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/module.h>
14 #include <linux/kthread.h>
15 #include <linux/stop_machine.h>
16 #include <linux/mutex.h>
19 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
20 static DEFINE_MUTEX(cpu_add_remove_lock);
22 static __cpuinitdata RAW_NOTIFIER_HEAD(cpu_chain);
24 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
25 * Should always be manipulated under cpu_add_remove_lock
27 static int cpu_hotplug_disabled;
30 struct task_struct *active_writer;
31 struct mutex lock; /* Synchronizes accesses to refcount, */
33 * Also blocks the new readers during
34 * an ongoing cpu hotplug operation.
38 .active_writer = NULL,
39 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
43 #ifdef CONFIG_HOTPLUG_CPU
45 void get_online_cpus(void)
48 if (cpu_hotplug.active_writer == current)
50 mutex_lock(&cpu_hotplug.lock);
51 cpu_hotplug.refcount++;
52 mutex_unlock(&cpu_hotplug.lock);
55 EXPORT_SYMBOL_GPL(get_online_cpus);
57 void put_online_cpus(void)
59 if (cpu_hotplug.active_writer == current)
61 mutex_lock(&cpu_hotplug.lock);
62 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
63 wake_up_process(cpu_hotplug.active_writer);
64 mutex_unlock(&cpu_hotplug.lock);
67 EXPORT_SYMBOL_GPL(put_online_cpus);
69 #endif /* CONFIG_HOTPLUG_CPU */
72 * The following two API's must be used when attempting
73 * to serialize the updates to cpu_online_mask, cpu_present_mask.
75 void cpu_maps_update_begin(void)
77 mutex_lock(&cpu_add_remove_lock);
80 void cpu_maps_update_done(void)
82 mutex_unlock(&cpu_add_remove_lock);
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);
126 /* Need to know about CPUs going up/down? */
127 int __ref register_cpu_notifier(struct notifier_block *nb)
130 cpu_maps_update_begin();
131 ret = raw_notifier_chain_register(&cpu_chain, nb);
132 cpu_maps_update_done();
136 #ifdef CONFIG_HOTPLUG_CPU
138 EXPORT_SYMBOL(register_cpu_notifier);
140 void __ref unregister_cpu_notifier(struct notifier_block *nb)
142 cpu_maps_update_begin();
143 raw_notifier_chain_unregister(&cpu_chain, nb);
144 cpu_maps_update_done();
146 EXPORT_SYMBOL(unregister_cpu_notifier);
148 static inline void check_for_tasks(int cpu)
150 struct task_struct *p;
152 write_lock_irq(&tasklist_lock);
153 for_each_process(p) {
154 if (task_cpu(p) == cpu && p->state == TASK_RUNNING &&
155 (!cputime_eq(p->utime, cputime_zero) ||
156 !cputime_eq(p->stime, cputime_zero)))
157 printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d "
158 "(state = %ld, flags = %x)\n",
159 p->comm, task_pid_nr(p), cpu,
162 write_unlock_irq(&tasklist_lock);
165 struct take_cpu_down_param {
166 struct task_struct *caller;
171 /* Take this CPU down. */
172 static int __ref take_cpu_down(void *_param)
174 struct take_cpu_down_param *param = _param;
175 unsigned int cpu = (unsigned long)param->hcpu;
178 /* Ensure this CPU doesn't handle any more interrupts. */
179 err = __cpu_disable();
183 raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
186 if (task_cpu(param->caller) == cpu)
187 move_task_off_dead_cpu(cpu, param->caller);
188 /* Force idle task to run as soon as we yield: it should
189 immediately notice cpu is offline and die quickly. */
194 /* Requires cpu_add_remove_lock to be held */
195 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
197 int err, nr_calls = 0;
198 void *hcpu = (void *)(long)cpu;
199 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
200 struct take_cpu_down_param tcd_param = {
206 if (num_online_cpus() == 1)
209 if (!cpu_online(cpu))
213 set_cpu_active(cpu, false);
214 err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
215 hcpu, -1, &nr_calls);
216 if (err == NOTIFY_BAD) {
217 set_cpu_active(cpu, true);
220 __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
221 hcpu, nr_calls, NULL);
222 printk("%s: attempt to take down CPU %u failed\n",
228 err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
230 set_cpu_active(cpu, true);
231 /* CPU didn't die: tell everyone. Can't complain. */
232 if (raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
238 BUG_ON(cpu_online(cpu));
240 /* Wait for it to sleep (leaving idle task). */
241 while (!idle_cpu(cpu))
244 /* This actually kills the CPU. */
247 /* CPU is completely dead: tell everyone. Too late to complain. */
248 if (raw_notifier_call_chain(&cpu_chain, CPU_DEAD | mod,
252 check_for_tasks(cpu);
257 if (raw_notifier_call_chain(&cpu_chain, CPU_POST_DEAD | mod,
264 int __ref cpu_down(unsigned int cpu)
268 err = stop_machine_create();
271 cpu_maps_update_begin();
273 if (cpu_hotplug_disabled) {
278 err = _cpu_down(cpu, 0);
281 cpu_maps_update_done();
282 stop_machine_destroy();
285 EXPORT_SYMBOL(cpu_down);
286 #endif /*CONFIG_HOTPLUG_CPU*/
288 /* Requires cpu_add_remove_lock to be held */
289 static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
291 int ret, nr_calls = 0;
292 void *hcpu = (void *)(long)cpu;
293 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
295 if (cpu_online(cpu) || !cpu_present(cpu))
299 ret = __raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE | mod, hcpu,
301 if (ret == NOTIFY_BAD) {
303 printk("%s: attempt to bring up CPU %u failed\n",
309 /* Arch-specific enabling code. */
313 BUG_ON(!cpu_online(cpu));
315 set_cpu_active(cpu, true);
317 /* Now call notifier in preparation. */
318 raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu);
322 __raw_notifier_call_chain(&cpu_chain,
323 CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
329 int __cpuinit cpu_up(unsigned int cpu)
332 if (!cpu_possible(cpu)) {
333 printk(KERN_ERR "can't online cpu %d because it is not "
334 "configured as may-hotadd at boot time\n", cpu);
335 #if defined(CONFIG_IA64)
336 printk(KERN_ERR "please check additional_cpus= boot "
342 cpu_maps_update_begin();
344 if (cpu_hotplug_disabled) {
349 err = _cpu_up(cpu, 0);
352 cpu_maps_update_done();
356 #ifdef CONFIG_PM_SLEEP_SMP
357 static cpumask_var_t frozen_cpus;
359 int disable_nonboot_cpus(void)
361 int cpu, first_cpu, error;
363 error = stop_machine_create();
366 cpu_maps_update_begin();
367 first_cpu = cpumask_first(cpu_online_mask);
369 * We take down all of the non-boot CPUs in one shot to avoid races
370 * with the userspace trying to use the CPU hotplug at the same time
372 cpumask_clear(frozen_cpus);
374 printk("Disabling non-boot CPUs ...\n");
375 for_each_online_cpu(cpu) {
376 if (cpu == first_cpu)
378 error = _cpu_down(cpu, 1);
380 cpumask_set_cpu(cpu, frozen_cpus);
382 printk(KERN_ERR "Error taking CPU%d down: %d\n",
389 BUG_ON(num_online_cpus() > 1);
390 /* Make sure the CPUs won't be enabled by someone else */
391 cpu_hotplug_disabled = 1;
393 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
395 cpu_maps_update_done();
396 stop_machine_destroy();
400 void __weak arch_enable_nonboot_cpus_begin(void)
404 void __weak arch_enable_nonboot_cpus_end(void)
408 void __ref enable_nonboot_cpus(void)
412 /* Allow everyone to use the CPU hotplug again */
413 cpu_maps_update_begin();
414 cpu_hotplug_disabled = 0;
415 if (cpumask_empty(frozen_cpus))
418 printk("Enabling non-boot CPUs ...\n");
420 arch_enable_nonboot_cpus_begin();
422 for_each_cpu(cpu, frozen_cpus) {
423 error = _cpu_up(cpu, 1);
425 printk("CPU%d is up\n", cpu);
428 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
431 arch_enable_nonboot_cpus_end();
433 cpumask_clear(frozen_cpus);
435 cpu_maps_update_done();
438 static int alloc_frozen_cpus(void)
440 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
444 core_initcall(alloc_frozen_cpus);
445 #endif /* CONFIG_PM_SLEEP_SMP */
448 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
449 * @cpu: cpu that just started
451 * This function calls the cpu_chain notifiers with CPU_STARTING.
452 * It must be called by the arch code on the new cpu, before the new cpu
453 * enables interrupts and before the "boot" cpu returns from __cpu_up().
455 void __cpuinit notify_cpu_starting(unsigned int cpu)
457 unsigned long val = CPU_STARTING;
459 #ifdef CONFIG_PM_SLEEP_SMP
460 if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus))
461 val = CPU_STARTING_FROZEN;
462 #endif /* CONFIG_PM_SLEEP_SMP */
463 raw_notifier_call_chain(&cpu_chain, val, (void *)(long)cpu);
466 #endif /* CONFIG_SMP */
469 * cpu_bit_bitmap[] is a special, "compressed" data structure that
470 * represents all NR_CPUS bits binary values of 1<<nr.
472 * It is used by cpumask_of() to get a constant address to a CPU
473 * mask value that has a single bit set only.
476 /* cpu_bit_bitmap[0] is empty - so we can back into it */
477 #define MASK_DECLARE_1(x) [x+1][0] = 1UL << (x)
478 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
479 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
480 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
482 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
484 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
485 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
486 #if BITS_PER_LONG > 32
487 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
488 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
491 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
493 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
494 EXPORT_SYMBOL(cpu_all_bits);
496 #ifdef CONFIG_INIT_ALL_POSSIBLE
497 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
500 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
502 const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
503 EXPORT_SYMBOL(cpu_possible_mask);
505 static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
506 const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
507 EXPORT_SYMBOL(cpu_online_mask);
509 static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
510 const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
511 EXPORT_SYMBOL(cpu_present_mask);
513 static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
514 const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
515 EXPORT_SYMBOL(cpu_active_mask);
517 void set_cpu_possible(unsigned int cpu, bool possible)
520 cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
522 cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
525 void set_cpu_present(unsigned int cpu, bool present)
528 cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
530 cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
533 void set_cpu_online(unsigned int cpu, bool online)
536 cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
538 cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
541 void set_cpu_active(unsigned int cpu, bool active)
544 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
546 cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
549 void init_cpu_present(const struct cpumask *src)
551 cpumask_copy(to_cpumask(cpu_present_bits), src);
554 void init_cpu_possible(const struct cpumask *src)
556 cpumask_copy(to_cpumask(cpu_possible_bits), src);
559 void init_cpu_online(const struct cpumask *src)
561 cpumask_copy(to_cpumask(cpu_online_bits), src);