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>
23 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
24 static DEFINE_MUTEX(cpu_add_remove_lock
);
27 * The following two API's must be used when attempting
28 * to serialize the updates to cpu_online_mask, cpu_present_mask.
30 void cpu_maps_update_begin(void)
32 mutex_lock(&cpu_add_remove_lock
);
35 void cpu_maps_update_done(void)
37 mutex_unlock(&cpu_add_remove_lock
);
40 static RAW_NOTIFIER_HEAD(cpu_chain
);
42 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
43 * Should always be manipulated under cpu_add_remove_lock
45 static int cpu_hotplug_disabled
;
47 #ifdef CONFIG_HOTPLUG_CPU
50 struct task_struct
*active_writer
;
51 struct mutex lock
; /* Synchronizes accesses to refcount, */
53 * Also blocks the new readers during
54 * an ongoing cpu hotplug operation.
58 .active_writer
= NULL
,
59 .lock
= __MUTEX_INITIALIZER(cpu_hotplug
.lock
),
63 void get_online_cpus(void)
66 if (cpu_hotplug
.active_writer
== current
)
68 mutex_lock(&cpu_hotplug
.lock
);
69 cpu_hotplug
.refcount
++;
70 mutex_unlock(&cpu_hotplug
.lock
);
73 EXPORT_SYMBOL_GPL(get_online_cpus
);
75 void put_online_cpus(void)
77 if (cpu_hotplug
.active_writer
== current
)
79 mutex_lock(&cpu_hotplug
.lock
);
80 if (!--cpu_hotplug
.refcount
&& unlikely(cpu_hotplug
.active_writer
))
81 wake_up_process(cpu_hotplug
.active_writer
);
82 mutex_unlock(&cpu_hotplug
.lock
);
85 EXPORT_SYMBOL_GPL(put_online_cpus
);
88 * This ensures that the hotplug operation can begin only when the
89 * refcount goes to zero.
91 * Note that during a cpu-hotplug operation, the new readers, if any,
92 * will be blocked by the cpu_hotplug.lock
94 * Since cpu_hotplug_begin() is always called after invoking
95 * cpu_maps_update_begin(), we can be sure that only one writer is active.
97 * Note that theoretically, there is a possibility of a livelock:
98 * - Refcount goes to zero, last reader wakes up the sleeping
100 * - Last reader unlocks the cpu_hotplug.lock.
101 * - A new reader arrives at this moment, bumps up the refcount.
102 * - The writer acquires the cpu_hotplug.lock finds the refcount
103 * non zero and goes to sleep again.
105 * However, this is very difficult to achieve in practice since
106 * get_online_cpus() not an api which is called all that often.
109 static void cpu_hotplug_begin(void)
111 cpu_hotplug
.active_writer
= current
;
114 mutex_lock(&cpu_hotplug
.lock
);
115 if (likely(!cpu_hotplug
.refcount
))
117 __set_current_state(TASK_UNINTERRUPTIBLE
);
118 mutex_unlock(&cpu_hotplug
.lock
);
123 static void cpu_hotplug_done(void)
125 cpu_hotplug
.active_writer
= NULL
;
126 mutex_unlock(&cpu_hotplug
.lock
);
129 #else /* #if CONFIG_HOTPLUG_CPU */
130 static void cpu_hotplug_begin(void) {}
131 static void cpu_hotplug_done(void) {}
132 #endif /* #else #if CONFIG_HOTPLUG_CPU */
134 /* Need to know about CPUs going up/down? */
135 int __ref
register_cpu_notifier(struct notifier_block
*nb
)
138 cpu_maps_update_begin();
139 ret
= raw_notifier_chain_register(&cpu_chain
, nb
);
140 cpu_maps_update_done();
144 static int __cpu_notify(unsigned long val
, void *v
, int nr_to_call
,
149 ret
= __raw_notifier_call_chain(&cpu_chain
, val
, v
, nr_to_call
,
152 return notifier_to_errno(ret
);
155 static int cpu_notify(unsigned long val
, void *v
)
157 return __cpu_notify(val
, v
, -1, NULL
);
160 #ifdef CONFIG_HOTPLUG_CPU
162 static void cpu_notify_nofail(unsigned long val
, void *v
)
164 BUG_ON(cpu_notify(val
, v
));
166 EXPORT_SYMBOL(register_cpu_notifier
);
168 void __ref
unregister_cpu_notifier(struct notifier_block
*nb
)
170 cpu_maps_update_begin();
171 raw_notifier_chain_unregister(&cpu_chain
, nb
);
172 cpu_maps_update_done();
174 EXPORT_SYMBOL(unregister_cpu_notifier
);
176 static inline void check_for_tasks(int cpu
)
178 struct task_struct
*p
;
180 write_lock_irq(&tasklist_lock
);
181 for_each_process(p
) {
182 if (task_cpu(p
) == cpu
&& p
->state
== TASK_RUNNING
&&
183 (p
->utime
|| p
->stime
))
184 printk(KERN_WARNING
"Task %s (pid = %d) is on cpu %d "
185 "(state = %ld, flags = %x)\n",
186 p
->comm
, task_pid_nr(p
), cpu
,
189 write_unlock_irq(&tasklist_lock
);
192 struct take_cpu_down_param
{
197 /* Take this CPU down. */
198 static int __ref
take_cpu_down(void *_param
)
200 struct take_cpu_down_param
*param
= _param
;
203 /* Ensure this CPU doesn't handle any more interrupts. */
204 err
= __cpu_disable();
208 cpu_notify(CPU_DYING
| param
->mod
, param
->hcpu
);
212 /* Requires cpu_add_remove_lock to be held */
213 static int __ref
_cpu_down(unsigned int cpu
, int tasks_frozen
)
215 int err
, nr_calls
= 0;
216 void *hcpu
= (void *)(long)cpu
;
217 unsigned long mod
= tasks_frozen
? CPU_TASKS_FROZEN
: 0;
218 struct take_cpu_down_param tcd_param
= {
223 if (num_online_cpus() == 1)
226 if (!cpu_online(cpu
))
231 err
= __cpu_notify(CPU_DOWN_PREPARE
| mod
, hcpu
, -1, &nr_calls
);
234 __cpu_notify(CPU_DOWN_FAILED
| mod
, hcpu
, nr_calls
, NULL
);
235 printk("%s: attempt to take down CPU %u failed\n",
240 err
= __stop_machine(take_cpu_down
, &tcd_param
, cpumask_of(cpu
));
242 /* CPU didn't die: tell everyone. Can't complain. */
243 cpu_notify_nofail(CPU_DOWN_FAILED
| mod
, hcpu
);
247 BUG_ON(cpu_online(cpu
));
250 * The migration_call() CPU_DYING callback will have removed all
251 * runnable tasks from the cpu, there's only the idle task left now
252 * that the migration thread is done doing the stop_machine thing.
254 * Wait for the stop thread to go away.
256 while (!idle_cpu(cpu
))
259 /* This actually kills the CPU. */
262 /* CPU is completely dead: tell everyone. Too late to complain. */
263 cpu_notify_nofail(CPU_DEAD
| mod
, hcpu
);
265 check_for_tasks(cpu
);
270 cpu_notify_nofail(CPU_POST_DEAD
| mod
, hcpu
);
274 int __ref
cpu_down(unsigned int cpu
)
278 cpu_maps_update_begin();
280 if (cpu_hotplug_disabled
) {
285 err
= _cpu_down(cpu
, 0);
288 cpu_maps_update_done();
291 EXPORT_SYMBOL(cpu_down
);
292 #endif /*CONFIG_HOTPLUG_CPU*/
294 /* Requires cpu_add_remove_lock to be held */
295 static int __cpuinit
_cpu_up(unsigned int cpu
, int tasks_frozen
)
297 int ret
, nr_calls
= 0;
298 void *hcpu
= (void *)(long)cpu
;
299 unsigned long mod
= tasks_frozen
? CPU_TASKS_FROZEN
: 0;
300 struct task_struct
*idle
;
302 if (cpu_online(cpu
) || !cpu_present(cpu
))
307 idle
= idle_thread_get(cpu
);
313 ret
= __cpu_notify(CPU_UP_PREPARE
| mod
, hcpu
, -1, &nr_calls
);
316 printk(KERN_WARNING
"%s: attempt to bring up CPU %u failed\n",
321 /* Arch-specific enabling code. */
322 ret
= __cpu_up(cpu
, idle
);
325 BUG_ON(!cpu_online(cpu
));
327 /* Now call notifier in preparation. */
328 cpu_notify(CPU_ONLINE
| mod
, hcpu
);
332 __cpu_notify(CPU_UP_CANCELED
| mod
, hcpu
, nr_calls
, NULL
);
339 int __cpuinit
cpu_up(unsigned int cpu
)
343 #ifdef CONFIG_MEMORY_HOTPLUG
348 if (!cpu_possible(cpu
)) {
349 printk(KERN_ERR
"can't online cpu %d because it is not "
350 "configured as may-hotadd at boot time\n", cpu
);
351 #if defined(CONFIG_IA64)
352 printk(KERN_ERR
"please check additional_cpus= boot "
358 #ifdef CONFIG_MEMORY_HOTPLUG
359 nid
= cpu_to_node(cpu
);
360 if (!node_online(nid
)) {
361 err
= mem_online_node(nid
);
366 pgdat
= NODE_DATA(nid
);
369 "Can't online cpu %d due to NULL pgdat\n", cpu
);
373 if (pgdat
->node_zonelists
->_zonerefs
->zone
== NULL
) {
374 mutex_lock(&zonelists_mutex
);
375 build_all_zonelists(NULL
);
376 mutex_unlock(&zonelists_mutex
);
380 cpu_maps_update_begin();
382 if (cpu_hotplug_disabled
) {
387 err
= _cpu_up(cpu
, 0);
390 cpu_maps_update_done();
393 EXPORT_SYMBOL_GPL(cpu_up
);
395 #ifdef CONFIG_PM_SLEEP_SMP
396 static cpumask_var_t frozen_cpus
;
398 void __weak
arch_disable_nonboot_cpus_begin(void)
402 void __weak
arch_disable_nonboot_cpus_end(void)
406 int disable_nonboot_cpus(void)
408 int cpu
, first_cpu
, error
= 0;
410 cpu_maps_update_begin();
411 first_cpu
= cpumask_first(cpu_online_mask
);
413 * We take down all of the non-boot CPUs in one shot to avoid races
414 * with the userspace trying to use the CPU hotplug at the same time
416 cpumask_clear(frozen_cpus
);
417 arch_disable_nonboot_cpus_begin();
419 printk("Disabling non-boot CPUs ...\n");
420 for_each_online_cpu(cpu
) {
421 if (cpu
== first_cpu
)
423 error
= _cpu_down(cpu
, 1);
425 cpumask_set_cpu(cpu
, frozen_cpus
);
427 printk(KERN_ERR
"Error taking CPU%d down: %d\n",
433 arch_disable_nonboot_cpus_end();
436 BUG_ON(num_online_cpus() > 1);
437 /* Make sure the CPUs won't be enabled by someone else */
438 cpu_hotplug_disabled
= 1;
440 printk(KERN_ERR
"Non-boot CPUs are not disabled\n");
442 cpu_maps_update_done();
446 void __weak
arch_enable_nonboot_cpus_begin(void)
450 void __weak
arch_enable_nonboot_cpus_end(void)
454 void __ref
enable_nonboot_cpus(void)
458 /* Allow everyone to use the CPU hotplug again */
459 cpu_maps_update_begin();
460 cpu_hotplug_disabled
= 0;
461 if (cpumask_empty(frozen_cpus
))
464 printk(KERN_INFO
"Enabling non-boot CPUs ...\n");
466 arch_enable_nonboot_cpus_begin();
468 for_each_cpu(cpu
, frozen_cpus
) {
469 error
= _cpu_up(cpu
, 1);
471 printk(KERN_INFO
"CPU%d is up\n", cpu
);
474 printk(KERN_WARNING
"Error taking CPU%d up: %d\n", cpu
, error
);
477 arch_enable_nonboot_cpus_end();
479 cpumask_clear(frozen_cpus
);
481 cpu_maps_update_done();
484 static int __init
alloc_frozen_cpus(void)
486 if (!alloc_cpumask_var(&frozen_cpus
, GFP_KERNEL
|__GFP_ZERO
))
490 core_initcall(alloc_frozen_cpus
);
493 * Prevent regular CPU hotplug from racing with the freezer, by disabling CPU
494 * hotplug when tasks are about to be frozen. Also, don't allow the freezer
495 * to continue until any currently running CPU hotplug operation gets
497 * To modify the 'cpu_hotplug_disabled' flag, we need to acquire the
498 * 'cpu_add_remove_lock'. And this same lock is also taken by the regular
499 * CPU hotplug path and released only after it is complete. Thus, we
500 * (and hence the freezer) will block here until any currently running CPU
501 * hotplug operation gets completed.
503 void cpu_hotplug_disable_before_freeze(void)
505 cpu_maps_update_begin();
506 cpu_hotplug_disabled
= 1;
507 cpu_maps_update_done();
512 * When tasks have been thawed, re-enable regular CPU hotplug (which had been
513 * disabled while beginning to freeze tasks).
515 void cpu_hotplug_enable_after_thaw(void)
517 cpu_maps_update_begin();
518 cpu_hotplug_disabled
= 0;
519 cpu_maps_update_done();
523 * When callbacks for CPU hotplug notifications are being executed, we must
524 * ensure that the state of the system with respect to the tasks being frozen
525 * or not, as reported by the notification, remains unchanged *throughout the
526 * duration* of the execution of the callbacks.
527 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
529 * This synchronization is implemented by mutually excluding regular CPU
530 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
531 * Hibernate notifications.
534 cpu_hotplug_pm_callback(struct notifier_block
*nb
,
535 unsigned long action
, void *ptr
)
539 case PM_SUSPEND_PREPARE
:
540 case PM_HIBERNATION_PREPARE
:
541 cpu_hotplug_disable_before_freeze();
544 case PM_POST_SUSPEND
:
545 case PM_POST_HIBERNATION
:
546 cpu_hotplug_enable_after_thaw();
557 static int __init
cpu_hotplug_pm_sync_init(void)
559 pm_notifier(cpu_hotplug_pm_callback
, 0);
562 core_initcall(cpu_hotplug_pm_sync_init
);
564 #endif /* CONFIG_PM_SLEEP_SMP */
567 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
568 * @cpu: cpu that just started
570 * This function calls the cpu_chain notifiers with CPU_STARTING.
571 * It must be called by the arch code on the new cpu, before the new cpu
572 * enables interrupts and before the "boot" cpu returns from __cpu_up().
574 void __cpuinit
notify_cpu_starting(unsigned int cpu
)
576 unsigned long val
= CPU_STARTING
;
578 #ifdef CONFIG_PM_SLEEP_SMP
579 if (frozen_cpus
!= NULL
&& cpumask_test_cpu(cpu
, frozen_cpus
))
580 val
= CPU_STARTING_FROZEN
;
581 #endif /* CONFIG_PM_SLEEP_SMP */
582 cpu_notify(val
, (void *)(long)cpu
);
585 #endif /* CONFIG_SMP */
588 * cpu_bit_bitmap[] is a special, "compressed" data structure that
589 * represents all NR_CPUS bits binary values of 1<<nr.
591 * It is used by cpumask_of() to get a constant address to a CPU
592 * mask value that has a single bit set only.
595 /* cpu_bit_bitmap[0] is empty - so we can back into it */
596 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
597 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
598 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
599 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
601 const unsigned long cpu_bit_bitmap
[BITS_PER_LONG
+1][BITS_TO_LONGS(NR_CPUS
)] = {
603 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
604 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
605 #if BITS_PER_LONG > 32
606 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
607 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
610 EXPORT_SYMBOL_GPL(cpu_bit_bitmap
);
612 const DECLARE_BITMAP(cpu_all_bits
, NR_CPUS
) = CPU_BITS_ALL
;
613 EXPORT_SYMBOL(cpu_all_bits
);
615 #ifdef CONFIG_INIT_ALL_POSSIBLE
616 static DECLARE_BITMAP(cpu_possible_bits
, CONFIG_NR_CPUS
) __read_mostly
619 static DECLARE_BITMAP(cpu_possible_bits
, CONFIG_NR_CPUS
) __read_mostly
;
621 const struct cpumask
*const cpu_possible_mask
= to_cpumask(cpu_possible_bits
);
622 EXPORT_SYMBOL(cpu_possible_mask
);
624 static DECLARE_BITMAP(cpu_online_bits
, CONFIG_NR_CPUS
) __read_mostly
;
625 const struct cpumask
*const cpu_online_mask
= to_cpumask(cpu_online_bits
);
626 EXPORT_SYMBOL(cpu_online_mask
);
628 static DECLARE_BITMAP(cpu_present_bits
, CONFIG_NR_CPUS
) __read_mostly
;
629 const struct cpumask
*const cpu_present_mask
= to_cpumask(cpu_present_bits
);
630 EXPORT_SYMBOL(cpu_present_mask
);
632 static DECLARE_BITMAP(cpu_active_bits
, CONFIG_NR_CPUS
) __read_mostly
;
633 const struct cpumask
*const cpu_active_mask
= to_cpumask(cpu_active_bits
);
634 EXPORT_SYMBOL(cpu_active_mask
);
636 void set_cpu_possible(unsigned int cpu
, bool possible
)
639 cpumask_set_cpu(cpu
, to_cpumask(cpu_possible_bits
));
641 cpumask_clear_cpu(cpu
, to_cpumask(cpu_possible_bits
));
644 void set_cpu_present(unsigned int cpu
, bool present
)
647 cpumask_set_cpu(cpu
, to_cpumask(cpu_present_bits
));
649 cpumask_clear_cpu(cpu
, to_cpumask(cpu_present_bits
));
652 void set_cpu_online(unsigned int cpu
, bool online
)
655 cpumask_set_cpu(cpu
, to_cpumask(cpu_online_bits
));
657 cpumask_clear_cpu(cpu
, to_cpumask(cpu_online_bits
));
660 void set_cpu_active(unsigned int cpu
, bool active
)
663 cpumask_set_cpu(cpu
, to_cpumask(cpu_active_bits
));
665 cpumask_clear_cpu(cpu
, to_cpumask(cpu_active_bits
));
668 void init_cpu_present(const struct cpumask
*src
)
670 cpumask_copy(to_cpumask(cpu_present_bits
), src
);
673 void init_cpu_possible(const struct cpumask
*src
)
675 cpumask_copy(to_cpumask(cpu_possible_bits
), src
);
678 void init_cpu_online(const struct cpumask
*src
)
680 cpumask_copy(to_cpumask(cpu_online_bits
), src
);