +483,8 @@ static int hctx_tags_bitmap_show(void *data, struct seq_file *m) res =mutex_lock_interruptible (&q->sysfs_lock);if (res) goto out; -if (hctx->tags) -sbitmap_bitmap_show (&hctx->tags->bitmap_tags.sb, m); +if (hctx->tags)/* We should just iterate the relevant bits for this hctx FIXME */
Bart's solution to this problem seemed ok, if he doesn't mind us borrowing his idea:
https://lore.kernel.org/linux-block/[email protected]/T/#m24394fe70b1ea79a154dfd9620f5e553c3e7e7da
See hctx_tags_bitmap_show().
It might be also reasonable to put that in another follow on patch, as there would be no enablers of the "shared" bitmap until later patches.
Quoted 71 lines
block/blk-mq-debugfs.c +2
-1+sbitmap_bitmap_show (&hctx->tags->bitmap_tags->sb, m);mutex_unlock (&q->sysfs_lock); out:@@ -518,7 +518,7 @@ static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)if (res) goto out;if (hctx->sched_tags) -sbitmap_bitmap_show (&hctx->sched_tags->bitmap_tags.sb, m); +sbitmap_bitmap_show (&hctx->sched_tags->bitmap_tags->sb, m);mutex_unlock (&q->sysfs_lock); out:block/blk-mq-sched.c +14
-0@@ -492,6 +492,7 @@ static void blk_mq_sched_tags_teardown(struct request_queue *q)int blk_mq_init_sched (struct request_queue *q,struct elevator_type *e) { +struct blk_mq_tag_set *tag_set = q->tag_set;struct blk_mq_hw_ctx *hctx;struct elevator_queue *eq;unsigned int i;@@ -537,6 +538,19 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)blk_mq_debugfs_register_sched_hctx (q, hctx); } +if (blk_mq_is_sbitmap_shared (tag_set)) { +if (!blk_mq_init_sched_shared_sbitmap (tag_set, q->nr_requests)) { + ret = -ENOMEM ; + goto err; + } +queue_for_each_hw_ctx (q, hctx, i) { +struct blk_mq_tags *tags = hctx->sched_tags; + + tags->bitmap_tags = &tag_set->__sched_bitmap_tags; + tags->breserved_tags = &tag_set->__sched_breserved_tags; + } + } +return 0 ; err:block/blk-mq-tag.c +3
-3@@ -20,7 +20,7 @@ bool blk_mq_has_free_tags(struct blk_mq_tags *tags)if (!tags)return true; -return sbitmap_any_bit_clear (&tags->bitmap_tags.sb); +return sbitmap_any_bit_clear (&tags->bitmap_tags->sb); }/* @@ -43,9 +43,9 @@ bool __blk_mq_tag_busy(struct blk_mq_hw_ctx *hctx)*/ void blk_mq_tag_wakeup_all (struct blk_mq_tags *tags,bool include_reserve) { -sbitmap_queue_wake_all (&tags->bitmap_tags); +sbitmap_queue_wake_all (tags->bitmap_tags);if (include_reserve) -sbitmap_queue_wake_all (&tags->breserved_tags); +sbitmap_queue_wake_all (tags->breserved_tags); }/*
block/blk-mq-tag.c unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data) · line 124
Quoted 5 lines
block/blk-mq-tag.c +1
-1@@ -121,10 +121,10 @@ unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data)WARN_ON_ONCE (1 );return BLK_MQ_TAG_FAIL ; } - bt = &tags->breserved_tags; + bt = tags->breserved_tags;
We could put all of this in an earlier patch (as you had in v4, modulo dynamic memory part), which would be easier to review and get accepted.
Quoted 6 lines
block/blk-mq-tag.c +1
-1tag_offset =0 ; }else { - bt = &tags->bitmap_tags; + bt = tags->bitmap_tags; tag_offset = tags->nr_reserved_tags; }
[...]
Quoted 10 lines
if (!set)
block/blk-mq-tag.c +1
-0@@ -3160,6 +3179,7 @@ int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr) ret =blk_mq_tag_update_depth (hctx, &hctx->tags, nr, false); }else { + sched_tags = true; ret =blk_mq_tag_update_depth (hctx, &hctx->sched_tags, nr, true); }
block/blk-mq-tag.c int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr) · line 3196
Quoted 9 lines
block/blk-mq-tag.c +5
-1@@ -3169,8 +3189,43 @@ int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr) q->elevator->type->ops.depth_updated (hctx); } -if (!ret) +/* +* if ret is 0, all queues should have been updated to the same depth +* if not, then maybe some have been updated - yuk, need to handle this for shared sbitmap... +* if some are updated, we should probably roll back the change altogether. FIXME +*/
If you don't have a shared sched bitmap - which I didn't think we needed - then all we need is a simple sbitmap_queue_resize(&tagset->__bitmap_tags)
Otherwise it's horrible to resize shared sched bitmaps...
Quoted 40 lines
block/blk-mq-tag.c +31
-0+if (!ret) { +if (blk_mq_is_sbitmap_shared (set)) { +if (sched_tags) { +blk_mq_exit_shared_sched_sbitmap (set); +if (!blk_mq_init_sched_shared_sbitmap (set, nr)) +return -ENOMEM ;/* fixup error handling */ + +queue_for_each_hw_ctx (q, hctx, i) { + hctx->sched_tags->bitmap_tags = + &set->__sched_bitmap_tags; + hctx->sched_tags->breserved_tags = + &set->__sched_breserved_tags; + } + }else { +blk_mq_exit_shared_sbitmap (set); +if (!blk_mq_init_shared_sbitmap (set)) +return -ENOMEM ;/* fixup error handling */ + +queue_for_each_hw_ctx (q, hctx, i) { + hctx->tags->bitmap_tags = + &set->__bitmap_tags; + hctx->tags->breserved_tags = + &set->__breserved_tags; + } + } + } q->nr_requests = nr; + } +/* +* if ret != 0, q->nr_requests would not be updated, yet the depth +* for some hctx may have changed - is that right? +*/ blk_mq_unquiesce_queue (q);blk_mq_unfreeze_queue (q);block/blk-mq.h +0
-0
block/blk-mq.h struct blk_mq_alloc_data { · line 171
Quoted 6 lines
block/blk-mq.h +3
-0@@ -166,6 +166,11 @@ struct blk_mq_alloc_data {struct blk_mq_hw_ctx *hctx; }; +static inline bool blk_mq_is_sbitmap_shared (struct blk_mq_tag_set *tag_set) +{ +return !!(tag_set->flags &BLK_MQ_F_TAG_HCTX_SHARED );
Bart already gave some comments on this
Quoted 55 lines
block/blk-mq.h +2
-0+} +static inline struct blk_mq_tags *blk_mq_tags_from_data (struct blk_mq_alloc_data *data) {if (data->flags &BLK_MQ_REQ_INTERNAL )block/kyber-iosched.c +2
-2@@ -359,7 +359,7 @@ static unsigned int kyber_sched_tags_shift(struct request_queue *q)* All of the hardware queues have the same depth, so we can just grab * the shift of the first one. */ -return q->queue_hw_ctx[0 ]->sched_tags->bitmap_tags.sb.shift; +return q->queue_hw_ctx[0 ]->sched_tags->bitmap_tags->sb.shift; }static struct kyber_queue_data *kyber_queue_data_alloc (struct request_queue *q)@@ -502,7 +502,7 @@ static int kyber_init_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) khd->batching =0 ; hctx->sched_data = khd; -sbitmap_queue_min_shallow_depth (&hctx->sched_tags->bitmap_tags, +sbitmap_queue_min_shallow_depth (hctx->sched_tags->bitmap_tags, kqd->async_depth);return 0 ;include/linux/blk-mq.h +9
-0@@ -109,6 +109,12 @@ struct blk_mq_tag_set {unsigned int flags;/* BLK_MQ_F_* */ void *driver_data; +struct sbitmap_queue __bitmap_tags; +struct sbitmap_queue __breserved_tags; + +struct sbitmap_queue __sched_bitmap_tags; +struct sbitmap_queue __sched_breserved_tags; +struct blk_mq_tags **tags;struct mutex tag_list_lock;@@ -226,6 +232,9 @@ struct blk_mq_ops {enum {BLK_MQ_F_SHOULD_MERGE =1 <<0 ,BLK_MQ_F_TAG_QUEUE_SHARED =1 <<1 , +BLK_MQ_F_TAG_HCTX_SHARED =1 <<2 , +BLK_MQ_F_TAG_BITMAP_ALLOCATED =1 <<3 , +BLK_MQ_F_TAG_SCHED_BITMAP_ALLOCATED =1 <<4 ,BLK_MQ_F_BLOCKING =1 <<5 ,BLK_MQ_F_NO_SCHED =1 <<6 ,BLK_MQ_F_ALLOC_POLICY_START_BIT =8 ,
Re: [PATCH 04/11] blk-mq: Facilitate a shared sbitmap per tagset
Quoted 13 lines · John Garry
Quoted 9 lines
@@ -483,8 +483,8 @@ static int
hctx_tags_bitmap_show(void *data, struct seq_file *m)
res =mutex_lock_interruptible(&q->sysfs_lock);
if (res)
goto out;
- if (hctx->tags)
-sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m); + if (hctx->tags) /* We should just iterate the relevant bits for this hctx FIXME */Bart's solution to this problem seemed ok, if he doesn't mind us borrowing his idea:
Quoted 8 lines
https://lore.kernel.org/linux-block/[email protected]/T/#m24394fe70b1ea79a154dfd9620f5e553c3e7e7da
See
hctx_tags_bitmap_show().It might be also reasonable to put that in another follow on patch, as there would be no enablers of the "shared" bitmap until later patches.
Yeah, that was my plan, too.
But then I'd rather wait for feedback on the general approach here; no point is wasting perfectly good bits if no-one's wanting them ...
[ .. ]
Quoted 11 lines
block/blk-mq-tag.c unsigned int blk_mq_get_tag(struct · line 125
Quoted 7 lines
block/blk-mq-tag.c +1
-1@@ -121,10 +121,10 @@ unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data) WARN_ON_ONCE(1); return BLK_MQ_TAG_FAIL; } - bt = &tags->breserved_tags; + bt = tags->breserved_tags;We could put all of this in an earlier patch (as you had in v4, modulo dynamic memory part), which would be easier to review and get accepted.
Yeah, but I felt it a bit odd, just having pointers to an existing structure element.
But yes, will be doing for the next round.
Quoted 37 lines
Quoted 6 lines
block/blk-mq-tag.c +1
-1tag_offset = 0; } else { - bt = &tags->bitmap_tags; + bt = tags->bitmap_tags; tag_offset = tags->nr_reserved_tags; }[...]
Quoted 17 lines
if (!set)
block/blk-mq.c +1
-0@@ -3160,6 +3179,7 @@ int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr) ret = blk_mq_tag_update_depth(hctx, &hctx->tags, nr, false); } else { + sched_tags = true; ret = blk_mq_tag_update_depth(hctx, &hctx->sched_tags, nr, true);}
block/blk-mq.c +2
-0@@ -3169,8 +3189,43 @@ int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr) q->elevator->type->ops.depth_updated(hctx); } - if (!ret) + /* +* if ret is 0, all queues should have been updated to the same Quoted 6 lines
depth
+ * if not, then maybe some have been updated - yuk, need to handle this for shared sbitmap...
+ * if some are updated, we should probably roll back the change altogether. FIXME
+ */If you don't have a shared sched bitmap - which I didn't think we needed
- then all we need is a simplesbitmap_queue_resize(&tagset->__bitmap_tags)
Quoted 2 lines
Otherwise it's horrible to resize shared sched bitmaps...
Resizing shared sched bitmaps is done in patch 6/11.
General idea is to move the scheduler bitmap into the request queue (well, actually the elevator), as this gives us a per-request_queue bitmap. Which is actually what we want here, as the scheduler will need to look at all requests, hence needing to access to the same bitmap. And it also gives us an easy way of resizing the sched tag bitmap, as then we can resize the bitmap on a per-queue basis, and leave the underlying tagset bitmap untouched.
[ .. ]
Quoted 5 lines
Quoted 5 lines
block/blk-mq.h +0
-0@@ -166,6 +166,11 @@ struct blk_mq_alloc_data {
Quoted 9 lines
Quoted 6 lines
block/blk-mq.h +0
-0struct blk_mq_hw_ctx *hctx; };+static inline bool
blk_mq_is_sbitmap_shared(struct blk_mq_tag_set *tag_set)
+{
+ return !!(tag_set->flags &BLK_MQ_F_TAG_HCTX_SHARED);Bart already gave some comments on this
Ah. Missed that one. Will be including in the next round.
Thanks for the review!
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Teamlead Storage & Networking [email protected] +49 911 74053 688 SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
Re: [PATCH 04/11] blk-mq: Facilitate a shared sbitmap per tagset
Quoted 6 lines · Hannes Reinecke
Quoted 6 lines
Quoted 1 line
+ */
If you don't have a shared sched bitmap - which I didn't think we needed
- then all we need is a simplesbitmap_queue_resize(&tagset->__bitmap_tags)Otherwise it's horrible to resize shared sched bitmaps...
Quoted 8 lines
Resizing shared sched bitmaps is done in patch 6/11.
General idea is to move the scheduler bitmap into the request queue (well, actually the elevator), as this gives us a per-request_queue bitmap. Which is actually what we want here, as the scheduler will need to look at all requests, hence needing to access to the same bitmap. And it also gives us an easy way of resizing the sched tag bitmap, as then we can resize the bitmap on a per-queue basis, and leave the underlying tagset bitmap untouched.
OK, but I am just concerned if that is really required in this series and whether it is just another obstacle to getting it accepted.
Thanks,
John
Quoted 6 lines
[ .. ]
Quoted 4 lines
Quoted 4 lines
block/blk-mq.h +0
-0
Re: [PATCH 04/11] blk-mq: Facilitate a shared sbitmap per tagset
In reply to John GarryQuoted 12 lines · John Garry
Quoted 9 lines
block/blk-mq-debugfs.c +1
-2@@ -483,8 +483,8 @@ static int hctx_tags_bitmap_show(void *data, struct seq_file *m) res = mutex_lock_interruptible(&q->sysfs_lock); if (res) goto out; - if (hctx->tags) - sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m); + if (hctx->tags) /* We should just iterate the relevant bits forthis hctx FIXME */
Bart's solution to this problem seemed ok, if he doesn't mind us borrowing his idea:
Quoted 3 lines
https://lore.kernel.org/linux-block/[email protected]/T/#m24394fe70b1ea79a154dfd9620f5e553c3e7e7da
See
hctx_tags_bitmap_show().
Hi John,
Sure, borrowing that code is fine with me.
Bart.
PATCH v5 5/11 blk-mq: add WARN_ON in blk_mq_free_rqs()
In reply to Hannes ReineckeBefore freeing up the requests we should ensure that none of those requests are still present in the ->rqs array; this could lead to an use-after free error.
--- block/blk-mq.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
block/blk-mq.c +7 -3
@@ -2049,10 +2049,14 @@ void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags, unsigned int hctx_idx) {struct page *page; +int i; -if (tags->rqs && set->ops->exit_request) { -int i; - +if (tags->rqs) { +for (i =0 ; i < tags->nr_tags; i++) +if (WARN_ON (tags->rqs[i])) + tags->rqs[i] =NULL ; + } +if (tags->static_rqs && set->ops->exit_request) {for (i =0 ; i < tags->nr_tags; i++) {struct request *rq = tags->static_rqs[i];
Signature · 1 line
2.16.4
PATCH v5 6/11 blk-mq: move shared sbitmap into elevator queue
In reply to Hannes ReineckeWhen an elevator is present the sbitmap is actually per request queue, so modifying the tagset queue depth in the shared sbitmap case will be doing the wrong thing as it'll change the queue depth for all request queues.
So this patch moves the shared scheduler sbitmap into struct elevator queue, thereby insulating any modifications to this request queue only.
And with that we can increase the number of requests for the shared sbitmap case to the queue depth times the number of queues, as now all tags are allocated from the same bitmap.
This also solves the problem of sbitmap resizing in the shared sbitmap case; we can simply require an elevator to be present if the queue depth needs to be modified, as then the queue depth will be modified for this particular request queue only.
--- block/blk-mq-sched.c | 14 +++++++++++--- block/blk-mq-tag.c | 30 ++++++++++++++++-------------- block/blk-mq-tag.h | 8 +++++--- block/blk-mq.c | 41 +++++++++++++++-------------------------- block/blk-sysfs.c | 7 +++++++ include/linux/blk-mq.h | 4 ---- include/linux/elevator.h | 3 +++ 7 files changed, 57 insertions(+), 50 deletions(-)
block/blk-mq-sched.c +11 -3
@@ -511,6 +511,12 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)*/ q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth,BLKDEV_MAX_RQ ); +/* +* For the shared sbitmap case it's per request queue, so multiply +* with the number of hw queues +*/ +if (blk_mq_is_sbitmap_shared (tag_set)) + q->nr_requests *= q->nr_hw_queues;queue_for_each_hw_ctx (q, hctx, i) { ret =blk_mq_sched_alloc_tags (q, hctx, i);
@@ -539,15 +545,16 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) }if (blk_mq_is_sbitmap_shared (tag_set)) { -if (!blk_mq_init_sched_shared_sbitmap (tag_set, q->nr_requests)) { + eq = q->elevator; +if (!blk_mq_init_elevator_sbitmap (q, eq, q->nr_requests)) { ret = -ENOMEM ; goto err; }queue_for_each_hw_ctx (q, hctx, i) {struct blk_mq_tags *tags = hctx->sched_tags; - tags->bitmap_tags = &tag_set->__sched_bitmap_tags; - tags->breserved_tags = &tag_set->__sched_breserved_tags; + tags->bitmap_tags = &eq->__bitmap_tags; + tags->breserved_tags = &eq->__breserved_tags; } }
@@ -591,5 +598,6 @@ void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e)if (e->type->ops.exit_sched) e->type->ops.exit_sched (e);blk_mq_sched_tags_teardown (q); +blk_mq_exit_elevator_sbitmap (q, e); q->elevator =NULL ; }
block/blk-mq-tag.c +16 -14
@@ -491,35 +491,37 @@ void blk_mq_exit_shared_sbitmap(struct blk_mq_tag_set *tag_set) } } -bool blk_mq_init_sched_shared_sbitmap (struct blk_mq_tag_set *tag_set, -unsigned long nr_requests) +bool blk_mq_init_elevator_sbitmap (struct request_queue *q, +struct elevator_queue *eq, +unsigned int nr_requests) { -unsigned int depth = nr_requests -tag_set->reserved_tags; +struct blk_mq_tag_set *tag_set = q->tag_set; +unsigned int depth = nr_requests - tag_set->reserved_tags;int alloc_policy =BLK_MQ_FLAG_TO_ALLOC_POLICY (tag_set->flags);bool round_robin = alloc_policy ==BLK_TAG_ALLOC_RR ;int node = tag_set->numa_node; -if (tag_set->flags &BLK_MQ_F_TAG_SCHED_BITMAP_ALLOCATED ) -return false; -if (bt_alloc (&tag_set->__sched_bitmap_tags, depth, round_robin, node)) +if (!blk_mq_is_sbitmap_shared (q->tag_set)) +return true; + +if (bt_alloc (&eq->__bitmap_tags, depth, round_robin, node))return false; -if (bt_alloc (&tag_set->__sched_breserved_tags, tag_set->reserved_tags, +if (bt_alloc (&eq->__breserved_tags, tag_set->reserved_tags, round_robin, node)) goto free_bitmap_tags; - tag_set->flags |=BLK_MQ_F_TAG_SCHED_BITMAP_ALLOCATED ;return true; free_bitmap_tags: -sbitmap_queue_free (&tag_set->__sched_bitmap_tags); +sbitmap_queue_free (&eq->__bitmap_tags);return false; } -void blk_mq_exit_shared_sched_sbitmap (struct blk_mq_tag_set *tag_set) +void blk_mq_exit_elevator_sbitmap (struct request_queue *q, +struct elevator_queue *eq) { -if (tag_set->flags &BLK_MQ_F_TAG_SCHED_BITMAP_ALLOCATED ) { -sbitmap_queue_free (&tag_set->__sched_bitmap_tags); -sbitmap_queue_free (&tag_set->__sched_breserved_tags); - tag_set->flags &= ~BLK_MQ_F_TAG_SCHED_BITMAP_ALLOCATED ; +if (blk_mq_is_sbitmap_shared (q->tag_set)) { +sbitmap_queue_free (&eq->__bitmap_tags); +sbitmap_queue_free (&eq->__breserved_tags); } }
block/blk-mq-tag.h +5 -3
@@ -27,9 +27,11 @@ struct blk_mq_tags {extern bool blk_mq_init_shared_sbitmap (struct blk_mq_tag_set *tag_set);extern void blk_mq_exit_shared_sbitmap (struct blk_mq_tag_set *tag_set); -extern bool blk_mq_init_sched_shared_sbitmap (struct blk_mq_tag_set *tag_set, -unsigned long nr_requests); -extern void blk_mq_exit_shared_sched_sbitmap (struct blk_mq_tag_set *tag_set); +extern bool blk_mq_init_elevator_sbitmap (struct request_queue *q, +struct elevator_queue *eq, +unsigned int nr_tags); +extern void blk_mq_exit_elevator_sbitmap (struct request_queue *q, +struct elevator_queue *eq); extern struct blk_mq_tags *blk_mq_init_tags(struct blk_mq_tag_set *tag_set, unsigned int nr_tags, unsigned int reserved_tags,
block/blk-mq.c +15 -26
70 lines not on this page · raw message
block/blk-sysfs.c +7 -0
@@ -78,6 +78,13 @@ queue_requests_store(struct request_queue *q, const char *page, size_t count)if (nr <BLKDEV_MIN_RQ ) nr =BLKDEV_MIN_RQ ; +/* +* We can only modify the queue depth for shared sbitmaps +* if an I/O scheduler is set. +*/ +if (blk_mq_is_sbitmap_shared (q->tag_set) && !q->elevator) +return -EINVAL ; + err =blk_mq_update_nr_requests (q, nr);if (err)return err;
include/linux/blk-mq.h +0 -4
@@ -112,9 +112,6 @@ struct blk_mq_tag_set {struct sbitmap_queue __bitmap_tags;struct sbitmap_queue __breserved_tags; -struct sbitmap_queue __sched_bitmap_tags; -struct sbitmap_queue __sched_breserved_tags; -struct blk_mq_tags **tags;struct mutex tag_list_lock;
@@ -234,7 +231,6 @@ enum {BLK_MQ_F_TAG_QUEUE_SHARED =1 <<1 ,BLK_MQ_F_TAG_HCTX_SHARED =1 <<2 ,BLK_MQ_F_TAG_BITMAP_ALLOCATED =1 <<3 , -BLK_MQ_F_TAG_SCHED_BITMAP_ALLOCATED =1 <<4 ,BLK_MQ_F_BLOCKING =1 <<5 ,BLK_MQ_F_NO_SCHED =1 <<6 , BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
include/linux/elevator.h +3 -0
@@ -4,6 +4,7 @@#include <linux/percpu.h> #include <linux/hashtable.h> +#include <linux/sbitmap.h> #ifdef CONFIG_BLOCK
@@ -104,6 +105,8 @@ struct elevator_queuevoid *elevator_data;struct kobject kobj;struct mutex sysfs_lock; +struct sbitmap_queue __bitmap_tags; +struct sbitmap_queue __breserved_tags; unsigned int registered:1;DECLARE_HASHTABLE (hash,ELV_HASH_BITS ); };
Signature · 1 line
2.16.4
PATCH v5 7/11 scsi: Add template flag 'host_tagset'
In reply to Hannes ReineckeFrom: Hannes Reinecke <[email protected]>
Add a host template flag 'host_tagset' so hostwide tagset can be shared on multiple reply queues after the SCSI device's reply queue is converted to blk-mq hw queue.
--- drivers/scsi/scsi_lib.c | 2 ++ include/scsi/scsi_host.h | 3 +++ 2 files changed, 5 insertions(+)
drivers/scsi/scsi_lib.c +2 -0
@@ -1899,6 +1899,8 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) shost->tag_set.flags =BLK_MQ_F_SHOULD_MERGE ; shost->tag_set.flags |=BLK_ALLOC_POLICY_TO_MQ_FLAG (shost->hostt->tag_alloc_policy); +if (shost->hostt->host_tagset) + shost->tag_set.flags |=BLK_MQ_F_TAG_HCTX_SHARED ; shost->tag_set.driver_data = shost;return blk_mq_alloc_tag_set (&shost->tag_set);
include/scsi/scsi_host.h +3 -0
@@ -429,6 +429,9 @@ struct scsi_host_template {/* True if the low-level driver supports blk-mq only */ unsigned force_blk_mq:1; +/* True if the host uses host-wide tagspace */ + unsigned host_tagset:1; +/* * Countdown for host blocking with no commands outstanding. */
Signature · 1 line
2.16.4
PATCH v5 8/11 scsi: hisi_sas: Switch v3 hw to MQ
In reply to Hannes ReineckeFrom: John Garry <[email protected]>
Now that the block layer provides a shared tag, we can switch the driver to expose all HW queues.
--- drivers/scsi/hisi_sas/hisi_sas.h | 3 +- drivers/scsi/hisi_sas/hisi_sas_main.c | 36 +++++++------- drivers/scsi/hisi_sas/hisi_sas_v3_hw.c | 86 ++++++++++++++-------------------- 3 files changed, 56 insertions(+), 69 deletions(-)
drivers/scsi/hisi_sas/hisi_sas.h +2 -1
@@ -8,6 +8,8 @@#define _HISI_SAS_H_#include <linux/acpi.h> +#include <linux/blk-mq.h> +#include <linux/blk-mq-pci.h> #include <linux/clk.h> #include <linux/debugfs.h> #include <linux/dmapool.h>
@@ -431,7 +433,6 @@ struct hisi_hba {u32 intr_coal_count;/* Interrupt count to coalesce */ int cq_nvecs; -unsigned int *reply_map;/* bist */ enum sas_linkrate debugfs_bist_linkrate;
drivers/scsi/hisi_sas/hisi_sas_main.c +19 -17
@@ -421,6 +421,7 @@ static int hisi_sas_task_prep(struct sas_task *task,struct device *dev = hisi_hba->dev;int dlvry_queue_slot, dlvry_queue, rc, slot_idx;int n_elem =0 , n_elem_dif =0 , n_elem_req =0 ; +struct scsi_cmnd *scmd =NULL ;struct hisi_sas_dq *dq;unsigned long flags;int wr_q_index;
@@ -436,10 +437,23 @@ static int hisi_sas_task_prep(struct sas_task *task,return -ECOMM ; } -if (hisi_hba->reply_map) { -int cpu =raw_smp_processor_id (); -unsigned int dq_index = hisi_hba->reply_map[cpu]; +if (task->uldd_task) { +struct ata_queued_cmd *qc; +if (dev_is_sata (device)) { + qc = task->uldd_task; + scmd = qc->scsicmd; + }else { + scmd = task->uldd_task; + } + } + +if (scmd) { +unsigned int dq_index; +u32 blk_tag; + + blk_tag =blk_mq_unique_tag (scmd->request); + dq_index =blk_mq_unique_tag_to_hwq (blk_tag); *dq_pointer = dq = &hisi_hba->dq[dq_index]; }else { *dq_pointer = dq = sas_dev->dq;
@@ -468,21 +482,9 @@ static int hisi_sas_task_prep(struct sas_task *task,if (hisi_hba->hw->slot_index_alloc) rc = hisi_hba->hw->slot_index_alloc (hisi_hba, device); -else { -struct scsi_cmnd *scsi_cmnd =NULL ; - -if (task->uldd_task) { -struct ata_queued_cmd *qc; +else + rc =hisi_sas_slot_index_alloc (hisi_hba, scmd); -if (dev_is_sata (device)) { - qc = task->uldd_task; - scsi_cmnd = qc->scsicmd; - }else { - scsi_cmnd = task->uldd_task; - } - } - rc =hisi_sas_slot_index_alloc (hisi_hba, scsi_cmnd); - }if (rc <0 ) goto err_out_dif_dma_unmap;
drivers/scsi/hisi_sas/hisi_sas_v3_hw.c +35 -51
125 lines not on this page · raw message
Signature · 1 line
2.16.4
PATCH v5 9/11 megaraid_sas: switch fusion adapters to MQ
In reply to Hannes ReineckeFusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.
--- drivers/scsi/megaraid/megaraid_sas.h | 1 - drivers/scsi/megaraid/megaraid_sas_base.c | 65 +++++++++-------------------- drivers/scsi/megaraid/megaraid_sas_fusion.c | 14 ++++--- 3 files changed, 28 insertions(+), 52 deletions(-)
drivers/scsi/megaraid/megaraid_sas.h +0 -1
@@ -2261,7 +2261,6 @@ enum MR_PERF_MODE {struct megasas_instance { -unsigned int *reply_map;__le32 *producer;dma_addr_t producer_h;__le32 *consumer;
drivers/scsi/megaraid/megaraid_sas_base.c +19 -46
134 lines not on this page · raw message
drivers/scsi/megaraid/megaraid_sas_fusion.c +9 -5
@@ -2731,6 +2731,7 @@ megasas_build_ldio_fusion(struct megasas_instance *instance,struct MR_PRIV_DEVICE *mrdev_priv;struct RAID_CONTEXT *rctx;struct RAID_CONTEXT_G35 *rctx_g35; +u32 tag =blk_mq_unique_tag (scp->request); device_id =MEGASAS_DEV_INDEX (scp);
@@ -2837,7 +2838,7 @@ megasas_build_ldio_fusion(struct megasas_instance *instance, instance->msix_vectors)); else cmd->request_desc->SCSIIO.MSIxIndex = - instance->reply_map[raw_smp_processor_id ()]; +blk_mq_unique_tag_to_hwq (tag);if (instance->adapter_type >=VENTURA_SERIES ) {/* FP for Optimal raid level 1.
@@ -3080,6 +3081,7 @@ megasas_build_syspd_fusion(struct megasas_instance *instance,u16 pd_index =0 ;u16 os_timeout_value;u16 timeout_limit; +u32 tag =blk_mq_unique_tag (scmd->request);struct MR_DRV_RAID_MAP_ALL *local_map_ptr;struct RAID_CONTEXT *pRAID_Context;struct MR_PD_CFG_SEQ_NUM_SYNC *pd_sync;
@@ -3169,7 +3171,7 @@ megasas_build_syspd_fusion(struct megasas_instance *instance, instance->msix_vectors)); else cmd->request_desc->SCSIIO.MSIxIndex = - instance->reply_map[raw_smp_processor_id ()]; +blk_mq_unique_tag_to_hwq (tag);if (!fp_possible) {/* system pd firmware path */
@@ -3373,7 +3375,7 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance, {struct megasas_cmd_fusion *cmd, *r1_cmd =NULL ;union MEGASAS_REQUEST_DESCRIPTOR_UNION *req_desc; -u32 index; +u32 index, blk_tag, unique_tag; if ((megasas_cmd_type(scmd) == READ_WRITE_LDIO) && instance->ldio_threshold &&
@@ -3389,7 +3391,9 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance,return SCSI_MLQUEUE_HOST_BUSY ; } - cmd =megasas_get_cmd_fusion (instance, scmd->request->tag); + unique_tag =blk_mq_unique_tag (scmd->request); + blk_tag =blk_mq_unique_tag_to_tag (unique_tag); + cmd =megasas_get_cmd_fusion (instance, blk_tag);if (!cmd) {atomic_dec (&instance->fw_outstanding);
@@ -3430,7 +3434,7 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance,*/ if (cmd->r1_alt_dev_handle !=MR_DEVHANDLE_INVALID ) { r1_cmd =megasas_get_cmd_fusion (instance, - (scmd->request->tag + instance->max_fw_cmds)); + (blk_tag + instance->max_fw_cmds));megasas_prepare_secondRaid1_IO (instance, cmd, r1_cmd); }
Signature · 1 line
2.16.4
Sumit Saxena
maintainerRe: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.
Hi Hannes,
Ming Lei also proposed similar changes in megaraid_sas driver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/
So, we will do some performance tests with this patch and update you.
Thanks,
Sumit
Quoted 239 lines
Signed-off-by Hannes Reinecke <[email protected]>--- drivers/scsi/megaraid/megaraid_sas.h | 1 - drivers/scsi/megaraid/megaraid_sas_base.c | 65 +++++++++-------------------- drivers/scsi/megaraid/megaraid_sas_fusion.c | 14 ++++--- 3 files changed, 28 insertions(+), 52 deletions(-)drivers/scsi/megaraid/megaraid_sas.h +0
-1@@ -2261,7 +2261,6 @@ enum MR_PERF_MODE {struct megasas_instance { -unsigned int *reply_map;__le32 *producer;dma_addr_t producer_h;__le32 *consumer;drivers/scsi/megaraid/megaraid_sas_base.c +19
-46@@ -37,6 +37,7 @@#include <linux/poll.h> #include <linux/vmalloc.h> #include <linux/irq_poll.h> +#include <linux/blk-mq-pci.h> #include <scsi/scsi.h> #include <scsi/scsi_cmnd.h> @@ -3106,6 +3107,19 @@ megasas_bios_param(struct scsi_device *sdev, struct block_device *bdev,return 0 ; } +static int megasas_map_queues (struct Scsi_Host *shost) +{ +struct megasas_instance *instance; + + instance = (struct megasas_instance *)shost->hostdata; + +if (!instance->smp_affinity_enable) +return 0 ; + +return blk_mq_pci_map_queues (&shost->tag_set.map[HCTX_TYPE_DEFAULT ], + instance->pdev, instance->low_latency_index_start); +} +static void megasas_aen_polling (struct work_struct *work);/** @@ -3414,9 +3428,11 @@ static struct scsi_host_template megasas_template = { .eh_timed_out = megasas_reset_timer, .shost_attrs = megaraid_host_attrs, .bios_param = megasas_bios_param, + .map_queues = megasas_map_queues, .change_queue_depth = scsi_change_queue_depth, .max_segment_size = 0xffffffff, .no_write_same = 1, + .host_tagset = 1, };/** @@ -5695,34 +5711,6 @@ megasas_setup_jbod_map(struct megasas_instance *instance) instance->use_seqnum_jbod_fp = false; } -static void megasas_setup_reply_map (struct megasas_instance *instance) -{ -const struct cpumask *mask; -unsigned int queue, cpu, low_latency_index_start; - - low_latency_index_start = instance->low_latency_index_start; - -for (queue = low_latency_index_start; queue < instance->msix_vectors; queue++) { - mask =pci_irq_get_affinity (instance->pdev, queue); -if (!mask) - goto fallback; - -for_each_cpu (cpu, mask) - instance->reply_map[cpu] = queue; - } -return ; - -fallback: - queue = low_latency_index_start; -for_each_possible_cpu (cpu) { - instance->reply_map[cpu] = queue; -if (queue == (instance->msix_vectors -1 )) - queue = low_latency_index_start; -else - queue++; - } -} -/** * megasas_get_device_list - Get the PD and LD device list from FW. * @instance: Adapter soft state @@ -6021,12 +6009,6 @@ static int megasas_init_fw(struct megasas_instance *instance) instance->is_rdpq = (scratch_pad_1 &MR_RDPQ_MODE_OFFSET ) ?1 :0 ; -if (instance->adapter_type >=INVADER_SERIES && - !instance->msix_combined) { - instance->msix_load_balance = true; - instance->smp_affinity_enable = false; - } -/* Save 1-15 reply post index address to local memory * Index 0 is already saved from reg offset * MPI2_REPLY_POST_HOST_INDEX_OFFSET @@ -6145,8 +6127,6 @@ static int megasas_init_fw(struct megasas_instance *instance) goto fail_init_adapter; } -megasas_setup_reply_map (instance); -dev_info (&instance->pdev->dev,"current msix/online cpus\t: (%d/%d)\n" , instance->msix_vectors, (unsigned int )num_online_cpus ());@@ -6780,6 +6760,9 @@ static int megasas_io_attach(struct megasas_instance *instance) host->max_id =MEGASAS_MAX_DEV_PER_CHANNEL ; host->max_lun =MEGASAS_MAX_LUN ; host->max_cmd_len =16 ; +if (instance->adapter_type !=MFI_SERIES && instance->msix_vectors >0 ) + host->nr_hw_queues = instance->msix_vectors - + instance->low_latency_index_start;/* * Notify the mid-layer about the new controller @@ -6947,11 +6930,6 @@ static inline int megasas_alloc_mfi_ctrl_mem(struct megasas_instance *instance)*/ static int megasas_alloc_ctrl_mem (struct megasas_instance *instance) { - instance->reply_map =kcalloc (nr_cpu_ids,sizeof (unsigned int ), -GFP_KERNEL ); -if (!instance->reply_map) -return -ENOMEM ; -switch (instance->adapter_type) {case MFI_SERIES :if (megasas_alloc_mfi_ctrl_mem (instance))@@ -6968,8 +6946,6 @@ static int megasas_alloc_ctrl_mem(struct megasas_instance *instance)return 0 ; fail: -kfree (instance->reply_map); - instance->reply_map =NULL ;return -ENOMEM ; }@@ -6982,7 +6958,6 @@ static int megasas_alloc_ctrl_mem(struct megasas_instance *instance)*/ static inline void megasas_free_ctrl_mem (struct megasas_instance *instance) { -kfree (instance->reply_map);if (instance->adapter_type ==MFI_SERIES ) {if (instance->producer) dma_free_coherent(&instance->pdev->dev, sizeof(u32),@@ -7645,8 +7620,6 @@ megasas_resume(struct pci_dev *pdev)if (rval <0 ) goto fail_reenable_msix; -megasas_setup_reply_map (instance); -if (instance->adapter_type !=MFI_SERIES ) {megasas_reset_reply_desc (instance);if (megasas_ioc_init_fusion (instance)) {drivers/scsi/megaraid/megaraid_sas_fusion.c +9
-6@@ -2731,6 +2731,7 @@ megasas_build_ldio_fusion(struct megasas_instance *instance,struct MR_PRIV_DEVICE *mrdev_priv;struct RAID_CONTEXT *rctx;struct RAID_CONTEXT_G35 *rctx_g35; +u32 tag =blk_mq_unique_tag (scp->request); device_id =MEGASAS_DEV_INDEX (scp);@@ -2837,7 +2838,7 @@ megasas_build_ldio_fusion(struct megasas_instance *instance, instance->msix_vectors)); else cmd->request_desc->SCSIIO.MSIxIndex = - instance->reply_map[raw_smp_processor_id ()]; +blk_mq_unique_tag_to_hwq (tag);if (instance->adapter_type >=VENTURA_SERIES ) {/* FP for Optimal raid level 1. @@ -3080,6 +3081,7 @@ megasas_build_syspd_fusion(struct megasas_instance *instance,u16 pd_index =0 ;u16 os_timeout_value;u16 timeout_limit; +u32 tag =blk_mq_unique_tag (scmd->request);struct MR_DRV_RAID_MAP_ALL *local_map_ptr;struct RAID_CONTEXT *pRAID_Context;struct MR_PD_CFG_SEQ_NUM_SYNC *pd_sync;@@ -3169,7 +3171,7 @@ megasas_build_syspd_fusion(struct megasas_instance *instance, instance->msix_vectors)); else cmd->request_desc->SCSIIO.MSIxIndex = - instance->reply_map[raw_smp_processor_id ()]; +blk_mq_unique_tag_to_hwq (tag);if (!fp_possible) {/* system pd firmware path */ @@ -3373,7 +3375,7 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance, {struct megasas_cmd_fusion *cmd, *r1_cmd =NULL ;union MEGASAS_REQUEST_DESCRIPTOR_UNION *req_desc; -u32 index; +u32 index, blk_tag, unique_tag; if ((megasas_cmd_type(scmd) == READ_WRITE_LDIO) && instance->ldio_threshold &&@@ -3389,7 +3391,9 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance,return SCSI_MLQUEUE_HOST_BUSY ; } - cmd =megasas_get_cmd_fusion (instance, scmd->request->tag); + unique_tag =blk_mq_unique_tag (scmd->request); + blk_tag =blk_mq_unique_tag_to_tag (unique_tag); + cmd =megasas_get_cmd_fusion (instance, blk_tag);if (!cmd) {atomic_dec (&instance->fw_outstanding);@@ -3430,7 +3434,7 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance,*/ if (cmd->r1_alt_dev_handle !=MR_DEVHANDLE_INVALID ) { r1_cmd =megasas_get_cmd_fusion (instance, - (scmd->request->tag + instance->max_fw_cmds)); + (blk_tag + instance->max_fw_cmds));megasas_prepare_secondRaid1_IO (instance, cmd, r1_cmd); } --2.16.4
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.
Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you. Thank you.
I'm aware of the results of Ming Leis work, but I do hope this patchset performs better.
And when you do performance measurements, can you please run with both, 'none' I/O scheduler and 'mq-deadline' I/O scheduler?
I've measured quite a performance improvements when using mq-deadline, up to the point where I've gotten on-par performance with the original, non-mq, implementation. (As a data point, on my setup I've measured about 270k IOPS and 1092 MB/s througput, running on just 2 SSDs).
But thanks for doing a performance test here.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Teamlead Storage & Networking [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Sumit Saxena
maintainerRe: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 29 lines · Hannes Reinecke
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you. Thank you.
I'm aware of the results of Ming Leis work, but I do hope this patchset performs better.
And when you do performance measurements, can you please run with both, 'none' I/O scheduler and 'mq-deadline' I/O scheduler?
I've measured quite a performance improvements when using mq-deadline, up to the point where I've gotten on-par performance with the original, non-mq, implementation.
(As a data point, on my setup I've measured about 270k IOPS and 1092 MB/s througput, running on just 2 SSDs).
Quoted 1 line
But thanks for doing a performance test here.
Hi Hannes,
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.
lspcu:
# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 72 On-line CPU(s) list: 0-71 Thread(s) per core: 2 Core(s) per socket: 18 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 85 Model name: Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz Stepping: 4 CPU MHz: 3204.246 CPU max MHz: 3700.0000 CPU min MHz: 1200.0000 BogoMIPS: 5400.00 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 1024K L3 cache: 25344K NUMA node0 CPU(s): 0-17,36-53 NUMA node1 CPU(s): 18-35,54-71 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe s yscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin mba tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_lo
fio test:
#numactl -N 0 fio jbod.fio
# cat jbod.fio [global] allow_mounted_write=0 ioengine=libaio buffered=0 direct=1 group_reporting iodepth=1 bs=4096 readwrite=randread ..
In this test, IOs are submitted on Node 0 but observed IO completions on Node 1.
IRQs / 1 second(s)
IRQ# TOTAL NODE0 NODE1 NAME
176 48387 48387 0 IR-PCI-MSI 34603073-edge megasas14-msix65 178 47966 47966 0 IR-PCI-MSI 34603075-edge megasas14-msix67 170 47706 47706 0 IR-PCI-MSI 34603067-edge megasas14-msix59 180 47291 47291 0 IR-PCI-MSI 34603077-edge megasas14-msix69 181 47155 47155 0 IR-PCI-MSI 34603078-edge megasas14-msix70 173 46806 46806 0 IR-PCI-MSI 34603070-edge megasas14-msix62 179 46773 46773 0 IR-PCI-MSI 34603076-edge megasas14-msix68 169 46600 46600 0 IR-PCI-MSI 34603066-edge megasas14-msix58 175 46447 46447 0 IR-PCI-MSI 34603072-edge megasas14-msix64 172 46184 46184 0 IR-PCI-MSI 34603069-edge megasas14-msix61 182 46117 46117 0 IR-PCI-MSI 34603079-edge megasas14-msix71 165 46070 46070 0 IR-PCI-MSI 34603062-edge megasas14-msix54 164 45892 45892 0 IR-PCI-MSI 34603061-edge megasas14-msix53 174 45864 45864 0 IR-PCI-MSI 34603071-edge megasas14-msix63 156 45348 45348 0 IR-PCI-MSI 34603053-edge megasas14-msix45 147 45302 0 45302 IR-PCI-MSI 34603044-edge megasas14-msix36 151 44922 0 44922 IR-PCI-MSI 34603048-edge megasas14-msix40 171 44876 44876 0 IR-PCI-MSI 34603068-edge megasas14-msix60 159 44755 44755 0 IR-PCI-MSI 34603056-edge megasas14-msix48 148 44695 0 44695 IR-PCI-MSI 34603045-edge megasas14-msix37 157 44304 44304 0 IR-PCI-MSI 34603054-edge megasas14-msix46 167 42552 42552 0 IR-PCI-MSI 34603064-edge megasas14-msix56 154 35937 0 35937 IR-PCI-MSI 34603051-edge megasas14-msix43 166 16004 16004 0 IR-PCI-MSI 34603063-edge megasas14-msix55
IRQ-CPU affinity:
Ran below script to get IRQ-CPU affinity:
--
#!/bin/bash
PCID=lspci | grep "SAS39xx" | cut -c1-7
PCIP=find /sys/devices -name *$PCID | grep pci
IRQS=ls $PCIP/msi_irqs
echo "kernel version: "
uname -a
for IRQ in $IRQS; do
CPUS=cat /proc/irq/$IRQ/smp_affinity_list
echo "irq $IRQ, cpu list $CPUS"
done
--
irq 103, cpu list 0-17,36-53
irq 112, cpu list 0-17,36-53
irq 113, cpu list 0-17,36-53
irq 114, cpu list 0-17,36-53
irq 115, cpu list 0-17,36-53
irq 116, cpu list 0-17,36-53
irq 117, cpu list 0-17,36-53
irq 118, cpu list 0-17,36-53
irq 119, cpu list 18
irq 120, cpu list 19
irq 121, cpu list 20
irq 122, cpu list 21
irq 123, cpu list 22
irq 124, cpu list 23
irq 125, cpu list 24
irq 126, cpu list 25
irq 127, cpu list 26
irq 128, cpu list 27
irq 129, cpu list 28
irq 130, cpu list 29
irq 131, cpu list 30
irq 132, cpu list 31
irq 133, cpu list 32
irq 134, cpu list 33
irq 135, cpu list 34
irq 136, cpu list 35
irq 137, cpu list 54
irq 138, cpu list 55
irq 139, cpu list 56
irq 140, cpu list 57
irq 141, cpu list 58
irq 142, cpu list 59
irq 143, cpu list 60
irq 144, cpu list 61
irq 145, cpu list 62
irq 146, cpu list 63
irq 147, cpu list 64
irq 148, cpu list 65
irq 149, cpu list 66
irq 150, cpu list 67
irq 151, cpu list 68
irq 152, cpu list 69
irq 153, cpu list 70
irq 154, cpu list 71
irq 155, cpu list 0
irq 156, cpu list 1
irq 157, cpu list 2
irq 158, cpu list 3
irq 159, cpu list 4
irq 160, cpu list 5
irq 161, cpu list 6
irq 162, cpu list 7
irq 163, cpu list 8
irq 164, cpu list 9
irq 165, cpu list 10
irq 166, cpu list 11
irq 167, cpu list 12
irq 168, cpu list 13
irq 169, cpu list 14
irq 170, cpu list 15
irq 171, cpu list 16
irq 172, cpu list 17
irq 173, cpu list 36
irq 174, cpu list 37
irq 175, cpu list 38
irq 176, cpu list 39
irq 177, cpu list 40
irq 178, cpu list 41
irq 179, cpu list 42
irq 180, cpu list 43
irq 181, cpu list 44
irq 182, cpu list 45
irq 183, cpu list 46
irq 184, cpu list 47
irq 185, cpu list 48
irq 186, cpu list 49
irq 187, cpu list 50
irq 188, cpu list 51
irq 189, cpu list 52
irq 190, cpu list 53
I added prints in megaraid_sas driver's IO path to catch when MSI-x affined to IO submitter CPU does not match with what is returned by API "blk_mq_unique_tag_to_hwq(tag)".
I have copied below few prints from dmesg logs- in these prints CPU is submitter CPU, calculated MSX-x is what is returned by "blk_mq_unique_tag_to_hwq(tag)" and affined MSI-x is actual MSI-x to which submitting
CPU is affined to:
[2536843.629877] BRCM DBG: CPU:6 calculated MSI-x: 153 affined MSIx: 161 [2536843.641674] BRCM DBG: CPU:39 calculated MSI-x: 168 affined MSIx: 176 [2536843.641674] BRCM DBG: CPU:13 calculated MSI-x: 160 affined MSIx: 168
2. Seeing below stack traces/messages in dmesg during driver unload –
[2565601.054366] Call Trace: [2565601.054368] blk_mq_free_map_and_requests+0x28/0x50 [2565601.054369] blk_mq_free_tag_set+0x1d/0x90 [2565601.054370] scsi_host_dev_release+0x8a/0xf0 [2565601.054370] device_release+0x27/0x80 [2565601.054371] kobject_cleanup+0x61/0x190 [2565601.054373] megasas_detach_one+0x4c1/0x650 [megaraid_sas] [2565601.054374] pci_device_remove+0x3b/0xc0 [2565601.054375] device_release_driver_internal+0xec/0x1b0 [2565601.054376] driver_detach+0x46/0x90 [2565601.054377] bus_remove_driver+0x58/0xd0 [2565601.054378] pci_unregister_driver+0x26/0xa0 [2565601.054379] megasas_exit+0x91/0x882 [megaraid_sas] [2565601.054381] __x64_sys_delete_module+0x16c/0x250 [2565601.054382] do_syscall_64+0x5b/0x1b0 [2565601.054383] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [2565601.054383] RIP: 0033:0x7f7212a82837 [2565601.054384] RSP: 002b:00007ffdfa2dcea8 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0 [2565601.054385] RAX: ffffffffffffffda RBX: 0000000000b6e2e0 RCX: 00007f7212a82837 [2565601.054385] RDX: 00007f7212af3ac0 RSI: 0000000000000800 RDI: 0000000000b6e348 [2565601.054386] RBP: 0000000000000000 R08: 00007f7212d47060 R09: 00007f7212af3ac0 [2565601.054386] R10: 00007ffdfa2dcbc0 R11: 0000000000000202 R12: 00007ffdfa2dd71c [2565601.054387] R13: 0000000000000000 R14: 0000000000b6e2e0 R15: 0000000000b6e010 [2565601.054387] ---[ end trace 38899303bd85e838 ]--- [2565601.054390] ------------[ cut here ]------------ [2565601.054391] WARNING: CPU: 31 PID: 50996 at block/blk-mq.c:2056 blk_mq_free_rqs+0x10b/0x120 [2565601.054391] Modules linked in: megaraid_sas(-) ses enclosure
scsi_transport_sas xt_CHECKSUM xt_MASQUERADE tun bridge stp llc ip6t_rpfilter ipt_REJECT nf_reject_ipv4 ip6t_REJECT nf_reject_ipv6 xt_conntrack ip_set nfnetlink ebtable_nat ebtable_broute ip6table_nat ip6table_mangle ip6table_security ip6table_raw iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 iptable_mangle iptable_security iptable_raw ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter intel_rapl_msr intel_rapl_common skx_edac nfit libnvdimm ftdi_sio x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel snd_hda_codec_realtek kvm snd_hda_codec_generic ledtrig_audio snd_hda_intel snd_intel_nhlt snd_hda_codec snd_hda_core irqbypass snd_hwdep crct10dif_pclmul snd_seq crc32_pclmul ghash_clmulni_intel snd_seq_device snd_pcm iTCO_wdt mei_me iTCO_vendor_support cryptd snd_timer joydev snd mei soundcore ioatdma sg pcspkr ipmi_si ipmi_devintf ipmi_msghandler dca i2c_i801 lpc_ich acpi_power_meter wmi
[2565601.054400] acpi_pad nfsd auth_rpcgss nfs_acl lockd grace sunrpc
ip_tables xfs libcrc32c sd_mod ast i2c_algo_bit drm_vram_helper ttm drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops drm e1000e ahci libahci crc32c_intel nvme libata nvme_core i2c_core dm_mirror dm_region_hash dm_log dm_mod [last unloaded: megaraid_sas]
[2565601.054404] CPU: 31 PID: 50996 Comm: rmmod Kdump: loaded Tainted: G W OE 5.4.0-rc1+ #2 [2565601.054405] Hardware name: Supermicro Super Server/X11DPG-QT, BIOS 1.0 06/22/2017 [2565601.054406] RIP: 0010:blk_mq_free_rqs+0x10b/0x120 [2565601.054407] Code: 89 10 48 8b 73 20 48 89 1b 4c 89 e7 48 89 5b 08 e8 2a 54 e7 ff 48 8b 85 b0 00 00 00 49 39 c5 75 bc 5b 5d 41 5c 41 5d 41 5e c3 <0f> 0b 48 c7 02 00 00 00 00 e9 2b ff ff ff 0f 1f 80 00 00 00 00 0f [2565601.054407] RSP: 0018:ffffb37446a6bd58 EFLAGS: 00010286 [2565601.054408] RAX: 0000000000000747 RBX: ffff92219cb280a8 RCX: 0000000000000747 [2565601.054408] RDX: ffff92219b153a38 RSI: ffff9221692bb5c0 RDI: ffff92219cb280a8 [2565601.054408] RBP: ffff9221692bb5c0 R08: 0000000000000001 R09: 0000000000000000 [2565601.054409] R10: ffff9221692bb680 R11: ffff9221bffd5f60 R12: ffff9221970dd678 [2565601.054409] R13: 0000000000000030 R14: ffff92219cb280a8 R15: ffff922199ada010 [2565601.054410] FS: 00007f72135a1740(0000) GS:ffff92299f540000(0000) knlGS:0000000000000000 [2565601.054410] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [2565601.054410] CR2: 0000000000b77c78 CR3: 0000000f27db0006 CR4: 00000000007606e0 [2565601.054412] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [2565601.054412] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [2565601.054413] PKRU: 55555554
3. For High IOs(outstanding IOs per physical disk > 8) oriented workloads, performance numbers are good(no performance drop) as in that case driver uses non-managed affinity high IOPs reply queues and this patchset does not touch driver's high IOPs IO path.
4. This patch removes below code from driver so what this piece of code does is broken-
- if (instance->adapter_type >= INVADER_SERIES &&
- !instance->msix_combined) {
- instance->msix_load_balance = true;
- instance->smp_affinity_enable = false;
- }
Thanks,
Sumit
Quoted 9 lines
Cheers,
Hannes
--
Dr. Hannes Reinecke Teamlead Storage & Networking
[email protected] +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 30 lines · Sumit Saxena
Quoted 29 lines · Hannes Reinecke
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you. Thank you.
I'm aware of the results of Ming Leis work, but I do hope this patchset performs better.
And when you do performance measurements, can you please run with both, 'none' I/O scheduler and 'mq-deadline' I/O scheduler?
I've measured quite a performance improvements when using mq-deadline, up to the point where I've gotten on-par performance with the original, non-mq, implementation.
(As a data point, on my setup I've measured about 270k IOPS and 1092 MB/s througput, running on just 2 SSDs).
Quoted 2 lines
Quoted 1 line
But thanks for doing a performance test here.
Hi Sumit,
Thanks for this. We'll have a look ...
Cheers,
John
EOM
Quoted 332 lines
Hi Hannes,
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.lspcu:
# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 72 On-line CPU(s) list: 0-71 Thread(s) per core: 2 Core(s) per socket: 18 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 85 Model name: Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz Stepping: 4 CPU MHz: 3204.246 CPU max MHz: 3700.0000 CPU min MHz: 1200.0000 BogoMIPS: 5400.00 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 1024K L3 cache: 25344K NUMA node0 CPU(s): 0-17,36-53 NUMA node1 CPU(s): 18-35,54-71 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe s yscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin mba tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_lofio test:
#numactl -N 0 fio jbod.fio
# cat jbod.fio [global] allow_mounted_write=0 ioengine=libaio buffered=0 direct=1 group_reporting iodepth=1 bs=4096 readwrite=randread ..In this test, IOs are submitted on Node 0 but observed IO completions on Node 1.
IRQs / 1 second(s)
IRQ# TOTAL NODE0 NODE1 NAME176 48387 48387 0 IR-PCI-MSI 34603073-edge megasas14-msix65 178 47966 47966 0 IR-PCI-MSI 34603075-edge megasas14-msix67 170 47706 47706 0 IR-PCI-MSI 34603067-edge megasas14-msix59 180 47291 47291 0 IR-PCI-MSI 34603077-edge megasas14-msix69 181 47155 47155 0 IR-PCI-MSI 34603078-edge megasas14-msix70 173 46806 46806 0 IR-PCI-MSI 34603070-edge megasas14-msix62 179 46773 46773 0 IR-PCI-MSI 34603076-edge megasas14-msix68 169 46600 46600 0 IR-PCI-MSI 34603066-edge megasas14-msix58 175 46447 46447 0 IR-PCI-MSI 34603072-edge megasas14-msix64 172 46184 46184 0 IR-PCI-MSI 34603069-edge megasas14-msix61 182 46117 46117 0 IR-PCI-MSI 34603079-edge megasas14-msix71 165 46070 46070 0 IR-PCI-MSI 34603062-edge megasas14-msix54 164 45892 45892 0 IR-PCI-MSI 34603061-edge megasas14-msix53 174 45864 45864 0 IR-PCI-MSI 34603071-edge megasas14-msix63 156 45348 45348 0 IR-PCI-MSI 34603053-edge megasas14-msix45 147 45302 0 45302 IR-PCI-MSI 34603044-edge megasas14-msix36 151 44922 0 44922 IR-PCI-MSI 34603048-edge megasas14-msix40 171 44876 44876 0 IR-PCI-MSI 34603068-edge megasas14-msix60 159 44755 44755 0 IR-PCI-MSI 34603056-edge megasas14-msix48 148 44695 0 44695 IR-PCI-MSI 34603045-edge megasas14-msix37 157 44304 44304 0 IR-PCI-MSI 34603054-edge megasas14-msix46 167 42552 42552 0 IR-PCI-MSI 34603064-edge megasas14-msix56 154 35937 0 35937 IR-PCI-MSI 34603051-edge megasas14-msix43 166 16004 16004 0 IR-PCI-MSI 34603063-edge megasas14-msix55IRQ-CPU affinity:
Ran below script to get IRQ-CPU affinity:
--
#!/bin/bash
PCID=lspci | grep "SAS39xx" | cut -c1-7
PCIP=find /sys/devices -name *$PCID | grep pci
IRQS=ls $PCIP/msi_irqsecho "kernel version: "
uname -afor IRQ in $IRQS; do
CPUS=cat /proc/irq/$IRQ/smp_affinity_list
echo "irq $IRQ, cpu list $CPUS"
done--
irq 103, cpu list 0-17,36-53
irq 112, cpu list 0-17,36-53
irq 113, cpu list 0-17,36-53
irq 114, cpu list 0-17,36-53
irq 115, cpu list 0-17,36-53
irq 116, cpu list 0-17,36-53
irq 117, cpu list 0-17,36-53
irq 118, cpu list 0-17,36-53
irq 119, cpu list 18
irq 120, cpu list 19
irq 121, cpu list 20
irq 122, cpu list 21
irq 123, cpu list 22
irq 124, cpu list 23
irq 125, cpu list 24
irq 126, cpu list 25
irq 127, cpu list 26
irq 128, cpu list 27
irq 129, cpu list 28
irq 130, cpu list 29
irq 131, cpu list 30
irq 132, cpu list 31
irq 133, cpu list 32
irq 134, cpu list 33
irq 135, cpu list 34
irq 136, cpu list 35
irq 137, cpu list 54
irq 138, cpu list 55
irq 139, cpu list 56
irq 140, cpu list 57
irq 141, cpu list 58
irq 142, cpu list 59
irq 143, cpu list 60
irq 144, cpu list 61
irq 145, cpu list 62
irq 146, cpu list 63
irq 147, cpu list 64
irq 148, cpu list 65
irq 149, cpu list 66
irq 150, cpu list 67
irq 151, cpu list 68
irq 152, cpu list 69
irq 153, cpu list 70
irq 154, cpu list 71
irq 155, cpu list 0
irq 156, cpu list 1
irq 157, cpu list 2
irq 158, cpu list 3
irq 159, cpu list 4
irq 160, cpu list 5
irq 161, cpu list 6
irq 162, cpu list 7
irq 163, cpu list 8
irq 164, cpu list 9
irq 165, cpu list 10
irq 166, cpu list 11
irq 167, cpu list 12
irq 168, cpu list 13
irq 169, cpu list 14
irq 170, cpu list 15
irq 171, cpu list 16
irq 172, cpu list 17
irq 173, cpu list 36
irq 174, cpu list 37
irq 175, cpu list 38
irq 176, cpu list 39
irq 177, cpu list 40
irq 178, cpu list 41
irq 179, cpu list 42
irq 180, cpu list 43
irq 181, cpu list 44
irq 182, cpu list 45
irq 183, cpu list 46
irq 184, cpu list 47
irq 185, cpu list 48
irq 186, cpu list 49
irq 187, cpu list 50
irq 188, cpu list 51
irq 189, cpu list 52
irq 190, cpu list 53I added prints in
megaraid_sasdriver's IO path to catch when MSI-x affined to IO submitter CPU does not match with what is returned by API "blk_mq_unique_tag_to_hwq(tag)".
I have copied below few prints from dmesg logs- in these prints CPU is submitter CPU, calculated MSX-x is what is returned by "blk_mq_unique_tag_to_hwq(tag)" and affined MSI-x is actual MSI-x to which submitting
CPU is affined to:[2536843.629877] BRCM DBG: CPU:6 calculated MSI-x: 153 affined MSIx: 161 [2536843.641674] BRCM DBG: CPU:39 calculated MSI-x: 168 affined MSIx: 176 [2536843.641674] BRCM DBG: CPU:13 calculated MSI-x: 160 affined MSIx: 1682. Seeing below stack traces/messages in dmesg during driver unload –
[2565601.054366] Call Trace: [2565601.054368] blk_mq_free_map_and_requests+0x28/0x50 [2565601.054369] blk_mq_free_tag_set+0x1d/0x90 [2565601.054370] scsi_host_dev_release+0x8a/0xf0 [2565601.054370] device_release+0x27/0x80 [2565601.054371] kobject_cleanup+0x61/0x190 [2565601.054373] megasas_detach_one+0x4c1/0x650 [megaraid_sas] [2565601.054374] pci_device_remove+0x3b/0xc0 [2565601.054375] device_release_driver_internal+0xec/0x1b0 [2565601.054376] driver_detach+0x46/0x90 [2565601.054377] bus_remove_driver+0x58/0xd0 [2565601.054378] pci_unregister_driver+0x26/0xa0 [2565601.054379] megasas_exit+0x91/0x882 [megaraid_sas] [2565601.054381] __x64_sys_delete_module+0x16c/0x250 [2565601.054382] do_syscall_64+0x5b/0x1b0 [2565601.054383] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [2565601.054383] RIP: 0033:0x7f7212a82837 [2565601.054384] RSP: 002b:00007ffdfa2dcea8 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0 [2565601.054385] RAX: ffffffffffffffda RBX: 0000000000b6e2e0 RCX: 00007f7212a82837 [2565601.054385] RDX: 00007f7212af3ac0 RSI: 0000000000000800 RDI: 0000000000b6e348 [2565601.054386] RBP: 0000000000000000 R08: 00007f7212d47060 R09: 00007f7212af3ac0 [2565601.054386] R10: 00007ffdfa2dcbc0 R11: 0000000000000202 R12: 00007ffdfa2dd71c [2565601.054387] R13: 0000000000000000 R14: 0000000000b6e2e0 R15: 0000000000b6e010 [2565601.054387] ---[ end trace 38899303bd85e838 ]--- [2565601.054390] ------------[ cut here ]------------ [2565601.054391] WARNING: CPU: 31 PID: 50996 at block/blk-mq.c:2056 blk_mq_free_rqs+0x10b/0x120 [2565601.054391] Modules linked in: megaraid_sas(-) ses enclosure
scsi_transport_sasxt_CHECKSUMxt_MASQUERADEtun bridge stp llcip6t_rpfilteript_REJECTnf_reject_ipv4ip6t_REJECTnf_reject_ipv6xt_conntrackip_setnfnetlinkebtable_natebtable_brouteip6table_natip6table_mangleip6table_securityip6table_rawiptable_natnf_natnf_conntracknf_defrag_ipv6nf_defrag_ipv4iptable_mangleiptable_securityiptable_rawebtable_filterebtablesip6table_filterip6_tablesiptable_filterintel_rapl_msrintel_rapl_commonskx_edacnfit libnvdimmftdi_siox86_pkg_temp_thermalintel_powerclampcoretempkvm_intelsnd_hda_codec_realtekkvmsnd_hda_codec_genericledtrig_audiosnd_hda_intelsnd_intel_nhltsnd_hda_codecsnd_hda_coreirqbypasssnd_hwdepcrct10dif_pclmulsnd_seqcrc32_pclmulghash_clmulni_intelsnd_seq_devicesnd_pcmiTCO_wdtmei_meiTCO_vendor_supportcryptdsnd_timerjoydev snd mei soundcore ioatdma sg pcspkripmi_siipmi_devintfipmi_msghandlerdcai2c_i801lpc_ichacpi_power_meterwmi[2565601.054400] acpi_pad nfsd auth_rpcgss nfs_acl lockd grace sunrpc
ip_tablesxfs libcrc32csd_modasti2c_algo_bitdrm_vram_helperttmdrm_kms_helpersyscopyarea sysfillrect sysimgbltfb_sys_fopsdrm e1000e ahci libahcicrc32c_intelnvme libatanvme_corei2c_coredm_mirrordm_region_hashdm_logdm_mod[last unloaded:megaraid_sas][2565601.054404] CPU: 31 PID: 50996 Comm: rmmod Kdump: loaded Tainted: G W OE 5.4.0-rc1+ #2 [2565601.054405] Hardware name: Supermicro Super Server/X11DPG-QT, BIOS 1.0 06/22/2017 [2565601.054406] RIP: 0010:blk_mq_free_rqs+0x10b/0x120 [2565601.054407] Code: 89 10 48 8b 73 20 48 89 1b 4c 89 e7 48 89 5b 08 e8 2a 54 e7 ff 48 8b 85 b0 00 00 00 49 39 c5 75 bc 5b 5d 41 5c 41 5d 41 5e c3 <0f> 0b 48 c7 02 00 00 00 00 e9 2b ff ff ff 0f 1f 80 00 00 00 00 0f [2565601.054407] RSP: 0018:ffffb37446a6bd58 EFLAGS: 00010286 [2565601.054408] RAX: 0000000000000747 RBX: ffff92219cb280a8 RCX: 0000000000000747 [2565601.054408] RDX: ffff92219b153a38 RSI: ffff9221692bb5c0 RDI: ffff92219cb280a8 [2565601.054408] RBP: ffff9221692bb5c0 R08: 0000000000000001 R09: 0000000000000000 [2565601.054409] R10: ffff9221692bb680 R11: ffff9221bffd5f60 R12: ffff9221970dd678 [2565601.054409] R13: 0000000000000030 R14: ffff92219cb280a8 R15: ffff922199ada010 [2565601.054410] FS: 00007f72135a1740(0000) GS:ffff92299f540000(0000) knlGS:0000000000000000 [2565601.054410] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [2565601.054410] CR2: 0000000000b77c78 CR3: 0000000f27db0006 CR4: 00000000007606e0 [2565601.054412] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [2565601.054412] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [2565601.054413] PKRU: 555555543. For High IOs(outstanding IOs per physical disk > 8) oriented workloads, performance numbers are good(no performance drop) as in that case driver uses non-managed affinity high IOPs reply queues and this patchset does not touch driver's high IOPs IO path.
4. This patch removes below code from driver so what this piece of code does is broken-
drivers/scsi/megaraid/megaraid_sas_base.c +0
-5-if (instance->adapter_type >=INVADER_SERIES && - !instance->msix_combined) { - instance->msix_load_balance = true; - instance->smp_affinity_enable = false; - }Thanks,
SumitQuoted 9 lines
Cheers,
Hannes
--
Dr. Hannes Reinecke Teamlead Storage & Networking
[email protected] +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer.
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
In reply to Sumit SaxenaQuoted 32 lines · Sumit Saxena
Quoted 21 lines · Hannes Reinecke
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you. Thank you.
I'm aware of the results of Ming Leis work, but I do hope this patchset performs better.
Quoted 9 lines
And when you do performance measurements, can you please run with both, 'none' I/O scheduler and 'mq-deadline' I/O scheduler?
I've measured quite a performance improvements when using mq-deadline, up to the point where I've gotten on-par performance with the original, non-mq, implementation.
(As a data point, on my setup I've measured about 270k IOPS and 1092 MB/s througput, running on just 2 SSDs).asas_build_ldio_fusion
But thanks for doing a performance test here.
Quoted 8 lines
Hi Hannes,
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.
Hi Sumit,
So the new code has:
megasas_build_ldio_fusion()
{
cmd->request_desc->SCSIIO.MSIxIndex =blk_mq_unique_tag_to_hwq(tag);
}
So the value here is hw queue index from blk-mq point of view, and not megaraid_sas msix index, as you alluded to.
So we get 80 msix, 8 are reserved for low_latency_index_start (that's how it seems to me), and we report other 72 as #hw queues = 72 to SCSI midlayer.
So I think that this should be:
cmd->request_desc->SCSIIO.MSIxIndex = blk_mq_unique_tag_to_hwq(tag) + low_latency_index_start;
Quoted 47 lines
lspcu:
# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 72 On-line CPU(s) list: 0-71 Thread(s) per core: 2 Core(s) per socket: 18 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 85 Model name: Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz Stepping: 4 CPU MHz: 3204.246 CPU max MHz: 3700.0000 CPU min MHz: 1200.0000 BogoMIPS: 5400.00 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 1024K L3 cache: 25344K NUMA node0 CPU(s): 0-17,36-53 NUMA node1 CPU(s): 18-35,54-71 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe s yscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin mba tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_lo
[snip]
drivers/scsi/megaraid/megaraid_sas_base.c static int megasas_init_fw(struct megasas_instance *instance) · line 6028
Quoted 9 lines
4. This patch removes below code from driver so what this piece of code does is broken-
drivers/scsi/megaraid/megaraid_sas_base.c +0
-5-if (instance->adapter_type >=INVADER_SERIES && - !instance->msix_combined) { - instance->msix_load_balance = true; - instance->smp_affinity_enable = false; - }
Does this code need to be re-added? Would this have affected your test?
Thanks,
John
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 58 lines · John Garry
Quoted 34 lines · Sumit Saxena
Quoted 21 lines · Hannes Reinecke
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you. Thank you.
I'm aware of the results of Ming Leis work, but I do hope this patchset performs better.
Quoted 9 lines
And when you do performance measurements, can you please run with both, 'none' I/O scheduler and 'mq-deadline' I/O scheduler?
I've measured quite a performance improvements when using mq-deadline, up to the point where I've gotten on-par performance with the original, non-mq, implementation.
(As a data point, on my setup I've measured about 270k IOPS and 1092 MB/s througput, running on just 2 SSDs).asas_build_ldio_fusion
But thanks for doing a performance test here.Hi Hannes,
Quoted 7 lines
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.Hi Sumit,
So the new code has:
megasas_build_ldio_fusion()
{
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag);}
So the value here is hw queue index from blk-mq point of view, and not
megaraid_sasmsix index, as you alluded to.
Quoted 10 lines
So we get 80 msix, 8 are reserved for
low_latency_index_start(that's how it seems to me), and we report other 72 as #hw queues = 72 to SCSI midlayer.So I think that this should be:
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag)+low_latency_index_start;
Indeed, that sounds reasonable.
(The whole queue mapping stuff isn't exactly well documented :-( )
I'll be updating the patch.
Quoted 63 lines
Quoted 47 lines
lspcu:
# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 72 On-line CPU(s) list: 0-71 Thread(s) per core: 2 Core(s) per socket: 18 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 85 Model name: Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz Stepping: 4 CPU MHz: 3204.246 CPU max MHz: 3700.0000 CPU min MHz: 1200.0000 BogoMIPS: 5400.00 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 1024K L3 cache: 25344K NUMA node0 CPU(s): 0-17,36-53 NUMA node1 CPU(s): 18-35,54-71 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe s yscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin mba tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_lo[snip]
Quoted 4 lines
4. This patch removes below code from driver so what this piece of code does is broken-
Quoted 8 lines
- if (
instance->adapter_type>=INVADER_SERIES&&
- !instance->msix_combined) { -instance->msix_load_balance= true;
-instance->smp_affinity_enable= false;
- }
Quoted 2 lines
Does this code need to be re-added? Would this have affected your test? Primarily this patch was required to enable interrupt affinity on my
machine (Lenovo RAID 930-8i).
Can you give me some information why the code is present in the first place? Some hardware limitation, maybe?
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Teamlead Storage & Networking [email protected] +49 911 74053 688 SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 31 lines
Quoted 31 lines
Quoted 11 lines
Hi Hannes,
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.Hi Sumit,
So the new code has:
megasas_build_ldio_fusion()
{
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag);}
So the value here is hw queue index from blk-mq point of view, and not
megaraid_sasmsix index, as you alluded to.So we get 80 msix, 8 are reserved for
low_latency_index_start(that's how it seems to me), and we report other 72 as #hw queues = 72 to SCSI midlayer.
Quoted 9 lines
Quoted 6 lines
So I think that this should be:
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag)+low_latency_index_start;Indeed, that sounds reasonable.
(The whole queue mapping stuff isn't exactly well documented :-( )
Yeah, there's certainly lots of knobs and levers in this driver.
Quoted 1 line
I'll be updating the patch.
About this one:
Quoted 24 lines
2. Seeing below stack traces/messages in dmesg during driver unload –
[2565601.054366] Call Trace: [2565601.054368] blk_mq_free_map_and_requests+0x28/0x50 [2565601.054369] blk_mq_free_tag_set+0x1d/0x90 [2565601.054370] scsi_host_dev_release+0x8a/0xf0 [2565601.054370] device_release+0x27/0x80 [2565601.054371] kobject_cleanup+0x61/0x190 [2565601.054373] megasas_detach_one+0x4c1/0x650 [megaraid_sas] [2565601.054374] pci_device_remove+0x3b/0xc0 [2565601.054375] device_release_driver_internal+0xec/0x1b0 [2565601.054376] driver_detach+0x46/0x90 [2565601.054377] bus_remove_driver+0x58/0xd0 [2565601.054378] pci_unregister_driver+0x26/0xa0 [2565601.054379] megasas_exit+0x91/0x882 [megaraid_sas] [2565601.054381] __x64_sys_delete_module+0x16c/0x250 [2565601.054382] do_syscall_64+0x5b/0x1b0 [2565601.054383] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [2565601.054383] RIP: 0033:0x7f7212a82837 [2565601.054384] RSP: 002b:00007ffdfa2dcea8 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0 [2565601.054385] RAX: ffffffffffffffda RBX: 0000000000b6e2e0 RCX: 00007f7212a82837 [2565601.054385] RDX: 00007f7212af3ac0 RSI: 0000000000000800 RDI:
Quoted 8 lines
0000000000b6e348[2565601.054386] RBP: 0000000000000000 R08: 00007f7212d47060 R09: 00007f7212af3ac0 [2565601.054386] R10: 00007ffdfa2dcbc0 R11: 0000000000000202 R12: 00007ffdfa2dd71c [2565601.054387] R13: 0000000000000000 R14: 0000000000b6e2e0 R15: 0000000000b6e010 [2565601.054387] ---[ end trace 38899303bd85e838 ]---
I see it also for hisi_sas_v3_hw.
And so I don't understand the code change here, specifically where the WARN is generated:
void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,
unsigned int hctx_idx)
{
struct page *page;
int i;
if (tags->rqs) {
for (i = 0; i < tags->nr_tags; i++)
if (WARN_ON(tags->rqs[i]))
tags->rqs[i] = NULL; <--- here
}I thought that tags->rqs[i] was just a holder for a pointer to a static tag, like assigned here:
static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data, unsigned int tag, unsigned int op, u64 alloc_time_ns)
{
struct blk_mq_tags *tags = blk_mq_tags_from_data(data);
struct request *rq = tags->static_rqs[tag];
...
rq->tag = tag;
rq->internal_tag = -1;
data->hctx->tags->rqs[rq->tag] = rq;
...
}So I don't know why we need to WARN if unset, and then also clear it. The memory is freed pretty soon after this anyway.
Thanks,
John
Quoted 74 lines
Quoted 64 lines
Quoted 48 lines
lspcu:
# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 72 On-line CPU(s) list: 0-71 Thread(s) per core: 2 Core(s) per socket: 18 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 85 Model name: Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz Stepping: 4 CPU MHz: 3204.246 CPU max MHz: 3700.0000 CPU min MHz: 1200.0000 BogoMIPS: 5400.00 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 1024K L3 cache: 25344K NUMA node0 CPU(s): 0-17,36-53 NUMA node1 CPU(s): 18-35,54-71 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe s yscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin mba tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_lo[snip]
Quoted 4 lines
4. This patch removes below code from driver so what this piece of code does is broken-
Quoted 8 lines
- if (
instance->adapter_type>=INVADER_SERIES&&
- !instance->msix_combined) { -instance->msix_load_balance= true;
-instance->smp_affinity_enable= false;
- }Quoted 2 lines
Does this code need to be re-added? Would this have affected your test? Primarily this patch was required to enable interrupt affinity on my
machine (Lenovo RAID 930-8i).
Can you give me some information why the code is present in the first place? Some hardware limitation, maybe?Cheers,
Hannes
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 108 lines · John Garry
Quoted 31 lines
Quoted 31 lines
Quoted 11 lines
Hi Hannes,
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.Hi Sumit,
So the new code has:
megasas_build_ldio_fusion()
{
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag);}
So the value here is hw queue index from blk-mq point of view, and not
megaraid_sasmsix index, as you alluded to.So we get 80 msix, 8 are reserved for
low_latency_index_start(that's how it seems to me), and we report other 72 as #hw queues = 72 to SCSI midlayer.Quoted 9 lines
Quoted 6 lines
So I think that this should be:
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag)+low_latency_index_start;Indeed, that sounds reasonable.
(The whole queue mapping stuff isn't exactly well documented :-( )Yeah, there's certainly lots of knobs and levers in this driver.
Quoted 1 line
I'll be updating the patch.
About this one:
Quoted 24 lines
2. Seeing below stack traces/messages in dmesg during driver unload –
[2565601.054366] Call Trace: [2565601.054368] blk_mq_free_map_and_requests+0x28/0x50 [2565601.054369] blk_mq_free_tag_set+0x1d/0x90 [2565601.054370] scsi_host_dev_release+0x8a/0xf0 [2565601.054370] device_release+0x27/0x80 [2565601.054371] kobject_cleanup+0x61/0x190 [2565601.054373] megasas_detach_one+0x4c1/0x650 [megaraid_sas] [2565601.054374] pci_device_remove+0x3b/0xc0 [2565601.054375] device_release_driver_internal+0xec/0x1b0 [2565601.054376] driver_detach+0x46/0x90 [2565601.054377] bus_remove_driver+0x58/0xd0 [2565601.054378] pci_unregister_driver+0x26/0xa0 [2565601.054379] megasas_exit+0x91/0x882 [megaraid_sas] [2565601.054381] __x64_sys_delete_module+0x16c/0x250 [2565601.054382] do_syscall_64+0x5b/0x1b0 [2565601.054383] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [2565601.054383] RIP: 0033:0x7f7212a82837 [2565601.054384] RSP: 002b:00007ffdfa2dcea8 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0 [2565601.054385] RAX: ffffffffffffffda RBX: 0000000000b6e2e0 RCX: 00007f7212a82837 [2565601.054385] RDX: 00007f7212af3ac0 RSI: 0000000000000800 RDI:Quoted 8 lines
0000000000b6e348[2565601.054386] RBP: 0000000000000000 R08: 00007f7212d47060 R09: 00007f7212af3ac0 [2565601.054386] R10: 00007ffdfa2dcbc0 R11: 0000000000000202 R12: 00007ffdfa2dd71c [2565601.054387] R13: 0000000000000000 R14: 0000000000b6e2e0 R15: 0000000000b6e010 [2565601.054387] ---[ end trace 38899303bd85e838 ]---I see it also for
hisi_sas_v3_hw.And so I don't understand the code change here, specifically where the WARN is generated:
void blk_mq_free_rqs(struct
blk_mq_tag_set*set, structblk_mq_tags*tags,
unsigned inthctx_idx)
{
struct page *page;
int i;if (
tags->rqs) {
for (i = 0; i <tags->nr_tags; i++)
if (WARN_ON(tags->rqs[i]))
tags->rqs[i] = NULL; <--- here
}I thought that tags->rqs[i] was just a holder for a pointer to a static tag, like assigned here:
static struct request *blk_mq_rq_ctx_init(struct
blk_mq_alloc_data*data, unsigned int tag, unsigned int op, u64alloc_time_ns)
{
structblk_mq_tags*tags =blk_mq_tags_from_data(data); struct request *rq = tags->static_rqs[tag];...
Quoted 9 lines
rq->tag = tag;
rq->internal_tag = -1;
data->hctx->tags->rqs[rq->tag] = rq;...
}So I don't know why we need to WARN if unset, and then also clear it. The memory is freed pretty soon after this anyway.
Indeed, ->rqs is a holder, referencing an entry in ->static_rqs.
Point here is that ->rqs is set when allocating a request, and should be zeroed when freeing the request.
And then this above patch would warn us if there's an imbalance, ie an allocated request didn't get freed.
But apparently the latter part didn't happen, leaving us with stale entries in ->rqs.
Either we fix that, or we drop the WARN_ON.
Personally I like clearing of the ->rqs pointer (as then it's easier to track use-after-free issues), but then this might have performance implications, and Jens might have some views about it. So I'm fine with dropping it.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Teamlead Storage & Networking [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Sumit Saxena
maintainerRe: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
In reply to Hannes ReineckeQuoted 138 lines · Hannes Reinecke
Quoted 58 lines · John Garry
Quoted 34 lines · Sumit Saxena
Quoted 21 lines · Hannes Reinecke
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you. Thank you.
I'm aware of the results of Ming Leis work, but I do hope this patchset performs better.
Quoted 9 lines
And when you do performance measurements, can you please run with both, 'none' I/O scheduler and 'mq-deadline' I/O scheduler?
I've measured quite a performance improvements when using mq-deadline, up to the point where I've gotten on-par performance with the original, non-mq, implementation.
(As a data point, on my setup I've measured about 270k IOPS and 1092 MB/s througput, running on just 2 SSDs).asas_build_ldio_fusion
But thanks for doing a performance test here.Hi Hannes,
Quoted 7 lines
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.Hi Sumit,
So the new code has:
megasas_build_ldio_fusion()
{
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag);}
So the value here is hw queue index from blk-mq point of view, and not
megaraid_sasmsix index, as you alluded to.Quoted 10 lines
So we get 80 msix, 8 are reserved for
low_latency_index_start(that's how it seems to me), and we report other 72 as #hw queues = 72 to SCSI midlayer.So I think that this should be:
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag)+low_latency_index_start;Indeed, that sounds reasonable.
(The whole queue mapping stuff isn't exactly well documented :-( )I'll be updating the patch.
Quoted 63 lines
Quoted 47 lines
lspcu:
# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 72 On-line CPU(s) list: 0-71 Thread(s) per core: 2 Core(s) per socket: 18 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 85 Model name: Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz Stepping: 4 CPU MHz: 3204.246 CPU max MHz: 3700.0000 CPU min MHz: 1200.0000 BogoMIPS: 5400.00 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 1024K L3 cache: 25344K NUMA node0 CPU(s): 0-17,36-53 NUMA node1 CPU(s): 18-35,54-71 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe s yscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin mba tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_lo[snip]
Quoted 12 lines
4. This patch removes below code from driver so what this piece of code does is broken-
- if (
instance->adapter_type>=INVADER_SERIES&&(quoted diff) +0
-2- !instance->msix_combined) { - instance->msix_load_balance =true;
-instance->smp_affinity_enable
= false;
- }
Quoted 5 lines
Quoted 2 lines
Does this code need to be re-added? Would this have affected your test? Primarily this patch was required to enable interrupt affinity on my
machine (Lenovo RAID 930-8i).
Can you give me some information why the code is present in the first place? Some hardware limitation, maybe?
Hi Hannes,
This code is for specific family of megaraid_sas adapters and Lenovo RAID 930-8i belongs to them. For those adapters, we want to use available HW queues in round robin fashion instead of using interrupt affinity. It helps to improve performance and fixes soft lockups. For interrupt affinity test purpose on Lenovo RAID 930-8i, you can disable this code to use affinity. But in the final patch, this code has to be present. This code was added through below commit:
commit 1d15d9098ad12b0021ac5a6b851f26d1ab021e5a
Author: Shivasharan S <[email protected]> Date: Tue May 7 10:05:36 2019 -0700
scsi: megaraid_sas: Load balance completions across all MSI-X
Driver will use "reply descriptor post queues" in round robin fashion when the combined MSI-X mode is not enabled. With this IO completions are distributed and load balanced across all the available reply descriptor post queues equally.
This is enabled only if combined MSI-X mode is not enabled in firmware. This improves performance and also fixes soft lockups.
When load balancing is enabled, IRQ affinity from driver needs to be disabled.
Thanks,
Sumit
Quoted 9 lines
Cheers,
Hannes
--
Dr. Hannes Reinecke Teamlead Storage & Networking
[email protected] +49 911 74053 688
SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
Sumit Saxena
maintainerRe: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
In reply to John GarryQuoted 50 lines · John Garry
Quoted 32 lines · Sumit Saxena
Quoted 21 lines · Hannes Reinecke
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you. Thank you.
I'm aware of the results of Ming Leis work, but I do hope this patchset performs better.
Quoted 9 lines
And when you do performance measurements, can you please run with both, 'none' I/O scheduler and 'mq-deadline' I/O scheduler?
I've measured quite a performance improvements when using mq-deadline, up to the point where I've gotten on-par performance with the original, non-mq, implementation.
(As a data point, on my setup I've measured about 270k IOPS and 1092 MB/s througput, running on just 2 SSDs).asas_build_ldio_fusion
But thanks for doing a performance test here.Quoted 8 lines
Hi Hannes,
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.Hi Sumit,
So the new code has:
megasas_build_ldio_fusion()
{
Quoted 7 lines
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag);}
So the value here is hw queue index from blk-mq point of view, and not
megaraid_sasmsix index, as you alluded to.
Yes John, value filled in "cmd->request_desc->SCSIIO.MSIxIndex" is HW queue index.
Quoted 9 lines
So we get 80 msix, 8 are reserved for
low_latency_index_start(that's how it seems to me), and we report other 72 as #hw queues = 72 to SCSI midlayer.So I think that this should be:
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag)+low_latency_index_start;
Agreed, this should return correct HW queue index.
Quoted 56 lines
Quoted 49 lines
lspcu:
# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 72 On-line CPU(s) list: 0-71 Thread(s) per core: 2 Core(s) per socket: 18 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 85 Model name: Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz Stepping: 4 CPU MHz: 3204.246 CPU max MHz: 3700.0000 CPU min MHz: 1200.0000 BogoMIPS: 5400.00 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 1024K L3 cache: 25344K NUMA node0 CPU(s): 0-17,36-53 NUMA node1 CPU(s): 18-35,54-71 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe s yscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin mba tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_lo[snip]
Quoted 4 lines
4. This patch removes below code from driver so what this piece of code does is broken-
Quoted 7 lines
drivers/scsi/megaraid/megaraid_sas_base.c static int megasas_init_fw(struct megasas_instance *instance) · line 6028
Quoted 5 lines
drivers/scsi/megaraid/megaraid_sas_base.c +0
-5-if (instance->adapter_type >=INVADER_SERIES && - !instance->msix_combined) { - instance->msix_load_balance = true; - instance->smp_affinity_enable = false; - }Does this code need to be re-added? Would this have affected your test?
This code did not affect my test but has to be re-added for affected hardware. There are few megaraid_sas adapters for which "instance->msix_combined" would be "0" and we need this code for those adapters.
Thanks,
Sumit
Quoted 3 lines
Thanks,
John
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 3 lines · Sumit Saxena
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
or doing a performance test here.
Hi Hannes,
Hi Sumit,
Quoted 16 lines
Quoted 16 lines
Quoted 6 lines
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.Hi Sumit,
So the new code has:
megasas_build_ldio_fusion()
{
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag);
Quoted 15 lines
Quoted 4 lines
}
So the value here is hw queue index from blk-mq point of view, and not
megaraid_sasmsix index, as you alluded to.Yes John, value filled in "
cmd->request_desc->SCSIIO.MSIxIndex" is HW queue index.Quoted 8 lines
So we get 80 msix, 8 are reserved for
low_latency_index_start(that's how it seems to me), and we report other 72 as #hw queues = 72 to SCSI midlayer.So I think that this should be:
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag)+low_latency_index_start;
Can you possibly test performance again with this local change, or would you rather an updated patchset be sent?
Quoted 2 lines
Agreed, this should return correct HW queue index.
Quoted 1 line
Thanks,
John
Sumit Saxena
maintainerRe: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 35 lines · John Garry
Quoted 3 lines · Sumit Saxena
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
or doing a performance test here.
Hi Hannes,
Hi Sumit,
Quoted 27 lines
Quoted 13 lines
Quoted 6 lines
Sorry for the delay in replying, I observed a few issues with this patchset:
1. "
blk_mq_unique_tag_to_hwq(tag)" does not return MSI-x vector to which IO submitter CPU is affined with. Due to this IO submission and completion CPUs are different which causes performance drop for low latency workloads.Hi Sumit,
So the new code has:
megasas_build_ldio_fusion()
{Quoted 7 lines
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag);}
So the value here is hw queue index from blk-mq point of view, and not
megaraid_sasmsix index, as you alluded to.Yes John, value filled in "
cmd->request_desc->SCSIIO.MSIxIndex" is HW queue index.Quoted 4 lines
So we get 80 msix, 8 are reserved for
low_latency_index_start(that's how it seems to me), and we report other 72 as #hw queues = 72 to SCSI midlayer.
Quoted 7 lines
Quoted 4 lines
Quoted 4 lines
So I think that this should be:
cmd->request_desc->SCSIIO.MSIxIndex=blk_mq_unique_tag_to_hwq(tag)+low_latency_index_start;Can you possibly test performance again with this local change, or would you rather an updated patchset be sent?
Updated patchset is not required. I will do performance run with this change and update.
Thanks,
Sumit
Quoted 7 lines
Quoted 3 lines
Agreed, this should return correct HW queue index.
Quoted 1 line
Thanks,
John
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
In reply to Sumit SaxenaQuoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.
Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you.
Hi Sumit,
I was wondering if you had a chance to do this test yet?
It would be good to know, so we can try to progress this work.
@Hannes, This shared sbitmap work now seems to conflict with Jens work on tag caching https://lore.kernel.org/linux-block/[email protected]/T/#t, but should be resolvable AFAICS (v1, anyway, which I checked). Anway, we seem to have stalled, which I feared...
Thanks,
John
Quoted 243 lines
Thanks,
SumitQuoted 237 lines
Signed-off-by Hannes Reinecke <[email protected]>--- drivers/scsi/megaraid/megaraid_sas.h | 1 - drivers/scsi/megaraid/megaraid_sas_base.c | 65 +++++++++-------------------- drivers/scsi/megaraid/megaraid_sas_fusion.c | 14 ++++--- 3 files changed, 28 insertions(+), 52 deletions(-)drivers/scsi/megaraid/megaraid_sas.h +0
-1@@ -2261,7 +2261,6 @@ enum MR_PERF_MODE {struct megasas_instance { -unsigned int *reply_map;__le32 *producer;dma_addr_t producer_h;__le32 *consumer;drivers/scsi/megaraid/megaraid_sas_base.c +19
-46@@ -37,6 +37,7 @@#include <linux/poll.h> #include <linux/vmalloc.h> #include <linux/irq_poll.h> +#include <linux/blk-mq-pci.h> #include <scsi/scsi.h> #include <scsi/scsi_cmnd.h> @@ -3106,6 +3107,19 @@ megasas_bios_param(struct scsi_device *sdev, struct block_device *bdev,return 0 ; } +static int megasas_map_queues (struct Scsi_Host *shost) +{ +struct megasas_instance *instance; + + instance = (struct megasas_instance *)shost->hostdata; + +if (!instance->smp_affinity_enable) +return 0 ; + +return blk_mq_pci_map_queues (&shost->tag_set.map[HCTX_TYPE_DEFAULT ], + instance->pdev, instance->low_latency_index_start); +} +static void megasas_aen_polling (struct work_struct *work);/** @@ -3414,9 +3428,11 @@ static struct scsi_host_template megasas_template = { .eh_timed_out = megasas_reset_timer, .shost_attrs = megaraid_host_attrs, .bios_param = megasas_bios_param, + .map_queues = megasas_map_queues, .change_queue_depth = scsi_change_queue_depth, .max_segment_size = 0xffffffff, .no_write_same = 1, + .host_tagset = 1, };/** @@ -5695,34 +5711,6 @@ megasas_setup_jbod_map(struct megasas_instance *instance) instance->use_seqnum_jbod_fp = false; } -static void megasas_setup_reply_map (struct megasas_instance *instance) -{ -const struct cpumask *mask; -unsigned int queue, cpu, low_latency_index_start; - - low_latency_index_start = instance->low_latency_index_start; - -for (queue = low_latency_index_start; queue < instance->msix_vectors; queue++) { - mask =pci_irq_get_affinity (instance->pdev, queue); -if (!mask) - goto fallback; - -for_each_cpu (cpu, mask) - instance->reply_map[cpu] = queue; - } -return ; - -fallback: - queue = low_latency_index_start; -for_each_possible_cpu (cpu) { - instance->reply_map[cpu] = queue; -if (queue == (instance->msix_vectors -1 )) - queue = low_latency_index_start; -else - queue++; - } -} -/** * megasas_get_device_list - Get the PD and LD device list from FW. * @instance: Adapter soft state @@ -6021,12 +6009,6 @@ static int megasas_init_fw(struct megasas_instance *instance) instance->is_rdpq = (scratch_pad_1 &MR_RDPQ_MODE_OFFSET ) ?1 :0 ; -if (instance->adapter_type >=INVADER_SERIES && - !instance->msix_combined) { - instance->msix_load_balance = true; - instance->smp_affinity_enable = false; - } -/* Save 1-15 reply post index address to local memory * Index 0 is already saved from reg offset * MPI2_REPLY_POST_HOST_INDEX_OFFSET @@ -6145,8 +6127,6 @@ static int megasas_init_fw(struct megasas_instance *instance) goto fail_init_adapter; } -megasas_setup_reply_map (instance); -dev_info (&instance->pdev->dev,"current msix/online cpus\t: (%d/%d)\n" , instance->msix_vectors, (unsigned int )num_online_cpus ());@@ -6780,6 +6760,9 @@ static int megasas_io_attach(struct megasas_instance *instance) host->max_id =MEGASAS_MAX_DEV_PER_CHANNEL ; host->max_lun =MEGASAS_MAX_LUN ; host->max_cmd_len =16 ; +if (instance->adapter_type !=MFI_SERIES && instance->msix_vectors >0 ) + host->nr_hw_queues = instance->msix_vectors - + instance->low_latency_index_start;/* * Notify the mid-layer about the new controller @@ -6947,11 +6930,6 @@ static inline int megasas_alloc_mfi_ctrl_mem(struct megasas_instance *instance)*/ static int megasas_alloc_ctrl_mem (struct megasas_instance *instance) { - instance->reply_map =kcalloc (nr_cpu_ids,sizeof (unsigned int ), -GFP_KERNEL ); -if (!instance->reply_map) -return -ENOMEM ; -switch (instance->adapter_type) {case MFI_SERIES :if (megasas_alloc_mfi_ctrl_mem (instance))@@ -6968,8 +6946,6 @@ static int megasas_alloc_ctrl_mem(struct megasas_instance *instance)return 0 ; fail: -kfree (instance->reply_map); - instance->reply_map =NULL ;return -ENOMEM ; }@@ -6982,7 +6958,6 @@ static int megasas_alloc_ctrl_mem(struct megasas_instance *instance)*/ static inline void megasas_free_ctrl_mem (struct megasas_instance *instance) { -kfree (instance->reply_map);if (instance->adapter_type ==MFI_SERIES ) {if (instance->producer) dma_free_coherent(&instance->pdev->dev, sizeof(u32),@@ -7645,8 +7620,6 @@ megasas_resume(struct pci_dev *pdev)if (rval <0 ) goto fail_reenable_msix; -megasas_setup_reply_map (instance); -if (instance->adapter_type !=MFI_SERIES ) {megasas_reset_reply_desc (instance);if (megasas_ioc_init_fusion (instance)) {drivers/scsi/megaraid/megaraid_sas_fusion.c +9
-6@@ -2731,6 +2731,7 @@ megasas_build_ldio_fusion(struct megasas_instance *instance,struct MR_PRIV_DEVICE *mrdev_priv;struct RAID_CONTEXT *rctx;struct RAID_CONTEXT_G35 *rctx_g35; +u32 tag =blk_mq_unique_tag (scp->request); device_id =MEGASAS_DEV_INDEX (scp);@@ -2837,7 +2838,7 @@ megasas_build_ldio_fusion(struct megasas_instance *instance, instance->msix_vectors)); else cmd->request_desc->SCSIIO.MSIxIndex = - instance->reply_map[raw_smp_processor_id ()]; +blk_mq_unique_tag_to_hwq (tag);if (instance->adapter_type >=VENTURA_SERIES ) {/* FP for Optimal raid level 1. @@ -3080,6 +3081,7 @@ megasas_build_syspd_fusion(struct megasas_instance *instance,u16 pd_index =0 ;u16 os_timeout_value;u16 timeout_limit; +u32 tag =blk_mq_unique_tag (scmd->request);struct MR_DRV_RAID_MAP_ALL *local_map_ptr;struct RAID_CONTEXT *pRAID_Context;struct MR_PD_CFG_SEQ_NUM_SYNC *pd_sync;@@ -3169,7 +3171,7 @@ megasas_build_syspd_fusion(struct megasas_instance *instance, instance->msix_vectors)); else cmd->request_desc->SCSIIO.MSIxIndex = - instance->reply_map[raw_smp_processor_id ()]; +blk_mq_unique_tag_to_hwq (tag);if (!fp_possible) {/* system pd firmware path */ @@ -3373,7 +3375,7 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance, {struct megasas_cmd_fusion *cmd, *r1_cmd =NULL ;union MEGASAS_REQUEST_DESCRIPTOR_UNION *req_desc; -u32 index; +u32 index, blk_tag, unique_tag; if ((megasas_cmd_type(scmd) == READ_WRITE_LDIO) && instance->ldio_threshold &&@@ -3389,7 +3391,9 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance,return SCSI_MLQUEUE_HOST_BUSY ; } - cmd =megasas_get_cmd_fusion (instance, scmd->request->tag); + unique_tag =blk_mq_unique_tag (scmd->request); + blk_tag =blk_mq_unique_tag_to_tag (unique_tag); + cmd =megasas_get_cmd_fusion (instance, blk_tag);if (!cmd) {atomic_dec (&instance->fw_outstanding);@@ -3430,7 +3434,7 @@ megasas_build_and_issue_cmd_fusion(struct megasas_instance *instance,*/ if (cmd->r1_alt_dev_handle !=MR_DEVHANDLE_INVALID ) { r1_cmd =megasas_get_cmd_fusion (instance, - (scmd->request->tag + instance->max_fw_cmds)); + (blk_tag + instance->max_fw_cmds));megasas_prepare_secondRaid1_IO (instance, cmd, r1_cmd); } --Quoted 2 lines
2.16.4
.
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 21 lines · John Garry
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you.
Hi Sumit,
I was wondering if you had a chance to do this test yet?
Quoted 8 lines
It would be good to know, so we can try to progress this work.
@Hannes, This shared sbitmap work now seems to conflict with Jens work on tag caching
https://lore.kernel.org/linux-block/[email protected]/T/#t,
but should be resolvable AFAICS (v1, anyway, which I checked). Anway, we seem to have stalled, which I feared...
Thanks for the reminder.
That was a topic I was wanting to discuss at LSF; will be sending a topic then.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Teamlead Storage & Networking [email protected] +49 911 74053 688 SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), GF: Felix Imendörffer
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 30 lines · Hannes Reinecke
Quoted 21 lines · John Garry
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you.
Hi Sumit,
I was wondering if you had a chance to do this test yet?
Quoted 8 lines
It would be good to know, so we can try to progress this work.
@Hannes, This shared sbitmap work now seems to conflict with Jens work on tag caching
https://lore.kernel.org/linux-block/[email protected]/T/#t,
but should be resolvable AFAICS (v1, anyway, which I checked). Anway, we seem to have stalled, which I feared...
Quoted 3 lines
Thanks for the reminder.
That was a topic I was wanting to discuss at LSF; will be sending a topic then.
Alright, but I am not really sure what else we need to wait months to discuss, unless this shared sbitmap approach is rejected and/or testing on other HBAs shows unsatisfactory performance.
To summarize, as I see, we have 3 topics to tackle:
- shared tags
- block hotplug improvement
- Ming Lei had said that he will post another version of
https://lore.kernel.org/linux-block/[email protected]/
- https://patchwork.kernel.org/cover/10967071/
- I'm not sure what's happening on that, but thought that it was
somewhat straightforward.
If there's something else I've missed, then let me know.
Cheers,
John
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
In reply to John GarryQuoted 14 lines · John Garry
Quoted 13 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/
Quoted 8 lines
Quoted 2 lines
So, we will do some performance tests with this patch and update you.
Hi Sumit,
I was wondering if you had a chance to do this test yet?
It would be good to know, so we can try to progress this work.
Looks most of the comment in the following link isn't addressed:
https://lore.kernel.org/linux-block/[email protected]/
Quoted 4 lines
Firstly too
much((nr_hw_queues - 1) times)memory is wasted. Secondly IO latency could be increased by too deep scheduler queue depth. Finally CPU could be wasted in the retrying of running busy hw queue.
Quoted 8 lines
Wrt. driver tags, this patch may be worse, given the average limit for each LUN is reduced by (
nr_hw_queues) times, seehctx_may_queue().Another change is
bt_wait_ptr(). Before your patches, there is single .wait_index, now the number of .wait_index is changed tonr_hw_queues.Also the run queue number is increased a lot in SCSI's IO completion, see
scsi_end_request().
I guess memory waste won't be a blocker.
But it may not be one accepted behavior to reduce average active queue depth for each LUN by nr_hw_queues times, meantime scheduler queue depth is increased by nr_hw_queues times, compared with single queue.
thanks,
Ming
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 20 lines · Ming Lei
Quoted 19 lines · John Garry
Quoted 7 lines · Sumit Saxena
Quoted 5 lines · Hannes Reinecke
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you.
Hi Sumit,
Quoted 7 lines
Quoted 3 lines
I was wondering if you had a chance to do this test yet?
It would be good to know, so we can try to progress this work.
Looks most of the comment in the following link isn't addressed:
OK, but I was waiting for results first, which I hoped would not take too long. They weren't forgotten, for sure. Let me check them now.
Quoted 8 lines
Quoted 8 lines
Firstly too
much((nr_hw_queues - 1) times)memory is wasted. Secondly IO latency could be increased by too deep scheduler queue depth. Finally CPU could be wasted in the retrying of running busy hw queue.Wrt. driver tags, this patch may be worse, given the average limit for each LUN is reduced by (
nr_hw_queues) times, seehctx_may_queue().
Quoted 7 lines
Quoted 5 lines
Another change is
bt_wait_ptr(). Before your patches, there is single .wait_index, now the number of .wait_index is changed tonr_hw_queues.Also the run queue number is increased a lot in SCSI's IO completion, see
scsi_end_request().I guess memory waste won't be a blocker.
Yeah, that's a trade-off. And, as I remember, memory waste does not seem extreme.
Quoted 5 lines
But it may not be one accepted behavior to reduce average active queue depth for each LUN by
nr_hw_queuestimes, meantime scheduler queue depth is increased bynr_hw_queuestimes, compared with single queue.
Thanks,
John
Re: [PATCH 09/11] megaraid_sas: switch fusion adapters to MQ
Quoted 19 lines · John Garry
Quoted 19 lines
Quoted 19 lines
Quoted 6 lines
Quoted 5 lines
Fusion adapters can steer completions to individual queues, and we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters and drop the hand-crafted interrupt affinity settings.Quoted 8 lines
Hi Hannes,
Ming Lei also proposed similar changes in
megaraid_sasdriver some time back and it had resulted in performance drop-
https://patchwork.kernel.org/patch/10969511/So, we will do some performance tests with this patch and update you.
Hi Sumit,
I was wondering if you had a chance to do this test yet?
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
It would be good to know, so we can try to progress this work.
Looks most of the comment in the following link isn't addressed:
OK, but I was waiting for results first, which I hoped would not take too long. They weren't forgotten, for sure. Let me check them now.
Hi Ming,
I think that your questions here were related to the shared scheduler tags, which was Hannes' proposal (I initially had it in v2 series, but dropped it for v3).
I was just content to maintain the concept of shared driver tags.
Thanks,
John
Quoted 27 lines
Quoted 9 lines
Quoted 9 lines
Firstly too
much((nr_hw_queues - 1) times)memory is wasted. Secondly IO latency could be increased by too deep scheduler queue depth. Finally CPU could be wasted in the retrying of running busy hw queue.Wrt. driver tags, this patch may be worse, given the average limit for each LUN is reduced by (
nr_hw_queues) times, seehctx_may_queue().Quoted 7 lines
Quoted 5 lines
Another change is
bt_wait_ptr(). Before your patches, there is single .wait_index, now the number of .wait_index is changed tonr_hw_queues.Also the run queue number is increased a lot in SCSI's IO completion, see
scsi_end_request().I guess memory waste won't be a blocker.
Yeah, that's a trade-off. And, as I remember, memory waste does not seem extreme.
Quoted 5 lines
But it may not be one accepted behavior to reduce average active queue depth for each LUN by
nr_hw_queuestimes, meantime scheduler queue depth is increased bynr_hw_queuestimes, compared with single queue.Thanks,
John
PATCH v5 10/11 smartpqi: enable host tagset
In reply to Hannes ReineckeEnable host tagset for smartpqi; with this we can use the request tag to look command from the pool avoiding the list iteration in the hot path.
--- drivers/scsi/smartpqi/smartpqi_init.c | 38 ++++++++++++++++++++++++++--------- 1 file changed, 28 insertions(+), 10 deletions(-)
drivers/scsi/smartpqi/smartpqi_init.c +28 -10
@@ -575,17 +575,29 @@ static inline void pqi_reinit_io_request(struct pqi_io_request *io_request) }static struct pqi_io_request *pqi_alloc_io_request ( -struct pqi_ctrl_info *ctrl_info) +struct pqi_ctrl_info *ctrl_info,struct scsi_cmnd *scmd) {struct pqi_io_request *io_request; +unsigned int limit =PQI_RESERVED_IO_SLOTS ;u16 i = ctrl_info->next_io_request_slot;/* benignly racy */ -while (1 ) { +if (scmd) { +u32 blk_tag =blk_mq_unique_tag (scmd->request); + + i =blk_mq_unique_tag_to_tag (blk_tag) + limit; io_request = &ctrl_info->io_request_pool[i]; -if (atomic_inc_return (&io_request->refcount) ==1 ) -break ; -atomic_dec (&io_request->refcount); - i = (i +1 ) % ctrl_info->max_io_slots; +if (WARN_ON (atomic_inc_return (&io_request->refcount) >1 )) { +atomic_dec (&io_request->refcount); +return NULL ; + } + }else { +while (1 ) { + io_request = &ctrl_info->io_request_pool[i]; +if (atomic_inc_return (&io_request->refcount) ==1 ) +break ; +atomic_dec (&io_request->refcount); + i = (i +1 ) % limit; + } }/* benignly racy */
@@ -4075,7 +4087,7 @@ static int pqi_submit_raid_request_synchronous(struct pqi_ctrl_info *ctrl_info,atomic_inc (&ctrl_info->sync_cmds_outstanding); - io_request =pqi_alloc_io_request (ctrl_info); + io_request =pqi_alloc_io_request (ctrl_info,NULL );put_unaligned_le16 (io_request->index, &(((struct pqi_raid_path_request *)request)->request_id));
@@ -5032,7 +5044,9 @@ static inline int pqi_raid_submit_scsi_cmd(struct pqi_ctrl_info *ctrl_info, {struct pqi_io_request *io_request; - io_request =pqi_alloc_io_request (ctrl_info); + io_request =pqi_alloc_io_request (ctrl_info, scmd); +if (!io_request) +return SCSI_MLQUEUE_HOST_BUSY ;return pqi_raid_submit_scsi_cmd_with_io_request (ctrl_info, io_request, device, scmd, queue_group);
@@ -5230,7 +5244,10 @@ static int pqi_aio_submit_io(struct pqi_ctrl_info *ctrl_info,struct pqi_io_request *io_request;struct pqi_aio_path_request *request; - io_request =pqi_alloc_io_request (ctrl_info); + io_request =pqi_alloc_io_request (ctrl_info, scmd); +if (!io_request) +return SCSI_MLQUEUE_HOST_BUSY ; + io_request->io_complete_callback = pqi_aio_io_complete; io_request->scmd = scmd; io_request->raid_bypass = raid_bypass;
@@ -5657,7 +5674,7 @@ static int pqi_lun_reset(struct pqi_ctrl_info *ctrl_info,DECLARE_COMPLETION_ONSTACK (wait);struct pqi_task_management_request *request; - io_request =pqi_alloc_io_request (ctrl_info); + io_request =pqi_alloc_io_request (ctrl_info,NULL ); io_request->io_complete_callback = pqi_lun_reset_complete; io_request->context = &wait;
@@ -6504,6 +6521,7 @@ static struct scsi_host_template pqi_driver_template = { .map_queues = pqi_map_queues, .sdev_attrs = pqi_sdev_attrs, .shost_attrs = pqi_shost_attrs, + .host_tagset = 1, };static int pqi_register_scsi (struct pqi_ctrl_info *ctrl_info)
Signature · 1 line
2.16.4
PATCH v5 11/11 hpsa: enable host_tagset and switch to MQ
In reply to Hannes ReineckeThe smart array HBAs can steer interrupt completion, so this patch switches the implementation to use multiqueue and enables 'host_tagset' as the HBA has a shared host-wide tagset.
--- drivers/scsi/hpsa.c | 44 +++++++------------------------------------- drivers/scsi/hpsa.h | 1 - 2 files changed, 7 insertions(+), 38 deletions(-)
drivers/scsi/hpsa.c +7 -37
@@ -974,6 +974,7 @@ static struct scsi_host_template hpsa_driver_template = { .shost_attrs = hpsa_shost_attrs, .max_sectors = 2048, .no_write_same = 1, + .host_tagset = 1, };static inline u32 next_command (struct ctlr_info *h,u8 q)
@@ -1138,12 +1139,14 @@ static void dial_up_lockup_detection_on_fw_flash_complete(struct ctlr_info *h,static void __enqueue_cmd_and_start_io (struct ctlr_info *h,struct CommandList *c,int reply_queue) { +u32 blk_tag =blk_mq_unique_tag (c->scsi_cmd->request); +dial_down_lockup_detection_during_fw_flash (h, c);atomic_inc (&h->commands_outstanding);if (c->device)atomic_inc (&c->device->commands_outstanding); - reply_queue = h->reply_map[raw_smp_processor_id ()]; + reply_queue =blk_mq_unique_tag_to_hwq (blk_tag);switch (c->cmd_type) {case CMD_IOACCEL1 :set_ioaccel1_performant_mode (h, c, reply_queue);
@@ -5628,8 +5631,6 @@ static int hpsa_scsi_queue_command(struct Scsi_Host *sh, struct scsi_cmnd *cmd)/* Get the ptr to our adapter structure out of cmd->host. */ h =sdev_to_hba (cmd->device); -BUG_ON (cmd->request->tag <0 ); - dev = cmd->device->hostdata;if (!dev) { cmd->result =DID_NO_CONNECT <<16 ;
@@ -5805,7 +5806,7 @@ static int hpsa_scsi_host_alloc(struct ctlr_info *h) sh->hostdata[0 ] = (unsigned long ) h; sh->irq =pci_irq_vector (h->pdev,0 ); sh->unique_id = sh->irq; - + sh->nr_hw_queues = h->msix_vectors >0 ? h->msix_vectors :1 ; h->scsi_host = sh;return 0 ; }
@@ -5831,7 +5832,8 @@ static int hpsa_scsi_add_host(struct ctlr_info *h)*/ static int hpsa_get_cmd_index (struct scsi_cmnd *scmd) { -int idx = scmd->request->tag; +u32 blk_tag =blk_mq_unique_tag (scmd->request); +int idx =blk_mq_unique_tag_to_tag (blk_tag);if (idx <0 )return idx;
@@ -7431,26 +7433,6 @@ static void hpsa_disable_interrupt_mode(struct ctlr_info *h) h->msix_vectors =0 ; } -static void hpsa_setup_reply_map (struct ctlr_info *h) -{ -const struct cpumask *mask; -unsigned int queue, cpu; - -for (queue =0 ; queue < h->msix_vectors; queue++) { - mask =pci_irq_get_affinity (h->pdev, queue); -if (!mask) - goto fallback; - -for_each_cpu (cpu, mask) - h->reply_map[cpu] = queue; - } -return ; - -fallback: -for_each_possible_cpu (cpu) - h->reply_map[cpu] =0 ; -} -/* If MSI/MSI-X is supported by the kernel we will try to enable it on * controllers that are capable. If not, we use legacy INTx mode. */
@@ -7847,9 +7829,6 @@ static int hpsa_pci_init(struct ctlr_info *h)if (err) goto clean1; -/* setup mapping between CPU and reply queue */ -hpsa_setup_reply_map (h); - err =hpsa_pci_find_memory_BAR (h->pdev, &h->paddr);if (err) goto clean2;/* intmode+region, pci */
@@ -8575,7 +8554,6 @@ static struct workqueue_struct *hpsa_create_controller_wq(struct ctlr_info *h,static void hpda_free_ctlr_info (struct ctlr_info *h) { -kfree (h->reply_map);kfree (h); }
@@ -8584,14 +8562,6 @@ static struct ctlr_info *hpda_alloc_ctlr_info(void)struct ctlr_info *h; h =kzalloc (sizeof (*h),GFP_KERNEL ); -if (!h) -return NULL ; - - h->reply_map =kcalloc (nr_cpu_ids,sizeof (*h->reply_map),GFP_KERNEL ); -if (!h->reply_map) { -kfree (h); -return NULL ; - }return h; }
drivers/scsi/hpsa.h +0 -1
@@ -161,7 +161,6 @@ struct bmic_controller_parameters {#pragma pack()struct ctlr_info { -unsigned int *reply_map;int ctlr;char devname[8 ];char *product_name;
Signature · 1 line
2.16.4
Quoted 2 lines · Hannes Reinecke
Hi all,
JFYI, Sumit requested that we rebase this patchset for testing against the latest kernel - it no longer applies for 5.6-rc. I'm going to do that now and repost.
Thanks,
John
Quoted 88 lines
here now is an updated version of the v2 patchset from John Garry, including the suggestions and reviews from the mailing list. John, apologies for hijacking your work :-)
For this version I've only added some slight modifications to Johns original patch (renaming variables etc); the contentious separate sbitmap allocation has been dropped in favour of Johns original version with pointers to the embedded sbitmap.
But more importantly I've reworked the scheduler tag allocation after discussions with Ming Lei.
Point is, hostwide shared tags can't really be resized; they surely cannot be increased (as it's a hardware limitation), and even decreasing is questionable as any modification here would affect all devices served by this HBA.
Scheduler tags, OTOH, can be considered as per-queue, as the I/O scheduler might want to look at all requests on all queues. As such the queue depth is distinct from the actual queue depth of the tagset.
Seeing that it is distinct the depth can now be changed independently of the underlying tagset, and there's no need ever to change the tagset itself.I've also modified
megaraid_sas, smartpqi and hpsa to take advantage ofhost_tags.Performance for
megaraid_sasis on par with the original implementation, with the added benefit that with this we should be able to handle cpu hotplug properly.Differences to v4:
- Rework scheduler tag allocations
- Revert back to the original implementation from JohnDifferences to v3:
- Include reviews from Ming LeiDifferences to v2:
- Drop embedded tag bitmaps
- Do not share scheduling tags
- Add patches for hpsa and smartpqiDifferences to v1:
- Use a shared sbitmap, and not a separate shared tags (a big change!)
- Drop request.shared_tag
- Add RB tagsHannes Reinecke (7):
blk-mq: renameblk_mq_update_tag_set_depth()
blk-mq: addWARN_ONinblk_mq_free_rqs()
blk-mq: move shared sbitmap into elevator queue
scsi: Add template flag 'host_tagset'
megaraid_sas: switch fusion adapters to MQ
smartpqi: enable host tagset
hpsa: enablehost_tagsetand switch to MQJohn Garry (3):
blk-mq: Remove some unused function arguments
blk-mq: Facilitate a shared sbitmap per tagset
scsi:hisi_sas: Switch v3 hw to MQMing Lei (1):
blk-mq: renameBLK_MQ_F_TAG_SHAREDasBLK_MQ_F_TAG_QUEUE_SHAREDblock/bfq-iosched.c | 4 +- block/blk-mq-debugfs.c | 12 +-- block/blk-mq-sched.c | 22 +++++ block/blk-mq-tag.c | 140 +++++++++++++++++++++------- block/blk-mq-tag.h | 27 ++++-- block/blk-mq.c | 104 +++++++++++++++------ block/blk-mq.h | 7 +- block/blk-sysfs.c | 7 ++ block/kyber-iosched.c | 4 +- drivers/scsi/hisi_sas/hisi_sas.h | 3 +- drivers/scsi/hisi_sas/hisi_sas_main.c | 36 +++---- drivers/scsi/hisi_sas/hisi_sas_v3_hw.c | 86 +++++++---------- drivers/scsi/hpsa.c | 44 ++------- drivers/scsi/hpsa.h | 1 - drivers/scsi/megaraid/megaraid_sas.h | 1 - drivers/scsi/megaraid/megaraid_sas_base.c | 65 ++++--------- drivers/scsi/megaraid/megaraid_sas_fusion.c | 14 ++- drivers/scsi/scsi_lib.c | 2 + drivers/scsi/smartpqi/smartpqi_init.c | 38 ++++++-- include/linux/blk-mq.h | 7 +- include/linux/elevator.h | 3 + include/scsi/scsi_host.h | 3 + 22 files changed, 380 insertions(+), 250 deletions(-)