sched: rework of "prioritize non-migratable tasks over migratable ones"
[linux-2.6-block.git] / kernel / cpu.c
CommitLineData
1da177e4
LT
1/* CPU control.
2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
3 *
4 * This code is licenced under the GPL.
5 */
6#include <linux/proc_fs.h>
7#include <linux/smp.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>
81615b62 16#include <linux/mutex.h>
1da177e4 17
68f4f1ec
MK
18/*
19 * Represents all cpu's present in the system
20 * In systems capable of hotplug, this map could dynamically grow
21 * as new cpu's are detected in the system via any platform specific
22 * method, such as ACPI for e.g.
23 */
24cpumask_t cpu_present_map __read_mostly;
25EXPORT_SYMBOL(cpu_present_map);
26
27#ifndef CONFIG_SMP
28
29/*
30 * Represents all cpu's that are currently online.
31 */
32cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL;
33EXPORT_SYMBOL(cpu_online_map);
34
35cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL;
36EXPORT_SYMBOL(cpu_possible_map);
37
38#else /* CONFIG_SMP */
39
d221938c 40/* Serializes the updates to cpu_online_map, cpu_present_map */
aa953877 41static DEFINE_MUTEX(cpu_add_remove_lock);
1da177e4 42
bd5349cf 43static __cpuinitdata RAW_NOTIFIER_HEAD(cpu_chain);
1da177e4 44
e3920fb4
RW
45/* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
46 * Should always be manipulated under cpu_add_remove_lock
47 */
48static int cpu_hotplug_disabled;
49
d221938c
GS
50static struct {
51 struct task_struct *active_writer;
52 struct mutex lock; /* Synchronizes accesses to refcount, */
53 /*
54 * Also blocks the new readers during
55 * an ongoing cpu hotplug operation.
56 */
57 int refcount;
d221938c 58} cpu_hotplug;
90d45d17 59
d221938c
GS
60void __init cpu_hotplug_init(void)
61{
62 cpu_hotplug.active_writer = NULL;
63 mutex_init(&cpu_hotplug.lock);
64 cpu_hotplug.refcount = 0;
d221938c
GS
65}
66
67#ifdef CONFIG_HOTPLUG_CPU
90d45d17 68
86ef5c9a 69void get_online_cpus(void)
a9d9baa1 70{
d221938c
GS
71 might_sleep();
72 if (cpu_hotplug.active_writer == current)
aa953877 73 return;
d221938c
GS
74 mutex_lock(&cpu_hotplug.lock);
75 cpu_hotplug.refcount++;
76 mutex_unlock(&cpu_hotplug.lock);
77
a9d9baa1 78}
86ef5c9a 79EXPORT_SYMBOL_GPL(get_online_cpus);
90d45d17 80
86ef5c9a 81void put_online_cpus(void)
a9d9baa1 82{
d221938c 83 if (cpu_hotplug.active_writer == current)
aa953877 84 return;
d221938c 85 mutex_lock(&cpu_hotplug.lock);
d2ba7e2a
ON
86 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
87 wake_up_process(cpu_hotplug.active_writer);
d221938c
GS
88 mutex_unlock(&cpu_hotplug.lock);
89
a9d9baa1 90}
86ef5c9a 91EXPORT_SYMBOL_GPL(put_online_cpus);
a9d9baa1 92
a9d9baa1 93#endif /* CONFIG_HOTPLUG_CPU */
90d45d17 94
d221938c
GS
95/*
96 * The following two API's must be used when attempting
97 * to serialize the updates to cpu_online_map, cpu_present_map.
98 */
99void cpu_maps_update_begin(void)
100{
101 mutex_lock(&cpu_add_remove_lock);
102}
103
104void cpu_maps_update_done(void)
105{
106 mutex_unlock(&cpu_add_remove_lock);
107}
108
109/*
110 * This ensures that the hotplug operation can begin only when the
111 * refcount goes to zero.
112 *
113 * Note that during a cpu-hotplug operation, the new readers, if any,
114 * will be blocked by the cpu_hotplug.lock
115 *
d2ba7e2a
ON
116 * Since cpu_hotplug_begin() is always called after invoking
117 * cpu_maps_update_begin(), we can be sure that only one writer is active.
d221938c
GS
118 *
119 * Note that theoretically, there is a possibility of a livelock:
120 * - Refcount goes to zero, last reader wakes up the sleeping
121 * writer.
122 * - Last reader unlocks the cpu_hotplug.lock.
123 * - A new reader arrives at this moment, bumps up the refcount.
124 * - The writer acquires the cpu_hotplug.lock finds the refcount
125 * non zero and goes to sleep again.
126 *
127 * However, this is very difficult to achieve in practice since
86ef5c9a 128 * get_online_cpus() not an api which is called all that often.
d221938c
GS
129 *
130 */
131static void cpu_hotplug_begin(void)
132{
d221938c 133 cpu_hotplug.active_writer = current;
d2ba7e2a
ON
134
135 for (;;) {
136 mutex_lock(&cpu_hotplug.lock);
137 if (likely(!cpu_hotplug.refcount))
138 break;
139 __set_current_state(TASK_UNINTERRUPTIBLE);
d221938c
GS
140 mutex_unlock(&cpu_hotplug.lock);
141 schedule();
d221938c 142 }
d221938c
GS
143}
144
145static void cpu_hotplug_done(void)
146{
147 cpu_hotplug.active_writer = NULL;
148 mutex_unlock(&cpu_hotplug.lock);
149}
1da177e4 150/* Need to know about CPUs going up/down? */
f7b16c10 151int __ref register_cpu_notifier(struct notifier_block *nb)
1da177e4 152{
bd5349cf 153 int ret;
d221938c 154 cpu_maps_update_begin();
bd5349cf 155 ret = raw_notifier_chain_register(&cpu_chain, nb);
d221938c 156 cpu_maps_update_done();
bd5349cf 157 return ret;
1da177e4 158}
65edc68c
CS
159
160#ifdef CONFIG_HOTPLUG_CPU
161
1da177e4
LT
162EXPORT_SYMBOL(register_cpu_notifier);
163
9647155f 164void __ref unregister_cpu_notifier(struct notifier_block *nb)
1da177e4 165{
d221938c 166 cpu_maps_update_begin();
bd5349cf 167 raw_notifier_chain_unregister(&cpu_chain, nb);
d221938c 168 cpu_maps_update_done();
1da177e4
LT
169}
170EXPORT_SYMBOL(unregister_cpu_notifier);
171
1da177e4
LT
172static inline void check_for_tasks(int cpu)
173{
174 struct task_struct *p;
175
176 write_lock_irq(&tasklist_lock);
177 for_each_process(p) {
178 if (task_cpu(p) == cpu &&
179 (!cputime_eq(p->utime, cputime_zero) ||
180 !cputime_eq(p->stime, cputime_zero)))
181 printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d\
e7407dcc 182 (state = %ld, flags = %x) \n",
ba25f9dc
PE
183 p->comm, task_pid_nr(p), cpu,
184 p->state, p->flags);
1da177e4
LT
185 }
186 write_unlock_irq(&tasklist_lock);
187}
188
db912f96
AK
189struct take_cpu_down_param {
190 unsigned long mod;
191 void *hcpu;
192};
193
1da177e4 194/* Take this CPU down. */
514a20a5 195static int __ref take_cpu_down(void *_param)
1da177e4 196{
db912f96 197 struct take_cpu_down_param *param = _param;
1da177e4
LT
198 int err;
199
db912f96
AK
200 raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
201 param->hcpu);
1da177e4
LT
202 /* Ensure this CPU doesn't handle any more interrupts. */
203 err = __cpu_disable();
204 if (err < 0)
f3705136 205 return err;
1da177e4 206
f3705136
ZM
207 /* Force idle task to run as soon as we yield: it should
208 immediately notice cpu is offline and die quickly. */
209 sched_idle_next();
210 return 0;
1da177e4
LT
211}
212
e3920fb4 213/* Requires cpu_add_remove_lock to be held */
514a20a5 214static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
1da177e4 215{
e7407dcc 216 int err, nr_calls = 0;
1da177e4
LT
217 struct task_struct *p;
218 cpumask_t old_allowed, tmp;
e7407dcc 219 void *hcpu = (void *)(long)cpu;
8bb78442 220 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
db912f96
AK
221 struct take_cpu_down_param tcd_param = {
222 .mod = mod,
223 .hcpu = hcpu,
224 };
1da177e4 225
e3920fb4
RW
226 if (num_online_cpus() == 1)
227 return -EBUSY;
1da177e4 228
e3920fb4
RW
229 if (!cpu_online(cpu))
230 return -EINVAL;
1da177e4 231
d221938c 232 cpu_hotplug_begin();
8bb78442 233 err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
e7407dcc 234 hcpu, -1, &nr_calls);
1da177e4 235 if (err == NOTIFY_BAD) {
a0d8cdb6 236 nr_calls--;
8bb78442
RW
237 __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
238 hcpu, nr_calls, NULL);
1da177e4 239 printk("%s: attempt to take down CPU %u failed\n",
af1f16d0 240 __func__, cpu);
baaca49f
GS
241 err = -EINVAL;
242 goto out_release;
1da177e4
LT
243 }
244
245 /* Ensure that we are not runnable on dying cpu */
246 old_allowed = current->cpus_allowed;
f70316da 247 cpus_setall(tmp);
1da177e4 248 cpu_clear(cpu, tmp);
f70316da 249 set_cpus_allowed_ptr(current, &tmp);
1da177e4 250
db912f96 251 p = __stop_machine_run(take_cpu_down, &tcd_param, cpu);
aa953877 252
8fa1d7d3 253 if (IS_ERR(p) || cpu_online(cpu)) {
1da177e4 254 /* CPU didn't die: tell everyone. Can't complain. */
8bb78442 255 if (raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
e7407dcc 256 hcpu) == NOTIFY_BAD)
1da177e4
LT
257 BUG();
258
8fa1d7d3
ST
259 if (IS_ERR(p)) {
260 err = PTR_ERR(p);
261 goto out_allowed;
262 }
1da177e4 263 goto out_thread;
8fa1d7d3 264 }
1da177e4
LT
265
266 /* Wait for it to sleep (leaving idle task). */
267 while (!idle_cpu(cpu))
268 yield();
269
270 /* This actually kills the CPU. */
271 __cpu_die(cpu);
272
1da177e4 273 /* CPU is completely dead: tell everyone. Too late to complain. */
8bb78442
RW
274 if (raw_notifier_call_chain(&cpu_chain, CPU_DEAD | mod,
275 hcpu) == NOTIFY_BAD)
1da177e4
LT
276 BUG();
277
278 check_for_tasks(cpu);
279
280out_thread:
281 err = kthread_stop(p);
282out_allowed:
f70316da 283 set_cpus_allowed_ptr(current, &old_allowed);
baaca49f 284out_release:
d221938c 285 cpu_hotplug_done();
e3920fb4
RW
286 return err;
287}
288
514a20a5 289int __ref cpu_down(unsigned int cpu)
e3920fb4
RW
290{
291 int err = 0;
292
d221938c 293 cpu_maps_update_begin();
e3920fb4
RW
294 if (cpu_hotplug_disabled)
295 err = -EBUSY;
296 else
8bb78442 297 err = _cpu_down(cpu, 0);
e3920fb4 298
d221938c 299 cpu_maps_update_done();
1da177e4
LT
300 return err;
301}
b62b8ef9 302EXPORT_SYMBOL(cpu_down);
1da177e4
LT
303#endif /*CONFIG_HOTPLUG_CPU*/
304
e3920fb4 305/* Requires cpu_add_remove_lock to be held */
8bb78442 306static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
1da177e4 307{
baaca49f 308 int ret, nr_calls = 0;
1da177e4 309 void *hcpu = (void *)(long)cpu;
8bb78442 310 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
1da177e4 311
e3920fb4
RW
312 if (cpu_online(cpu) || !cpu_present(cpu))
313 return -EINVAL;
90d45d17 314
d221938c 315 cpu_hotplug_begin();
8bb78442 316 ret = __raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE | mod, hcpu,
baaca49f 317 -1, &nr_calls);
1da177e4 318 if (ret == NOTIFY_BAD) {
a0d8cdb6 319 nr_calls--;
1da177e4 320 printk("%s: attempt to bring up CPU %u failed\n",
af1f16d0 321 __func__, cpu);
1da177e4
LT
322 ret = -EINVAL;
323 goto out_notify;
324 }
325
326 /* Arch-specific enabling code. */
327 ret = __cpu_up(cpu);
328 if (ret != 0)
329 goto out_notify;
6978c705 330 BUG_ON(!cpu_online(cpu));
1da177e4
LT
331
332 /* Now call notifier in preparation. */
8bb78442 333 raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu);
1da177e4
LT
334
335out_notify:
336 if (ret != 0)
baaca49f 337 __raw_notifier_call_chain(&cpu_chain,
8bb78442 338 CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
d221938c 339 cpu_hotplug_done();
e3920fb4
RW
340
341 return ret;
342}
343
b282b6f8 344int __cpuinit cpu_up(unsigned int cpu)
e3920fb4
RW
345{
346 int err = 0;
73e753a5
KH
347 if (!cpu_isset(cpu, cpu_possible_map)) {
348 printk(KERN_ERR "can't online cpu %d because it is not "
349 "configured as may-hotadd at boot time\n", cpu);
350#if defined(CONFIG_IA64) || defined(CONFIG_X86_64) || defined(CONFIG_S390)
351 printk(KERN_ERR "please check additional_cpus= boot "
352 "parameter\n");
353#endif
354 return -EINVAL;
355 }
e3920fb4 356
d221938c 357 cpu_maps_update_begin();
e3920fb4
RW
358 if (cpu_hotplug_disabled)
359 err = -EBUSY;
360 else
8bb78442 361 err = _cpu_up(cpu, 0);
e3920fb4 362
d221938c 363 cpu_maps_update_done();
e3920fb4
RW
364 return err;
365}
366
f3de4be9 367#ifdef CONFIG_PM_SLEEP_SMP
e3920fb4
RW
368static cpumask_t frozen_cpus;
369
370int disable_nonboot_cpus(void)
371{
e1d9fd2e 372 int cpu, first_cpu, error = 0;
e3920fb4 373
d221938c 374 cpu_maps_update_begin();
1d64b9cb 375 first_cpu = first_cpu(cpu_online_map);
e3920fb4
RW
376 /* We take down all of the non-boot CPUs in one shot to avoid races
377 * with the userspace trying to use the CPU hotplug at the same time
378 */
379 cpus_clear(frozen_cpus);
380 printk("Disabling non-boot CPUs ...\n");
381 for_each_online_cpu(cpu) {
382 if (cpu == first_cpu)
383 continue;
8bb78442 384 error = _cpu_down(cpu, 1);
e3920fb4
RW
385 if (!error) {
386 cpu_set(cpu, frozen_cpus);
387 printk("CPU%d is down\n", cpu);
388 } else {
389 printk(KERN_ERR "Error taking CPU%d down: %d\n",
390 cpu, error);
391 break;
392 }
393 }
394 if (!error) {
395 BUG_ON(num_online_cpus() > 1);
396 /* Make sure the CPUs won't be enabled by someone else */
397 cpu_hotplug_disabled = 1;
398 } else {
e1d9fd2e 399 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
e3920fb4 400 }
d221938c 401 cpu_maps_update_done();
e3920fb4
RW
402 return error;
403}
404
fa7303e2 405void __ref enable_nonboot_cpus(void)
e3920fb4
RW
406{
407 int cpu, error;
408
409 /* Allow everyone to use the CPU hotplug again */
d221938c 410 cpu_maps_update_begin();
e3920fb4 411 cpu_hotplug_disabled = 0;
ed746e3b 412 if (cpus_empty(frozen_cpus))
1d64b9cb 413 goto out;
e3920fb4
RW
414
415 printk("Enabling non-boot CPUs ...\n");
416 for_each_cpu_mask(cpu, frozen_cpus) {
8bb78442 417 error = _cpu_up(cpu, 1);
e3920fb4
RW
418 if (!error) {
419 printk("CPU%d is up\n", cpu);
420 continue;
421 }
1d64b9cb 422 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
e3920fb4
RW
423 }
424 cpus_clear(frozen_cpus);
1d64b9cb 425out:
d221938c 426 cpu_maps_update_done();
1da177e4 427}
f3de4be9 428#endif /* CONFIG_PM_SLEEP_SMP */
68f4f1ec
MK
429
430#endif /* CONFIG_SMP */