blk-mq: improvement on handling IO during CPU hotplug
Subsystems (3)
v3 cover blk-mq: improvement on handling IO during CPU hotplug
Hi,
Thomas mentioned:
"
That was the constraint of managed interrupts from the very beginning:
The driver/subsystem has to quiesce the interrupt line and the associated queue _before_ it gets shutdown in CPU unplug and not fiddle with it until it's restarted by the core when the CPU is plugged in again.
"
But no drivers or blk-mq do that before one hctx becomes dead(all CPUs for one hctx are offline), and even it is worse, blk-mq stills tries to run hw queue after hctx is dead, see blk_mq_hctx_notify_dead().
This patchset tries to address the issue by two stages:
1) add one new cpuhp state of CPUHP_AP_BLK_MQ_ONLINE
- mark the hctx as internal stopped, and drain all in-flight requests if the hctx is going to be dead.
2) re-submit IO in the state of CPUHP_BLK_MQ_DEAD after the hctx becomes dead
- steal bios from the request, and resubmit them via generic_make_request(), then these IO will be mapped to other live hctx for dispatch
Please comment & review, thanks!
John, I don't add your tested-by tag since V3 have some changes, and I appreciate if you may run your test on V3.
V3:
- re-organize patch 2 & 3 a bit for addressing Hannes's comment
- fix patch 4 for avoiding potential deadlock, as found by Hannes
V2:
- patch4 & patch 5 in V1 have been merged to block tree, so remove them
- address comments from John Garry and Minwoo
Ming Lei (5):
blk-mq: add new state of BLK_MQ_S_INTERNAL_STOPPED
blk-mq: prepare for draining IO when hctx's all CPUs are offline
blk-mq: stop to handle IO and drain IO before hctx becomes dead
blk-mq: re-submit IO in case that hctx is dead
blk-mq: handle requests dispatched from IO scheduler in case that hctx
is dead
block/blk-mq-debugfs.c | 2 + block/blk-mq-tag.c | 2 +- block/blk-mq-tag.h | 2 + block/blk-mq.c | 135 +++++++++++++++++++++++++++++++++---- block/blk-mq.h | 3 +- drivers/block/loop.c | 2 +- drivers/md/dm-rq.c | 2 +- include/linux/blk-mq.h | 5 ++ include/linux/cpuhotplug.h | 1 + 9 files changed, 138 insertions(+), 16 deletions(-)
Signature · 1 line
2.20.1
PATCH v3 1/5 blk-mq: add new state of BLK_MQ_S_INTERNAL_STOPPED
Add a new hw queue state of BLK_MQ_S_INTERNAL_STOPPED, which prepares for stopping hw queue before all CPUs of this hctx become offline.
We can't reuse BLK_MQ_S_STOPPED because that state can be cleared during IO completion.
--- block/blk-mq-debugfs.c | 1 + block/blk-mq.h | 3 ++- include/linux/blk-mq.h | 3 +++ 3 files changed, 6 insertions(+), 1 deletion(-)
block/blk-mq-debugfs.c +1 -0
@@ -213,6 +213,7 @@ static const char *const hctx_state_name[] = {HCTX_STATE_NAME (STOPPED ),HCTX_STATE_NAME (TAG_ACTIVE ),HCTX_STATE_NAME (SCHED_RESTART ), + HCTX_STATE_NAME(INTERNAL_STOPPED), };#undef HCTX_STATE_NAME
block/blk-mq.h +2 -1
@@ -176,7 +176,8 @@ static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_datastatic inline bool blk_mq_hctx_stopped (struct blk_mq_hw_ctx *hctx) { -return test_bit (BLK_MQ_S_STOPPED , &hctx->state); +return test_bit (BLK_MQ_S_STOPPED , &hctx->state) || +test_bit (BLK_MQ_S_INTERNAL_STOPPED , &hctx->state); }static inline bool blk_mq_hw_queue_mapped (struct blk_mq_hw_ctx *hctx)
include/linux/blk-mq.h +3 -0
@@ -235,6 +235,9 @@ enum {BLK_MQ_S_TAG_ACTIVE =1 ,BLK_MQ_S_SCHED_RESTART =2 , +/* hw queue is internal stopped, driver do not use it */ +BLK_MQ_S_INTERNAL_STOPPED =3 , +BLK_MQ_MAX_DEPTH =10240 , BLK_MQ_CPU_WORK_BATCH = 8,
Signature · 1 line
2.20.1
PATCH v3 2/5 blk-mq: prepare for draining IO when hctx's all CPUs are offline
In reply to Ming LeiMost of blk-mq drivers depend on managed IRQ's auto-affinity to setup up queue mapping. Thomas mentioned the following point[1]:
"
That was the constraint of managed interrupts from the very beginning:
The driver/subsystem has to quiesce the interrupt line and the associated queue _before_ it gets shutdown in CPU unplug and not fiddle with it until it's restarted by the core when the CPU is plugged in again.
"
However, current blk-mq implementation doesn't quiesce hw queue before the last CPU in the hctx is shutdown. Even worse, CPUHP_BLK_MQ_DEAD is one cpuhp state handled after the CPU is down, so there isn't any chance to quiesce hctx for blk-mq wrt. CPU hotplug.
Add new cpuhp state of CPUHP_AP_BLK_MQ_ONLINE for blk-mq to stop queues and wait for completion of in-flight requests.
We will stop hw queue and wait for completion of in-flight requests when one hctx is becoming dead in the following patch. This way may cause dead-lock for some stacking blk-mq drivers, such as dm-rq and loop.
Add blk-mq flag of BLK_MQ_F_NO_MANAGED_IRQ and mark it for dm-rq and loop, so we needn't to wait for completion of in-flight requests from dm-rq & loop, then the potential dead-lock can be avoided.
[1] https://lore.kernel.org/linux-block/[email protected]/
--- block/blk-mq-debugfs.c | 1 + block/blk-mq.c | 13 +++++++++++++ drivers/block/loop.c | 2 +- drivers/md/dm-rq.c | 2 +- include/linux/blk-mq.h | 2 ++ include/linux/cpuhotplug.h | 1 + 6 files changed, 19 insertions(+), 2 deletions(-)
block/blk-mq-debugfs.c +1 -0
@@ -240,6 +240,7 @@ static const char *const hctx_flag_name[] = {HCTX_FLAG_NAME (TAG_SHARED ),HCTX_FLAG_NAME (BLOCKING ),HCTX_FLAG_NAME (NO_SCHED ), + HCTX_FLAG_NAME(NO_MANAGED_IRQ), };#undef HCTX_FLAG_NAME
block/blk-mq.c +13 -0
@@ -2225,6 +2225,11 @@ int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,return -ENOMEM ; } +static int blk_mq_hctx_notify_online (unsigned int cpu,struct hlist_node *node) +{ +return 0 ; +} +/* * 'cpu' is going away. splice any existing rq_list entries from this * software queue to the hw queue dispatch list, and ensure that it
@@ -2261,6 +2266,9 @@ static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node)static void blk_mq_remove_cpuhp (struct blk_mq_hw_ctx *hctx) { +if (!(hctx->flags &BLK_MQ_F_NO_MANAGED_IRQ )) +cpuhp_state_remove_instance_nocalls (CPUHP_AP_BLK_MQ_ONLINE , + &hctx->cpuhp_online);cpuhp_state_remove_instance_nocalls (CPUHP_BLK_MQ_DEAD , &hctx->cpuhp_dead); }
@@ -2320,6 +2328,9 @@ static int blk_mq_init_hctx(struct request_queue *q, { hctx->queue_num = hctx_idx; +if (!(hctx->flags &BLK_MQ_F_NO_MANAGED_IRQ )) +cpuhp_state_add_instance_nocalls (CPUHP_AP_BLK_MQ_ONLINE , + &hctx->cpuhp_online);cpuhp_state_add_instance_nocalls (CPUHP_BLK_MQ_DEAD , &hctx->cpuhp_dead); hctx->tags = set->tags[hctx_idx];
@@ -3547,6 +3558,8 @@ static int __init blk_mq_init(void) {cpuhp_setup_state_multi (CPUHP_BLK_MQ_DEAD ,"block/mq:dead" ,NULL , blk_mq_hctx_notify_dead); +cpuhp_setup_state_multi (CPUHP_AP_BLK_MQ_ONLINE ,"block/mq:online" , +NULL , blk_mq_hctx_notify_online);return 0 ; }subsys_initcall (blk_mq_init);
drivers/block/loop.c +1 -1
@@ -1999,7 +1999,7 @@ static int loop_add(struct loop_device **l, int i) lo->tag_set.queue_depth =128 ; lo->tag_set.numa_node =NUMA_NO_NODE ; lo->tag_set.cmd_size =sizeof (struct loop_cmd ); - lo->tag_set.flags =BLK_MQ_F_SHOULD_MERGE ; + lo->tag_set.flags =BLK_MQ_F_SHOULD_MERGE |BLK_MQ_F_NO_MANAGED_IRQ ; lo->tag_set.driver_data = lo; err =blk_mq_alloc_tag_set (&lo->tag_set);
drivers/md/dm-rq.c +1 -1
@@ -547,7 +547,7 @@ int dm_mq_init_request_queue(struct mapped_device *md, struct dm_table *t) md->tag_set->ops = &dm_mq_ops; md->tag_set->queue_depth =dm_get_blk_mq_queue_depth (); md->tag_set->numa_node = md->numa_node_id; - md->tag_set->flags =BLK_MQ_F_SHOULD_MERGE ; + md->tag_set->flags =BLK_MQ_F_SHOULD_MERGE |BLK_MQ_F_NO_MANAGED_IRQ ; md->tag_set->nr_hw_queues =dm_get_blk_mq_nr_hw_queues (); md->tag_set->driver_data = md;
include/linux/blk-mq.h +2 -0
@@ -58,6 +58,7 @@ struct blk_mq_hw_ctx {atomic_t nr_active; +struct hlist_node cpuhp_online;struct hlist_node cpuhp_dead;struct kobject kobj;
@@ -226,6 +227,7 @@ struct blk_mq_ops {enum {BLK_MQ_F_SHOULD_MERGE =1 <<0 ,BLK_MQ_F_TAG_SHARED =1 <<1 , +BLK_MQ_F_NO_MANAGED_IRQ =1 <<2 ,BLK_MQ_F_BLOCKING =1 <<5 ,BLK_MQ_F_NO_SCHED =1 <<6 ,BLK_MQ_F_ALLOC_POLICY_START_BIT =8 ,
include/linux/cpuhotplug.h +1 -0
@@ -147,6 +147,7 @@ enum cpuhp_state {CPUHP_AP_SMPBOOT_THREADS ,CPUHP_AP_X86_VDSO_VMA_ONLINE ,CPUHP_AP_IRQ_AFFINITY_ONLINE , +CPUHP_AP_BLK_MQ_ONLINE ,CPUHP_AP_ARM_MVEBU_SYNC_CLOCKS ,CPUHP_AP_X86_INTEL_EPB_ONLINE , CPUHP_AP_PERF_ONLINE,
Signature · 1 line
2.20.1
PATCH v3 3/5 blk-mq: stop to handle IO and drain IO before hctx becomes dead
In reply to Ming LeiBefore one CPU becomes offline, check if it is the last online CPU of hctx. If yes, mark this hctx as BLK_MQ_S_INTERNAL_STOPPED, meantime wait for completion of all in-flight IOs originated from this hctx.
This way guarantees that there isn't any inflight IO before shutdowning the managed IRQ line.
--- block/blk-mq-tag.c | 2 +- block/blk-mq-tag.h | 2 ++ block/blk-mq.c | 40 ++++++++++++++++++++++++++++++++++++++++ 3 files changed, 43 insertions(+), 1 deletion(-)
block/blk-mq-tag.c +1 -1
@@ -325,7 +325,7 @@ static void bt_tags_for_each(struct blk_mq_tags *tags, struct sbitmap_queue *bt,* true to continue iterating tags, false to stop. * @priv: Will be passed as second argument to @fn. */ -static void blk_mq_all_tag_busy_iter (struct blk_mq_tags *tags, +void blk_mq_all_tag_busy_iter (struct blk_mq_tags *tags,busy_tag_iter_fn *fn,void *priv) {if (tags->nr_reserved_tags)
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_all_tag_busy_iter (struct blk_mq_tags *tags, +busy_tag_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 +40 -0
@@ -2225,8 +2225,46 @@ int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,return -ENOMEM ; } +static bool blk_mq_count_inflight_rq (struct request *rq,void *data, +bool reserved) +{ +unsigned *count = data; + +if ((blk_mq_rq_state (rq) ==MQ_RQ_IN_FLIGHT )) + (*count)++; + +return true; +} + +static unsigned blk_mq_tags_inflight_rqs (struct blk_mq_tags *tags) +{ +unsigned count =0 ; + +blk_mq_all_tag_busy_iter (tags, blk_mq_count_inflight_rq, &count); + +return count; +} + +static void blk_mq_hctx_drain_inflight_rqs (struct blk_mq_hw_ctx *hctx) +{ +while (1 ) { +if (!blk_mq_tags_inflight_rqs (hctx->tags)) +break ; +msleep (5 ); + } +} +static int blk_mq_hctx_notify_online (unsigned int cpu,struct hlist_node *node) { +struct blk_mq_hw_ctx *hctx =hlist_entry_safe (node, + struct blk_mq_hw_ctx, cpuhp_online); + +if ((cpumask_next_and (-1 , hctx->cpumask, cpu_online_mask) == cpu) && + (cpumask_next_and (cpu, hctx->cpumask, cpu_online_mask) >= + nr_cpu_ids)) { +set_bit (BLK_MQ_S_INTERNAL_STOPPED , &hctx->state); +blk_mq_hctx_drain_inflight_rqs (hctx); + }return 0 ; }
@@ -2246,6 +2284,8 @@ static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node) ctx =__blk_mq_get_ctx (hctx->queue, cpu); type = hctx->type; +clear_bit (BLK_MQ_S_INTERNAL_STOPPED , &hctx->state); +spin_lock (&ctx->lock);if (!list_empty (&ctx->rq_lists[type])) {list_splice_init (&ctx->rq_lists[type], &tmp);
Signature · 1 line
2.20.1
Re: [PATCH V3 3/5] blk-mq: stop to handle IO and drain IO before hctx becomes dead
Quoted 7 lines · Ming Lei
Before one CPU becomes offline, check if it is the last online CPU of hctx. If yes, mark this hctx as
BLK_MQ_S_INTERNAL_STOPPED, meantime wait for completion of all in-flight IOs originated from this hctx.This way guarantees that there isn't any inflight IO before shutdowning the managed IRQ line.
Quoted 6 lines
Cc John Garry <[email protected]>Cc Bart Van Assche <[email protected]>Cc Hannes Reinecke <[email protected]>Cc Christoph Hellwig <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Keith Busch <[email protected]>
Apart from nits, below, FWIW:
Quoted 38 lines
Signed-off-by Ming Lei <[email protected]>--- block/blk-mq-tag.c | 2 +- block/blk-mq-tag.h | 2 ++ block/blk-mq.c | 40 ++++++++++++++++++++++++++++++++++++++++ 3 files changed, 43 insertions(+), 1 deletion(-)block/blk-mq-tag.c +1
-1@@ -325,7 +325,7 @@ static void bt_tags_for_each(struct blk_mq_tags *tags, struct sbitmap_queue *bt,* true to continue iterating tags, false to stop. * @priv: Will be passed as second argument to @fn. */ -static void blk_mq_all_tag_busy_iter (struct blk_mq_tags *tags, +void blk_mq_all_tag_busy_iter (struct blk_mq_tags *tags,busy_tag_iter_fn *fn,void *priv) {if (tags->nr_reserved_tags)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_all_tag_busy_iter (struct blk_mq_tags *tags, +busy_tag_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 +0
-0
block/blk-mq.c int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags, · line 2233
Quoted 9 lines
block/blk-mq.c +6
-0@@ -2225,8 +2225,46 @@ int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,return -ENOMEM ; } +static bool blk_mq_count_inflight_rq (struct request *rq,void *data, +bool reserved) +{ +unsigned *count = data; + +if ((blk_mq_rq_state (rq) ==MQ_RQ_IN_FLIGHT ))
nit: extra parentheses
Quoted 18 lines
block/blk-mq.c +18
-0+ (*count)++; + +return true; +} + +static unsigned blk_mq_tags_inflight_rqs (struct blk_mq_tags *tags) +{ +unsigned count =0 ; + +blk_mq_all_tag_busy_iter (tags, blk_mq_count_inflight_rq, &count); + +return count; +} + +static void blk_mq_hctx_drain_inflight_rqs (struct blk_mq_hw_ctx *hctx) +{ +while (1 ) { +if (!blk_mq_tags_inflight_rqs (hctx->tags))
block/blk-mq.c int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags, · line 2261
Quoted 10 lines
block/blk-mq.c +8
-0+break ; +msleep (5 ); + } +} +static int blk_mq_hctx_notify_online (unsigned int cpu,struct hlist_node *node) { +struct blk_mq_hw_ctx *hctx =hlist_entry_safe (node, + struct blk_mq_hw_ctx, cpuhp_online); +
nit: we could make this a little neater by using an intermediate variable for hctx->cpumask:
struct cpumask *cpumask = hctx->cpumask; if ((cpumask_next_and(-1, cpumask, cpu_online_mask) == cpu) && ...
Quoted 19 lines
block/blk-mq.c +8
-0+if ((cpumask_next_and (-1 , hctx->cpumask, cpu_online_mask) == cpu) && + (cpumask_next_and (cpu, hctx->cpumask, cpu_online_mask) >= + nr_cpu_ids)) { +set_bit (BLK_MQ_S_INTERNAL_STOPPED , &hctx->state); +blk_mq_hctx_drain_inflight_rqs (hctx); + }return 0 ; }@@ -2246,6 +2284,8 @@ static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node) ctx =__blk_mq_get_ctx (hctx->queue, cpu); type = hctx->type; +clear_bit (BLK_MQ_S_INTERNAL_STOPPED , &hctx->state); +spin_lock (&ctx->lock);if (!list_empty (&ctx->rq_lists[type])) {list_splice_init (&ctx->rq_lists[type], &tmp);
PATCH v3 4/5 blk-mq: re-submit IO in case that hctx is dead
In reply to Ming LeiWhen all CPUs in one hctx are offline, we shouldn't run this hw queue for completing request any more.
So steal bios from the request, and resubmit them, and finally free the request in blk_mq_hctx_notify_dead().
--- block/blk-mq.c | 52 +++++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 45 insertions(+), 7 deletions(-)
block/blk-mq.c +45 -7
@@ -2268,10 +2268,34 @@ static int blk_mq_hctx_notify_online(unsigned int cpu, struct hlist_node *node)return 0 ; } +static void blk_mq_resubmit_io (struct request *rq) +{ +struct bio_list list; +struct bio *bio; + +bio_list_init (&list); +blk_steal_bios (&list, rq); + +/* +* Free the old empty request before submitting bio for avoiding +* potential deadlock +*/ +blk_mq_cleanup_rq (rq); +blk_mq_end_request (rq,0 ); + +while (true) { + bio =bio_list_pop (&list); +if (!bio) +break ; + +generic_make_request (bio); + } +} +/* -* 'cpu' is going away. splice any existing rq_list entries from this -* software queue to the hw queue dispatch list, and ensure that it -* gets run. +* 'cpu' has gone away. If this hctx is dead, we can't dispatch request +* to the hctx any more, so steal bios from requests of this hctx, and +* re-submit them to the request queue, and free these requests finally. */ static int blk_mq_hctx_notify_dead (unsigned int cpu,struct hlist_node *node) {
@@ -2279,6 +2303,8 @@ static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node)struct blk_mq_ctx *ctx;LIST_HEAD (tmp);enum hctx_type type; +bool hctx_dead; +struct request *rq; hctx = hlist_entry_safe(node, struct blk_mq_hw_ctx, cpuhp_dead); ctx =__blk_mq_get_ctx (hctx->queue, cpu);
@@ -2286,6 +2312,9 @@ static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node)clear_bit (BLK_MQ_S_INTERNAL_STOPPED , &hctx->state); + hctx_dead =cpumask_first_and (hctx->cpumask, cpu_online_mask) >= + nr_cpu_ids; +spin_lock (&ctx->lock);if (!list_empty (&ctx->rq_lists[type])) {list_splice_init (&ctx->rq_lists[type], &tmp);
@@ -2296,11 +2325,20 @@ static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node)if (list_empty (&tmp))return 0 ; -spin_lock (&hctx->lock); -list_splice_tail_init (&tmp, &hctx->dispatch); -spin_unlock (&hctx->lock); +if (!hctx_dead) { +spin_lock (&hctx->lock); +list_splice_tail_init (&tmp, &hctx->dispatch); +spin_unlock (&hctx->lock); +blk_mq_run_hw_queue (hctx, true); +return 0 ; + } + +while (!list_empty (&tmp)) { + rq = list_entry(tmp.next, struct request, queuelist); +list_del_init (&rq->queuelist); +blk_mq_resubmit_io (rq); + } -blk_mq_run_hw_queue (hctx, true);return 0 ; }
Signature · 1 line
2.20.1
PATCH v3 5/5 blk-mq: handle requests dispatched from IO scheduler in case that hctx is dead
In reply to Ming LeiIf hctx becomes dead, all in-queue IO requests aimed at this hctx have to be re-submitted, so cover requests queued in scheduler queue.
--- block/blk-mq.c | 30 +++++++++++++++++++++++++----- 1 file changed, 25 insertions(+), 5 deletions(-)
block/blk-mq.c +25 -5
@@ -2305,6 +2305,7 @@ static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node)enum hctx_type type;bool hctx_dead;struct request *rq; +struct elevator_queue *e; hctx = hlist_entry_safe(node, struct blk_mq_hw_ctx, cpuhp_dead); ctx =__blk_mq_get_ctx (hctx->queue, cpu);
@@ -2315,12 +2316,31 @@ static int blk_mq_hctx_notify_dead(unsigned int cpu, struct hlist_node *node) hctx_dead =cpumask_first_and (hctx->cpumask, cpu_online_mask) >= nr_cpu_ids; -spin_lock (&ctx->lock); -if (!list_empty (&ctx->rq_lists[type])) { -list_splice_init (&ctx->rq_lists[type], &tmp); -blk_mq_hctx_clear_pending (hctx, ctx); + e = hctx->queue->elevator; +if (!e) { +spin_lock (&ctx->lock); +if (!list_empty (&ctx->rq_lists[type])) { +list_splice_init (&ctx->rq_lists[type], &tmp); +blk_mq_hctx_clear_pending (hctx, ctx); + } +spin_unlock (&ctx->lock); + }else if (hctx_dead) { +LIST_HEAD (sched_tmp); + +while ((rq = e->type->ops.dispatch_request (hctx))) { +if (rq->mq_hctx != hctx) +list_add (&rq->queuelist, &sched_tmp); +else +list_add (&rq->queuelist, &tmp); + } + +while (!list_empty (&sched_tmp)) { + rq = list_entry(sched_tmp.next, struct request, + queuelist); +list_del_init (&rq->queuelist); +blk_mq_sched_insert_request (rq, true, true, true); + } } -spin_unlock (&ctx->lock);if (list_empty (&tmp))return 0 ;
Signature · 1 line
2.20.1
Quoted 24 lines · Ming Lei
Hi,
Thomas mentioned:
"
That was the constraint of managed interrupts from the very beginning:The driver/subsystem has to quiesce the interrupt line and the associated queue _before_ it gets shutdown in CPU unplug and not fiddle with it until it's restarted by the core when the CPU is plugged in again.
"But no drivers or blk-mq do that before one hctx becomes dead(all CPUs for one hctx are offline), and even it is worse, blk-mq stills tries to run hw queue after hctx is dead, see
blk_mq_hctx_notify_dead().This patchset tries to address the issue by two stages:
1) add one new cpuhp state of
CPUHP_AP_BLK_MQ_ONLINE- mark the hctx as internal stopped, and drain all in-flight requests if the hctx is going to be dead.
2) re-submit IO in the state of
CPUHP_BLK_MQ_DEADafter the hctx becomes dead
Quoted 8 lines
- steal bios from the request, and resubmit them via
generic_make_request(), then these IO will be mapped to other live hctx for dispatchPlease comment & review, thanks!
John, I don't add your tested-by tag since V3 have some changes, and I appreciate if you may run your test on V3.
Will do, Thanks
Quoted 8 lines
V3:
- re-organize patch 2 & 3 a bit for addressing Hannes's comment
- fix patch 4 for avoiding potential deadlock, as found by HannesV2:
- patch4 & patch 5 in V1 have been merged to block tree, so remove them
- addres
Quoted 33 lines · John Garry
Quoted 32 lines · Ming Lei
Hi,
Thomas mentioned:
"
That was the constraint of managed interrupts from the very
beginning:The driver/subsystem has to quiesce the interrupt line and the
associated
queue _before_ it gets shutdown in CPU unplug and not fiddle
with it
until it's restarted by the core when the CPU is plugged in again.
"But no drivers or blk-mq do that before one hctx becomes dead(all CPUs for one hctx are offline), and even it is worse, blk-mq stills tries to run hw queue after hctx is dead, see
blk_mq_hctx_notify_dead().This patchset tries to address the issue by two stages:
1) add one new cpuhp state of
CPUHP_AP_BLK_MQ_ONLINE- mark the hctx as internal stopped, and drain all in-flight requests if the hctx is going to be dead.
2) re-submit IO in the state of
CPUHP_BLK_MQ_DEADafter the hctx becomes dead- steal bios from the request, and resubmit them via
generic_make_request(),
then these IO will be mapped to other live hctx for dispatch
Quoted 7 lines
Quoted 5 lines
Please comment & review, thanks!
John, I don't add your tested-by tag since V3 have some changes, and I appreciate if you may run your test on V3.
Will do, Thanks
Hi Ming,
I got this warning once:
[ 162.558185] CPU10: shutdown [ 162.560994] psci: CPU10 killed. [ 162.593939] CPU9: shutdown [ 162.596645] psci: CPU9 killed. [ 162.625838] CPU8: shutdown [ 162.628550] psci: CPU8 killed. [ 162.685790] CPU7: shutdown [ 162.688496] psci: CPU7 killed. [ 162.725771] CPU6: shutdown [ 162.728486] psci: CPU6 killed. [ 162.753884] CPU5: shutdown [ 162.756591] psci: CPU5 killed. [ 162.785584] irq_shutdown [ 162.788277] IRQ 800: no longer affine to CPU4 [ 162.793267] CPU4: shutdown [ 162.795975] psci: CPU4 killed. [ 162.849680] run queue from wrong CPU 13, hctx active [ 162.849692] CPU3: shutdown [ 162.854649] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 162.854653] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 162.857362] psci: CPU3 killed. [ 162.866039] Workqueue: kblockd blk_mq_run_work_fn [ 162.882281] Call trace: [ 162.884716] dump_backtrace+0x0/0x150 [ 162.888365] show_stack+0x14/0x20 [ 162.891668] dump_stack+0xb0/0xf8 [ 162.894970] __blk_mq_run_hw_queue+0x11c/0x128 [ 162.899400] blk_mq_run_work_fn+0x1c/0x28 [ 162.903397] process_one_work+0x1e0/0x358 [ 162.907393] worker_thread+0x40/0x488 [ 162.911042] kthread+0x118/0x120 [ 162.914257] ret_from_fork+0x10/0x18 [ 162.917834] run queue from wrong CPU 13, hctx active [ 162.922789] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 162.931472] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 162.939983] Workqueue: kblockd blk_mq_run_work_fn [ 162.944674] Call trace: [ 162.947107] dump_backtrace+0x0/0x150 [ 162.950755] show_stack+0x14/0x20 [ 162.954057] dump_stack+0xb0/0xf8 [ 162.957359] __blk_mq_run_hw_queue+0x11c/0x128 [ 162.961788] blk_mq_run_work_fn+0x1c/0x28 [ 162.965784] process_one_work+0x1e0/0x358 [ 162.969780] worker_thread+0x40/0x488 [ 162.973429] kthread+0x118/0x120 [ 162.976644] ret_from_fork+0x10/0x18 [ 162.980214] run queue from wrong CPU 13, hctx active [ 162.985171] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 162.993853] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.002366] Workqueue: kblockd blk_mq_run_work_fn [ 163.007057] Call trace: [ 163.009490] dump_backtrace+0x0/0x150 [ 163.013138] show_stack+0x14/0x20 [ 163.016440] dump_stack+0xb0/0xf8 [ 163.019742] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.024172] blk_mq_run_work_fn+0x1c/0x28 [ 163.028167] process_one_work+0x1e0/0x358 [ 163.032163] worker_thread+0x238/0x488 [ 163.035899] kthread+0x118/0x120 [ 163.039113] ret_from_fork+0x10/0x18 [ 163.042736] run queue from wrong CPU 13, hctx active [ 163.047692] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.056374] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.064885] Workqueue: kblockd blk_mq_run_work_fn [ 163.069575] Call trace: [ 163.072008] dump_backtrace+0x0/0x150 [ 163.075656] show_stack+0x14/0x20 [ 163.078958] dump_stack+0xb0/0xf8 [ 163.082260] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.086690] blk_mq_run_work_fn+0x1c/0x28 [ 163.090686] process_one_work+0x1e0/0x358 [ 163.094681] worker_thread+0x238/0x488 [ 163.098417] kthread+0x118/0x120 [ 163.101631] ret_from_fork+0x10/0x18 [ 163.105200] run queue from wrong CPU 13, hctx active [ 163.110534] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.111852] CPU2: shutdown [ 163.119218] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.119222] Workqueue: kblockd blk_mq_run_work_fn [ 163.119223] Call trace: [ 163.119224] dump_backtrace+0x0/0x150 [ 163.119226] show_stack+0x14/0x20 [ 163.119228] dump_stack+0xb0/0xf8 [ 163.119230] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.119234] blk_mq_run_work_fn+0x1c/0x28 [ 163.121943] psci: CPU2 killed. [ 163.130439] process_one_work+0x1e0/0x358 [ 163.130441] worker_thread+0x238/0x488 [ 163.130443] kthread+0x118/0x120 [ 163.130447] ret_from_fork+0x10/0x18 [ 163.173789] run queue from wrong CPU 13, hctx active [ 163.178743] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.187425] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.195937] Workqueue: kblockd blk_mq_run_work_fn [ 163.200627] Call trace: [ 163.203061] dump_backtrace+0x0/0x150 [ 163.206709] show_stack+0x14/0x20 [ 163.210011] dump_stack+0xb0/0xf8 [ 163.213312] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.217742] blk_mq_run_work_fn+0x1c/0x28 [ 163.221738] process_one_work+0x1e0/0x358 [ 163.225733] worker_thread+0x238/0x488 [ 163.229469] kthread+0x118/0x120 [ 163.232684] ret_from_fork+0x10/0x18 [ 163.236253] run queue from wrong CPU 13, hctx active [ 163.241597] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.241691] CPU1: shutdown [ 163.250281] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.250285] Workqueue: kblockd blk_mq_run_work_fn [ 163.250287] Call trace: [ 163.250291] dump_backtrace+0x0/0x150 [ 163.252998] psci: CPU1 killed. [ 163.261496] show_stack+0x14/0x20 [ 163.261499] dump_stack+0xb0/0xf8 [ 163.261503] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.279008] process 870 (fio) no longer affine to cpu0 [ 163.291463] blk_mq_run_work_fn+0x1c/0x28 [ 163.295458] process_one_work+0x1e0/0x358 [ 163.299454] worker_thread+0x238/0x488 [ 163.303189] kthread+0x118/0x120 [ 163.306404] ret_from_fork+0x10/0x18 [ 163.309975] run queue from wrong CPU 13, hctx active [ 163.314929] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.323611] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.332122] Workqueue: kblockd blk_mq_run_work_fn [ 163.336812] Call trace: [ 163.339245] dump_backtrace+0x0/0x150 [ 163.342894] show_stack+0x14/0x20 [ 163.346195] dump_stack+0xb0/0xf8 [ 163.349497] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.353927] blk_mq_run_work_fn+0x1c/0x28 [ 163.357923] process_one_work+0x1e0/0x358 [ 163.361918] worker_thread+0x238/0x488 [ 163.365654] kthread+0x118/0x120 [ 163.368868] ret_from_fork+0x10/0x18 [ 163.372437] run queue from wrong CPU 13, hctx active [ 163.377391] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.386073] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.394583] Workqueue: kblockd blk_mq_run_work_fn [ 163.399273] Call trace: [ 163.401706] dump_backtrace+0x0/0x150 [ 163.405354] show_stack+0x14/0x20 [ 163.408656] dump_stack+0xb0/0xf8 [ 163.411958] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.416388] blk_mq_run_work_fn+0x1c/0x28 [ 163.420384] process_one_work+0x1e0/0x358 [ 163.424379] worker_thread+0x238/0x488 [ 163.428115] kthread+0x118/0x120 [ 163.431329] ret_from_fork+0x10/0x18 [ 163.434934] run queue from wrong CPU 13, hctx active [ 163.439887] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.448570] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.457080] Workqueue: kblockd blk_mq_run_work_fn [ 163.461770] Call trace: [ 163.464203] dump_backtrace+0x0/0x150 [ 163.467851] show_stack+0x14/0x20 [ 163.471153] dump_stack+0xb0/0xf8 [ 163.474455] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.478885] blk_mq_run_work_fn+0x1c/0x28 [ 163.482881] process_one_work+0x1e0/0x358 [ 163.486877] worker_thread+0x238/0x488 [ 163.490613] kthread+0x118/0x120 [ 163.493828] ret_from_fork+0x10/0x18 [ 163.497424] run queue from wrong CPU 13, hctx active [ 163.502378] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.511061] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.519572] Workqueue: kblockd blk_mq_run_work_fn [ 163.524262] Call trace: [ 163.526696] dump_backtrace+0x0/0x150 [ 163.530344] show_stack+0x14/0x20 [ 163.533646] dump_stack+0xb0/0xf8 [ 163.536948] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.541378] blk_mq_run_work_fn+0x1c/0x28 [ 163.545375] process_one_work+0x1e0/0x358 [ 163.549370] worker_thread+0x238/0x488 [ 163.553107] kthread+0x118/0x120 [ 163.556321] ret_from_fork+0x10/0x18 [ 163.559908] run queue from wrong CPU 24, hctx active [ 163.564871] CPU: 24 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.573554] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.582072] Workqueue: kblockd blk_mq_run_work_fn [ 163.586764] Call trace: [ 163.589199] dump_backtrace+0x0/0x150 [ 163.592848] show_stack+0x14/0x20 [ 163.596153] dump_stack+0xb0/0xf8 [ 163.599455] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.603885] blk_mq_run_work_fn+0x1c/0x28 [ 163.607882] process_one_work+0x1e0/0x358 [ 163.611877] worker_thread+0x238/0x488 [ 163.615613] kthread+0x118/0x120 [ 163.618828] ret_from_fork+0x10/0x18 [ 163.622404] run queue from wrong CPU 24, hctx active [ 163.627358] CPU: 24 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.636041] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.644552] Workqueue: kblockd blk_mq_run_work_fn [ 163.649242] Call trace: [ 163.651674] dump_backtrace+0x0/0x150 [ 163.655322] show_stack+0x14/0x20 [ 163.658623] dump_stack+0xb0/0xf8 [ 163.661924] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.666354] blk_mq_run_work_fn+0x1c/0x28 [ 163.670349] process_one_work+0x1e0/0x358 [ 163.674345] worker_thread+0x238/0x488 [ 163.678081] kthread+0x118/0x120 [ 163.681295] ret_from_fork+0x10/0x18 [ 163.684864] run queue from wrong CPU 24, hctx active [ 163.689819] CPU: 24 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.698501] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.707011] Workqueue: kblockd blk_mq_run_work_fn [ 163.711701] Call trace: [ 163.714133] dump_backtrace+0x0/0x150 [ 163.717781] show_stack+0x14/0x20 [ 163.721082] dump_stack+0xb0/0xf8 [ 163.724383] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.728813] blk_mq_run_work_fn+0x1c/0x28 [ 163.732808] process_one_work+0x1e0/0x358 [ 163.736803] worker_thread+0x238/0x488 [ 163.740539] kthread+0x118/0x120 [ 163.743753] ret_from_fork+0x10/0x18 [ 163.747342] run queue from wrong CPU 48, hctx active [ 163.752311] CPU: 48 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.760995] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.769516] Workqueue: kblockd blk_mq_run_work_fn [ 163.774208] Call trace: [ 163.776644] dump_backtrace+0x0/0x150 [ 163.780294] show_stack+0x14/0x20 [ 163.783600] dump_stack+0xb0/0xf8 [ 163.786902] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.791332] blk_mq_run_work_fn+0x1c/0x28 [ 163.795330] process_one_work+0x1e0/0x358 [ 163.799327] worker_thread+0x238/0x488 [ 163.803064] kthread+0x118/0x120 [ 163.806279] ret_from_fork+0x10/0x18 [ 163.809855] run queue from wrong CPU 48, hctx active [ 163.814811] CPU: 48 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.823496] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.832008] Workqueue: kblockd blk_mq_run_work_fn [ 163.836698] Call trace: [ 163.839132] dump_backtrace+0x0/0x150 [ 163.842782] show_stack+0x14/0x20 [ 163.846084] dump_stack+0xb0/0xf8 [ 163.849386] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.853817] blk_mq_run_work_fn+0x1c/0x28 [ 163.857813] process_one_work+0x1e0/0x358 [ 163.861810] worker_thread+0x238/0x488 [ 163.865546] kthread+0x118/0x120 [ 163.868762] ret_from_fork+0x10/0x18 [ 163.872454] run queue from wrong CPU 48, hctx active [ 163.877411] CPU: 48 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.886095] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.894606] Workqueue: kblockd blk_mq_run_work_fn [ 163.899297] Call trace: [ 163.901731] dump_backtrace+0x0/0x150 [ 163.905379] show_stack+0x14/0x20 [ 163.908681] dump_stack+0xb0/0xf8 [ 163.911983] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.916414] blk_mq_run_work_fn+0x1c/0x28 [ 163.920411] process_one_work+0x1e0/0x358 [ 163.924407] worker_thread+0x238/0x488 [ 163.928143] kthread+0x118/0x120 [ 163.931359] ret_from_fork+0x10/0x18 [ 163.934932] run queue from wrong CPU 48, hctx active [ 163.939888] CPU: 48 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 163.948571] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 163.957082] Workqueue: kblockd blk_mq_run_work_fn [ 163.961773] Call trace: [ 163.964207] dump_backtrace+0x0/0x150 [ 163.967855] show_stack+0x14/0x20 [ 163.971157] dump_stack+0xb0/0xf8 [ 163.974459] __blk_mq_run_hw_queue+0x11c/0x128 [ 163.978890] blk_mq_run_work_fn+0x1c/0x28 [ 163.982886] process_one_work+0x1e0/0x358 [ 163.986882] worker_thread+0x238/0x488 [ 163.990618] kthread+0x118/0x120 [ 163.993833] ret_from_fork+0x10/0x18 [ 163.997406] run queue from wrong CPU 48, hctx active [ 164.002361] CPU: 48 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 164.011045] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 164.019556] Workqueue: kblockd blk_mq_run_work_fn [ 164.024247] Call trace: [ 164.026681] dump_backtrace+0x0/0x150 [ 164.030329] show_stack+0x14/0x20 [ 164.033631] dump_stack+0xb0/0xf8 [ 164.036934] __blk_mq_run_hw_queue+0x11c/0x128 [ 164.041364] blk_mq_run_work_fn+0x1c/0x28 [ 164.045360] process_one_work+0x1e0/0x358 [ 164.049357] worker_thread+0x238/0x488 [ 164.053093] kthread+0x118/0x120 [ 164.056308] ret_from_fork+0x10/0x18 [ 164.059885] run queue from wrong CPU 48, hctx active [ 164.064841] CPU: 48 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 164.073525] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 164.082036] Workqueue: kblockd blk_mq_run_work_fn [ 164.086726] Call trace: [ 164.089160] dump_backtrace+0x0/0x150 [ 164.092809] show_stack+0x14/0x20 [ 164.096111] dump_stack+0xb0/0xf8 [ 164.099413] __blk_mq_run_hw_queue+0x11c/0x128 [ 164.103844] blk_mq_run_work_fn+0x1c/0x28 [ 164.107840] process_one_work+0x1e0/0x358 [ 164.111836] worker_thread+0x238/0x488 [ 164.115573] kthread+0x118/0x120 [ 164.118788] ret_from_fork+0x10/0x18 [ 164.122361] run queue from wrong CPU 48, hctx active [ 164.127317] CPU: 48 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 164.136000] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 164.144512] Workqueue: kblockd blk_mq_run_work_fn [ 164.149203] Call trace: [ 164.151636] dump_backtrace+0x0/0x150 [ 164.155285] show_stack+0x14/0x20 [ 164.158587] dump_stack+0xb0/0xf8 [ 164.161889] __blk_mq_run_hw_queue+0x11c/0x128 [ 164.166319] blk_mq_run_work_fn+0x1c/0x28 [ 164.170315] process_one_work+0x1e0/0x358 [ 164.174312] worker_thread+0x238/0x488 [ 164.178048] kthread+0x118/0x120 [ 164.181263] ret_from_fork+0x10/0x18 [ 164.184839] run queue from wrong CPU 48, hctx active [ 164.189794] CPU: 48 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 164.198478] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 164.206989] Workqueue: kblockd blk_mq_run_work_fn [ 164.211680] Call trace: [ 164.214114] dump_backtrace+0x0/0x150 [ 164.217762] show_stack+0x14/0x20 [ 164.221064] dump_stack+0xb0/0xf8 [ 164.224367] __blk_mq_run_hw_queue+0x11c/0x128 [ 164.228797] blk_mq_run_work_fn+0x1c/0x28 [ 164.232793] process_one_work+0x1e0/0x358 [ 164.236789] worker_thread+0x238/0x488 [ 164.240525] kthread+0x118/0x120 [ 164.243740] ret_from_fork+0x10/0x18 [ 164.247348] run queue from wrong CPU 11, hctx active [ 164.252324] CPU: 11 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 164.261008] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 164.269524] Workqueue: kblockd blk_mq_run_work_fn [ 164.274215] Call trace: [ 164.276649] dump_backtrace+0x0/0x150 [ 164.280299] show_stack+0x14/0x20 [ 164.283603] dump_stack+0xb0/0xf8 [ 164.286904] __blk_mq_run_hw_queue+0x11c/0x128 [ 164.291335] blk_mq_run_work_fn+0x1c/0x28 [ 164.295332] process_one_work+0x1e0/0x358 [ 164.299328] worker_thread+0x238/0x488 [ 164.303065] kthread+0x118/0x120 [ 164.306279] ret_from_fork+0x10/0x18 [ 164.857365] irq_shutdown [ 164.859957] irq_shutdown [ 164.862676] IRQ 799: no longer affine to CPU0 [ 164.867332] CPU0: shutdown [ 164.870048] psci: CPU0 killed.
root@(none)$
[I manually added the irq_shutdown print]
From looking at 7df938fbc4ee641, appearantly it's harmless...
I'll continue to test.
John
Quoted 14 lines
Quoted 9 lines
V3:
- re-organize patch 2 & 3 a bit for addressing Hannes's comment
- fix patch 4 for avoiding potential deadlock, as found by HannesV2:
- patch4 & patch 5 in V1 have been merged to block tree, so remove them
- addres.
Quoted 72 lines · John Garry
Quoted 33 lines · John Garry
Quoted 32 lines · Ming Lei
Hi,
Thomas mentioned:
"
That was the constraint of managed interrupts from the very
beginning:The driver/subsystem has to quiesce the interrupt line and the
associated
queue _before_ it gets shutdown in CPU unplug and not fiddle
with it
until it's restarted by the core when the CPU is plugged in again.
"But no drivers or blk-mq do that before one hctx becomes dead(all CPUs for one hctx are offline), and even it is worse, blk-mq stills tries to run hw queue after hctx is dead, see
blk_mq_hctx_notify_dead().This patchset tries to address the issue by two stages:
1) add one new cpuhp state of
CPUHP_AP_BLK_MQ_ONLINE- mark the hctx as internal stopped, and drain all in-flight requests if the hctx is going to be dead.
2) re-submit IO in the state of
CPUHP_BLK_MQ_DEADafter the hctx becomes dead- steal bios from the request, and resubmit them via
generic_make_request(),
then these IO will be mapped to other live hctx for dispatchQuoted 7 lines
Quoted 5 lines
Please comment & review, thanks!
John, I don't add your tested-by tag since V3 have some changes, and I appreciate if you may run your test on V3.
Will do, Thanks
Hi Ming,
I got this warning once:
[ 162.558185] CPU10: shutdown [ 162.560994] psci: CPU10 killed. [ 162.593939] CPU9: shutdown [ 162.596645] psci: CPU9 killed. [ 162.625838] CPU8: shutdown [ 162.628550] psci: CPU8 killed. [ 162.685790] CPU7: shutdown [ 162.688496] psci: CPU7 killed. [ 162.725771] CPU6: shutdown [ 162.728486] psci: CPU6 killed. [ 162.753884] CPU5: shutdown [ 162.756591] psci: CPU5 killed. [ 162.785584] irq_shutdown [ 162.788277] IRQ 800: no longer affine to CPU4 [ 162.793267] CPU4: shutdown [ 162.795975] psci: CPU4 killed. [ 162.849680] run queue from wrong CPU 13, hctx active [ 162.849692] CPU3: shutdown [ 162.854649] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 162.854653] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 162.857362] psci: CPU3 killed. [ 162.866039] Workqueue: kblockd blk_mq_run_work_fn [ 162.882281] Call trace: [ 162.884716] dump_backtrace+0x0/0x150
Quoted 8 lines
[ 162.888365] show_stack+0x14/0x20 [ 162.891668] dump_stack+0xb0/0xf8 [ 162.894970] __blk_mq_run_hw_queue+0x11c/0x128 [ 162.899400] blk_mq_run_work_fn+0x1c/0x28 [ 162.903397] process_one_work+0x1e0/0x358 [ 162.907393] worker_thread+0x40/0x488 [ 162.911042] kthread+0x118/0x120 [ 162.914257] ret_from_fork+0x10/0x18
What is the HBA? If it is Hisilicon SAS, it isn't strange, because this patch can't fix single hw queue with multiple private reply queue yet, that can be one follow-up job of this patchset.
Thanks,
Ming
Quoted 51 lines
Quoted 42 lines
Quoted 11 lines
Quoted 9 lines
- steal bios from the request, and resubmit them via
generic_make_request(),
then these IO will be mapped to other live hctx for dispatchPlease comment & review, thanks!
John, I don't add your tested-by tag since V3 have some changes, and I appreciate if you may run your test on V3.
Will do, Thanks
Hi Ming,
I got this warning once:
[ 162.558185] CPU10: shutdown [ 162.560994] psci: CPU10 killed. [ 162.593939] CPU9: shutdown [ 162.596645] psci: CPU9 killed. [ 162.625838] CPU8: shutdown [ 162.628550] psci: CPU8 killed. [ 162.685790] CPU7: shutdown [ 162.688496] psci: CPU7 killed. [ 162.725771] CPU6: shutdown [ 162.728486] psci: CPU6 killed. [ 162.753884] CPU5: shutdown [ 162.756591] psci: CPU5 killed. [ 162.785584] irq_shutdown [ 162.788277] IRQ 800: no longer affine to CPU4 [ 162.793267] CPU4: shutdown [ 162.795975] psci: CPU4 killed. [ 162.849680] run queue from wrong CPU 13, hctx active [ 162.849692] CPU3: shutdown [ 162.854649] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 162.854653] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 162.857362] psci: CPU3 killed. [ 162.866039] Workqueue: kblockd blk_mq_run_work_fn [ 162.882281] Call trace: [ 162.884716] dump_backtrace+0x0/0x150Quoted 8 lines
[ 162.888365] show_stack+0x14/0x20 [ 162.891668] dump_stack+0xb0/0xf8 [ 162.894970] __blk_mq_run_hw_queue+0x11c/0x128 [ 162.899400] blk_mq_run_work_fn+0x1c/0x28 [ 162.903397] process_one_work+0x1e0/0x358 [ 162.907393] worker_thread+0x40/0x488 [ 162.911042] kthread+0x118/0x120 [ 162.914257] ret_from_fork+0x10/0x18
Quoted 4 lines
What is the HBA? If it is Hisilicon SAS, it isn't strange, because this patch can't fix single hw queue with multiple private reply queue yet, that can be one follow-up job of this patchset.
Yes, hisi_sas. So, right, it is single queue today on mainline, but I manually made it multiqueue on my dev branch just to test this series. Otherwise I could not test it for that driver.
My dev branch is here, if interested:
https://github.com/hisilicon/kernel-dev/commits/private-topic-sas-5.4-mq
I am also going to retest NVMe.
Thanks,
John
Quoted 5 lines
Thanks,
Ming.
Quoted 56 lines · John Garry
Quoted 51 lines
Quoted 42 lines
Quoted 11 lines
Quoted 9 lines
- steal bios from the request, and resubmit them via
generic_make_request(),
then these IO will be mapped to other live hctx for dispatchPlease comment & review, thanks!
John, I don't add your tested-by tag since V3 have some changes, and I appreciate if you may run your test on V3.
Will do, Thanks
Hi Ming,
I got this warning once:
[ 162.558185] CPU10: shutdown [ 162.560994] psci: CPU10 killed. [ 162.593939] CPU9: shutdown [ 162.596645] psci: CPU9 killed. [ 162.625838] CPU8: shutdown [ 162.628550] psci: CPU8 killed. [ 162.685790] CPU7: shutdown [ 162.688496] psci: CPU7 killed. [ 162.725771] CPU6: shutdown [ 162.728486] psci: CPU6 killed. [ 162.753884] CPU5: shutdown [ 162.756591] psci: CPU5 killed. [ 162.785584] irq_shutdown [ 162.788277] IRQ 800: no longer affine to CPU4 [ 162.793267] CPU4: shutdown [ 162.795975] psci: CPU4 killed. [ 162.849680] run queue from wrong CPU 13, hctx active [ 162.849692] CPU3: shutdown [ 162.854649] CPU: 13 PID: 874 Comm: kworker/3:2H Not tainted 5.4.0-rc1-00012-gad025dd3d001 #1098 [ 162.854653] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019 [ 162.857362] psci: CPU3 killed. [ 162.866039] Workqueue: kblockd blk_mq_run_work_fn [ 162.882281] Call trace: [ 162.884716] dump_backtrace+0x0/0x150Quoted 8 lines
[ 162.888365] show_stack+0x14/0x20 [ 162.891668] dump_stack+0xb0/0xf8 [ 162.894970] __blk_mq_run_hw_queue+0x11c/0x128 [ 162.899400] blk_mq_run_work_fn+0x1c/0x28 [ 162.903397] process_one_work+0x1e0/0x358 [ 162.907393] worker_thread+0x40/0x488 [ 162.911042] kthread+0x118/0x120 [ 162.914257] ret_from_fork+0x10/0x18Quoted 4 lines
What is the HBA? If it is Hisilicon SAS, it isn't strange, because this patch can't fix single hw queue with multiple private reply queue yet, that can be one follow-up job of this patchset.
Quoted 6 lines
Yes,
hisi_sas. So, right, it is single queue today on mainline, but I manually made it multiqueue on my dev branch just to test this series. Otherwise I could not test it for that driver.My dev branch is here, if interested:
https://github.com/hisilicon/kernel-dev/commits/private-topic-sas-5.4-mq
Your conversion shouldn't work given you do not change .can_queue in the patch of 'hisi_sas_v3: multiqueue support'.
As discussed before, tags of hisilicon V3 is HBA wide. If you switch to real hw queue, each hw queue has to own its independent tags. However, that isn't supported by V3 hardware.
See previous discussion:
https://marc.info/?t=155928863000001&r=1&w=2
Thanks,
Ming
Quoted 8 lines · Ming Lei
Quoted 6 lines
Yes,
hisi_sas. So, right, it is single queue today on mainline, but IQuoted 5 lines
manually made it multiqueue on my dev branch just to test this series. Otherwise I could not test it for that driver.
My dev branch is here, if interested:
https://github.com/hisilicon/kernel-dev/commits/private-topic-sas-5.4-mqYour conversion shouldn't work given you do not change .can_queue in the patch of '
hisi_sas_v3: multiqueue support'.
Ah, I missed that, but I don't think that it will make a difference really since I'm only using a single disk, so I don't think that can_queue really comes into play. But....
Quoted 4 lines
As discussed before, tags of hisilicon V3 is HBA wide. If you switch to real hw queue, each hw queue has to own its independent tags. However, that isn't supported by V3 hardware.
I am generating the tag internally in the driver now, so that hostwide tags issue should not be an issue.
And, to be clear, I am not paying too much attention to performance, but rather just hotplugging while running IO.
An update on testing:
I did some scripted overnight testing. The script essentially loops like this:
- online all CPUS
- run fio binded on a limited bunch of CPUs to cover a hctx mask for 1 minute
- offline those CPUs
- wait 1 minute (> SCSI or NVMe timeout)
- and repeat
SCSI is actually quite stable, but NVMe isn't. For NVMe I am finding some fio processes never dying with IOPS @ 0. I don't see any NVMe timeout reported. Did you do any NVMe testing of this sort?
Thanks,
John
Quoted 11 lines
See previous discussion:
https://marc.info/?t=155928863000001&r=1&w=2
Thanks,
Ming.
Quoted 22 lines · John Garry
Quoted 5 lines
As discussed before, tags of hisilicon V3 is HBA wide. If you switch to real hw queue, each hw queue has to own its independent tags. However, that isn't supported by V3 hardware.
I am generating the tag internally in the driver now, so that hostwide tags issue should not be an issue.
And, to be clear, I am not paying too much attention to performance, but rather just hotplugging while running IO.
An update on testing:
I did some scripted overnight testing. The script essentially loops like this:
- online all CPUS
- run fio binded on a limited bunch of CPUs to cover a hctx mask for 1 minute
- offline those CPUs
- wait 1 minute (> SCSI or NVMe timeout)
- and repeat
Quoted 4 lines
SCSI is actually quite stable, but NVMe isn't. For NVMe I am finding some fio processes never dying with IOPS @ 0. I don't see any NVMe timeout reported. Did you do any NVMe testing of this sort?
Yeah, so for NVMe, I see some sort of regression, like this:
Jobs: 1 (f=1): [_R] [0.0% done] [0KB/0KB/0KB /s] [0/0/0 iops] [eta 1158037877d:17h:18m:22s]
I have tested against vanilla 5.4 rc1 without problem.
If you can advise some debug to add, then I'd appreciate it. If not, I'll try to add some debug to the new paths introduced in this series to see if anything I hit coincides with the error state, so will at least have a hint...
Thanks,
John
Quoted 9 lines
Quoted 9 lines
See previous discussion:
https://marc.info/?t=155928863000001&r=1&w=2
Thanks,
Ming
Quoted 24 lines · John Garry
Quoted 22 lines · John Garry
Quoted 5 lines
As discussed before, tags of hisilicon V3 is HBA wide. If you switch to real hw queue, each hw queue has to own its independent tags. However, that isn't supported by V3 hardware.
I am generating the tag internally in the driver now, so that hostwide tags issue should not be an issue.
And, to be clear, I am not paying too much attention to performance, but rather just hotplugging while running IO.
An update on testing:
I did some scripted overnight testing. The script essentially loops like this:
- online all CPUS
- run fio binded on a limited bunch of CPUs to cover a hctx mask for 1 minute
- offline those CPUs
- wait 1 minute (> SCSI or NVMe timeout)
- and repeat
Quoted 8 lines
Quoted 4 lines
SCSI is actually quite stable, but NVMe isn't. For NVMe I am finding some fio processes never dying with IOPS @ 0. I don't see any NVMe timeout reported. Did you do any NVMe testing of this sort?
Yeah, so for NVMe, I see some sort of regression, like this:
Jobs: 1 (f=1): [_R] [0.0% done] [0KB/0KB/0KB /s] [0/0/0 iops] [eta 1158037877d:17h:18m:22s]
I can reproduce this issue, and looks there are requests in ->dispatch. I am a bit busy this week, please feel free to investigate it and debugfs can help you much. I may have time next week for looking this issue.
Thanks,
Ming Lei
Quoted 30 lines · Ming Lei
Quoted 29 lines · John Garry
Quoted 26 lines · John Garry
Quoted 5 lines
As discussed before, tags of hisilicon V3 is HBA wide. If you switch to real hw queue, each hw queue has to own its independent tags. However, that isn't supported by V3 hardware.
I am generating the tag internally in the driver now, so that hostwide tags issue should not be an issue.
And, to be clear, I am not paying too much attention to performance, but rather just hotplugging while running IO.
An update on testing:
I did some scripted overnight testing. The script essentially loops like this:
- online all CPUS
- run fio binded on a limited bunch of CPUs to cover a hctx mask for 1 minute
- offline those CPUs
- wait 1 minute (> SCSI or NVMe timeout)
- and repeatSCSI is actually quite stable, but NVMe isn't. For NVMe I am finding some fio processes never dying with IOPS @ 0. I don't see any NVMe timeout reported. Did you do any NVMe testing of this sort?
Quoted 5 lines
Quoted 3 lines
Yeah, so for NVMe, I see some sort of regression, like this:
Jobs: 1 (f=1): [_R] [0.0% done] [0KB/0KB/0KB /s] [0/0/0 iops] [eta 1158037877d:17h:18m:22s]I can reproduce this issue, and looks there are requests in ->dispatch.
OK, that may match with what I see:
- the problem occuring coincides with this callpath with BLK_MQ_S_INTERNAL_STOPPED set:
blk_mq_request_bypass_insert
(__)blk_mq_try_issue_list_directlyblk_mq_sched_insert_requestsblk_mq_flush_plug_listblk_flush_plug_listblk_finish_plugblkdev_direct_IOgeneric_file_read_iterblkdev_read_iteraio_readio_submit_one
blk_mq_request_bypass_insert() adds to the dispatch list, and looking at debugfs, could this be that dispatched request sitting:
root@(none)$ more /sys/kernel/debug/block/nvme0n1/hctx18/dispatch
00000000ac28511d {.op=READ, .cmd_flags=, .rq_flags=IO_STAT, .state=idle, .tag=56, .internal_tag=-1}So could there be some race here?
Quoted 3 lines
I am a bit busy this week, please feel free to investigate it and debugfs can help you much. I may have time next week for looking this issue.
OK, appreciated
John
Quoted 4 lines
Thanks,
Ming Lei
Quoted 36 lines · John Garry
Quoted 34 lines · Ming Lei
Quoted 32 lines · John Garry
Quoted 22 lines · John Garry
Quoted 5 lines
As discussed before, tags of hisilicon V3 is HBA wide. If you switch to real hw queue, each hw queue has to own its independent tags. However, that isn't supported by V3 hardware.
I am generating the tag internally in the driver now, so that hostwide tags issue should not be an issue.
And, to be clear, I am not paying too much attention to performance, but rather just hotplugging while running IO.
An update on testing:
I did some scripted overnight testing. The script essentially loops like this:
- online all CPUS
- run fio binded on a limited bunch of CPUs to cover a hctx mask for 1 minute
- offline those CPUs
- wait 1 minute (> SCSI or NVMe timeout)
- and repeatQuoted 4 lines
SCSI is actually quite stable, but NVMe isn't. For NVMe I am finding some fio processes never dying with IOPS @ 0. I don't see any NVMe timeout reported. Did you do any NVMe testing of this sort?
Yeah, so for NVMe, I see some sort of regression, like this:
Jobs: 1 (f=1): [_R] [0.0% done] [0KB/0KB/0KB /s] [0/0/0 iops] [eta 1158037877d:17h:18m:22s]
Quoted 5 lines
Quoted 1 line
I can reproduce this issue, and looks there are requests in ->dispatch.
OK, that may match with what I see:
- the problem occuring coincides with this callpath withBLK_MQ_S_INTERNAL_STOPPEDset:
Good catch, these requests should have been re-submitted in blk_mq_hctx_notify_dead() too.
Will do it in V4.
Thanks,
Ming Lei
Quoted 39 lines · Ming Lei
Quoted 38 lines · John Garry
Quoted 35 lines · Ming Lei
Quoted 24 lines · John Garry
Quoted 22 lines · John Garry
Quoted 5 lines
As discussed before, tags of hisilicon V3 is HBA wide. If you switch to real hw queue, each hw queue has to own its independent tags. However, that isn't supported by V3 hardware.
I am generating the tag internally in the driver now, so that hostwide tags issue should not be an issue.
And, to be clear, I am not paying too much attention to performance, but rather just hotplugging while running IO.
An update on testing:
I did some scripted overnight testing. The script essentially loops like this:
- online all CPUS
- run fio binded on a limited bunch of CPUs to cover a hctx mask for 1 minute
- offline those CPUs
- wait 1 minute (> SCSI or NVMe timeout)
- and repeatQuoted 8 lines
Quoted 4 lines
SCSI is actually quite stable, but NVMe isn't. For NVMe I am finding some fio processes never dying with IOPS @ 0. I don't see any NVMe timeout reported. Did you do any NVMe testing of this sort?
Yeah, so for NVMe, I see some sort of regression, like this:
Jobs: 1 (f=1): [_R] [0.0% done] [0KB/0KB/0KB /s] [0/0/0 iops] [eta 1158037877d:17h:18m:22s]I can reproduce this issue, and looks there are requests in ->dispatch.
Quoted 8 lines
Quoted 3 lines
OK, that may match with what I see:
- the problem occuring coincides with this callpath withBLK_MQ_S_INTERNAL_STOPPEDset:Good catch, these requests should have been re-submitted in
blk_mq_hctx_notify_dead()too.Will do it in V4.
OK, I'll have a look at v4 and retest - it may take a while as testing this is slow...
All the best,
John
Quoted 6 lines
Thanks,
Ming Lei.