2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
4 * This code is licenced under the GPL.
6 #include <linux/proc_fs.h>
8 #include <linux/init.h>
9 #include <linux/notifier.h>
10 #include <linux/sched.h>
11 #include <linux/unistd.h>
12 #include <linux/cpu.h>
13 #include <linux/export.h>
14 #include <linux/kthread.h>
15 #include <linux/stop_machine.h>
16 #include <linux/mutex.h>
17 #include <linux/gfp.h>
18 #include <linux/suspend.h>
21 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
22 static DEFINE_MUTEX(cpu_add_remove_lock);
25 * The following two API's must be used when attempting
26 * to serialize the updates to cpu_online_mask, cpu_present_mask.
28 void cpu_maps_update_begin(void)
30 mutex_lock(&cpu_add_remove_lock);
33 void cpu_maps_update_done(void)
35 mutex_unlock(&cpu_add_remove_lock);
38 static RAW_NOTIFIER_HEAD(cpu_chain);
40 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
41 * Should always be manipulated under cpu_add_remove_lock
43 static int cpu_hotplug_disabled;
45 #ifdef CONFIG_HOTPLUG_CPU
48 struct task_struct *active_writer;
49 struct mutex lock; /* Synchronizes accesses to refcount, */
51 * Also blocks the new readers during
52 * an ongoing cpu hotplug operation.
56 .active_writer = NULL,
57 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
61 void get_online_cpus(void)
64 if (cpu_hotplug.active_writer == current)
66 mutex_lock(&cpu_hotplug.lock);
67 cpu_hotplug.refcount++;
68 mutex_unlock(&cpu_hotplug.lock);
71 EXPORT_SYMBOL_GPL(get_online_cpus);
73 void put_online_cpus(void)
75 if (cpu_hotplug.active_writer == current)
77 mutex_lock(&cpu_hotplug.lock);
78 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
79 wake_up_process(cpu_hotplug.active_writer);
80 mutex_unlock(&cpu_hotplug.lock);
83 EXPORT_SYMBOL_GPL(put_online_cpus);
86 * This ensures that the hotplug operation can begin only when the
87 * refcount goes to zero.
89 * Note that during a cpu-hotplug operation, the new readers, if any,
90 * will be blocked by the cpu_hotplug.lock
92 * Since cpu_hotplug_begin() is always called after invoking
93 * cpu_maps_update_begin(), we can be sure that only one writer is active.
95 * Note that theoretically, there is a possibility of a livelock:
96 * - Refcount goes to zero, last reader wakes up the sleeping
98 * - Last reader unlocks the cpu_hotplug.lock.
99 * - A new reader arrives at this moment, bumps up the refcount.
100 * - The writer acquires the cpu_hotplug.lock finds the refcount
101 * non zero and goes to sleep again.
103 * However, this is very difficult to achieve in practice since
104 * get_online_cpus() not an api which is called all that often.
107 static void cpu_hotplug_begin(void)
109 cpu_hotplug.active_writer = current;
112 mutex_lock(&cpu_hotplug.lock);
113 if (likely(!cpu_hotplug.refcount))
115 __set_current_state(TASK_UNINTERRUPTIBLE);
116 mutex_unlock(&cpu_hotplug.lock);
121 static void cpu_hotplug_done(void)
123 cpu_hotplug.active_writer = NULL;
124 mutex_unlock(&cpu_hotplug.lock);
128 * Wait for currently running CPU hotplug operations to complete (if any) and
129 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
130 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
131 * hotplug path before performing hotplug operations. So acquiring that lock
132 * guarantees mutual exclusion from any currently running hotplug operations.
134 void cpu_hotplug_disable(void)
136 cpu_maps_update_begin();
137 cpu_hotplug_disabled = 1;
138 cpu_maps_update_done();
141 void cpu_hotplug_enable(void)
143 cpu_maps_update_begin();
144 cpu_hotplug_disabled = 0;
145 cpu_maps_update_done();
148 #else /* #if CONFIG_HOTPLUG_CPU */
149 static void cpu_hotplug_begin(void) {}
150 static void cpu_hotplug_done(void) {}
151 #endif /* #else #if CONFIG_HOTPLUG_CPU */
153 /* Need to know about CPUs going up/down? */
154 int __ref register_cpu_notifier(struct notifier_block *nb)
157 cpu_maps_update_begin();
158 ret = raw_notifier_chain_register(&cpu_chain, nb);
159 cpu_maps_update_done();
163 static int __cpu_notify(unsigned long val, void *v, int nr_to_call,
168 ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call,
171 return notifier_to_errno(ret);
174 static int cpu_notify(unsigned long val, void *v)
176 return __cpu_notify(val, v, -1, NULL);
179 EXPORT_SYMBOL(register_cpu_notifier);
181 void __ref unregister_cpu_notifier(struct notifier_block *nb)
183 cpu_maps_update_begin();
184 raw_notifier_chain_unregister(&cpu_chain, nb);
185 cpu_maps_update_done();
187 EXPORT_SYMBOL(unregister_cpu_notifier);
189 #ifdef CONFIG_HOTPLUG_CPU
191 static void cpu_notify_nofail(unsigned long val, void *v)
193 BUG_ON(cpu_notify(val, v));
196 static inline void check_for_tasks(int cpu)
198 struct task_struct *p;
200 write_lock_irq(&tasklist_lock);
201 for_each_process(p) {
202 if (task_cpu(p) == cpu && p->state == TASK_RUNNING &&
203 (!cputime_eq(p->utime, cputime_zero) ||
204 !cputime_eq(p->stime, cputime_zero)))
205 printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d "
206 "(state = %ld, flags = %x)\n",
207 p->comm, task_pid_nr(p), cpu,
210 write_unlock_irq(&tasklist_lock);
213 struct take_cpu_down_param {
218 /* Take this CPU down. */
219 static int __ref take_cpu_down(void *_param)
221 struct take_cpu_down_param *param = _param;
224 /* Ensure this CPU doesn't handle any more interrupts. */
225 err = __cpu_disable();
229 cpu_notify(CPU_DYING | param->mod, param->hcpu);
233 /* Requires cpu_add_remove_lock to be held */
234 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
236 int err, nr_calls = 0;
237 void *hcpu = (void *)(long)cpu;
238 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
239 struct take_cpu_down_param tcd_param = {
244 if (num_online_cpus() == 1)
247 if (!cpu_online(cpu))
252 err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
255 __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL);
256 printk("%s: attempt to take down CPU %u failed\n",
261 err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
263 /* CPU didn't die: tell everyone. Can't complain. */
264 cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
268 BUG_ON(cpu_online(cpu));
271 * The migration_call() CPU_DYING callback will have removed all
272 * runnable tasks from the cpu, there's only the idle task left now
273 * that the migration thread is done doing the stop_machine thing.
275 * Wait for the stop thread to go away.
277 while (!idle_cpu(cpu))
280 /* This actually kills the CPU. */
283 /* CPU is completely dead: tell everyone. Too late to complain. */
284 cpu_notify_nofail(CPU_DEAD | mod, hcpu);
286 check_for_tasks(cpu);
291 cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
295 int __ref cpu_down(unsigned int cpu)
299 cpu_maps_update_begin();
301 if (cpu_hotplug_disabled) {
306 err = _cpu_down(cpu, 0);
309 cpu_maps_update_done();
312 EXPORT_SYMBOL(cpu_down);
313 #endif /*CONFIG_HOTPLUG_CPU*/
315 /* Requires cpu_add_remove_lock to be held */
316 static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
318 int ret, nr_calls = 0;
319 void *hcpu = (void *)(long)cpu;
320 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
322 if (cpu_online(cpu) || !cpu_present(cpu))
326 ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls);
329 printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
334 /* Arch-specific enabling code. */
338 BUG_ON(!cpu_online(cpu));
340 /* Now call notifier in preparation. */
341 cpu_notify(CPU_ONLINE | mod, hcpu);
345 __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
351 int __cpuinit cpu_up(unsigned int cpu)
355 #ifdef CONFIG_MEMORY_HOTPLUG
360 if (!cpu_possible(cpu)) {
361 printk(KERN_ERR "can't online cpu %d because it is not "
362 "configured as may-hotadd at boot time\n", cpu);
363 #if defined(CONFIG_IA64)
364 printk(KERN_ERR "please check additional_cpus= boot "
370 #ifdef CONFIG_MEMORY_HOTPLUG
371 nid = cpu_to_node(cpu);
372 if (!node_online(nid)) {
373 err = mem_online_node(nid);
378 pgdat = NODE_DATA(nid);
381 "Can't online cpu %d due to NULL pgdat\n", cpu);
385 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
386 mutex_lock(&zonelists_mutex);
387 build_all_zonelists(NULL);
388 mutex_unlock(&zonelists_mutex);
392 cpu_maps_update_begin();
394 if (cpu_hotplug_disabled) {
399 err = _cpu_up(cpu, 0);
402 cpu_maps_update_done();
406 #ifdef CONFIG_PM_SLEEP_SMP
407 static cpumask_var_t frozen_cpus;
409 void __weak arch_disable_nonboot_cpus_begin(void)
413 void __weak arch_disable_nonboot_cpus_end(void)
417 int disable_nonboot_cpus(void)
419 int cpu, first_cpu, error = 0;
421 cpu_maps_update_begin();
422 first_cpu = cpumask_first(cpu_online_mask);
424 * We take down all of the non-boot CPUs in one shot to avoid races
425 * with the userspace trying to use the CPU hotplug at the same time
427 cpumask_clear(frozen_cpus);
428 arch_disable_nonboot_cpus_begin();
430 printk("Disabling non-boot CPUs ...\n");
431 for_each_online_cpu(cpu) {
432 if (cpu == first_cpu)
434 error = _cpu_down(cpu, 1);
436 cpumask_set_cpu(cpu, frozen_cpus);
438 printk(KERN_ERR "Error taking CPU%d down: %d\n",
444 arch_disable_nonboot_cpus_end();
447 BUG_ON(num_online_cpus() > 1);
448 /* Make sure the CPUs won't be enabled by someone else */
449 cpu_hotplug_disabled = 1;
451 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
453 cpu_maps_update_done();
457 void __weak arch_enable_nonboot_cpus_begin(void)
461 void __weak arch_enable_nonboot_cpus_end(void)
465 void __ref enable_nonboot_cpus(void)
469 /* Allow everyone to use the CPU hotplug again */
470 cpu_maps_update_begin();
471 cpu_hotplug_disabled = 0;
472 if (cpumask_empty(frozen_cpus))
475 printk(KERN_INFO "Enabling non-boot CPUs ...\n");
477 arch_enable_nonboot_cpus_begin();
479 for_each_cpu(cpu, frozen_cpus) {
480 error = _cpu_up(cpu, 1);
482 printk(KERN_INFO "CPU%d is up\n", cpu);
485 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
488 arch_enable_nonboot_cpus_end();
490 cpumask_clear(frozen_cpus);
492 cpu_maps_update_done();
495 static int alloc_frozen_cpus(void)
497 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
501 core_initcall(alloc_frozen_cpus);
504 * When callbacks for CPU hotplug notifications are being executed, we must
505 * ensure that the state of the system with respect to the tasks being frozen
506 * or not, as reported by the notification, remains unchanged *throughout the
507 * duration* of the execution of the callbacks.
508 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
510 * This synchronization is implemented by mutually excluding regular CPU
511 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
512 * Hibernate notifications.
515 cpu_hotplug_pm_callback(struct notifier_block *nb,
516 unsigned long action, void *ptr)
520 case PM_SUSPEND_PREPARE:
521 case PM_HIBERNATION_PREPARE:
522 cpu_hotplug_disable();
525 case PM_POST_SUSPEND:
526 case PM_POST_HIBERNATION:
527 cpu_hotplug_enable();
538 int cpu_hotplug_pm_sync_init(void)
540 pm_notifier(cpu_hotplug_pm_callback, 0);
543 core_initcall(cpu_hotplug_pm_sync_init);
545 #endif /* CONFIG_PM_SLEEP_SMP */
548 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
549 * @cpu: cpu that just started
551 * This function calls the cpu_chain notifiers with CPU_STARTING.
552 * It must be called by the arch code on the new cpu, before the new cpu
553 * enables interrupts and before the "boot" cpu returns from __cpu_up().
555 void __cpuinit notify_cpu_starting(unsigned int cpu)
557 unsigned long val = CPU_STARTING;
559 #ifdef CONFIG_PM_SLEEP_SMP
560 if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus))
561 val = CPU_STARTING_FROZEN;
562 #endif /* CONFIG_PM_SLEEP_SMP */
563 cpu_notify(val, (void *)(long)cpu);
566 #endif /* CONFIG_SMP */
569 * cpu_bit_bitmap[] is a special, "compressed" data structure that
570 * represents all NR_CPUS bits binary values of 1<<nr.
572 * It is used by cpumask_of() to get a constant address to a CPU
573 * mask value that has a single bit set only.
576 /* cpu_bit_bitmap[0] is empty - so we can back into it */
577 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
578 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
579 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
580 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
582 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
584 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
585 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
586 #if BITS_PER_LONG > 32
587 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
588 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
591 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
593 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
594 EXPORT_SYMBOL(cpu_all_bits);
596 #ifdef CONFIG_INIT_ALL_POSSIBLE
597 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
600 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
602 const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
603 EXPORT_SYMBOL(cpu_possible_mask);
605 static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
606 const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
607 EXPORT_SYMBOL(cpu_online_mask);
609 static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
610 const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
611 EXPORT_SYMBOL(cpu_present_mask);
613 static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
614 const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
615 EXPORT_SYMBOL(cpu_active_mask);
617 void set_cpu_possible(unsigned int cpu, bool possible)
620 cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
622 cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
625 void set_cpu_present(unsigned int cpu, bool present)
628 cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
630 cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
633 void set_cpu_online(unsigned int cpu, bool online)
636 cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
638 cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
641 void set_cpu_active(unsigned int cpu, bool active)
644 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
646 cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
649 void init_cpu_present(const struct cpumask *src)
651 cpumask_copy(to_cpumask(cpu_present_bits), src);
654 void init_cpu_possible(const struct cpumask *src)
656 cpumask_copy(to_cpumask(cpu_possible_bits), src);
659 void init_cpu_online(const struct cpumask *src)
661 cpumask_copy(to_cpumask(cpu_online_bits), src);