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 #ifdef CONFIG_HOTPLUG_CPU
139 EXPORT_SYMBOL(register_cpu_notifier);
141 void __ref unregister_cpu_notifier(struct notifier_block *nb)
143 cpu_maps_update_begin();
144 raw_notifier_chain_unregister(&cpu_chain, nb);
145 cpu_maps_update_done();
147 EXPORT_SYMBOL(unregister_cpu_notifier);
149 static inline void check_for_tasks(int cpu)
151 struct task_struct *p;
153 write_lock_irq(&tasklist_lock);
154 for_each_process(p) {
155 if (task_cpu(p) == cpu && p->state == TASK_RUNNING &&
156 (!cputime_eq(p->utime, cputime_zero) ||
157 !cputime_eq(p->stime, cputime_zero)))
158 printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d "
159 "(state = %ld, flags = %x)\n",
160 p->comm, task_pid_nr(p), cpu,
163 write_unlock_irq(&tasklist_lock);
166 struct take_cpu_down_param {
167 struct task_struct *caller;
172 /* Take this CPU down. */
173 static int __ref take_cpu_down(void *_param)
175 struct take_cpu_down_param *param = _param;
176 unsigned int cpu = (unsigned long)param->hcpu;
179 /* Ensure this CPU doesn't handle any more interrupts. */
180 err = __cpu_disable();
184 raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
187 if (task_cpu(param->caller) == cpu)
188 move_task_off_dead_cpu(cpu, param->caller);
189 /* Force idle task to run as soon as we yield: it should
190 immediately notice cpu is offline and die quickly. */
195 /* Requires cpu_add_remove_lock to be held */
196 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
198 int err, nr_calls = 0;
199 void *hcpu = (void *)(long)cpu;
200 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
201 struct take_cpu_down_param tcd_param = {
207 if (num_online_cpus() == 1)
210 if (!cpu_online(cpu))
214 set_cpu_active(cpu, false);
215 err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
216 hcpu, -1, &nr_calls);
217 if (err == NOTIFY_BAD) {
218 set_cpu_active(cpu, true);
221 __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
222 hcpu, nr_calls, NULL);
223 printk("%s: attempt to take down CPU %u failed\n",
229 err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
231 set_cpu_active(cpu, true);
232 /* CPU didn't die: tell everyone. Can't complain. */
233 if (raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
239 BUG_ON(cpu_online(cpu));
241 /* Wait for it to sleep (leaving idle task). */
242 while (!idle_cpu(cpu))
245 /* This actually kills the CPU. */
248 /* CPU is completely dead: tell everyone. Too late to complain. */
249 if (raw_notifier_call_chain(&cpu_chain, CPU_DEAD | mod,
253 check_for_tasks(cpu);
258 if (raw_notifier_call_chain(&cpu_chain, CPU_POST_DEAD | mod,
265 int __ref cpu_down(unsigned int cpu)
269 cpu_maps_update_begin();
271 if (cpu_hotplug_disabled) {
276 err = _cpu_down(cpu, 0);
279 cpu_maps_update_done();
282 EXPORT_SYMBOL(cpu_down);
283 #endif /*CONFIG_HOTPLUG_CPU*/
285 /* Requires cpu_add_remove_lock to be held */
286 static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
288 int ret, nr_calls = 0;
289 void *hcpu = (void *)(long)cpu;
290 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
292 if (cpu_online(cpu) || !cpu_present(cpu))
296 ret = __raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE | mod, hcpu,
298 if (ret == NOTIFY_BAD) {
300 printk("%s: attempt to bring up CPU %u failed\n",
306 /* Arch-specific enabling code. */
310 BUG_ON(!cpu_online(cpu));
312 set_cpu_active(cpu, true);
314 /* Now call notifier in preparation. */
315 raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu);
319 __raw_notifier_call_chain(&cpu_chain,
320 CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
326 int __cpuinit cpu_up(unsigned int cpu)
330 #ifdef CONFIG_MEMORY_HOTPLUG
335 if (!cpu_possible(cpu)) {
336 printk(KERN_ERR "can't online cpu %d because it is not "
337 "configured as may-hotadd at boot time\n", cpu);
338 #if defined(CONFIG_IA64)
339 printk(KERN_ERR "please check additional_cpus= boot "
345 #ifdef CONFIG_MEMORY_HOTPLUG
346 nid = cpu_to_node(cpu);
347 if (!node_online(nid)) {
348 err = mem_online_node(nid);
353 pgdat = NODE_DATA(nid);
356 "Can't online cpu %d due to NULL pgdat\n", cpu);
360 if (pgdat->node_zonelists->_zonerefs->zone == NULL)
361 build_all_zonelists(NULL);
364 cpu_maps_update_begin();
366 if (cpu_hotplug_disabled) {
371 err = _cpu_up(cpu, 0);
374 cpu_maps_update_done();
378 #ifdef CONFIG_PM_SLEEP_SMP
379 static cpumask_var_t frozen_cpus;
381 int disable_nonboot_cpus(void)
383 int cpu, first_cpu, error;
385 cpu_maps_update_begin();
386 first_cpu = cpumask_first(cpu_online_mask);
388 * We take down all of the non-boot CPUs in one shot to avoid races
389 * with the userspace trying to use the CPU hotplug at the same time
391 cpumask_clear(frozen_cpus);
393 printk("Disabling non-boot CPUs ...\n");
394 for_each_online_cpu(cpu) {
395 if (cpu == first_cpu)
397 error = _cpu_down(cpu, 1);
399 cpumask_set_cpu(cpu, frozen_cpus);
401 printk(KERN_ERR "Error taking CPU%d down: %d\n",
408 BUG_ON(num_online_cpus() > 1);
409 /* Make sure the CPUs won't be enabled by someone else */
410 cpu_hotplug_disabled = 1;
412 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
414 cpu_maps_update_done();
418 void __weak arch_enable_nonboot_cpus_begin(void)
422 void __weak arch_enable_nonboot_cpus_end(void)
426 void __ref enable_nonboot_cpus(void)
430 /* Allow everyone to use the CPU hotplug again */
431 cpu_maps_update_begin();
432 cpu_hotplug_disabled = 0;
433 if (cpumask_empty(frozen_cpus))
436 printk("Enabling non-boot CPUs ...\n");
438 arch_enable_nonboot_cpus_begin();
440 for_each_cpu(cpu, frozen_cpus) {
441 error = _cpu_up(cpu, 1);
443 printk("CPU%d is up\n", cpu);
446 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
449 arch_enable_nonboot_cpus_end();
451 cpumask_clear(frozen_cpus);
453 cpu_maps_update_done();
456 static int alloc_frozen_cpus(void)
458 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
462 core_initcall(alloc_frozen_cpus);
463 #endif /* CONFIG_PM_SLEEP_SMP */
466 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
467 * @cpu: cpu that just started
469 * This function calls the cpu_chain notifiers with CPU_STARTING.
470 * It must be called by the arch code on the new cpu, before the new cpu
471 * enables interrupts and before the "boot" cpu returns from __cpu_up().
473 void __cpuinit notify_cpu_starting(unsigned int cpu)
475 unsigned long val = CPU_STARTING;
477 #ifdef CONFIG_PM_SLEEP_SMP
478 if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus))
479 val = CPU_STARTING_FROZEN;
480 #endif /* CONFIG_PM_SLEEP_SMP */
481 raw_notifier_call_chain(&cpu_chain, val, (void *)(long)cpu);
484 #endif /* CONFIG_SMP */
487 * cpu_bit_bitmap[] is a special, "compressed" data structure that
488 * represents all NR_CPUS bits binary values of 1<<nr.
490 * It is used by cpumask_of() to get a constant address to a CPU
491 * mask value that has a single bit set only.
494 /* cpu_bit_bitmap[0] is empty - so we can back into it */
495 #define MASK_DECLARE_1(x) [x+1][0] = 1UL << (x)
496 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
497 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
498 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
500 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
502 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
503 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
504 #if BITS_PER_LONG > 32
505 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
506 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
509 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
511 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
512 EXPORT_SYMBOL(cpu_all_bits);
514 #ifdef CONFIG_INIT_ALL_POSSIBLE
515 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
518 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
520 const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
521 EXPORT_SYMBOL(cpu_possible_mask);
523 static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
524 const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
525 EXPORT_SYMBOL(cpu_online_mask);
527 static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
528 const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
529 EXPORT_SYMBOL(cpu_present_mask);
531 static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
532 const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
533 EXPORT_SYMBOL(cpu_active_mask);
535 void set_cpu_possible(unsigned int cpu, bool possible)
538 cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
540 cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
543 void set_cpu_present(unsigned int cpu, bool present)
546 cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
548 cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
551 void set_cpu_online(unsigned int cpu, bool online)
554 cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
556 cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
559 void set_cpu_active(unsigned int cpu, bool active)
562 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
564 cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
567 void init_cpu_present(const struct cpumask *src)
569 cpumask_copy(to_cpumask(cpu_present_bits), src);
572 void init_cpu_possible(const struct cpumask *src)
574 cpumask_copy(to_cpumask(cpu_possible_bits), src);
577 void init_cpu_online(const struct cpumask *src)
579 cpumask_copy(to_cpumask(cpu_online_bits), src);