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