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>
17 #include <linux/gfp.h>
20 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
21 static DEFINE_MUTEX(cpu_add_remove_lock);
23 static __cpuinitdata RAW_NOTIFIER_HEAD(cpu_chain);
25 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
26 * Should always be manipulated under cpu_add_remove_lock
28 static int cpu_hotplug_disabled;
31 struct task_struct *active_writer;
32 struct mutex lock; /* Synchronizes accesses to refcount, */
34 * Also blocks the new readers during
35 * an ongoing cpu hotplug operation.
39 .active_writer = NULL,
40 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
44 #ifdef CONFIG_HOTPLUG_CPU
46 void get_online_cpus(void)
49 if (cpu_hotplug.active_writer == current)
51 mutex_lock(&cpu_hotplug.lock);
52 cpu_hotplug.refcount++;
53 mutex_unlock(&cpu_hotplug.lock);
56 EXPORT_SYMBOL_GPL(get_online_cpus);
58 void put_online_cpus(void)
60 if (cpu_hotplug.active_writer == current)
62 mutex_lock(&cpu_hotplug.lock);
63 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
64 wake_up_process(cpu_hotplug.active_writer);
65 mutex_unlock(&cpu_hotplug.lock);
68 EXPORT_SYMBOL_GPL(put_online_cpus);
70 #endif /* CONFIG_HOTPLUG_CPU */
73 * The following two API's must be used when attempting
74 * to serialize the updates to cpu_online_mask, cpu_present_mask.
76 void cpu_maps_update_begin(void)
78 mutex_lock(&cpu_add_remove_lock);
81 void cpu_maps_update_done(void)
83 mutex_unlock(&cpu_add_remove_lock);
87 * This ensures that the hotplug operation can begin only when the
88 * refcount goes to zero.
90 * Note that during a cpu-hotplug operation, the new readers, if any,
91 * will be blocked by the cpu_hotplug.lock
93 * Since cpu_hotplug_begin() is always called after invoking
94 * cpu_maps_update_begin(), we can be sure that only one writer is active.
96 * Note that theoretically, there is a possibility of a livelock:
97 * - Refcount goes to zero, last reader wakes up the sleeping
99 * - Last reader unlocks the cpu_hotplug.lock.
100 * - A new reader arrives at this moment, bumps up the refcount.
101 * - The writer acquires the cpu_hotplug.lock finds the refcount
102 * non zero and goes to sleep again.
104 * However, this is very difficult to achieve in practice since
105 * get_online_cpus() not an api which is called all that often.
108 static void cpu_hotplug_begin(void)
110 cpu_hotplug.active_writer = current;
113 mutex_lock(&cpu_hotplug.lock);
114 if (likely(!cpu_hotplug.refcount))
116 __set_current_state(TASK_UNINTERRUPTIBLE);
117 mutex_unlock(&cpu_hotplug.lock);
122 static void cpu_hotplug_done(void)
124 cpu_hotplug.active_writer = NULL;
125 mutex_unlock(&cpu_hotplug.lock);
127 /* Need to know about CPUs going up/down? */
128 int __ref register_cpu_notifier(struct notifier_block *nb)
131 cpu_maps_update_begin();
132 ret = raw_notifier_chain_register(&cpu_chain, nb);
133 cpu_maps_update_done();
137 static int __cpu_notify(unsigned long val, void *v, int nr_to_call,
142 ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call,
145 return notifier_to_errno(ret);
148 static int cpu_notify(unsigned long val, void *v)
150 return __cpu_notify(val, v, -1, NULL);
153 static void cpu_notify_nofail(unsigned long val, void *v)
157 err = cpu_notify(val, v);
161 #ifdef CONFIG_HOTPLUG_CPU
163 EXPORT_SYMBOL(register_cpu_notifier);
165 void __ref unregister_cpu_notifier(struct notifier_block *nb)
167 cpu_maps_update_begin();
168 raw_notifier_chain_unregister(&cpu_chain, nb);
169 cpu_maps_update_done();
171 EXPORT_SYMBOL(unregister_cpu_notifier);
173 static inline void check_for_tasks(int cpu)
175 struct task_struct *p;
177 write_lock_irq(&tasklist_lock);
178 for_each_process(p) {
179 if (task_cpu(p) == cpu && p->state == TASK_RUNNING &&
180 (!cputime_eq(p->utime, cputime_zero) ||
181 !cputime_eq(p->stime, cputime_zero)))
182 printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d "
183 "(state = %ld, flags = %x)\n",
184 p->comm, task_pid_nr(p), cpu,
187 write_unlock_irq(&tasklist_lock);
190 struct take_cpu_down_param {
191 struct task_struct *caller;
196 /* Take this CPU down. */
197 static int __ref take_cpu_down(void *_param)
199 struct take_cpu_down_param *param = _param;
200 unsigned int cpu = (unsigned long)param->hcpu;
203 /* Ensure this CPU doesn't handle any more interrupts. */
204 err = __cpu_disable();
208 cpu_notify(CPU_DYING | param->mod, param->hcpu);
210 if (task_cpu(param->caller) == cpu)
211 move_task_off_dead_cpu(cpu, param->caller);
212 /* Force idle task to run as soon as we yield: it should
213 immediately notice cpu is offline and die quickly. */
218 /* Requires cpu_add_remove_lock to be held */
219 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
221 int err, nr_calls = 0;
222 void *hcpu = (void *)(long)cpu;
223 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
224 struct take_cpu_down_param tcd_param = {
230 if (num_online_cpus() == 1)
233 if (!cpu_online(cpu))
237 set_cpu_active(cpu, false);
238 err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
240 set_cpu_active(cpu, true);
243 __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL);
244 printk("%s: attempt to take down CPU %u failed\n",
249 err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
251 set_cpu_active(cpu, true);
252 /* CPU didn't die: tell everyone. Can't complain. */
253 cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
257 BUG_ON(cpu_online(cpu));
259 /* Wait for it to sleep (leaving idle task). */
260 while (!idle_cpu(cpu))
263 /* This actually kills the CPU. */
266 /* CPU is completely dead: tell everyone. Too late to complain. */
267 cpu_notify_nofail(CPU_DEAD | mod, hcpu);
269 check_for_tasks(cpu);
274 cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
278 int __ref cpu_down(unsigned int cpu)
282 cpu_maps_update_begin();
284 if (cpu_hotplug_disabled) {
289 err = _cpu_down(cpu, 0);
292 cpu_maps_update_done();
295 EXPORT_SYMBOL(cpu_down);
296 #endif /*CONFIG_HOTPLUG_CPU*/
298 /* Requires cpu_add_remove_lock to be held */
299 static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
301 int ret, nr_calls = 0;
302 void *hcpu = (void *)(long)cpu;
303 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
305 if (cpu_online(cpu) || !cpu_present(cpu))
309 ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls);
312 printk("%s: attempt to bring up CPU %u failed\n",
317 /* Arch-specific enabling code. */
321 BUG_ON(!cpu_online(cpu));
323 set_cpu_active(cpu, true);
325 /* Now call notifier in preparation. */
326 cpu_notify(CPU_ONLINE | mod, hcpu);
330 __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
336 int __cpuinit cpu_up(unsigned int cpu)
340 #ifdef CONFIG_MEMORY_HOTPLUG
345 if (!cpu_possible(cpu)) {
346 printk(KERN_ERR "can't online cpu %d because it is not "
347 "configured as may-hotadd at boot time\n", cpu);
348 #if defined(CONFIG_IA64)
349 printk(KERN_ERR "please check additional_cpus= boot "
355 #ifdef CONFIG_MEMORY_HOTPLUG
356 nid = cpu_to_node(cpu);
357 if (!node_online(nid)) {
358 err = mem_online_node(nid);
363 pgdat = NODE_DATA(nid);
366 "Can't online cpu %d due to NULL pgdat\n", cpu);
370 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
371 mutex_lock(&zonelists_mutex);
372 build_all_zonelists(NULL);
373 mutex_unlock(&zonelists_mutex);
377 cpu_maps_update_begin();
379 if (cpu_hotplug_disabled) {
384 err = _cpu_up(cpu, 0);
387 cpu_maps_update_done();
391 #ifdef CONFIG_PM_SLEEP_SMP
392 static cpumask_var_t frozen_cpus;
394 int disable_nonboot_cpus(void)
396 int cpu, first_cpu, error;
398 cpu_maps_update_begin();
399 first_cpu = cpumask_first(cpu_online_mask);
401 * We take down all of the non-boot CPUs in one shot to avoid races
402 * with the userspace trying to use the CPU hotplug at the same time
404 cpumask_clear(frozen_cpus);
406 printk("Disabling non-boot CPUs ...\n");
407 for_each_online_cpu(cpu) {
408 if (cpu == first_cpu)
410 error = _cpu_down(cpu, 1);
412 cpumask_set_cpu(cpu, frozen_cpus);
414 printk(KERN_ERR "Error taking CPU%d down: %d\n",
421 BUG_ON(num_online_cpus() > 1);
422 /* Make sure the CPUs won't be enabled by someone else */
423 cpu_hotplug_disabled = 1;
425 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
427 cpu_maps_update_done();
431 void __weak arch_enable_nonboot_cpus_begin(void)
435 void __weak arch_enable_nonboot_cpus_end(void)
439 void __ref enable_nonboot_cpus(void)
443 /* Allow everyone to use the CPU hotplug again */
444 cpu_maps_update_begin();
445 cpu_hotplug_disabled = 0;
446 if (cpumask_empty(frozen_cpus))
449 printk("Enabling non-boot CPUs ...\n");
451 arch_enable_nonboot_cpus_begin();
453 for_each_cpu(cpu, frozen_cpus) {
454 error = _cpu_up(cpu, 1);
456 printk("CPU%d is up\n", cpu);
459 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
462 arch_enable_nonboot_cpus_end();
464 cpumask_clear(frozen_cpus);
466 cpu_maps_update_done();
469 static int alloc_frozen_cpus(void)
471 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
475 core_initcall(alloc_frozen_cpus);
476 #endif /* CONFIG_PM_SLEEP_SMP */
479 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
480 * @cpu: cpu that just started
482 * This function calls the cpu_chain notifiers with CPU_STARTING.
483 * It must be called by the arch code on the new cpu, before the new cpu
484 * enables interrupts and before the "boot" cpu returns from __cpu_up().
486 void __cpuinit notify_cpu_starting(unsigned int cpu)
488 unsigned long val = CPU_STARTING;
490 #ifdef CONFIG_PM_SLEEP_SMP
491 if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus))
492 val = CPU_STARTING_FROZEN;
493 #endif /* CONFIG_PM_SLEEP_SMP */
494 cpu_notify(val, (void *)(long)cpu);
497 #endif /* CONFIG_SMP */
500 * cpu_bit_bitmap[] is a special, "compressed" data structure that
501 * represents all NR_CPUS bits binary values of 1<<nr.
503 * It is used by cpumask_of() to get a constant address to a CPU
504 * mask value that has a single bit set only.
507 /* cpu_bit_bitmap[0] is empty - so we can back into it */
508 #define MASK_DECLARE_1(x) [x+1][0] = 1UL << (x)
509 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
510 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
511 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
513 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
515 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
516 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
517 #if BITS_PER_LONG > 32
518 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
519 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
522 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
524 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
525 EXPORT_SYMBOL(cpu_all_bits);
527 #ifdef CONFIG_INIT_ALL_POSSIBLE
528 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
531 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
533 const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
534 EXPORT_SYMBOL(cpu_possible_mask);
536 static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
537 const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
538 EXPORT_SYMBOL(cpu_online_mask);
540 static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
541 const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
542 EXPORT_SYMBOL(cpu_present_mask);
544 static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
545 const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
546 EXPORT_SYMBOL(cpu_active_mask);
548 void set_cpu_possible(unsigned int cpu, bool possible)
551 cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
553 cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
556 void set_cpu_present(unsigned int cpu, bool present)
559 cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
561 cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
564 void set_cpu_online(unsigned int cpu, bool online)
567 cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
569 cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
572 void set_cpu_active(unsigned int cpu, bool active)
575 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
577 cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
580 void init_cpu_present(const struct cpumask *src)
582 cpumask_copy(to_cpumask(cpu_present_bits), src);
585 void init_cpu_possible(const struct cpumask *src)
587 cpumask_copy(to_cpumask(cpu_possible_bits), src);
590 void init_cpu_online(const struct cpumask *src)
592 cpumask_copy(to_cpumask(cpu_online_bits), src);