Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / 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
32 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
33                                                 "cpufreq-core", msg)
34
35 /**
36  * The "cpufreq driver" - the arch- or hardware-dependent low
37  * level driver of CPUFreq support, and its spinlock. This lock
38  * also protects the cpufreq_cpu_data array.
39  */
40 static struct cpufreq_driver *cpufreq_driver;
41 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
42 #ifdef CONFIG_HOTPLUG_CPU
43 /* This one keeps track of the previously set governor of a removed CPU */
44 static struct cpufreq_governor *cpufreq_cpu_governor[NR_CPUS];
45 #endif
46 static DEFINE_SPINLOCK(cpufreq_driver_lock);
47
48 /*
49  * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
50  * all cpufreq/hotplug/workqueue/etc related lock issues.
51  *
52  * The rules for this semaphore:
53  * - Any routine that wants to read from the policy structure will
54  *   do a down_read on this semaphore.
55  * - Any routine that will write to the policy structure and/or may take away
56  *   the policy altogether (eg. CPU hotplug), will hold this lock in write
57  *   mode before doing so.
58  *
59  * Additional rules:
60  * - All holders of the lock should check to make sure that the CPU they
61  *   are concerned with are online after they get the lock.
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  */
65 static DEFINE_PER_CPU(int, policy_cpu);
66 static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
67
68 #define lock_policy_rwsem(mode, cpu)                                    \
69 int lock_policy_rwsem_##mode                                            \
70 (int cpu)                                                               \
71 {                                                                       \
72         int policy_cpu = per_cpu(policy_cpu, cpu);                      \
73         BUG_ON(policy_cpu == -1);                                       \
74         down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));            \
75         if (unlikely(!cpu_online(cpu))) {                               \
76                 up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));      \
77                 return -1;                                              \
78         }                                                               \
79                                                                         \
80         return 0;                                                       \
81 }
82
83 lock_policy_rwsem(read, cpu);
84 EXPORT_SYMBOL_GPL(lock_policy_rwsem_read);
85
86 lock_policy_rwsem(write, cpu);
87 EXPORT_SYMBOL_GPL(lock_policy_rwsem_write);
88
89 void unlock_policy_rwsem_read(int cpu)
90 {
91         int policy_cpu = per_cpu(policy_cpu, cpu);
92         BUG_ON(policy_cpu == -1);
93         up_read(&per_cpu(cpu_policy_rwsem, policy_cpu));
94 }
95 EXPORT_SYMBOL_GPL(unlock_policy_rwsem_read);
96
97 void unlock_policy_rwsem_write(int cpu)
98 {
99         int policy_cpu = per_cpu(policy_cpu, cpu);
100         BUG_ON(policy_cpu == -1);
101         up_write(&per_cpu(cpu_policy_rwsem, policy_cpu));
102 }
103 EXPORT_SYMBOL_GPL(unlock_policy_rwsem_write);
104
105
106 /* internal prototypes */
107 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
108 static unsigned int __cpufreq_get(unsigned int cpu);
109 static void handle_update(struct work_struct *work);
110
111 /**
112  * Two notifier lists: the "policy" list is involved in the
113  * validation process for a new CPU frequency policy; the
114  * "transition" list for kernel code that needs to handle
115  * changes to devices when the CPU clock speed changes.
116  * The mutex locks both lists.
117  */
118 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
119 static struct srcu_notifier_head cpufreq_transition_notifier_list;
120
121 static int __init init_cpufreq_transition_notifier_list(void)
122 {
123         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
124         return 0;
125 }
126 pure_initcall(init_cpufreq_transition_notifier_list);
127
128 static LIST_HEAD(cpufreq_governor_list);
129 static DEFINE_MUTEX (cpufreq_governor_mutex);
130
131 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
132 {
133         struct cpufreq_policy *data;
134         unsigned long flags;
135
136         if (cpu >= NR_CPUS)
137                 goto err_out;
138
139         /* get the cpufreq driver */
140         spin_lock_irqsave(&cpufreq_driver_lock, flags);
141
142         if (!cpufreq_driver)
143                 goto err_out_unlock;
144
145         if (!try_module_get(cpufreq_driver->owner))
146                 goto err_out_unlock;
147
148
149         /* get the CPU */
150         data = cpufreq_cpu_data[cpu];
151
152         if (!data)
153                 goto err_out_put_module;
154
155         if (!kobject_get(&data->kobj))
156                 goto err_out_put_module;
157
158         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
159         return data;
160
161 err_out_put_module:
162         module_put(cpufreq_driver->owner);
163 err_out_unlock:
164         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
165 err_out:
166         return NULL;
167 }
168 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
169
170
171 void cpufreq_cpu_put(struct cpufreq_policy *data)
172 {
173         kobject_put(&data->kobj);
174         module_put(cpufreq_driver->owner);
175 }
176 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
177
178
179 /*********************************************************************
180  *                     UNIFIED DEBUG HELPERS                         *
181  *********************************************************************/
182 #ifdef CONFIG_CPU_FREQ_DEBUG
183
184 /* what part(s) of the CPUfreq subsystem are debugged? */
185 static unsigned int debug;
186
187 /* is the debug output ratelimit'ed using printk_ratelimit? User can
188  * set or modify this value.
189  */
190 static unsigned int debug_ratelimit = 1;
191
192 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
193  * loading of a cpufreq driver, temporarily disabled when a new policy
194  * is set, and disabled upon cpufreq driver removal
195  */
196 static unsigned int disable_ratelimit = 1;
197 static DEFINE_SPINLOCK(disable_ratelimit_lock);
198
199 static void cpufreq_debug_enable_ratelimit(void)
200 {
201         unsigned long flags;
202
203         spin_lock_irqsave(&disable_ratelimit_lock, flags);
204         if (disable_ratelimit)
205                 disable_ratelimit--;
206         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
207 }
208
209 static void cpufreq_debug_disable_ratelimit(void)
210 {
211         unsigned long flags;
212
213         spin_lock_irqsave(&disable_ratelimit_lock, flags);
214         disable_ratelimit++;
215         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
216 }
217
218 void cpufreq_debug_printk(unsigned int type, const char *prefix,
219                                                         const char *fmt, ...)
220 {
221         char s[256];
222         va_list args;
223         unsigned int len;
224         unsigned long flags;
225
226         WARN_ON(!prefix);
227         if (type & debug) {
228                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
229                 if (!disable_ratelimit && debug_ratelimit
230                                         && !printk_ratelimit()) {
231                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
232                         return;
233                 }
234                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
235
236                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
237
238                 va_start(args, fmt);
239                 len += vsnprintf(&s[len], (256 - len), fmt, args);
240                 va_end(args);
241
242                 printk(s);
243
244                 WARN_ON(len < 5);
245         }
246 }
247 EXPORT_SYMBOL(cpufreq_debug_printk);
248
249
250 module_param(debug, uint, 0644);
251 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core,"
252                         " 2 to debug drivers, and 4 to debug governors.");
253
254 module_param(debug_ratelimit, uint, 0644);
255 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging:"
256                                         " set to 0 to disable ratelimiting.");
257
258 #else /* !CONFIG_CPU_FREQ_DEBUG */
259
260 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
261 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
262
263 #endif /* CONFIG_CPU_FREQ_DEBUG */
264
265
266 /*********************************************************************
267  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
268  *********************************************************************/
269
270 /**
271  * adjust_jiffies - adjust the system "loops_per_jiffy"
272  *
273  * This function alters the system "loops_per_jiffy" for the clock
274  * speed change. Note that loops_per_jiffy cannot be updated on SMP
275  * systems as each CPU might be scaled differently. So, use the arch
276  * per-CPU loops_per_jiffy value wherever possible.
277  */
278 #ifndef CONFIG_SMP
279 static unsigned long l_p_j_ref;
280 static unsigned int  l_p_j_ref_freq;
281
282 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
283 {
284         if (ci->flags & CPUFREQ_CONST_LOOPS)
285                 return;
286
287         if (!l_p_j_ref_freq) {
288                 l_p_j_ref = loops_per_jiffy;
289                 l_p_j_ref_freq = ci->old;
290                 dprintk("saving %lu as reference value for loops_per_jiffy; "
291                         "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
292         }
293         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
294             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
295             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
296                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
297                                                                 ci->new);
298                 dprintk("scaling loops_per_jiffy to %lu "
299                         "for frequency %u kHz\n", loops_per_jiffy, ci->new);
300         }
301 }
302 #else
303 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
304 {
305         return;
306 }
307 #endif
308
309
310 /**
311  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
312  * on frequency transition.
313  *
314  * This function calls the transition notifiers and the "adjust_jiffies"
315  * function. It is called twice on all CPU frequency changes that have
316  * external effects.
317  */
318 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
319 {
320         struct cpufreq_policy *policy;
321
322         BUG_ON(irqs_disabled());
323
324         freqs->flags = cpufreq_driver->flags;
325         dprintk("notification %u of frequency transition to %u kHz\n",
326                 state, freqs->new);
327
328         policy = cpufreq_cpu_data[freqs->cpu];
329         switch (state) {
330
331         case CPUFREQ_PRECHANGE:
332                 /* detect if the driver reported a value as "old frequency"
333                  * which is not equal to what the cpufreq core thinks is
334                  * "old frequency".
335                  */
336                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
337                         if ((policy) && (policy->cpu == freqs->cpu) &&
338                             (policy->cur) && (policy->cur != freqs->old)) {
339                                 dprintk("Warning: CPU frequency is"
340                                         " %u, cpufreq assumed %u kHz.\n",
341                                         freqs->old, policy->cur);
342                                 freqs->old = policy->cur;
343                         }
344                 }
345                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
346                                 CPUFREQ_PRECHANGE, freqs);
347                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
348                 break;
349
350         case CPUFREQ_POSTCHANGE:
351                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
352                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
353                                 CPUFREQ_POSTCHANGE, freqs);
354                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
355                         policy->cur = freqs->new;
356                 break;
357         }
358 }
359 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
360
361
362
363 /*********************************************************************
364  *                          SYSFS INTERFACE                          *
365  *********************************************************************/
366
367 static struct cpufreq_governor *__find_governor(const char *str_governor)
368 {
369         struct cpufreq_governor *t;
370
371         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
372                 if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN))
373                         return t;
374
375         return NULL;
376 }
377
378 /**
379  * cpufreq_parse_governor - parse a governor string
380  */
381 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
382                                 struct cpufreq_governor **governor)
383 {
384         int err = -EINVAL;
385
386         if (!cpufreq_driver)
387                 goto out;
388
389         if (cpufreq_driver->setpolicy) {
390                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
391                         *policy = CPUFREQ_POLICY_PERFORMANCE;
392                         err = 0;
393                 } else if (!strnicmp(str_governor, "powersave",
394                                                 CPUFREQ_NAME_LEN)) {
395                         *policy = CPUFREQ_POLICY_POWERSAVE;
396                         err = 0;
397                 }
398         } else if (cpufreq_driver->target) {
399                 struct cpufreq_governor *t;
400
401                 mutex_lock(&cpufreq_governor_mutex);
402
403                 t = __find_governor(str_governor);
404
405                 if (t == NULL) {
406                         char *name = kasprintf(GFP_KERNEL, "cpufreq_%s",
407                                                                 str_governor);
408
409                         if (name) {
410                                 int ret;
411
412                                 mutex_unlock(&cpufreq_governor_mutex);
413                                 ret = request_module(name);
414                                 mutex_lock(&cpufreq_governor_mutex);
415
416                                 if (ret == 0)
417                                         t = __find_governor(str_governor);
418                         }
419
420                         kfree(name);
421                 }
422
423                 if (t != NULL) {
424                         *governor = t;
425                         err = 0;
426                 }
427
428                 mutex_unlock(&cpufreq_governor_mutex);
429         }
430   out:
431         return err;
432 }
433
434
435 /* drivers/base/cpu.c */
436 extern struct sysdev_class cpu_sysdev_class;
437
438
439 /**
440  * cpufreq_per_cpu_attr_read() / show_##file_name() -
441  * print out cpufreq information
442  *
443  * Write out information from cpufreq_driver->policy[cpu]; object must be
444  * "unsigned int".
445  */
446
447 #define show_one(file_name, object)                     \
448 static ssize_t show_##file_name                         \
449 (struct cpufreq_policy * policy, char *buf)             \
450 {                                                       \
451         return sprintf (buf, "%u\n", policy->object);   \
452 }
453
454 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
455 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
456 show_one(scaling_min_freq, min);
457 show_one(scaling_max_freq, max);
458 show_one(scaling_cur_freq, cur);
459
460 static int __cpufreq_set_policy(struct cpufreq_policy *data,
461                                 struct cpufreq_policy *policy);
462
463 /**
464  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
465  */
466 #define store_one(file_name, object)                    \
467 static ssize_t store_##file_name                                        \
468 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
469 {                                                                       \
470         unsigned int ret = -EINVAL;                                     \
471         struct cpufreq_policy new_policy;                               \
472                                                                         \
473         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
474         if (ret)                                                        \
475                 return -EINVAL;                                         \
476                                                                         \
477         ret = sscanf (buf, "%u", &new_policy.object);                   \
478         if (ret != 1)                                                   \
479                 return -EINVAL;                                         \
480                                                                         \
481         ret = __cpufreq_set_policy(policy, &new_policy);                \
482         policy->user_policy.object = policy->object;                    \
483                                                                         \
484         return ret ? ret : count;                                       \
485 }
486
487 store_one(scaling_min_freq,min);
488 store_one(scaling_max_freq,max);
489
490 /**
491  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
492  */
493 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy,
494                                                         char *buf)
495 {
496         unsigned int cur_freq = __cpufreq_get(policy->cpu);
497         if (!cur_freq)
498                 return sprintf(buf, "<unknown>");
499         return sprintf(buf, "%u\n", cur_freq);
500 }
501
502
503 /**
504  * show_scaling_governor - show the current policy for the specified CPU
505  */
506 static ssize_t show_scaling_governor (struct cpufreq_policy * policy,
507                                                         char *buf)
508 {
509         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
510                 return sprintf(buf, "powersave\n");
511         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
512                 return sprintf(buf, "performance\n");
513         else if (policy->governor)
514                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
515         return -EINVAL;
516 }
517
518
519 /**
520  * store_scaling_governor - store policy for the specified CPU
521  */
522 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
523                                        const char *buf, size_t count)
524 {
525         unsigned int ret = -EINVAL;
526         char    str_governor[16];
527         struct cpufreq_policy new_policy;
528
529         ret = cpufreq_get_policy(&new_policy, policy->cpu);
530         if (ret)
531                 return ret;
532
533         ret = sscanf (buf, "%15s", str_governor);
534         if (ret != 1)
535                 return -EINVAL;
536
537         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
538                                                 &new_policy.governor))
539                 return -EINVAL;
540
541         /* Do not use cpufreq_set_policy here or the user_policy.max
542            will be wrongly overridden */
543         ret = __cpufreq_set_policy(policy, &new_policy);
544
545         policy->user_policy.policy = policy->policy;
546         policy->user_policy.governor = policy->governor;
547
548         if (ret)
549                 return ret;
550         else
551                 return count;
552 }
553
554 /**
555  * show_scaling_driver - show the cpufreq driver currently loaded
556  */
557 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
558 {
559         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
560 }
561
562 /**
563  * show_scaling_available_governors - show the available CPUfreq governors
564  */
565 static ssize_t show_scaling_available_governors (struct cpufreq_policy *policy,
566                                 char *buf)
567 {
568         ssize_t i = 0;
569         struct cpufreq_governor *t;
570
571         if (!cpufreq_driver->target) {
572                 i += sprintf(buf, "performance powersave");
573                 goto out;
574         }
575
576         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
577                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
578                         goto out;
579                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
580         }
581 out:
582         i += sprintf(&buf[i], "\n");
583         return i;
584 }
585 /**
586  * show_affected_cpus - show the CPUs affected by each transition
587  */
588 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
589 {
590         ssize_t i = 0;
591         unsigned int cpu;
592
593         for_each_cpu_mask(cpu, policy->cpus) {
594                 if (i)
595                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
596                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
597                 if (i >= (PAGE_SIZE - 5))
598                     break;
599         }
600         i += sprintf(&buf[i], "\n");
601         return i;
602 }
603
604 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
605                 const char *buf, size_t count)
606 {
607         unsigned int freq = 0;
608         unsigned int ret;
609
610         if (!policy->governor->store_setspeed)
611                 return -EINVAL;
612
613         ret = sscanf(buf, "%u", &freq);
614         if (ret != 1)
615                 return -EINVAL;
616
617         policy->governor->store_setspeed(policy, freq);
618
619         return count;
620 }
621
622 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
623 {
624         if (!policy->governor->show_setspeed)
625                 return sprintf(buf, "<unsupported>\n");
626
627         return policy->governor->show_setspeed(policy, buf);
628 }
629
630 #define define_one_ro(_name) \
631 static struct freq_attr _name = \
632 __ATTR(_name, 0444, show_##_name, NULL)
633
634 #define define_one_ro0400(_name) \
635 static struct freq_attr _name = \
636 __ATTR(_name, 0400, show_##_name, NULL)
637
638 #define define_one_rw(_name) \
639 static struct freq_attr _name = \
640 __ATTR(_name, 0644, show_##_name, store_##_name)
641
642 define_one_ro0400(cpuinfo_cur_freq);
643 define_one_ro(cpuinfo_min_freq);
644 define_one_ro(cpuinfo_max_freq);
645 define_one_ro(scaling_available_governors);
646 define_one_ro(scaling_driver);
647 define_one_ro(scaling_cur_freq);
648 define_one_ro(affected_cpus);
649 define_one_rw(scaling_min_freq);
650 define_one_rw(scaling_max_freq);
651 define_one_rw(scaling_governor);
652 define_one_rw(scaling_setspeed);
653
654 static struct attribute * default_attrs[] = {
655         &cpuinfo_min_freq.attr,
656         &cpuinfo_max_freq.attr,
657         &scaling_min_freq.attr,
658         &scaling_max_freq.attr,
659         &affected_cpus.attr,
660         &scaling_governor.attr,
661         &scaling_driver.attr,
662         &scaling_available_governors.attr,
663         &scaling_setspeed.attr,
664         NULL
665 };
666
667 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
668 #define to_attr(a) container_of(a,struct freq_attr,attr)
669
670 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
671 {
672         struct cpufreq_policy * policy = to_policy(kobj);
673         struct freq_attr * fattr = to_attr(attr);
674         ssize_t ret;
675         policy = cpufreq_cpu_get(policy->cpu);
676         if (!policy)
677                 return -EINVAL;
678
679         if (lock_policy_rwsem_read(policy->cpu) < 0)
680                 return -EINVAL;
681
682         if (fattr->show)
683                 ret = fattr->show(policy, buf);
684         else
685                 ret = -EIO;
686
687         unlock_policy_rwsem_read(policy->cpu);
688
689         cpufreq_cpu_put(policy);
690         return ret;
691 }
692
693 static ssize_t store(struct kobject * kobj, struct attribute * attr,
694                      const char * buf, size_t count)
695 {
696         struct cpufreq_policy * policy = to_policy(kobj);
697         struct freq_attr * fattr = to_attr(attr);
698         ssize_t ret;
699         policy = cpufreq_cpu_get(policy->cpu);
700         if (!policy)
701                 return -EINVAL;
702
703         if (lock_policy_rwsem_write(policy->cpu) < 0)
704                 return -EINVAL;
705
706         if (fattr->store)
707                 ret = fattr->store(policy, buf, count);
708         else
709                 ret = -EIO;
710
711         unlock_policy_rwsem_write(policy->cpu);
712
713         cpufreq_cpu_put(policy);
714         return ret;
715 }
716
717 static void cpufreq_sysfs_release(struct kobject * kobj)
718 {
719         struct cpufreq_policy * policy = to_policy(kobj);
720         dprintk("last reference is dropped\n");
721         complete(&policy->kobj_unregister);
722 }
723
724 static struct sysfs_ops sysfs_ops = {
725         .show   = show,
726         .store  = store,
727 };
728
729 static struct kobj_type ktype_cpufreq = {
730         .sysfs_ops      = &sysfs_ops,
731         .default_attrs  = default_attrs,
732         .release        = cpufreq_sysfs_release,
733 };
734
735
736 /**
737  * cpufreq_add_dev - add a CPU device
738  *
739  * Adds the cpufreq interface for a CPU device.
740  */
741 static int cpufreq_add_dev (struct sys_device * sys_dev)
742 {
743         unsigned int cpu = sys_dev->id;
744         int ret = 0;
745         struct cpufreq_policy new_policy;
746         struct cpufreq_policy *policy;
747         struct freq_attr **drv_attr;
748         struct sys_device *cpu_sys_dev;
749         unsigned long flags;
750         unsigned int j;
751 #ifdef CONFIG_SMP
752         struct cpufreq_policy *managed_policy;
753 #endif
754
755         if (cpu_is_offline(cpu))
756                 return 0;
757
758         cpufreq_debug_disable_ratelimit();
759         dprintk("adding CPU %u\n", cpu);
760
761 #ifdef CONFIG_SMP
762         /* check whether a different CPU already registered this
763          * CPU because it is in the same boat. */
764         policy = cpufreq_cpu_get(cpu);
765         if (unlikely(policy)) {
766                 cpufreq_cpu_put(policy);
767                 cpufreq_debug_enable_ratelimit();
768                 return 0;
769         }
770 #endif
771
772         if (!try_module_get(cpufreq_driver->owner)) {
773                 ret = -EINVAL;
774                 goto module_out;
775         }
776
777         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
778         if (!policy) {
779                 ret = -ENOMEM;
780                 goto nomem_out;
781         }
782
783         policy->cpu = cpu;
784         policy->cpus = cpumask_of_cpu(cpu);
785
786         /* Initially set CPU itself as the policy_cpu */
787         per_cpu(policy_cpu, cpu) = cpu;
788         lock_policy_rwsem_write(cpu);
789
790         init_completion(&policy->kobj_unregister);
791         INIT_WORK(&policy->update, handle_update);
792
793         /* Set governor before ->init, so that driver could check it */
794         policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
795         /* call driver. From then on the cpufreq must be able
796          * to accept all calls to ->verify and ->setpolicy for this CPU
797          */
798         ret = cpufreq_driver->init(policy);
799         if (ret) {
800                 dprintk("initialization failed\n");
801                 unlock_policy_rwsem_write(cpu);
802                 goto err_out;
803         }
804         policy->user_policy.min = policy->cpuinfo.min_freq;
805         policy->user_policy.max = policy->cpuinfo.max_freq;
806
807 #ifdef CONFIG_SMP
808
809 #ifdef CONFIG_HOTPLUG_CPU
810         if (cpufreq_cpu_governor[cpu]){
811                 policy->governor = cpufreq_cpu_governor[cpu];
812                 dprintk("Restoring governor %s for cpu %d\n",
813                        policy->governor->name, cpu);
814         }
815 #endif
816
817         for_each_cpu_mask(j, policy->cpus) {
818                 if (cpu == j)
819                         continue;
820
821                 /* check for existing affected CPUs.  They may not be aware
822                  * of it due to CPU Hotplug.
823                  */
824                 managed_policy = cpufreq_cpu_get(j);
825                 if (unlikely(managed_policy)) {
826
827                         /* Set proper policy_cpu */
828                         unlock_policy_rwsem_write(cpu);
829                         per_cpu(policy_cpu, cpu) = managed_policy->cpu;
830
831                         if (lock_policy_rwsem_write(cpu) < 0)
832                                 goto err_out_driver_exit;
833
834                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
835                         managed_policy->cpus = policy->cpus;
836                         cpufreq_cpu_data[cpu] = managed_policy;
837                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
838
839                         dprintk("CPU already managed, adding link\n");
840                         ret = sysfs_create_link(&sys_dev->kobj,
841                                                 &managed_policy->kobj,
842                                                 "cpufreq");
843                         if (ret) {
844                                 unlock_policy_rwsem_write(cpu);
845                                 goto err_out_driver_exit;
846                         }
847
848                         cpufreq_debug_enable_ratelimit();
849                         ret = 0;
850                         unlock_policy_rwsem_write(cpu);
851                         goto err_out_driver_exit; /* call driver->exit() */
852                 }
853         }
854 #endif
855         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
856
857         /* prepare interface data */
858         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq, &sys_dev->kobj,
859                                    "cpufreq");
860         if (ret) {
861                 unlock_policy_rwsem_write(cpu);
862                 goto err_out_driver_exit;
863         }
864         /* set up files for this cpu device */
865         drv_attr = cpufreq_driver->attr;
866         while ((drv_attr) && (*drv_attr)) {
867                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
868                 if (ret) {
869                         unlock_policy_rwsem_write(cpu);
870                         goto err_out_driver_exit;
871                 }
872                 drv_attr++;
873         }
874         if (cpufreq_driver->get){
875                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
876                 if (ret) {
877                         unlock_policy_rwsem_write(cpu);
878                         goto err_out_driver_exit;
879                 }
880         }
881         if (cpufreq_driver->target){
882                 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
883                 if (ret) {
884                         unlock_policy_rwsem_write(cpu);
885                         goto err_out_driver_exit;
886                 }
887         }
888
889         spin_lock_irqsave(&cpufreq_driver_lock, flags);
890         for_each_cpu_mask(j, policy->cpus) {
891                 cpufreq_cpu_data[j] = policy;
892                 per_cpu(policy_cpu, j) = policy->cpu;
893         }
894         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
895
896         /* symlink affected CPUs */
897         for_each_cpu_mask(j, policy->cpus) {
898                 if (j == cpu)
899                         continue;
900                 if (!cpu_online(j))
901                         continue;
902
903                 dprintk("CPU %u already managed, adding link\n", j);
904                 cpufreq_cpu_get(cpu);
905                 cpu_sys_dev = get_cpu_sysdev(j);
906                 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
907                                         "cpufreq");
908                 if (ret) {
909                         unlock_policy_rwsem_write(cpu);
910                         goto err_out_unregister;
911                 }
912         }
913
914         policy->governor = NULL; /* to assure that the starting sequence is
915                                   * run in cpufreq_set_policy */
916
917         /* set default policy */
918         ret = __cpufreq_set_policy(policy, &new_policy);
919         policy->user_policy.policy = policy->policy;
920         policy->user_policy.governor = policy->governor;
921
922         unlock_policy_rwsem_write(cpu);
923
924         if (ret) {
925                 dprintk("setting policy failed\n");
926                 goto err_out_unregister;
927         }
928
929         kobject_uevent(&policy->kobj, KOBJ_ADD);
930         module_put(cpufreq_driver->owner);
931         dprintk("initialization complete\n");
932         cpufreq_debug_enable_ratelimit();
933
934         return 0;
935
936
937 err_out_unregister:
938         spin_lock_irqsave(&cpufreq_driver_lock, flags);
939         for_each_cpu_mask(j, policy->cpus)
940                 cpufreq_cpu_data[j] = NULL;
941         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
942
943         kobject_put(&policy->kobj);
944         wait_for_completion(&policy->kobj_unregister);
945
946 err_out_driver_exit:
947         if (cpufreq_driver->exit)
948                 cpufreq_driver->exit(policy);
949
950 err_out:
951         kfree(policy);
952
953 nomem_out:
954         module_put(cpufreq_driver->owner);
955 module_out:
956         cpufreq_debug_enable_ratelimit();
957         return ret;
958 }
959
960
961 /**
962  * __cpufreq_remove_dev - remove a CPU device
963  *
964  * Removes the cpufreq interface for a CPU device.
965  * Caller should already have policy_rwsem in write mode for this CPU.
966  * This routine frees the rwsem before returning.
967  */
968 static int __cpufreq_remove_dev (struct sys_device * sys_dev)
969 {
970         unsigned int cpu = sys_dev->id;
971         unsigned long flags;
972         struct cpufreq_policy *data;
973 #ifdef CONFIG_SMP
974         struct sys_device *cpu_sys_dev;
975         unsigned int j;
976 #endif
977
978         cpufreq_debug_disable_ratelimit();
979         dprintk("unregistering CPU %u\n", cpu);
980
981         spin_lock_irqsave(&cpufreq_driver_lock, flags);
982         data = cpufreq_cpu_data[cpu];
983
984         if (!data) {
985                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
986                 cpufreq_debug_enable_ratelimit();
987                 unlock_policy_rwsem_write(cpu);
988                 return -EINVAL;
989         }
990         cpufreq_cpu_data[cpu] = NULL;
991
992
993 #ifdef CONFIG_SMP
994         /* if this isn't the CPU which is the parent of the kobj, we
995          * only need to unlink, put and exit
996          */
997         if (unlikely(cpu != data->cpu)) {
998                 dprintk("removing link\n");
999                 cpu_clear(cpu, data->cpus);
1000                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1001                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
1002                 cpufreq_cpu_put(data);
1003                 cpufreq_debug_enable_ratelimit();
1004                 unlock_policy_rwsem_write(cpu);
1005                 return 0;
1006         }
1007 #endif
1008
1009 #ifdef CONFIG_SMP
1010
1011 #ifdef CONFIG_HOTPLUG_CPU
1012         cpufreq_cpu_governor[cpu] = data->governor;
1013 #endif
1014
1015         /* if we have other CPUs still registered, we need to unlink them,
1016          * or else wait_for_completion below will lock up. Clean the
1017          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
1018          * links afterwards.
1019          */
1020         if (unlikely(cpus_weight(data->cpus) > 1)) {
1021                 for_each_cpu_mask(j, data->cpus) {
1022                         if (j == cpu)
1023                                 continue;
1024                         cpufreq_cpu_data[j] = NULL;
1025                 }
1026         }
1027
1028         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1029
1030         if (unlikely(cpus_weight(data->cpus) > 1)) {
1031                 for_each_cpu_mask(j, data->cpus) {
1032                         if (j == cpu)
1033                                 continue;
1034                         dprintk("removing link for cpu %u\n", j);
1035 #ifdef CONFIG_HOTPLUG_CPU
1036                         cpufreq_cpu_governor[j] = data->governor;
1037 #endif
1038                         cpu_sys_dev = get_cpu_sysdev(j);
1039                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
1040                         cpufreq_cpu_put(data);
1041                 }
1042         }
1043 #else
1044         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1045 #endif
1046
1047         if (cpufreq_driver->target)
1048                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1049
1050         unlock_policy_rwsem_write(cpu);
1051
1052         kobject_put(&data->kobj);
1053
1054         /* we need to make sure that the underlying kobj is actually
1055          * not referenced anymore by anybody before we proceed with
1056          * unloading.
1057          */
1058         dprintk("waiting for dropping of refcount\n");
1059         wait_for_completion(&data->kobj_unregister);
1060         dprintk("wait complete\n");
1061
1062         if (cpufreq_driver->exit)
1063                 cpufreq_driver->exit(data);
1064
1065         kfree(data);
1066
1067         cpufreq_debug_enable_ratelimit();
1068         return 0;
1069 }
1070
1071
1072 static int cpufreq_remove_dev (struct sys_device * sys_dev)
1073 {
1074         unsigned int cpu = sys_dev->id;
1075         int retval;
1076
1077         if (cpu_is_offline(cpu))
1078                 return 0;
1079
1080         if (unlikely(lock_policy_rwsem_write(cpu)))
1081                 BUG();
1082
1083         retval = __cpufreq_remove_dev(sys_dev);
1084         return retval;
1085 }
1086
1087
1088 static void handle_update(struct work_struct *work)
1089 {
1090         struct cpufreq_policy *policy =
1091                 container_of(work, struct cpufreq_policy, update);
1092         unsigned int cpu = policy->cpu;
1093         dprintk("handle_update for cpu %u called\n", cpu);
1094         cpufreq_update_policy(cpu);
1095 }
1096
1097 /**
1098  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1099  *      @cpu: cpu number
1100  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
1101  *      @new_freq: CPU frequency the CPU actually runs at
1102  *
1103  *      We adjust to current frequency first, and need to clean up later. So either call
1104  *      to cpufreq_update_policy() or schedule handle_update()).
1105  */
1106 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1107                                 unsigned int new_freq)
1108 {
1109         struct cpufreq_freqs freqs;
1110
1111         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
1112                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1113
1114         freqs.cpu = cpu;
1115         freqs.old = old_freq;
1116         freqs.new = new_freq;
1117         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1118         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1119 }
1120
1121
1122 /**
1123  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1124  * @cpu: CPU number
1125  *
1126  * This is the last known freq, without actually getting it from the driver.
1127  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1128  */
1129 unsigned int cpufreq_quick_get(unsigned int cpu)
1130 {
1131         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1132         unsigned int ret_freq = 0;
1133
1134         if (policy) {
1135                 ret_freq = policy->cur;
1136                 cpufreq_cpu_put(policy);
1137         }
1138
1139         return (ret_freq);
1140 }
1141 EXPORT_SYMBOL(cpufreq_quick_get);
1142
1143
1144 static unsigned int __cpufreq_get(unsigned int cpu)
1145 {
1146         struct cpufreq_policy *policy = cpufreq_cpu_data[cpu];
1147         unsigned int ret_freq = 0;
1148
1149         if (!cpufreq_driver->get)
1150                 return (ret_freq);
1151
1152         ret_freq = cpufreq_driver->get(cpu);
1153
1154         if (ret_freq && policy->cur &&
1155                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1156                 /* verify no discrepancy between actual and
1157                                         saved value exists */
1158                 if (unlikely(ret_freq != policy->cur)) {
1159                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1160                         schedule_work(&policy->update);
1161                 }
1162         }
1163
1164         return (ret_freq);
1165 }
1166
1167 /**
1168  * cpufreq_get - get the current CPU frequency (in kHz)
1169  * @cpu: CPU number
1170  *
1171  * Get the CPU current (static) CPU frequency
1172  */
1173 unsigned int cpufreq_get(unsigned int cpu)
1174 {
1175         unsigned int ret_freq = 0;
1176         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1177
1178         if (!policy)
1179                 goto out;
1180
1181         if (unlikely(lock_policy_rwsem_read(cpu)))
1182                 goto out_policy;
1183
1184         ret_freq = __cpufreq_get(cpu);
1185
1186         unlock_policy_rwsem_read(cpu);
1187
1188 out_policy:
1189         cpufreq_cpu_put(policy);
1190 out:
1191         return (ret_freq);
1192 }
1193 EXPORT_SYMBOL(cpufreq_get);
1194
1195
1196 /**
1197  *      cpufreq_suspend - let the low level driver prepare for suspend
1198  */
1199
1200 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
1201 {
1202         int cpu = sysdev->id;
1203         int ret = 0;
1204         unsigned int cur_freq = 0;
1205         struct cpufreq_policy *cpu_policy;
1206
1207         dprintk("suspending cpu %u\n", cpu);
1208
1209         if (!cpu_online(cpu))
1210                 return 0;
1211
1212         /* we may be lax here as interrupts are off. Nonetheless
1213          * we need to grab the correct cpu policy, as to check
1214          * whether we really run on this CPU.
1215          */
1216
1217         cpu_policy = cpufreq_cpu_get(cpu);
1218         if (!cpu_policy)
1219                 return -EINVAL;
1220
1221         /* only handle each CPU group once */
1222         if (unlikely(cpu_policy->cpu != cpu)) {
1223                 cpufreq_cpu_put(cpu_policy);
1224                 return 0;
1225         }
1226
1227         if (cpufreq_driver->suspend) {
1228                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1229                 if (ret) {
1230                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1231                                         "step on CPU %u\n", cpu_policy->cpu);
1232                         cpufreq_cpu_put(cpu_policy);
1233                         return ret;
1234                 }
1235         }
1236
1237
1238         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1239                 goto out;
1240
1241         if (cpufreq_driver->get)
1242                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1243
1244         if (!cur_freq || !cpu_policy->cur) {
1245                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1246                        "frequency is what timing core thinks it is.\n");
1247                 goto out;
1248         }
1249
1250         if (unlikely(cur_freq != cpu_policy->cur)) {
1251                 struct cpufreq_freqs freqs;
1252
1253                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1254                         dprintk("Warning: CPU frequency is %u, "
1255                                "cpufreq assumed %u kHz.\n",
1256                                cur_freq, cpu_policy->cur);
1257
1258                 freqs.cpu = cpu;
1259                 freqs.old = cpu_policy->cur;
1260                 freqs.new = cur_freq;
1261
1262                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
1263                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1264                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1265
1266                 cpu_policy->cur = cur_freq;
1267         }
1268
1269 out:
1270         cpufreq_cpu_put(cpu_policy);
1271         return 0;
1272 }
1273
1274 /**
1275  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1276  *
1277  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1278  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1279  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1280  *          restored.
1281  */
1282 static int cpufreq_resume(struct sys_device * sysdev)
1283 {
1284         int cpu = sysdev->id;
1285         int ret = 0;
1286         struct cpufreq_policy *cpu_policy;
1287
1288         dprintk("resuming cpu %u\n", cpu);
1289
1290         if (!cpu_online(cpu))
1291                 return 0;
1292
1293         /* we may be lax here as interrupts are off. Nonetheless
1294          * we need to grab the correct cpu policy, as to check
1295          * whether we really run on this CPU.
1296          */
1297
1298         cpu_policy = cpufreq_cpu_get(cpu);
1299         if (!cpu_policy)
1300                 return -EINVAL;
1301
1302         /* only handle each CPU group once */
1303         if (unlikely(cpu_policy->cpu != cpu)) {
1304                 cpufreq_cpu_put(cpu_policy);
1305                 return 0;
1306         }
1307
1308         if (cpufreq_driver->resume) {
1309                 ret = cpufreq_driver->resume(cpu_policy);
1310                 if (ret) {
1311                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1312                                         "step on CPU %u\n", cpu_policy->cpu);
1313                         cpufreq_cpu_put(cpu_policy);
1314                         return ret;
1315                 }
1316         }
1317
1318         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1319                 unsigned int cur_freq = 0;
1320
1321                 if (cpufreq_driver->get)
1322                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1323
1324                 if (!cur_freq || !cpu_policy->cur) {
1325                         printk(KERN_ERR "cpufreq: resume failed to assert "
1326                                         "current frequency is what timing core "
1327                                         "thinks it is.\n");
1328                         goto out;
1329                 }
1330
1331                 if (unlikely(cur_freq != cpu_policy->cur)) {
1332                         struct cpufreq_freqs freqs;
1333
1334                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1335                                 dprintk("Warning: CPU frequency "
1336                                        "is %u, cpufreq assumed %u kHz.\n",
1337                                        cur_freq, cpu_policy->cur);
1338
1339                         freqs.cpu = cpu;
1340                         freqs.old = cpu_policy->cur;
1341                         freqs.new = cur_freq;
1342
1343                         srcu_notifier_call_chain(
1344                                         &cpufreq_transition_notifier_list,
1345                                         CPUFREQ_RESUMECHANGE, &freqs);
1346                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1347
1348                         cpu_policy->cur = cur_freq;
1349                 }
1350         }
1351
1352 out:
1353         schedule_work(&cpu_policy->update);
1354         cpufreq_cpu_put(cpu_policy);
1355         return ret;
1356 }
1357
1358 static struct sysdev_driver cpufreq_sysdev_driver = {
1359         .add            = cpufreq_add_dev,
1360         .remove         = cpufreq_remove_dev,
1361         .suspend        = cpufreq_suspend,
1362         .resume         = cpufreq_resume,
1363 };
1364
1365
1366 /*********************************************************************
1367  *                     NOTIFIER LISTS INTERFACE                      *
1368  *********************************************************************/
1369
1370 /**
1371  *      cpufreq_register_notifier - register a driver with cpufreq
1372  *      @nb: notifier function to register
1373  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1374  *
1375  *      Add a driver to one of two lists: either a list of drivers that
1376  *      are notified about clock rate changes (once before and once after
1377  *      the transition), or a list of drivers that are notified about
1378  *      changes in cpufreq policy.
1379  *
1380  *      This function may sleep, and has the same return conditions as
1381  *      blocking_notifier_chain_register.
1382  */
1383 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1384 {
1385         int ret;
1386
1387         switch (list) {
1388         case CPUFREQ_TRANSITION_NOTIFIER:
1389                 ret = srcu_notifier_chain_register(
1390                                 &cpufreq_transition_notifier_list, nb);
1391                 break;
1392         case CPUFREQ_POLICY_NOTIFIER:
1393                 ret = blocking_notifier_chain_register(
1394                                 &cpufreq_policy_notifier_list, nb);
1395                 break;
1396         default:
1397                 ret = -EINVAL;
1398         }
1399
1400         return ret;
1401 }
1402 EXPORT_SYMBOL(cpufreq_register_notifier);
1403
1404
1405 /**
1406  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1407  *      @nb: notifier block to be unregistered
1408  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1409  *
1410  *      Remove a driver from the CPU frequency notifier list.
1411  *
1412  *      This function may sleep, and has the same return conditions as
1413  *      blocking_notifier_chain_unregister.
1414  */
1415 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1416 {
1417         int ret;
1418
1419         switch (list) {
1420         case CPUFREQ_TRANSITION_NOTIFIER:
1421                 ret = srcu_notifier_chain_unregister(
1422                                 &cpufreq_transition_notifier_list, nb);
1423                 break;
1424         case CPUFREQ_POLICY_NOTIFIER:
1425                 ret = blocking_notifier_chain_unregister(
1426                                 &cpufreq_policy_notifier_list, nb);
1427                 break;
1428         default:
1429                 ret = -EINVAL;
1430         }
1431
1432         return ret;
1433 }
1434 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1435
1436
1437 /*********************************************************************
1438  *                              GOVERNORS                            *
1439  *********************************************************************/
1440
1441
1442 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1443                             unsigned int target_freq,
1444                             unsigned int relation)
1445 {
1446         int retval = -EINVAL;
1447
1448         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1449                 target_freq, relation);
1450         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1451                 retval = cpufreq_driver->target(policy, target_freq, relation);
1452
1453         return retval;
1454 }
1455 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1456
1457 int cpufreq_driver_target(struct cpufreq_policy *policy,
1458                           unsigned int target_freq,
1459                           unsigned int relation)
1460 {
1461         int ret;
1462
1463         policy = cpufreq_cpu_get(policy->cpu);
1464         if (!policy)
1465                 return -EINVAL;
1466
1467         if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1468                 return -EINVAL;
1469
1470         ret = __cpufreq_driver_target(policy, target_freq, relation);
1471
1472         unlock_policy_rwsem_write(policy->cpu);
1473
1474         cpufreq_cpu_put(policy);
1475         return ret;
1476 }
1477 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1478
1479 int __cpufreq_driver_getavg(struct cpufreq_policy *policy)
1480 {
1481         int ret = 0;
1482
1483         policy = cpufreq_cpu_get(policy->cpu);
1484         if (!policy)
1485                 return -EINVAL;
1486
1487         if (cpu_online(policy->cpu) && cpufreq_driver->getavg)
1488                 ret = cpufreq_driver->getavg(policy->cpu);
1489
1490         cpufreq_cpu_put(policy);
1491         return ret;
1492 }
1493 EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1494
1495 /*
1496  * when "event" is CPUFREQ_GOV_LIMITS
1497  */
1498
1499 static int __cpufreq_governor(struct cpufreq_policy *policy,
1500                                         unsigned int event)
1501 {
1502         int ret;
1503
1504         /* Only must be defined when default governor is known to have latency
1505            restrictions, like e.g. conservative or ondemand.
1506            That this is the case is already ensured in Kconfig
1507         */
1508 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1509         struct cpufreq_governor *gov = &cpufreq_gov_performance;
1510 #else
1511         struct cpufreq_governor *gov = NULL;
1512 #endif
1513
1514         if (policy->governor->max_transition_latency &&
1515             policy->cpuinfo.transition_latency >
1516             policy->governor->max_transition_latency) {
1517                 if (!gov)
1518                         return -EINVAL;
1519                 else {
1520                         printk(KERN_WARNING "%s governor failed, too long"
1521                                " transition latency of HW, fallback"
1522                                " to %s governor\n",
1523                                policy->governor->name,
1524                                gov->name);
1525                         policy->governor = gov;
1526                 }
1527         }
1528
1529         if (!try_module_get(policy->governor->owner))
1530                 return -EINVAL;
1531
1532         dprintk("__cpufreq_governor for CPU %u, event %u\n",
1533                                                 policy->cpu, event);
1534         ret = policy->governor->governor(policy, event);
1535
1536         /* we keep one module reference alive for
1537                         each CPU governed by this CPU */
1538         if ((event != CPUFREQ_GOV_START) || ret)
1539                 module_put(policy->governor->owner);
1540         if ((event == CPUFREQ_GOV_STOP) && !ret)
1541                 module_put(policy->governor->owner);
1542
1543         return ret;
1544 }
1545
1546
1547 int cpufreq_register_governor(struct cpufreq_governor *governor)
1548 {
1549         int err;
1550
1551         if (!governor)
1552                 return -EINVAL;
1553
1554         mutex_lock(&cpufreq_governor_mutex);
1555
1556         err = -EBUSY;
1557         if (__find_governor(governor->name) == NULL) {
1558                 err = 0;
1559                 list_add(&governor->governor_list, &cpufreq_governor_list);
1560         }
1561
1562         mutex_unlock(&cpufreq_governor_mutex);
1563         return err;
1564 }
1565 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1566
1567
1568 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1569 {
1570         if (!governor)
1571                 return;
1572
1573         mutex_lock(&cpufreq_governor_mutex);
1574         list_del(&governor->governor_list);
1575         mutex_unlock(&cpufreq_governor_mutex);
1576         return;
1577 }
1578 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1579
1580
1581
1582 /*********************************************************************
1583  *                          POLICY INTERFACE                         *
1584  *********************************************************************/
1585
1586 /**
1587  * cpufreq_get_policy - get the current cpufreq_policy
1588  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1589  *
1590  * Reads the current cpufreq policy.
1591  */
1592 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1593 {
1594         struct cpufreq_policy *cpu_policy;
1595         if (!policy)
1596                 return -EINVAL;
1597
1598         cpu_policy = cpufreq_cpu_get(cpu);
1599         if (!cpu_policy)
1600                 return -EINVAL;
1601
1602         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1603
1604         cpufreq_cpu_put(cpu_policy);
1605         return 0;
1606 }
1607 EXPORT_SYMBOL(cpufreq_get_policy);
1608
1609
1610 /*
1611  * data   : current policy.
1612  * policy : policy to be set.
1613  */
1614 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1615                                 struct cpufreq_policy *policy)
1616 {
1617         int ret = 0;
1618
1619         cpufreq_debug_disable_ratelimit();
1620         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1621                 policy->min, policy->max);
1622
1623         memcpy(&policy->cpuinfo, &data->cpuinfo,
1624                                 sizeof(struct cpufreq_cpuinfo));
1625
1626         if (policy->min > data->max || policy->max < data->min) {
1627                 ret = -EINVAL;
1628                 goto error_out;
1629         }
1630
1631         /* verify the cpu speed can be set within this limit */
1632         ret = cpufreq_driver->verify(policy);
1633         if (ret)
1634                 goto error_out;
1635
1636         /* adjust if necessary - all reasons */
1637         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1638                         CPUFREQ_ADJUST, policy);
1639
1640         /* adjust if necessary - hardware incompatibility*/
1641         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1642                         CPUFREQ_INCOMPATIBLE, policy);
1643
1644         /* verify the cpu speed can be set within this limit,
1645            which might be different to the first one */
1646         ret = cpufreq_driver->verify(policy);
1647         if (ret)
1648                 goto error_out;
1649
1650         /* notification of the new policy */
1651         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1652                         CPUFREQ_NOTIFY, policy);
1653
1654         data->min = policy->min;
1655         data->max = policy->max;
1656
1657         dprintk("new min and max freqs are %u - %u kHz\n",
1658                                         data->min, data->max);
1659
1660         if (cpufreq_driver->setpolicy) {
1661                 data->policy = policy->policy;
1662                 dprintk("setting range\n");
1663                 ret = cpufreq_driver->setpolicy(policy);
1664         } else {
1665                 if (policy->governor != data->governor) {
1666                         /* save old, working values */
1667                         struct cpufreq_governor *old_gov = data->governor;
1668
1669                         dprintk("governor switch\n");
1670
1671                         /* end old governor */
1672                         if (data->governor)
1673                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1674
1675                         /* start new governor */
1676                         data->governor = policy->governor;
1677                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1678                                 /* new governor failed, so re-start old one */
1679                                 dprintk("starting governor %s failed\n",
1680                                                         data->governor->name);
1681                                 if (old_gov) {
1682                                         data->governor = old_gov;
1683                                         __cpufreq_governor(data,
1684                                                            CPUFREQ_GOV_START);
1685                                 }
1686                                 ret = -EINVAL;
1687                                 goto error_out;
1688                         }
1689                         /* might be a policy change, too, so fall through */
1690                 }
1691                 dprintk("governor: change or update limits\n");
1692                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1693         }
1694
1695 error_out:
1696         cpufreq_debug_enable_ratelimit();
1697         return ret;
1698 }
1699
1700 /**
1701  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1702  *      @cpu: CPU which shall be re-evaluated
1703  *
1704  *      Usefull for policy notifiers which have different necessities
1705  *      at different times.
1706  */
1707 int cpufreq_update_policy(unsigned int cpu)
1708 {
1709         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1710         struct cpufreq_policy policy;
1711         int ret = 0;
1712
1713         if (!data)
1714                 return -ENODEV;
1715
1716         if (unlikely(lock_policy_rwsem_write(cpu)))
1717                 return -EINVAL;
1718
1719         dprintk("updating policy for CPU %u\n", cpu);
1720         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1721         policy.min = data->user_policy.min;
1722         policy.max = data->user_policy.max;
1723         policy.policy = data->user_policy.policy;
1724         policy.governor = data->user_policy.governor;
1725
1726         /* BIOS might change freq behind our back
1727           -> ask driver for current freq and notify governors about a change */
1728         if (cpufreq_driver->get) {
1729                 policy.cur = cpufreq_driver->get(cpu);
1730                 if (!data->cur) {
1731                         dprintk("Driver did not initialize current freq");
1732                         data->cur = policy.cur;
1733                 } else {
1734                         if (data->cur != policy.cur)
1735                                 cpufreq_out_of_sync(cpu, data->cur,
1736                                                                 policy.cur);
1737                 }
1738         }
1739
1740         ret = __cpufreq_set_policy(data, &policy);
1741
1742         unlock_policy_rwsem_write(cpu);
1743
1744         cpufreq_cpu_put(data);
1745         return ret;
1746 }
1747 EXPORT_SYMBOL(cpufreq_update_policy);
1748
1749 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1750                                         unsigned long action, void *hcpu)
1751 {
1752         unsigned int cpu = (unsigned long)hcpu;
1753         struct sys_device *sys_dev;
1754
1755         sys_dev = get_cpu_sysdev(cpu);
1756         if (sys_dev) {
1757                 switch (action) {
1758                 case CPU_ONLINE:
1759                 case CPU_ONLINE_FROZEN:
1760                         cpufreq_add_dev(sys_dev);
1761                         break;
1762                 case CPU_DOWN_PREPARE:
1763                 case CPU_DOWN_PREPARE_FROZEN:
1764                         if (unlikely(lock_policy_rwsem_write(cpu)))
1765                                 BUG();
1766
1767                         __cpufreq_remove_dev(sys_dev);
1768                         break;
1769                 case CPU_DOWN_FAILED:
1770                 case CPU_DOWN_FAILED_FROZEN:
1771                         cpufreq_add_dev(sys_dev);
1772                         break;
1773                 }
1774         }
1775         return NOTIFY_OK;
1776 }
1777
1778 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1779 {
1780     .notifier_call = cpufreq_cpu_callback,
1781 };
1782
1783 /*********************************************************************
1784  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1785  *********************************************************************/
1786
1787 /**
1788  * cpufreq_register_driver - register a CPU Frequency driver
1789  * @driver_data: A struct cpufreq_driver containing the values#
1790  * submitted by the CPU Frequency driver.
1791  *
1792  *   Registers a CPU Frequency driver to this core code. This code
1793  * returns zero on success, -EBUSY when another driver got here first
1794  * (and isn't unregistered in the meantime).
1795  *
1796  */
1797 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1798 {
1799         unsigned long flags;
1800         int ret;
1801
1802         if (!driver_data || !driver_data->verify || !driver_data->init ||
1803             ((!driver_data->setpolicy) && (!driver_data->target)))
1804                 return -EINVAL;
1805
1806         dprintk("trying to register driver %s\n", driver_data->name);
1807
1808         if (driver_data->setpolicy)
1809                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1810
1811         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1812         if (cpufreq_driver) {
1813                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1814                 return -EBUSY;
1815         }
1816         cpufreq_driver = driver_data;
1817         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1818
1819         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1820
1821         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1822                 int i;
1823                 ret = -ENODEV;
1824
1825                 /* check for at least one working CPU */
1826                 for (i=0; i<NR_CPUS; i++)
1827                         if (cpufreq_cpu_data[i])
1828                                 ret = 0;
1829
1830                 /* if all ->init() calls failed, unregister */
1831                 if (ret) {
1832                         dprintk("no CPU initialized for driver %s\n",
1833                                                         driver_data->name);
1834                         sysdev_driver_unregister(&cpu_sysdev_class,
1835                                                 &cpufreq_sysdev_driver);
1836
1837                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1838                         cpufreq_driver = NULL;
1839                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1840                 }
1841         }
1842
1843         if (!ret) {
1844                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1845                 dprintk("driver %s up and running\n", driver_data->name);
1846                 cpufreq_debug_enable_ratelimit();
1847         }
1848
1849         return (ret);
1850 }
1851 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1852
1853
1854 /**
1855  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1856  *
1857  *    Unregister the current CPUFreq driver. Only call this if you have
1858  * the right to do so, i.e. if you have succeeded in initialising before!
1859  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1860  * currently not initialised.
1861  */
1862 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1863 {
1864         unsigned long flags;
1865
1866         cpufreq_debug_disable_ratelimit();
1867
1868         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1869                 cpufreq_debug_enable_ratelimit();
1870                 return -EINVAL;
1871         }
1872
1873         dprintk("unregistering driver %s\n", driver->name);
1874
1875         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1876         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1877
1878         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1879         cpufreq_driver = NULL;
1880         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1881
1882         return 0;
1883 }
1884 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1885
1886 static int __init cpufreq_core_init(void)
1887 {
1888         int cpu;
1889
1890         for_each_possible_cpu(cpu) {
1891                 per_cpu(policy_cpu, cpu) = -1;
1892                 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1893         }
1894         return 0;
1895 }
1896
1897 core_initcall(cpufreq_core_init);