cgroup: Add generic cgroup subsystem permission checks.
[linux-2.6.git] / kernel / cpu.c
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>
16 #include <linux/mutex.h>
17 #include <linux/gfp.h>
18
19 #ifdef CONFIG_SMP
20 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
21 static DEFINE_MUTEX(cpu_add_remove_lock);
22
23 /*
24  * The following two API's must be used when attempting
25  * to serialize the updates to cpu_online_mask, cpu_present_mask.
26  */
27 void cpu_maps_update_begin(void)
28 {
29         mutex_lock(&cpu_add_remove_lock);
30 }
31
32 void cpu_maps_update_done(void)
33 {
34         mutex_unlock(&cpu_add_remove_lock);
35 }
36
37 static RAW_NOTIFIER_HEAD(cpu_chain);
38
39 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
40  * Should always be manipulated under cpu_add_remove_lock
41  */
42 static int cpu_hotplug_disabled;
43
44 #ifdef CONFIG_HOTPLUG_CPU
45
46 static struct {
47         struct task_struct *active_writer;
48         struct mutex lock; /* Synchronizes accesses to refcount, */
49         /*
50          * Also blocks the new readers during
51          * an ongoing cpu hotplug operation.
52          */
53         int refcount;
54 } cpu_hotplug = {
55         .active_writer = NULL,
56         .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
57         .refcount = 0,
58 };
59
60 void get_online_cpus(void)
61 {
62         might_sleep();
63         if (cpu_hotplug.active_writer == current)
64                 return;
65         mutex_lock(&cpu_hotplug.lock);
66         cpu_hotplug.refcount++;
67         mutex_unlock(&cpu_hotplug.lock);
68
69 }
70 EXPORT_SYMBOL_GPL(get_online_cpus);
71
72 void put_online_cpus(void)
73 {
74         if (cpu_hotplug.active_writer == current)
75                 return;
76         mutex_lock(&cpu_hotplug.lock);
77         if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
78                 wake_up_process(cpu_hotplug.active_writer);
79         mutex_unlock(&cpu_hotplug.lock);
80
81 }
82 EXPORT_SYMBOL_GPL(put_online_cpus);
83
84 /*
85  * This ensures that the hotplug operation can begin only when the
86  * refcount goes to zero.
87  *
88  * Note that during a cpu-hotplug operation, the new readers, if any,
89  * will be blocked by the cpu_hotplug.lock
90  *
91  * Since cpu_hotplug_begin() is always called after invoking
92  * cpu_maps_update_begin(), we can be sure that only one writer is active.
93  *
94  * Note that theoretically, there is a possibility of a livelock:
95  * - Refcount goes to zero, last reader wakes up the sleeping
96  *   writer.
97  * - Last reader unlocks the cpu_hotplug.lock.
98  * - A new reader arrives at this moment, bumps up the refcount.
99  * - The writer acquires the cpu_hotplug.lock finds the refcount
100  *   non zero and goes to sleep again.
101  *
102  * However, this is very difficult to achieve in practice since
103  * get_online_cpus() not an api which is called all that often.
104  *
105  */
106 static void cpu_hotplug_begin(void)
107 {
108         cpu_hotplug.active_writer = current;
109
110         for (;;) {
111                 mutex_lock(&cpu_hotplug.lock);
112                 if (likely(!cpu_hotplug.refcount))
113                         break;
114                 __set_current_state(TASK_UNINTERRUPTIBLE);
115                 mutex_unlock(&cpu_hotplug.lock);
116                 schedule();
117         }
118 }
119
120 static void cpu_hotplug_done(void)
121 {
122         cpu_hotplug.active_writer = NULL;
123         mutex_unlock(&cpu_hotplug.lock);
124 }
125
126 #else /* #if CONFIG_HOTPLUG_CPU */
127 static void cpu_hotplug_begin(void) {}
128 static void cpu_hotplug_done(void) {}
129 #endif  /* #else #if CONFIG_HOTPLUG_CPU */
130
131 /* Need to know about CPUs going up/down? */
132 int __ref register_cpu_notifier(struct notifier_block *nb)
133 {
134         int ret;
135         cpu_maps_update_begin();
136         ret = raw_notifier_chain_register(&cpu_chain, nb);
137         cpu_maps_update_done();
138         return ret;
139 }
140
141 static int __cpu_notify(unsigned long val, void *v, int nr_to_call,
142                         int *nr_calls)
143 {
144         int ret;
145
146         ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call,
147                                         nr_calls);
148
149         return notifier_to_errno(ret);
150 }
151
152 static int cpu_notify(unsigned long val, void *v)
153 {
154         return __cpu_notify(val, v, -1, NULL);
155 }
156
157 #ifdef CONFIG_HOTPLUG_CPU
158
159 static void cpu_notify_nofail(unsigned long val, void *v)
160 {
161         BUG_ON(cpu_notify(val, v));
162 }
163 EXPORT_SYMBOL(register_cpu_notifier);
164
165 void __ref unregister_cpu_notifier(struct notifier_block *nb)
166 {
167         cpu_maps_update_begin();
168         raw_notifier_chain_unregister(&cpu_chain, nb);
169         cpu_maps_update_done();
170 }
171 EXPORT_SYMBOL(unregister_cpu_notifier);
172
173 static inline void check_for_tasks(int cpu)
174 {
175         struct task_struct *p;
176
177         write_lock_irq(&tasklist_lock);
178         for_each_process(p) {
179                 if (task_cpu(p) == cpu && p->state == TASK_RUNNING &&
180                     (!cputime_eq(p->utime, cputime_zero) ||
181                      !cputime_eq(p->stime, cputime_zero)))
182                         printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d "
183                                 "(state = %ld, flags = %x)\n",
184                                 p->comm, task_pid_nr(p), cpu,
185                                 p->state, p->flags);
186         }
187         write_unlock_irq(&tasklist_lock);
188 }
189
190 struct take_cpu_down_param {
191         unsigned long mod;
192         void *hcpu;
193 };
194
195 /* Take this CPU down. */
196 static int __ref take_cpu_down(void *_param)
197 {
198         struct take_cpu_down_param *param = _param;
199         int err;
200
201         /* Ensure this CPU doesn't handle any more interrupts. */
202         err = __cpu_disable();
203         if (err < 0)
204                 return err;
205
206         cpu_notify(CPU_DYING | param->mod, param->hcpu);
207         return 0;
208 }
209
210 /* Requires cpu_add_remove_lock to be held */
211 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
212 {
213         int err, nr_calls = 0;
214         void *hcpu = (void *)(long)cpu;
215         unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
216         struct take_cpu_down_param tcd_param = {
217                 .mod = mod,
218                 .hcpu = hcpu,
219         };
220
221         if (num_online_cpus() == 1)
222                 return -EBUSY;
223
224         if (!cpu_online(cpu))
225                 return -EINVAL;
226
227         cpu_hotplug_begin();
228
229         err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
230         if (err) {
231                 nr_calls--;
232                 __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL);
233                 printk("%s: attempt to take down CPU %u failed\n",
234                                 __func__, cpu);
235                 goto out_release;
236         }
237
238         err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
239         if (err) {
240                 /* CPU didn't die: tell everyone.  Can't complain. */
241                 cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
242
243                 goto out_release;
244         }
245         BUG_ON(cpu_online(cpu));
246
247         /*
248          * The migration_call() CPU_DYING callback will have removed all
249          * runnable tasks from the cpu, there's only the idle task left now
250          * that the migration thread is done doing the stop_machine thing.
251          *
252          * Wait for the stop thread to go away.
253          */
254         while (!idle_cpu(cpu))
255                 cpu_relax();
256
257         /* This actually kills the CPU. */
258         __cpu_die(cpu);
259
260         /* CPU is completely dead: tell everyone.  Too late to complain. */
261         cpu_notify_nofail(CPU_DEAD | mod, hcpu);
262
263         check_for_tasks(cpu);
264
265 out_release:
266         cpu_hotplug_done();
267         if (!err)
268                 cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
269         return err;
270 }
271
272 int __ref cpu_down(unsigned int cpu)
273 {
274         int err;
275
276         cpu_maps_update_begin();
277
278         if (cpu_hotplug_disabled) {
279                 err = -EBUSY;
280                 goto out;
281         }
282
283         err = _cpu_down(cpu, 0);
284
285 out:
286         cpu_maps_update_done();
287         return err;
288 }
289 EXPORT_SYMBOL(cpu_down);
290 #endif /*CONFIG_HOTPLUG_CPU*/
291
292 /* Requires cpu_add_remove_lock to be held */
293 static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
294 {
295         int ret, nr_calls = 0;
296         void *hcpu = (void *)(long)cpu;
297         unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
298
299         if (cpu_online(cpu) || !cpu_present(cpu))
300                 return -EINVAL;
301
302         cpu_hotplug_begin();
303         ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls);
304         if (ret) {
305                 nr_calls--;
306                 printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
307                                 __func__, cpu);
308                 goto out_notify;
309         }
310
311         /* Arch-specific enabling code. */
312         ret = __cpu_up(cpu);
313         if (ret != 0)
314                 goto out_notify;
315         BUG_ON(!cpu_online(cpu));
316
317         /* Now call notifier in preparation. */
318         cpu_notify(CPU_ONLINE | mod, hcpu);
319
320 out_notify:
321         if (ret != 0)
322                 __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
323         cpu_hotplug_done();
324
325         return ret;
326 }
327
328 int __cpuinit cpu_up(unsigned int cpu)
329 {
330         int err = 0;
331
332 #ifdef  CONFIG_MEMORY_HOTPLUG
333         int nid;
334         pg_data_t       *pgdat;
335 #endif
336
337         if (!cpu_possible(cpu)) {
338                 printk(KERN_ERR "can't online cpu %d because it is not "
339                         "configured as may-hotadd at boot time\n", cpu);
340 #if defined(CONFIG_IA64)
341                 printk(KERN_ERR "please check additional_cpus= boot "
342                                 "parameter\n");
343 #endif
344                 return -EINVAL;
345         }
346
347 #ifdef  CONFIG_MEMORY_HOTPLUG
348         nid = cpu_to_node(cpu);
349         if (!node_online(nid)) {
350                 err = mem_online_node(nid);
351                 if (err)
352                         return err;
353         }
354
355         pgdat = NODE_DATA(nid);
356         if (!pgdat) {
357                 printk(KERN_ERR
358                         "Can't online cpu %d due to NULL pgdat\n", cpu);
359                 return -ENOMEM;
360         }
361
362         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
363                 mutex_lock(&zonelists_mutex);
364                 build_all_zonelists(NULL);
365                 mutex_unlock(&zonelists_mutex);
366         }
367 #endif
368
369         cpu_maps_update_begin();
370
371         if (cpu_hotplug_disabled) {
372                 err = -EBUSY;
373                 goto out;
374         }
375
376         err = _cpu_up(cpu, 0);
377
378 out:
379         cpu_maps_update_done();
380         return err;
381 }
382
383 #ifdef CONFIG_PM_SLEEP_SMP
384 static cpumask_var_t frozen_cpus;
385
386 void __weak arch_disable_nonboot_cpus_begin(void)
387 {
388 }
389
390 void __weak arch_disable_nonboot_cpus_end(void)
391 {
392 }
393
394 int disable_nonboot_cpus(void)
395 {
396         int cpu, first_cpu, error = 0;
397
398         cpu_maps_update_begin();
399         first_cpu = cpumask_first(cpu_online_mask);
400         /*
401          * We take down all of the non-boot CPUs in one shot to avoid races
402          * with the userspace trying to use the CPU hotplug at the same time
403          */
404         cpumask_clear(frozen_cpus);
405         arch_disable_nonboot_cpus_begin();
406
407         printk("Disabling non-boot CPUs ...\n");
408         for_each_online_cpu(cpu) {
409                 if (cpu == first_cpu)
410                         continue;
411                 error = _cpu_down(cpu, 1);
412                 if (!error)
413                         cpumask_set_cpu(cpu, frozen_cpus);
414                 else {
415                         printk(KERN_ERR "Error taking CPU%d down: %d\n",
416                                 cpu, error);
417                         break;
418                 }
419         }
420
421         arch_disable_nonboot_cpus_end();
422
423         if (!error) {
424                 BUG_ON(num_online_cpus() > 1);
425                 /* Make sure the CPUs won't be enabled by someone else */
426                 cpu_hotplug_disabled = 1;
427         } else {
428                 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
429         }
430         cpu_maps_update_done();
431         return error;
432 }
433
434 void __weak arch_enable_nonboot_cpus_begin(void)
435 {
436 }
437
438 void __weak arch_enable_nonboot_cpus_end(void)
439 {
440 }
441
442 void __ref enable_nonboot_cpus(void)
443 {
444         int cpu, error;
445
446         /* Allow everyone to use the CPU hotplug again */
447         cpu_maps_update_begin();
448         cpu_hotplug_disabled = 0;
449         if (cpumask_empty(frozen_cpus))
450                 goto out;
451
452         printk(KERN_INFO "Enabling non-boot CPUs ...\n");
453
454         arch_enable_nonboot_cpus_begin();
455
456         for_each_cpu(cpu, frozen_cpus) {
457                 error = _cpu_up(cpu, 1);
458                 if (!error) {
459                         printk(KERN_INFO "CPU%d is up\n", cpu);
460                         continue;
461                 }
462                 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
463         }
464
465         arch_enable_nonboot_cpus_end();
466
467         cpumask_clear(frozen_cpus);
468 out:
469         cpu_maps_update_done();
470 }
471
472 static int alloc_frozen_cpus(void)
473 {
474         if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
475                 return -ENOMEM;
476         return 0;
477 }
478 core_initcall(alloc_frozen_cpus);
479 #endif /* CONFIG_PM_SLEEP_SMP */
480
481 /**
482  * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
483  * @cpu: cpu that just started
484  *
485  * This function calls the cpu_chain notifiers with CPU_STARTING.
486  * It must be called by the arch code on the new cpu, before the new cpu
487  * enables interrupts and before the "boot" cpu returns from __cpu_up().
488  */
489 void __cpuinit notify_cpu_starting(unsigned int cpu)
490 {
491         unsigned long val = CPU_STARTING;
492
493 #ifdef CONFIG_PM_SLEEP_SMP
494         if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus))
495                 val = CPU_STARTING_FROZEN;
496 #endif /* CONFIG_PM_SLEEP_SMP */
497         cpu_notify(val, (void *)(long)cpu);
498 }
499
500 #endif /* CONFIG_SMP */
501
502 /*
503  * cpu_bit_bitmap[] is a special, "compressed" data structure that
504  * represents all NR_CPUS bits binary values of 1<<nr.
505  *
506  * It is used by cpumask_of() to get a constant address to a CPU
507  * mask value that has a single bit set only.
508  */
509
510 /* cpu_bit_bitmap[0] is empty - so we can back into it */
511 #define MASK_DECLARE_1(x)       [x+1][0] = (1UL << (x))
512 #define MASK_DECLARE_2(x)       MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
513 #define MASK_DECLARE_4(x)       MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
514 #define MASK_DECLARE_8(x)       MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
515
516 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
517
518         MASK_DECLARE_8(0),      MASK_DECLARE_8(8),
519         MASK_DECLARE_8(16),     MASK_DECLARE_8(24),
520 #if BITS_PER_LONG > 32
521         MASK_DECLARE_8(32),     MASK_DECLARE_8(40),
522         MASK_DECLARE_8(48),     MASK_DECLARE_8(56),
523 #endif
524 };
525 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
526
527 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
528 EXPORT_SYMBOL(cpu_all_bits);
529
530 #ifdef CONFIG_INIT_ALL_POSSIBLE
531 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
532         = CPU_BITS_ALL;
533 #else
534 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
535 #endif
536 const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
537 EXPORT_SYMBOL(cpu_possible_mask);
538
539 static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
540 const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
541 EXPORT_SYMBOL(cpu_online_mask);
542
543 static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
544 const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
545 EXPORT_SYMBOL(cpu_present_mask);
546
547 static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
548 const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
549 EXPORT_SYMBOL(cpu_active_mask);
550
551 void set_cpu_possible(unsigned int cpu, bool possible)
552 {
553         if (possible)
554                 cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
555         else
556                 cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
557 }
558
559 void set_cpu_present(unsigned int cpu, bool present)
560 {
561         if (present)
562                 cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
563         else
564                 cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
565 }
566
567 void set_cpu_online(unsigned int cpu, bool online)
568 {
569         if (online)
570                 cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
571         else
572                 cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
573 }
574
575 void set_cpu_active(unsigned int cpu, bool active)
576 {
577         if (active)
578                 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
579         else
580                 cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
581 }
582
583 void init_cpu_present(const struct cpumask *src)
584 {
585         cpumask_copy(to_cpumask(cpu_present_bits), src);
586 }
587
588 void init_cpu_possible(const struct cpumask *src)
589 {
590         cpumask_copy(to_cpumask(cpu_possible_bits), src);
591 }
592
593 void init_cpu_online(const struct cpumask *src)
594 {
595         cpumask_copy(to_cpumask(cpu_online_bits), src);
596 }