Merge branch 'x86-uv-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[pandora-kernel.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  /* #esle #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
164 EXPORT_SYMBOL(register_cpu_notifier);
165
166 void __ref unregister_cpu_notifier(struct notifier_block *nb)
167 {
168         cpu_maps_update_begin();
169         raw_notifier_chain_unregister(&cpu_chain, nb);
170         cpu_maps_update_done();
171 }
172 EXPORT_SYMBOL(unregister_cpu_notifier);
173
174 static 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 && p->state == TASK_RUNNING &&
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 "
184                                 "(state = %ld, flags = %x)\n",
185                                 p->comm, task_pid_nr(p), cpu,
186                                 p->state, p->flags);
187         }
188         write_unlock_irq(&tasklist_lock);
189 }
190
191 struct take_cpu_down_param {
192         unsigned long mod;
193         void *hcpu;
194 };
195
196 /* Take this CPU down. */
197 static int __ref take_cpu_down(void *_param)
198 {
199         struct take_cpu_down_param *param = _param;
200         int err;
201
202         /* Ensure this CPU doesn't handle any more interrupts. */
203         err = __cpu_disable();
204         if (err < 0)
205                 return err;
206
207         cpu_notify(CPU_DYING | param->mod, param->hcpu);
208
209         return 0;
210 }
211
212 /* Requires cpu_add_remove_lock to be held */
213 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
214 {
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 = {
219                 .mod = mod,
220                 .hcpu = hcpu,
221         };
222
223         if (num_online_cpus() == 1)
224                 return -EBUSY;
225
226         if (!cpu_online(cpu))
227                 return -EINVAL;
228
229         cpu_hotplug_begin();
230         err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
231         if (err) {
232                 nr_calls--;
233                 __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL);
234                 printk("%s: attempt to take down CPU %u failed\n",
235                                 __func__, cpu);
236                 goto out_release;
237         }
238
239         err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
240         if (err) {
241                 /* CPU didn't die: tell everyone.  Can't complain. */
242                 cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
243
244                 goto out_release;
245         }
246         BUG_ON(cpu_online(cpu));
247
248         /*
249          * The migration_call() CPU_DYING callback will have removed all
250          * runnable tasks from the cpu, there's only the idle task left now
251          * that the migration thread is done doing the stop_machine thing.
252          *
253          * Wait for the stop thread to go away.
254          */
255         while (!idle_cpu(cpu))
256                 cpu_relax();
257
258         /* This actually kills the CPU. */
259         __cpu_die(cpu);
260
261         /* CPU is completely dead: tell everyone.  Too late to complain. */
262         cpu_notify_nofail(CPU_DEAD | mod, hcpu);
263
264         check_for_tasks(cpu);
265
266 out_release:
267         cpu_hotplug_done();
268         if (!err)
269                 cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
270         return err;
271 }
272
273 int __ref cpu_down(unsigned int cpu)
274 {
275         int err;
276
277         cpu_maps_update_begin();
278
279         if (cpu_hotplug_disabled) {
280                 err = -EBUSY;
281                 goto out;
282         }
283
284         err = _cpu_down(cpu, 0);
285
286 out:
287         cpu_maps_update_done();
288         return err;
289 }
290 EXPORT_SYMBOL(cpu_down);
291 #endif /*CONFIG_HOTPLUG_CPU*/
292
293 /* Requires cpu_add_remove_lock to be held */
294 static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
295 {
296         int ret, nr_calls = 0;
297         void *hcpu = (void *)(long)cpu;
298         unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
299
300         if (cpu_online(cpu) || !cpu_present(cpu))
301                 return -EINVAL;
302
303         cpu_hotplug_begin();
304         ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls);
305         if (ret) {
306                 nr_calls--;
307                 printk("%s: attempt to bring up CPU %u failed\n",
308                                 __func__, cpu);
309                 goto out_notify;
310         }
311
312         /* Arch-specific enabling code. */
313         ret = __cpu_up(cpu);
314         if (ret != 0)
315                 goto out_notify;
316         BUG_ON(!cpu_online(cpu));
317
318         /* Now call notifier in preparation. */
319         cpu_notify(CPU_ONLINE | mod, hcpu);
320
321 out_notify:
322         if (ret != 0)
323                 __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
324         cpu_hotplug_done();
325
326         return ret;
327 }
328
329 int __cpuinit cpu_up(unsigned int cpu)
330 {
331         int err = 0;
332
333 #ifdef  CONFIG_MEMORY_HOTPLUG
334         int nid;
335         pg_data_t       *pgdat;
336 #endif
337
338         if (!cpu_possible(cpu)) {
339                 printk(KERN_ERR "can't online cpu %d because it is not "
340                         "configured as may-hotadd at boot time\n", cpu);
341 #if defined(CONFIG_IA64)
342                 printk(KERN_ERR "please check additional_cpus= boot "
343                                 "parameter\n");
344 #endif
345                 return -EINVAL;
346         }
347
348 #ifdef  CONFIG_MEMORY_HOTPLUG
349         nid = cpu_to_node(cpu);
350         if (!node_online(nid)) {
351                 err = mem_online_node(nid);
352                 if (err)
353                         return err;
354         }
355
356         pgdat = NODE_DATA(nid);
357         if (!pgdat) {
358                 printk(KERN_ERR
359                         "Can't online cpu %d due to NULL pgdat\n", cpu);
360                 return -ENOMEM;
361         }
362
363         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
364                 mutex_lock(&zonelists_mutex);
365                 build_all_zonelists(NULL);
366                 mutex_unlock(&zonelists_mutex);
367         }
368 #endif
369
370         cpu_maps_update_begin();
371
372         if (cpu_hotplug_disabled) {
373                 err = -EBUSY;
374                 goto out;
375         }
376
377         err = _cpu_up(cpu, 0);
378
379 out:
380         cpu_maps_update_done();
381         return err;
382 }
383
384 #ifdef CONFIG_PM_SLEEP_SMP
385 static cpumask_var_t frozen_cpus;
386
387 int disable_nonboot_cpus(void)
388 {
389         int cpu, first_cpu, error = 0;
390
391         cpu_maps_update_begin();
392         first_cpu = cpumask_first(cpu_online_mask);
393         /*
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         cpumask_clear(frozen_cpus);
398
399         printk("Disabling non-boot CPUs ...\n");
400         for_each_online_cpu(cpu) {
401                 if (cpu == first_cpu)
402                         continue;
403                 error = _cpu_down(cpu, 1);
404                 if (!error)
405                         cpumask_set_cpu(cpu, frozen_cpus);
406                 else {
407                         printk(KERN_ERR "Error taking CPU%d down: %d\n",
408                                 cpu, error);
409                         break;
410                 }
411         }
412
413         if (!error) {
414                 BUG_ON(num_online_cpus() > 1);
415                 /* Make sure the CPUs won't be enabled by someone else */
416                 cpu_hotplug_disabled = 1;
417         } else {
418                 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
419         }
420         cpu_maps_update_done();
421         return error;
422 }
423
424 void __weak arch_enable_nonboot_cpus_begin(void)
425 {
426 }
427
428 void __weak arch_enable_nonboot_cpus_end(void)
429 {
430 }
431
432 void __ref enable_nonboot_cpus(void)
433 {
434         int cpu, error;
435
436         /* Allow everyone to use the CPU hotplug again */
437         cpu_maps_update_begin();
438         cpu_hotplug_disabled = 0;
439         if (cpumask_empty(frozen_cpus))
440                 goto out;
441
442         printk("Enabling non-boot CPUs ...\n");
443
444         arch_enable_nonboot_cpus_begin();
445
446         for_each_cpu(cpu, frozen_cpus) {
447                 error = _cpu_up(cpu, 1);
448                 if (!error) {
449                         printk("CPU%d is up\n", cpu);
450                         continue;
451                 }
452                 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
453         }
454
455         arch_enable_nonboot_cpus_end();
456
457         cpumask_clear(frozen_cpus);
458 out:
459         cpu_maps_update_done();
460 }
461
462 static int alloc_frozen_cpus(void)
463 {
464         if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
465                 return -ENOMEM;
466         return 0;
467 }
468 core_initcall(alloc_frozen_cpus);
469 #endif /* CONFIG_PM_SLEEP_SMP */
470
471 /**
472  * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
473  * @cpu: cpu that just started
474  *
475  * This function calls the cpu_chain notifiers with CPU_STARTING.
476  * It must be called by the arch code on the new cpu, before the new cpu
477  * enables interrupts and before the "boot" cpu returns from __cpu_up().
478  */
479 void __cpuinit notify_cpu_starting(unsigned int cpu)
480 {
481         unsigned long val = CPU_STARTING;
482
483 #ifdef CONFIG_PM_SLEEP_SMP
484         if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus))
485                 val = CPU_STARTING_FROZEN;
486 #endif /* CONFIG_PM_SLEEP_SMP */
487         cpu_notify(val, (void *)(long)cpu);
488 }
489
490 #endif /* CONFIG_SMP */
491
492 /*
493  * cpu_bit_bitmap[] is a special, "compressed" data structure that
494  * represents all NR_CPUS bits binary values of 1<<nr.
495  *
496  * It is used by cpumask_of() to get a constant address to a CPU
497  * mask value that has a single bit set only.
498  */
499
500 /* cpu_bit_bitmap[0] is empty - so we can back into it */
501 #define MASK_DECLARE_1(x)       [x+1][0] = 1UL << (x)
502 #define MASK_DECLARE_2(x)       MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
503 #define MASK_DECLARE_4(x)       MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
504 #define MASK_DECLARE_8(x)       MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
505
506 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
507
508         MASK_DECLARE_8(0),      MASK_DECLARE_8(8),
509         MASK_DECLARE_8(16),     MASK_DECLARE_8(24),
510 #if BITS_PER_LONG > 32
511         MASK_DECLARE_8(32),     MASK_DECLARE_8(40),
512         MASK_DECLARE_8(48),     MASK_DECLARE_8(56),
513 #endif
514 };
515 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
516
517 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
518 EXPORT_SYMBOL(cpu_all_bits);
519
520 #ifdef CONFIG_INIT_ALL_POSSIBLE
521 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
522         = CPU_BITS_ALL;
523 #else
524 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
525 #endif
526 const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
527 EXPORT_SYMBOL(cpu_possible_mask);
528
529 static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
530 const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
531 EXPORT_SYMBOL(cpu_online_mask);
532
533 static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
534 const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
535 EXPORT_SYMBOL(cpu_present_mask);
536
537 static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
538 const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
539 EXPORT_SYMBOL(cpu_active_mask);
540
541 void set_cpu_possible(unsigned int cpu, bool possible)
542 {
543         if (possible)
544                 cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
545         else
546                 cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
547 }
548
549 void set_cpu_present(unsigned int cpu, bool present)
550 {
551         if (present)
552                 cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
553         else
554                 cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
555 }
556
557 void set_cpu_online(unsigned int cpu, bool online)
558 {
559         if (online)
560                 cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
561         else
562                 cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
563 }
564
565 void set_cpu_active(unsigned int cpu, bool active)
566 {
567         if (active)
568                 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
569         else
570                 cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
571 }
572
573 void init_cpu_present(const struct cpumask *src)
574 {
575         cpumask_copy(to_cpumask(cpu_present_bits), src);
576 }
577
578 void init_cpu_possible(const struct cpumask *src)
579 {
580         cpumask_copy(to_cpumask(cpu_possible_bits), src);
581 }
582
583 void init_cpu_online(const struct cpumask *src)
584 {
585         cpumask_copy(to_cpumask(cpu_online_bits), src);
586 }