cpu hotplug, sched: Introduce cpu_active_map and redo sched domain managment (take 2)
[deliverable/linux.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
e761b772
MK
67cpumask_t cpu_active_map;
68
d221938c 69#ifdef CONFIG_HOTPLUG_CPU
90d45d17 70
86ef5c9a 71void get_online_cpus(void)
a9d9baa1 72{
d221938c
GS
73 might_sleep();
74 if (cpu_hotplug.active_writer == current)
aa953877 75 return;
d221938c
GS
76 mutex_lock(&cpu_hotplug.lock);
77 cpu_hotplug.refcount++;
78 mutex_unlock(&cpu_hotplug.lock);
79
a9d9baa1 80}
86ef5c9a 81EXPORT_SYMBOL_GPL(get_online_cpus);
90d45d17 82
86ef5c9a 83void put_online_cpus(void)
a9d9baa1 84{
d221938c 85 if (cpu_hotplug.active_writer == current)
aa953877 86 return;
d221938c 87 mutex_lock(&cpu_hotplug.lock);
d2ba7e2a
ON
88 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
89 wake_up_process(cpu_hotplug.active_writer);
d221938c
GS
90 mutex_unlock(&cpu_hotplug.lock);
91
a9d9baa1 92}
86ef5c9a 93EXPORT_SYMBOL_GPL(put_online_cpus);
a9d9baa1 94
a9d9baa1 95#endif /* CONFIG_HOTPLUG_CPU */
90d45d17 96
d221938c
GS
97/*
98 * The following two API's must be used when attempting
99 * to serialize the updates to cpu_online_map, cpu_present_map.
100 */
101void cpu_maps_update_begin(void)
102{
103 mutex_lock(&cpu_add_remove_lock);
104}
105
106void cpu_maps_update_done(void)
107{
108 mutex_unlock(&cpu_add_remove_lock);
109}
110
111/*
112 * This ensures that the hotplug operation can begin only when the
113 * refcount goes to zero.
114 *
115 * Note that during a cpu-hotplug operation, the new readers, if any,
116 * will be blocked by the cpu_hotplug.lock
117 *
d2ba7e2a
ON
118 * Since cpu_hotplug_begin() is always called after invoking
119 * cpu_maps_update_begin(), we can be sure that only one writer is active.
d221938c
GS
120 *
121 * Note that theoretically, there is a possibility of a livelock:
122 * - Refcount goes to zero, last reader wakes up the sleeping
123 * writer.
124 * - Last reader unlocks the cpu_hotplug.lock.
125 * - A new reader arrives at this moment, bumps up the refcount.
126 * - The writer acquires the cpu_hotplug.lock finds the refcount
127 * non zero and goes to sleep again.
128 *
129 * However, this is very difficult to achieve in practice since
86ef5c9a 130 * get_online_cpus() not an api which is called all that often.
d221938c
GS
131 *
132 */
133static void cpu_hotplug_begin(void)
134{
d221938c 135 cpu_hotplug.active_writer = current;
d2ba7e2a
ON
136
137 for (;;) {
138 mutex_lock(&cpu_hotplug.lock);
139 if (likely(!cpu_hotplug.refcount))
140 break;
141 __set_current_state(TASK_UNINTERRUPTIBLE);
d221938c
GS
142 mutex_unlock(&cpu_hotplug.lock);
143 schedule();
d221938c 144 }
d221938c
GS
145}
146
147static void cpu_hotplug_done(void)
148{
149 cpu_hotplug.active_writer = NULL;
150 mutex_unlock(&cpu_hotplug.lock);
151}
1da177e4 152/* Need to know about CPUs going up/down? */
f7b16c10 153int __ref register_cpu_notifier(struct notifier_block *nb)
1da177e4 154{
bd5349cf 155 int ret;
d221938c 156 cpu_maps_update_begin();
bd5349cf 157 ret = raw_notifier_chain_register(&cpu_chain, nb);
d221938c 158 cpu_maps_update_done();
bd5349cf 159 return ret;
1da177e4 160}
65edc68c
CS
161
162#ifdef CONFIG_HOTPLUG_CPU
163
1da177e4
LT
164EXPORT_SYMBOL(register_cpu_notifier);
165
9647155f 166void __ref unregister_cpu_notifier(struct notifier_block *nb)
1da177e4 167{
d221938c 168 cpu_maps_update_begin();
bd5349cf 169 raw_notifier_chain_unregister(&cpu_chain, nb);
d221938c 170 cpu_maps_update_done();
1da177e4
LT
171}
172EXPORT_SYMBOL(unregister_cpu_notifier);
173
1da177e4
LT
174static inline void check_for_tasks(int cpu)
175{
176 struct task_struct *p;
177
178 write_lock_irq(&tasklist_lock);
179 for_each_process(p) {
180 if (task_cpu(p) == cpu &&
181 (!cputime_eq(p->utime, cputime_zero) ||
182 !cputime_eq(p->stime, cputime_zero)))
183 printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d\
e7407dcc 184 (state = %ld, flags = %x) \n",
ba25f9dc
PE
185 p->comm, task_pid_nr(p), cpu,
186 p->state, p->flags);
1da177e4
LT
187 }
188 write_unlock_irq(&tasklist_lock);
189}
190
db912f96
AK
191struct take_cpu_down_param {
192 unsigned long mod;
193 void *hcpu;
194};
195
1da177e4 196/* Take this CPU down. */
514a20a5 197static int __ref take_cpu_down(void *_param)
1da177e4 198{
db912f96 199 struct take_cpu_down_param *param = _param;
1da177e4
LT
200 int err;
201
db912f96
AK
202 raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
203 param->hcpu);
1da177e4
LT
204 /* Ensure this CPU doesn't handle any more interrupts. */
205 err = __cpu_disable();
206 if (err < 0)
f3705136 207 return err;
1da177e4 208
f3705136
ZM
209 /* Force idle task to run as soon as we yield: it should
210 immediately notice cpu is offline and die quickly. */
211 sched_idle_next();
212 return 0;
1da177e4
LT
213}
214
e3920fb4 215/* Requires cpu_add_remove_lock to be held */
514a20a5 216static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
1da177e4 217{
e7407dcc 218 int err, nr_calls = 0;
1da177e4
LT
219 struct task_struct *p;
220 cpumask_t old_allowed, tmp;
e7407dcc 221 void *hcpu = (void *)(long)cpu;
8bb78442 222 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
db912f96
AK
223 struct take_cpu_down_param tcd_param = {
224 .mod = mod,
225 .hcpu = hcpu,
226 };
1da177e4 227
e3920fb4
RW
228 if (num_online_cpus() == 1)
229 return -EBUSY;
1da177e4 230
e3920fb4
RW
231 if (!cpu_online(cpu))
232 return -EINVAL;
1da177e4 233
d221938c 234 cpu_hotplug_begin();
8bb78442 235 err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
e7407dcc 236 hcpu, -1, &nr_calls);
1da177e4 237 if (err == NOTIFY_BAD) {
a0d8cdb6 238 nr_calls--;
8bb78442
RW
239 __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
240 hcpu, nr_calls, NULL);
1da177e4 241 printk("%s: attempt to take down CPU %u failed\n",
af1f16d0 242 __func__, cpu);
baaca49f
GS
243 err = -EINVAL;
244 goto out_release;
1da177e4
LT
245 }
246
247 /* Ensure that we are not runnable on dying cpu */
248 old_allowed = current->cpus_allowed;
f70316da 249 cpus_setall(tmp);
1da177e4 250 cpu_clear(cpu, tmp);
f70316da 251 set_cpus_allowed_ptr(current, &tmp);
1da177e4 252
db912f96 253 p = __stop_machine_run(take_cpu_down, &tcd_param, cpu);
aa953877 254
8fa1d7d3 255 if (IS_ERR(p) || cpu_online(cpu)) {
1da177e4 256 /* CPU didn't die: tell everyone. Can't complain. */
8bb78442 257 if (raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
e7407dcc 258 hcpu) == NOTIFY_BAD)
1da177e4
LT
259 BUG();
260
8fa1d7d3
ST
261 if (IS_ERR(p)) {
262 err = PTR_ERR(p);
263 goto out_allowed;
264 }
1da177e4 265 goto out_thread;
8fa1d7d3 266 }
1da177e4
LT
267
268 /* Wait for it to sleep (leaving idle task). */
269 while (!idle_cpu(cpu))
270 yield();
271
272 /* This actually kills the CPU. */
273 __cpu_die(cpu);
274
1da177e4 275 /* CPU is completely dead: tell everyone. Too late to complain. */
8bb78442
RW
276 if (raw_notifier_call_chain(&cpu_chain, CPU_DEAD | mod,
277 hcpu) == NOTIFY_BAD)
1da177e4
LT
278 BUG();
279
280 check_for_tasks(cpu);
281
282out_thread:
283 err = kthread_stop(p);
284out_allowed:
f70316da 285 set_cpus_allowed_ptr(current, &old_allowed);
baaca49f 286out_release:
d221938c 287 cpu_hotplug_done();
e3920fb4
RW
288 return err;
289}
290
514a20a5 291int __ref cpu_down(unsigned int cpu)
e3920fb4
RW
292{
293 int err = 0;
294
d221938c 295 cpu_maps_update_begin();
e761b772
MK
296
297 if (cpu_hotplug_disabled) {
e3920fb4 298 err = -EBUSY;
e761b772
MK
299 goto out;
300 }
301
302 cpu_clear(cpu, cpu_active_map);
303
304 err = _cpu_down(cpu, 0);
e3920fb4 305
e761b772
MK
306 if (cpu_online(cpu))
307 cpu_set(cpu, cpu_active_map);
308
309out:
d221938c 310 cpu_maps_update_done();
1da177e4
LT
311 return err;
312}
b62b8ef9 313EXPORT_SYMBOL(cpu_down);
1da177e4
LT
314#endif /*CONFIG_HOTPLUG_CPU*/
315
e3920fb4 316/* Requires cpu_add_remove_lock to be held */
8bb78442 317static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
1da177e4 318{
baaca49f 319 int ret, nr_calls = 0;
1da177e4 320 void *hcpu = (void *)(long)cpu;
8bb78442 321 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
1da177e4 322
e3920fb4
RW
323 if (cpu_online(cpu) || !cpu_present(cpu))
324 return -EINVAL;
90d45d17 325
d221938c 326 cpu_hotplug_begin();
8bb78442 327 ret = __raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE | mod, hcpu,
baaca49f 328 -1, &nr_calls);
1da177e4 329 if (ret == NOTIFY_BAD) {
a0d8cdb6 330 nr_calls--;
1da177e4 331 printk("%s: attempt to bring up CPU %u failed\n",
af1f16d0 332 __func__, cpu);
1da177e4
LT
333 ret = -EINVAL;
334 goto out_notify;
335 }
336
337 /* Arch-specific enabling code. */
338 ret = __cpu_up(cpu);
339 if (ret != 0)
340 goto out_notify;
6978c705 341 BUG_ON(!cpu_online(cpu));
1da177e4
LT
342
343 /* Now call notifier in preparation. */
8bb78442 344 raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu);
1da177e4
LT
345
346out_notify:
347 if (ret != 0)
baaca49f 348 __raw_notifier_call_chain(&cpu_chain,
8bb78442 349 CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
d221938c 350 cpu_hotplug_done();
e3920fb4
RW
351
352 return ret;
353}
354
b282b6f8 355int __cpuinit cpu_up(unsigned int cpu)
e3920fb4
RW
356{
357 int err = 0;
73e753a5
KH
358 if (!cpu_isset(cpu, cpu_possible_map)) {
359 printk(KERN_ERR "can't online cpu %d because it is not "
360 "configured as may-hotadd at boot time\n", cpu);
361#if defined(CONFIG_IA64) || defined(CONFIG_X86_64) || defined(CONFIG_S390)
362 printk(KERN_ERR "please check additional_cpus= boot "
363 "parameter\n");
364#endif
365 return -EINVAL;
366 }
e3920fb4 367
d221938c 368 cpu_maps_update_begin();
e761b772
MK
369
370 if (cpu_hotplug_disabled) {
e3920fb4 371 err = -EBUSY;
e761b772
MK
372 goto out;
373 }
374
375 err = _cpu_up(cpu, 0);
376
377 if (cpu_online(cpu))
378 cpu_set(cpu, cpu_active_map);
e3920fb4 379
e761b772 380out:
d221938c 381 cpu_maps_update_done();
e3920fb4
RW
382 return err;
383}
384
f3de4be9 385#ifdef CONFIG_PM_SLEEP_SMP
e3920fb4
RW
386static cpumask_t frozen_cpus;
387
388int disable_nonboot_cpus(void)
389{
e1d9fd2e 390 int cpu, first_cpu, error = 0;
e3920fb4 391
d221938c 392 cpu_maps_update_begin();
1d64b9cb 393 first_cpu = first_cpu(cpu_online_map);
e3920fb4
RW
394 /* We take down all of the non-boot CPUs in one shot to avoid races
395 * with the userspace trying to use the CPU hotplug at the same time
396 */
397 cpus_clear(frozen_cpus);
398 printk("Disabling non-boot CPUs ...\n");
399 for_each_online_cpu(cpu) {
400 if (cpu == first_cpu)
401 continue;
8bb78442 402 error = _cpu_down(cpu, 1);
e3920fb4
RW
403 if (!error) {
404 cpu_set(cpu, frozen_cpus);
405 printk("CPU%d is down\n", cpu);
406 } else {
407 printk(KERN_ERR "Error taking CPU%d down: %d\n",
408 cpu, error);
409 break;
410 }
411 }
412 if (!error) {
413 BUG_ON(num_online_cpus() > 1);
414 /* Make sure the CPUs won't be enabled by someone else */
415 cpu_hotplug_disabled = 1;
416 } else {
e1d9fd2e 417 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
e3920fb4 418 }
d221938c 419 cpu_maps_update_done();
e3920fb4
RW
420 return error;
421}
422
fa7303e2 423void __ref enable_nonboot_cpus(void)
e3920fb4
RW
424{
425 int cpu, error;
426
427 /* Allow everyone to use the CPU hotplug again */
d221938c 428 cpu_maps_update_begin();
e3920fb4 429 cpu_hotplug_disabled = 0;
ed746e3b 430 if (cpus_empty(frozen_cpus))
1d64b9cb 431 goto out;
e3920fb4
RW
432
433 printk("Enabling non-boot CPUs ...\n");
434 for_each_cpu_mask(cpu, frozen_cpus) {
8bb78442 435 error = _cpu_up(cpu, 1);
e3920fb4
RW
436 if (!error) {
437 printk("CPU%d is up\n", cpu);
438 continue;
439 }
1d64b9cb 440 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
e3920fb4
RW
441 }
442 cpus_clear(frozen_cpus);
1d64b9cb 443out:
d221938c 444 cpu_maps_update_done();
1da177e4 445}
f3de4be9 446#endif /* CONFIG_PM_SLEEP_SMP */
68f4f1ec
MK
447
448#endif /* CONFIG_SMP */
This page took 0.351093 seconds and 5 git commands to generate.