[patch 0/3] block/mq: Convert to the new hotplug state machine

[patch 0/3] block/mq: Convert to the new hotplug state machine

[Thomas Gleixner]

The following series converts block/mq to the new hotplug state
machine. Patch 1/3 reserves the states for the block layer and is already applied to

   git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git smp/for-block

to avoid merge conflicts. This branch can be pulled into the block layer
instead of applying patch 1/3 manually,

Thanks,

	tglx
---
 include/linux/blk-mq.h     |    2 
 block/blk-mq-cpu.c         |   15 ++----
 block/blk-mq.c             |  108 ++++++++++++++++++++-------------------------
 block/blk-mq.h             |    2 
 include/linux/cpuhotplug.h |    2 
 5 files changed, 59 insertions(+), 70 deletions(-)

[patch 1/3] blk/mq: Reserve hotplug states for block multiqueue

[Thomas Gleixner]

This patch only reserves two CPU hotplug states for block/mq so the block tree
can apply the conversion patches.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: rt@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---
 include/linux/cpuhotplug.h |    2 ++
 1 file changed, 2 insertions(+)

--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -14,6 +14,7 @@ enum cpuhp_state {
 	CPUHP_PERF_SUPERH,
 	CPUHP_X86_HPET_DEAD,
 	CPUHP_X86_APB_DEAD,
+	CPUHP_BLK_MQ_DEAD,
 	CPUHP_WORKQUEUE_PREP,
 	CPUHP_POWER_NUMA_PREPARE,
 	CPUHP_HRTIMERS_PREPARE,
@@ -22,6 +23,7 @@ enum cpuhp_state {
 	CPUHP_SMPCFD_PREPARE,
 	CPUHP_RCUTREE_PREP,
 	CPUHP_NOTIFY_PREPARE,
+	CPUHP_BLK_MQ_PREPARE,
 	CPUHP_TIMERS_DEAD,
 	CPUHP_BRINGUP_CPU,
 	CPUHP_AP_IDLE_DEAD,

[patch 1/3] blk/mq: Reserve hotplug states for block multiqueue

[Thomas Gleixner]

[-- Attachment #1: blkmq_Reserve_hotplug_ID_states_for_block.patch --]
[-- Type: text/plain, Size: 867 bytes --]

This patch only reserves two CPU hotplug states for block/mq so the block tree
can apply the conversion patches.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: rt@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---
 include/linux/cpuhotplug.h |    2 ++
 1 file changed, 2 insertions(+)

--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -14,6 +14,7 @@ enum cpuhp_state {
 	CPUHP_PERF_SUPERH,
 	CPUHP_X86_HPET_DEAD,
 	CPUHP_X86_APB_DEAD,
+	CPUHP_BLK_MQ_DEAD,
 	CPUHP_WORKQUEUE_PREP,
 	CPUHP_POWER_NUMA_PREPARE,
 	CPUHP_HRTIMERS_PREPARE,
@@ -22,6 +23,7 @@ enum cpuhp_state {
 	CPUHP_SMPCFD_PREPARE,
 	CPUHP_RCUTREE_PREP,
 	CPUHP_NOTIFY_PREPARE,
+	CPUHP_BLK_MQ_PREPARE,
 	CPUHP_TIMERS_DEAD,
 	CPUHP_BRINGUP_CPU,
 	CPUHP_AP_IDLE_DEAD,

[patch 2/3] blk/mq/cpu-notif: Convert to hotplug state machine

[Thomas Gleixner]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Install the callbacks via the state machine so we can phase out the cpu
hotplug notifiers..

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: rt@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---

 block/blk-mq-cpu.c     |   15 +++++++--------
 block/blk-mq.c         |   21 +++++----------------
 block/blk-mq.h         |    2 +-
 include/linux/blk-mq.h |    2 +-
 4 files changed, 14 insertions(+), 26 deletions(-)

--- a/block/blk-mq-cpu.c
+++ b/block/blk-mq-cpu.c
@@ -18,18 +18,16 @@
 static LIST_HEAD(blk_mq_cpu_notify_list);
 static DEFINE_RAW_SPINLOCK(blk_mq_cpu_notify_lock);
 
-static int blk_mq_main_cpu_notify(struct notifier_block *self,
-				  unsigned long action, void *hcpu)
+static int blk_mq_cpu_dead(unsigned int cpu)
 {
-	unsigned int cpu = (unsigned long) hcpu;
 	struct blk_mq_cpu_notifier *notify;
-	int ret = NOTIFY_OK;
+	int ret;
 
 	raw_spin_lock(&blk_mq_cpu_notify_lock);
 
 	list_for_each_entry(notify, &blk_mq_cpu_notify_list, list) {
-		ret = notify->notify(notify->data, action, cpu);
-		if (ret != NOTIFY_OK)
+		ret = notify->notify(notify->data, cpu);
+		if (ret)
 			break;
 	}
 
@@ -54,7 +52,7 @@ void blk_mq_unregister_cpu_notifier(stru
 }
 
 void blk_mq_init_cpu_notifier(struct blk_mq_cpu_notifier *notifier,
-			      int (*fn)(void *, unsigned long, unsigned int),
+			      int (*fn)(void *, unsigned int),
 			      void *data)
 {
 	notifier->notify = fn;
@@ -63,5 +61,6 @@ void blk_mq_init_cpu_notifier(struct blk
 
 void __init blk_mq_cpu_init(void)
 {
-	hotcpu_notifier(blk_mq_main_cpu_notify, 0);
+	cpuhp_setup_state_nocalls(CPUHP_BLK_MQ_DEAD, "block/mq:dead", NULL,
+				  blk_mq_cpu_dead);
 }
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1590,30 +1590,19 @@ static int blk_mq_hctx_cpu_offline(struc
 	spin_unlock(&ctx->lock);
 
 	if (list_empty(&tmp))
-		return NOTIFY_OK;
+		return 0;
 
 	spin_lock(&hctx->lock);
 	list_splice_tail_init(&tmp, &hctx->dispatch);
 	spin_unlock(&hctx->lock);
 
 	blk_mq_run_hw_queue(hctx, true);
-	return NOTIFY_OK;
+	return 0;
 }
 
-static int blk_mq_hctx_notify(void *data, unsigned long action,
-			      unsigned int cpu)
+static int blk_mq_hctx_notify_dead(void *hctx, unsigned int cpu)
 {
-	struct blk_mq_hw_ctx *hctx = data;
-
-	if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
-		return blk_mq_hctx_cpu_offline(hctx, cpu);
-
-	/*
-	 * In case of CPU online, tags may be reallocated
-	 * in blk_mq_map_swqueue() after mapping is updated.
-	 */
-
-	return NOTIFY_OK;
+	return blk_mq_hctx_cpu_offline(hctx, cpu);
 }
 
 /* hctx->ctxs will be freed in queue's release handler */
@@ -1681,7 +1670,7 @@ static int blk_mq_init_hctx(struct reque
 	hctx->flags = set->flags & ~BLK_MQ_F_TAG_SHARED;
 
 	blk_mq_init_cpu_notifier(&hctx->cpu_notifier,
-					blk_mq_hctx_notify, hctx);
+					blk_mq_hctx_notify_dead, hctx);
 	blk_mq_register_cpu_notifier(&hctx->cpu_notifier);
 
 	hctx->tags = set->tags[hctx_idx];
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -34,7 +34,7 @@ void blk_mq_wake_waiters(struct request_
  */
 struct blk_mq_cpu_notifier;
 void blk_mq_init_cpu_notifier(struct blk_mq_cpu_notifier *notifier,
-			      int (*fn)(void *, unsigned long, unsigned int),
+			      int (*fn)(void *, unsigned int),
 			      void *data);
 void blk_mq_register_cpu_notifier(struct blk_mq_cpu_notifier *notifier);
 void blk_mq_unregister_cpu_notifier(struct blk_mq_cpu_notifier *notifier);
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -10,7 +10,7 @@ struct blk_flush_queue;
 struct blk_mq_cpu_notifier {
 	struct list_head list;
 	void *data;
-	int (*notify)(void *data, unsigned long action, unsigned int cpu);
+	int (*notify)(void *data, unsigned int cpu);
 };
 
 struct blk_mq_hw_ctx {

[patch 2/3] blk/mq/cpu-notif: Convert to hotplug state machine

[Thomas Gleixner]

[-- Attachment #1: blkmqcpu-notif_Convert_to_hotplug_state_machine.patch --]
[-- Type: text/plain, Size: 3857 bytes --]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Install the callbacks via the state machine so we can phase out the cpu
hotplug notifiers..

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: rt@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---

 block/blk-mq-cpu.c     |   15 +++++++--------
 block/blk-mq.c         |   21 +++++----------------
 block/blk-mq.h         |    2 +-
 include/linux/blk-mq.h |    2 +-
 4 files changed, 14 insertions(+), 26 deletions(-)

--- a/block/blk-mq-cpu.c
+++ b/block/blk-mq-cpu.c
@@ -18,18 +18,16 @@
 static LIST_HEAD(blk_mq_cpu_notify_list);
 static DEFINE_RAW_SPINLOCK(blk_mq_cpu_notify_lock);
 
-static int blk_mq_main_cpu_notify(struct notifier_block *self,
-				  unsigned long action, void *hcpu)
+static int blk_mq_cpu_dead(unsigned int cpu)
 {
-	unsigned int cpu = (unsigned long) hcpu;
 	struct blk_mq_cpu_notifier *notify;
-	int ret = NOTIFY_OK;
+	int ret;
 
 	raw_spin_lock(&blk_mq_cpu_notify_lock);
 
 	list_for_each_entry(notify, &blk_mq_cpu_notify_list, list) {
-		ret = notify->notify(notify->data, action, cpu);
-		if (ret != NOTIFY_OK)
+		ret = notify->notify(notify->data, cpu);
+		if (ret)
 			break;
 	}
 
@@ -54,7 +52,7 @@ void blk_mq_unregister_cpu_notifier(stru
 }
 
 void blk_mq_init_cpu_notifier(struct blk_mq_cpu_notifier *notifier,
-			      int (*fn)(void *, unsigned long, unsigned int),
+			      int (*fn)(void *, unsigned int),
 			      void *data)
 {
 	notifier->notify = fn;
@@ -63,5 +61,6 @@ void blk_mq_init_cpu_notifier(struct blk
 
 void __init blk_mq_cpu_init(void)
 {
-	hotcpu_notifier(blk_mq_main_cpu_notify, 0);
+	cpuhp_setup_state_nocalls(CPUHP_BLK_MQ_DEAD, "block/mq:dead", NULL,
+				  blk_mq_cpu_dead);
 }
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1590,30 +1590,19 @@ static int blk_mq_hctx_cpu_offline(struc
 	spin_unlock(&ctx->lock);
 
 	if (list_empty(&tmp))
-		return NOTIFY_OK;
+		return 0;
 
 	spin_lock(&hctx->lock);
 	list_splice_tail_init(&tmp, &hctx->dispatch);
 	spin_unlock(&hctx->lock);
 
 	blk_mq_run_hw_queue(hctx, true);
-	return NOTIFY_OK;
+	return 0;
 }
 
-static int blk_mq_hctx_notify(void *data, unsigned long action,
-			      unsigned int cpu)
+static int blk_mq_hctx_notify_dead(void *hctx, unsigned int cpu)
 {
-	struct blk_mq_hw_ctx *hctx = data;
-
-	if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
-		return blk_mq_hctx_cpu_offline(hctx, cpu);
-
-	/*
-	 * In case of CPU online, tags may be reallocated
-	 * in blk_mq_map_swqueue() after mapping is updated.
-	 */
-
-	return NOTIFY_OK;
+	return blk_mq_hctx_cpu_offline(hctx, cpu);
 }
 
 /* hctx->ctxs will be freed in queue's release handler */
@@ -1681,7 +1670,7 @@ static int blk_mq_init_hctx(struct reque
 	hctx->flags = set->flags & ~BLK_MQ_F_TAG_SHARED;
 
 	blk_mq_init_cpu_notifier(&hctx->cpu_notifier,
-					blk_mq_hctx_notify, hctx);
+					blk_mq_hctx_notify_dead, hctx);
 	blk_mq_register_cpu_notifier(&hctx->cpu_notifier);
 
 	hctx->tags = set->tags[hctx_idx];
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -34,7 +34,7 @@ void blk_mq_wake_waiters(struct request_
  */
 struct blk_mq_cpu_notifier;
 void blk_mq_init_cpu_notifier(struct blk_mq_cpu_notifier *notifier,
-			      int (*fn)(void *, unsigned long, unsigned int),
+			      int (*fn)(void *, unsigned int),
 			      void *data);
 void blk_mq_register_cpu_notifier(struct blk_mq_cpu_notifier *notifier);
 void blk_mq_unregister_cpu_notifier(struct blk_mq_cpu_notifier *notifier);
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -10,7 +10,7 @@ struct blk_flush_queue;
 struct blk_mq_cpu_notifier {
 	struct list_head list;
 	void *data;
-	int (*notify)(void *data, unsigned long action, unsigned int cpu);
+	int (*notify)(void *data, unsigned int cpu);
 };
 
 struct blk_mq_hw_ctx {

[patch 3/3] blk/mq: Convert to hotplug state machine

[Thomas Gleixner]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Install the callbacks via the state machine so we can phase out the cpu
hotplug notifiers mess.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: rt@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---

 block/blk-mq.c |   87 ++++++++++++++++++++++++++++-----------------------------
 1 file changed, 43 insertions(+), 44 deletions(-)

--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2090,50 +2090,18 @@ static void blk_mq_queue_reinit(struct r
 	blk_mq_sysfs_register(q);
 }
 
-static int blk_mq_queue_reinit_notify(struct notifier_block *nb,
-				      unsigned long action, void *hcpu)
+/*
+ * New online cpumask which is going to be set in this hotplug event.
+ * Declare this cpumasks as global as cpu-hotplug operation is invoked
+ * one-by-one and dynamically allocating this could result in a failure.
+ */
+static struct cpumask cpuhp_online_new;
+
+static void blk_mq_queue_reinit_work(void)
 {
 	struct request_queue *q;
-	int cpu = (unsigned long)hcpu;
-	/*
-	 * New online cpumask which is going to be set in this hotplug event.
-	 * Declare this cpumasks as global as cpu-hotplug operation is invoked
-	 * one-by-one and dynamically allocating this could result in a failure.
-	 */
-	static struct cpumask online_new;
-
-	/*
-	 * Before hotadded cpu starts handling requests, new mappings must
-	 * be established.  Otherwise, these requests in hw queue might
-	 * never be dispatched.
-	 *
-	 * For example, there is a single hw queue (hctx) and two CPU queues
-	 * (ctx0 for CPU0, and ctx1 for CPU1).
-	 *
-	 * Now CPU1 is just onlined and a request is inserted into
-	 * ctx1->rq_list and set bit0 in pending bitmap as ctx1->index_hw is
-	 * still zero.
-	 *
-	 * And then while running hw queue, flush_busy_ctxs() finds bit0 is
-	 * set in pending bitmap and tries to retrieve requests in
-	 * hctx->ctxs[0]->rq_list.  But htx->ctxs[0] is a pointer to ctx0,
-	 * so the request in ctx1->rq_list is ignored.
-	 */
-	switch (action & ~CPU_TASKS_FROZEN) {
-	case CPU_DEAD:
-	case CPU_UP_CANCELED:
-		cpumask_copy(&online_new, cpu_online_mask);
-		break;
-	case CPU_UP_PREPARE:
-		cpumask_copy(&online_new, cpu_online_mask);
-		cpumask_set_cpu(cpu, &online_new);
-		break;
-	default:
-		return NOTIFY_OK;
-	}
 
 	mutex_lock(&all_q_mutex);
-
 	/*
 	 * We need to freeze and reinit all existing queues.  Freezing
 	 * involves synchronous wait for an RCU grace period and doing it
@@ -2154,13 +2122,43 @@ static int blk_mq_queue_reinit_notify(st
 	}
 
 	list_for_each_entry(q, &all_q_list, all_q_node)
-		blk_mq_queue_reinit(q, &online_new);
+		blk_mq_queue_reinit(q, &cpuhp_online_new);
 
 	list_for_each_entry(q, &all_q_list, all_q_node)
 		blk_mq_unfreeze_queue(q);
 
 	mutex_unlock(&all_q_mutex);
-	return NOTIFY_OK;
+}
+
+static int blk_mq_queue_reinit_dead(unsigned int cpu)
+{
+	cpumask_clear_cpu(cpu, &cpuhp_online_new);
+	blk_mq_queue_reinit_work();
+	return 0;
+}
+
+/*
+ * Before hotadded cpu starts handling requests, new mappings must be
+ * established.  Otherwise, these requests in hw queue might never be
+ * dispatched.
+ *
+ * For example, there is a single hw queue (hctx) and two CPU queues (ctx0
+ * for CPU0, and ctx1 for CPU1).
+ *
+ * Now CPU1 is just onlined and a request is inserted into ctx1->rq_list
+ * and set bit0 in pending bitmap as ctx1->index_hw is still zero.
+ *
+ * And then while running hw queue, flush_busy_ctxs() finds bit0 is set in
+ * pending bitmap and tries to retrieve requests in hctx->ctxs[0]->rq_list.
+ * But htx->ctxs[0] is a pointer to ctx0, so the request in ctx1->rq_list
+ * is ignored.
+ */
+static int blk_mq_queue_reinit_prepare(unsigned int cpu)
+{
+	cpumask_copy(&cpuhp_online_new, cpu_online_mask);
+	cpumask_set_cpu(cpu, &cpuhp_online_new);
+	blk_mq_queue_reinit_work();
+	return 0;
 }
 
 static int __blk_mq_alloc_rq_maps(struct blk_mq_tag_set *set)
@@ -2381,8 +2379,9 @@ static int __init blk_mq_init(void)
 {
 	blk_mq_cpu_init();
 
-	hotcpu_notifier(blk_mq_queue_reinit_notify, 0);
-
+	cpuhp_setup_state_nocalls(CPUHP_BLK_MQ_PREPARE, "block/mq:prepare",
+				  blk_mq_queue_reinit_prepare,
+				  blk_mq_queue_reinit_dead);
 	return 0;
 }
 subsys_initcall(blk_mq_init);

[patch 3/3] blk/mq: Convert to hotplug state machine

[Thomas Gleixner]

[-- Attachment #1: blkmq_Convert_to_hotplug_state_machine.patch --]
[-- Type: text/plain, Size: 4319 bytes --]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Install the callbacks via the state machine so we can phase out the cpu
hotplug notifiers mess.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: rt@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---

 block/blk-mq.c |   87 ++++++++++++++++++++++++++++-----------------------------
 1 file changed, 43 insertions(+), 44 deletions(-)

--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2090,50 +2090,18 @@ static void blk_mq_queue_reinit(struct r
 	blk_mq_sysfs_register(q);
 }
 
-static int blk_mq_queue_reinit_notify(struct notifier_block *nb,
-				      unsigned long action, void *hcpu)
+/*
+ * New online cpumask which is going to be set in this hotplug event.
+ * Declare this cpumasks as global as cpu-hotplug operation is invoked
+ * one-by-one and dynamically allocating this could result in a failure.
+ */
+static struct cpumask cpuhp_online_new;
+
+static void blk_mq_queue_reinit_work(void)
 {
 	struct request_queue *q;
-	int cpu = (unsigned long)hcpu;
-	/*
-	 * New online cpumask which is going to be set in this hotplug event.
-	 * Declare this cpumasks as global as cpu-hotplug operation is invoked
-	 * one-by-one and dynamically allocating this could result in a failure.
-	 */
-	static struct cpumask online_new;
-
-	/*
-	 * Before hotadded cpu starts handling requests, new mappings must
-	 * be established.  Otherwise, these requests in hw queue might
-	 * never be dispatched.
-	 *
-	 * For example, there is a single hw queue (hctx) and two CPU queues
-	 * (ctx0 for CPU0, and ctx1 for CPU1).
-	 *
-	 * Now CPU1 is just onlined and a request is inserted into
-	 * ctx1->rq_list and set bit0 in pending bitmap as ctx1->index_hw is
-	 * still zero.
-	 *
-	 * And then while running hw queue, flush_busy_ctxs() finds bit0 is
-	 * set in pending bitmap and tries to retrieve requests in
-	 * hctx->ctxs[0]->rq_list.  But htx->ctxs[0] is a pointer to ctx0,
-	 * so the request in ctx1->rq_list is ignored.
-	 */
-	switch (action & ~CPU_TASKS_FROZEN) {
-	case CPU_DEAD:
-	case CPU_UP_CANCELED:
-		cpumask_copy(&online_new, cpu_online_mask);
-		break;
-	case CPU_UP_PREPARE:
-		cpumask_copy(&online_new, cpu_online_mask);
-		cpumask_set_cpu(cpu, &online_new);
-		break;
-	default:
-		return NOTIFY_OK;
-	}
 
 	mutex_lock(&all_q_mutex);
-
 	/*
 	 * We need to freeze and reinit all existing queues.  Freezing
 	 * involves synchronous wait for an RCU grace period and doing it
@@ -2154,13 +2122,43 @@ static int blk_mq_queue_reinit_notify(st
 	}
 
 	list_for_each_entry(q, &all_q_list, all_q_node)
-		blk_mq_queue_reinit(q, &online_new);
+		blk_mq_queue_reinit(q, &cpuhp_online_new);
 
 	list_for_each_entry(q, &all_q_list, all_q_node)
 		blk_mq_unfreeze_queue(q);
 
 	mutex_unlock(&all_q_mutex);
-	return NOTIFY_OK;
+}
+
+static int blk_mq_queue_reinit_dead(unsigned int cpu)
+{
+	cpumask_clear_cpu(cpu, &cpuhp_online_new);
+	blk_mq_queue_reinit_work();
+	return 0;
+}
+
+/*
+ * Before hotadded cpu starts handling requests, new mappings must be
+ * established.  Otherwise, these requests in hw queue might never be
+ * dispatched.
+ *
+ * For example, there is a single hw queue (hctx) and two CPU queues (ctx0
+ * for CPU0, and ctx1 for CPU1).
+ *
+ * Now CPU1 is just onlined and a request is inserted into ctx1->rq_list
+ * and set bit0 in pending bitmap as ctx1->index_hw is still zero.
+ *
+ * And then while running hw queue, flush_busy_ctxs() finds bit0 is set in
+ * pending bitmap and tries to retrieve requests in hctx->ctxs[0]->rq_list.
+ * But htx->ctxs[0] is a pointer to ctx0, so the request in ctx1->rq_list
+ * is ignored.
+ */
+static int blk_mq_queue_reinit_prepare(unsigned int cpu)
+{
+	cpumask_copy(&cpuhp_online_new, cpu_online_mask);
+	cpumask_set_cpu(cpu, &cpuhp_online_new);
+	blk_mq_queue_reinit_work();
+	return 0;
 }
 
 static int __blk_mq_alloc_rq_maps(struct blk_mq_tag_set *set)
@@ -2381,8 +2379,9 @@ static int __init blk_mq_init(void)
 {
 	blk_mq_cpu_init();
 
-	hotcpu_notifier(blk_mq_queue_reinit_notify, 0);
-
+	cpuhp_setup_state_nocalls(CPUHP_BLK_MQ_PREPARE, "block/mq:prepare",
+				  blk_mq_queue_reinit_prepare,
+				  blk_mq_queue_reinit_dead);
 	return 0;
 }
 subsys_initcall(blk_mq_init);

Re: [patch 2/3] blk/mq/cpu-notif: Convert to hotplug state machine

[Christoph Hellwing]

On Mon, Sep 19, 2016 at 09:28:20PM -0000, Thomas Gleixner wrote:
> From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> 
> Install the callbacks via the state machine so we can phase out the cpu
> hotplug notifiers..

Didn't Sebastian come up with a version of the hotplug state callbacks
that can be per-object?  This seems to be a perfect candidate for that.

Re: [patch 2/3] blk/mq/cpu-notif: Convert to hotplug state machine

[Thomas Gleixner]

On Tue, 20 Sep 2016, Christoph Hellwing wrote:

> On Mon, Sep 19, 2016 at 09:28:20PM -0000, Thomas Gleixner wrote:
> > From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> > 
> > Install the callbacks via the state machine so we can phase out the cpu
> > hotplug notifiers..
> 
> Didn't Sebastian come up with a version of the hotplug state callbacks
> that can be per-object?  This seems to be a perfect candidate for that.

Indeed. I wrote that myself and forgot about it already..... :(

So yes, we can use that and get rid of blk-mq-cpu.c completely.

Thanks,

	tglx