Merge branch 'smp-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull smp/hotplug fixes from Thomas Gleixner:
 "This addresses the fallout of the new lockdep mechanism which covers
  completions in the CPU hotplug code.

  The lockdep splats are false positives, but there is no way to
  annotate that reliably. The solution is to split the completions for
  CPU up and down, which requires some reshuffling of the failure
  rollback handling as well"

* 'smp-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  smp/hotplug: Hotplug state fail injection
  smp/hotplug: Differentiate the AP completion between up and down
  smp/hotplug: Differentiate the AP-work lockdep class between up and down
  smp/hotplug: Callback vs state-machine consistency
  smp/hotplug: Rewrite AP state machine core
  smp/hotplug: Allow external multi-instance rollback
  smp/hotplug: Add state diagram

include/linux/cpuhotplug.h

@@ -3,8 +3,27 @@
 
 #include <linux/types.h>
 
+/*
+ * CPU-up			CPU-down
+ *
+ * BP		AP		BP		AP
+ *
+ * OFFLINE			OFFLINE
+ *   |				  ^
+ *   v				  |
+ * BRINGUP_CPU->AP_OFFLINE	BRINGUP_CPU  <- AP_IDLE_DEAD (idle thread/play_dead)
+ *		  |				AP_OFFLINE
+ *		  v (IRQ-off)	  ,---------------^
+ *		AP_ONLNE	  | (stop_machine)
+ *		  |		TEARDOWN_CPU <-	AP_ONLINE_IDLE
+ *		  |				  ^
+ *		  v				  |
+ *              AP_ACTIVE			AP_ACTIVE
+ */
+
 enum cpuhp_state {
-	CPUHP_OFFLINE,
+	CPUHP_INVALID = -1,
+	CPUHP_OFFLINE = 0,
 	CPUHP_CREATE_THREADS,
 	CPUHP_PERF_PREPARE,
 	CPUHP_PERF_X86_PREPARE,

kernel/cpu.c

@@ -46,11 +46,13 @@
  * @bringup:	Single callback bringup or teardown selector
  * @cb_state:	The state for a single callback (install/uninstall)
  * @result:	Result of the operation
- * @done:	Signal completion to the issuer of the task
+ * @done_up:	Signal completion to the issuer of the task for cpu-up
+ * @done_down:	Signal completion to the issuer of the task for cpu-down
  */
 struct cpuhp_cpu_state {
 	enum cpuhp_state	state;
 	enum cpuhp_state	target;
+	enum cpuhp_state	fail;
 #ifdef CONFIG_SMP
 	struct task_struct	*thread;
 	bool			should_run;
@@ -58,18 +60,39 @@ struct cpuhp_cpu_state {
 	bool			single;
 	bool			bringup;
 	struct hlist_node	*node;
+	struct hlist_node	*last;
 	enum cpuhp_state	cb_state;
 	int			result;
-	struct completion	done;
+	struct completion	done_up;
+	struct completion	done_down;
 #endif
 };
 
-static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state);
+static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
+	.fail = CPUHP_INVALID,
+};
 
 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
-static struct lock_class_key cpuhp_state_key;
-static struct lockdep_map cpuhp_state_lock_map =
-	STATIC_LOCKDEP_MAP_INIT("cpuhp_state", &cpuhp_state_key);
+static struct lockdep_map cpuhp_state_up_map =
+	STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
+static struct lockdep_map cpuhp_state_down_map =
+	STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
+
+
+static void inline cpuhp_lock_acquire(bool bringup)
+{
+	lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
+}
+
+static void inline cpuhp_lock_release(bool bringup)
+{
+	lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
+}
+#else
+
+static void inline cpuhp_lock_acquire(bool bringup) { }
+static void inline cpuhp_lock_release(bool bringup) { }
+
 #endif
 
 /**
@@ -123,13 +146,16 @@ static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
 /**
  * cpuhp_invoke_callback _ Invoke the callbacks for a given state
  * @cpu:	The cpu for which the callback should be invoked
- * @step:	The step in the state machine
+ * @state:	The state to do callbacks for
  * @bringup:	True if the bringup callback should be invoked
+ * @node:	For multi-instance, do a single entry callback for install/remove
+ * @lastp:	For multi-instance rollback, remember how far we got
  *
  * Called from cpu hotplug and from the state register machinery.
  */
 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
-				 bool bringup, struct hlist_node *node)
+				 bool bringup, struct hlist_node *node,
+				 struct hlist_node **lastp)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 	struct cpuhp_step *step = cpuhp_get_step(state);
@@ -137,7 +163,17 @@ static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
 	int (*cb)(unsigned int cpu);
 	int ret, cnt;
 
+	if (st->fail == state) {
+		st->fail = CPUHP_INVALID;
+
+		if (!(bringup ? step->startup.single : step->teardown.single))
+			return 0;
+
+		return -EAGAIN;
+	}
+
 	if (!step->multi_instance) {
+		WARN_ON_ONCE(lastp && *lastp);
 		cb = bringup ? step->startup.single : step->teardown.single;
 		if (!cb)
 			return 0;
@@ -152,6 +188,7 @@ static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
 
 	/* Single invocation for instance add/remove */
 	if (node) {
+		WARN_ON_ONCE(lastp && *lastp);
 		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
 		ret = cbm(cpu, node);
 		trace_cpuhp_exit(cpu, st->state, state, ret);
@@ -161,13 +198,23 @@ static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
 	/* State transition. Invoke on all instances */
 	cnt = 0;
 	hlist_for_each(node, &step->list) {
+		if (lastp && node == *lastp)
+			break;
+
 		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
 		ret = cbm(cpu, node);
 		trace_cpuhp_exit(cpu, st->state, state, ret);
-		if (ret)
+		if (ret) {
+			if (!lastp)
 				goto err;
+
+			*lastp = node;
+			return ret;
+		}
 		cnt++;
 	}
+	if (lastp)
+		*lastp = NULL;
 	return 0;
 err:
 	/* Rollback the instances if one failed */
@@ -178,12 +225,39 @@ static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
 	hlist_for_each(node, &step->list) {
 		if (!cnt--)
 			break;
-		cbm(cpu, node);
+
+		trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
+		ret = cbm(cpu, node);
+		trace_cpuhp_exit(cpu, st->state, state, ret);
+		/*
+		 * Rollback must not fail,
+		 */
+		WARN_ON_ONCE(ret);
 	}
 	return ret;
 }
 
 #ifdef CONFIG_SMP
+static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
+{
+	struct completion *done = bringup ? &st->done_up : &st->done_down;
+	wait_for_completion(done);
+}
+
+static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
+{
+	struct completion *done = bringup ? &st->done_up : &st->done_down;
+	complete(done);
+}
+
+/*
+ * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
+ */
+static bool cpuhp_is_atomic_state(enum cpuhp_state state)
+{
+	return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
+}
+
 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
 static DEFINE_MUTEX(cpu_add_remove_lock);
 bool cpuhp_tasks_frozen;
@@ -271,14 +345,79 @@ void cpu_hotplug_enable(void)
 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
 #endif	/* CONFIG_HOTPLUG_CPU */
 
-static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st);
+static inline enum cpuhp_state
+cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
+{
+	enum cpuhp_state prev_state = st->state;
+
+	st->rollback = false;
+	st->last = NULL;
+
+	st->target = target;
+	st->single = false;
+	st->bringup = st->state < target;
+
+	return prev_state;
+}
+
+static inline void
+cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
+{
+	st->rollback = true;
+
+	/*
+	 * If we have st->last we need to undo partial multi_instance of this
+	 * state first. Otherwise start undo at the previous state.
+	 */
+	if (!st->last) {
+		if (st->bringup)
+			st->state--;
+		else
+			st->state++;
+	}
+
+	st->target = prev_state;
+	st->bringup = !st->bringup;
+}
+
+/* Regular hotplug invocation of the AP hotplug thread */
+static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
+{
+	if (!st->single && st->state == st->target)
+		return;
+
+	st->result = 0;
+	/*
+	 * Make sure the above stores are visible before should_run becomes
+	 * true. Paired with the mb() above in cpuhp_thread_fun()
+	 */
+	smp_mb();
+	st->should_run = true;
+	wake_up_process(st->thread);
+	wait_for_ap_thread(st, st->bringup);
+}
+
+static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
+{
+	enum cpuhp_state prev_state;
+	int ret;
+
+	prev_state = cpuhp_set_state(st, target);
+	__cpuhp_kick_ap(st);
+	if ((ret = st->result)) {
+		cpuhp_reset_state(st, prev_state);
+		__cpuhp_kick_ap(st);
+	}
+
+	return ret;
+}
 
 static int bringup_wait_for_ap(unsigned int cpu)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 
 	/* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
-	wait_for_completion(&st->done);
+	wait_for_ap_thread(st, true);
 	if (WARN_ON_ONCE((!cpu_online(cpu))))
 		return -ECANCELED;
 
@@ -286,12 +425,10 @@ static int bringup_wait_for_ap(unsigned int cpu)
 	stop_machine_unpark(cpu);
 	kthread_unpark(st->thread);
 
-	/* Should we go further up ? */
-	if (st->target > CPUHP_AP_ONLINE_IDLE) {
-		__cpuhp_kick_ap_work(st);
-		wait_for_completion(&st->done);
-	}
-	return st->result;
+	if (st->target <= CPUHP_AP_ONLINE_IDLE)
+		return 0;
+
+	return cpuhp_kick_ap(st, st->target);
 }
 
 static int bringup_cpu(unsigned int cpu)
@@ -317,32 +454,6 @@ static int bringup_cpu(unsigned int cpu)
 /*
  * Hotplug state machine related functions
  */
-static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
-{
-	for (st->state++; st->state < st->target; st->state++) {
-		struct cpuhp_step *step = cpuhp_get_step(st->state);
-
-		if (!step->skip_onerr)
-			cpuhp_invoke_callback(cpu, st->state, true, NULL);
-	}
-}
-
-static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
-				enum cpuhp_state target)
-{
-	enum cpuhp_state prev_state = st->state;
-	int ret = 0;
-
-	for (; st->state > target; st->state--) {
-		ret = cpuhp_invoke_callback(cpu, st->state, false, NULL);
-		if (ret) {
-			st->target = prev_state;
-			undo_cpu_down(cpu, st);
-			break;
-		}
-	}
-	return ret;
-}
 
 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
 {
@@ -350,7 +461,7 @@ static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
 		struct cpuhp_step *step = cpuhp_get_step(st->state);
 
 		if (!step->skip_onerr)
-			cpuhp_invoke_callback(cpu, st->state, false, NULL);
+			cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
 	}
 }
 
@@ -362,7 +473,7 @@ static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
 
 	while (st->state < target) {
 		st->state++;
-		ret = cpuhp_invoke_callback(cpu, st->state, true, NULL);
+		ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
 		if (ret) {
 			st->target = prev_state;
 			undo_cpu_up(cpu, st);
@@ -379,7 +490,8 @@ static void cpuhp_create(unsigned int cpu)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 
-	init_completion(&st->done);
+	init_completion(&st->done_up);
+	init_completion(&st->done_down);
 }
 
 static int cpuhp_should_run(unsigned int cpu)
@@ -389,69 +501,90 @@ static int cpuhp_should_run(unsigned int cpu)
 	return st->should_run;
 }
 
-/* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */
-static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st)
-{
-	enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU);
-
-	return cpuhp_down_callbacks(cpu, st, target);
-}
-
-/* Execute the online startup callbacks. Used to be CPU_ONLINE */
-static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st)
-{
-	return cpuhp_up_callbacks(cpu, st, st->target);
-}
-
 /*
  * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
  * callbacks when a state gets [un]installed at runtime.
+ *
+ * Each invocation of this function by the smpboot thread does a single AP
+ * state callback.
+ *
+ * It has 3 modes of operation:
+ *  - single: runs st->cb_state
+ *  - up:     runs ++st->state, while st->state < st->target
+ *  - down:   runs st->state--, while st->state > st->target
+ *
+ * When complete or on error, should_run is cleared and the completion is fired.
  */
 static void cpuhp_thread_fun(unsigned int cpu)
 {
 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
-	int ret = 0;
+	bool bringup = st->bringup;
+	enum cpuhp_state state;
 
 	/*
-	 * Paired with the mb() in cpuhp_kick_ap_work and
-	 * cpuhp_invoke_ap_callback, so the work set is consistent visible.
+	 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
+	 * that if we see ->should_run we also see the rest of the state.
 	 */
 	smp_mb();
-	if (!st->should_run)
+
+	if (WARN_ON_ONCE(!st->should_run))
 		return;
 
-	st->should_run = false;
+	cpuhp_lock_acquire(bringup);
 
-	lock_map_acquire(&cpuhp_state_lock_map);
-	/* Single callback invocation for [un]install ? */
 	if (st->single) {
-		if (st->cb_state < CPUHP_AP_ONLINE) {
+		state = st->cb_state;
+		st->should_run = false;
+	} else {
+		if (bringup) {
+			st->state++;
+			state = st->state;
+			st->should_run = (st->state < st->target);
+			WARN_ON_ONCE(st->state > st->target);
+		} else {
+			state = st->state;
+			st->state--;
+			st->should_run = (st->state > st->target);
+			WARN_ON_ONCE(st->state < st->target);
+		}
+	}
+
+	WARN_ON_ONCE(!cpuhp_is_ap_state(state));
+
+	if (st->rollback) {
+		struct cpuhp_step *step = cpuhp_get_step(state);
+		if (step->skip_onerr)
+			goto next;
+	}
+
+	if (cpuhp_is_atomic_state(state)) {
 		local_irq_disable();
-			ret = cpuhp_invoke_callback(cpu, st->cb_state,
-						    st->bringup, st->node);
+		st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
 		local_irq_enable();
+
+		/*
+		 * STARTING/DYING must not fail!
+		 */
+		WARN_ON_ONCE(st->result);
 	} else {
-			ret = cpuhp_invoke_callback(cpu, st->cb_state,
-						    st->bringup, st->node);
+		st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
 	}
-	} else if (st->rollback) {
-		BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
 
-		undo_cpu_down(cpu, st);
-		st->rollback = false;
-	} else {
-		/* Cannot happen .... */
-		BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
-
-		/* Regular hotplug work */
-		if (st->state < st->target)
-			ret = cpuhp_ap_online(cpu, st);
-		else if (st->state > st->target)
-			ret = cpuhp_ap_offline(cpu, st);
+	if (st->result) {
+		/*
+		 * If we fail on a rollback, we're up a creek without no
+		 * paddle, no way forward, no way back. We loose, thanks for
+		 * playing.
+		 */
+		WARN_ON_ONCE(st->rollback);
+		st->should_run = false;
 	}
-	lock_map_release(&cpuhp_state_lock_map);
-	st->result = ret;
-	complete(&st->done);
+
+next:
+	cpuhp_lock_release(bringup);
+
+	if (!st->should_run)
+		complete_ap_thread(st, bringup);
 }
 
 /* Invoke a single callback on a remote cpu */
@@ -460,62 +593,64 @@ cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
 			 struct hlist_node *node)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
+	int ret;
 
 	if (!cpu_online(cpu))
 		return 0;
 
-	lock_map_acquire(&cpuhp_state_lock_map);
-	lock_map_release(&cpuhp_state_lock_map);
+	cpuhp_lock_acquire(false);
+	cpuhp_lock_release(false);
+
+	cpuhp_lock_acquire(true);
+	cpuhp_lock_release(true);
 
 	/*
 	 * If we are up and running, use the hotplug thread. For early calls
 	 * we invoke the thread function directly.
 	 */
 	if (!st->thread)
-		return cpuhp_invoke_callback(cpu, state, bringup, node);
+		return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
+
+	st->rollback = false;
+	st->last = NULL;
 
+	st->node = node;
+	st->bringup = bringup;
 	st->cb_state = state;
 	st->single = true;
-	st->bringup = bringup;
-	st->node = node;
 
-	/*
-	 * Make sure the above stores are visible before should_run becomes
-	 * true. Paired with the mb() above in cpuhp_thread_fun()
-	 */
-	smp_mb();
-	st->should_run = true;
-	wake_up_process(st->thread);
-	wait_for_completion(&st->done);
-	return st->result;
-}
+	__cpuhp_kick_ap(st);
 
-/* Regular hotplug invocation of the AP hotplug thread */
-static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st)
-{
-	st->result = 0;
-	st->single = false;
 	/*
-	 * Make sure the above stores are visible before should_run becomes
-	 * true. Paired with the mb() above in cpuhp_thread_fun()
+	 * If we failed and did a partial, do a rollback.
 	 */
-	smp_mb();
-	st->should_run = true;
-	wake_up_process(st->thread);
+	if ((ret = st->result) && st->last) {
+		st->rollback = true;
+		st->bringup = !bringup;
+
+		__cpuhp_kick_ap(st);
+	}
+
+	return ret;
 }
 
 static int cpuhp_kick_ap_work(unsigned int cpu)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
-	enum cpuhp_state state = st->state;
+	enum cpuhp_state prev_state = st->state;
+	int ret;
+
+	cpuhp_lock_acquire(false);
+	cpuhp_lock_release(false);
 
-	trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work);
-	lock_map_acquire(&cpuhp_state_lock_map);
-	lock_map_release(&cpuhp_state_lock_map);
-	__cpuhp_kick_ap_work(st);
-	wait_for_completion(&st->done);
-	trace_cpuhp_exit(cpu, st->state, state, st->result);
-	return st->result;
+	cpuhp_lock_acquire(true);
+	cpuhp_lock_release(true);
+
+	trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
+	ret = cpuhp_kick_ap(st, st->target);
+	trace_cpuhp_exit(cpu, st->state, prev_state, ret);
+
+	return ret;
 }
 
 static struct smp_hotplug_thread cpuhp_threads = {
@@ -581,6 +716,7 @@ static int take_cpu_down(void *_param)
 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
 	enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
 	int err, cpu = smp_processor_id();
+	int ret;
 
 	/* Ensure this CPU doesn't handle any more interrupts. */
 	err = __cpu_disable();
@@ -594,8 +730,13 @@ static int take_cpu_down(void *_param)
 	WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
 	st->state--;
 	/* Invoke the former CPU_DYING callbacks */
-	for (; st->state > target; st->state--)
-		cpuhp_invoke_callback(cpu, st->state, false, NULL);
+	for (; st->state > target; st->state--) {
+		ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
+		/*
+		 * DYING must not fail!
+		 */
+		WARN_ON_ONCE(ret);
+	}
 
 	/* Give up timekeeping duties */
 	tick_handover_do_timer();
@@ -639,7 +780,7 @@ static int takedown_cpu(unsigned int cpu)
 	 *
 	 * Wait for the stop thread to go away.
 	 */
-	wait_for_completion(&st->done);
+	wait_for_ap_thread(st, false);
 	BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
 
 	/* Interrupts are moved away from the dying cpu, reenable alloc/free */
@@ -658,7 +799,7 @@ static void cpuhp_complete_idle_dead(void *arg)
 {
 	struct cpuhp_cpu_state *st = arg;
 
-	complete(&st->done);
+	complete_ap_thread(st, false);
 }
 
 void cpuhp_report_idle_dead(void)
@@ -676,11 +817,32 @@ void cpuhp_report_idle_dead(void)
 				 cpuhp_complete_idle_dead, st, 0);
 }
 
-#else
-#define takedown_cpu		NULL
-#endif
+static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
+{
+	for (st->state++; st->state < st->target; st->state++) {
+		struct cpuhp_step *step = cpuhp_get_step(st->state);
 
-#ifdef CONFIG_HOTPLUG_CPU
+		if (!step->skip_onerr)
+			cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
+	}
+}
+
+static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
+				enum cpuhp_state target)
+{
+	enum cpuhp_state prev_state = st->state;
+	int ret = 0;
+
+	for (; st->state > target; st->state--) {
+		ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
+		if (ret) {
+			st->target = prev_state;
+			undo_cpu_down(cpu, st);
+			break;
+		}
+	}
+	return ret;
+}
 
 /* Requires cpu_add_remove_lock to be held */
 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
@@ -699,13 +861,13 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
 
 	cpuhp_tasks_frozen = tasks_frozen;
 
-	prev_state = st->state;
-	st->target = target;
+	prev_state = cpuhp_set_state(st, target);
 	/*
 	 * If the current CPU state is in the range of the AP hotplug thread,
 	 * then we need to kick the thread.
 	 */
 	if (st->state > CPUHP_TEARDOWN_CPU) {
+		st->target = max((int)target, CPUHP_TEARDOWN_CPU);
 		ret = cpuhp_kick_ap_work(cpu);
 		/*
 		 * The AP side has done the error rollback already. Just
@@ -720,6 +882,8 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
 		 */
 		if (st->state > CPUHP_TEARDOWN_CPU)
 			goto out;
+
+		st->target = target;
 	}
 	/*
 	 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
@@ -727,9 +891,8 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
 	 */
 	ret = cpuhp_down_callbacks(cpu, st, target);
 	if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
-		st->target = prev_state;
-		st->rollback = true;
-		cpuhp_kick_ap_work(cpu);
+		cpuhp_reset_state(st, prev_state);
+		__cpuhp_kick_ap(st);
 	}
 
 out:
@@ -754,11 +917,15 @@ static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
 	cpu_maps_update_done();
 	return err;
 }
+
 int cpu_down(unsigned int cpu)
 {
 	return do_cpu_down(cpu, CPUHP_OFFLINE);
 }
 EXPORT_SYMBOL(cpu_down);
+
+#else
+#define takedown_cpu		NULL
 #endif /*CONFIG_HOTPLUG_CPU*/
 
 /**
@@ -772,11 +939,16 @@ void notify_cpu_starting(unsigned int cpu)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 	enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
+	int ret;
 
 	rcu_cpu_starting(cpu);	/* Enables RCU usage on this CPU. */
 	while (st->state < target) {
 		st->state++;
-		cpuhp_invoke_callback(cpu, st->state, true, NULL);
+		ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
+		/*
+		 * STARTING must not fail!
+		 */
+		WARN_ON_ONCE(ret);
 	}
 }
 
@@ -794,7 +966,7 @@ void cpuhp_online_idle(enum cpuhp_state state)
 		return;
 
 	st->state = CPUHP_AP_ONLINE_IDLE;
-	complete(&st->done);
+	complete_ap_thread(st, true);
 }
 
 /* Requires cpu_add_remove_lock to be held */
@@ -829,7 +1001,7 @@ static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
 
 	cpuhp_tasks_frozen = tasks_frozen;
 
-	st->target = target;
+	cpuhp_set_state(st, target);
 	/*
 	 * If the current CPU state is in the range of the AP hotplug thread,
 	 * then we need to kick the thread once more.
@@ -1296,6 +1468,10 @@ static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
 	struct cpuhp_step *sp = cpuhp_get_step(state);
 	int ret;
 
+	/*
+	 * If there's nothing to do, we done.
+	 * Relies on the union for multi_instance.
+	 */
 	if ((bringup && !sp->startup.single) ||
 	    (!bringup && !sp->teardown.single))
 		return 0;
@@ -1307,9 +1483,9 @@ static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
 	if (cpuhp_is_ap_state(state))
 		ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
 	else
-		ret = cpuhp_invoke_callback(cpu, state, bringup, node);
+		ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
 #else
-	ret = cpuhp_invoke_callback(cpu, state, bringup, node);
+	ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
 #endif
 	BUG_ON(ret && !bringup);
 	return ret;
@@ -1641,9 +1817,55 @@ static ssize_t show_cpuhp_target(struct device *dev,
 }
 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
 
+
+static ssize_t write_cpuhp_fail(struct device *dev,
+				struct device_attribute *attr,
+				const char *buf, size_t count)
+{
+	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
+	struct cpuhp_step *sp;
+	int fail, ret;
+
+	ret = kstrtoint(buf, 10, &fail);
+	if (ret)
+		return ret;
+
+	/*
+	 * Cannot fail STARTING/DYING callbacks.
+	 */
+	if (cpuhp_is_atomic_state(fail))
+		return -EINVAL;
+
+	/*
+	 * Cannot fail anything that doesn't have callbacks.
+	 */
+	mutex_lock(&cpuhp_state_mutex);
+	sp = cpuhp_get_step(fail);
+	if (!sp->startup.single && !sp->teardown.single)
+		ret = -EINVAL;
+	mutex_unlock(&cpuhp_state_mutex);
+	if (ret)
+		return ret;
+
+	st->fail = fail;
+
+	return count;
+}
+
+static ssize_t show_cpuhp_fail(struct device *dev,
+			       struct device_attribute *attr, char *buf)
+{
+	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
+
+	return sprintf(buf, "%d\n", st->fail);
+}
+
+static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
+
 static struct attribute *cpuhp_cpu_attrs[] = {
 	&dev_attr_state.attr,
 	&dev_attr_target.attr,
+	&dev_attr_fail.attr,
 	NULL
 };