Merge branches 'x86-fixes-for-linus', 'sched-fixes-for-linus', 'timers-fixes-for...
[pandora-kernel.git] / drivers / cpufreq / cpufreq.c
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *
7  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
8  *      Added handling for CPU hotplug
9  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
10  *      Fix handling for CPU hotplug -- affected CPUs
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License version 2 as
14  * published by the Free Software Foundation.
15  *
16  */
17
18 #include <linux/kernel.h>
19 #include <linux/module.h>
20 #include <linux/init.h>
21 #include <linux/notifier.h>
22 #include <linux/cpufreq.h>
23 #include <linux/delay.h>
24 #include <linux/interrupt.h>
25 #include <linux/spinlock.h>
26 #include <linux/device.h>
27 #include <linux/slab.h>
28 #include <linux/cpu.h>
29 #include <linux/completion.h>
30 #include <linux/mutex.h>
31 #include <linux/syscore_ops.h>
32
33 #include <trace/events/power.h>
34
35 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
36                                                 "cpufreq-core", msg)
37
38 /**
39  * The "cpufreq driver" - the arch- or hardware-dependent low
40  * level driver of CPUFreq support, and its spinlock. This lock
41  * also protects the cpufreq_cpu_data array.
42  */
43 static struct cpufreq_driver *cpufreq_driver;
44 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
45 #ifdef CONFIG_HOTPLUG_CPU
46 /* This one keeps track of the previously set governor of a removed CPU */
47 static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
48 #endif
49 static DEFINE_SPINLOCK(cpufreq_driver_lock);
50
51 /*
52  * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
53  * all cpufreq/hotplug/workqueue/etc related lock issues.
54  *
55  * The rules for this semaphore:
56  * - Any routine that wants to read from the policy structure will
57  *   do a down_read on this semaphore.
58  * - Any routine that will write to the policy structure and/or may take away
59  *   the policy altogether (eg. CPU hotplug), will hold this lock in write
60  *   mode before doing so.
61  *
62  * Additional rules:
63  * - All holders of the lock should check to make sure that the CPU they
64  *   are concerned with are online after they get the lock.
65  * - Governor routines that can be called in cpufreq hotplug path should not
66  *   take this sem as top level hotplug notifier handler takes this.
67  * - Lock should not be held across
68  *     __cpufreq_governor(data, CPUFREQ_GOV_STOP);
69  */
70 static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
71 static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
72
73 #define lock_policy_rwsem(mode, cpu)                                    \
74 static int lock_policy_rwsem_##mode                                     \
75 (int cpu)                                                               \
76 {                                                                       \
77         int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);              \
78         BUG_ON(policy_cpu == -1);                                       \
79         down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));            \
80         if (unlikely(!cpu_online(cpu))) {                               \
81                 up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));      \
82                 return -1;                                              \
83         }                                                               \
84                                                                         \
85         return 0;                                                       \
86 }
87
88 lock_policy_rwsem(read, cpu);
89
90 lock_policy_rwsem(write, cpu);
91
92 static void unlock_policy_rwsem_read(int cpu)
93 {
94         int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
95         BUG_ON(policy_cpu == -1);
96         up_read(&per_cpu(cpu_policy_rwsem, policy_cpu));
97 }
98
99 static void unlock_policy_rwsem_write(int cpu)
100 {
101         int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
102         BUG_ON(policy_cpu == -1);
103         up_write(&per_cpu(cpu_policy_rwsem, policy_cpu));
104 }
105
106
107 /* internal prototypes */
108 static int __cpufreq_governor(struct cpufreq_policy *policy,
109                 unsigned int event);
110 static unsigned int __cpufreq_get(unsigned int cpu);
111 static void handle_update(struct work_struct *work);
112
113 /**
114  * Two notifier lists: the "policy" list is involved in the
115  * validation process for a new CPU frequency policy; the
116  * "transition" list for kernel code that needs to handle
117  * changes to devices when the CPU clock speed changes.
118  * The mutex locks both lists.
119  */
120 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
121 static struct srcu_notifier_head cpufreq_transition_notifier_list;
122
123 static bool init_cpufreq_transition_notifier_list_called;
124 static int __init init_cpufreq_transition_notifier_list(void)
125 {
126         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
127         init_cpufreq_transition_notifier_list_called = true;
128         return 0;
129 }
130 pure_initcall(init_cpufreq_transition_notifier_list);
131
132 static LIST_HEAD(cpufreq_governor_list);
133 static DEFINE_MUTEX(cpufreq_governor_mutex);
134
135 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
136 {
137         struct cpufreq_policy *data;
138         unsigned long flags;
139
140         if (cpu >= nr_cpu_ids)
141                 goto err_out;
142
143         /* get the cpufreq driver */
144         spin_lock_irqsave(&cpufreq_driver_lock, flags);
145
146         if (!cpufreq_driver)
147                 goto err_out_unlock;
148
149         if (!try_module_get(cpufreq_driver->owner))
150                 goto err_out_unlock;
151
152
153         /* get the CPU */
154         data = per_cpu(cpufreq_cpu_data, cpu);
155
156         if (!data)
157                 goto err_out_put_module;
158
159         if (!kobject_get(&data->kobj))
160                 goto err_out_put_module;
161
162         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
163         return data;
164
165 err_out_put_module:
166         module_put(cpufreq_driver->owner);
167 err_out_unlock:
168         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
169 err_out:
170         return NULL;
171 }
172 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
173
174
175 void cpufreq_cpu_put(struct cpufreq_policy *data)
176 {
177         kobject_put(&data->kobj);
178         module_put(cpufreq_driver->owner);
179 }
180 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
181
182
183 /*********************************************************************
184  *                     UNIFIED DEBUG HELPERS                         *
185  *********************************************************************/
186 #ifdef CONFIG_CPU_FREQ_DEBUG
187
188 /* what part(s) of the CPUfreq subsystem are debugged? */
189 static unsigned int debug;
190
191 /* is the debug output ratelimit'ed using printk_ratelimit? User can
192  * set or modify this value.
193  */
194 static unsigned int debug_ratelimit = 1;
195
196 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
197  * loading of a cpufreq driver, temporarily disabled when a new policy
198  * is set, and disabled upon cpufreq driver removal
199  */
200 static unsigned int disable_ratelimit = 1;
201 static DEFINE_SPINLOCK(disable_ratelimit_lock);
202
203 static void cpufreq_debug_enable_ratelimit(void)
204 {
205         unsigned long flags;
206
207         spin_lock_irqsave(&disable_ratelimit_lock, flags);
208         if (disable_ratelimit)
209                 disable_ratelimit--;
210         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
211 }
212
213 static void cpufreq_debug_disable_ratelimit(void)
214 {
215         unsigned long flags;
216
217         spin_lock_irqsave(&disable_ratelimit_lock, flags);
218         disable_ratelimit++;
219         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
220 }
221
222 void cpufreq_debug_printk(unsigned int type, const char *prefix,
223                         const char *fmt, ...)
224 {
225         char s[256];
226         va_list args;
227         unsigned int len;
228         unsigned long flags;
229
230         WARN_ON(!prefix);
231         if (type & debug) {
232                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
233                 if (!disable_ratelimit && debug_ratelimit
234                                         && !printk_ratelimit()) {
235                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
236                         return;
237                 }
238                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
239
240                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
241
242                 va_start(args, fmt);
243                 len += vsnprintf(&s[len], (256 - len), fmt, args);
244                 va_end(args);
245
246                 printk(s);
247
248                 WARN_ON(len < 5);
249         }
250 }
251 EXPORT_SYMBOL(cpufreq_debug_printk);
252
253
254 module_param(debug, uint, 0644);
255 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core,"
256                         " 2 to debug drivers, and 4 to debug governors.");
257
258 module_param(debug_ratelimit, uint, 0644);
259 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging:"
260                                         " set to 0 to disable ratelimiting.");
261
262 #else /* !CONFIG_CPU_FREQ_DEBUG */
263
264 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
265 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
266
267 #endif /* CONFIG_CPU_FREQ_DEBUG */
268
269
270 /*********************************************************************
271  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
272  *********************************************************************/
273
274 /**
275  * adjust_jiffies - adjust the system "loops_per_jiffy"
276  *
277  * This function alters the system "loops_per_jiffy" for the clock
278  * speed change. Note that loops_per_jiffy cannot be updated on SMP
279  * systems as each CPU might be scaled differently. So, use the arch
280  * per-CPU loops_per_jiffy value wherever possible.
281  */
282 #ifndef CONFIG_SMP
283 static unsigned long l_p_j_ref;
284 static unsigned int  l_p_j_ref_freq;
285
286 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
287 {
288         if (ci->flags & CPUFREQ_CONST_LOOPS)
289                 return;
290
291         if (!l_p_j_ref_freq) {
292                 l_p_j_ref = loops_per_jiffy;
293                 l_p_j_ref_freq = ci->old;
294                 dprintk("saving %lu as reference value for loops_per_jiffy; "
295                         "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
296         }
297         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
298             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
299             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
300                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
301                                                                 ci->new);
302                 dprintk("scaling loops_per_jiffy to %lu "
303                         "for frequency %u kHz\n", loops_per_jiffy, ci->new);
304         }
305 }
306 #else
307 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
308 {
309         return;
310 }
311 #endif
312
313
314 /**
315  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
316  * on frequency transition.
317  *
318  * This function calls the transition notifiers and the "adjust_jiffies"
319  * function. It is called twice on all CPU frequency changes that have
320  * external effects.
321  */
322 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
323 {
324         struct cpufreq_policy *policy;
325
326         BUG_ON(irqs_disabled());
327
328         freqs->flags = cpufreq_driver->flags;
329         dprintk("notification %u of frequency transition to %u kHz\n",
330                 state, freqs->new);
331
332         policy = per_cpu(cpufreq_cpu_data, freqs->cpu);
333         switch (state) {
334
335         case CPUFREQ_PRECHANGE:
336                 /* detect if the driver reported a value as "old frequency"
337                  * which is not equal to what the cpufreq core thinks is
338                  * "old frequency".
339                  */
340                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
341                         if ((policy) && (policy->cpu == freqs->cpu) &&
342                             (policy->cur) && (policy->cur != freqs->old)) {
343                                 dprintk("Warning: CPU frequency is"
344                                         " %u, cpufreq assumed %u kHz.\n",
345                                         freqs->old, policy->cur);
346                                 freqs->old = policy->cur;
347                         }
348                 }
349                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
350                                 CPUFREQ_PRECHANGE, freqs);
351                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
352                 break;
353
354         case CPUFREQ_POSTCHANGE:
355                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
356                 dprintk("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
357                         (unsigned long)freqs->cpu);
358                 trace_power_frequency(POWER_PSTATE, freqs->new, freqs->cpu);
359                 trace_cpu_frequency(freqs->new, freqs->cpu);
360                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
361                                 CPUFREQ_POSTCHANGE, freqs);
362                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
363                         policy->cur = freqs->new;
364                 break;
365         }
366 }
367 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
368
369
370
371 /*********************************************************************
372  *                          SYSFS INTERFACE                          *
373  *********************************************************************/
374
375 static struct cpufreq_governor *__find_governor(const char *str_governor)
376 {
377         struct cpufreq_governor *t;
378
379         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
380                 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
381                         return t;
382
383         return NULL;
384 }
385
386 /**
387  * cpufreq_parse_governor - parse a governor string
388  */
389 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
390                                 struct cpufreq_governor **governor)
391 {
392         int err = -EINVAL;
393
394         if (!cpufreq_driver)
395                 goto out;
396
397         if (cpufreq_driver->setpolicy) {
398                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
399                         *policy = CPUFREQ_POLICY_PERFORMANCE;
400                         err = 0;
401                 } else if (!strnicmp(str_governor, "powersave",
402                                                 CPUFREQ_NAME_LEN)) {
403                         *policy = CPUFREQ_POLICY_POWERSAVE;
404                         err = 0;
405                 }
406         } else if (cpufreq_driver->target) {
407                 struct cpufreq_governor *t;
408
409                 mutex_lock(&cpufreq_governor_mutex);
410
411                 t = __find_governor(str_governor);
412
413                 if (t == NULL) {
414                         char *name = kasprintf(GFP_KERNEL, "cpufreq_%s",
415                                                                 str_governor);
416
417                         if (name) {
418                                 int ret;
419
420                                 mutex_unlock(&cpufreq_governor_mutex);
421                                 ret = request_module("%s", name);
422                                 mutex_lock(&cpufreq_governor_mutex);
423
424                                 if (ret == 0)
425                                         t = __find_governor(str_governor);
426                         }
427
428                         kfree(name);
429                 }
430
431                 if (t != NULL) {
432                         *governor = t;
433                         err = 0;
434                 }
435
436                 mutex_unlock(&cpufreq_governor_mutex);
437         }
438 out:
439         return err;
440 }
441
442
443 /**
444  * cpufreq_per_cpu_attr_read() / show_##file_name() -
445  * print out cpufreq information
446  *
447  * Write out information from cpufreq_driver->policy[cpu]; object must be
448  * "unsigned int".
449  */
450
451 #define show_one(file_name, object)                     \
452 static ssize_t show_##file_name                         \
453 (struct cpufreq_policy *policy, char *buf)              \
454 {                                                       \
455         return sprintf(buf, "%u\n", policy->object);    \
456 }
457
458 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
459 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
460 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
461 show_one(scaling_min_freq, min);
462 show_one(scaling_max_freq, max);
463 show_one(scaling_cur_freq, cur);
464
465 static int __cpufreq_set_policy(struct cpufreq_policy *data,
466                                 struct cpufreq_policy *policy);
467
468 /**
469  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
470  */
471 #define store_one(file_name, object)                    \
472 static ssize_t store_##file_name                                        \
473 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
474 {                                                                       \
475         unsigned int ret = -EINVAL;                                     \
476         struct cpufreq_policy new_policy;                               \
477                                                                         \
478         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
479         if (ret)                                                        \
480                 return -EINVAL;                                         \
481                                                                         \
482         ret = sscanf(buf, "%u", &new_policy.object);                    \
483         if (ret != 1)                                                   \
484                 return -EINVAL;                                         \
485                                                                         \
486         ret = __cpufreq_set_policy(policy, &new_policy);                \
487         policy->user_policy.object = policy->object;                    \
488                                                                         \
489         return ret ? ret : count;                                       \
490 }
491
492 store_one(scaling_min_freq, min);
493 store_one(scaling_max_freq, max);
494
495 /**
496  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
497  */
498 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
499                                         char *buf)
500 {
501         unsigned int cur_freq = __cpufreq_get(policy->cpu);
502         if (!cur_freq)
503                 return sprintf(buf, "<unknown>");
504         return sprintf(buf, "%u\n", cur_freq);
505 }
506
507
508 /**
509  * show_scaling_governor - show the current policy for the specified CPU
510  */
511 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
512 {
513         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
514                 return sprintf(buf, "powersave\n");
515         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
516                 return sprintf(buf, "performance\n");
517         else if (policy->governor)
518                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n",
519                                 policy->governor->name);
520         return -EINVAL;
521 }
522
523
524 /**
525  * store_scaling_governor - store policy for the specified CPU
526  */
527 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
528                                         const char *buf, size_t count)
529 {
530         unsigned int ret = -EINVAL;
531         char    str_governor[16];
532         struct cpufreq_policy new_policy;
533
534         ret = cpufreq_get_policy(&new_policy, policy->cpu);
535         if (ret)
536                 return ret;
537
538         ret = sscanf(buf, "%15s", str_governor);
539         if (ret != 1)
540                 return -EINVAL;
541
542         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
543                                                 &new_policy.governor))
544                 return -EINVAL;
545
546         /* Do not use cpufreq_set_policy here or the user_policy.max
547            will be wrongly overridden */
548         ret = __cpufreq_set_policy(policy, &new_policy);
549
550         policy->user_policy.policy = policy->policy;
551         policy->user_policy.governor = policy->governor;
552
553         if (ret)
554                 return ret;
555         else
556                 return count;
557 }
558
559 /**
560  * show_scaling_driver - show the cpufreq driver currently loaded
561  */
562 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
563 {
564         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
565 }
566
567 /**
568  * show_scaling_available_governors - show the available CPUfreq governors
569  */
570 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
571                                                 char *buf)
572 {
573         ssize_t i = 0;
574         struct cpufreq_governor *t;
575
576         if (!cpufreq_driver->target) {
577                 i += sprintf(buf, "performance powersave");
578                 goto out;
579         }
580
581         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
582                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
583                     - (CPUFREQ_NAME_LEN + 2)))
584                         goto out;
585                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
586         }
587 out:
588         i += sprintf(&buf[i], "\n");
589         return i;
590 }
591
592 static ssize_t show_cpus(const struct cpumask *mask, char *buf)
593 {
594         ssize_t i = 0;
595         unsigned int cpu;
596
597         for_each_cpu(cpu, mask) {
598                 if (i)
599                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
600                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
601                 if (i >= (PAGE_SIZE - 5))
602                         break;
603         }
604         i += sprintf(&buf[i], "\n");
605         return i;
606 }
607
608 /**
609  * show_related_cpus - show the CPUs affected by each transition even if
610  * hw coordination is in use
611  */
612 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
613 {
614         if (cpumask_empty(policy->related_cpus))
615                 return show_cpus(policy->cpus, buf);
616         return show_cpus(policy->related_cpus, buf);
617 }
618
619 /**
620  * show_affected_cpus - show the CPUs affected by each transition
621  */
622 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
623 {
624         return show_cpus(policy->cpus, buf);
625 }
626
627 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
628                                         const char *buf, size_t count)
629 {
630         unsigned int freq = 0;
631         unsigned int ret;
632
633         if (!policy->governor || !policy->governor->store_setspeed)
634                 return -EINVAL;
635
636         ret = sscanf(buf, "%u", &freq);
637         if (ret != 1)
638                 return -EINVAL;
639
640         policy->governor->store_setspeed(policy, freq);
641
642         return count;
643 }
644
645 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
646 {
647         if (!policy->governor || !policy->governor->show_setspeed)
648                 return sprintf(buf, "<unsupported>\n");
649
650         return policy->governor->show_setspeed(policy, buf);
651 }
652
653 /**
654  * show_scaling_driver - show the current cpufreq HW/BIOS limitation
655  */
656 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
657 {
658         unsigned int limit;
659         int ret;
660         if (cpufreq_driver->bios_limit) {
661                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
662                 if (!ret)
663                         return sprintf(buf, "%u\n", limit);
664         }
665         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
666 }
667
668 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
669 cpufreq_freq_attr_ro(cpuinfo_min_freq);
670 cpufreq_freq_attr_ro(cpuinfo_max_freq);
671 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
672 cpufreq_freq_attr_ro(scaling_available_governors);
673 cpufreq_freq_attr_ro(scaling_driver);
674 cpufreq_freq_attr_ro(scaling_cur_freq);
675 cpufreq_freq_attr_ro(bios_limit);
676 cpufreq_freq_attr_ro(related_cpus);
677 cpufreq_freq_attr_ro(affected_cpus);
678 cpufreq_freq_attr_rw(scaling_min_freq);
679 cpufreq_freq_attr_rw(scaling_max_freq);
680 cpufreq_freq_attr_rw(scaling_governor);
681 cpufreq_freq_attr_rw(scaling_setspeed);
682
683 static struct attribute *default_attrs[] = {
684         &cpuinfo_min_freq.attr,
685         &cpuinfo_max_freq.attr,
686         &cpuinfo_transition_latency.attr,
687         &scaling_min_freq.attr,
688         &scaling_max_freq.attr,
689         &affected_cpus.attr,
690         &related_cpus.attr,
691         &scaling_governor.attr,
692         &scaling_driver.attr,
693         &scaling_available_governors.attr,
694         &scaling_setspeed.attr,
695         NULL
696 };
697
698 struct kobject *cpufreq_global_kobject;
699 EXPORT_SYMBOL(cpufreq_global_kobject);
700
701 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
702 #define to_attr(a) container_of(a, struct freq_attr, attr)
703
704 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
705 {
706         struct cpufreq_policy *policy = to_policy(kobj);
707         struct freq_attr *fattr = to_attr(attr);
708         ssize_t ret = -EINVAL;
709         policy = cpufreq_cpu_get(policy->cpu);
710         if (!policy)
711                 goto no_policy;
712
713         if (lock_policy_rwsem_read(policy->cpu) < 0)
714                 goto fail;
715
716         if (fattr->show)
717                 ret = fattr->show(policy, buf);
718         else
719                 ret = -EIO;
720
721         unlock_policy_rwsem_read(policy->cpu);
722 fail:
723         cpufreq_cpu_put(policy);
724 no_policy:
725         return ret;
726 }
727
728 static ssize_t store(struct kobject *kobj, struct attribute *attr,
729                      const char *buf, size_t count)
730 {
731         struct cpufreq_policy *policy = to_policy(kobj);
732         struct freq_attr *fattr = to_attr(attr);
733         ssize_t ret = -EINVAL;
734         policy = cpufreq_cpu_get(policy->cpu);
735         if (!policy)
736                 goto no_policy;
737
738         if (lock_policy_rwsem_write(policy->cpu) < 0)
739                 goto fail;
740
741         if (fattr->store)
742                 ret = fattr->store(policy, buf, count);
743         else
744                 ret = -EIO;
745
746         unlock_policy_rwsem_write(policy->cpu);
747 fail:
748         cpufreq_cpu_put(policy);
749 no_policy:
750         return ret;
751 }
752
753 static void cpufreq_sysfs_release(struct kobject *kobj)
754 {
755         struct cpufreq_policy *policy = to_policy(kobj);
756         dprintk("last reference is dropped\n");
757         complete(&policy->kobj_unregister);
758 }
759
760 static const struct sysfs_ops sysfs_ops = {
761         .show   = show,
762         .store  = store,
763 };
764
765 static struct kobj_type ktype_cpufreq = {
766         .sysfs_ops      = &sysfs_ops,
767         .default_attrs  = default_attrs,
768         .release        = cpufreq_sysfs_release,
769 };
770
771 /*
772  * Returns:
773  *   Negative: Failure
774  *   0:        Success
775  *   Positive: When we have a managed CPU and the sysfs got symlinked
776  */
777 static int cpufreq_add_dev_policy(unsigned int cpu,
778                                   struct cpufreq_policy *policy,
779                                   struct sys_device *sys_dev)
780 {
781         int ret = 0;
782 #ifdef CONFIG_SMP
783         unsigned long flags;
784         unsigned int j;
785 #ifdef CONFIG_HOTPLUG_CPU
786         struct cpufreq_governor *gov;
787
788         gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
789         if (gov) {
790                 policy->governor = gov;
791                 dprintk("Restoring governor %s for cpu %d\n",
792                        policy->governor->name, cpu);
793         }
794 #endif
795
796         for_each_cpu(j, policy->cpus) {
797                 struct cpufreq_policy *managed_policy;
798
799                 if (cpu == j)
800                         continue;
801
802                 /* Check for existing affected CPUs.
803                  * They may not be aware of it due to CPU Hotplug.
804                  * cpufreq_cpu_put is called when the device is removed
805                  * in __cpufreq_remove_dev()
806                  */
807                 managed_policy = cpufreq_cpu_get(j);
808                 if (unlikely(managed_policy)) {
809
810                         /* Set proper policy_cpu */
811                         unlock_policy_rwsem_write(cpu);
812                         per_cpu(cpufreq_policy_cpu, cpu) = managed_policy->cpu;
813
814                         if (lock_policy_rwsem_write(cpu) < 0) {
815                                 /* Should not go through policy unlock path */
816                                 if (cpufreq_driver->exit)
817                                         cpufreq_driver->exit(policy);
818                                 cpufreq_cpu_put(managed_policy);
819                                 return -EBUSY;
820                         }
821
822                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
823                         cpumask_copy(managed_policy->cpus, policy->cpus);
824                         per_cpu(cpufreq_cpu_data, cpu) = managed_policy;
825                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
826
827                         dprintk("CPU already managed, adding link\n");
828                         ret = sysfs_create_link(&sys_dev->kobj,
829                                                 &managed_policy->kobj,
830                                                 "cpufreq");
831                         if (ret)
832                                 cpufreq_cpu_put(managed_policy);
833                         /*
834                          * Success. We only needed to be added to the mask.
835                          * Call driver->exit() because only the cpu parent of
836                          * the kobj needed to call init().
837                          */
838                         if (cpufreq_driver->exit)
839                                 cpufreq_driver->exit(policy);
840
841                         if (!ret)
842                                 return 1;
843                         else
844                                 return ret;
845                 }
846         }
847 #endif
848         return ret;
849 }
850
851
852 /* symlink affected CPUs */
853 static int cpufreq_add_dev_symlink(unsigned int cpu,
854                                    struct cpufreq_policy *policy)
855 {
856         unsigned int j;
857         int ret = 0;
858
859         for_each_cpu(j, policy->cpus) {
860                 struct cpufreq_policy *managed_policy;
861                 struct sys_device *cpu_sys_dev;
862
863                 if (j == cpu)
864                         continue;
865                 if (!cpu_online(j))
866                         continue;
867
868                 dprintk("CPU %u already managed, adding link\n", j);
869                 managed_policy = cpufreq_cpu_get(cpu);
870                 cpu_sys_dev = get_cpu_sysdev(j);
871                 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
872                                         "cpufreq");
873                 if (ret) {
874                         cpufreq_cpu_put(managed_policy);
875                         return ret;
876                 }
877         }
878         return ret;
879 }
880
881 static int cpufreq_add_dev_interface(unsigned int cpu,
882                                      struct cpufreq_policy *policy,
883                                      struct sys_device *sys_dev)
884 {
885         struct cpufreq_policy new_policy;
886         struct freq_attr **drv_attr;
887         unsigned long flags;
888         int ret = 0;
889         unsigned int j;
890
891         /* prepare interface data */
892         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
893                                    &sys_dev->kobj, "cpufreq");
894         if (ret)
895                 return ret;
896
897         /* set up files for this cpu device */
898         drv_attr = cpufreq_driver->attr;
899         while ((drv_attr) && (*drv_attr)) {
900                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
901                 if (ret)
902                         goto err_out_kobj_put;
903                 drv_attr++;
904         }
905         if (cpufreq_driver->get) {
906                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
907                 if (ret)
908                         goto err_out_kobj_put;
909         }
910         if (cpufreq_driver->target) {
911                 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
912                 if (ret)
913                         goto err_out_kobj_put;
914         }
915         if (cpufreq_driver->bios_limit) {
916                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
917                 if (ret)
918                         goto err_out_kobj_put;
919         }
920
921         spin_lock_irqsave(&cpufreq_driver_lock, flags);
922         for_each_cpu(j, policy->cpus) {
923                 if (!cpu_online(j))
924                         continue;
925                 per_cpu(cpufreq_cpu_data, j) = policy;
926                 per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
927         }
928         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
929
930         ret = cpufreq_add_dev_symlink(cpu, policy);
931         if (ret)
932                 goto err_out_kobj_put;
933
934         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
935         /* assure that the starting sequence is run in __cpufreq_set_policy */
936         policy->governor = NULL;
937
938         /* set default policy */
939         ret = __cpufreq_set_policy(policy, &new_policy);
940         policy->user_policy.policy = policy->policy;
941         policy->user_policy.governor = policy->governor;
942
943         if (ret) {
944                 dprintk("setting policy failed\n");
945                 if (cpufreq_driver->exit)
946                         cpufreq_driver->exit(policy);
947         }
948         return ret;
949
950 err_out_kobj_put:
951         kobject_put(&policy->kobj);
952         wait_for_completion(&policy->kobj_unregister);
953         return ret;
954 }
955
956
957 /**
958  * cpufreq_add_dev - add a CPU device
959  *
960  * Adds the cpufreq interface for a CPU device.
961  *
962  * The Oracle says: try running cpufreq registration/unregistration concurrently
963  * with with cpu hotplugging and all hell will break loose. Tried to clean this
964  * mess up, but more thorough testing is needed. - Mathieu
965  */
966 static int cpufreq_add_dev(struct sys_device *sys_dev)
967 {
968         unsigned int cpu = sys_dev->id;
969         int ret = 0, found = 0;
970         struct cpufreq_policy *policy;
971         unsigned long flags;
972         unsigned int j;
973 #ifdef CONFIG_HOTPLUG_CPU
974         int sibling;
975 #endif
976
977         if (cpu_is_offline(cpu))
978                 return 0;
979
980         cpufreq_debug_disable_ratelimit();
981         dprintk("adding CPU %u\n", cpu);
982
983 #ifdef CONFIG_SMP
984         /* check whether a different CPU already registered this
985          * CPU because it is in the same boat. */
986         policy = cpufreq_cpu_get(cpu);
987         if (unlikely(policy)) {
988                 cpufreq_cpu_put(policy);
989                 cpufreq_debug_enable_ratelimit();
990                 return 0;
991         }
992 #endif
993
994         if (!try_module_get(cpufreq_driver->owner)) {
995                 ret = -EINVAL;
996                 goto module_out;
997         }
998
999         ret = -ENOMEM;
1000         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
1001         if (!policy)
1002                 goto nomem_out;
1003
1004         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1005                 goto err_free_policy;
1006
1007         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1008                 goto err_free_cpumask;
1009
1010         policy->cpu = cpu;
1011         cpumask_copy(policy->cpus, cpumask_of(cpu));
1012
1013         /* Initially set CPU itself as the policy_cpu */
1014         per_cpu(cpufreq_policy_cpu, cpu) = cpu;
1015         ret = (lock_policy_rwsem_write(cpu) < 0);
1016         WARN_ON(ret);
1017
1018         init_completion(&policy->kobj_unregister);
1019         INIT_WORK(&policy->update, handle_update);
1020
1021         /* Set governor before ->init, so that driver could check it */
1022 #ifdef CONFIG_HOTPLUG_CPU
1023         for_each_online_cpu(sibling) {
1024                 struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
1025                 if (cp && cp->governor &&
1026                     (cpumask_test_cpu(cpu, cp->related_cpus))) {
1027                         policy->governor = cp->governor;
1028                         found = 1;
1029                         break;
1030                 }
1031         }
1032 #endif
1033         if (!found)
1034                 policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
1035         /* call driver. From then on the cpufreq must be able
1036          * to accept all calls to ->verify and ->setpolicy for this CPU
1037          */
1038         ret = cpufreq_driver->init(policy);
1039         if (ret) {
1040                 dprintk("initialization failed\n");
1041                 goto err_unlock_policy;
1042         }
1043         policy->user_policy.min = policy->min;
1044         policy->user_policy.max = policy->max;
1045
1046         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1047                                      CPUFREQ_START, policy);
1048
1049         ret = cpufreq_add_dev_policy(cpu, policy, sys_dev);
1050         if (ret) {
1051                 if (ret > 0)
1052                         /* This is a managed cpu, symlink created,
1053                            exit with 0 */
1054                         ret = 0;
1055                 goto err_unlock_policy;
1056         }
1057
1058         ret = cpufreq_add_dev_interface(cpu, policy, sys_dev);
1059         if (ret)
1060                 goto err_out_unregister;
1061
1062         unlock_policy_rwsem_write(cpu);
1063
1064         kobject_uevent(&policy->kobj, KOBJ_ADD);
1065         module_put(cpufreq_driver->owner);
1066         dprintk("initialization complete\n");
1067         cpufreq_debug_enable_ratelimit();
1068
1069         return 0;
1070
1071
1072 err_out_unregister:
1073         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1074         for_each_cpu(j, policy->cpus)
1075                 per_cpu(cpufreq_cpu_data, j) = NULL;
1076         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1077
1078         kobject_put(&policy->kobj);
1079         wait_for_completion(&policy->kobj_unregister);
1080
1081 err_unlock_policy:
1082         unlock_policy_rwsem_write(cpu);
1083         free_cpumask_var(policy->related_cpus);
1084 err_free_cpumask:
1085         free_cpumask_var(policy->cpus);
1086 err_free_policy:
1087         kfree(policy);
1088 nomem_out:
1089         module_put(cpufreq_driver->owner);
1090 module_out:
1091         cpufreq_debug_enable_ratelimit();
1092         return ret;
1093 }
1094
1095
1096 /**
1097  * __cpufreq_remove_dev - remove a CPU device
1098  *
1099  * Removes the cpufreq interface for a CPU device.
1100  * Caller should already have policy_rwsem in write mode for this CPU.
1101  * This routine frees the rwsem before returning.
1102  */
1103 static int __cpufreq_remove_dev(struct sys_device *sys_dev)
1104 {
1105         unsigned int cpu = sys_dev->id;
1106         unsigned long flags;
1107         struct cpufreq_policy *data;
1108         struct kobject *kobj;
1109         struct completion *cmp;
1110 #ifdef CONFIG_SMP
1111         struct sys_device *cpu_sys_dev;
1112         unsigned int j;
1113 #endif
1114
1115         cpufreq_debug_disable_ratelimit();
1116         dprintk("unregistering CPU %u\n", cpu);
1117
1118         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1119         data = per_cpu(cpufreq_cpu_data, cpu);
1120
1121         if (!data) {
1122                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1123                 cpufreq_debug_enable_ratelimit();
1124                 unlock_policy_rwsem_write(cpu);
1125                 return -EINVAL;
1126         }
1127         per_cpu(cpufreq_cpu_data, cpu) = NULL;
1128
1129
1130 #ifdef CONFIG_SMP
1131         /* if this isn't the CPU which is the parent of the kobj, we
1132          * only need to unlink, put and exit
1133          */
1134         if (unlikely(cpu != data->cpu)) {
1135                 dprintk("removing link\n");
1136                 cpumask_clear_cpu(cpu, data->cpus);
1137                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1138                 kobj = &sys_dev->kobj;
1139                 cpufreq_cpu_put(data);
1140                 cpufreq_debug_enable_ratelimit();
1141                 unlock_policy_rwsem_write(cpu);
1142                 sysfs_remove_link(kobj, "cpufreq");
1143                 return 0;
1144         }
1145 #endif
1146
1147 #ifdef CONFIG_SMP
1148
1149 #ifdef CONFIG_HOTPLUG_CPU
1150         strncpy(per_cpu(cpufreq_cpu_governor, cpu), data->governor->name,
1151                         CPUFREQ_NAME_LEN);
1152 #endif
1153
1154         /* if we have other CPUs still registered, we need to unlink them,
1155          * or else wait_for_completion below will lock up. Clean the
1156          * per_cpu(cpufreq_cpu_data) while holding the lock, and remove
1157          * the sysfs links afterwards.
1158          */
1159         if (unlikely(cpumask_weight(data->cpus) > 1)) {
1160                 for_each_cpu(j, data->cpus) {
1161                         if (j == cpu)
1162                                 continue;
1163                         per_cpu(cpufreq_cpu_data, j) = NULL;
1164                 }
1165         }
1166
1167         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1168
1169         if (unlikely(cpumask_weight(data->cpus) > 1)) {
1170                 for_each_cpu(j, data->cpus) {
1171                         if (j == cpu)
1172                                 continue;
1173                         dprintk("removing link for cpu %u\n", j);
1174 #ifdef CONFIG_HOTPLUG_CPU
1175                         strncpy(per_cpu(cpufreq_cpu_governor, j),
1176                                 data->governor->name, CPUFREQ_NAME_LEN);
1177 #endif
1178                         cpu_sys_dev = get_cpu_sysdev(j);
1179                         kobj = &cpu_sys_dev->kobj;
1180                         unlock_policy_rwsem_write(cpu);
1181                         sysfs_remove_link(kobj, "cpufreq");
1182                         lock_policy_rwsem_write(cpu);
1183                         cpufreq_cpu_put(data);
1184                 }
1185         }
1186 #else
1187         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1188 #endif
1189
1190         if (cpufreq_driver->target)
1191                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1192
1193         kobj = &data->kobj;
1194         cmp = &data->kobj_unregister;
1195         unlock_policy_rwsem_write(cpu);
1196         kobject_put(kobj);
1197
1198         /* we need to make sure that the underlying kobj is actually
1199          * not referenced anymore by anybody before we proceed with
1200          * unloading.
1201          */
1202         dprintk("waiting for dropping of refcount\n");
1203         wait_for_completion(cmp);
1204         dprintk("wait complete\n");
1205
1206         lock_policy_rwsem_write(cpu);
1207         if (cpufreq_driver->exit)
1208                 cpufreq_driver->exit(data);
1209         unlock_policy_rwsem_write(cpu);
1210
1211         free_cpumask_var(data->related_cpus);
1212         free_cpumask_var(data->cpus);
1213         kfree(data);
1214         per_cpu(cpufreq_cpu_data, cpu) = NULL;
1215
1216         cpufreq_debug_enable_ratelimit();
1217         return 0;
1218 }
1219
1220
1221 static int cpufreq_remove_dev(struct sys_device *sys_dev)
1222 {
1223         unsigned int cpu = sys_dev->id;
1224         int retval;
1225
1226         if (cpu_is_offline(cpu))
1227                 return 0;
1228
1229         if (unlikely(lock_policy_rwsem_write(cpu)))
1230                 BUG();
1231
1232         retval = __cpufreq_remove_dev(sys_dev);
1233         return retval;
1234 }
1235
1236
1237 static void handle_update(struct work_struct *work)
1238 {
1239         struct cpufreq_policy *policy =
1240                 container_of(work, struct cpufreq_policy, update);
1241         unsigned int cpu = policy->cpu;
1242         dprintk("handle_update for cpu %u called\n", cpu);
1243         cpufreq_update_policy(cpu);
1244 }
1245
1246 /**
1247  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1248  *      @cpu: cpu number
1249  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
1250  *      @new_freq: CPU frequency the CPU actually runs at
1251  *
1252  *      We adjust to current frequency first, and need to clean up later.
1253  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1254  */
1255 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1256                                 unsigned int new_freq)
1257 {
1258         struct cpufreq_freqs freqs;
1259
1260         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
1261                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1262
1263         freqs.cpu = cpu;
1264         freqs.old = old_freq;
1265         freqs.new = new_freq;
1266         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1267         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1268 }
1269
1270
1271 /**
1272  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1273  * @cpu: CPU number
1274  *
1275  * This is the last known freq, without actually getting it from the driver.
1276  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1277  */
1278 unsigned int cpufreq_quick_get(unsigned int cpu)
1279 {
1280         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1281         unsigned int ret_freq = 0;
1282
1283         if (policy) {
1284                 ret_freq = policy->cur;
1285                 cpufreq_cpu_put(policy);
1286         }
1287
1288         return ret_freq;
1289 }
1290 EXPORT_SYMBOL(cpufreq_quick_get);
1291
1292
1293 static unsigned int __cpufreq_get(unsigned int cpu)
1294 {
1295         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1296         unsigned int ret_freq = 0;
1297
1298         if (!cpufreq_driver->get)
1299                 return ret_freq;
1300
1301         ret_freq = cpufreq_driver->get(cpu);
1302
1303         if (ret_freq && policy->cur &&
1304                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1305                 /* verify no discrepancy between actual and
1306                                         saved value exists */
1307                 if (unlikely(ret_freq != policy->cur)) {
1308                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1309                         schedule_work(&policy->update);
1310                 }
1311         }
1312
1313         return ret_freq;
1314 }
1315
1316 /**
1317  * cpufreq_get - get the current CPU frequency (in kHz)
1318  * @cpu: CPU number
1319  *
1320  * Get the CPU current (static) CPU frequency
1321  */
1322 unsigned int cpufreq_get(unsigned int cpu)
1323 {
1324         unsigned int ret_freq = 0;
1325         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1326
1327         if (!policy)
1328                 goto out;
1329
1330         if (unlikely(lock_policy_rwsem_read(cpu)))
1331                 goto out_policy;
1332
1333         ret_freq = __cpufreq_get(cpu);
1334
1335         unlock_policy_rwsem_read(cpu);
1336
1337 out_policy:
1338         cpufreq_cpu_put(policy);
1339 out:
1340         return ret_freq;
1341 }
1342 EXPORT_SYMBOL(cpufreq_get);
1343
1344 static struct sysdev_driver cpufreq_sysdev_driver = {
1345         .add            = cpufreq_add_dev,
1346         .remove         = cpufreq_remove_dev,
1347 };
1348
1349
1350 /**
1351  * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1352  *
1353  * This function is only executed for the boot processor.  The other CPUs
1354  * have been put offline by means of CPU hotplug.
1355  */
1356 static int cpufreq_bp_suspend(void)
1357 {
1358         int ret = 0;
1359
1360         int cpu = smp_processor_id();
1361         struct cpufreq_policy *cpu_policy;
1362
1363         dprintk("suspending cpu %u\n", cpu);
1364
1365         /* If there's no policy for the boot CPU, we have nothing to do. */
1366         cpu_policy = cpufreq_cpu_get(cpu);
1367         if (!cpu_policy)
1368                 return 0;
1369
1370         if (cpufreq_driver->suspend) {
1371                 ret = cpufreq_driver->suspend(cpu_policy);
1372                 if (ret)
1373                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1374                                         "step on CPU %u\n", cpu_policy->cpu);
1375         }
1376
1377         cpufreq_cpu_put(cpu_policy);
1378         return ret;
1379 }
1380
1381 /**
1382  * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1383  *
1384  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1385  *      2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1386  *          restored. It will verify that the current freq is in sync with
1387  *          what we believe it to be. This is a bit later than when it
1388  *          should be, but nonethteless it's better than calling
1389  *          cpufreq_driver->get() here which might re-enable interrupts...
1390  *
1391  * This function is only executed for the boot CPU.  The other CPUs have not
1392  * been turned on yet.
1393  */
1394 static void cpufreq_bp_resume(void)
1395 {
1396         int ret = 0;
1397
1398         int cpu = smp_processor_id();
1399         struct cpufreq_policy *cpu_policy;
1400
1401         dprintk("resuming cpu %u\n", cpu);
1402
1403         /* If there's no policy for the boot CPU, we have nothing to do. */
1404         cpu_policy = cpufreq_cpu_get(cpu);
1405         if (!cpu_policy)
1406                 return;
1407
1408         if (cpufreq_driver->resume) {
1409                 ret = cpufreq_driver->resume(cpu_policy);
1410                 if (ret) {
1411                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1412                                         "step on CPU %u\n", cpu_policy->cpu);
1413                         goto fail;
1414                 }
1415         }
1416
1417         schedule_work(&cpu_policy->update);
1418
1419 fail:
1420         cpufreq_cpu_put(cpu_policy);
1421 }
1422
1423 static struct syscore_ops cpufreq_syscore_ops = {
1424         .suspend        = cpufreq_bp_suspend,
1425         .resume         = cpufreq_bp_resume,
1426 };
1427
1428
1429 /*********************************************************************
1430  *                     NOTIFIER LISTS INTERFACE                      *
1431  *********************************************************************/
1432
1433 /**
1434  *      cpufreq_register_notifier - register a driver with cpufreq
1435  *      @nb: notifier function to register
1436  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1437  *
1438  *      Add a driver to one of two lists: either a list of drivers that
1439  *      are notified about clock rate changes (once before and once after
1440  *      the transition), or a list of drivers that are notified about
1441  *      changes in cpufreq policy.
1442  *
1443  *      This function may sleep, and has the same return conditions as
1444  *      blocking_notifier_chain_register.
1445  */
1446 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1447 {
1448         int ret;
1449
1450         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1451
1452         switch (list) {
1453         case CPUFREQ_TRANSITION_NOTIFIER:
1454                 ret = srcu_notifier_chain_register(
1455                                 &cpufreq_transition_notifier_list, nb);
1456                 break;
1457         case CPUFREQ_POLICY_NOTIFIER:
1458                 ret = blocking_notifier_chain_register(
1459                                 &cpufreq_policy_notifier_list, nb);
1460                 break;
1461         default:
1462                 ret = -EINVAL;
1463         }
1464
1465         return ret;
1466 }
1467 EXPORT_SYMBOL(cpufreq_register_notifier);
1468
1469
1470 /**
1471  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1472  *      @nb: notifier block to be unregistered
1473  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1474  *
1475  *      Remove a driver from the CPU frequency notifier list.
1476  *
1477  *      This function may sleep, and has the same return conditions as
1478  *      blocking_notifier_chain_unregister.
1479  */
1480 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1481 {
1482         int ret;
1483
1484         switch (list) {
1485         case CPUFREQ_TRANSITION_NOTIFIER:
1486                 ret = srcu_notifier_chain_unregister(
1487                                 &cpufreq_transition_notifier_list, nb);
1488                 break;
1489         case CPUFREQ_POLICY_NOTIFIER:
1490                 ret = blocking_notifier_chain_unregister(
1491                                 &cpufreq_policy_notifier_list, nb);
1492                 break;
1493         default:
1494                 ret = -EINVAL;
1495         }
1496
1497         return ret;
1498 }
1499 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1500
1501
1502 /*********************************************************************
1503  *                              GOVERNORS                            *
1504  *********************************************************************/
1505
1506
1507 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1508                             unsigned int target_freq,
1509                             unsigned int relation)
1510 {
1511         int retval = -EINVAL;
1512
1513         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1514                 target_freq, relation);
1515         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1516                 retval = cpufreq_driver->target(policy, target_freq, relation);
1517
1518         return retval;
1519 }
1520 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1521
1522 int cpufreq_driver_target(struct cpufreq_policy *policy,
1523                           unsigned int target_freq,
1524                           unsigned int relation)
1525 {
1526         int ret = -EINVAL;
1527
1528         policy = cpufreq_cpu_get(policy->cpu);
1529         if (!policy)
1530                 goto no_policy;
1531
1532         if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1533                 goto fail;
1534
1535         ret = __cpufreq_driver_target(policy, target_freq, relation);
1536
1537         unlock_policy_rwsem_write(policy->cpu);
1538
1539 fail:
1540         cpufreq_cpu_put(policy);
1541 no_policy:
1542         return ret;
1543 }
1544 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1545
1546 int __cpufreq_driver_getavg(struct cpufreq_policy *policy, unsigned int cpu)
1547 {
1548         int ret = 0;
1549
1550         policy = cpufreq_cpu_get(policy->cpu);
1551         if (!policy)
1552                 return -EINVAL;
1553
1554         if (cpu_online(cpu) && cpufreq_driver->getavg)
1555                 ret = cpufreq_driver->getavg(policy, cpu);
1556
1557         cpufreq_cpu_put(policy);
1558         return ret;
1559 }
1560 EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1561
1562 /*
1563  * when "event" is CPUFREQ_GOV_LIMITS
1564  */
1565
1566 static int __cpufreq_governor(struct cpufreq_policy *policy,
1567                                         unsigned int event)
1568 {
1569         int ret;
1570
1571         /* Only must be defined when default governor is known to have latency
1572            restrictions, like e.g. conservative or ondemand.
1573            That this is the case is already ensured in Kconfig
1574         */
1575 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1576         struct cpufreq_governor *gov = &cpufreq_gov_performance;
1577 #else
1578         struct cpufreq_governor *gov = NULL;
1579 #endif
1580
1581         if (policy->governor->max_transition_latency &&
1582             policy->cpuinfo.transition_latency >
1583             policy->governor->max_transition_latency) {
1584                 if (!gov)
1585                         return -EINVAL;
1586                 else {
1587                         printk(KERN_WARNING "%s governor failed, too long"
1588                                " transition latency of HW, fallback"
1589                                " to %s governor\n",
1590                                policy->governor->name,
1591                                gov->name);
1592                         policy->governor = gov;
1593                 }
1594         }
1595
1596         if (!try_module_get(policy->governor->owner))
1597                 return -EINVAL;
1598
1599         dprintk("__cpufreq_governor for CPU %u, event %u\n",
1600                                                 policy->cpu, event);
1601         ret = policy->governor->governor(policy, event);
1602
1603         /* we keep one module reference alive for
1604                         each CPU governed by this CPU */
1605         if ((event != CPUFREQ_GOV_START) || ret)
1606                 module_put(policy->governor->owner);
1607         if ((event == CPUFREQ_GOV_STOP) && !ret)
1608                 module_put(policy->governor->owner);
1609
1610         return ret;
1611 }
1612
1613
1614 int cpufreq_register_governor(struct cpufreq_governor *governor)
1615 {
1616         int err;
1617
1618         if (!governor)
1619                 return -EINVAL;
1620
1621         mutex_lock(&cpufreq_governor_mutex);
1622
1623         err = -EBUSY;
1624         if (__find_governor(governor->name) == NULL) {
1625                 err = 0;
1626                 list_add(&governor->governor_list, &cpufreq_governor_list);
1627         }
1628
1629         mutex_unlock(&cpufreq_governor_mutex);
1630         return err;
1631 }
1632 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1633
1634
1635 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1636 {
1637 #ifdef CONFIG_HOTPLUG_CPU
1638         int cpu;
1639 #endif
1640
1641         if (!governor)
1642                 return;
1643
1644 #ifdef CONFIG_HOTPLUG_CPU
1645         for_each_present_cpu(cpu) {
1646                 if (cpu_online(cpu))
1647                         continue;
1648                 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1649                         strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1650         }
1651 #endif
1652
1653         mutex_lock(&cpufreq_governor_mutex);
1654         list_del(&governor->governor_list);
1655         mutex_unlock(&cpufreq_governor_mutex);
1656         return;
1657 }
1658 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1659
1660
1661
1662 /*********************************************************************
1663  *                          POLICY INTERFACE                         *
1664  *********************************************************************/
1665
1666 /**
1667  * cpufreq_get_policy - get the current cpufreq_policy
1668  * @policy: struct cpufreq_policy into which the current cpufreq_policy
1669  *      is written
1670  *
1671  * Reads the current cpufreq policy.
1672  */
1673 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1674 {
1675         struct cpufreq_policy *cpu_policy;
1676         if (!policy)
1677                 return -EINVAL;
1678
1679         cpu_policy = cpufreq_cpu_get(cpu);
1680         if (!cpu_policy)
1681                 return -EINVAL;
1682
1683         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1684
1685         cpufreq_cpu_put(cpu_policy);
1686         return 0;
1687 }
1688 EXPORT_SYMBOL(cpufreq_get_policy);
1689
1690
1691 /*
1692  * data   : current policy.
1693  * policy : policy to be set.
1694  */
1695 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1696                                 struct cpufreq_policy *policy)
1697 {
1698         int ret = 0;
1699
1700         cpufreq_debug_disable_ratelimit();
1701         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1702                 policy->min, policy->max);
1703
1704         memcpy(&policy->cpuinfo, &data->cpuinfo,
1705                                 sizeof(struct cpufreq_cpuinfo));
1706
1707         if (policy->min > data->max || policy->max < data->min) {
1708                 ret = -EINVAL;
1709                 goto error_out;
1710         }
1711
1712         /* verify the cpu speed can be set within this limit */
1713         ret = cpufreq_driver->verify(policy);
1714         if (ret)
1715                 goto error_out;
1716
1717         /* adjust if necessary - all reasons */
1718         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1719                         CPUFREQ_ADJUST, policy);
1720
1721         /* adjust if necessary - hardware incompatibility*/
1722         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1723                         CPUFREQ_INCOMPATIBLE, policy);
1724
1725         /* verify the cpu speed can be set within this limit,
1726            which might be different to the first one */
1727         ret = cpufreq_driver->verify(policy);
1728         if (ret)
1729                 goto error_out;
1730
1731         /* notification of the new policy */
1732         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1733                         CPUFREQ_NOTIFY, policy);
1734
1735         data->min = policy->min;
1736         data->max = policy->max;
1737
1738         dprintk("new min and max freqs are %u - %u kHz\n",
1739                                         data->min, data->max);
1740
1741         if (cpufreq_driver->setpolicy) {
1742                 data->policy = policy->policy;
1743                 dprintk("setting range\n");
1744                 ret = cpufreq_driver->setpolicy(policy);
1745         } else {
1746                 if (policy->governor != data->governor) {
1747                         /* save old, working values */
1748                         struct cpufreq_governor *old_gov = data->governor;
1749
1750                         dprintk("governor switch\n");
1751
1752                         /* end old governor */
1753                         if (data->governor)
1754                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1755
1756                         /* start new governor */
1757                         data->governor = policy->governor;
1758                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1759                                 /* new governor failed, so re-start old one */
1760                                 dprintk("starting governor %s failed\n",
1761                                                         data->governor->name);
1762                                 if (old_gov) {
1763                                         data->governor = old_gov;
1764                                         __cpufreq_governor(data,
1765                                                            CPUFREQ_GOV_START);
1766                                 }
1767                                 ret = -EINVAL;
1768                                 goto error_out;
1769                         }
1770                         /* might be a policy change, too, so fall through */
1771                 }
1772                 dprintk("governor: change or update limits\n");
1773                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1774         }
1775
1776 error_out:
1777         cpufreq_debug_enable_ratelimit();
1778         return ret;
1779 }
1780
1781 /**
1782  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1783  *      @cpu: CPU which shall be re-evaluated
1784  *
1785  *      Useful for policy notifiers which have different necessities
1786  *      at different times.
1787  */
1788 int cpufreq_update_policy(unsigned int cpu)
1789 {
1790         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1791         struct cpufreq_policy policy;
1792         int ret;
1793
1794         if (!data) {
1795                 ret = -ENODEV;
1796                 goto no_policy;
1797         }
1798
1799         if (unlikely(lock_policy_rwsem_write(cpu))) {
1800                 ret = -EINVAL;
1801                 goto fail;
1802         }
1803
1804         dprintk("updating policy for CPU %u\n", cpu);
1805         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1806         policy.min = data->user_policy.min;
1807         policy.max = data->user_policy.max;
1808         policy.policy = data->user_policy.policy;
1809         policy.governor = data->user_policy.governor;
1810
1811         /* BIOS might change freq behind our back
1812           -> ask driver for current freq and notify governors about a change */
1813         if (cpufreq_driver->get) {
1814                 policy.cur = cpufreq_driver->get(cpu);
1815                 if (!data->cur) {
1816                         dprintk("Driver did not initialize current freq");
1817                         data->cur = policy.cur;
1818                 } else {
1819                         if (data->cur != policy.cur)
1820                                 cpufreq_out_of_sync(cpu, data->cur,
1821                                                                 policy.cur);
1822                 }
1823         }
1824
1825         ret = __cpufreq_set_policy(data, &policy);
1826
1827         unlock_policy_rwsem_write(cpu);
1828
1829 fail:
1830         cpufreq_cpu_put(data);
1831 no_policy:
1832         return ret;
1833 }
1834 EXPORT_SYMBOL(cpufreq_update_policy);
1835
1836 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1837                                         unsigned long action, void *hcpu)
1838 {
1839         unsigned int cpu = (unsigned long)hcpu;
1840         struct sys_device *sys_dev;
1841
1842         sys_dev = get_cpu_sysdev(cpu);
1843         if (sys_dev) {
1844                 switch (action) {
1845                 case CPU_ONLINE:
1846                 case CPU_ONLINE_FROZEN:
1847                         cpufreq_add_dev(sys_dev);
1848                         break;
1849                 case CPU_DOWN_PREPARE:
1850                 case CPU_DOWN_PREPARE_FROZEN:
1851                         if (unlikely(lock_policy_rwsem_write(cpu)))
1852                                 BUG();
1853
1854                         __cpufreq_remove_dev(sys_dev);
1855                         break;
1856                 case CPU_DOWN_FAILED:
1857                 case CPU_DOWN_FAILED_FROZEN:
1858                         cpufreq_add_dev(sys_dev);
1859                         break;
1860                 }
1861         }
1862         return NOTIFY_OK;
1863 }
1864
1865 static struct notifier_block __refdata cpufreq_cpu_notifier = {
1866     .notifier_call = cpufreq_cpu_callback,
1867 };
1868
1869 /*********************************************************************
1870  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1871  *********************************************************************/
1872
1873 /**
1874  * cpufreq_register_driver - register a CPU Frequency driver
1875  * @driver_data: A struct cpufreq_driver containing the values#
1876  * submitted by the CPU Frequency driver.
1877  *
1878  *   Registers a CPU Frequency driver to this core code. This code
1879  * returns zero on success, -EBUSY when another driver got here first
1880  * (and isn't unregistered in the meantime).
1881  *
1882  */
1883 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1884 {
1885         unsigned long flags;
1886         int ret;
1887
1888         if (!driver_data || !driver_data->verify || !driver_data->init ||
1889             ((!driver_data->setpolicy) && (!driver_data->target)))
1890                 return -EINVAL;
1891
1892         dprintk("trying to register driver %s\n", driver_data->name);
1893
1894         if (driver_data->setpolicy)
1895                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1896
1897         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1898         if (cpufreq_driver) {
1899                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1900                 return -EBUSY;
1901         }
1902         cpufreq_driver = driver_data;
1903         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1904
1905         ret = sysdev_driver_register(&cpu_sysdev_class,
1906                                         &cpufreq_sysdev_driver);
1907         if (ret)
1908                 goto err_null_driver;
1909
1910         if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1911                 int i;
1912                 ret = -ENODEV;
1913
1914                 /* check for at least one working CPU */
1915                 for (i = 0; i < nr_cpu_ids; i++)
1916                         if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
1917                                 ret = 0;
1918                                 break;
1919                         }
1920
1921                 /* if all ->init() calls failed, unregister */
1922                 if (ret) {
1923                         dprintk("no CPU initialized for driver %s\n",
1924                                                         driver_data->name);
1925                         goto err_sysdev_unreg;
1926                 }
1927         }
1928
1929         register_hotcpu_notifier(&cpufreq_cpu_notifier);
1930         dprintk("driver %s up and running\n", driver_data->name);
1931         cpufreq_debug_enable_ratelimit();
1932
1933         return 0;
1934 err_sysdev_unreg:
1935         sysdev_driver_unregister(&cpu_sysdev_class,
1936                         &cpufreq_sysdev_driver);
1937 err_null_driver:
1938         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1939         cpufreq_driver = NULL;
1940         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1941         return ret;
1942 }
1943 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1944
1945
1946 /**
1947  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1948  *
1949  *    Unregister the current CPUFreq driver. Only call this if you have
1950  * the right to do so, i.e. if you have succeeded in initialising before!
1951  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1952  * currently not initialised.
1953  */
1954 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1955 {
1956         unsigned long flags;
1957
1958         cpufreq_debug_disable_ratelimit();
1959
1960         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1961                 cpufreq_debug_enable_ratelimit();
1962                 return -EINVAL;
1963         }
1964
1965         dprintk("unregistering driver %s\n", driver->name);
1966
1967         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1968         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1969
1970         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1971         cpufreq_driver = NULL;
1972         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1973
1974         return 0;
1975 }
1976 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1977
1978 static int __init cpufreq_core_init(void)
1979 {
1980         int cpu;
1981
1982         for_each_possible_cpu(cpu) {
1983                 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1984                 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1985         }
1986
1987         cpufreq_global_kobject = kobject_create_and_add("cpufreq",
1988                                                 &cpu_sysdev_class.kset.kobj);
1989         BUG_ON(!cpufreq_global_kobject);
1990         register_syscore_ops(&cpufreq_syscore_ops);
1991
1992         return 0;
1993 }
1994 core_initcall(cpufreq_core_init);