]> rtime.felk.cvut.cz Git - linux-imx.git/blob - drivers/cpufreq/cpufreq.c
cpufreq: Revert incorrect commit 5800043
[linux-imx.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/init.h>
23 #include <linux/notifier.h>
24 #include <linux/cpufreq.h>
25 #include <linux/delay.h>
26 #include <linux/interrupt.h>
27 #include <linux/spinlock.h>
28 #include <linux/device.h>
29 #include <linux/slab.h>
30 #include <linux/cpu.h>
31 #include <linux/completion.h>
32 #include <linux/mutex.h>
33 #include <linux/syscore_ops.h>
34
35 #include <trace/events/power.h>
36
37 /**
38  * The "cpufreq driver" - the arch- or hardware-dependent low
39  * level driver of CPUFreq support, and its spinlock. This lock
40  * also protects the cpufreq_cpu_data array.
41  */
42 static struct cpufreq_driver *cpufreq_driver;
43 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
44 #ifdef CONFIG_HOTPLUG_CPU
45 /* This one keeps track of the previously set governor of a removed CPU */
46 static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
47 #endif
48 static DEFINE_RWLOCK(cpufreq_driver_lock);
49
50 /*
51  * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
52  * all cpufreq/hotplug/workqueue/etc related lock issues.
53  *
54  * The rules for this semaphore:
55  * - Any routine that wants to read from the policy structure will
56  *   do a down_read on this semaphore.
57  * - Any routine that will write to the policy structure and/or may take away
58  *   the policy altogether (eg. CPU hotplug), will hold this lock in write
59  *   mode before doing so.
60  *
61  * Additional rules:
62  * - Governor routines that can be called in cpufreq hotplug path should not
63  *   take this sem as top level hotplug notifier handler takes this.
64  * - Lock should not be held across
65  *     __cpufreq_governor(data, CPUFREQ_GOV_STOP);
66  */
67 static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
68 static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
69
70 #define lock_policy_rwsem(mode, cpu)                                    \
71 static int lock_policy_rwsem_##mode(int cpu)                            \
72 {                                                                       \
73         int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);              \
74         BUG_ON(policy_cpu == -1);                                       \
75         down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));            \
76                                                                         \
77         return 0;                                                       \
78 }
79
80 lock_policy_rwsem(read, cpu);
81 lock_policy_rwsem(write, cpu);
82
83 #define unlock_policy_rwsem(mode, cpu)                                  \
84 static void unlock_policy_rwsem_##mode(int cpu)                         \
85 {                                                                       \
86         int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);              \
87         BUG_ON(policy_cpu == -1);                                       \
88         up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));              \
89 }
90
91 unlock_policy_rwsem(read, cpu);
92 unlock_policy_rwsem(write, cpu);
93
94 /* internal prototypes */
95 static int __cpufreq_governor(struct cpufreq_policy *policy,
96                 unsigned int event);
97 static unsigned int __cpufreq_get(unsigned int cpu);
98 static void handle_update(struct work_struct *work);
99
100 /**
101  * Two notifier lists: the "policy" list is involved in the
102  * validation process for a new CPU frequency policy; the
103  * "transition" list for kernel code that needs to handle
104  * changes to devices when the CPU clock speed changes.
105  * The mutex locks both lists.
106  */
107 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
108 static struct srcu_notifier_head cpufreq_transition_notifier_list;
109
110 static bool init_cpufreq_transition_notifier_list_called;
111 static int __init init_cpufreq_transition_notifier_list(void)
112 {
113         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
114         init_cpufreq_transition_notifier_list_called = true;
115         return 0;
116 }
117 pure_initcall(init_cpufreq_transition_notifier_list);
118
119 static int off __read_mostly;
120 static int cpufreq_disabled(void)
121 {
122         return off;
123 }
124 void disable_cpufreq(void)
125 {
126         off = 1;
127 }
128 static LIST_HEAD(cpufreq_governor_list);
129 static DEFINE_MUTEX(cpufreq_governor_mutex);
130
131 bool have_governor_per_policy(void)
132 {
133         return cpufreq_driver->have_governor_per_policy;
134 }
135
136 static struct cpufreq_policy *__cpufreq_cpu_get(unsigned int cpu, bool sysfs)
137 {
138         struct cpufreq_policy *data;
139         unsigned long flags;
140
141         if (cpu >= nr_cpu_ids)
142                 goto err_out;
143
144         /* get the cpufreq driver */
145         read_lock_irqsave(&cpufreq_driver_lock, flags);
146
147         if (!cpufreq_driver)
148                 goto err_out_unlock;
149
150         if (!try_module_get(cpufreq_driver->owner))
151                 goto err_out_unlock;
152
153
154         /* get the CPU */
155         data = per_cpu(cpufreq_cpu_data, cpu);
156
157         if (!data)
158                 goto err_out_put_module;
159
160         if (!sysfs && !kobject_get(&data->kobj))
161                 goto err_out_put_module;
162
163         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
164         return data;
165
166 err_out_put_module:
167         module_put(cpufreq_driver->owner);
168 err_out_unlock:
169         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
170 err_out:
171         return NULL;
172 }
173
174 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
175 {
176         if (cpufreq_disabled())
177                 return NULL;
178
179         return __cpufreq_cpu_get(cpu, false);
180 }
181 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
182
183 static struct cpufreq_policy *cpufreq_cpu_get_sysfs(unsigned int cpu)
184 {
185         return __cpufreq_cpu_get(cpu, true);
186 }
187
188 static void __cpufreq_cpu_put(struct cpufreq_policy *data, bool sysfs)
189 {
190         if (!sysfs)
191                 kobject_put(&data->kobj);
192         module_put(cpufreq_driver->owner);
193 }
194
195 void cpufreq_cpu_put(struct cpufreq_policy *data)
196 {
197         if (cpufreq_disabled())
198                 return;
199
200         __cpufreq_cpu_put(data, false);
201 }
202 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
203
204 static void cpufreq_cpu_put_sysfs(struct cpufreq_policy *data)
205 {
206         __cpufreq_cpu_put(data, true);
207 }
208
209 /*********************************************************************
210  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
211  *********************************************************************/
212
213 /**
214  * adjust_jiffies - adjust the system "loops_per_jiffy"
215  *
216  * This function alters the system "loops_per_jiffy" for the clock
217  * speed change. Note that loops_per_jiffy cannot be updated on SMP
218  * systems as each CPU might be scaled differently. So, use the arch
219  * per-CPU loops_per_jiffy value wherever possible.
220  */
221 #ifndef CONFIG_SMP
222 static unsigned long l_p_j_ref;
223 static unsigned int  l_p_j_ref_freq;
224
225 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
226 {
227         if (ci->flags & CPUFREQ_CONST_LOOPS)
228                 return;
229
230         if (!l_p_j_ref_freq) {
231                 l_p_j_ref = loops_per_jiffy;
232                 l_p_j_ref_freq = ci->old;
233                 pr_debug("saving %lu as reference value for loops_per_jiffy; "
234                         "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
235         }
236         if ((val == CPUFREQ_POSTCHANGE  && ci->old != ci->new) ||
237             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
238                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
239                                                                 ci->new);
240                 pr_debug("scaling loops_per_jiffy to %lu "
241                         "for frequency %u kHz\n", loops_per_jiffy, ci->new);
242         }
243 }
244 #else
245 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
246 {
247         return;
248 }
249 #endif
250
251
252 void __cpufreq_notify_transition(struct cpufreq_policy *policy,
253                 struct cpufreq_freqs *freqs, unsigned int state)
254 {
255         BUG_ON(irqs_disabled());
256
257         if (cpufreq_disabled())
258                 return;
259
260         freqs->flags = cpufreq_driver->flags;
261         pr_debug("notification %u of frequency transition to %u kHz\n",
262                 state, freqs->new);
263
264         switch (state) {
265
266         case CPUFREQ_PRECHANGE:
267                 /* detect if the driver reported a value as "old frequency"
268                  * which is not equal to what the cpufreq core thinks is
269                  * "old frequency".
270                  */
271                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
272                         if ((policy) && (policy->cpu == freqs->cpu) &&
273                             (policy->cur) && (policy->cur != freqs->old)) {
274                                 pr_debug("Warning: CPU frequency is"
275                                         " %u, cpufreq assumed %u kHz.\n",
276                                         freqs->old, policy->cur);
277                                 freqs->old = policy->cur;
278                         }
279                 }
280                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
281                                 CPUFREQ_PRECHANGE, freqs);
282                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
283                 break;
284
285         case CPUFREQ_POSTCHANGE:
286                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
287                 pr_debug("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
288                         (unsigned long)freqs->cpu);
289                 trace_cpu_frequency(freqs->new, freqs->cpu);
290                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
291                                 CPUFREQ_POSTCHANGE, freqs);
292                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
293                         policy->cur = freqs->new;
294                 break;
295         }
296 }
297 /**
298  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
299  * on frequency transition.
300  *
301  * This function calls the transition notifiers and the "adjust_jiffies"
302  * function. It is called twice on all CPU frequency changes that have
303  * external effects.
304  */
305 void cpufreq_notify_transition(struct cpufreq_policy *policy,
306                 struct cpufreq_freqs *freqs, unsigned int state)
307 {
308         for_each_cpu(freqs->cpu, policy->cpus)
309                 __cpufreq_notify_transition(policy, freqs, state);
310 }
311 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
312
313
314
315 /*********************************************************************
316  *                          SYSFS INTERFACE                          *
317  *********************************************************************/
318
319 static struct cpufreq_governor *__find_governor(const char *str_governor)
320 {
321         struct cpufreq_governor *t;
322
323         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
324                 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
325                         return t;
326
327         return NULL;
328 }
329
330 /**
331  * cpufreq_parse_governor - parse a governor string
332  */
333 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
334                                 struct cpufreq_governor **governor)
335 {
336         int err = -EINVAL;
337
338         if (!cpufreq_driver)
339                 goto out;
340
341         if (cpufreq_driver->setpolicy) {
342                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
343                         *policy = CPUFREQ_POLICY_PERFORMANCE;
344                         err = 0;
345                 } else if (!strnicmp(str_governor, "powersave",
346                                                 CPUFREQ_NAME_LEN)) {
347                         *policy = CPUFREQ_POLICY_POWERSAVE;
348                         err = 0;
349                 }
350         } else if (cpufreq_driver->target) {
351                 struct cpufreq_governor *t;
352
353                 mutex_lock(&cpufreq_governor_mutex);
354
355                 t = __find_governor(str_governor);
356
357                 if (t == NULL) {
358                         int ret;
359
360                         mutex_unlock(&cpufreq_governor_mutex);
361                         ret = request_module("cpufreq_%s", str_governor);
362                         mutex_lock(&cpufreq_governor_mutex);
363
364                         if (ret == 0)
365                                 t = __find_governor(str_governor);
366                 }
367
368                 if (t != NULL) {
369                         *governor = t;
370                         err = 0;
371                 }
372
373                 mutex_unlock(&cpufreq_governor_mutex);
374         }
375 out:
376         return err;
377 }
378
379
380 /**
381  * cpufreq_per_cpu_attr_read() / show_##file_name() -
382  * print out cpufreq information
383  *
384  * Write out information from cpufreq_driver->policy[cpu]; object must be
385  * "unsigned int".
386  */
387
388 #define show_one(file_name, object)                     \
389 static ssize_t show_##file_name                         \
390 (struct cpufreq_policy *policy, char *buf)              \
391 {                                                       \
392         return sprintf(buf, "%u\n", policy->object);    \
393 }
394
395 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
396 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
397 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
398 show_one(scaling_min_freq, min);
399 show_one(scaling_max_freq, max);
400 show_one(scaling_cur_freq, cur);
401
402 static int __cpufreq_set_policy(struct cpufreq_policy *data,
403                                 struct cpufreq_policy *policy);
404
405 /**
406  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
407  */
408 #define store_one(file_name, object)                    \
409 static ssize_t store_##file_name                                        \
410 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
411 {                                                                       \
412         unsigned int ret;                                               \
413         struct cpufreq_policy new_policy;                               \
414                                                                         \
415         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
416         if (ret)                                                        \
417                 return -EINVAL;                                         \
418                                                                         \
419         ret = sscanf(buf, "%u", &new_policy.object);                    \
420         if (ret != 1)                                                   \
421                 return -EINVAL;                                         \
422                                                                         \
423         ret = __cpufreq_set_policy(policy, &new_policy);                \
424         policy->user_policy.object = policy->object;                    \
425                                                                         \
426         return ret ? ret : count;                                       \
427 }
428
429 store_one(scaling_min_freq, min);
430 store_one(scaling_max_freq, max);
431
432 /**
433  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
434  */
435 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
436                                         char *buf)
437 {
438         unsigned int cur_freq = __cpufreq_get(policy->cpu);
439         if (!cur_freq)
440                 return sprintf(buf, "<unknown>");
441         return sprintf(buf, "%u\n", cur_freq);
442 }
443
444
445 /**
446  * show_scaling_governor - show the current policy for the specified CPU
447  */
448 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
449 {
450         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
451                 return sprintf(buf, "powersave\n");
452         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
453                 return sprintf(buf, "performance\n");
454         else if (policy->governor)
455                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
456                                 policy->governor->name);
457         return -EINVAL;
458 }
459
460
461 /**
462  * store_scaling_governor - store policy for the specified CPU
463  */
464 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
465                                         const char *buf, size_t count)
466 {
467         unsigned int ret;
468         char    str_governor[16];
469         struct cpufreq_policy new_policy;
470
471         ret = cpufreq_get_policy(&new_policy, policy->cpu);
472         if (ret)
473                 return ret;
474
475         ret = sscanf(buf, "%15s", str_governor);
476         if (ret != 1)
477                 return -EINVAL;
478
479         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
480                                                 &new_policy.governor))
481                 return -EINVAL;
482
483         /* Do not use cpufreq_set_policy here or the user_policy.max
484            will be wrongly overridden */
485         ret = __cpufreq_set_policy(policy, &new_policy);
486
487         policy->user_policy.policy = policy->policy;
488         policy->user_policy.governor = policy->governor;
489
490         if (ret)
491                 return ret;
492         else
493                 return count;
494 }
495
496 /**
497  * show_scaling_driver - show the cpufreq driver currently loaded
498  */
499 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
500 {
501         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
502 }
503
504 /**
505  * show_scaling_available_governors - show the available CPUfreq governors
506  */
507 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
508                                                 char *buf)
509 {
510         ssize_t i = 0;
511         struct cpufreq_governor *t;
512
513         if (!cpufreq_driver->target) {
514                 i += sprintf(buf, "performance powersave");
515                 goto out;
516         }
517
518         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
519                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
520                     - (CPUFREQ_NAME_LEN + 2)))
521                         goto out;
522                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
523         }
524 out:
525         i += sprintf(&buf[i], "\n");
526         return i;
527 }
528
529 static ssize_t show_cpus(const struct cpumask *mask, char *buf)
530 {
531         ssize_t i = 0;
532         unsigned int cpu;
533
534         for_each_cpu(cpu, mask) {
535                 if (i)
536                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
537                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
538                 if (i >= (PAGE_SIZE - 5))
539                         break;
540         }
541         i += sprintf(&buf[i], "\n");
542         return i;
543 }
544
545 /**
546  * show_related_cpus - show the CPUs affected by each transition even if
547  * hw coordination is in use
548  */
549 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
550 {
551         return show_cpus(policy->related_cpus, buf);
552 }
553
554 /**
555  * show_affected_cpus - show the CPUs affected by each transition
556  */
557 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
558 {
559         return show_cpus(policy->cpus, buf);
560 }
561
562 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
563                                         const char *buf, size_t count)
564 {
565         unsigned int freq = 0;
566         unsigned int ret;
567
568         if (!policy->governor || !policy->governor->store_setspeed)
569                 return -EINVAL;
570
571         ret = sscanf(buf, "%u", &freq);
572         if (ret != 1)
573                 return -EINVAL;
574
575         policy->governor->store_setspeed(policy, freq);
576
577         return count;
578 }
579
580 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
581 {
582         if (!policy->governor || !policy->governor->show_setspeed)
583                 return sprintf(buf, "<unsupported>\n");
584
585         return policy->governor->show_setspeed(policy, buf);
586 }
587
588 /**
589  * show_bios_limit - show the current cpufreq HW/BIOS limitation
590  */
591 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
592 {
593         unsigned int limit;
594         int ret;
595         if (cpufreq_driver->bios_limit) {
596                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
597                 if (!ret)
598                         return sprintf(buf, "%u\n", limit);
599         }
600         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
601 }
602
603 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
604 cpufreq_freq_attr_ro(cpuinfo_min_freq);
605 cpufreq_freq_attr_ro(cpuinfo_max_freq);
606 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
607 cpufreq_freq_attr_ro(scaling_available_governors);
608 cpufreq_freq_attr_ro(scaling_driver);
609 cpufreq_freq_attr_ro(scaling_cur_freq);
610 cpufreq_freq_attr_ro(bios_limit);
611 cpufreq_freq_attr_ro(related_cpus);
612 cpufreq_freq_attr_ro(affected_cpus);
613 cpufreq_freq_attr_rw(scaling_min_freq);
614 cpufreq_freq_attr_rw(scaling_max_freq);
615 cpufreq_freq_attr_rw(scaling_governor);
616 cpufreq_freq_attr_rw(scaling_setspeed);
617
618 static struct attribute *default_attrs[] = {
619         &cpuinfo_min_freq.attr,
620         &cpuinfo_max_freq.attr,
621         &cpuinfo_transition_latency.attr,
622         &scaling_min_freq.attr,
623         &scaling_max_freq.attr,
624         &affected_cpus.attr,
625         &related_cpus.attr,
626         &scaling_governor.attr,
627         &scaling_driver.attr,
628         &scaling_available_governors.attr,
629         &scaling_setspeed.attr,
630         NULL
631 };
632
633 struct kobject *cpufreq_global_kobject;
634 EXPORT_SYMBOL(cpufreq_global_kobject);
635
636 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
637 #define to_attr(a) container_of(a, struct freq_attr, attr)
638
639 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
640 {
641         struct cpufreq_policy *policy = to_policy(kobj);
642         struct freq_attr *fattr = to_attr(attr);
643         ssize_t ret = -EINVAL;
644         policy = cpufreq_cpu_get_sysfs(policy->cpu);
645         if (!policy)
646                 goto no_policy;
647
648         if (lock_policy_rwsem_read(policy->cpu) < 0)
649                 goto fail;
650
651         if (fattr->show)
652                 ret = fattr->show(policy, buf);
653         else
654                 ret = -EIO;
655
656         unlock_policy_rwsem_read(policy->cpu);
657 fail:
658         cpufreq_cpu_put_sysfs(policy);
659 no_policy:
660         return ret;
661 }
662
663 static ssize_t store(struct kobject *kobj, struct attribute *attr,
664                      const char *buf, size_t count)
665 {
666         struct cpufreq_policy *policy = to_policy(kobj);
667         struct freq_attr *fattr = to_attr(attr);
668         ssize_t ret = -EINVAL;
669         policy = cpufreq_cpu_get_sysfs(policy->cpu);
670         if (!policy)
671                 goto no_policy;
672
673         if (lock_policy_rwsem_write(policy->cpu) < 0)
674                 goto fail;
675
676         if (fattr->store)
677                 ret = fattr->store(policy, buf, count);
678         else
679                 ret = -EIO;
680
681         unlock_policy_rwsem_write(policy->cpu);
682 fail:
683         cpufreq_cpu_put_sysfs(policy);
684 no_policy:
685         return ret;
686 }
687
688 static void cpufreq_sysfs_release(struct kobject *kobj)
689 {
690         struct cpufreq_policy *policy = to_policy(kobj);
691         pr_debug("last reference is dropped\n");
692         complete(&policy->kobj_unregister);
693 }
694
695 static const struct sysfs_ops sysfs_ops = {
696         .show   = show,
697         .store  = store,
698 };
699
700 static struct kobj_type ktype_cpufreq = {
701         .sysfs_ops      = &sysfs_ops,
702         .default_attrs  = default_attrs,
703         .release        = cpufreq_sysfs_release,
704 };
705
706 /* symlink affected CPUs */
707 static int cpufreq_add_dev_symlink(unsigned int cpu,
708                                    struct cpufreq_policy *policy)
709 {
710         unsigned int j;
711         int ret = 0;
712
713         for_each_cpu(j, policy->cpus) {
714                 struct cpufreq_policy *managed_policy;
715                 struct device *cpu_dev;
716
717                 if (j == cpu)
718                         continue;
719
720                 pr_debug("CPU %u already managed, adding link\n", j);
721                 managed_policy = cpufreq_cpu_get(cpu);
722                 cpu_dev = get_cpu_device(j);
723                 ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
724                                         "cpufreq");
725                 if (ret) {
726                         cpufreq_cpu_put(managed_policy);
727                         return ret;
728                 }
729         }
730         return ret;
731 }
732
733 static int cpufreq_add_dev_interface(unsigned int cpu,
734                                      struct cpufreq_policy *policy,
735                                      struct device *dev)
736 {
737         struct cpufreq_policy new_policy;
738         struct freq_attr **drv_attr;
739         unsigned long flags;
740         int ret = 0;
741         unsigned int j;
742
743         /* prepare interface data */
744         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
745                                    &dev->kobj, "cpufreq");
746         if (ret)
747                 return ret;
748
749         /* set up files for this cpu device */
750         drv_attr = cpufreq_driver->attr;
751         while ((drv_attr) && (*drv_attr)) {
752                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
753                 if (ret)
754                         goto err_out_kobj_put;
755                 drv_attr++;
756         }
757         if (cpufreq_driver->get) {
758                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
759                 if (ret)
760                         goto err_out_kobj_put;
761         }
762         if (cpufreq_driver->target) {
763                 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
764                 if (ret)
765                         goto err_out_kobj_put;
766         }
767         if (cpufreq_driver->bios_limit) {
768                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
769                 if (ret)
770                         goto err_out_kobj_put;
771         }
772
773         write_lock_irqsave(&cpufreq_driver_lock, flags);
774         for_each_cpu(j, policy->cpus) {
775                 per_cpu(cpufreq_cpu_data, j) = policy;
776                 per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
777         }
778         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
779
780         ret = cpufreq_add_dev_symlink(cpu, policy);
781         if (ret)
782                 goto err_out_kobj_put;
783
784         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
785         /* assure that the starting sequence is run in __cpufreq_set_policy */
786         policy->governor = NULL;
787
788         /* set default policy */
789         ret = __cpufreq_set_policy(policy, &new_policy);
790         policy->user_policy.policy = policy->policy;
791         policy->user_policy.governor = policy->governor;
792
793         if (ret) {
794                 pr_debug("setting policy failed\n");
795                 if (cpufreq_driver->exit)
796                         cpufreq_driver->exit(policy);
797         }
798         return ret;
799
800 err_out_kobj_put:
801         kobject_put(&policy->kobj);
802         wait_for_completion(&policy->kobj_unregister);
803         return ret;
804 }
805
806 #ifdef CONFIG_HOTPLUG_CPU
807 static int cpufreq_add_policy_cpu(unsigned int cpu, unsigned int sibling,
808                                   struct device *dev)
809 {
810         struct cpufreq_policy *policy;
811         int ret = 0, has_target = !!cpufreq_driver->target;
812         unsigned long flags;
813
814         policy = cpufreq_cpu_get(sibling);
815         WARN_ON(!policy);
816
817         if (has_target)
818                 __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
819
820         lock_policy_rwsem_write(sibling);
821
822         write_lock_irqsave(&cpufreq_driver_lock, flags);
823
824         cpumask_set_cpu(cpu, policy->cpus);
825         per_cpu(cpufreq_policy_cpu, cpu) = policy->cpu;
826         per_cpu(cpufreq_cpu_data, cpu) = policy;
827         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
828
829         unlock_policy_rwsem_write(sibling);
830
831         if (has_target) {
832                 __cpufreq_governor(policy, CPUFREQ_GOV_START);
833                 __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
834         }
835
836         ret = sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
837         if (ret) {
838                 cpufreq_cpu_put(policy);
839                 return ret;
840         }
841
842         return 0;
843 }
844 #endif
845
846 /**
847  * cpufreq_add_dev - add a CPU device
848  *
849  * Adds the cpufreq interface for a CPU device.
850  *
851  * The Oracle says: try running cpufreq registration/unregistration concurrently
852  * with with cpu hotplugging and all hell will break loose. Tried to clean this
853  * mess up, but more thorough testing is needed. - Mathieu
854  */
855 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
856 {
857         unsigned int j, cpu = dev->id;
858         int ret = -ENOMEM;
859         struct cpufreq_policy *policy;
860         unsigned long flags;
861 #ifdef CONFIG_HOTPLUG_CPU
862         struct cpufreq_governor *gov;
863         int sibling;
864 #endif
865
866         if (cpu_is_offline(cpu))
867                 return 0;
868
869         pr_debug("adding CPU %u\n", cpu);
870
871 #ifdef CONFIG_SMP
872         /* check whether a different CPU already registered this
873          * CPU because it is in the same boat. */
874         policy = cpufreq_cpu_get(cpu);
875         if (unlikely(policy)) {
876                 cpufreq_cpu_put(policy);
877                 return 0;
878         }
879
880 #ifdef CONFIG_HOTPLUG_CPU
881         /* Check if this cpu was hot-unplugged earlier and has siblings */
882         read_lock_irqsave(&cpufreq_driver_lock, flags);
883         for_each_online_cpu(sibling) {
884                 struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
885                 if (cp && cpumask_test_cpu(cpu, cp->related_cpus)) {
886                         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
887                         return cpufreq_add_policy_cpu(cpu, sibling, dev);
888                 }
889         }
890         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
891 #endif
892 #endif
893
894         if (!try_module_get(cpufreq_driver->owner)) {
895                 ret = -EINVAL;
896                 goto module_out;
897         }
898
899         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
900         if (!policy)
901                 goto nomem_out;
902
903         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
904                 goto err_free_policy;
905
906         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
907                 goto err_free_cpumask;
908
909         policy->cpu = cpu;
910         policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
911         cpumask_copy(policy->cpus, cpumask_of(cpu));
912
913         /* Initially set CPU itself as the policy_cpu */
914         per_cpu(cpufreq_policy_cpu, cpu) = cpu;
915
916         init_completion(&policy->kobj_unregister);
917         INIT_WORK(&policy->update, handle_update);
918
919         /* call driver. From then on the cpufreq must be able
920          * to accept all calls to ->verify and ->setpolicy for this CPU
921          */
922         ret = cpufreq_driver->init(policy);
923         if (ret) {
924                 pr_debug("initialization failed\n");
925                 goto err_set_policy_cpu;
926         }
927
928         /* related cpus should atleast have policy->cpus */
929         cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
930
931         /*
932          * affected cpus must always be the one, which are online. We aren't
933          * managing offline cpus here.
934          */
935         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
936
937         policy->user_policy.min = policy->min;
938         policy->user_policy.max = policy->max;
939
940         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
941                                      CPUFREQ_START, policy);
942
943 #ifdef CONFIG_HOTPLUG_CPU
944         gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
945         if (gov) {
946                 policy->governor = gov;
947                 pr_debug("Restoring governor %s for cpu %d\n",
948                        policy->governor->name, cpu);
949         }
950 #endif
951
952         ret = cpufreq_add_dev_interface(cpu, policy, dev);
953         if (ret)
954                 goto err_out_unregister;
955
956         kobject_uevent(&policy->kobj, KOBJ_ADD);
957         module_put(cpufreq_driver->owner);
958         pr_debug("initialization complete\n");
959
960         return 0;
961
962 err_out_unregister:
963         write_lock_irqsave(&cpufreq_driver_lock, flags);
964         for_each_cpu(j, policy->cpus)
965                 per_cpu(cpufreq_cpu_data, j) = NULL;
966         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
967
968         kobject_put(&policy->kobj);
969         wait_for_completion(&policy->kobj_unregister);
970
971 err_set_policy_cpu:
972         per_cpu(cpufreq_policy_cpu, cpu) = -1;
973         free_cpumask_var(policy->related_cpus);
974 err_free_cpumask:
975         free_cpumask_var(policy->cpus);
976 err_free_policy:
977         kfree(policy);
978 nomem_out:
979         module_put(cpufreq_driver->owner);
980 module_out:
981         return ret;
982 }
983
984 static void update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
985 {
986         int j;
987
988         policy->last_cpu = policy->cpu;
989         policy->cpu = cpu;
990
991         for_each_cpu(j, policy->cpus)
992                 per_cpu(cpufreq_policy_cpu, j) = cpu;
993
994 #ifdef CONFIG_CPU_FREQ_TABLE
995         cpufreq_frequency_table_update_policy_cpu(policy);
996 #endif
997         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
998                         CPUFREQ_UPDATE_POLICY_CPU, policy);
999 }
1000
1001 /**
1002  * __cpufreq_remove_dev - remove a CPU device
1003  *
1004  * Removes the cpufreq interface for a CPU device.
1005  * Caller should already have policy_rwsem in write mode for this CPU.
1006  * This routine frees the rwsem before returning.
1007  */
1008 static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1009 {
1010         unsigned int cpu = dev->id, ret, cpus;
1011         unsigned long flags;
1012         struct cpufreq_policy *data;
1013         struct kobject *kobj;
1014         struct completion *cmp;
1015         struct device *cpu_dev;
1016
1017         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1018
1019         write_lock_irqsave(&cpufreq_driver_lock, flags);
1020
1021         data = per_cpu(cpufreq_cpu_data, cpu);
1022         per_cpu(cpufreq_cpu_data, cpu) = NULL;
1023
1024         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1025
1026         if (!data) {
1027                 pr_debug("%s: No cpu_data found\n", __func__);
1028                 return -EINVAL;
1029         }
1030
1031         if (cpufreq_driver->target)
1032                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1033
1034 #ifdef CONFIG_HOTPLUG_CPU
1035         if (!cpufreq_driver->setpolicy)
1036                 strncpy(per_cpu(cpufreq_cpu_governor, cpu),
1037                         data->governor->name, CPUFREQ_NAME_LEN);
1038 #endif
1039
1040         WARN_ON(lock_policy_rwsem_write(cpu));
1041         cpus = cpumask_weight(data->cpus);
1042
1043         if (cpus > 1)
1044                 cpumask_clear_cpu(cpu, data->cpus);
1045         unlock_policy_rwsem_write(cpu);
1046
1047         if (cpu != data->cpu) {
1048                 sysfs_remove_link(&dev->kobj, "cpufreq");
1049         } else if (cpus > 1) {
1050                 /* first sibling now owns the new sysfs dir */
1051                 cpu_dev = get_cpu_device(cpumask_first(data->cpus));
1052                 sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
1053                 ret = kobject_move(&data->kobj, &cpu_dev->kobj);
1054                 if (ret) {
1055                         pr_err("%s: Failed to move kobj: %d", __func__, ret);
1056
1057                         WARN_ON(lock_policy_rwsem_write(cpu));
1058                         cpumask_set_cpu(cpu, data->cpus);
1059
1060                         write_lock_irqsave(&cpufreq_driver_lock, flags);
1061                         per_cpu(cpufreq_cpu_data, cpu) = data;
1062                         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1063
1064                         unlock_policy_rwsem_write(cpu);
1065
1066                         ret = sysfs_create_link(&cpu_dev->kobj, &data->kobj,
1067                                         "cpufreq");
1068                         return -EINVAL;
1069                 }
1070
1071                 WARN_ON(lock_policy_rwsem_write(cpu));
1072                 update_policy_cpu(data, cpu_dev->id);
1073                 unlock_policy_rwsem_write(cpu);
1074                 pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
1075                                 __func__, cpu_dev->id, cpu);
1076         }
1077
1078         pr_debug("%s: removing link, cpu: %d\n", __func__, cpu);
1079         cpufreq_cpu_put(data);
1080
1081         /* If cpu is last user of policy, free policy */
1082         if (cpus == 1) {
1083                 if (cpufreq_driver->target)
1084                         __cpufreq_governor(data, CPUFREQ_GOV_POLICY_EXIT);
1085
1086                 lock_policy_rwsem_read(cpu);
1087                 kobj = &data->kobj;
1088                 cmp = &data->kobj_unregister;
1089                 unlock_policy_rwsem_read(cpu);
1090                 kobject_put(kobj);
1091
1092                 /* we need to make sure that the underlying kobj is actually
1093                  * not referenced anymore by anybody before we proceed with
1094                  * unloading.
1095                  */
1096                 pr_debug("waiting for dropping of refcount\n");
1097                 wait_for_completion(cmp);
1098                 pr_debug("wait complete\n");
1099
1100                 if (cpufreq_driver->exit)
1101                         cpufreq_driver->exit(data);
1102
1103                 free_cpumask_var(data->related_cpus);
1104                 free_cpumask_var(data->cpus);
1105                 kfree(data);
1106         } else if (cpufreq_driver->target) {
1107                 __cpufreq_governor(data, CPUFREQ_GOV_START);
1108                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1109         }
1110
1111         per_cpu(cpufreq_policy_cpu, cpu) = -1;
1112         return 0;
1113 }
1114
1115
1116 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1117 {
1118         unsigned int cpu = dev->id;
1119         int retval;
1120
1121         if (cpu_is_offline(cpu))
1122                 return 0;
1123
1124         retval = __cpufreq_remove_dev(dev, sif);
1125         return retval;
1126 }
1127
1128
1129 static void handle_update(struct work_struct *work)
1130 {
1131         struct cpufreq_policy *policy =
1132                 container_of(work, struct cpufreq_policy, update);
1133         unsigned int cpu = policy->cpu;
1134         pr_debug("handle_update for cpu %u called\n", cpu);
1135         cpufreq_update_policy(cpu);
1136 }
1137
1138 /**
1139  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1140  *      @cpu: cpu number
1141  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
1142  *      @new_freq: CPU frequency the CPU actually runs at
1143  *
1144  *      We adjust to current frequency first, and need to clean up later.
1145  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1146  */
1147 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1148                                 unsigned int new_freq)
1149 {
1150         struct cpufreq_policy *policy;
1151         struct cpufreq_freqs freqs;
1152         unsigned long flags;
1153
1154
1155         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1156                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1157
1158         freqs.old = old_freq;
1159         freqs.new = new_freq;
1160
1161         read_lock_irqsave(&cpufreq_driver_lock, flags);
1162         policy = per_cpu(cpufreq_cpu_data, cpu);
1163         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1164
1165         cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
1166         cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
1167 }
1168
1169
1170 /**
1171  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1172  * @cpu: CPU number
1173  *
1174  * This is the last known freq, without actually getting it from the driver.
1175  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1176  */
1177 unsigned int cpufreq_quick_get(unsigned int cpu)
1178 {
1179         struct cpufreq_policy *policy;
1180         unsigned int ret_freq = 0;
1181
1182         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1183                 return cpufreq_driver->get(cpu);
1184
1185         policy = cpufreq_cpu_get(cpu);
1186         if (policy) {
1187                 ret_freq = policy->cur;
1188                 cpufreq_cpu_put(policy);
1189         }
1190
1191         return ret_freq;
1192 }
1193 EXPORT_SYMBOL(cpufreq_quick_get);
1194
1195 /**
1196  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1197  * @cpu: CPU number
1198  *
1199  * Just return the max possible frequency for a given CPU.
1200  */
1201 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1202 {
1203         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1204         unsigned int ret_freq = 0;
1205
1206         if (policy) {
1207                 ret_freq = policy->max;
1208                 cpufreq_cpu_put(policy);
1209         }
1210
1211         return ret_freq;
1212 }
1213 EXPORT_SYMBOL(cpufreq_quick_get_max);
1214
1215
1216 static unsigned int __cpufreq_get(unsigned int cpu)
1217 {
1218         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1219         unsigned int ret_freq = 0;
1220
1221         if (!cpufreq_driver->get)
1222                 return ret_freq;
1223
1224         ret_freq = cpufreq_driver->get(cpu);
1225
1226         if (ret_freq && policy->cur &&
1227                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1228                 /* verify no discrepancy between actual and
1229                                         saved value exists */
1230                 if (unlikely(ret_freq != policy->cur)) {
1231                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1232                         schedule_work(&policy->update);
1233                 }
1234         }
1235
1236         return ret_freq;
1237 }
1238
1239 /**
1240  * cpufreq_get - get the current CPU frequency (in kHz)
1241  * @cpu: CPU number
1242  *
1243  * Get the CPU current (static) CPU frequency
1244  */
1245 unsigned int cpufreq_get(unsigned int cpu)
1246 {
1247         unsigned int ret_freq = 0;
1248         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1249
1250         if (!policy)
1251                 goto out;
1252
1253         if (unlikely(lock_policy_rwsem_read(cpu)))
1254                 goto out_policy;
1255
1256         ret_freq = __cpufreq_get(cpu);
1257
1258         unlock_policy_rwsem_read(cpu);
1259
1260 out_policy:
1261         cpufreq_cpu_put(policy);
1262 out:
1263         return ret_freq;
1264 }
1265 EXPORT_SYMBOL(cpufreq_get);
1266
1267 static struct subsys_interface cpufreq_interface = {
1268         .name           = "cpufreq",
1269         .subsys         = &cpu_subsys,
1270         .add_dev        = cpufreq_add_dev,
1271         .remove_dev     = cpufreq_remove_dev,
1272 };
1273
1274
1275 /**
1276  * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1277  *
1278  * This function is only executed for the boot processor.  The other CPUs
1279  * have been put offline by means of CPU hotplug.
1280  */
1281 static int cpufreq_bp_suspend(void)
1282 {
1283         int ret = 0;
1284
1285         int cpu = smp_processor_id();
1286         struct cpufreq_policy *cpu_policy;
1287
1288         pr_debug("suspending cpu %u\n", cpu);
1289
1290         /* If there's no policy for the boot CPU, we have nothing to do. */
1291         cpu_policy = cpufreq_cpu_get(cpu);
1292         if (!cpu_policy)
1293                 return 0;
1294
1295         if (cpufreq_driver->suspend) {
1296                 ret = cpufreq_driver->suspend(cpu_policy);
1297                 if (ret)
1298                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1299                                         "step on CPU %u\n", cpu_policy->cpu);
1300         }
1301
1302         cpufreq_cpu_put(cpu_policy);
1303         return ret;
1304 }
1305
1306 /**
1307  * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1308  *
1309  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1310  *      2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1311  *          restored. It will verify that the current freq is in sync with
1312  *          what we believe it to be. This is a bit later than when it
1313  *          should be, but nonethteless it's better than calling
1314  *          cpufreq_driver->get() here which might re-enable interrupts...
1315  *
1316  * This function is only executed for the boot CPU.  The other CPUs have not
1317  * been turned on yet.
1318  */
1319 static void cpufreq_bp_resume(void)
1320 {
1321         int ret = 0;
1322
1323         int cpu = smp_processor_id();
1324         struct cpufreq_policy *cpu_policy;
1325
1326         pr_debug("resuming cpu %u\n", cpu);
1327
1328         /* If there's no policy for the boot CPU, we have nothing to do. */
1329         cpu_policy = cpufreq_cpu_get(cpu);
1330         if (!cpu_policy)
1331                 return;
1332
1333         if (cpufreq_driver->resume) {
1334                 ret = cpufreq_driver->resume(cpu_policy);
1335                 if (ret) {
1336                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1337                                         "step on CPU %u\n", cpu_policy->cpu);
1338                         goto fail;
1339                 }
1340         }
1341
1342         schedule_work(&cpu_policy->update);
1343
1344 fail:
1345         cpufreq_cpu_put(cpu_policy);
1346 }
1347
1348 static struct syscore_ops cpufreq_syscore_ops = {
1349         .suspend        = cpufreq_bp_suspend,
1350         .resume         = cpufreq_bp_resume,
1351 };
1352
1353 /**
1354  *      cpufreq_get_current_driver - return current driver's name
1355  *
1356  *      Return the name string of the currently loaded cpufreq driver
1357  *      or NULL, if none.
1358  */
1359 const char *cpufreq_get_current_driver(void)
1360 {
1361         if (cpufreq_driver)
1362                 return cpufreq_driver->name;
1363
1364         return NULL;
1365 }
1366 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1367
1368 /*********************************************************************
1369  *                     NOTIFIER LISTS INTERFACE                      *
1370  *********************************************************************/
1371
1372 /**
1373  *      cpufreq_register_notifier - register a driver with cpufreq
1374  *      @nb: notifier function to register
1375  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1376  *
1377  *      Add a driver to one of two lists: either a list of drivers that
1378  *      are notified about clock rate changes (once before and once after
1379  *      the transition), or a list of drivers that are notified about
1380  *      changes in cpufreq policy.
1381  *
1382  *      This function may sleep, and has the same return conditions as
1383  *      blocking_notifier_chain_register.
1384  */
1385 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1386 {
1387         int ret;
1388
1389         if (cpufreq_disabled())
1390                 return -EINVAL;
1391
1392         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1393
1394         switch (list) {
1395         case CPUFREQ_TRANSITION_NOTIFIER:
1396                 ret = srcu_notifier_chain_register(
1397                                 &cpufreq_transition_notifier_list, nb);
1398                 break;
1399         case CPUFREQ_POLICY_NOTIFIER:
1400                 ret = blocking_notifier_chain_register(
1401                                 &cpufreq_policy_notifier_list, nb);
1402                 break;
1403         default:
1404                 ret = -EINVAL;
1405         }
1406
1407         return ret;
1408 }
1409 EXPORT_SYMBOL(cpufreq_register_notifier);
1410
1411
1412 /**
1413  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1414  *      @nb: notifier block to be unregistered
1415  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1416  *
1417  *      Remove a driver from the CPU frequency notifier list.
1418  *
1419  *      This function may sleep, and has the same return conditions as
1420  *      blocking_notifier_chain_unregister.
1421  */
1422 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1423 {
1424         int ret;
1425
1426         if (cpufreq_disabled())
1427                 return -EINVAL;
1428
1429         switch (list) {
1430         case CPUFREQ_TRANSITION_NOTIFIER:
1431                 ret = srcu_notifier_chain_unregister(
1432                                 &cpufreq_transition_notifier_list, nb);
1433                 break;
1434         case CPUFREQ_POLICY_NOTIFIER:
1435                 ret = blocking_notifier_chain_unregister(
1436                                 &cpufreq_policy_notifier_list, nb);
1437                 break;
1438         default:
1439                 ret = -EINVAL;
1440         }
1441
1442         return ret;
1443 }
1444 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1445
1446
1447 /*********************************************************************
1448  *                              GOVERNORS                            *
1449  *********************************************************************/
1450
1451
1452 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1453                             unsigned int target_freq,
1454                             unsigned int relation)
1455 {
1456         int retval = -EINVAL;
1457         unsigned int old_target_freq = target_freq;
1458
1459         if (cpufreq_disabled())
1460                 return -ENODEV;
1461
1462         /* Make sure that target_freq is within supported range */
1463         if (target_freq > policy->max)
1464                 target_freq = policy->max;
1465         if (target_freq < policy->min)
1466                 target_freq = policy->min;
1467
1468         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1469                         policy->cpu, target_freq, relation, old_target_freq);
1470
1471         if (target_freq == policy->cur)
1472                 return 0;
1473
1474         if (cpufreq_driver->target)
1475                 retval = cpufreq_driver->target(policy, target_freq, relation);
1476
1477         return retval;
1478 }
1479 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1480
1481 int cpufreq_driver_target(struct cpufreq_policy *policy,
1482                           unsigned int target_freq,
1483                           unsigned int relation)
1484 {
1485         int ret = -EINVAL;
1486
1487         policy = cpufreq_cpu_get(policy->cpu);
1488         if (!policy)
1489                 goto no_policy;
1490
1491         if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1492                 goto fail;
1493
1494         ret = __cpufreq_driver_target(policy, target_freq, relation);
1495
1496         unlock_policy_rwsem_write(policy->cpu);
1497
1498 fail:
1499         cpufreq_cpu_put(policy);
1500 no_policy:
1501         return ret;
1502 }
1503 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1504
1505 int __cpufreq_driver_getavg(struct cpufreq_policy *policy, unsigned int cpu)
1506 {
1507         int ret = 0;
1508
1509         if (cpufreq_disabled())
1510                 return ret;
1511
1512         if (!cpufreq_driver->getavg)
1513                 return 0;
1514
1515         policy = cpufreq_cpu_get(policy->cpu);
1516         if (!policy)
1517                 return -EINVAL;
1518
1519         ret = cpufreq_driver->getavg(policy, cpu);
1520
1521         cpufreq_cpu_put(policy);
1522         return ret;
1523 }
1524 EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1525
1526 /*
1527  * when "event" is CPUFREQ_GOV_LIMITS
1528  */
1529
1530 static int __cpufreq_governor(struct cpufreq_policy *policy,
1531                                         unsigned int event)
1532 {
1533         int ret;
1534
1535         /* Only must be defined when default governor is known to have latency
1536            restrictions, like e.g. conservative or ondemand.
1537            That this is the case is already ensured in Kconfig
1538         */
1539 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1540         struct cpufreq_governor *gov = &cpufreq_gov_performance;
1541 #else
1542         struct cpufreq_governor *gov = NULL;
1543 #endif
1544
1545         if (policy->governor->max_transition_latency &&
1546             policy->cpuinfo.transition_latency >
1547             policy->governor->max_transition_latency) {
1548                 if (!gov)
1549                         return -EINVAL;
1550                 else {
1551                         printk(KERN_WARNING "%s governor failed, too long"
1552                                " transition latency of HW, fallback"
1553                                " to %s governor\n",
1554                                policy->governor->name,
1555                                gov->name);
1556                         policy->governor = gov;
1557                 }
1558         }
1559
1560         if (!try_module_get(policy->governor->owner))
1561                 return -EINVAL;
1562
1563         pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1564                                                 policy->cpu, event);
1565         ret = policy->governor->governor(policy, event);
1566
1567         if (!ret) {
1568                 if (event == CPUFREQ_GOV_POLICY_INIT)
1569                         policy->governor->initialized++;
1570                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
1571                         policy->governor->initialized--;
1572         }
1573
1574         /* we keep one module reference alive for
1575                         each CPU governed by this CPU */
1576         if ((event != CPUFREQ_GOV_START) || ret)
1577                 module_put(policy->governor->owner);
1578         if ((event == CPUFREQ_GOV_STOP) && !ret)
1579                 module_put(policy->governor->owner);
1580
1581         return ret;
1582 }
1583
1584
1585 int cpufreq_register_governor(struct cpufreq_governor *governor)
1586 {
1587         int err;
1588
1589         if (!governor)
1590                 return -EINVAL;
1591
1592         if (cpufreq_disabled())
1593                 return -ENODEV;
1594
1595         mutex_lock(&cpufreq_governor_mutex);
1596
1597         governor->initialized = 0;
1598         err = -EBUSY;
1599         if (__find_governor(governor->name) == NULL) {
1600                 err = 0;
1601                 list_add(&governor->governor_list, &cpufreq_governor_list);
1602         }
1603
1604         mutex_unlock(&cpufreq_governor_mutex);
1605         return err;
1606 }
1607 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1608
1609
1610 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1611 {
1612 #ifdef CONFIG_HOTPLUG_CPU
1613         int cpu;
1614 #endif
1615
1616         if (!governor)
1617                 return;
1618
1619         if (cpufreq_disabled())
1620                 return;
1621
1622 #ifdef CONFIG_HOTPLUG_CPU
1623         for_each_present_cpu(cpu) {
1624                 if (cpu_online(cpu))
1625                         continue;
1626                 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1627                         strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1628         }
1629 #endif
1630
1631         mutex_lock(&cpufreq_governor_mutex);
1632         list_del(&governor->governor_list);
1633         mutex_unlock(&cpufreq_governor_mutex);
1634         return;
1635 }
1636 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1637
1638
1639
1640 /*********************************************************************
1641  *                          POLICY INTERFACE                         *
1642  *********************************************************************/
1643
1644 /**
1645  * cpufreq_get_policy - get the current cpufreq_policy
1646  * @policy: struct cpufreq_policy into which the current cpufreq_policy
1647  *      is written
1648  *
1649  * Reads the current cpufreq policy.
1650  */
1651 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1652 {
1653         struct cpufreq_policy *cpu_policy;
1654         if (!policy)
1655                 return -EINVAL;
1656
1657         cpu_policy = cpufreq_cpu_get(cpu);
1658         if (!cpu_policy)
1659                 return -EINVAL;
1660
1661         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1662
1663         cpufreq_cpu_put(cpu_policy);
1664         return 0;
1665 }
1666 EXPORT_SYMBOL(cpufreq_get_policy);
1667
1668
1669 /*
1670  * data   : current policy.
1671  * policy : policy to be set.
1672  */
1673 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1674                                 struct cpufreq_policy *policy)
1675 {
1676         int ret = 0, failed = 1;
1677
1678         pr_debug("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1679                 policy->min, policy->max);
1680
1681         memcpy(&policy->cpuinfo, &data->cpuinfo,
1682                                 sizeof(struct cpufreq_cpuinfo));
1683
1684         if (policy->min > data->max || policy->max < data->min) {
1685                 ret = -EINVAL;
1686                 goto error_out;
1687         }
1688
1689         /* verify the cpu speed can be set within this limit */
1690         ret = cpufreq_driver->verify(policy);
1691         if (ret)
1692                 goto error_out;
1693
1694         /* adjust if necessary - all reasons */
1695         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1696                         CPUFREQ_ADJUST, policy);
1697
1698         /* adjust if necessary - hardware incompatibility*/
1699         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1700                         CPUFREQ_INCOMPATIBLE, policy);
1701
1702         /* verify the cpu speed can be set within this limit,
1703            which might be different to the first one */
1704         ret = cpufreq_driver->verify(policy);
1705         if (ret)
1706                 goto error_out;
1707
1708         /* notification of the new policy */
1709         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1710                         CPUFREQ_NOTIFY, policy);
1711
1712         data->min = policy->min;
1713         data->max = policy->max;
1714
1715         pr_debug("new min and max freqs are %u - %u kHz\n",
1716                                         data->min, data->max);
1717
1718         if (cpufreq_driver->setpolicy) {
1719                 data->policy = policy->policy;
1720                 pr_debug("setting range\n");
1721                 ret = cpufreq_driver->setpolicy(policy);
1722         } else {
1723                 if (policy->governor != data->governor) {
1724                         /* save old, working values */
1725                         struct cpufreq_governor *old_gov = data->governor;
1726
1727                         pr_debug("governor switch\n");
1728
1729                         /* end old governor */
1730                         if (data->governor) {
1731                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1732                                 __cpufreq_governor(data,
1733                                                 CPUFREQ_GOV_POLICY_EXIT);
1734                         }
1735
1736                         /* start new governor */
1737                         data->governor = policy->governor;
1738                         if (!__cpufreq_governor(data, CPUFREQ_GOV_POLICY_INIT)) {
1739                                 if (!__cpufreq_governor(data, CPUFREQ_GOV_START))
1740                                         failed = 0;
1741                                 else
1742                                         __cpufreq_governor(data,
1743                                                         CPUFREQ_GOV_POLICY_EXIT);
1744                         }
1745
1746                         if (failed) {
1747                                 /* new governor failed, so re-start old one */
1748                                 pr_debug("starting governor %s failed\n",
1749                                                         data->governor->name);
1750                                 if (old_gov) {
1751                                         data->governor = old_gov;
1752                                         __cpufreq_governor(data,
1753                                                         CPUFREQ_GOV_POLICY_INIT);
1754                                         __cpufreq_governor(data,
1755                                                            CPUFREQ_GOV_START);
1756                                 }
1757                                 ret = -EINVAL;
1758                                 goto error_out;
1759                         }
1760                         /* might be a policy change, too, so fall through */
1761                 }
1762                 pr_debug("governor: change or update limits\n");
1763                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1764         }
1765
1766 error_out:
1767         return ret;
1768 }
1769
1770 /**
1771  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1772  *      @cpu: CPU which shall be re-evaluated
1773  *
1774  *      Useful for policy notifiers which have different necessities
1775  *      at different times.
1776  */
1777 int cpufreq_update_policy(unsigned int cpu)
1778 {
1779         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1780         struct cpufreq_policy policy;
1781         int ret;
1782
1783         if (!data) {
1784                 ret = -ENODEV;
1785                 goto no_policy;
1786         }
1787
1788         if (unlikely(lock_policy_rwsem_write(cpu))) {
1789                 ret = -EINVAL;
1790                 goto fail;
1791         }
1792
1793         pr_debug("updating policy for CPU %u\n", cpu);
1794         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1795         policy.min = data->user_policy.min;
1796         policy.max = data->user_policy.max;
1797         policy.policy = data->user_policy.policy;
1798         policy.governor = data->user_policy.governor;
1799
1800         /* BIOS might change freq behind our back
1801           -> ask driver for current freq and notify governors about a change */
1802         if (cpufreq_driver->get) {
1803                 policy.cur = cpufreq_driver->get(cpu);
1804                 if (!data->cur) {
1805                         pr_debug("Driver did not initialize current freq");
1806                         data->cur = policy.cur;
1807                 } else {
1808                         if (data->cur != policy.cur && cpufreq_driver->target)
1809                                 cpufreq_out_of_sync(cpu, data->cur,
1810                                                                 policy.cur);
1811                 }
1812         }
1813
1814         ret = __cpufreq_set_policy(data, &policy);
1815
1816         unlock_policy_rwsem_write(cpu);
1817
1818 fail:
1819         cpufreq_cpu_put(data);
1820 no_policy:
1821         return ret;
1822 }
1823 EXPORT_SYMBOL(cpufreq_update_policy);
1824
1825 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1826                                         unsigned long action, void *hcpu)
1827 {
1828         unsigned int cpu = (unsigned long)hcpu;
1829         struct device *dev;
1830
1831         dev = get_cpu_device(cpu);
1832         if (dev) {
1833                 switch (action) {
1834                 case CPU_ONLINE:
1835                 case CPU_ONLINE_FROZEN:
1836                         cpufreq_add_dev(dev, NULL);
1837                         break;
1838                 case CPU_DOWN_PREPARE:
1839                 case CPU_DOWN_PREPARE_FROZEN:
1840                         __cpufreq_remove_dev(dev, NULL);
1841                         break;
1842                 case CPU_DOWN_FAILED:
1843                 case CPU_DOWN_FAILED_FROZEN:
1844                         cpufreq_add_dev(dev, NULL);
1845                         break;
1846                 }
1847         }
1848         return NOTIFY_OK;
1849 }
1850
1851 static struct notifier_block __refdata cpufreq_cpu_notifier = {
1852     .notifier_call = cpufreq_cpu_callback,
1853 };
1854
1855 /*********************************************************************
1856  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1857  *********************************************************************/
1858
1859 /**
1860  * cpufreq_register_driver - register a CPU Frequency driver
1861  * @driver_data: A struct cpufreq_driver containing the values#
1862  * submitted by the CPU Frequency driver.
1863  *
1864  *   Registers a CPU Frequency driver to this core code. This code
1865  * returns zero on success, -EBUSY when another driver got here first
1866  * (and isn't unregistered in the meantime).
1867  *
1868  */
1869 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1870 {
1871         unsigned long flags;
1872         int ret;
1873
1874         if (cpufreq_disabled())
1875                 return -ENODEV;
1876
1877         if (!driver_data || !driver_data->verify || !driver_data->init ||
1878             ((!driver_data->setpolicy) && (!driver_data->target)))
1879                 return -EINVAL;
1880
1881         pr_debug("trying to register driver %s\n", driver_data->name);
1882
1883         if (driver_data->setpolicy)
1884                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1885
1886         write_lock_irqsave(&cpufreq_driver_lock, flags);
1887         if (cpufreq_driver) {
1888                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1889                 return -EBUSY;
1890         }
1891         cpufreq_driver = driver_data;
1892         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1893
1894         ret = subsys_interface_register(&cpufreq_interface);
1895         if (ret)
1896                 goto err_null_driver;
1897
1898         if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1899                 int i;
1900                 ret = -ENODEV;
1901
1902                 /* check for at least one working CPU */
1903                 for (i = 0; i < nr_cpu_ids; i++)
1904                         if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
1905                                 ret = 0;
1906                                 break;
1907                         }
1908
1909                 /* if all ->init() calls failed, unregister */
1910                 if (ret) {
1911                         pr_debug("no CPU initialized for driver %s\n",
1912                                                         driver_data->name);
1913                         goto err_if_unreg;
1914                 }
1915         }
1916
1917         register_hotcpu_notifier(&cpufreq_cpu_notifier);
1918         pr_debug("driver %s up and running\n", driver_data->name);
1919
1920         return 0;
1921 err_if_unreg:
1922         subsys_interface_unregister(&cpufreq_interface);
1923 err_null_driver:
1924         write_lock_irqsave(&cpufreq_driver_lock, flags);
1925         cpufreq_driver = NULL;
1926         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1927         return ret;
1928 }
1929 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1930
1931
1932 /**
1933  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1934  *
1935  *    Unregister the current CPUFreq driver. Only call this if you have
1936  * the right to do so, i.e. if you have succeeded in initialising before!
1937  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1938  * currently not initialised.
1939  */
1940 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1941 {
1942         unsigned long flags;
1943
1944         if (!cpufreq_driver || (driver != cpufreq_driver))
1945                 return -EINVAL;
1946
1947         pr_debug("unregistering driver %s\n", driver->name);
1948
1949         subsys_interface_unregister(&cpufreq_interface);
1950         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1951
1952         write_lock_irqsave(&cpufreq_driver_lock, flags);
1953         cpufreq_driver = NULL;
1954         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1955
1956         return 0;
1957 }
1958 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1959
1960 static int __init cpufreq_core_init(void)
1961 {
1962         int cpu;
1963
1964         if (cpufreq_disabled())
1965                 return -ENODEV;
1966
1967         for_each_possible_cpu(cpu) {
1968                 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1969                 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1970         }
1971
1972         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
1973         BUG_ON(!cpufreq_global_kobject);
1974         register_syscore_ops(&cpufreq_syscore_ops);
1975
1976         return 0;
1977 }
1978 core_initcall(cpufreq_core_init);