[PATCH] cpu hotplug: make [un]register_cpu_notifier init time only
[linux-2.6.git] / drivers / cpufreq / cpufreq.c
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *
7  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
8  *      Added handling for CPU hotplug
9  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
10  *      Fix handling for CPU hotplug -- affected CPUs
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License version 2 as
14  * published by the Free Software Foundation.
15  *
16  */
17
18 #include <linux/config.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/init.h>
22 #include <linux/notifier.h>
23 #include <linux/cpufreq.h>
24 #include <linux/delay.h>
25 #include <linux/interrupt.h>
26 #include <linux/spinlock.h>
27 #include <linux/device.h>
28 #include <linux/slab.h>
29 #include <linux/cpu.h>
30 #include <linux/completion.h>
31 #include <linux/mutex.h>
32
33 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
34
35 /**
36  * The "cpufreq driver" - the arch- or hardware-dependend 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 static DEFINE_SPINLOCK(cpufreq_driver_lock);
43
44 /* internal prototypes */
45 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
46 static void handle_update(void *data);
47
48 /**
49  * Two notifier lists: the "policy" list is involved in the
50  * validation process for a new CPU frequency policy; the
51  * "transition" list for kernel code that needs to handle
52  * changes to devices when the CPU clock speed changes.
53  * The mutex locks both lists.
54  */
55 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
56 static BLOCKING_NOTIFIER_HEAD(cpufreq_transition_notifier_list);
57
58
59 static LIST_HEAD(cpufreq_governor_list);
60 static DEFINE_MUTEX (cpufreq_governor_mutex);
61
62 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
63 {
64         struct cpufreq_policy *data;
65         unsigned long flags;
66
67         if (cpu >= NR_CPUS)
68                 goto err_out;
69
70         /* get the cpufreq driver */
71         spin_lock_irqsave(&cpufreq_driver_lock, flags);
72
73         if (!cpufreq_driver)
74                 goto err_out_unlock;
75
76         if (!try_module_get(cpufreq_driver->owner))
77                 goto err_out_unlock;
78
79
80         /* get the CPU */
81         data = cpufreq_cpu_data[cpu];
82
83         if (!data)
84                 goto err_out_put_module;
85
86         if (!kobject_get(&data->kobj))
87                 goto err_out_put_module;
88
89         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
90         return data;
91
92 err_out_put_module:
93         module_put(cpufreq_driver->owner);
94 err_out_unlock:
95         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
96 err_out:
97         return NULL;
98 }
99 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
100
101
102 void cpufreq_cpu_put(struct cpufreq_policy *data)
103 {
104         kobject_put(&data->kobj);
105         module_put(cpufreq_driver->owner);
106 }
107 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
108
109
110 /*********************************************************************
111  *                     UNIFIED DEBUG HELPERS                         *
112  *********************************************************************/
113 #ifdef CONFIG_CPU_FREQ_DEBUG
114
115 /* what part(s) of the CPUfreq subsystem are debugged? */
116 static unsigned int debug;
117
118 /* is the debug output ratelimit'ed using printk_ratelimit? User can
119  * set or modify this value.
120  */
121 static unsigned int debug_ratelimit = 1;
122
123 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
124  * loading of a cpufreq driver, temporarily disabled when a new policy
125  * is set, and disabled upon cpufreq driver removal
126  */
127 static unsigned int disable_ratelimit = 1;
128 static DEFINE_SPINLOCK(disable_ratelimit_lock);
129
130 static void cpufreq_debug_enable_ratelimit(void)
131 {
132         unsigned long flags;
133
134         spin_lock_irqsave(&disable_ratelimit_lock, flags);
135         if (disable_ratelimit)
136                 disable_ratelimit--;
137         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
138 }
139
140 static void cpufreq_debug_disable_ratelimit(void)
141 {
142         unsigned long flags;
143
144         spin_lock_irqsave(&disable_ratelimit_lock, flags);
145         disable_ratelimit++;
146         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
147 }
148
149 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
150 {
151         char s[256];
152         va_list args;
153         unsigned int len;
154         unsigned long flags;
155
156         WARN_ON(!prefix);
157         if (type & debug) {
158                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
159                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
160                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
161                         return;
162                 }
163                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
164
165                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
166
167                 va_start(args, fmt);
168                 len += vsnprintf(&s[len], (256 - len), fmt, args);
169                 va_end(args);
170
171                 printk(s);
172
173                 WARN_ON(len < 5);
174         }
175 }
176 EXPORT_SYMBOL(cpufreq_debug_printk);
177
178
179 module_param(debug, uint, 0644);
180 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
181
182 module_param(debug_ratelimit, uint, 0644);
183 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
184
185 #else /* !CONFIG_CPU_FREQ_DEBUG */
186
187 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
188 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
189
190 #endif /* CONFIG_CPU_FREQ_DEBUG */
191
192
193 /*********************************************************************
194  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
195  *********************************************************************/
196
197 /**
198  * adjust_jiffies - adjust the system "loops_per_jiffy"
199  *
200  * This function alters the system "loops_per_jiffy" for the clock
201  * speed change. Note that loops_per_jiffy cannot be updated on SMP
202  * systems as each CPU might be scaled differently. So, use the arch
203  * per-CPU loops_per_jiffy value wherever possible.
204  */
205 #ifndef CONFIG_SMP
206 static unsigned long l_p_j_ref;
207 static unsigned int  l_p_j_ref_freq;
208
209 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
210 {
211         if (ci->flags & CPUFREQ_CONST_LOOPS)
212                 return;
213
214         if (!l_p_j_ref_freq) {
215                 l_p_j_ref = loops_per_jiffy;
216                 l_p_j_ref_freq = ci->old;
217                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
218         }
219         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
220             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
221             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
222                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
223                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
224         }
225 }
226 #else
227 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
228 #endif
229
230
231 /**
232  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
233  * on frequency transition.
234  *
235  * This function calls the transition notifiers and the "adjust_jiffies"
236  * function. It is called twice on all CPU frequency changes that have
237  * external effects.
238  */
239 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
240 {
241         struct cpufreq_policy *policy;
242
243         BUG_ON(irqs_disabled());
244
245         freqs->flags = cpufreq_driver->flags;
246         dprintk("notification %u of frequency transition to %u kHz\n",
247                 state, freqs->new);
248
249         policy = cpufreq_cpu_data[freqs->cpu];
250         switch (state) {
251
252         case CPUFREQ_PRECHANGE:
253                 /* detect if the driver reported a value as "old frequency"
254                  * which is not equal to what the cpufreq core thinks is
255                  * "old frequency".
256                  */
257                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
258                         if ((policy) && (policy->cpu == freqs->cpu) &&
259                             (policy->cur) && (policy->cur != freqs->old)) {
260                                 dprintk("Warning: CPU frequency is"
261                                         " %u, cpufreq assumed %u kHz.\n",
262                                         freqs->old, policy->cur);
263                                 freqs->old = policy->cur;
264                         }
265                 }
266                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
267                                 CPUFREQ_PRECHANGE, freqs);
268                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
269                 break;
270
271         case CPUFREQ_POSTCHANGE:
272                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
273                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
274                                 CPUFREQ_POSTCHANGE, freqs);
275                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
276                         policy->cur = freqs->new;
277                 break;
278         }
279 }
280 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
281
282
283
284 /*********************************************************************
285  *                          SYSFS INTERFACE                          *
286  *********************************************************************/
287
288 /**
289  * cpufreq_parse_governor - parse a governor string
290  */
291 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
292                                 struct cpufreq_governor **governor)
293 {
294         if (!cpufreq_driver)
295                 return -EINVAL;
296         if (cpufreq_driver->setpolicy) {
297                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
298                         *policy = CPUFREQ_POLICY_PERFORMANCE;
299                         return 0;
300                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
301                         *policy = CPUFREQ_POLICY_POWERSAVE;
302                         return 0;
303                 }
304                 return -EINVAL;
305         } else {
306                 struct cpufreq_governor *t;
307                 mutex_lock(&cpufreq_governor_mutex);
308                 if (!cpufreq_driver || !cpufreq_driver->target)
309                         goto out;
310                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
311                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
312                                 *governor = t;
313                                 mutex_unlock(&cpufreq_governor_mutex);
314                                 return 0;
315                         }
316                 }
317 out:
318                 mutex_unlock(&cpufreq_governor_mutex);
319         }
320         return -EINVAL;
321 }
322
323
324 /* drivers/base/cpu.c */
325 extern struct sysdev_class cpu_sysdev_class;
326
327
328 /**
329  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
330  *
331  * Write out information from cpufreq_driver->policy[cpu]; object must be
332  * "unsigned int".
333  */
334
335 #define show_one(file_name, object)                     \
336 static ssize_t show_##file_name                         \
337 (struct cpufreq_policy * policy, char *buf)             \
338 {                                                       \
339         return sprintf (buf, "%u\n", policy->object);   \
340 }
341
342 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
343 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
344 show_one(scaling_min_freq, min);
345 show_one(scaling_max_freq, max);
346 show_one(scaling_cur_freq, cur);
347
348 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
349
350 /**
351  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
352  */
353 #define store_one(file_name, object)                    \
354 static ssize_t store_##file_name                                        \
355 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
356 {                                                                       \
357         unsigned int ret = -EINVAL;                                     \
358         struct cpufreq_policy new_policy;                               \
359                                                                         \
360         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
361         if (ret)                                                        \
362                 return -EINVAL;                                         \
363                                                                         \
364         ret = sscanf (buf, "%u", &new_policy.object);                   \
365         if (ret != 1)                                                   \
366                 return -EINVAL;                                         \
367                                                                         \
368         mutex_lock(&policy->lock);                                      \
369         ret = __cpufreq_set_policy(policy, &new_policy);                \
370         policy->user_policy.object = policy->object;                    \
371         mutex_unlock(&policy->lock);                                    \
372                                                                         \
373         return ret ? ret : count;                                       \
374 }
375
376 store_one(scaling_min_freq,min);
377 store_one(scaling_max_freq,max);
378
379 /**
380  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
381  */
382 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
383 {
384         unsigned int cur_freq = cpufreq_get(policy->cpu);
385         if (!cur_freq)
386                 return sprintf(buf, "<unknown>");
387         return sprintf(buf, "%u\n", cur_freq);
388 }
389
390
391 /**
392  * show_scaling_governor - show the current policy for the specified CPU
393  */
394 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
395 {
396         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
397                 return sprintf(buf, "powersave\n");
398         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
399                 return sprintf(buf, "performance\n");
400         else if (policy->governor)
401                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
402         return -EINVAL;
403 }
404
405
406 /**
407  * store_scaling_governor - store policy for the specified CPU
408  */
409 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
410                                        const char *buf, size_t count)
411 {
412         unsigned int ret = -EINVAL;
413         char    str_governor[16];
414         struct cpufreq_policy new_policy;
415
416         ret = cpufreq_get_policy(&new_policy, policy->cpu);
417         if (ret)
418                 return ret;
419
420         ret = sscanf (buf, "%15s", str_governor);
421         if (ret != 1)
422                 return -EINVAL;
423
424         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
425                 return -EINVAL;
426
427         /* Do not use cpufreq_set_policy here or the user_policy.max
428            will be wrongly overridden */
429         mutex_lock(&policy->lock);
430         ret = __cpufreq_set_policy(policy, &new_policy);
431
432         policy->user_policy.policy = policy->policy;
433         policy->user_policy.governor = policy->governor;
434         mutex_unlock(&policy->lock);
435
436         return ret ? ret : count;
437 }
438
439 /**
440  * show_scaling_driver - show the cpufreq driver currently loaded
441  */
442 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
443 {
444         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
445 }
446
447 /**
448  * show_scaling_available_governors - show the available CPUfreq governors
449  */
450 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
451                                 char *buf)
452 {
453         ssize_t i = 0;
454         struct cpufreq_governor *t;
455
456         if (!cpufreq_driver->target) {
457                 i += sprintf(buf, "performance powersave");
458                 goto out;
459         }
460
461         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
462                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
463                         goto out;
464                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
465         }
466 out:
467         i += sprintf(&buf[i], "\n");
468         return i;
469 }
470 /**
471  * show_affected_cpus - show the CPUs affected by each transition
472  */
473 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
474 {
475         ssize_t i = 0;
476         unsigned int cpu;
477
478         for_each_cpu_mask(cpu, policy->cpus) {
479                 if (i)
480                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
481                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
482                 if (i >= (PAGE_SIZE - 5))
483                     break;
484         }
485         i += sprintf(&buf[i], "\n");
486         return i;
487 }
488
489
490 #define define_one_ro(_name) \
491 static struct freq_attr _name = \
492 __ATTR(_name, 0444, show_##_name, NULL)
493
494 #define define_one_ro0400(_name) \
495 static struct freq_attr _name = \
496 __ATTR(_name, 0400, show_##_name, NULL)
497
498 #define define_one_rw(_name) \
499 static struct freq_attr _name = \
500 __ATTR(_name, 0644, show_##_name, store_##_name)
501
502 define_one_ro0400(cpuinfo_cur_freq);
503 define_one_ro(cpuinfo_min_freq);
504 define_one_ro(cpuinfo_max_freq);
505 define_one_ro(scaling_available_governors);
506 define_one_ro(scaling_driver);
507 define_one_ro(scaling_cur_freq);
508 define_one_ro(affected_cpus);
509 define_one_rw(scaling_min_freq);
510 define_one_rw(scaling_max_freq);
511 define_one_rw(scaling_governor);
512
513 static struct attribute * default_attrs[] = {
514         &cpuinfo_min_freq.attr,
515         &cpuinfo_max_freq.attr,
516         &scaling_min_freq.attr,
517         &scaling_max_freq.attr,
518         &affected_cpus.attr,
519         &scaling_governor.attr,
520         &scaling_driver.attr,
521         &scaling_available_governors.attr,
522         NULL
523 };
524
525 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
526 #define to_attr(a) container_of(a,struct freq_attr,attr)
527
528 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
529 {
530         struct cpufreq_policy * policy = to_policy(kobj);
531         struct freq_attr * fattr = to_attr(attr);
532         ssize_t ret;
533         policy = cpufreq_cpu_get(policy->cpu);
534         if (!policy)
535                 return -EINVAL;
536         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
537         cpufreq_cpu_put(policy);
538         return ret;
539 }
540
541 static ssize_t store(struct kobject * kobj, struct attribute * attr,
542                      const char * buf, size_t count)
543 {
544         struct cpufreq_policy * policy = to_policy(kobj);
545         struct freq_attr * fattr = to_attr(attr);
546         ssize_t ret;
547         policy = cpufreq_cpu_get(policy->cpu);
548         if (!policy)
549                 return -EINVAL;
550         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
551         cpufreq_cpu_put(policy);
552         return ret;
553 }
554
555 static void cpufreq_sysfs_release(struct kobject * kobj)
556 {
557         struct cpufreq_policy * policy = to_policy(kobj);
558         dprintk("last reference is dropped\n");
559         complete(&policy->kobj_unregister);
560 }
561
562 static struct sysfs_ops sysfs_ops = {
563         .show   = show,
564         .store  = store,
565 };
566
567 static struct kobj_type ktype_cpufreq = {
568         .sysfs_ops      = &sysfs_ops,
569         .default_attrs  = default_attrs,
570         .release        = cpufreq_sysfs_release,
571 };
572
573
574 /**
575  * cpufreq_add_dev - add a CPU device
576  *
577  * Adds the cpufreq interface for a CPU device.
578  */
579 static int cpufreq_add_dev (struct sys_device * sys_dev)
580 {
581         unsigned int cpu = sys_dev->id;
582         int ret = 0;
583         struct cpufreq_policy new_policy;
584         struct cpufreq_policy *policy;
585         struct freq_attr **drv_attr;
586         struct sys_device *cpu_sys_dev;
587         unsigned long flags;
588         unsigned int j;
589 #ifdef CONFIG_SMP
590         struct cpufreq_policy *managed_policy;
591 #endif
592
593         if (cpu_is_offline(cpu))
594                 return 0;
595
596         cpufreq_debug_disable_ratelimit();
597         dprintk("adding CPU %u\n", cpu);
598
599 #ifdef CONFIG_SMP
600         /* check whether a different CPU already registered this
601          * CPU because it is in the same boat. */
602         policy = cpufreq_cpu_get(cpu);
603         if (unlikely(policy)) {
604                 cpufreq_cpu_put(policy);
605                 cpufreq_debug_enable_ratelimit();
606                 return 0;
607         }
608 #endif
609
610         if (!try_module_get(cpufreq_driver->owner)) {
611                 ret = -EINVAL;
612                 goto module_out;
613         }
614
615         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
616         if (!policy) {
617                 ret = -ENOMEM;
618                 goto nomem_out;
619         }
620
621         policy->cpu = cpu;
622         policy->cpus = cpumask_of_cpu(cpu);
623
624         mutex_init(&policy->lock);
625         mutex_lock(&policy->lock);
626         init_completion(&policy->kobj_unregister);
627         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
628
629         /* call driver. From then on the cpufreq must be able
630          * to accept all calls to ->verify and ->setpolicy for this CPU
631          */
632         ret = cpufreq_driver->init(policy);
633         if (ret) {
634                 dprintk("initialization failed\n");
635                 mutex_unlock(&policy->lock);
636                 goto err_out;
637         }
638
639 #ifdef CONFIG_SMP
640         for_each_cpu_mask(j, policy->cpus) {
641                 if (cpu == j)
642                         continue;
643
644                 /* check for existing affected CPUs.  They may not be aware
645                  * of it due to CPU Hotplug.
646                  */
647                 managed_policy = cpufreq_cpu_get(j);
648                 if (unlikely(managed_policy)) {
649                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
650                         managed_policy->cpus = policy->cpus;
651                         cpufreq_cpu_data[cpu] = managed_policy;
652                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
653
654                         dprintk("CPU already managed, adding link\n");
655                         sysfs_create_link(&sys_dev->kobj,
656                                           &managed_policy->kobj, "cpufreq");
657
658                         cpufreq_debug_enable_ratelimit();
659                         mutex_unlock(&policy->lock);
660                         ret = 0;
661                         goto err_out_driver_exit; /* call driver->exit() */
662                 }
663         }
664 #endif
665         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
666
667         /* prepare interface data */
668         policy->kobj.parent = &sys_dev->kobj;
669         policy->kobj.ktype = &ktype_cpufreq;
670         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
671
672         ret = kobject_register(&policy->kobj);
673         if (ret) {
674                 mutex_unlock(&policy->lock);
675                 goto err_out_driver_exit;
676         }
677         /* set up files for this cpu device */
678         drv_attr = cpufreq_driver->attr;
679         while ((drv_attr) && (*drv_attr)) {
680                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
681                 drv_attr++;
682         }
683         if (cpufreq_driver->get)
684                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
685         if (cpufreq_driver->target)
686                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
687
688         spin_lock_irqsave(&cpufreq_driver_lock, flags);
689         for_each_cpu_mask(j, policy->cpus)
690                 cpufreq_cpu_data[j] = policy;
691         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
692
693         /* symlink affected CPUs */
694         for_each_cpu_mask(j, policy->cpus) {
695                 if (j == cpu)
696                         continue;
697                 if (!cpu_online(j))
698                         continue;
699
700                 dprintk("CPU %u already managed, adding link\n", j);
701                 cpufreq_cpu_get(cpu);
702                 cpu_sys_dev = get_cpu_sysdev(j);
703                 sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
704                                   "cpufreq");
705         }
706
707         policy->governor = NULL; /* to assure that the starting sequence is
708                                   * run in cpufreq_set_policy */
709         mutex_unlock(&policy->lock);
710
711         /* set default policy */
712         ret = cpufreq_set_policy(&new_policy);
713         if (ret) {
714                 dprintk("setting policy failed\n");
715                 goto err_out_unregister;
716         }
717
718         module_put(cpufreq_driver->owner);
719         dprintk("initialization complete\n");
720         cpufreq_debug_enable_ratelimit();
721
722         return 0;
723
724
725 err_out_unregister:
726         spin_lock_irqsave(&cpufreq_driver_lock, flags);
727         for_each_cpu_mask(j, policy->cpus)
728                 cpufreq_cpu_data[j] = NULL;
729         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
730
731         kobject_unregister(&policy->kobj);
732         wait_for_completion(&policy->kobj_unregister);
733
734 err_out_driver_exit:
735         if (cpufreq_driver->exit)
736                 cpufreq_driver->exit(policy);
737
738 err_out:
739         kfree(policy);
740
741 nomem_out:
742         module_put(cpufreq_driver->owner);
743 module_out:
744         cpufreq_debug_enable_ratelimit();
745         return ret;
746 }
747
748
749 /**
750  * cpufreq_remove_dev - remove a CPU device
751  *
752  * Removes the cpufreq interface for a CPU device.
753  */
754 static int cpufreq_remove_dev (struct sys_device * sys_dev)
755 {
756         unsigned int cpu = sys_dev->id;
757         unsigned long flags;
758         struct cpufreq_policy *data;
759 #ifdef CONFIG_SMP
760         struct sys_device *cpu_sys_dev;
761         unsigned int j;
762 #endif
763
764         cpufreq_debug_disable_ratelimit();
765         dprintk("unregistering CPU %u\n", cpu);
766
767         spin_lock_irqsave(&cpufreq_driver_lock, flags);
768         data = cpufreq_cpu_data[cpu];
769
770         if (!data) {
771                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
772                 cpufreq_debug_enable_ratelimit();
773                 return -EINVAL;
774         }
775         cpufreq_cpu_data[cpu] = NULL;
776
777
778 #ifdef CONFIG_SMP
779         /* if this isn't the CPU which is the parent of the kobj, we
780          * only need to unlink, put and exit
781          */
782         if (unlikely(cpu != data->cpu)) {
783                 dprintk("removing link\n");
784                 cpu_clear(cpu, data->cpus);
785                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
786                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
787                 cpufreq_cpu_put(data);
788                 cpufreq_debug_enable_ratelimit();
789                 return 0;
790         }
791 #endif
792
793
794         if (!kobject_get(&data->kobj)) {
795                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
796                 cpufreq_debug_enable_ratelimit();
797                 return -EFAULT;
798         }
799
800 #ifdef CONFIG_SMP
801         /* if we have other CPUs still registered, we need to unlink them,
802          * or else wait_for_completion below will lock up. Clean the
803          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
804          * links afterwards.
805          */
806         if (unlikely(cpus_weight(data->cpus) > 1)) {
807                 for_each_cpu_mask(j, data->cpus) {
808                         if (j == cpu)
809                                 continue;
810                         cpufreq_cpu_data[j] = NULL;
811                 }
812         }
813
814         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
815
816         if (unlikely(cpus_weight(data->cpus) > 1)) {
817                 for_each_cpu_mask(j, data->cpus) {
818                         if (j == cpu)
819                                 continue;
820                         dprintk("removing link for cpu %u\n", j);
821                         cpu_sys_dev = get_cpu_sysdev(j);
822                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
823                         cpufreq_cpu_put(data);
824                 }
825         }
826 #else
827         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
828 #endif
829
830         mutex_lock(&data->lock);
831         if (cpufreq_driver->target)
832                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
833         mutex_unlock(&data->lock);
834
835         kobject_unregister(&data->kobj);
836
837         kobject_put(&data->kobj);
838
839         /* we need to make sure that the underlying kobj is actually
840          * not referenced anymore by anybody before we proceed with
841          * unloading.
842          */
843         dprintk("waiting for dropping of refcount\n");
844         wait_for_completion(&data->kobj_unregister);
845         dprintk("wait complete\n");
846
847         if (cpufreq_driver->exit)
848                 cpufreq_driver->exit(data);
849
850         kfree(data);
851
852         cpufreq_debug_enable_ratelimit();
853         return 0;
854 }
855
856
857 static void handle_update(void *data)
858 {
859         unsigned int cpu = (unsigned int)(long)data;
860         dprintk("handle_update for cpu %u called\n", cpu);
861         cpufreq_update_policy(cpu);
862 }
863
864 /**
865  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
866  *      @cpu: cpu number
867  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
868  *      @new_freq: CPU frequency the CPU actually runs at
869  *
870  *      We adjust to current frequency first, and need to clean up later. So either call
871  *      to cpufreq_update_policy() or schedule handle_update()).
872  */
873 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
874 {
875         struct cpufreq_freqs freqs;
876
877         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
878                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
879
880         freqs.cpu = cpu;
881         freqs.old = old_freq;
882         freqs.new = new_freq;
883         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
884         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
885 }
886
887
888 /**
889  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
890  * @cpu: CPU number
891  *
892  * This is the last known freq, without actually getting it from the driver.
893  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
894  */
895 unsigned int cpufreq_quick_get(unsigned int cpu)
896 {
897         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
898         unsigned int ret = 0;
899
900         if (policy) {
901                 mutex_lock(&policy->lock);
902                 ret = policy->cur;
903                 mutex_unlock(&policy->lock);
904                 cpufreq_cpu_put(policy);
905         }
906
907         return (ret);
908 }
909 EXPORT_SYMBOL(cpufreq_quick_get);
910
911
912 /**
913  * cpufreq_get - get the current CPU frequency (in kHz)
914  * @cpu: CPU number
915  *
916  * Get the CPU current (static) CPU frequency
917  */
918 unsigned int cpufreq_get(unsigned int cpu)
919 {
920         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
921         unsigned int ret = 0;
922
923         if (!policy)
924                 return 0;
925
926         if (!cpufreq_driver->get)
927                 goto out;
928
929         mutex_lock(&policy->lock);
930
931         ret = cpufreq_driver->get(cpu);
932
933         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
934                 /* verify no discrepancy between actual and saved value exists */
935                 if (unlikely(ret != policy->cur)) {
936                         cpufreq_out_of_sync(cpu, policy->cur, ret);
937                         schedule_work(&policy->update);
938                 }
939         }
940
941         mutex_unlock(&policy->lock);
942
943 out:
944         cpufreq_cpu_put(policy);
945
946         return (ret);
947 }
948 EXPORT_SYMBOL(cpufreq_get);
949
950
951 /**
952  *      cpufreq_suspend - let the low level driver prepare for suspend
953  */
954
955 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
956 {
957         int cpu = sysdev->id;
958         unsigned int ret = 0;
959         unsigned int cur_freq = 0;
960         struct cpufreq_policy *cpu_policy;
961
962         dprintk("resuming cpu %u\n", cpu);
963
964         if (!cpu_online(cpu))
965                 return 0;
966
967         /* we may be lax here as interrupts are off. Nonetheless
968          * we need to grab the correct cpu policy, as to check
969          * whether we really run on this CPU.
970          */
971
972         cpu_policy = cpufreq_cpu_get(cpu);
973         if (!cpu_policy)
974                 return -EINVAL;
975
976         /* only handle each CPU group once */
977         if (unlikely(cpu_policy->cpu != cpu)) {
978                 cpufreq_cpu_put(cpu_policy);
979                 return 0;
980         }
981
982         if (cpufreq_driver->suspend) {
983                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
984                 if (ret) {
985                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
986                                         "step on CPU %u\n", cpu_policy->cpu);
987                         cpufreq_cpu_put(cpu_policy);
988                         return ret;
989                 }
990         }
991
992
993         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
994                 goto out;
995
996         if (cpufreq_driver->get)
997                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
998
999         if (!cur_freq || !cpu_policy->cur) {
1000                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1001                        "frequency is what timing core thinks it is.\n");
1002                 goto out;
1003         }
1004
1005         if (unlikely(cur_freq != cpu_policy->cur)) {
1006                 struct cpufreq_freqs freqs;
1007
1008                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1009                         dprintk("Warning: CPU frequency is %u, "
1010                                "cpufreq assumed %u kHz.\n",
1011                                cur_freq, cpu_policy->cur);
1012
1013                 freqs.cpu = cpu;
1014                 freqs.old = cpu_policy->cur;
1015                 freqs.new = cur_freq;
1016
1017                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
1018                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1019                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1020
1021                 cpu_policy->cur = cur_freq;
1022         }
1023
1024 out:
1025         cpufreq_cpu_put(cpu_policy);
1026         return 0;
1027 }
1028
1029 /**
1030  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1031  *
1032  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1033  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1034  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1035  *          restored.
1036  */
1037 static int cpufreq_resume(struct sys_device * sysdev)
1038 {
1039         int cpu = sysdev->id;
1040         unsigned int ret = 0;
1041         struct cpufreq_policy *cpu_policy;
1042
1043         dprintk("resuming cpu %u\n", cpu);
1044
1045         if (!cpu_online(cpu))
1046                 return 0;
1047
1048         /* we may be lax here as interrupts are off. Nonetheless
1049          * we need to grab the correct cpu policy, as to check
1050          * whether we really run on this CPU.
1051          */
1052
1053         cpu_policy = cpufreq_cpu_get(cpu);
1054         if (!cpu_policy)
1055                 return -EINVAL;
1056
1057         /* only handle each CPU group once */
1058         if (unlikely(cpu_policy->cpu != cpu)) {
1059                 cpufreq_cpu_put(cpu_policy);
1060                 return 0;
1061         }
1062
1063         if (cpufreq_driver->resume) {
1064                 ret = cpufreq_driver->resume(cpu_policy);
1065                 if (ret) {
1066                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1067                                         "step on CPU %u\n", cpu_policy->cpu);
1068                         cpufreq_cpu_put(cpu_policy);
1069                         return ret;
1070                 }
1071         }
1072
1073         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1074                 unsigned int cur_freq = 0;
1075
1076                 if (cpufreq_driver->get)
1077                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1078
1079                 if (!cur_freq || !cpu_policy->cur) {
1080                         printk(KERN_ERR "cpufreq: resume failed to assert "
1081                                         "current frequency is what timing core "
1082                                         "thinks it is.\n");
1083                         goto out;
1084                 }
1085
1086                 if (unlikely(cur_freq != cpu_policy->cur)) {
1087                         struct cpufreq_freqs freqs;
1088
1089                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1090                                 dprintk("Warning: CPU frequency"
1091                                        "is %u, cpufreq assumed %u kHz.\n",
1092                                        cur_freq, cpu_policy->cur);
1093
1094                         freqs.cpu = cpu;
1095                         freqs.old = cpu_policy->cur;
1096                         freqs.new = cur_freq;
1097
1098                         blocking_notifier_call_chain(
1099                                         &cpufreq_transition_notifier_list,
1100                                         CPUFREQ_RESUMECHANGE, &freqs);
1101                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1102
1103                         cpu_policy->cur = cur_freq;
1104                 }
1105         }
1106
1107 out:
1108         schedule_work(&cpu_policy->update);
1109         cpufreq_cpu_put(cpu_policy);
1110         return ret;
1111 }
1112
1113 static struct sysdev_driver cpufreq_sysdev_driver = {
1114         .add            = cpufreq_add_dev,
1115         .remove         = cpufreq_remove_dev,
1116         .suspend        = cpufreq_suspend,
1117         .resume         = cpufreq_resume,
1118 };
1119
1120
1121 /*********************************************************************
1122  *                     NOTIFIER LISTS INTERFACE                      *
1123  *********************************************************************/
1124
1125 /**
1126  *      cpufreq_register_notifier - register a driver with cpufreq
1127  *      @nb: notifier function to register
1128  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1129  *
1130  *      Add a driver to one of two lists: either a list of drivers that
1131  *      are notified about clock rate changes (once before and once after
1132  *      the transition), or a list of drivers that are notified about
1133  *      changes in cpufreq policy.
1134  *
1135  *      This function may sleep, and has the same return conditions as
1136  *      blocking_notifier_chain_register.
1137  */
1138 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1139 {
1140         int ret;
1141
1142         switch (list) {
1143         case CPUFREQ_TRANSITION_NOTIFIER:
1144                 ret = blocking_notifier_chain_register(
1145                                 &cpufreq_transition_notifier_list, nb);
1146                 break;
1147         case CPUFREQ_POLICY_NOTIFIER:
1148                 ret = blocking_notifier_chain_register(
1149                                 &cpufreq_policy_notifier_list, nb);
1150                 break;
1151         default:
1152                 ret = -EINVAL;
1153         }
1154
1155         return ret;
1156 }
1157 EXPORT_SYMBOL(cpufreq_register_notifier);
1158
1159
1160 /**
1161  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1162  *      @nb: notifier block to be unregistered
1163  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1164  *
1165  *      Remove a driver from the CPU frequency notifier list.
1166  *
1167  *      This function may sleep, and has the same return conditions as
1168  *      blocking_notifier_chain_unregister.
1169  */
1170 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1171 {
1172         int ret;
1173
1174         switch (list) {
1175         case CPUFREQ_TRANSITION_NOTIFIER:
1176                 ret = blocking_notifier_chain_unregister(
1177                                 &cpufreq_transition_notifier_list, nb);
1178                 break;
1179         case CPUFREQ_POLICY_NOTIFIER:
1180                 ret = blocking_notifier_chain_unregister(
1181                                 &cpufreq_policy_notifier_list, nb);
1182                 break;
1183         default:
1184                 ret = -EINVAL;
1185         }
1186
1187         return ret;
1188 }
1189 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1190
1191
1192 /*********************************************************************
1193  *                              GOVERNORS                            *
1194  *********************************************************************/
1195
1196
1197 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1198                             unsigned int target_freq,
1199                             unsigned int relation)
1200 {
1201         int retval = -EINVAL;
1202
1203         lock_cpu_hotplug();
1204         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1205                 target_freq, relation);
1206         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1207                 retval = cpufreq_driver->target(policy, target_freq, relation);
1208
1209         unlock_cpu_hotplug();
1210
1211         return retval;
1212 }
1213 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1214
1215 int cpufreq_driver_target(struct cpufreq_policy *policy,
1216                           unsigned int target_freq,
1217                           unsigned int relation)
1218 {
1219         int ret;
1220
1221         policy = cpufreq_cpu_get(policy->cpu);
1222         if (!policy)
1223                 return -EINVAL;
1224
1225         mutex_lock(&policy->lock);
1226
1227         ret = __cpufreq_driver_target(policy, target_freq, relation);
1228
1229         mutex_unlock(&policy->lock);
1230
1231         cpufreq_cpu_put(policy);
1232         return ret;
1233 }
1234 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1235
1236
1237 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1238 {
1239         int ret;
1240
1241         if (!try_module_get(policy->governor->owner))
1242                 return -EINVAL;
1243
1244         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1245         ret = policy->governor->governor(policy, event);
1246
1247         /* we keep one module reference alive for each CPU governed by this CPU */
1248         if ((event != CPUFREQ_GOV_START) || ret)
1249                 module_put(policy->governor->owner);
1250         if ((event == CPUFREQ_GOV_STOP) && !ret)
1251                 module_put(policy->governor->owner);
1252
1253         return ret;
1254 }
1255
1256
1257 int cpufreq_governor(unsigned int cpu, unsigned int event)
1258 {
1259         int ret = 0;
1260         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1261
1262         if (!policy)
1263                 return -EINVAL;
1264
1265         mutex_lock(&policy->lock);
1266         ret = __cpufreq_governor(policy, event);
1267         mutex_unlock(&policy->lock);
1268
1269         cpufreq_cpu_put(policy);
1270         return ret;
1271 }
1272 EXPORT_SYMBOL_GPL(cpufreq_governor);
1273
1274
1275 int cpufreq_register_governor(struct cpufreq_governor *governor)
1276 {
1277         struct cpufreq_governor *t;
1278
1279         if (!governor)
1280                 return -EINVAL;
1281
1282         mutex_lock(&cpufreq_governor_mutex);
1283
1284         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1285                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1286                         mutex_unlock(&cpufreq_governor_mutex);
1287                         return -EBUSY;
1288                 }
1289         }
1290         list_add(&governor->governor_list, &cpufreq_governor_list);
1291
1292         mutex_unlock(&cpufreq_governor_mutex);
1293         return 0;
1294 }
1295 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1296
1297
1298 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1299 {
1300         if (!governor)
1301                 return;
1302
1303         mutex_lock(&cpufreq_governor_mutex);
1304         list_del(&governor->governor_list);
1305         mutex_unlock(&cpufreq_governor_mutex);
1306         return;
1307 }
1308 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1309
1310
1311
1312 /*********************************************************************
1313  *                          POLICY INTERFACE                         *
1314  *********************************************************************/
1315
1316 /**
1317  * cpufreq_get_policy - get the current cpufreq_policy
1318  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1319  *
1320  * Reads the current cpufreq policy.
1321  */
1322 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1323 {
1324         struct cpufreq_policy *cpu_policy;
1325         if (!policy)
1326                 return -EINVAL;
1327
1328         cpu_policy = cpufreq_cpu_get(cpu);
1329         if (!cpu_policy)
1330                 return -EINVAL;
1331
1332         mutex_lock(&cpu_policy->lock);
1333         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1334         mutex_unlock(&cpu_policy->lock);
1335
1336         cpufreq_cpu_put(cpu_policy);
1337         return 0;
1338 }
1339 EXPORT_SYMBOL(cpufreq_get_policy);
1340
1341
1342 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1343 {
1344         int ret = 0;
1345
1346         cpufreq_debug_disable_ratelimit();
1347         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1348                 policy->min, policy->max);
1349
1350         memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
1351
1352         /* verify the cpu speed can be set within this limit */
1353         ret = cpufreq_driver->verify(policy);
1354         if (ret)
1355                 goto error_out;
1356
1357         /* adjust if necessary - all reasons */
1358         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1359                         CPUFREQ_ADJUST, policy);
1360
1361         /* adjust if necessary - hardware incompatibility*/
1362         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1363                         CPUFREQ_INCOMPATIBLE, policy);
1364
1365         /* verify the cpu speed can be set within this limit,
1366            which might be different to the first one */
1367         ret = cpufreq_driver->verify(policy);
1368         if (ret)
1369                 goto error_out;
1370
1371         /* notification of the new policy */
1372         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1373                         CPUFREQ_NOTIFY, policy);
1374
1375         data->min = policy->min;
1376         data->max = policy->max;
1377
1378         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1379
1380         if (cpufreq_driver->setpolicy) {
1381                 data->policy = policy->policy;
1382                 dprintk("setting range\n");
1383                 ret = cpufreq_driver->setpolicy(policy);
1384         } else {
1385                 if (policy->governor != data->governor) {
1386                         /* save old, working values */
1387                         struct cpufreq_governor *old_gov = data->governor;
1388
1389                         dprintk("governor switch\n");
1390
1391                         /* end old governor */
1392                         if (data->governor)
1393                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1394
1395                         /* start new governor */
1396                         data->governor = policy->governor;
1397                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1398                                 /* new governor failed, so re-start old one */
1399                                 dprintk("starting governor %s failed\n", data->governor->name);
1400                                 if (old_gov) {
1401                                         data->governor = old_gov;
1402                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1403                                 }
1404                                 ret = -EINVAL;
1405                                 goto error_out;
1406                         }
1407                         /* might be a policy change, too, so fall through */
1408                 }
1409                 dprintk("governor: change or update limits\n");
1410                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1411         }
1412
1413 error_out:
1414         cpufreq_debug_enable_ratelimit();
1415         return ret;
1416 }
1417
1418 /**
1419  *      cpufreq_set_policy - set a new CPUFreq policy
1420  *      @policy: policy to be set.
1421  *
1422  *      Sets a new CPU frequency and voltage scaling policy.
1423  */
1424 int cpufreq_set_policy(struct cpufreq_policy *policy)
1425 {
1426         int ret = 0;
1427         struct cpufreq_policy *data;
1428
1429         if (!policy)
1430                 return -EINVAL;
1431
1432         data = cpufreq_cpu_get(policy->cpu);
1433         if (!data)
1434                 return -EINVAL;
1435
1436         /* lock this CPU */
1437         mutex_lock(&data->lock);
1438
1439         ret = __cpufreq_set_policy(data, policy);
1440         data->user_policy.min = data->min;
1441         data->user_policy.max = data->max;
1442         data->user_policy.policy = data->policy;
1443         data->user_policy.governor = data->governor;
1444
1445         mutex_unlock(&data->lock);
1446         cpufreq_cpu_put(data);
1447
1448         return ret;
1449 }
1450 EXPORT_SYMBOL(cpufreq_set_policy);
1451
1452
1453 /**
1454  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1455  *      @cpu: CPU which shall be re-evaluated
1456  *
1457  *      Usefull for policy notifiers which have different necessities
1458  *      at different times.
1459  */
1460 int cpufreq_update_policy(unsigned int cpu)
1461 {
1462         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1463         struct cpufreq_policy policy;
1464         int ret = 0;
1465
1466         if (!data)
1467                 return -ENODEV;
1468
1469         mutex_lock(&data->lock);
1470
1471         dprintk("updating policy for CPU %u\n", cpu);
1472         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1473         policy.min = data->user_policy.min;
1474         policy.max = data->user_policy.max;
1475         policy.policy = data->user_policy.policy;
1476         policy.governor = data->user_policy.governor;
1477
1478         /* BIOS might change freq behind our back
1479           -> ask driver for current freq and notify governors about a change */
1480         if (cpufreq_driver->get) {
1481                 policy.cur = cpufreq_driver->get(cpu);
1482                 if (!data->cur) {
1483                         dprintk("Driver did not initialize current freq");
1484                         data->cur = policy.cur;
1485                 } else {
1486                         if (data->cur != policy.cur)
1487                                 cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1488                 }
1489         }
1490
1491         ret = __cpufreq_set_policy(data, &policy);
1492
1493         mutex_unlock(&data->lock);
1494
1495         cpufreq_cpu_put(data);
1496         return ret;
1497 }
1498 EXPORT_SYMBOL(cpufreq_update_policy);
1499
1500 #ifdef CONFIG_HOTPLUG_CPU
1501 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1502                                         unsigned long action, void *hcpu)
1503 {
1504         unsigned int cpu = (unsigned long)hcpu;
1505         struct cpufreq_policy *policy;
1506         struct sys_device *sys_dev;
1507
1508         sys_dev = get_cpu_sysdev(cpu);
1509
1510         if (sys_dev) {
1511                 switch (action) {
1512                 case CPU_ONLINE:
1513                         cpufreq_add_dev(sys_dev);
1514                         break;
1515                 case CPU_DOWN_PREPARE:
1516                         /*
1517                          * We attempt to put this cpu in lowest frequency
1518                          * possible before going down. This will permit
1519                          * hardware-managed P-State to switch other related
1520                          * threads to min or higher speeds if possible.
1521                          */
1522                         policy = cpufreq_cpu_data[cpu];
1523                         if (policy) {
1524                                 cpufreq_driver_target(policy, policy->min,
1525                                                 CPUFREQ_RELATION_H);
1526                         }
1527                         break;
1528                 case CPU_DEAD:
1529                         cpufreq_remove_dev(sys_dev);
1530                         break;
1531                 }
1532         }
1533         return NOTIFY_OK;
1534 }
1535
1536 static struct notifier_block cpufreq_cpu_notifier =
1537 {
1538     .notifier_call = cpufreq_cpu_callback,
1539 };
1540 #endif /* CONFIG_HOTPLUG_CPU */
1541
1542 /*********************************************************************
1543  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1544  *********************************************************************/
1545
1546 /**
1547  * cpufreq_register_driver - register a CPU Frequency driver
1548  * @driver_data: A struct cpufreq_driver containing the values#
1549  * submitted by the CPU Frequency driver.
1550  *
1551  *   Registers a CPU Frequency driver to this core code. This code
1552  * returns zero on success, -EBUSY when another driver got here first
1553  * (and isn't unregistered in the meantime).
1554  *
1555  */
1556 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1557 {
1558         unsigned long flags;
1559         int ret;
1560
1561         if (!driver_data || !driver_data->verify || !driver_data->init ||
1562             ((!driver_data->setpolicy) && (!driver_data->target)))
1563                 return -EINVAL;
1564
1565         dprintk("trying to register driver %s\n", driver_data->name);
1566
1567         if (driver_data->setpolicy)
1568                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1569
1570         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1571         if (cpufreq_driver) {
1572                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1573                 return -EBUSY;
1574         }
1575         cpufreq_driver = driver_data;
1576         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1577
1578         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1579
1580         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1581                 int i;
1582                 ret = -ENODEV;
1583
1584                 /* check for at least one working CPU */
1585                 for (i=0; i<NR_CPUS; i++)
1586                         if (cpufreq_cpu_data[i])
1587                                 ret = 0;
1588
1589                 /* if all ->init() calls failed, unregister */
1590                 if (ret) {
1591                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1592                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1593
1594                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1595                         cpufreq_driver = NULL;
1596                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1597                 }
1598         }
1599
1600         if (!ret) {
1601                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1602                 dprintk("driver %s up and running\n", driver_data->name);
1603                 cpufreq_debug_enable_ratelimit();
1604         }
1605
1606         return (ret);
1607 }
1608 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1609
1610
1611 /**
1612  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1613  *
1614  *    Unregister the current CPUFreq driver. Only call this if you have
1615  * the right to do so, i.e. if you have succeeded in initialising before!
1616  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1617  * currently not initialised.
1618  */
1619 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1620 {
1621         unsigned long flags;
1622
1623         cpufreq_debug_disable_ratelimit();
1624
1625         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1626                 cpufreq_debug_enable_ratelimit();
1627                 return -EINVAL;
1628         }
1629
1630         dprintk("unregistering driver %s\n", driver->name);
1631
1632         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1633         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1634
1635         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1636         cpufreq_driver = NULL;
1637         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1638
1639         return 0;
1640 }
1641 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);