digi_acceleport: Drag the driver kicking and screaming into coding style
[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
605 #define define_one_ro(_name) \
606 static struct freq_attr _name = \
607 __ATTR(_name, 0444, show_##_name, NULL)
608
609 #define define_one_ro0400(_name) \
610 static struct freq_attr _name = \
611 __ATTR(_name, 0400, show_##_name, NULL)
612
613 #define define_one_rw(_name) \
614 static struct freq_attr _name = \
615 __ATTR(_name, 0644, show_##_name, store_##_name)
616
617 define_one_ro0400(cpuinfo_cur_freq);
618 define_one_ro(cpuinfo_min_freq);
619 define_one_ro(cpuinfo_max_freq);
620 define_one_ro(scaling_available_governors);
621 define_one_ro(scaling_driver);
622 define_one_ro(scaling_cur_freq);
623 define_one_ro(affected_cpus);
624 define_one_rw(scaling_min_freq);
625 define_one_rw(scaling_max_freq);
626 define_one_rw(scaling_governor);
627
628 static struct attribute * default_attrs[] = {
629         &cpuinfo_min_freq.attr,
630         &cpuinfo_max_freq.attr,
631         &scaling_min_freq.attr,
632         &scaling_max_freq.attr,
633         &affected_cpus.attr,
634         &scaling_governor.attr,
635         &scaling_driver.attr,
636         &scaling_available_governors.attr,
637         NULL
638 };
639
640 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
641 #define to_attr(a) container_of(a,struct freq_attr,attr)
642
643 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
644 {
645         struct cpufreq_policy * policy = to_policy(kobj);
646         struct freq_attr * fattr = to_attr(attr);
647         ssize_t ret;
648         policy = cpufreq_cpu_get(policy->cpu);
649         if (!policy)
650                 return -EINVAL;
651
652         if (lock_policy_rwsem_read(policy->cpu) < 0)
653                 return -EINVAL;
654
655         if (fattr->show)
656                 ret = fattr->show(policy, buf);
657         else
658                 ret = -EIO;
659
660         unlock_policy_rwsem_read(policy->cpu);
661
662         cpufreq_cpu_put(policy);
663         return ret;
664 }
665
666 static ssize_t store(struct kobject * kobj, struct attribute * attr,
667                      const char * buf, size_t count)
668 {
669         struct cpufreq_policy * policy = to_policy(kobj);
670         struct freq_attr * fattr = to_attr(attr);
671         ssize_t ret;
672         policy = cpufreq_cpu_get(policy->cpu);
673         if (!policy)
674                 return -EINVAL;
675
676         if (lock_policy_rwsem_write(policy->cpu) < 0)
677                 return -EINVAL;
678
679         if (fattr->store)
680                 ret = fattr->store(policy, buf, count);
681         else
682                 ret = -EIO;
683
684         unlock_policy_rwsem_write(policy->cpu);
685
686         cpufreq_cpu_put(policy);
687         return ret;
688 }
689
690 static void cpufreq_sysfs_release(struct kobject * kobj)
691 {
692         struct cpufreq_policy * policy = to_policy(kobj);
693         dprintk("last reference is dropped\n");
694         complete(&policy->kobj_unregister);
695 }
696
697 static struct sysfs_ops sysfs_ops = {
698         .show   = show,
699         .store  = store,
700 };
701
702 static struct kobj_type ktype_cpufreq = {
703         .sysfs_ops      = &sysfs_ops,
704         .default_attrs  = default_attrs,
705         .release        = cpufreq_sysfs_release,
706 };
707
708
709 /**
710  * cpufreq_add_dev - add a CPU device
711  *
712  * Adds the cpufreq interface for a CPU device.
713  */
714 static int cpufreq_add_dev (struct sys_device * sys_dev)
715 {
716         unsigned int cpu = sys_dev->id;
717         int ret = 0;
718         struct cpufreq_policy new_policy;
719         struct cpufreq_policy *policy;
720         struct freq_attr **drv_attr;
721         struct sys_device *cpu_sys_dev;
722         unsigned long flags;
723         unsigned int j;
724 #ifdef CONFIG_SMP
725         struct cpufreq_policy *managed_policy;
726 #endif
727
728         if (cpu_is_offline(cpu))
729                 return 0;
730
731         cpufreq_debug_disable_ratelimit();
732         dprintk("adding CPU %u\n", cpu);
733
734 #ifdef CONFIG_SMP
735         /* check whether a different CPU already registered this
736          * CPU because it is in the same boat. */
737         policy = cpufreq_cpu_get(cpu);
738         if (unlikely(policy)) {
739                 cpufreq_cpu_put(policy);
740                 cpufreq_debug_enable_ratelimit();
741                 return 0;
742         }
743 #endif
744
745         if (!try_module_get(cpufreq_driver->owner)) {
746                 ret = -EINVAL;
747                 goto module_out;
748         }
749
750         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
751         if (!policy) {
752                 ret = -ENOMEM;
753                 goto nomem_out;
754         }
755
756         policy->cpu = cpu;
757         policy->cpus = cpumask_of_cpu(cpu);
758
759         /* Initially set CPU itself as the policy_cpu */
760         per_cpu(policy_cpu, cpu) = cpu;
761         lock_policy_rwsem_write(cpu);
762
763         init_completion(&policy->kobj_unregister);
764         INIT_WORK(&policy->update, handle_update);
765
766         /* call driver. From then on the cpufreq must be able
767          * to accept all calls to ->verify and ->setpolicy for this CPU
768          */
769         ret = cpufreq_driver->init(policy);
770         if (ret) {
771                 dprintk("initialization failed\n");
772                 unlock_policy_rwsem_write(cpu);
773                 goto err_out;
774         }
775         policy->user_policy.min = policy->cpuinfo.min_freq;
776         policy->user_policy.max = policy->cpuinfo.max_freq;
777
778 #ifdef CONFIG_SMP
779
780 #ifdef CONFIG_HOTPLUG_CPU
781         if (cpufreq_cpu_governor[cpu]){
782                 policy->governor = cpufreq_cpu_governor[cpu];
783                 dprintk("Restoring governor %s for cpu %d\n",
784                        policy->governor->name, cpu);
785         }
786 #endif
787
788         for_each_cpu_mask(j, policy->cpus) {
789                 if (cpu == j)
790                         continue;
791
792                 /* check for existing affected CPUs.  They may not be aware
793                  * of it due to CPU Hotplug.
794                  */
795                 managed_policy = cpufreq_cpu_get(j);
796                 if (unlikely(managed_policy)) {
797
798                         /* Set proper policy_cpu */
799                         unlock_policy_rwsem_write(cpu);
800                         per_cpu(policy_cpu, cpu) = managed_policy->cpu;
801
802                         if (lock_policy_rwsem_write(cpu) < 0)
803                                 goto err_out_driver_exit;
804
805                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
806                         managed_policy->cpus = policy->cpus;
807                         cpufreq_cpu_data[cpu] = managed_policy;
808                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
809
810                         dprintk("CPU already managed, adding link\n");
811                         ret = sysfs_create_link(&sys_dev->kobj,
812                                                 &managed_policy->kobj,
813                                                 "cpufreq");
814                         if (ret) {
815                                 unlock_policy_rwsem_write(cpu);
816                                 goto err_out_driver_exit;
817                         }
818
819                         cpufreq_debug_enable_ratelimit();
820                         ret = 0;
821                         unlock_policy_rwsem_write(cpu);
822                         goto err_out_driver_exit; /* call driver->exit() */
823                 }
824         }
825 #endif
826         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
827
828         /* prepare interface data */
829         policy->kobj.parent = &sys_dev->kobj;
830         policy->kobj.ktype = &ktype_cpufreq;
831         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
832
833         ret = kobject_register(&policy->kobj);
834         if (ret) {
835                 unlock_policy_rwsem_write(cpu);
836                 goto err_out_driver_exit;
837         }
838         /* set up files for this cpu device */
839         drv_attr = cpufreq_driver->attr;
840         while ((drv_attr) && (*drv_attr)) {
841                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
842                 if (ret)
843                         goto err_out_driver_exit;
844                 drv_attr++;
845         }
846         if (cpufreq_driver->get){
847                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
848                 if (ret)
849                         goto err_out_driver_exit;
850         }
851         if (cpufreq_driver->target){
852                 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
853                 if (ret)
854                         goto err_out_driver_exit;
855         }
856
857         spin_lock_irqsave(&cpufreq_driver_lock, flags);
858         for_each_cpu_mask(j, policy->cpus) {
859                 cpufreq_cpu_data[j] = policy;
860                 per_cpu(policy_cpu, j) = policy->cpu;
861         }
862         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
863
864         /* symlink affected CPUs */
865         for_each_cpu_mask(j, policy->cpus) {
866                 if (j == cpu)
867                         continue;
868                 if (!cpu_online(j))
869                         continue;
870
871                 dprintk("CPU %u already managed, adding link\n", j);
872                 cpufreq_cpu_get(cpu);
873                 cpu_sys_dev = get_cpu_sysdev(j);
874                 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
875                                         "cpufreq");
876                 if (ret) {
877                         unlock_policy_rwsem_write(cpu);
878                         goto err_out_unregister;
879                 }
880         }
881
882         policy->governor = NULL; /* to assure that the starting sequence is
883                                   * run in cpufreq_set_policy */
884
885         /* set default policy */
886         ret = __cpufreq_set_policy(policy, &new_policy);
887         policy->user_policy.policy = policy->policy;
888         policy->user_policy.governor = policy->governor;
889
890         unlock_policy_rwsem_write(cpu);
891
892         if (ret) {
893                 dprintk("setting policy failed\n");
894                 goto err_out_unregister;
895         }
896
897         module_put(cpufreq_driver->owner);
898         dprintk("initialization complete\n");
899         cpufreq_debug_enable_ratelimit();
900
901         return 0;
902
903
904 err_out_unregister:
905         spin_lock_irqsave(&cpufreq_driver_lock, flags);
906         for_each_cpu_mask(j, policy->cpus)
907                 cpufreq_cpu_data[j] = NULL;
908         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
909
910         kobject_unregister(&policy->kobj);
911         wait_for_completion(&policy->kobj_unregister);
912
913 err_out_driver_exit:
914         if (cpufreq_driver->exit)
915                 cpufreq_driver->exit(policy);
916
917 err_out:
918         kfree(policy);
919
920 nomem_out:
921         module_put(cpufreq_driver->owner);
922 module_out:
923         cpufreq_debug_enable_ratelimit();
924         return ret;
925 }
926
927
928 /**
929  * __cpufreq_remove_dev - remove a CPU device
930  *
931  * Removes the cpufreq interface for a CPU device.
932  * Caller should already have policy_rwsem in write mode for this CPU.
933  * This routine frees the rwsem before returning.
934  */
935 static int __cpufreq_remove_dev (struct sys_device * sys_dev)
936 {
937         unsigned int cpu = sys_dev->id;
938         unsigned long flags;
939         struct cpufreq_policy *data;
940 #ifdef CONFIG_SMP
941         struct sys_device *cpu_sys_dev;
942         unsigned int j;
943 #endif
944
945         cpufreq_debug_disable_ratelimit();
946         dprintk("unregistering CPU %u\n", cpu);
947
948         spin_lock_irqsave(&cpufreq_driver_lock, flags);
949         data = cpufreq_cpu_data[cpu];
950
951         if (!data) {
952                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
953                 cpufreq_debug_enable_ratelimit();
954                 unlock_policy_rwsem_write(cpu);
955                 return -EINVAL;
956         }
957         cpufreq_cpu_data[cpu] = NULL;
958
959
960 #ifdef CONFIG_SMP
961         /* if this isn't the CPU which is the parent of the kobj, we
962          * only need to unlink, put and exit
963          */
964         if (unlikely(cpu != data->cpu)) {
965                 dprintk("removing link\n");
966                 cpu_clear(cpu, data->cpus);
967                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
968                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
969                 cpufreq_cpu_put(data);
970                 cpufreq_debug_enable_ratelimit();
971                 unlock_policy_rwsem_write(cpu);
972                 return 0;
973         }
974 #endif
975
976
977         if (!kobject_get(&data->kobj)) {
978                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
979                 cpufreq_debug_enable_ratelimit();
980                 unlock_policy_rwsem_write(cpu);
981                 return -EFAULT;
982         }
983
984 #ifdef CONFIG_SMP
985
986 #ifdef CONFIG_HOTPLUG_CPU
987         cpufreq_cpu_governor[cpu] = data->governor;
988 #endif
989
990         /* if we have other CPUs still registered, we need to unlink them,
991          * or else wait_for_completion below will lock up. Clean the
992          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
993          * links afterwards.
994          */
995         if (unlikely(cpus_weight(data->cpus) > 1)) {
996                 for_each_cpu_mask(j, data->cpus) {
997                         if (j == cpu)
998                                 continue;
999                         cpufreq_cpu_data[j] = NULL;
1000                 }
1001         }
1002
1003         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1004
1005         if (unlikely(cpus_weight(data->cpus) > 1)) {
1006                 for_each_cpu_mask(j, data->cpus) {
1007                         if (j == cpu)
1008                                 continue;
1009                         dprintk("removing link for cpu %u\n", j);
1010 #ifdef CONFIG_HOTPLUG_CPU
1011                         cpufreq_cpu_governor[j] = data->governor;
1012 #endif
1013                         cpu_sys_dev = get_cpu_sysdev(j);
1014                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
1015                         cpufreq_cpu_put(data);
1016                 }
1017         }
1018 #else
1019         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1020 #endif
1021
1022         if (cpufreq_driver->target)
1023                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1024
1025         unlock_policy_rwsem_write(cpu);
1026
1027         kobject_unregister(&data->kobj);
1028
1029         kobject_put(&data->kobj);
1030
1031         /* we need to make sure that the underlying kobj is actually
1032          * not referenced anymore by anybody before we proceed with
1033          * unloading.
1034          */
1035         dprintk("waiting for dropping of refcount\n");
1036         wait_for_completion(&data->kobj_unregister);
1037         dprintk("wait complete\n");
1038
1039         if (cpufreq_driver->exit)
1040                 cpufreq_driver->exit(data);
1041
1042         kfree(data);
1043
1044         cpufreq_debug_enable_ratelimit();
1045         return 0;
1046 }
1047
1048
1049 static int cpufreq_remove_dev (struct sys_device * sys_dev)
1050 {
1051         unsigned int cpu = sys_dev->id;
1052         int retval;
1053
1054         if (cpu_is_offline(cpu))
1055                 return 0;
1056
1057         if (unlikely(lock_policy_rwsem_write(cpu)))
1058                 BUG();
1059
1060         retval = __cpufreq_remove_dev(sys_dev);
1061         return retval;
1062 }
1063
1064
1065 static void handle_update(struct work_struct *work)
1066 {
1067         struct cpufreq_policy *policy =
1068                 container_of(work, struct cpufreq_policy, update);
1069         unsigned int cpu = policy->cpu;
1070         dprintk("handle_update for cpu %u called\n", cpu);
1071         cpufreq_update_policy(cpu);
1072 }
1073
1074 /**
1075  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1076  *      @cpu: cpu number
1077  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
1078  *      @new_freq: CPU frequency the CPU actually runs at
1079  *
1080  *      We adjust to current frequency first, and need to clean up later. So either call
1081  *      to cpufreq_update_policy() or schedule handle_update()).
1082  */
1083 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1084                                 unsigned int new_freq)
1085 {
1086         struct cpufreq_freqs freqs;
1087
1088         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
1089                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1090
1091         freqs.cpu = cpu;
1092         freqs.old = old_freq;
1093         freqs.new = new_freq;
1094         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1095         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1096 }
1097
1098
1099 /**
1100  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1101  * @cpu: CPU number
1102  *
1103  * This is the last known freq, without actually getting it from the driver.
1104  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1105  */
1106 unsigned int cpufreq_quick_get(unsigned int cpu)
1107 {
1108         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1109         unsigned int ret_freq = 0;
1110
1111         if (policy) {
1112                 if (unlikely(lock_policy_rwsem_read(cpu)))
1113                         return ret_freq;
1114
1115                 ret_freq = policy->cur;
1116
1117                 unlock_policy_rwsem_read(cpu);
1118                 cpufreq_cpu_put(policy);
1119         }
1120
1121         return (ret_freq);
1122 }
1123 EXPORT_SYMBOL(cpufreq_quick_get);
1124
1125
1126 static unsigned int __cpufreq_get(unsigned int cpu)
1127 {
1128         struct cpufreq_policy *policy = cpufreq_cpu_data[cpu];
1129         unsigned int ret_freq = 0;
1130
1131         if (!cpufreq_driver->get)
1132                 return (ret_freq);
1133
1134         ret_freq = cpufreq_driver->get(cpu);
1135
1136         if (ret_freq && policy->cur &&
1137                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1138                 /* verify no discrepancy between actual and
1139                                         saved value exists */
1140                 if (unlikely(ret_freq != policy->cur)) {
1141                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1142                         schedule_work(&policy->update);
1143                 }
1144         }
1145
1146         return (ret_freq);
1147 }
1148
1149 /**
1150  * cpufreq_get - get the current CPU frequency (in kHz)
1151  * @cpu: CPU number
1152  *
1153  * Get the CPU current (static) CPU frequency
1154  */
1155 unsigned int cpufreq_get(unsigned int cpu)
1156 {
1157         unsigned int ret_freq = 0;
1158         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1159
1160         if (!policy)
1161                 goto out;
1162
1163         if (unlikely(lock_policy_rwsem_read(cpu)))
1164                 goto out_policy;
1165
1166         ret_freq = __cpufreq_get(cpu);
1167
1168         unlock_policy_rwsem_read(cpu);
1169
1170 out_policy:
1171         cpufreq_cpu_put(policy);
1172 out:
1173         return (ret_freq);
1174 }
1175 EXPORT_SYMBOL(cpufreq_get);
1176
1177
1178 /**
1179  *      cpufreq_suspend - let the low level driver prepare for suspend
1180  */
1181
1182 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
1183 {
1184         int cpu = sysdev->id;
1185         int ret = 0;
1186         unsigned int cur_freq = 0;
1187         struct cpufreq_policy *cpu_policy;
1188
1189         dprintk("suspending cpu %u\n", cpu);
1190
1191         if (!cpu_online(cpu))
1192                 return 0;
1193
1194         /* we may be lax here as interrupts are off. Nonetheless
1195          * we need to grab the correct cpu policy, as to check
1196          * whether we really run on this CPU.
1197          */
1198
1199         cpu_policy = cpufreq_cpu_get(cpu);
1200         if (!cpu_policy)
1201                 return -EINVAL;
1202
1203         /* only handle each CPU group once */
1204         if (unlikely(cpu_policy->cpu != cpu)) {
1205                 cpufreq_cpu_put(cpu_policy);
1206                 return 0;
1207         }
1208
1209         if (cpufreq_driver->suspend) {
1210                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1211                 if (ret) {
1212                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1213                                         "step on CPU %u\n", cpu_policy->cpu);
1214                         cpufreq_cpu_put(cpu_policy);
1215                         return ret;
1216                 }
1217         }
1218
1219
1220         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1221                 goto out;
1222
1223         if (cpufreq_driver->get)
1224                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1225
1226         if (!cur_freq || !cpu_policy->cur) {
1227                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1228                        "frequency is what timing core thinks it is.\n");
1229                 goto out;
1230         }
1231
1232         if (unlikely(cur_freq != cpu_policy->cur)) {
1233                 struct cpufreq_freqs freqs;
1234
1235                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1236                         dprintk("Warning: CPU frequency is %u, "
1237                                "cpufreq assumed %u kHz.\n",
1238                                cur_freq, cpu_policy->cur);
1239
1240                 freqs.cpu = cpu;
1241                 freqs.old = cpu_policy->cur;
1242                 freqs.new = cur_freq;
1243
1244                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
1245                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1246                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1247
1248                 cpu_policy->cur = cur_freq;
1249         }
1250
1251 out:
1252         cpufreq_cpu_put(cpu_policy);
1253         return 0;
1254 }
1255
1256 /**
1257  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1258  *
1259  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1260  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1261  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1262  *          restored.
1263  */
1264 static int cpufreq_resume(struct sys_device * sysdev)
1265 {
1266         int cpu = sysdev->id;
1267         int ret = 0;
1268         struct cpufreq_policy *cpu_policy;
1269
1270         dprintk("resuming cpu %u\n", cpu);
1271
1272         if (!cpu_online(cpu))
1273                 return 0;
1274
1275         /* we may be lax here as interrupts are off. Nonetheless
1276          * we need to grab the correct cpu policy, as to check
1277          * whether we really run on this CPU.
1278          */
1279
1280         cpu_policy = cpufreq_cpu_get(cpu);
1281         if (!cpu_policy)
1282                 return -EINVAL;
1283
1284         /* only handle each CPU group once */
1285         if (unlikely(cpu_policy->cpu != cpu)) {
1286                 cpufreq_cpu_put(cpu_policy);
1287                 return 0;
1288         }
1289
1290         if (cpufreq_driver->resume) {
1291                 ret = cpufreq_driver->resume(cpu_policy);
1292                 if (ret) {
1293                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1294                                         "step on CPU %u\n", cpu_policy->cpu);
1295                         cpufreq_cpu_put(cpu_policy);
1296                         return ret;
1297                 }
1298         }
1299
1300         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1301                 unsigned int cur_freq = 0;
1302
1303                 if (cpufreq_driver->get)
1304                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1305
1306                 if (!cur_freq || !cpu_policy->cur) {
1307                         printk(KERN_ERR "cpufreq: resume failed to assert "
1308                                         "current frequency is what timing core "
1309                                         "thinks it is.\n");
1310                         goto out;
1311                 }
1312
1313                 if (unlikely(cur_freq != cpu_policy->cur)) {
1314                         struct cpufreq_freqs freqs;
1315
1316                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1317                                 dprintk("Warning: CPU frequency"
1318                                        "is %u, cpufreq assumed %u kHz.\n",
1319                                        cur_freq, cpu_policy->cur);
1320
1321                         freqs.cpu = cpu;
1322                         freqs.old = cpu_policy->cur;
1323                         freqs.new = cur_freq;
1324
1325                         srcu_notifier_call_chain(
1326                                         &cpufreq_transition_notifier_list,
1327                                         CPUFREQ_RESUMECHANGE, &freqs);
1328                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1329
1330                         cpu_policy->cur = cur_freq;
1331                 }
1332         }
1333
1334 out:
1335         schedule_work(&cpu_policy->update);
1336         cpufreq_cpu_put(cpu_policy);
1337         return ret;
1338 }
1339
1340 static struct sysdev_driver cpufreq_sysdev_driver = {
1341         .add            = cpufreq_add_dev,
1342         .remove         = cpufreq_remove_dev,
1343         .suspend        = cpufreq_suspend,
1344         .resume         = cpufreq_resume,
1345 };
1346
1347
1348 /*********************************************************************
1349  *                     NOTIFIER LISTS INTERFACE                      *
1350  *********************************************************************/
1351
1352 /**
1353  *      cpufreq_register_notifier - register a driver with cpufreq
1354  *      @nb: notifier function to register
1355  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1356  *
1357  *      Add a driver to one of two lists: either a list of drivers that
1358  *      are notified about clock rate changes (once before and once after
1359  *      the transition), or a list of drivers that are notified about
1360  *      changes in cpufreq policy.
1361  *
1362  *      This function may sleep, and has the same return conditions as
1363  *      blocking_notifier_chain_register.
1364  */
1365 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1366 {
1367         int ret;
1368
1369         switch (list) {
1370         case CPUFREQ_TRANSITION_NOTIFIER:
1371                 ret = srcu_notifier_chain_register(
1372                                 &cpufreq_transition_notifier_list, nb);
1373                 break;
1374         case CPUFREQ_POLICY_NOTIFIER:
1375                 ret = blocking_notifier_chain_register(
1376                                 &cpufreq_policy_notifier_list, nb);
1377                 break;
1378         default:
1379                 ret = -EINVAL;
1380         }
1381
1382         return ret;
1383 }
1384 EXPORT_SYMBOL(cpufreq_register_notifier);
1385
1386
1387 /**
1388  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1389  *      @nb: notifier block to be unregistered
1390  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1391  *
1392  *      Remove a driver from the CPU frequency notifier list.
1393  *
1394  *      This function may sleep, and has the same return conditions as
1395  *      blocking_notifier_chain_unregister.
1396  */
1397 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1398 {
1399         int ret;
1400
1401         switch (list) {
1402         case CPUFREQ_TRANSITION_NOTIFIER:
1403                 ret = srcu_notifier_chain_unregister(
1404                                 &cpufreq_transition_notifier_list, nb);
1405                 break;
1406         case CPUFREQ_POLICY_NOTIFIER:
1407                 ret = blocking_notifier_chain_unregister(
1408                                 &cpufreq_policy_notifier_list, nb);
1409                 break;
1410         default:
1411                 ret = -EINVAL;
1412         }
1413
1414         return ret;
1415 }
1416 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1417
1418
1419 /*********************************************************************
1420  *                              GOVERNORS                            *
1421  *********************************************************************/
1422
1423
1424 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1425                             unsigned int target_freq,
1426                             unsigned int relation)
1427 {
1428         int retval = -EINVAL;
1429
1430         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1431                 target_freq, relation);
1432         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1433                 retval = cpufreq_driver->target(policy, target_freq, relation);
1434
1435         return retval;
1436 }
1437 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1438
1439 int cpufreq_driver_target(struct cpufreq_policy *policy,
1440                           unsigned int target_freq,
1441                           unsigned int relation)
1442 {
1443         int ret;
1444
1445         policy = cpufreq_cpu_get(policy->cpu);
1446         if (!policy)
1447                 return -EINVAL;
1448
1449         if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1450                 return -EINVAL;
1451
1452         ret = __cpufreq_driver_target(policy, target_freq, relation);
1453
1454         unlock_policy_rwsem_write(policy->cpu);
1455
1456         cpufreq_cpu_put(policy);
1457         return ret;
1458 }
1459 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1460
1461 int __cpufreq_driver_getavg(struct cpufreq_policy *policy)
1462 {
1463         int ret = 0;
1464
1465         policy = cpufreq_cpu_get(policy->cpu);
1466         if (!policy)
1467                 return -EINVAL;
1468
1469         if (cpu_online(policy->cpu) && cpufreq_driver->getavg)
1470                 ret = cpufreq_driver->getavg(policy->cpu);
1471
1472         cpufreq_cpu_put(policy);
1473         return ret;
1474 }
1475 EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1476
1477 /*
1478  * when "event" is CPUFREQ_GOV_LIMITS
1479  */
1480
1481 static int __cpufreq_governor(struct cpufreq_policy *policy,
1482                                         unsigned int event)
1483 {
1484         int ret;
1485
1486         if (!try_module_get(policy->governor->owner))
1487                 return -EINVAL;
1488
1489         dprintk("__cpufreq_governor for CPU %u, event %u\n",
1490                                                 policy->cpu, event);
1491         ret = policy->governor->governor(policy, event);
1492
1493         /* we keep one module reference alive for
1494                         each CPU governed by this CPU */
1495         if ((event != CPUFREQ_GOV_START) || ret)
1496                 module_put(policy->governor->owner);
1497         if ((event == CPUFREQ_GOV_STOP) && !ret)
1498                 module_put(policy->governor->owner);
1499
1500         return ret;
1501 }
1502
1503
1504 int cpufreq_register_governor(struct cpufreq_governor *governor)
1505 {
1506         int err;
1507
1508         if (!governor)
1509                 return -EINVAL;
1510
1511         mutex_lock(&cpufreq_governor_mutex);
1512
1513         err = -EBUSY;
1514         if (__find_governor(governor->name) == NULL) {
1515                 err = 0;
1516                 list_add(&governor->governor_list, &cpufreq_governor_list);
1517         }
1518
1519         mutex_unlock(&cpufreq_governor_mutex);
1520         return err;
1521 }
1522 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1523
1524
1525 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1526 {
1527         if (!governor)
1528                 return;
1529
1530         mutex_lock(&cpufreq_governor_mutex);
1531         list_del(&governor->governor_list);
1532         mutex_unlock(&cpufreq_governor_mutex);
1533         return;
1534 }
1535 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1536
1537
1538
1539 /*********************************************************************
1540  *                          POLICY INTERFACE                         *
1541  *********************************************************************/
1542
1543 /**
1544  * cpufreq_get_policy - get the current cpufreq_policy
1545  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1546  *
1547  * Reads the current cpufreq policy.
1548  */
1549 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1550 {
1551         struct cpufreq_policy *cpu_policy;
1552         if (!policy)
1553                 return -EINVAL;
1554
1555         cpu_policy = cpufreq_cpu_get(cpu);
1556         if (!cpu_policy)
1557                 return -EINVAL;
1558
1559         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1560
1561         cpufreq_cpu_put(cpu_policy);
1562         return 0;
1563 }
1564 EXPORT_SYMBOL(cpufreq_get_policy);
1565
1566
1567 /*
1568  * data   : current policy.
1569  * policy : policy to be set.
1570  */
1571 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1572                                 struct cpufreq_policy *policy)
1573 {
1574         int ret = 0;
1575
1576         cpufreq_debug_disable_ratelimit();
1577         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1578                 policy->min, policy->max);
1579
1580         memcpy(&policy->cpuinfo, &data->cpuinfo,
1581                                 sizeof(struct cpufreq_cpuinfo));
1582
1583         if (policy->min > data->min && policy->min > policy->max) {
1584                 ret = -EINVAL;
1585                 goto error_out;
1586         }
1587
1588         /* verify the cpu speed can be set within this limit */
1589         ret = cpufreq_driver->verify(policy);
1590         if (ret)
1591                 goto error_out;
1592
1593         /* adjust if necessary - all reasons */
1594         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1595                         CPUFREQ_ADJUST, policy);
1596
1597         /* adjust if necessary - hardware incompatibility*/
1598         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1599                         CPUFREQ_INCOMPATIBLE, policy);
1600
1601         /* verify the cpu speed can be set within this limit,
1602            which might be different to the first one */
1603         ret = cpufreq_driver->verify(policy);
1604         if (ret)
1605                 goto error_out;
1606
1607         /* notification of the new policy */
1608         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1609                         CPUFREQ_NOTIFY, policy);
1610
1611         data->min = policy->min;
1612         data->max = policy->max;
1613
1614         dprintk("new min and max freqs are %u - %u kHz\n",
1615                                         data->min, data->max);
1616
1617         if (cpufreq_driver->setpolicy) {
1618                 data->policy = policy->policy;
1619                 dprintk("setting range\n");
1620                 ret = cpufreq_driver->setpolicy(policy);
1621         } else {
1622                 if (policy->governor != data->governor) {
1623                         /* save old, working values */
1624                         struct cpufreq_governor *old_gov = data->governor;
1625
1626                         dprintk("governor switch\n");
1627
1628                         /* end old governor */
1629                         if (data->governor)
1630                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1631
1632                         /* start new governor */
1633                         data->governor = policy->governor;
1634                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1635                                 /* new governor failed, so re-start old one */
1636                                 dprintk("starting governor %s failed\n",
1637                                                         data->governor->name);
1638                                 if (old_gov) {
1639                                         data->governor = old_gov;
1640                                         __cpufreq_governor(data,
1641                                                            CPUFREQ_GOV_START);
1642                                 }
1643                                 ret = -EINVAL;
1644                                 goto error_out;
1645                         }
1646                         /* might be a policy change, too, so fall through */
1647                 }
1648                 dprintk("governor: change or update limits\n");
1649                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1650         }
1651
1652 error_out:
1653         cpufreq_debug_enable_ratelimit();
1654         return ret;
1655 }
1656
1657 /**
1658  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1659  *      @cpu: CPU which shall be re-evaluated
1660  *
1661  *      Usefull for policy notifiers which have different necessities
1662  *      at different times.
1663  */
1664 int cpufreq_update_policy(unsigned int cpu)
1665 {
1666         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1667         struct cpufreq_policy policy;
1668         int ret = 0;
1669
1670         if (!data)
1671                 return -ENODEV;
1672
1673         if (unlikely(lock_policy_rwsem_write(cpu)))
1674                 return -EINVAL;
1675
1676         dprintk("updating policy for CPU %u\n", cpu);
1677         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1678         policy.min = data->user_policy.min;
1679         policy.max = data->user_policy.max;
1680         policy.policy = data->user_policy.policy;
1681         policy.governor = data->user_policy.governor;
1682
1683         /* BIOS might change freq behind our back
1684           -> ask driver for current freq and notify governors about a change */
1685         if (cpufreq_driver->get) {
1686                 policy.cur = cpufreq_driver->get(cpu);
1687                 if (!data->cur) {
1688                         dprintk("Driver did not initialize current freq");
1689                         data->cur = policy.cur;
1690                 } else {
1691                         if (data->cur != policy.cur)
1692                                 cpufreq_out_of_sync(cpu, data->cur,
1693                                                                 policy.cur);
1694                 }
1695         }
1696
1697         ret = __cpufreq_set_policy(data, &policy);
1698
1699         unlock_policy_rwsem_write(cpu);
1700
1701         cpufreq_cpu_put(data);
1702         return ret;
1703 }
1704 EXPORT_SYMBOL(cpufreq_update_policy);
1705
1706 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1707                                         unsigned long action, void *hcpu)
1708 {
1709         unsigned int cpu = (unsigned long)hcpu;
1710         struct sys_device *sys_dev;
1711
1712         sys_dev = get_cpu_sysdev(cpu);
1713         if (sys_dev) {
1714                 switch (action) {
1715                 case CPU_ONLINE:
1716                 case CPU_ONLINE_FROZEN:
1717                         cpufreq_add_dev(sys_dev);
1718                         break;
1719                 case CPU_DOWN_PREPARE:
1720                 case CPU_DOWN_PREPARE_FROZEN:
1721                         if (unlikely(lock_policy_rwsem_write(cpu)))
1722                                 BUG();
1723
1724                         __cpufreq_remove_dev(sys_dev);
1725                         break;
1726                 case CPU_DOWN_FAILED:
1727                 case CPU_DOWN_FAILED_FROZEN:
1728                         cpufreq_add_dev(sys_dev);
1729                         break;
1730                 }
1731         }
1732         return NOTIFY_OK;
1733 }
1734
1735 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1736 {
1737     .notifier_call = cpufreq_cpu_callback,
1738 };
1739
1740 /*********************************************************************
1741  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1742  *********************************************************************/
1743
1744 /**
1745  * cpufreq_register_driver - register a CPU Frequency driver
1746  * @driver_data: A struct cpufreq_driver containing the values#
1747  * submitted by the CPU Frequency driver.
1748  *
1749  *   Registers a CPU Frequency driver to this core code. This code
1750  * returns zero on success, -EBUSY when another driver got here first
1751  * (and isn't unregistered in the meantime).
1752  *
1753  */
1754 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1755 {
1756         unsigned long flags;
1757         int ret;
1758
1759         if (!driver_data || !driver_data->verify || !driver_data->init ||
1760             ((!driver_data->setpolicy) && (!driver_data->target)))
1761                 return -EINVAL;
1762
1763         dprintk("trying to register driver %s\n", driver_data->name);
1764
1765         if (driver_data->setpolicy)
1766                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1767
1768         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1769         if (cpufreq_driver) {
1770                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1771                 return -EBUSY;
1772         }
1773         cpufreq_driver = driver_data;
1774         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1775
1776         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1777
1778         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1779                 int i;
1780                 ret = -ENODEV;
1781
1782                 /* check for at least one working CPU */
1783                 for (i=0; i<NR_CPUS; i++)
1784                         if (cpufreq_cpu_data[i])
1785                                 ret = 0;
1786
1787                 /* if all ->init() calls failed, unregister */
1788                 if (ret) {
1789                         dprintk("no CPU initialized for driver %s\n",
1790                                                         driver_data->name);
1791                         sysdev_driver_unregister(&cpu_sysdev_class,
1792                                                 &cpufreq_sysdev_driver);
1793
1794                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1795                         cpufreq_driver = NULL;
1796                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1797                 }
1798         }
1799
1800         if (!ret) {
1801                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1802                 dprintk("driver %s up and running\n", driver_data->name);
1803                 cpufreq_debug_enable_ratelimit();
1804         }
1805
1806         return (ret);
1807 }
1808 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1809
1810
1811 /**
1812  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1813  *
1814  *    Unregister the current CPUFreq driver. Only call this if you have
1815  * the right to do so, i.e. if you have succeeded in initialising before!
1816  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1817  * currently not initialised.
1818  */
1819 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1820 {
1821         unsigned long flags;
1822
1823         cpufreq_debug_disable_ratelimit();
1824
1825         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1826                 cpufreq_debug_enable_ratelimit();
1827                 return -EINVAL;
1828         }
1829
1830         dprintk("unregistering driver %s\n", driver->name);
1831
1832         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1833         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1834
1835         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1836         cpufreq_driver = NULL;
1837         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1838
1839         return 0;
1840 }
1841 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1842
1843 static int __init cpufreq_core_init(void)
1844 {
1845         int cpu;
1846
1847         for_each_possible_cpu(cpu) {
1848                 per_cpu(policy_cpu, cpu) = -1;
1849                 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1850         }
1851         return 0;
1852 }
1853
1854 core_initcall(cpufreq_core_init);