SCSI: introduce per-host admin queue & enable runtime PM
Subsystems (7)
v2 cover SCSI: introduce per-host admin queue & enable runtime PM
Hi,
This patchset introduces per-host admin request queue for submitting admin request only, and uses this approach to implement both SCSI quiesce and runtime PM in one very simply way. Also runtime PM deadlock can be avoided in case that request pool is used up, such as when too many IO requests are allocated before resuming device.
The idea is borrowed from NVMe.
In this patchset, admin request(all requests submitted via __scsi_execute) will be submitted via one per-host admin queue, and the request is still associated with the same scsi_device as before, and respects this scsi_device's all kinds of limits too. Admin queue shares host tags with other IO queues.
One core idea is that for any admin request submitted from this admin queue, this request won't be called back to block layer via the associated IO queue(scsi_device). And this is done in the 3rd patch. So once IO queue is frozen, it can be observed as really frozen from block layer view.
SCSI quiesce is implemented by admin queue in very simple way, see patch 15.
Also runtime PM for legacy path is simplified too, see patch 16, and device resume is moved to blk_queue_enter().
blk-mq simply follows legacy's approach for supporting runtime PM.
Also the fast IO path is simplified much, see blk_queue_enter().
gitweb:
https://github.com/ming1/linux/commits/v4.18-rc-next-scsi_admin_queue_v2
Both runtime PM and system suspend on both legacy & blk-mq have been verified, and not see regression when running blktests.
Any comments are welcome!
Thanks,
Ming
V1->V2:
- convert NO_SCHED to ADMIN flag, don't allocate driver tag budget for admin queue, as pointed by Jianchao(4/17)
- fix one issue in run scsi queue: admin queue shares IO queue depth when sending one command to this scsi_device(10/17)
- fix one race between runtime PM and system suspend(16/17)
- iterate over scheduler tags instead of driver tags for counting allocated requests(17/17)
Ming Lei (17):
blk-mq: allow to pass default queue flags for creating & initializing
queue
blk-mq: convert BLK_MQ_F_NO_SCHED into per-queue flag
block: rename QUEUE_FLAG_NO_SCHED as QUEUE_FLAG_ADMIN
blk-mq: don't reserve tags for admin queue
SCSI: try to retrieve request_queue via 'scsi_cmnd' if possible
SCSI: pass 'scsi_device' instance from 'scsi_request'
SCSI: prepare for introducing admin queue for legacy path
SCSI: pass scsi_device to scsi_mq_prep_fn
SCSI: don't set .queuedata in scsi_mq_alloc_queue()
SCSI: deal with admin queue busy
SCSI: track pending admin commands
SCSI: create admin queue for each host
SCSI: use the dedicated admin queue to send admin commands
SCSI: transport_spi: resume a quiesced device
SCSI: use admin queue to implement queue QUIESCE
block: simplify runtime PM support
block: enable runtime PM for blk-mq
block/blk-core.c | 179 ++++++++++++----------- block/blk-mq-debugfs.c | 3 +- block/blk-mq-tag.c | 27 +++- block/blk-mq-tag.h | 2 + block/blk-mq.c | 44 ++++-- block/elevator.c | 28 +--- drivers/ata/libata-eh.c | 2 +- drivers/block/null_blk_main.c | 7 +- drivers/nvme/host/fc.c | 4 +- drivers/nvme/host/pci.c | 4 +- drivers/nvme/host/rdma.c | 4 +- drivers/nvme/target/loop.c | 4 +- drivers/scsi/hosts.c | 9 ++ drivers/scsi/libsas/sas_ata.c | 2 +- drivers/scsi/libsas/sas_scsi_host.c | 2 +- drivers/scsi/scsi_error.c | 2 +- drivers/scsi/scsi_lib.c | 278 ++++++++++++++++++++++++++---------- drivers/scsi/scsi_priv.h | 1 + drivers/scsi/scsi_scan.c | 1 + drivers/scsi/scsi_sysfs.c | 1 + drivers/scsi/scsi_transport_spi.c | 3 + include/linux/blk-mq.h | 22 ++- include/linux/blkdev.h | 14 +- include/scsi/scsi_device.h | 5 +- include/scsi/scsi_host.h | 2 + include/scsi/scsi_request.h | 5 +- 26 files changed, 423 insertions(+), 232 deletions(-)
Signature · 1 line
2.9.5
RFC PATCH v2 1/17 blk-mq: allow to pass default queue flags for creating & initializing queue
Prepare for converting the flag of BLK_MQ_F_NO_SCHED into per-queue flag, since the following patches need this way for supporting per-host admin queue.
--- block/blk-mq.c | 16 +++++++++------- include/linux/blk-mq.h | 19 ++++++++++++++++--- 2 files changed, 25 insertions(+), 10 deletions(-)
block/blk-mq.c +9 -7
@@ -2483,7 +2483,8 @@ void blk_mq_release(struct request_queue *q)free_percpu (q->queue_ctx); } -struct request_queue *blk_mq_init_queue (struct blk_mq_tag_set *set) +struct request_queue *__blk_mq_init_queue (struct blk_mq_tag_set *set, +unsigned long def_flags) {struct request_queue *uninit_q, *q;
@@ -2491,13 +2492,13 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)if (!uninit_q)return ERR_PTR (-ENOMEM ); - q =blk_mq_init_allocated_queue (set, uninit_q); + q =__blk_mq_init_allocated_queue (set, uninit_q, def_flags);if (IS_ERR (q))blk_cleanup_queue (uninit_q);return q; } -EXPORT_SYMBOL (blk_mq_init_queue); +EXPORT_SYMBOL (__blk_mq_init_queue);static int blk_mq_hw_ctx_size (struct blk_mq_tag_set *tag_set) {
@@ -2571,8 +2572,9 @@ static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set,blk_mq_sysfs_register (q); } -struct request_queue *blk_mq_init_allocated_queue (struct blk_mq_tag_set *set, -struct request_queue *q) +struct request_queue *__blk_mq_init_allocated_queue (struct blk_mq_tag_set *set, +struct request_queue *q, +unsigned long def_flags) {/* mark the queue as mq asap */ q->mq_ops = set->ops;
@@ -2606,7 +2608,7 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, q->nr_queues = nr_cpu_ids; - q->queue_flags |=QUEUE_FLAG_MQ_DEFAULT ; + q->queue_flags |= def_flags;if (!(set->flags &BLK_MQ_F_SG_MERGE ))queue_flag_set_unlocked (QUEUE_FLAG_NO_SG_MERGE , q);
@@ -2656,7 +2658,7 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, q->mq_ops =NULL ;return ERR_PTR (-ENOMEM ); } -EXPORT_SYMBOL (blk_mq_init_allocated_queue); +EXPORT_SYMBOL (__blk_mq_init_allocated_queue);void blk_mq_free_queue (struct request_queue *q) {
include/linux/blk-mq.h +16 -3
@@ -200,9 +200,22 @@ enum { ((policy & ((1 <<BLK_MQ_F_ALLOC_POLICY_BITS ) -1 )) \ << BLK_MQ_F_ALLOC_POLICY_START_BIT) -struct request_queue *blk_mq_init_queue (struct blk_mq_tag_set *); -struct request_queue *blk_mq_init_allocated_queue (struct blk_mq_tag_set *set, -struct request_queue *q); +struct request_queue *__blk_mq_init_queue (struct blk_mq_tag_set *,unsigned long ); +struct request_queue *__blk_mq_init_allocated_queue (struct blk_mq_tag_set *set, +struct request_queue *q, +unsigned long def_flags); + +static inline struct request_queue *blk_mq_init_queue (struct blk_mq_tag_set *set) +{ +return __blk_mq_init_queue (set,QUEUE_FLAG_MQ_DEFAULT ); +} + +static inline struct request_queue * +blk_mq_init_allocated_queue (struct blk_mq_tag_set *set,struct request_queue *q) +{ +return __blk_mq_init_allocated_queue (set, q,QUEUE_FLAG_MQ_DEFAULT ); +} +int blk_mq_register_dev (struct device *,struct request_queue *);void blk_mq_unregister_dev (struct device *,struct request_queue *);
Signature · 1 line
2.9.5
RFC PATCH v2 2/17 blk-mq: convert BLK_MQ_F_NO_SCHED into per-queue flag
In reply to Ming LeiWe need to support admin queue for scsi host, and not like NVMe, this support is only from logic view, and the admin queue still has to share same tags with IO queues.
Convert BLK_MQ_F_NO_SCHED into per-queue flag so that we can support admin queue for SCSI.
--- block/blk-mq-debugfs.c | 2 +- block/blk-mq.c | 2 +- block/elevator.c | 3 +-- drivers/block/null_blk_main.c | 7 ++++--- drivers/nvme/host/fc.c | 4 ++-- drivers/nvme/host/pci.c | 4 ++-- drivers/nvme/host/rdma.c | 4 ++-- drivers/nvme/target/loop.c | 4 ++-- include/linux/blk-mq.h | 1 - include/linux/blkdev.h | 5 +++++ 10 files changed, 20 insertions(+), 16 deletions(-)
block/blk-mq-debugfs.c +1 -1
@@ -133,6 +133,7 @@ static const char *const blk_queue_flag_name[] = {QUEUE_FLAG_NAME (SCSI_PASSTHROUGH ),QUEUE_FLAG_NAME (QUIESCED ),QUEUE_FLAG_NAME (PREEMPT_ONLY ), + QUEUE_FLAG_NAME(NO_SCHED), };#undef QUEUE_FLAG_NAME
@@ -246,7 +247,6 @@ static const char *const hctx_flag_name[] = {HCTX_FLAG_NAME (TAG_SHARED ),HCTX_FLAG_NAME (SG_MERGE ), HCTX_FLAG_NAME(BLOCKING), - HCTX_FLAG_NAME(NO_SCHED), };#undef HCTX_FLAG_NAME
block/blk-mq.c +1 -1
@@ -2640,7 +2640,7 @@ struct request_queue *__blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,blk_mq_add_queue_tag_set (set, q);blk_mq_map_swqueue (q); -if (!(set->flags &BLK_MQ_F_NO_SCHED )) { +if (!blk_queue_no_sched (q)) {int ret; ret =elevator_init_mq (q);
block/elevator.c +1 -2
@@ -1108,8 +1108,7 @@ static int __elevator_change(struct request_queue *q, const char *name)static inline bool elv_support_iosched (struct request_queue *q) { -if (q->mq_ops && q->tag_set && (q->tag_set->flags & -BLK_MQ_F_NO_SCHED )) +if (q->mq_ops &&blk_queue_no_sched (q))return false;return true; }
drivers/block/null_blk_main.c +4 -3
@@ -1617,8 +1617,6 @@ static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set) set->numa_node = nullb ? nullb->dev->home_node : g_home_node; set->cmd_size =sizeof (struct nullb_cmd ); set->flags =BLK_MQ_F_SHOULD_MERGE ; -if (g_no_sched) - set->flags |=BLK_MQ_F_NO_SCHED ; set->driver_data =NULL ;if ((nullb && nullb->dev->blocking) || g_blocking)
@@ -1703,6 +1701,9 @@ static int null_add_dev(struct nullb_device *dev) goto out_free_nullb;if (dev->queue_mode ==NULL_Q_MQ ) { +unsigned long q_flags = g_no_sched ? +QUEUE_FLAG_MQ_NO_SCHED_DEFAULT :QUEUE_FLAG_MQ_DEFAULT ; +if (shared_tags) { nullb->tag_set = &tag_set; rv =0 ;
@@ -1718,7 +1719,7 @@ static int null_add_dev(struct nullb_device *dev) goto out_cleanup_queues; nullb->tag_set->timeout =5 *HZ ; - nullb->q =blk_mq_init_queue (nullb->tag_set); + nullb->q =__blk_mq_init_queue (nullb->tag_set, q_flags);if (IS_ERR (nullb->q)) { rv = -ENOMEM ; goto out_cleanup_tags;
drivers/nvme/host/fc.c +2 -2
@@ -3034,14 +3034,14 @@ nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts, ctrl->admin_tag_set.driver_data = ctrl; ctrl->admin_tag_set.nr_hw_queues =1 ; ctrl->admin_tag_set.timeout =ADMIN_TIMEOUT ; - ctrl->admin_tag_set.flags =BLK_MQ_F_NO_SCHED ; ret =blk_mq_alloc_tag_set (&ctrl->admin_tag_set);if (ret) goto out_free_queues; ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set; - ctrl->ctrl.admin_q =blk_mq_init_queue (&ctrl->admin_tag_set); + ctrl->ctrl.admin_q =__blk_mq_init_queue (&ctrl->admin_tag_set, +QUEUE_FLAG_MQ_NO_SCHED_DEFAULT );if (IS_ERR (ctrl->ctrl.admin_q)) { ret =PTR_ERR (ctrl->ctrl.admin_q); goto out_free_admin_tag_set;
drivers/nvme/host/pci.c +2 -2
@@ -1484,14 +1484,14 @@ static int nvme_alloc_admin_tags(struct nvme_dev *dev) dev->admin_tagset.timeout =ADMIN_TIMEOUT ; dev->admin_tagset.numa_node =dev_to_node (dev->dev); dev->admin_tagset.cmd_size =nvme_pci_cmd_size (dev, false); - dev->admin_tagset.flags =BLK_MQ_F_NO_SCHED ; dev->admin_tagset.driver_data = dev;if (blk_mq_alloc_tag_set (&dev->admin_tagset))return -ENOMEM ; dev->ctrl.admin_tagset = &dev->admin_tagset; - dev->ctrl.admin_q =blk_mq_init_queue (&dev->admin_tagset); + dev->ctrl.admin_q =__blk_mq_init_queue (&dev->admin_tagset, +QUEUE_FLAG_MQ_NO_SCHED_DEFAULT );if (IS_ERR (dev->ctrl.admin_q)) {blk_mq_free_tag_set (&dev->admin_tagset);return -ENOMEM ;
drivers/nvme/host/rdma.c +2 -2
@@ -692,7 +692,6 @@ static struct blk_mq_tag_set *nvme_rdma_alloc_tagset(struct nvme_ctrl *nctrl, set->driver_data = ctrl; set->nr_hw_queues =1 ; set->timeout =ADMIN_TIMEOUT ; - set->flags =BLK_MQ_F_NO_SCHED ; }else { set = &ctrl->tag_set;memset (set,0 ,sizeof (*set));
@@ -770,7 +769,8 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl, goto out_free_async_qe; } - ctrl->ctrl.admin_q =blk_mq_init_queue (&ctrl->admin_tag_set); + ctrl->ctrl.admin_q =__blk_mq_init_queue (&ctrl->admin_tag_set, +QUEUE_FLAG_MQ_NO_SCHED_DEFAULT );if (IS_ERR (ctrl->ctrl.admin_q)) { error =PTR_ERR (ctrl->ctrl.admin_q); goto out_free_tagset;
drivers/nvme/target/loop.c +2 -2
@@ -368,7 +368,6 @@ static int nvme_loop_configure_admin_queue(struct nvme_loop_ctrl *ctrl) ctrl->admin_tag_set.driver_data = ctrl; ctrl->admin_tag_set.nr_hw_queues =1 ; ctrl->admin_tag_set.timeout =ADMIN_TIMEOUT ; - ctrl->admin_tag_set.flags =BLK_MQ_F_NO_SCHED ; ctrl->queues[0 ].ctrl = ctrl; error =nvmet_sq_init (&ctrl->queues[0 ].nvme_sq);
@@ -381,7 +380,8 @@ static int nvme_loop_configure_admin_queue(struct nvme_loop_ctrl *ctrl) goto out_free_sq; ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set; - ctrl->ctrl.admin_q =blk_mq_init_queue (&ctrl->admin_tag_set); + ctrl->ctrl.admin_q =__blk_mq_init_queue (&ctrl->admin_tag_set, +QUEUE_FLAG_MQ_NO_SCHED_DEFAULT );if (IS_ERR (ctrl->ctrl.admin_q)) { error =PTR_ERR (ctrl->ctrl.admin_q); goto out_free_tagset;
include/linux/blk-mq.h +0 -1
@@ -181,7 +181,6 @@ enum {BLK_MQ_F_TAG_SHARED =1 <<1 ,BLK_MQ_F_SG_MERGE =1 <<2 ,BLK_MQ_F_BLOCKING =1 <<5 , -BLK_MQ_F_NO_SCHED =1 <<6 ,BLK_MQ_F_ALLOC_POLICY_START_BIT =8 , BLK_MQ_F_ALLOC_POLICY_BITS = 1,
include/linux/blkdev.h +5 -0
@@ -699,6 +699,7 @@ struct request_queue {#define QUEUE_FLAG_SCSI_PASSTHROUGH 27/* queue supports SCSI commands */ #define QUEUE_FLAG_QUIESCED 28/* queue has been quiesced */ #define QUEUE_FLAG_PREEMPT_ONLY 29/* only process REQ_PREEMPT requests */ +#define QUEUE_FLAG_NO_SCHED 30/* no scheduler allowed */ #define QUEUE_FLAG_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \ (1 << QUEUE_FLAG_SAME_COMP) | \
@@ -708,6 +709,9 @@ struct request_queue { (1 << QUEUE_FLAG_SAME_COMP) | \ (1 << QUEUE_FLAG_POLL)) +#define QUEUE_FLAG_MQ_NO_SCHED_DEFAULT (QUEUE_FLAG_MQ_DEFAULT | \ + (1 << QUEUE_FLAG_NO_SCHED)) +void blk_queue_flag_set (unsigned int flag,struct request_queue *q);void blk_queue_flag_clear (unsigned int flag,struct request_queue *q);bool blk_queue_flag_test_and_set (unsigned int flag,struct request_queue *q);
@@ -739,6 +743,7 @@ bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q);#define blk_queue_preempt_only (q) \ test_bit(QUEUE_FLAG_PREEMPT_ONLY, &(q)->queue_flags)#define blk_queue_fua (q) test_bit(QUEUE_FLAG_FUA, &(q)->queue_flags) +#define blk_queue_no_sched (q) test_bit(QUEUE_FLAG_NO_SCHED, &(q)->queue_flags)extern int blk_set_preempt_only (struct request_queue *q);extern void blk_clear_preempt_only (struct request_queue *q);
Signature · 1 line
2.9.5
RFC PATCH v2 3/17 block: rename QUEUE_FLAG_NO_SCHED as QUEUE_FLAG_ADMIN
In reply to Ming LeiNow all users of QUEUE_FLAG_NO_SCHED is for admin queue only, and not see any drivers need this flag for IO queue.
So rename it as QUEUE_FLAG_ADMIN, which looks more straightforward.
--- block/blk-mq-debugfs.c | 2 +- block/blk-mq.c | 2 +- block/elevator.c | 2 +- drivers/block/null_blk_main.c | 2 +- drivers/nvme/host/fc.c | 2 +- drivers/nvme/host/pci.c | 2 +- drivers/nvme/host/rdma.c | 2 +- drivers/nvme/target/loop.c | 2 +- include/linux/blkdev.h | 8 ++++---- 9 files changed, 12 insertions(+), 12 deletions(-)
block/blk-mq-debugfs.c +1 -1
@@ -133,7 +133,7 @@ static const char *const blk_queue_flag_name[] = {QUEUE_FLAG_NAME (SCSI_PASSTHROUGH ),QUEUE_FLAG_NAME (QUIESCED ),QUEUE_FLAG_NAME (PREEMPT_ONLY ), - QUEUE_FLAG_NAME(NO_SCHED), + QUEUE_FLAG_NAME(ADMIN), };#undef QUEUE_FLAG_NAME
block/blk-mq.c +1 -1
@@ -2640,7 +2640,7 @@ struct request_queue *__blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,blk_mq_add_queue_tag_set (set, q);blk_mq_map_swqueue (q); -if (!blk_queue_no_sched (q)) { +if (!blk_queue_admin (q)) {int ret; ret =elevator_init_mq (q);
block/elevator.c +1 -1
@@ -1108,7 +1108,7 @@ static int __elevator_change(struct request_queue *q, const char *name)static inline bool elv_support_iosched (struct request_queue *q) { -if (q->mq_ops &&blk_queue_no_sched (q)) +if (q->mq_ops &&blk_queue_admin (q))return false;return true; }
drivers/block/null_blk_main.c +1 -1
@@ -1702,7 +1702,7 @@ static int null_add_dev(struct nullb_device *dev)if (dev->queue_mode ==NULL_Q_MQ ) {unsigned long q_flags = g_no_sched ? -QUEUE_FLAG_MQ_NO_SCHED_DEFAULT :QUEUE_FLAG_MQ_DEFAULT ; +QUEUE_FLAG_MQ_ADMIN_DEFAULT :QUEUE_FLAG_MQ_DEFAULT ;if (shared_tags) { nullb->tag_set = &tag_set;
drivers/nvme/host/fc.c +1 -1
@@ -3041,7 +3041,7 @@ nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts, ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set; ctrl->ctrl.admin_q =__blk_mq_init_queue (&ctrl->admin_tag_set, -QUEUE_FLAG_MQ_NO_SCHED_DEFAULT ); +QUEUE_FLAG_MQ_ADMIN_DEFAULT );if (IS_ERR (ctrl->ctrl.admin_q)) { ret =PTR_ERR (ctrl->ctrl.admin_q); goto out_free_admin_tag_set;
drivers/nvme/host/pci.c +1 -1
@@ -1491,7 +1491,7 @@ static int nvme_alloc_admin_tags(struct nvme_dev *dev) dev->ctrl.admin_tagset = &dev->admin_tagset; dev->ctrl.admin_q =__blk_mq_init_queue (&dev->admin_tagset, -QUEUE_FLAG_MQ_NO_SCHED_DEFAULT ); +QUEUE_FLAG_MQ_ADMIN_DEFAULT );if (IS_ERR (dev->ctrl.admin_q)) {blk_mq_free_tag_set (&dev->admin_tagset);return -ENOMEM ;
drivers/nvme/host/rdma.c +1 -1
@@ -770,7 +770,7 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl, } ctrl->ctrl.admin_q =__blk_mq_init_queue (&ctrl->admin_tag_set, -QUEUE_FLAG_MQ_NO_SCHED_DEFAULT ); +QUEUE_FLAG_MQ_ADMIN_DEFAULT );if (IS_ERR (ctrl->ctrl.admin_q)) { error =PTR_ERR (ctrl->ctrl.admin_q); goto out_free_tagset;
drivers/nvme/target/loop.c +1 -1
@@ -381,7 +381,7 @@ static int nvme_loop_configure_admin_queue(struct nvme_loop_ctrl *ctrl) ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set; ctrl->ctrl.admin_q =__blk_mq_init_queue (&ctrl->admin_tag_set, -QUEUE_FLAG_MQ_NO_SCHED_DEFAULT ); +QUEUE_FLAG_MQ_ADMIN_DEFAULT );if (IS_ERR (ctrl->ctrl.admin_q)) { error =PTR_ERR (ctrl->ctrl.admin_q); goto out_free_tagset;
include/linux/blkdev.h +4 -4
@@ -699,7 +699,7 @@ struct request_queue {#define QUEUE_FLAG_SCSI_PASSTHROUGH 27/* queue supports SCSI commands */ #define QUEUE_FLAG_QUIESCED 28/* queue has been quiesced */ #define QUEUE_FLAG_PREEMPT_ONLY 29/* only process REQ_PREEMPT requests */ -#define QUEUE_FLAG_NO_SCHED 30/* no scheduler allowed */ +#define QUEUE_FLAG_ADMIN 30/* admin queue */ #define QUEUE_FLAG_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \ (1 << QUEUE_FLAG_SAME_COMP) | \
@@ -709,8 +709,8 @@ struct request_queue { (1 << QUEUE_FLAG_SAME_COMP) | \ (1 << QUEUE_FLAG_POLL)) -#define QUEUE_FLAG_MQ_NO_SCHED_DEFAULT (QUEUE_FLAG_MQ_DEFAULT | \ - (1 << QUEUE_FLAG_NO_SCHED)) +#define QUEUE_FLAG_MQ_ADMIN_DEFAULT (QUEUE_FLAG_MQ_DEFAULT | \ + (1 << QUEUE_FLAG_ADMIN))void blk_queue_flag_set (unsigned int flag,struct request_queue *q);void blk_queue_flag_clear (unsigned int flag,struct request_queue *q);
@@ -743,7 +743,7 @@ bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q);#define blk_queue_preempt_only (q) \ test_bit(QUEUE_FLAG_PREEMPT_ONLY, &(q)->queue_flags)#define blk_queue_fua (q) test_bit(QUEUE_FLAG_FUA, &(q)->queue_flags) -#define blk_queue_no_sched (q) test_bit(QUEUE_FLAG_NO_SCHED, &(q)->queue_flags) +#define blk_queue_admin (q) test_bit(QUEUE_FLAG_ADMIN, &(q)->queue_flags)extern int blk_set_preempt_only (struct request_queue *q);extern void blk_clear_preempt_only (struct request_queue *q);
Signature · 1 line
2.9.5
RFC PATCH v2 4/17 blk-mq: don't reserve tags for admin queue
In reply to Ming LeiNot necessary to reserve tags for admin queue since there isn't many inflight commands in admin queue usually.
This change won't starve admin queue too because each blocked queue has equal priority to get one new tag when one driver tag is released, no matter it is freed from any queue.
So that IO performance won't be affected after admin queue(shared tags with IO queues) is introduced in the following patches.
--- block/blk-mq-tag.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
block/blk-mq-tag.c +4 -2
@@ -30,7 +30,8 @@ bool blk_mq_has_free_tags(struct blk_mq_tags *tags)bool __blk_mq_tag_busy (struct blk_mq_hw_ctx *hctx) {if (!test_bit (BLK_MQ_S_TAG_ACTIVE , &hctx->state) && - !test_and_set_bit (BLK_MQ_S_TAG_ACTIVE , &hctx->state)) + !test_and_set_bit (BLK_MQ_S_TAG_ACTIVE , &hctx->state) && + !blk_queue_admin (hctx->queue))atomic_inc (&hctx->tags->active_queues);return true;
@@ -57,7 +58,8 @@ void __blk_mq_tag_idle(struct blk_mq_hw_ctx *hctx)if (!test_and_clear_bit (BLK_MQ_S_TAG_ACTIVE , &hctx->state))return ; -atomic_dec (&tags->active_queues); +if (!blk_queue_admin (hctx->queue)) +atomic_dec (&tags->active_queues);blk_mq_tag_wakeup_all (tags, false); }
Signature · 1 line
2.9.5
Re: [RFC PATCH V2 04/17] blk-mq: don't reserve tags for admin queue
Hi Ming
Quoted 7 lines · Ming Lei
Not necessary to reserve tags for admin queue since there isn't many inflight commands in admin queue usually.
This change won't starve admin queue too because each blocked queue has equal priority to get one new tag when one driver tag is released, no matter it is freed from any queue.
We don't count the adminq into tags->active_queues, there maybe following side-effect following:
- if send a admin request for the LUN_a, it may cause LUN_b cannot get request even though its budget is allowed. And when the admin request for LUN_a is completed, LUN_b may not be in the first batch to be waked up.
- all the LUNs still have to share a limited budget of tags
Certainly, I'm not sure whether the performance will be affected, all of above comments are just concern in theory. ;)
Thanks
Jianchao
Re: [RFC PATCH V2 04/17] blk-mq: don't reserve tags for admin queue
Quoted 6 lines · jianchao.wang
Hi Ming
Quoted 3 lines · Ming Lei
Not necessary to reserve tags for admin queue since there isn't many inflight commands in admin queue usually.
Quoted 6 lines
Quoted 4 lines
This change won't starve admin queue too because each blocked queue has equal priority to get one new tag when one driver tag is released, no matter it is freed from any queue.
We don't count the adminq into
tags->active_queues,
Yes, just like without this patchset.
Quoted 6 lines
there maybe following side-effect following:
- if send a admin request for theLUN_a, it may causeLUN_bcannot get request even though its budget is allowed. And when the admin request forLUN_ais completed,LUN_bmay not be in the first batch to be waked up.- all the LUNs still have to share a limited budget of tags
It is nothing to do with where the admin request is sent, so no any difference wrt. this issue between with and without this patchset, right?
Thanks,
Ming
Re: [RFC PATCH V2 04/17] blk-mq: don't reserve tags for admin queue
Hi Ming
Quoted 3 lines · Ming Lei
It is nothing to do with where the admin request is sent, so no any difference wrt. this issue between with and without this patchset, right?
I'm afraid not.
For example:
A scsi host has 8 LUNs associated with it.
Before this patch set,
When we send out the admin command, the budget is _per_ LUN, 1/8 of the total tags. After this patch set,
When we send out the admin command, the budget is equal to _one_ LUN, 1/8 of the total tags.
However, the 1/8 above is different.
Before the patch set, every LUN's admin command has 1/8 budget to use which is per LUN. After this patch set, all the 8 LUNs admin command has to share the 1/8 budget.
Thanks
Jianchao
Re: [RFC PATCH V2 04/17] blk-mq: don't reserve tags for admin queue
Quoted 16 lines · jianchao.wang
Hi Ming
Quoted 3 lines · Ming Lei
It is nothing to do with where the admin request is sent, so no any difference wrt. this issue between with and without this patchset, right?
I'm afraid not.
For example:
A scsi host has 8 LUNs associated with it.
Before this patch set,
When we send out the admin command, the budget is _per_ LUN, 1/8 of the total tags. After this patch set,
When we send out the admin command, the budget is equal to _one_ LUN, 1/8 of the total tags.
Quoted 2 lines
However, the 1/8 above is different.
Before the patch set, every LUN's admin command has 1/8 budget to use which is per LUN.
Strictly speaking, it is that all admin command and all other IOs share the 1/8 budget if they aimed at same LUN.
Quoted 1 line
After this patch set, all the 8 LUNs admin command has to share the 1/8 budget.
That only means number of active admin commands won't be bigger than 1/8 budget, which is one extra implicit limit on admin queue. However, other LUN's budget is still 1/8.
So performance for IO queue won't be affected at all, will it?
scsi_execute_* can't be called often, it is really in slow path, so I don't think there is any possible performance effect with this patch, or do you have other performance concern wrt. this patch?
We still have q->queue_depth for enhancing any limit for admin queue, but up to now, not see it is necessary.
Thanks,
Ming
Re: [RFC PATCH V2 04/17] blk-mq: don't reserve tags for admin queue
Hi Ming
Quoted 17 lines · Ming Lei
Quoted 16 lines · jianchao.wang
Hi Ming
Quoted 3 lines · Ming Lei
It is nothing to do with where the admin request is sent, so no any difference wrt. this issue between with and without this patchset, right?
I'm afraid not.
For example:
A scsi host has 8 LUNs associated with it.
Before this patch set,
When we send out the admin command, the budget is _per_ LUN, 1/8 of the total tags. After this patch set,
When we send out the admin command, the budget is equal to _one_ LUN, 1/8 of the total tags.
Quoted 5 lines
Quoted 2 lines
However, the 1/8 above is different.
Before the patch set, every LUN's admin command has 1/8 budget to use which is per LUN.Strictly speaking, it is that all admin command and all other IOs share the 1/8 budget if they aimed at same LUN.
Yes.
Quoted 6 lines
Quoted 2 lines
After this patch set, all the 8 LUNs admin command has to share the 1/8 budget.
That only means number of active admin commands won't be bigger than 1/8 budget, which is one extra implicit limit on admin queue. However, other LUN's budget is still 1/8.
Quoted 9 lines
So performance for IO queue won't be affected at all, will it?
scsi_execute_* can't be called often, it is really in slow path, so I don't think there is any possible performance effect with this patch, or do you have other performance concern wrt. this patch?
We still have
q->queue_depthfor enhancing any limit for admin queue, but up to now, not see it is necessary.
I agree with you that the performance will not be affected.
But the adminq's budget here looks weird.
We don't reserve budget for admin queue (not count tag->active_queues for it). But the admin queue has to comply to the limit in hctx_may_queue.
Since the there isn't many in-flight admin commands usually, we could take admin queue here out of the limit of hctx_may_queue, then things could be clearer. :)
Thanks
Jianchao
Re: [RFC PATCH V2 04/17] blk-mq: don't reserve tags for admin queue
Quoted 40 lines · jianchao.wang
Hi Ming
Quoted 17 lines · Ming Lei
Quoted 16 lines · jianchao.wang
Hi Ming
Quoted 3 lines · Ming Lei
It is nothing to do with where the admin request is sent, so no any difference wrt. this issue between with and without this patchset, right?
I'm afraid not.
For example:
A scsi host has 8 LUNs associated with it.
Before this patch set,
When we send out the admin command, the budget is _per_ LUN, 1/8 of the total tags. After this patch set,
When we send out the admin command, the budget is equal to _one_ LUN, 1/8 of the total tags.Quoted 5 lines
Quoted 2 lines
However, the 1/8 above is different.
Before the patch set, every LUN's admin command has 1/8 budget to use which is per LUN.Strictly speaking, it is that all admin command and all other IOs share the 1/8 budget if they aimed at same LUN.
Yes.
Quoted 12 lines
Quoted 2 lines
After this patch set, all the 8 LUNs admin command has to share the 1/8 budget.
That only means number of active admin commands won't be bigger than 1/8 budget, which is one extra implicit limit on admin queue. However, other LUN's budget is still 1/8.
So performance for IO queue won't be affected at all, will it?
scsi_execute_* can't be called often, it is really in slow path, so I don't think there is any possible performance effect with this patch, or do you have other performance concern wrt. this patch?
Quoted 5 lines
Quoted 3 lines
We still have
q->queue_depthfor enhancing any limit for admin queue, but up to now, not see it is necessary.I agree with you that the performance will not be affected.
OK, thanks for your confirm.
Quoted 6 lines
But the adminq's budget here looks weird.
We don't reserve budget for admin queue (not counttag->active_queuesfor it). But the admin queue has to comply to the limit inhctx_may_queue.Since the there isn't many in-flight admin commands usually, we could take admin queue here out of the limit of
hctx_may_queue, then things could be clearer. :)
I just didn't want to add one line code in the fast path of hctx_may_queue() because it isn't necessary.
Now looks this way may have one implicit benefit: avoid too many in-flight admin requests.
I will leave the code in this way, but add comment like below to hctx_may_queue():
Needn't to deal with admin queue specially here even though we don't take it account to tags->active_queues, so blk_queue_admin() can be avoided to check in the fast path, also with implicit benefit of limiting too many in-flight admin requests.
Thanks,
Ming
RFC PATCH v2 5/17 SCSI: try to retrieve request_queue via 'scsi_cmnd' if possible
In reply to Ming LeiPrepare for introduing per-host admin queue.
The most important part is that the request originated from admin queue can't be called back to the IO queue associated with scsi_device, especially, one request may be requeued, timedout or completed via block layer helper, so what we should do is to use 'scsi_cmnd->request->q' to retrieve the request queue, and pass that to block layer helper, instead of sdev->request_queue.
Fortunately most of users of 'scsi_device->request_queue' aren't in related IO path(requeue, timeout, complete, run queue), so the audit isn't more difficult than I thought of.
--- drivers/ata/libata-eh.c | 2 +- drivers/scsi/libsas/sas_ata.c | 2 +- drivers/scsi/libsas/sas_scsi_host.c | 2 +- drivers/scsi/scsi_error.c | 2 +- drivers/scsi/scsi_lib.c | 6 +++--- 5 files changed, 7 insertions(+), 7 deletions(-)
drivers/ata/libata-eh.c +1 -1
@@ -919,7 +919,7 @@ static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)void ata_qc_schedule_eh (struct ata_queued_cmd *qc) {struct ata_port *ap = qc->ap; -struct request_queue *q = qc->scsicmd->device->request_queue; +struct request_queue *q = qc->scsicmd->request->q;unsigned long flags;WARN_ON (!ap->ops->error_handler);
drivers/scsi/libsas/sas_ata.c +1 -1
@@ -594,7 +594,7 @@ void sas_ata_task_abort(struct sas_task *task)/* Bounce SCSI-initiated commands to the SCSI EH */ if (qc->scsicmd) { -struct request_queue *q = qc->scsicmd->device->request_queue; +struct request_queue *q = qc->scsicmd->request->q;unsigned long flags;spin_lock_irqsave (q->queue_lock, flags);
drivers/scsi/libsas/sas_scsi_host.c +1 -1
@@ -933,7 +933,7 @@ void sas_task_abort(struct sas_task *task)if (dev_is_sata (task->dev)) {sas_ata_task_abort (task); }else { -struct request_queue *q = sc->device->request_queue; +struct request_queue *q = sc->request->q;unsigned long flags;spin_lock_irqsave (q->queue_lock, flags);
drivers/scsi/scsi_error.c +1 -1
@@ -1360,7 +1360,7 @@ static int scsi_eh_try_stu(struct scsi_cmnd *scmd)int i, rtn =NEEDS_RETRY ;for (i =0 ; rtn ==NEEDS_RETRY && i <2 ; i++) - rtn =scsi_send_eh_cmnd (scmd, stu_command,6 , scmd->device->request_queue->rq_timeout,0 ); + rtn =scsi_send_eh_cmnd (scmd, stu_command,6 , scmd->request->q->rq_timeout,0 );if (rtn ==SUCCESS )return 0 ;
drivers/scsi/scsi_lib.c +3 -3
@@ -168,7 +168,7 @@ static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)static void __scsi_queue_insert (struct scsi_cmnd *cmd,int reason,bool unbusy) {struct scsi_device *device = cmd->device; -struct request_queue *q = device->request_queue; +struct request_queue *q = cmd->request->q;unsigned long flags; SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
@@ -667,7 +667,7 @@ static bool scsi_end_request(struct request *req, blk_status_t error, {struct scsi_cmnd *cmd =blk_mq_rq_to_pdu (req);struct scsi_device *sdev = cmd->device; -struct request_queue *q = sdev->request_queue; +struct request_queue *q = cmd->request->q;if (blk_update_request (req, error, bytes))return true;
@@ -792,7 +792,7 @@ static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)void scsi_io_completion (struct scsi_cmnd *cmd,unsigned int good_bytes) {int result = cmd->result; -struct request_queue *q = cmd->device->request_queue; +struct request_queue *q = cmd->request->q;struct request *req = cmd->request;blk_status_t error =BLK_STS_OK ;struct scsi_sense_hdr sshdr;
Signature · 1 line
2.9.5
RFC PATCH v2 6/17 SCSI: pass 'scsi_device' instance from 'scsi_request'
In reply to Ming LeiThis patch prepares for introducing SCSI per-host admin queue, which is only used for queuing admin requests, which are now submitted via __scsi_execute().
--- drivers/scsi/scsi_lib.c | 2 ++ include/scsi/scsi_request.h | 5 ++++- 2 files changed, 6 insertions(+), 1 deletion(-)
drivers/scsi/scsi_lib.c +2 -0
@@ -279,6 +279,8 @@ int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd, rq->cmd_len =COMMAND_SIZE (cmd[0 ]);memcpy (rq->cmd, cmd, rq->cmd_len); rq->retries = retries; + rq->sdev = sdev;/* only valid in submit path */ + req->timeout = timeout; req->cmd_flags |= flags; req->rq_flags |= rq_flags |RQF_QUIET ;
include/scsi/scsi_request.h +4 -1
@@ -14,7 +14,10 @@ struct scsi_request {unsigned int sense_len;unsigned int resid_len;/* residual count */ int retries; -void *sense; +union { +void *sense; +struct scsi_device *sdev; + }; };static inline struct scsi_request *scsi_req (struct request *rq)
Signature · 1 line
2.9.5
RFC PATCH v2 7/17 SCSI: prepare for introducing admin queue for legacy path
In reply to Ming LeiUses scsi_is_admin_queue() and scsi_get_scsi_dev() to retrieve 'scsi_device' for legacy path.
The same approach can be used in SCSI_MQ path too, just not very efficiently, and will deal with that in the patch when introducing admin queue for SCSI_MQ.
--- drivers/scsi/scsi_lib.c | 37 +++++++++++++++++++++++++++++-------- 1 file changed, 29 insertions(+), 8 deletions(-)
drivers/scsi/scsi_lib.c +29 -8
@@ -46,6 +46,20 @@ static DEFINE_MUTEX(scsi_sense_cache_mutex);static void scsi_mq_uninit_cmd (struct scsi_cmnd *cmd); +/* For admin queue, its queuedata is NULL */ +static inline bool scsi_is_admin_queue (struct request_queue *q) +{ +return !q->queuedata; +} + +/* This helper can only be used in req prep stage */ +static inline struct scsi_device *scsi_get_scsi_dev (struct request *rq) +{ +if (scsi_is_admin_queue (rq->q)) +return scsi_req (rq)->sdev; +return rq->q->queuedata; +} +static inline struct kmem_cache *scsi_select_sense_cache (bool unchecked_isa_dma) {
@@ -1376,10 +1390,9 @@ scsi_prep_state_check(struct scsi_device *sdev, struct request *req) }static int -scsi_prep_return (struct request_queue *q,struct request *req,int ret) +scsi_prep_return (struct scsi_device *sdev,struct request_queue *q, +struct request *req,int ret) { -struct scsi_device *sdev = q->queuedata; -switch (ret) {case BLKPREP_KILL :case BLKPREP_INVALID :
@@ -1411,7 +1424,7 @@ scsi_prep_return(struct request_queue *q, struct request *req, int ret)static int scsi_prep_fn (struct request_queue *q,struct request *req) { -struct scsi_device *sdev = q->queuedata; +struct scsi_device *sdev =scsi_get_scsi_dev (req);struct scsi_cmnd *cmd =blk_mq_rq_to_pdu (req);int ret;
@@ -1436,7 +1449,7 @@ static int scsi_prep_fn(struct request_queue *q, struct request *req) ret =scsi_setup_cmnd (sdev, req); out: -return scsi_prep_return (q, req, ret); +return scsi_prep_return (sdev, q, req, ret); }static void scsi_unprep_fn (struct request_queue *q,struct request *req)
@@ -1613,6 +1626,9 @@ static int scsi_lld_busy(struct request_queue *q)if (blk_queue_dying (q))return 0 ; +if (WARN_ON_ONCE (scsi_is_admin_queue (q))) +return 0 ; + shost = sdev->host;/*
@@ -1816,7 +1832,7 @@ static void scsi_request_fn(struct request_queue *q) __releases(q->queue_lock) __acquires(q->queue_lock) { -struct scsi_device *sdev = q->queuedata; +struct scsi_device *sdev;struct Scsi_Host *shost;struct scsi_cmnd *cmd;struct request *req;
@@ -1825,7 +1841,6 @@ static void scsi_request_fn(struct request_queue *q)* To start with, we keep looping until the queue is empty, or until * the host is no longer able to accept any more requests. */ - shost = sdev->host;for (;;) {int rtn;/*
@@ -1837,6 +1852,10 @@ static void scsi_request_fn(struct request_queue *q)if (!req)break ; + cmd =blk_mq_rq_to_pdu (req); + sdev = cmd->device; + shost = sdev->host; +if (unlikely (!scsi_device_online (sdev))) {sdev_printk (KERN_ERR , sdev,"rejecting I/O to offline device\n" );
@@ -1854,7 +1873,6 @@ static void scsi_request_fn(struct request_queue *q)blk_start_request (req);spin_unlock_irq (q->queue_lock); - cmd =blk_mq_rq_to_pdu (req);if (cmd != req->special) {printk (KERN_CRIT "impossible request in %s.\n" "please mail a stack trace to "
@@ -2332,6 +2350,9 @@ struct scsi_device *scsi_device_from_queue(struct request_queue *q) {struct scsi_device *sdev =NULL ; +/* admin queue won't be exposed to external users */ +WARN_ON_ONCE (scsi_is_admin_queue (q)); +if (q->mq_ops) {if (q->mq_ops == &scsi_mq_ops) sdev = q->queuedata;
Signature · 1 line
2.9.5
RFC PATCH v2 8/17 SCSI: pass scsi_device to scsi_mq_prep_fn
In reply to Ming LeiThis patchset will introduce per-host admin queue, so it may not to get 'scsi_device' via q->queuedata.
--- drivers/scsi/scsi_lib.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/scsi/scsi_lib.c +2 -3
@@ -1970,10 +1970,9 @@ static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)sizeof (struct scatterlist ); } -static int scsi_mq_prep_fn (struct request *req) +static int scsi_mq_prep_fn (struct scsi_device *sdev,struct request *req) {struct scsi_cmnd *cmd =blk_mq_rq_to_pdu (req); -struct scsi_device *sdev = req->q->queuedata;struct Scsi_Host *shost = sdev->host;struct scatterlist *sg;
@@ -2069,7 +2068,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, goto out_dec_target_busy;if (!(req->rq_flags &RQF_DONTPREP )) { - ret =prep_to_mq (scsi_mq_prep_fn (req)); + ret =prep_to_mq (scsi_mq_prep_fn (sdev, req));if (ret !=BLK_STS_OK ) goto out_dec_host_busy; req->rq_flags |=RQF_DONTPREP ;
Signature · 1 line
2.9.5
RFC PATCH v2 9/17 SCSI: don't set .queuedata in scsi_mq_alloc_queue()
In reply to Ming Lei.queuedata is set in scsi_alloc_sdev() for both non-mq and scsi_mq, so not necessary to do it in scsi_mq_alloc_queue().
--- drivers/scsi/scsi_lib.c | 1 - 1 file changed, 1 deletion(-)
drivers/scsi/scsi_lib.c +0 -1
@@ -2304,7 +2304,6 @@ struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)if (IS_ERR (sdev->request_queue))return NULL ; - sdev->request_queue->queuedata = sdev;__scsi_init_queue (sdev->host, sdev->request_queue);blk_queue_flag_set (QUEUE_FLAG_SCSI_PASSTHROUGH , sdev->request_queue);return sdev->request_queue;
Signature · 1 line
2.9.5
RFC PATCH v2 10/17 SCSI: deal with admin queue busy
In reply to Ming LeiWhen request originated from admin queue isn't queued successfully, we deal with it just like for normal requests, that said the admin queue will be rerun after one request in this host is completed.
--- drivers/scsi/scsi_lib.c | 61 ++++++++++++++++++++++++++++++++++-------------- include/scsi/scsi_host.h | 2 ++ 2 files changed, 46 insertions(+), 17 deletions(-)
drivers/scsi/scsi_lib.c +44 -17
@@ -475,10 +475,14 @@ static void scsi_starved_list_run(struct Scsi_Host *shost)LIST_HEAD (starved_list);struct scsi_device *sdev;unsigned long flags; +bool run_admin;spin_lock_irqsave (shost->host_lock, flags);list_splice_init (&shost->starved_list, &starved_list); + run_admin = shost->run_admin_queue; + shost->run_admin_queue = false; +while (!list_empty (&starved_list)) {struct request_queue *slq;
@@ -527,6 +531,10 @@ static void scsi_starved_list_run(struct Scsi_Host *shost)/* put any unprocessed entries back */ list_splice (&starved_list, &shost->starved_list);spin_unlock_irqrestore (shost->host_lock, flags); + +/* no need to get queue for admin_q */ +if (run_admin) +scsi_kick_queue (shost->admin_q); }/*
@@ -534,26 +542,30 @@ static void scsi_starved_list_run(struct Scsi_Host *shost)* * Purpose: Select a proper request queue to serve next * -* Arguments: q - last request's queue +* Arguments: sdev - the last request's scsi_device +* q - last request's queue, which may points to +* host->admin_q * * Returns: Nothing * * Notes: The previous command was completely finished, start * a new one if possible. */ -static void scsi_run_queue (struct request_queue *q) +static void scsi_run_queue (struct scsi_device *sdev,struct request_queue *q) { -struct scsi_device *sdev = q->queuedata; +struct Scsi_Host *shost = sdev->host;if (scsi_target (sdev)->single_lun)scsi_single_lun_run (sdev); -if (!list_empty (&sdev->host->starved_list)) -scsi_starved_list_run (sdev->host); -if (q->mq_ops) -blk_mq_run_hw_queues (q, false); -else -blk_run_queue (q); +if (!list_empty (&shost->starved_list) || shost->run_admin_queue) +scsi_starved_list_run (shost); + +scsi_kick_queue (q); + +/* q may points to host->admin_queue */ +if (sdev->request_queue != q) +scsi_kick_queue (sdev->request_queue); }void scsi_requeue_run_queue (struct work_struct *work)
@@ -563,7 +575,7 @@ void scsi_requeue_run_queue(struct work_struct *work) sdev = container_of(work, struct scsi_device, requeue_work); q = sdev->request_queue; -scsi_run_queue (q); +scsi_run_queue (sdev, q); }/*
@@ -597,7 +609,7 @@ static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)blk_requeue_request (q, req);spin_unlock_irqrestore (q->queue_lock, flags); -scsi_run_queue (q); +scsi_run_queue (sdev, q);put_device (&sdev->sdev_gendev); }
@@ -607,7 +619,7 @@ void scsi_run_host_queues(struct Scsi_Host *shost)struct scsi_device *sdev;shost_for_each_device (sdev, shost) -scsi_run_queue (sdev->request_queue); +scsi_run_queue (sdev, sdev->request_queue); }static void scsi_uninit_cmd (struct scsi_cmnd *cmd)
@@ -714,8 +726,13 @@ static bool scsi_end_request(struct request *req, blk_status_t error,__blk_mq_end_request (req, error); +/* +* scsi_device is shared between host->admin_queue and +* sdev->request_queue +*/ if (scsi_target (sdev)->single_lun || - !list_empty (&sdev->host->starved_list)) + !list_empty (&sdev->host->starved_list) || + sdev->host->run_admin_queue ||scsi_is_admin_queue (q))kblockd_schedule_work (&sdev->requeue_work);else blk_mq_run_hw_queues (q, true);
@@ -731,7 +748,7 @@ static bool scsi_end_request(struct request *req, blk_status_t error,blk_finish_request (req, error);spin_unlock_irqrestore (q->queue_lock, flags); -scsi_run_queue (q); +scsi_run_queue (sdev, q); }put_device (&sdev->sdev_gendev);
@@ -1494,6 +1511,12 @@ static inline int scsi_dev_queue_ready(struct request_queue *q,return 1 ; out_dec:atomic_dec (&sdev->device_busy); + +if (unlikely (scsi_is_admin_queue (q))) { +spin_lock_irq (sdev->host->host_lock); + sdev->host->run_admin_queue = true; +spin_unlock_irq (sdev->host->host_lock); + }return 0 ; }
@@ -1502,7 +1525,7 @@ static inline int scsi_dev_queue_ready(struct request_queue *q,* @sdev: scsi device on starget to check. */ static inline int scsi_target_queue_ready (struct Scsi_Host *shost, -struct scsi_device *sdev) +struct scsi_device *sdev,bool admin) {struct scsi_target *starget =scsi_target (sdev);unsigned int busy;
@@ -1544,6 +1567,8 @@ static inline int scsi_target_queue_ready(struct Scsi_Host *shost, starved:spin_lock_irq (shost->host_lock);list_move_tail (&sdev->starved_entry, &shost->starved_list); +if (admin) + shost->run_admin_queue = true;spin_unlock_irq (shost->host_lock); out_dec:if (starget->can_queue >0 )
@@ -1600,6 +1625,8 @@ static inline int scsi_host_queue_ready(struct request_queue *q,spin_lock_irq (shost->host_lock);if (list_empty (&sdev->starved_entry))list_add_tail (&sdev->starved_entry, &shost->starved_list); +if (scsi_is_admin_queue (q)) + shost->run_admin_queue = true;spin_unlock_irq (shost->host_lock); out_dec:scsi_dec_host_busy (shost);
@@ -1899,7 +1926,7 @@ static void scsi_request_fn(struct request_queue *q) goto not_ready; } -if (!scsi_target_queue_ready (shost, sdev)) +if (!scsi_target_queue_ready (shost, sdev,scsi_is_admin_queue (q))) goto not_ready;if (!scsi_host_queue_ready (q, shost, sdev))
@@ -2062,7 +2089,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, goto out_put_budget; ret =BLK_STS_RESOURCE ; -if (!scsi_target_queue_ready (shost, sdev)) +if (!scsi_target_queue_ready (shost, sdev,scsi_is_admin_queue (q))) goto out_put_budget;if (!scsi_host_queue_ready (q, shost, sdev)) goto out_dec_target_busy;
include/scsi/scsi_host.h +2 -0
@@ -532,6 +532,8 @@ struct Scsi_Host {struct list_head __devices;struct list_head __targets; +struct request_queue *admin_q; +bool run_admin_queue;struct list_head starved_list;spinlock_t default_lock;
Signature · 1 line
2.9.5
RFC PATCH v2 11/17 SCSI: track pending admin commands
In reply to Ming LeiFirstly we have to make sure that all pending admin commands to one same scsi_device are completed before removing the scsi_device.
Secondly scsi_internal_device_block() needs this too.
So introduce one waitqueue and atomic counter for this purpose.
--- drivers/scsi/scsi_lib.c | 6 ++++++ drivers/scsi/scsi_scan.c | 1 + drivers/scsi/scsi_sysfs.c | 1 + include/scsi/scsi_device.h | 4 ++++ 4 files changed, 12 insertions(+)
drivers/scsi/scsi_lib.c +6 -0
@@ -299,6 +299,8 @@ int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd, req->cmd_flags |= flags; req->rq_flags |= rq_flags |RQF_QUIET ; +atomic_inc (&sdev->nr_admin_pending); +/* * head injection *required* here otherwise quiesce won't work */
@@ -323,6 +325,9 @@ int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd, out:blk_put_request (req); +atomic_dec (&sdev->nr_admin_pending); +wake_up_all (&sdev->admin_wq); +return ret; }EXPORT_SYMBOL (__scsi_execute);
@@ -3196,6 +3201,7 @@ static int scsi_internal_device_block(struct scsi_device *sdev)else scsi_wait_for_queuecommand (sdev); } +wait_event (sdev->admin_wq, !atomic_read (&sdev->nr_admin_pending));mutex_unlock (&sdev->state_mutex);return err;
drivers/scsi/scsi_scan.c +1 -0
@@ -243,6 +243,7 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget,mutex_init (&sdev->inquiry_mutex);INIT_WORK (&sdev->event_work, scsi_evt_thread);INIT_WORK (&sdev->requeue_work, scsi_requeue_run_queue); +init_waitqueue_head (&sdev->admin_wq); sdev->sdev_gendev.parent =get_device (&starget->dev); sdev->sdev_target = starget;
drivers/scsi/scsi_sysfs.c +1 -0
@@ -1377,6 +1377,7 @@ void __scsi_remove_device(struct scsi_device *sdev)blk_cleanup_queue (sdev->request_queue);cancel_work_sync (&sdev->requeue_work); +wait_event (sdev->admin_wq, !atomic_read (&sdev->nr_admin_pending));if (sdev->host->hostt->slave_destroy) sdev->host->hostt->slave_destroy (sdev);
include/scsi/scsi_device.h +4 -0
@@ -227,6 +227,10 @@ struct scsi_device {struct mutex state_mutex;enum scsi_device_state sdev_state;struct task_struct *quiesced_by; + +atomic_t nr_admin_pending; +wait_queue_head_t admin_wq; +unsigned long sdev_data[0 ]; } __attribute__((aligned(sizeof(unsigned long))));
Signature · 1 line
2.9.5
RFC PATCH v2 12/17 SCSI: create admin queue for each host
In reply to Ming LeiThe created admin queue will be used to send internal admin commands, so we can simplify the sync between some admin commands and IO requests, typical examples are system suspend and runtime PM.
--- drivers/scsi/hosts.c | 9 ++++ drivers/scsi/scsi_lib.c | 117 +++++++++++++++++++++++++++++++++++++++++------ drivers/scsi/scsi_priv.h | 1 + 3 files changed, 114 insertions(+), 13 deletions(-)
drivers/scsi/hosts.c +9 -0
@@ -242,6 +242,9 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, shost->dma_dev = dma_dev; +if (!scsi_init_admin_queue (shost)) + goto out_remove_tags; +/* * Increase usage count temporarily here so that calling * scsi_autopm_put_host() will trigger runtime idle if there is
@@ -309,6 +312,9 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,pm_runtime_disable (&shost->shost_gendev);pm_runtime_set_suspended (&shost->shost_gendev);pm_runtime_put_noidle (&shost->shost_gendev); +blk_cleanup_queue (shost->admin_q); +blk_put_queue (shost->admin_q); + out_remove_tags:if (shost_use_blk_mq (shost))scsi_mq_destroy_tags (shost); fail:
@@ -344,6 +350,9 @@ static void scsi_host_dev_release(struct device *dev)kfree (dev_name (&shost->shost_dev)); } +blk_cleanup_queue (shost->admin_q); +blk_put_queue (shost->admin_q); +if (shost_use_blk_mq (shost)) {if (shost->tag_set.tags)scsi_mq_destroy_tags (shost);
drivers/scsi/scsi_lib.c +104 -13
182 lines not on this page · raw message
drivers/scsi/scsi_priv.h +1 -0
@@ -94,6 +94,7 @@ extern void scsi_run_host_queues(struct Scsi_Host *shost);extern void scsi_requeue_run_queue (struct work_struct *work);extern struct request_queue *scsi_old_alloc_queue (struct scsi_device *sdev);extern struct request_queue *scsi_mq_alloc_queue (struct scsi_device *sdev); +extern struct request_queue *scsi_init_admin_queue (struct Scsi_Host *shost);extern void scsi_start_queue (struct scsi_device *sdev);extern int scsi_mq_setup_tags (struct Scsi_Host *shost);extern void scsi_mq_destroy_tags (struct Scsi_Host *shost);
Signature · 1 line
2.9.5
Re: [RFC PATCH V2 12/17] SCSI: create admin queue for each host
Hi Ming
drivers/scsi/scsi_lib.c static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, · line 2175
Quoted 12 lines · Ming Lei
drivers/scsi/scsi_lib.c +12
-0+static blk_status_t scsi_admin_queue_rq (struct blk_mq_hw_ctx *hctx, +const struct blk_mq_queue_data *bd) +{ +struct scsi_device *sdev =scsi_req (bd->rq)->sdev; + +WARN_ON_ONCE (hctx->queue == sdev->request_queue); + +if (!__scsi_mq_get_budget (hctx, sdev)) +return BLK_STS_RESOURCE ; + +return __scsi_queue_rq (hctx, bd, sdev); +}
Why does it return BLK_STS_RESOURCE here ?
Do you want to use the following code branch in blk_mq_dispatch_rq_list to get the admin queue rerun ?
else if (needs_restart && (ret == BLK_STS_RESOURCE)) blk_mq_delay_run_hw_queue(hctx, BLK_MQ_RESOURCE_DELAY);
In theory, BLK_STS_DEV_RESOURCE should be returned here. However, I guess it will not work. Because the rr fashion loop restart has gone, the completion of normal queue will not restart the admin queue. If yes, please add some comment here to describe it. :)
Thanks
Jianchao
Re: [RFC PATCH V2 12/17] SCSI: create admin queue for each host
Quoted 18 lines · jianchao.wang
Hi Ming
drivers/scsi/scsi_lib.c static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, · line 2175
Quoted 12 lines · Ming Lei
drivers/scsi/scsi_lib.c +12
-0+static blk_status_t scsi_admin_queue_rq (struct blk_mq_hw_ctx *hctx, +const struct blk_mq_queue_data *bd) +{ +struct scsi_device *sdev =scsi_req (bd->rq)->sdev; + +WARN_ON_ONCE (hctx->queue == sdev->request_queue); + +if (!__scsi_mq_get_budget (hctx, sdev)) +return BLK_STS_RESOURCE ; + +return __scsi_queue_rq (hctx, bd, sdev); +}Why does it return
BLK_STS_RESOURCEhere ?
Quoted 9 lines
Do you want to use the following code branch in
blk_mq_dispatch_rq_listto get the admin queue rerun ?else if (needs_restart && (ret == BLK_STS_RESOURCE)) blk_mq_delay_run_hw_queue(hctx, BLK_MQ_RESOURCE_DELAY);In theory,
BLK_STS_DEV_RESOURCEshould be returned here. However, I guess it will not work. Because the rr fashion loop restart has gone, the completion of normal queue will not restart the admin queue. If yes, please add some comment here to describe it. :)
Looks like the patch 10 could avoid the admin queue hung.
Thanks
Jianchao
Re: [RFC PATCH V2 12/17] SCSI: create admin queue for each host
Quoted 26 lines · jianchao.wang
Quoted 23 lines · jianchao.wang
Hi Ming
drivers/scsi/scsi_lib.c static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, · line 2175
Quoted 12 lines · Ming Lei
drivers/scsi/scsi_lib.c +12
-0+static blk_status_t scsi_admin_queue_rq (struct blk_mq_hw_ctx *hctx, +const struct blk_mq_queue_data *bd) +{ +struct scsi_device *sdev =scsi_req (bd->rq)->sdev; + +WARN_ON_ONCE (hctx->queue == sdev->request_queue); + +if (!__scsi_mq_get_budget (hctx, sdev)) +return BLK_STS_RESOURCE ; + +return __scsi_queue_rq (hctx, bd, sdev); +}Why does it return
BLK_STS_RESOURCEhere ?Do you want to use the following code branch in
blk_mq_dispatch_rq_listto get the admin queue rerun ?else if (needs_restart && (ret == BLK_STS_RESOURCE)) blk_mq_delay_run_hw_queue(hctx, BLK_MQ_RESOURCE_DELAY);
Quoted 6 lines
Quoted 4 lines
In theory,
BLK_STS_DEV_RESOURCEshould be returned here. However, I guess it will not work. Because the rr fashion loop restart has gone, the completion of normal queue will not restart the admin queue. If yes, please add some comment here to describe it. :)Looks like the patch 10 could avoid the admin queue hung.
Yes, this one is patch 12, :-)
Also not necessary to add comment on returning BLK_STS_RESOURCE because it is always the safe say, given blk-mq will handle the queue rerun.
However, if driver returns BLK_STS_DEV_RESOURCE, some comment have to be added, please see comment on BLK_STS_DEV_RESOURCE.
Thanks,
Ming
RFC PATCH v2 13/17 SCSI: use the dedicated admin queue to send admin commands
In reply to Ming LeiNow the per-host dedicated admin queue is ready, so use this queue to send admin commands only.
--- drivers/scsi/scsi_lib.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/scsi/scsi_lib.c +2 -2
@@ -279,14 +279,14 @@ int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,struct scsi_request *rq;int ret =DRIVER_ERROR <<24 ; - req =blk_get_request (sdev->request_queue, + req =blk_get_request (sdev->host->admin_q, data_direction ==DMA_TO_DEVICE ?REQ_OP_SCSI_OUT :REQ_OP_SCSI_IN ,BLK_MQ_REQ_PREEMPT );if (IS_ERR (req))return ret; rq =scsi_req (req); -if (bufflen &&blk_rq_map_kern (sdev->request_queue, req, +if (bufflen &&blk_rq_map_kern (req->q, req, buffer, bufflen,GFP_NOIO )) goto out;
Signature · 1 line
2.9.5
RFC PATCH v2 14/17 SCSI: transport_spi: resume a quiesced device
In reply to Ming LeiWe have to preempt freeze queue in scsi_device_quiesce(), and unfreeze in scsi_device_resume(), so call scsi_device_resume() for the device which is quiesced by scsi_device_quiesce().
--- drivers/scsi/scsi_transport_spi.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/scsi/scsi_transport_spi.c +3 -0
@@ -1052,6 +1052,9 @@ spi_dv_device(struct scsi_device *sdev)scsi_target_resume (starget); +/* undo what scsi_device_quiesce() did */ +scsi_device_resume (sdev); +spi_initial_dv (starget) =1 ; out_free:
Signature · 1 line
2.9.5
RFC PATCH v2 15/17 SCSI: use admin queue to implement queue QUIESCE
In reply to Ming LeiAll admin commands are sent via per-host admin queue, so we can simply freeze the IO queue for quiescing scsi device.
Also the current SCSI stack guarantees that any request originated from admin queue won't be called back to block layer via the associated IO queue, and it is always dealt with by the admin queue.
So it is safe to submit admin request via admin queue when the associated IO queue is frozen, and this way matches the PREEMPT flag perfectly.
Finally, we can remove the preempt_only approach for supporting SCSI quiesce, then the code in block fast path is simplified a lot.
--- block/blk-core.c | 44 ++------------------------------------------ block/blk-mq-debugfs.c | 1 - drivers/scsi/scsi_lib.c | 29 +++-------------------------- include/linux/blkdev.h | 6 ------ include/scsi/scsi_device.h | 1 - 5 files changed, 5 insertions(+), 76 deletions(-)
block/blk-core.c +2 -42
@@ -421,26 +421,6 @@ void blk_sync_queue(struct request_queue *q)EXPORT_SYMBOL (blk_sync_queue);/** -* blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY -* @q: request queue pointer -* -* Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not -* set and 1 if the flag was already set. -*/ -int blk_set_preempt_only (struct request_queue *q) -{ -return blk_queue_flag_test_and_set (QUEUE_FLAG_PREEMPT_ONLY , q); -} -EXPORT_SYMBOL_GPL (blk_set_preempt_only); - -void blk_clear_preempt_only (struct request_queue *q) -{ -blk_queue_flag_clear (QUEUE_FLAG_PREEMPT_ONLY , q); -wake_up_all (&q->mq_freeze_wq); -} -EXPORT_SYMBOL_GPL (blk_clear_preempt_only); - -/** * __blk_run_queue_uncond - run a queue whether or not it has been stopped * @q: The queue to run *
@@ -917,27 +897,8 @@ EXPORT_SYMBOL(blk_alloc_queue);*/ int blk_queue_enter (struct request_queue *q,blk_mq_req_flags_t flags) { -const bool preempt = flags &BLK_MQ_REQ_PREEMPT ; -while (true) { -bool success = false; - -rcu_read_lock (); -if (percpu_ref_tryget_live (&q->q_usage_counter)) { -/* -* The code that sets the PREEMPT_ONLY flag is -* responsible for ensuring that that flag is globally -* visible before the queue is unfrozen. -*/ -if (preempt || !blk_queue_preempt_only (q)) { - success = true; - }else { -percpu_ref_put (&q->q_usage_counter); - } - } -rcu_read_unlock (); - -if (success) +if (percpu_ref_tryget_live (&q->q_usage_counter))return 0 ;if (flags &BLK_MQ_REQ_NOWAIT )
@@ -953,8 +914,7 @@ int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)smp_rmb ();wait_event (q->mq_freeze_wq, - (atomic_read (&q->mq_freeze_depth) ==0 && - (preempt || !blk_queue_preempt_only (q))) || +atomic_read (&q->mq_freeze_depth) ==0 ||blk_queue_dying (q));if (blk_queue_dying (q))return -ENODEV ;
block/blk-mq-debugfs.c +0 -1
@@ -132,7 +132,6 @@ static const char *const blk_queue_flag_name[] = {QUEUE_FLAG_NAME (REGISTERED ),QUEUE_FLAG_NAME (SCSI_PASSTHROUGH ),QUEUE_FLAG_NAME (QUIESCED ), -QUEUE_FLAG_NAME (PREEMPT_ONLY ), QUEUE_FLAG_NAME(ADMIN), };#undef QUEUE_FLAG_NAME
drivers/scsi/scsi_lib.c +3 -26
@@ -3130,34 +3130,12 @@ static void scsi_wait_for_queuecommand(struct scsi_device *sdev)int scsi_device_quiesce (struct scsi_device *sdev) { -struct request_queue *q = sdev->request_queue;int err; -/* -* It is allowed to call scsi_device_quiesce() multiple times from -* the same context but concurrent scsi_device_quiesce() calls are -* not allowed. -*/ -WARN_ON_ONCE (sdev->quiesced_by && sdev->quiesced_by != current); - -blk_set_preempt_only (q); - -blk_mq_freeze_queue (q); -/* -* Ensure that the effect of blk_set_preempt_only() will be visible -* for percpu_ref_tryget() callers that occur after the queue -* unfreeze even if the queue was already frozen before this function -* was called. See also https://lwn.net/Articles/573497/. -*/ -synchronize_rcu (); -blk_mq_unfreeze_queue (q); +blk_mq_freeze_queue (sdev->request_queue);mutex_lock (&sdev->state_mutex); err =scsi_device_set_state (sdev,SDEV_QUIESCE ); -if (err ==0 ) - sdev->quiesced_by = current; -else -blk_clear_preempt_only (q);mutex_unlock (&sdev->state_mutex);return err;
@@ -3180,12 +3158,11 @@ void scsi_device_resume(struct scsi_device *sdev)* device deleted during suspend) */ mutex_lock (&sdev->state_mutex); -WARN_ON_ONCE (!sdev->quiesced_by); - sdev->quiesced_by =NULL ; -blk_clear_preempt_only (sdev->request_queue);if (sdev->sdev_state ==SDEV_QUIESCE )scsi_device_set_state (sdev,SDEV_RUNNING );mutex_unlock (&sdev->state_mutex); + +blk_mq_unfreeze_queue (sdev->request_queue); }EXPORT_SYMBOL (scsi_device_resume);
include/linux/blkdev.h +0 -6
@@ -698,7 +698,6 @@ struct request_queue {#define QUEUE_FLAG_REGISTERED 26/* queue has been registered to a disk */ #define QUEUE_FLAG_SCSI_PASSTHROUGH 27/* queue supports SCSI commands */ #define QUEUE_FLAG_QUIESCED 28/* queue has been quiesced */ -#define QUEUE_FLAG_PREEMPT_ONLY 29/* only process REQ_PREEMPT requests */ #define QUEUE_FLAG_ADMIN 30/* admin queue */ #define QUEUE_FLAG_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \
@@ -740,14 +739,9 @@ bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q); ((rq)->cmd_flags & (REQ_FAILFAST_DEV |REQ_FAILFAST_TRANSPORT | \ REQ_FAILFAST_DRIVER))#define blk_queue_quiesced (q) test_bit(QUEUE_FLAG_QUIESCED, &(q)->queue_flags) -#define blk_queue_preempt_only (q) \ - test_bit(QUEUE_FLAG_PREEMPT_ONLY, &(q)->queue_flags)#define blk_queue_fua (q) test_bit(QUEUE_FLAG_FUA, &(q)->queue_flags)#define blk_queue_admin (q) test_bit(QUEUE_FLAG_ADMIN, &(q)->queue_flags) -extern int blk_set_preempt_only (struct request_queue *q); -extern void blk_clear_preempt_only (struct request_queue *q); -static inline int queue_in_flight (struct request_queue *q) {return q->in_flight[0 ] + q->in_flight[1 ];
include/scsi/scsi_device.h +0 -1
@@ -226,7 +226,6 @@ struct scsi_device {unsigned char access_state;struct mutex state_mutex;enum scsi_device_state sdev_state; -struct task_struct *quiesced_by;atomic_t nr_admin_pending;wait_queue_head_t admin_wq;
Signature · 1 line
2.9.5
RFC PATCH v2 16/17 block: simplify runtime PM support
In reply to Ming LeiThis patch simplifies runtime PM support by the following approach:
1) resume device in blk_queue_enter() if this device is runtime-suspended or runtime-suspending
2) freeze queue in blk_pre_runtime_suspend()
3) unfreeze queue in blk_pre_runtime_resume()
4) remove checking on RRF_PM because now we requires out-of-band PM request to resume device
5) introduce blk_unfreeze_queue_lock() and blk_freeze_queue_lock() so that both runtime-PM and system-PM can use them to freeze/unfreeze queue and avoid freeze & unfreeze mismatch
Then we can remove blk_pm_allow_request(), and more importantly this way can be applied to blk-mq path too.
Finally the IO queue associated with scsi_device is kept as runtime resumed in __scsi_execute() when sending non-PM RQF_REQUEST, and this way makes sure that the LUN is active for handling non-PM RQF_PREEMPT.
--- block/blk-core.c | 106 ++++++++++++++++++++++++++++-------------------- block/blk-mq.c | 22 ++++++++++ block/elevator.c | 25 ------------ drivers/scsi/scsi_lib.c | 14 +++++-- include/linux/blk-mq.h | 2 + include/linux/blkdev.h | 3 ++ 6 files changed, 101 insertions(+), 71 deletions(-)
block/blk-core.c +63 -43
@@ -890,6 +890,28 @@ struct request_queue *blk_alloc_queue(gfp_t gfp_mask) }EXPORT_SYMBOL (blk_alloc_queue); +#ifdef CONFIG_PM +static void blk_resume_queue (struct request_queue *q) +{ +int rpm_status; + +if (!q->dev) +return ; + +spin_lock_irq (q->queue_lock); + rpm_status = q->rpm_status; +spin_unlock_irq (q->queue_lock); + +/* PM request needs to be dealt with out of band */ +if (rpm_status ==RPM_SUSPENDED || rpm_status ==RPM_SUSPENDING ) +pm_runtime_resume (q->dev); +} +#else +static void blk_resume_queue (struct request_queue *q) +{ +} +#endif +/** * blk_queue_enter() - try to increase q->q_usage_counter * @q: request queue pointer
@@ -913,11 +935,20 @@ int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)*/ smp_rmb (); +blk_resume_queue (q); +wait_event (q->mq_freeze_wq,atomic_read (&q->mq_freeze_depth) ==0 ||blk_queue_dying (q));if (blk_queue_dying (q))return -ENODEV ; + +/* +* This allocation may be blocked via queue freezing before +* the queue is suspended, so we have to resume queue again +* after waking up. +*/ +blk_resume_queue (q); } }
@@ -1024,6 +1055,7 @@ struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id, q->bypass_depth =1 ;queue_flag_set_unlocked (QUEUE_FLAG_BYPASS , q); +mutex_init (&q->freeze_lock);init_waitqueue_head (&q->mq_freeze_wq);/*
@@ -1471,6 +1503,23 @@ static struct request *__get_request(struct request_list *rl, unsigned int op,return ERR_PTR (-ENOMEM ); } +#ifdef CONFIG_PM +static void blk_pm_add_request (struct request_queue *q) +{ +if (q->dev) + q->nr_pending++; +} +static void blk_pm_put_request (struct request_queue *q) +{ +if (q->dev && !--q->nr_pending) +pm_runtime_mark_last_busy (q->dev); +} +#else +static inline void blk_pm_put_request (struct request_queue *q) {} +static inline void blk_pm_add_request (struct request_queue *q){} +#endif + +/** * get_request - get a free request * @q: request_queue to allocate request from
@@ -1499,16 +1548,19 @@ static struct request *get_request(struct request_queue *q, unsigned int op, rl =blk_get_rl (q, bio);/* transferred to @rq on success */ retry: +blk_pm_add_request (q); rq =__get_request (rl, op, bio, flags, gfp);if (!IS_ERR (rq))return rq;if (op &REQ_NOWAIT ) { +blk_pm_put_request (q);blk_put_rl (rl);return ERR_PTR (-EAGAIN ); }if ((flags &BLK_MQ_REQ_NOWAIT ) ||unlikely (blk_queue_dying (q))) { +blk_pm_put_request (q);blk_put_rl (rl);return rq; }
@@ -1519,6 +1571,7 @@ static struct request *get_request(struct request_queue *q, unsigned int op,trace_block_sleeprq (q, bio, op); +blk_pm_put_request (q);spin_unlock_irq (q->queue_lock);io_schedule ();
@@ -1687,16 +1740,6 @@ void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part) }EXPORT_SYMBOL_GPL (part_round_stats); -#ifdef CONFIG_PM -static void blk_pm_put_request (struct request *rq) -{ -if (rq->q->dev && !(rq->rq_flags &RQF_PM ) && !--rq->q->nr_pending) -pm_runtime_mark_last_busy (rq->q->dev); -} -#else -static inline void blk_pm_put_request (struct request *rq) {} -#endif -void __blk_put_request (struct request_queue *q,struct request *req) {req_flags_t rq_flags = req->rq_flags;
@@ -1712,7 +1755,7 @@ void __blk_put_request(struct request_queue *q, struct request *req)lockdep_assert_held (q->queue_lock);blk_req_zone_write_unlock (req); -blk_pm_put_request (req); +blk_pm_put_request (q);elv_completed_request (q, req);
@@ -2708,30 +2751,6 @@ void blk_account_io_done(struct request *req, u64 now) } } -#ifdef CONFIG_PM -/* -* Don't process normal requests when queue is suspended -* or in the process of suspending/resuming -*/ -static bool blk_pm_allow_request (struct request *rq) -{ -switch (rq->q->rpm_status) { -case RPM_RESUMING : -case RPM_SUSPENDING : -return rq->rq_flags &RQF_PM ; -case RPM_SUSPENDED : -return false; -default : -return true; - } -} -#else -static bool blk_pm_allow_request (struct request *rq) -{ -return true; -} -#endif -void blk_account_io_start (struct request *rq,bool new_io) {struct hd_struct *part;
@@ -2776,13 +2795,8 @@ static struct request *elv_next_request(struct request_queue *q)WARN_ON_ONCE (q->mq_ops);while (1 ) { -list_for_each_entry (rq, &q->queue_head, queuelist) { -if (blk_pm_allow_request (rq)) -return rq; - -if (rq->rq_flags &RQF_SOFTBARRIER ) -break ; - } +list_for_each_entry (rq, &q->queue_head, queuelist) +return rq;/* * Flush request is running and flush request isn't queueable
@@ -3786,6 +3800,10 @@ int blk_pre_runtime_suspend(struct request_queue *q) q->rpm_status =RPM_SUSPENDING ; }spin_unlock_irq (q->queue_lock); + +if (!ret) +blk_freeze_queue_lock (q); +return ret; }EXPORT_SYMBOL (blk_pre_runtime_suspend);
@@ -3863,13 +3881,15 @@ void blk_post_runtime_resume(struct request_queue *q, int err)spin_lock_irq (q->queue_lock);if (!err) { q->rpm_status =RPM_ACTIVE ; -__blk_run_queue (q);pm_runtime_mark_last_busy (q->dev);pm_request_autosuspend (q->dev); }else { q->rpm_status =RPM_SUSPENDED ; }spin_unlock_irq (q->queue_lock); + +if (!err) +blk_unfreeze_queue_lock (q); }EXPORT_SYMBOL (blk_post_runtime_resume);
block/blk-mq.c +22 -0
@@ -204,6 +204,28 @@ void blk_mq_unfreeze_queue(struct request_queue *q) }EXPORT_SYMBOL_GPL (blk_mq_unfreeze_queue); +void blk_unfreeze_queue_lock (struct request_queue *q) +{ +mutex_lock (&q->freeze_lock); +if (q->q_frozen) { +blk_mq_unfreeze_queue (q); + q->q_frozen = false; + } +mutex_unlock (&q->freeze_lock); +} +EXPORT_SYMBOL (blk_unfreeze_queue_lock); + +void blk_freeze_queue_lock (struct request_queue *q) +{ +mutex_lock (&q->freeze_lock); +if (!q->q_frozen) { +blk_mq_freeze_queue (q); + q->q_frozen = true; + } +mutex_unlock (&q->freeze_lock); +} +EXPORT_SYMBOL (blk_freeze_queue_lock); +/* * FIXME: replace the scsi_internal_device_*block_nowait() calls in the * mpt3sas driver such that this function can be removed.
block/elevator.c +0 -25
@@ -557,27 +557,6 @@ void elv_bio_merged(struct request_queue *q, struct request *rq, e->type->ops.sq.elevator_bio_merged_fn (q, rq, bio); } -#ifdef CONFIG_PM -static void blk_pm_requeue_request (struct request *rq) -{ -if (rq->q->dev && !(rq->rq_flags &RQF_PM )) - rq->q->nr_pending--; -} - -static void blk_pm_add_request (struct request_queue *q,struct request *rq) -{ -if (q->dev && !(rq->rq_flags &RQF_PM ) && q->nr_pending++ ==0 && - (q->rpm_status ==RPM_SUSPENDED || q->rpm_status ==RPM_SUSPENDING )) -pm_request_resume (q->dev); -} -#else -static inline void blk_pm_requeue_request (struct request *rq) {} -static inline void blk_pm_add_request (struct request_queue *q, -struct request *rq) -{ -} -#endif -void elv_requeue_request (struct request_queue *q,struct request *rq) {/*
@@ -592,8 +571,6 @@ void elv_requeue_request(struct request_queue *q, struct request *rq) rq->rq_flags &= ~RQF_STARTED ; -blk_pm_requeue_request (rq); -__elv_add_request (q, rq,ELEVATOR_INSERT_REQUEUE ); }
@@ -620,8 +597,6 @@ void __elv_add_request(struct request_queue *q, struct request *rq, int where) {trace_block_rq_insert (q, rq); -blk_pm_add_request (q, rq); - rq->q = q;if (rq->rq_flags &RQF_SOFTBARRIER ) {
drivers/scsi/scsi_lib.c +11 -3
@@ -278,12 +278,17 @@ int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,struct request *req;struct scsi_request *rq;int ret =DRIVER_ERROR <<24 ; +bool pm_rq = rq_flags &RQF_PM ; + +if (!pm_rq) +scsi_autopm_get_device (sdev); req =blk_get_request (sdev->host->admin_q, data_direction ==DMA_TO_DEVICE ?REQ_OP_SCSI_OUT :REQ_OP_SCSI_IN ,BLK_MQ_REQ_PREEMPT );if (IS_ERR (req)) -return ret; + goto fail; + rq =scsi_req (req); if (bufflen && blk_rq_map_kern(req->q, req,
@@ -327,6 +332,9 @@ int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,atomic_dec (&sdev->nr_admin_pending);wake_up_all (&sdev->admin_wq); + fail: +if (!pm_rq) +scsi_autopm_put_device (sdev);return ret; }
@@ -3132,7 +3140,7 @@ scsi_device_quiesce(struct scsi_device *sdev) {int err; -blk_mq_freeze_queue (sdev->request_queue); +blk_freeze_queue_lock (sdev->request_queue);mutex_lock (&sdev->state_mutex); err =scsi_device_set_state (sdev,SDEV_QUIESCE );
@@ -3162,7 +3170,7 @@ void scsi_device_resume(struct scsi_device *sdev)scsi_device_set_state (sdev,SDEV_RUNNING );mutex_unlock (&sdev->state_mutex); -blk_mq_unfreeze_queue (sdev->request_queue); +blk_unfreeze_queue_lock (sdev->request_queue); }EXPORT_SYMBOL (scsi_device_resume);
include/linux/blk-mq.h +2 -0
@@ -295,6 +295,8 @@ void blk_freeze_queue_start(struct request_queue *q);void blk_mq_freeze_queue_wait (struct request_queue *q);int blk_mq_freeze_queue_wait_timeout (struct request_queue *q,unsigned long timeout); +void blk_freeze_queue_lock (struct request_queue *q); +void blk_unfreeze_queue_lock (struct request_queue *q);int blk_mq_map_queues (struct blk_mq_tag_set *set);void blk_mq_update_nr_hw_queues (struct blk_mq_tag_set *set,int nr_hw_queues);
include/linux/blkdev.h +3 -0
@@ -635,6 +635,9 @@ struct request_queue {int bypass_depth;atomic_t mq_freeze_depth; +bool q_frozen; +struct mutex freeze_lock; +#if defined(CONFIG_BLK_DEV_BSG )bsg_job_fn *bsg_job_fn;struct bsg_class_device bsg_dev;
Signature · 1 line
2.9.5
Re: [RFC PATCH V2 16/17] block: simplify runtime PM support
Hi Ming
block/blk-core.c int blk_pre_runtime_suspend(struct request_queue *q) · line 3807
Quoted 9 lines · Ming Lei
block/blk-core.c +4
-0@@ -3786,6 +3800,10 @@ int blk_pre_runtime_suspend(struct request_queue *q) q->rpm_status =RPM_SUSPENDING ; }spin_unlock_irq (q->queue_lock); + +if (!ret) +blk_freeze_queue_lock (q); +return ret;
Requests seem to be able enter the request_queue while the rpm_status is RPM_SUSPENDING and miss the rpm_resume.
Even though blk_freeze_queue_lock will drain them before the real suspend start, but looks like we should stop the suspend at the moment.
Thanks
Jianchao
Re: [RFC PATCH V2 16/17] block: simplify runtime PM support
Quoted 15 lines · jianchao.wang
Hi Ming
Review int blk_pre_runtime_suspend(struct request_queue *q) · line 3807
Quoted 9 lines · Ming Lei
(quoted diff) +4
-0@@ -3786,6 +3800,10 @@ int blk_pre_runtime_suspend(struct request_queue *q) q->rpm_status = RPM_SUSPENDING; } spin_unlock_irq(q->queue_lock); + + if (!ret) + blk_freeze_queue_lock(q); + return ret;Requests seem to be able enter the
request_queuewhile therpm_statusisRPM_SUSPENDINGand miss therpm_resume.
Any requests which enters queue during RPM_SUSPENDING will be drained by blk_freeze_queue_lock().
Quoted 3 lines
Even though
blk_freeze_queue_lockwill drain them before the real suspend start, but looks like we should stop the suspend at the moment.
Good point, even though it doesn't affect the correctness of this runtime PM implementation.
We may improve this situation by the following way, what do you think of it?
block/blk-core.c +10 -0
@@ -3803,10 +3803,13 @@ int blk_pre_runtime_suspend(struct request_queue *q) {int ret =0 ;bool busy = true; +unsigned long last_busy;if (!q->dev)return ret; + last_busy =READ_ONCE (dev->power.last_busy); +if (q->mq_ops) busy =blk_mq_pm_queue_busy (q);
@@ -3823,6 +3826,13 @@ int blk_pre_runtime_suspend(struct request_queue *q)if (!ret)blk_freeze_queue_lock (q); +/* +* Any new IO during this window will prevent the current suspend +* from going on +*/ +if (unlikely (last_busy !=READ_ONCE (dev->power.last_busy))) + ret = -EBUSY ; +return ret; }EXPORT_SYMBOL (blk_pre_runtime_suspend);
Thanks,
Ming
Re: [RFC PATCH V2 16/17] block: simplify runtime PM support
Quoted 53 lines · Ming Lei
Quoted 15 lines · jianchao.wang
Hi Ming
Review int blk_pre_runtime_suspend(struct request_queue *q) · line 3807
Quoted 9 lines · Ming Lei
(quoted diff) +4
-0@@ -3786,6 +3800,10 @@ int blk_pre_runtime_suspend(struct request_queue *q) q->rpm_status = RPM_SUSPENDING; } spin_unlock_irq(q->queue_lock); + + if (!ret) + blk_freeze_queue_lock(q); + return ret;Requests seem to be able enter the
request_queuewhile therpm_statusisRPM_SUSPENDINGand miss therpm_resume.Any requests which enters queue during
RPM_SUSPENDINGwill be drained byblk_freeze_queue_lock().Quoted 3 lines
Even though
blk_freeze_queue_lockwill drain them before the real suspend start, but looks like we should stop the suspend at the moment.Good point, even though it doesn't affect the correctness of this runtime PM implementation.
We may improve this situation by the following way, what do you think of it?
block/blk-core.c +3
-0@@ -3803,10 +3803,13 @@ int blk_pre_runtime_suspend(struct request_queue *q) {int ret =0 ;bool busy = true; +unsigned long last_busy;if (!q->dev)return ret; + last_busy =READ_ONCE (dev->power.last_busy); +if (q->mq_ops) busy =blk_mq_pm_queue_busy (q);@@ -3823,6 +3826,13 @@ int blk_pre_runtime_suspend(struct request_queue *q)if (!ret)blk_freeze_queue_lock (q);
block/blk-core.c int blk_pre_runtime_suspend(struct request_queue *q) · line 3838
Quoted 10 lines
block/blk-core.c +7
-0+/* +* Any new IO during this window will prevent the current suspend +* from going on +*/ +if (unlikely (last_busy !=READ_ONCE (dev->power.last_busy))) + ret = -EBUSY ; +return ret; }EXPORT_SYMBOL (blk_pre_runtime_suspend);
Wow, this is great method !
Maybe we should add checking as following:
+ last_busy = READ_ONCE(dev->power.last_busy);
+ if (last_busy == jiffies)
+ return -EBUSY
if (q->mq_ops)
busy = blk_mq_pm_queue_busy(q);
Thanks
Jianchao
Quoted 3 lines
Thanks,
Ming
Re: [RFC PATCH V2 16/17] block: simplify runtime PM support
Quoted 71 lines · jianchao.wang
Quoted 53 lines · Ming Lei
Quoted 15 lines · jianchao.wang
Hi Ming
Review int blk_pre_runtime_suspend(struct request_queue *q) · line 3807
Quoted 9 lines · Ming Lei
(quoted diff) +4
-0@@ -3786,6 +3800,10 @@ int blk_pre_runtime_suspend(struct request_queue *q) q->rpm_status = RPM_SUSPENDING; } spin_unlock_irq(q->queue_lock); + + if (!ret) + blk_freeze_queue_lock(q); + return ret;Requests seem to be able enter the
request_queuewhile therpm_statusisRPM_SUSPENDINGand miss therpm_resume.Any requests which enters queue during
RPM_SUSPENDINGwill be drained byblk_freeze_queue_lock().Quoted 3 lines
Even though
blk_freeze_queue_lockwill drain them before the real suspend start, but looks like we should stop the suspend at the moment.Good point, even though it doesn't affect the correctness of this runtime PM implementation.
We may improve this situation by the following way, what do you think of it?
block/blk-core.c +3
-0@@ -3803,10 +3803,13 @@ int blk_pre_runtime_suspend(struct request_queue *q) {int ret =0 ;bool busy = true; +unsigned long last_busy;if (!q->dev)return ret; + last_busy =READ_ONCE (dev->power.last_busy); +if (q->mq_ops) busy =blk_mq_pm_queue_busy (q);@@ -3823,6 +3826,13 @@ int blk_pre_runtime_suspend(struct request_queue *q)if (!ret)blk_freeze_queue_lock (q);block/blk-core.c int blk_pre_runtime_suspend(struct request_queue *q) · line 3838
Quoted 10 lines
block/blk-core.c +7
-0+/* +* Any new IO during this window will prevent the current suspend +* from going on +*/ +if (unlikely (last_busy !=READ_ONCE (dev->power.last_busy))) + ret = -EBUSY ; +return ret; }EXPORT_SYMBOL (blk_pre_runtime_suspend);Wow, this is great method !
Maybe we should add checking as following:
Review
Quoted 3 lines
(quoted diff) +3
-0+ last_busy = READ_ONCE(dev->power.last_busy); + if (last_busy == jiffies) + return -EBUSY
We may do that, but seems not necessary given the .runtime_suspend() (scsi_runtime_suspend) is just called when the timer is expired.
As we know, even there is request entering queue during the suspend window, not a big deal, the current approach still works fine.
Thanks,
Ming
Re: [RFC PATCH V2 16/17] block: simplify runtime PM support
Quoted 5 lines · Ming Lei
We may do that, but seems not necessary given the .runtime_suspend() (
scsi_runtime_suspend) is just called when the timer is expired.As we know, even there is request entering queue during the suspend window, not a big deal, the current approach still works fine.
Yes, indeed. :)
Thanks
Jianchao
RFC PATCH v2 17/17 block: enable runtime PM for blk-mq
In reply to Ming LeiNow blk-mq can borrow the runtime PM approach from legacy path, so enable it simply. The only difference with legacy is that:
1) blk_mq_queue_sched_tag_busy_iter() is introduced for checking if queue is idle, instead of maintaining one counter.
2) we have to iterate over scheduler tags for counting how many requests entering queue because requests in hw tags don't cover these allocated and not dispatched.
--- block/blk-core.c | 29 ++++++++++++++++++++++++----- block/blk-mq-tag.c | 21 +++++++++++++++++++-- block/blk-mq-tag.h | 2 ++ block/blk-mq.c | 4 ++++ 4 files changed, 49 insertions(+), 7 deletions(-)
block/blk-core.c +24 -5
@@ -3751,11 +3751,8 @@ EXPORT_SYMBOL(blk_finish_plug);*/ void blk_pm_runtime_init (struct request_queue *q,struct device *dev) { -/* Don't enable runtime PM for blk-mq until it is ready */ -if (q->mq_ops) { -pm_runtime_disable (dev); +if (WARN_ON_ONCE (blk_queue_admin (q)))return ; - } q->dev = dev; q->rpm_status =RPM_ACTIVE ;
@@ -3764,6 +3761,23 @@ void blk_pm_runtime_init(struct request_queue *q, struct device *dev) }EXPORT_SYMBOL (blk_pm_runtime_init); +static void blk_mq_pm_count_req (struct blk_mq_hw_ctx *hctx, +struct request *rq,void *priv,bool reserved) +{ +unsigned long *cnt = priv; + + (*cnt)++; +} + +static bool blk_mq_pm_queue_busy (struct request_queue *q) +{ +unsigned long cnt =0 ; + +blk_mq_queue_sched_tag_busy_iter (q, blk_mq_pm_count_req, &cnt); + +return cnt >0 ; +} +/** * blk_pre_runtime_suspend - Pre runtime suspend check * @q: the queue of the device
@@ -3788,12 +3802,17 @@ EXPORT_SYMBOL(blk_pm_runtime_init);int blk_pre_runtime_suspend (struct request_queue *q) {int ret =0 ; +bool busy = true;if (!q->dev)return ret; +if (q->mq_ops) + busy =blk_mq_pm_queue_busy (q); +spin_lock_irq (q->queue_lock); -if (q->nr_pending) { + busy = q->mq_ops ? busy : !!q->nr_pending; +if (busy) { ret = -EBUSY ;pm_runtime_mark_last_busy (q->dev); }else {
block/blk-mq-tag.c +19 -2
@@ -316,8 +316,8 @@ void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset, }EXPORT_SYMBOL (blk_mq_tagset_busy_iter); -void blk_mq_queue_tag_busy_iter (struct request_queue *q,busy_iter_fn *fn, -void *priv) +static void __blk_mq_queue_tag_busy_iter (struct request_queue *q, +busy_iter_fn *fn,void *priv,bool sched_tag) {struct blk_mq_hw_ctx *hctx;int i;
@@ -326,6 +326,9 @@ void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,queue_for_each_hw_ctx (q, hctx, i) {struct blk_mq_tags *tags = hctx->tags; +if (sched_tag && hctx->sched_tags) + tags = hctx->sched_tags; +/* * If not software queues are currently mapped to this * hardware queue, there's nothing to check
@@ -340,6 +343,20 @@ void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn, } +void blk_mq_queue_tag_busy_iter (struct request_queue *q, +busy_iter_fn *fn,void *priv) +{ + +__blk_mq_queue_tag_busy_iter (q, fn, priv, false); +} + +void blk_mq_queue_sched_tag_busy_iter (struct request_queue *q, +busy_iter_fn *fn,void *priv) +{ + +__blk_mq_queue_tag_busy_iter (q, fn, priv, true); +} +static int bt_alloc (struct sbitmap_queue *bt,unsigned int depth,bool round_robin,int node) {
block/blk-mq-tag.h +2 -0
@@ -35,6 +35,8 @@ extern int blk_mq_tag_update_depth(struct blk_mq_hw_ctx *hctx,extern void blk_mq_tag_wakeup_all (struct blk_mq_tags *tags,bool );void blk_mq_queue_tag_busy_iter (struct request_queue *q,busy_iter_fn *fn,void *priv); +void blk_mq_queue_sched_tag_busy_iter (struct request_queue *q,busy_iter_fn *fn, +void *priv);static inline struct sbq_wait_state *bt_wait_ptr (struct sbitmap_queue *bt,struct blk_mq_hw_ctx *hctx)
block/blk-mq.c +4 -0
@@ -25,6 +25,7 @@#include <linux/delay.h> #include <linux/crash_dump.h> #include <linux/prefetch.h> +#include <linux/pm_runtime.h> #include <trace/events/block.h>
@@ -503,6 +504,9 @@ static void __blk_mq_free_request(struct request *rq)blk_mq_put_tag (hctx, hctx->sched_tags, ctx, sched_tag);blk_mq_sched_restart (hctx);blk_queue_exit (q); + +if (q->dev) +pm_runtime_mark_last_busy (q->dev); }void blk_mq_free_request (struct request *rq)
Signature · 1 line
2.9.5