[PATCH] Reorganize the cpufreq cpu hotplug locking to not be totally bizare
[pandora-kernel.git] / drivers / cpufreq / cpufreq.c
1 /*
2  *  linux/drivers/cpufreq/cpufreq.c
3  *
4  *  Copyright (C) 2001 Russell King
5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
6  *
7  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
8  *      Added handling for CPU hotplug
9  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
10  *      Fix handling for CPU hotplug -- affected CPUs
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License version 2 as
14  * published by the Free Software Foundation.
15  *
16  */
17
18 #include <linux/kernel.h>
19 #include <linux/module.h>
20 #include <linux/init.h>
21 #include <linux/notifier.h>
22 #include <linux/cpufreq.h>
23 #include <linux/delay.h>
24 #include <linux/interrupt.h>
25 #include <linux/spinlock.h>
26 #include <linux/device.h>
27 #include <linux/slab.h>
28 #include <linux/cpu.h>
29 #include <linux/completion.h>
30 #include <linux/mutex.h>
31
32 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
33
34 /**
35  * The "cpufreq driver" - the arch- or hardware-dependend low
36  * level driver of CPUFreq support, and its spinlock. This lock
37  * also protects the cpufreq_cpu_data array.
38  */
39 static struct cpufreq_driver *cpufreq_driver;
40 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
41 static DEFINE_SPINLOCK(cpufreq_driver_lock);
42
43 /* internal prototypes */
44 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
45 static void handle_update(void *data);
46
47 /**
48  * Two notifier lists: the "policy" list is involved in the
49  * validation process for a new CPU frequency policy; the
50  * "transition" list for kernel code that needs to handle
51  * changes to devices when the CPU clock speed changes.
52  * The mutex locks both lists.
53  */
54 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
55 static BLOCKING_NOTIFIER_HEAD(cpufreq_transition_notifier_list);
56
57
58 static LIST_HEAD(cpufreq_governor_list);
59 static DEFINE_MUTEX (cpufreq_governor_mutex);
60
61 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
62 {
63         struct cpufreq_policy *data;
64         unsigned long flags;
65
66         if (cpu >= NR_CPUS)
67                 goto err_out;
68
69         /* get the cpufreq driver */
70         spin_lock_irqsave(&cpufreq_driver_lock, flags);
71
72         if (!cpufreq_driver)
73                 goto err_out_unlock;
74
75         if (!try_module_get(cpufreq_driver->owner))
76                 goto err_out_unlock;
77
78
79         /* get the CPU */
80         data = cpufreq_cpu_data[cpu];
81
82         if (!data)
83                 goto err_out_put_module;
84
85         if (!kobject_get(&data->kobj))
86                 goto err_out_put_module;
87
88         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
89         return data;
90
91 err_out_put_module:
92         module_put(cpufreq_driver->owner);
93 err_out_unlock:
94         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
95 err_out:
96         return NULL;
97 }
98 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
99
100
101 void cpufreq_cpu_put(struct cpufreq_policy *data)
102 {
103         kobject_put(&data->kobj);
104         module_put(cpufreq_driver->owner);
105 }
106 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
107
108
109 /*********************************************************************
110  *                     UNIFIED DEBUG HELPERS                         *
111  *********************************************************************/
112 #ifdef CONFIG_CPU_FREQ_DEBUG
113
114 /* what part(s) of the CPUfreq subsystem are debugged? */
115 static unsigned int debug;
116
117 /* is the debug output ratelimit'ed using printk_ratelimit? User can
118  * set or modify this value.
119  */
120 static unsigned int debug_ratelimit = 1;
121
122 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
123  * loading of a cpufreq driver, temporarily disabled when a new policy
124  * is set, and disabled upon cpufreq driver removal
125  */
126 static unsigned int disable_ratelimit = 1;
127 static DEFINE_SPINLOCK(disable_ratelimit_lock);
128
129 static void cpufreq_debug_enable_ratelimit(void)
130 {
131         unsigned long flags;
132
133         spin_lock_irqsave(&disable_ratelimit_lock, flags);
134         if (disable_ratelimit)
135                 disable_ratelimit--;
136         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
137 }
138
139 static void cpufreq_debug_disable_ratelimit(void)
140 {
141         unsigned long flags;
142
143         spin_lock_irqsave(&disable_ratelimit_lock, flags);
144         disable_ratelimit++;
145         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
146 }
147
148 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
149 {
150         char s[256];
151         va_list args;
152         unsigned int len;
153         unsigned long flags;
154
155         WARN_ON(!prefix);
156         if (type & debug) {
157                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
158                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
159                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
160                         return;
161                 }
162                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
163
164                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
165
166                 va_start(args, fmt);
167                 len += vsnprintf(&s[len], (256 - len), fmt, args);
168                 va_end(args);
169
170                 printk(s);
171
172                 WARN_ON(len < 5);
173         }
174 }
175 EXPORT_SYMBOL(cpufreq_debug_printk);
176
177
178 module_param(debug, uint, 0644);
179 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
180
181 module_param(debug_ratelimit, uint, 0644);
182 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
183
184 #else /* !CONFIG_CPU_FREQ_DEBUG */
185
186 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
187 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
188
189 #endif /* CONFIG_CPU_FREQ_DEBUG */
190
191
192 /*********************************************************************
193  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
194  *********************************************************************/
195
196 /**
197  * adjust_jiffies - adjust the system "loops_per_jiffy"
198  *
199  * This function alters the system "loops_per_jiffy" for the clock
200  * speed change. Note that loops_per_jiffy cannot be updated on SMP
201  * systems as each CPU might be scaled differently. So, use the arch
202  * per-CPU loops_per_jiffy value wherever possible.
203  */
204 #ifndef CONFIG_SMP
205 static unsigned long l_p_j_ref;
206 static unsigned int  l_p_j_ref_freq;
207
208 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
209 {
210         if (ci->flags & CPUFREQ_CONST_LOOPS)
211                 return;
212
213         if (!l_p_j_ref_freq) {
214                 l_p_j_ref = loops_per_jiffy;
215                 l_p_j_ref_freq = ci->old;
216                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
217         }
218         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
219             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
220             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
221                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
222                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
223         }
224 }
225 #else
226 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
227 #endif
228
229
230 /**
231  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
232  * on frequency transition.
233  *
234  * This function calls the transition notifiers and the "adjust_jiffies"
235  * function. It is called twice on all CPU frequency changes that have
236  * external effects.
237  */
238 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
239 {
240         struct cpufreq_policy *policy;
241
242         BUG_ON(irqs_disabled());
243
244         freqs->flags = cpufreq_driver->flags;
245         dprintk("notification %u of frequency transition to %u kHz\n",
246                 state, freqs->new);
247
248         policy = cpufreq_cpu_data[freqs->cpu];
249         switch (state) {
250
251         case CPUFREQ_PRECHANGE:
252                 /* detect if the driver reported a value as "old frequency"
253                  * which is not equal to what the cpufreq core thinks is
254                  * "old frequency".
255                  */
256                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
257                         if ((policy) && (policy->cpu == freqs->cpu) &&
258                             (policy->cur) && (policy->cur != freqs->old)) {
259                                 dprintk("Warning: CPU frequency is"
260                                         " %u, cpufreq assumed %u kHz.\n",
261                                         freqs->old, policy->cur);
262                                 freqs->old = policy->cur;
263                         }
264                 }
265                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
266                                 CPUFREQ_PRECHANGE, freqs);
267                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
268                 break;
269
270         case CPUFREQ_POSTCHANGE:
271                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
272                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
273                                 CPUFREQ_POSTCHANGE, freqs);
274                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
275                         policy->cur = freqs->new;
276                 break;
277         }
278 }
279 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
280
281
282
283 /*********************************************************************
284  *                          SYSFS INTERFACE                          *
285  *********************************************************************/
286
287 /**
288  * cpufreq_parse_governor - parse a governor string
289  */
290 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
291                                 struct cpufreq_governor **governor)
292 {
293         if (!cpufreq_driver)
294                 return -EINVAL;
295         if (cpufreq_driver->setpolicy) {
296                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
297                         *policy = CPUFREQ_POLICY_PERFORMANCE;
298                         return 0;
299                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
300                         *policy = CPUFREQ_POLICY_POWERSAVE;
301                         return 0;
302                 }
303                 return -EINVAL;
304         } else {
305                 struct cpufreq_governor *t;
306                 mutex_lock(&cpufreq_governor_mutex);
307                 if (!cpufreq_driver || !cpufreq_driver->target)
308                         goto out;
309                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
310                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
311                                 *governor = t;
312                                 mutex_unlock(&cpufreq_governor_mutex);
313                                 return 0;
314                         }
315                 }
316 out:
317                 mutex_unlock(&cpufreq_governor_mutex);
318         }
319         return -EINVAL;
320 }
321
322
323 /* drivers/base/cpu.c */
324 extern struct sysdev_class cpu_sysdev_class;
325
326
327 /**
328  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
329  *
330  * Write out information from cpufreq_driver->policy[cpu]; object must be
331  * "unsigned int".
332  */
333
334 #define show_one(file_name, object)                     \
335 static ssize_t show_##file_name                         \
336 (struct cpufreq_policy * policy, char *buf)             \
337 {                                                       \
338         return sprintf (buf, "%u\n", policy->object);   \
339 }
340
341 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
342 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
343 show_one(scaling_min_freq, min);
344 show_one(scaling_max_freq, max);
345 show_one(scaling_cur_freq, cur);
346
347 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
348
349 /**
350  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
351  */
352 #define store_one(file_name, object)                    \
353 static ssize_t store_##file_name                                        \
354 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
355 {                                                                       \
356         unsigned int ret = -EINVAL;                                     \
357         struct cpufreq_policy new_policy;                               \
358                                                                         \
359         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
360         if (ret)                                                        \
361                 return -EINVAL;                                         \
362                                                                         \
363         ret = sscanf (buf, "%u", &new_policy.object);                   \
364         if (ret != 1)                                                   \
365                 return -EINVAL;                                         \
366                                                                         \
367         lock_cpu_hotplug();                                             \
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         unlock_cpu_hotplug();                                           \
373                                                                         \
374         return ret ? ret : count;                                       \
375 }
376
377 store_one(scaling_min_freq,min);
378 store_one(scaling_max_freq,max);
379
380 /**
381  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
382  */
383 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
384 {
385         unsigned int cur_freq = cpufreq_get(policy->cpu);
386         if (!cur_freq)
387                 return sprintf(buf, "<unknown>");
388         return sprintf(buf, "%u\n", cur_freq);
389 }
390
391
392 /**
393  * show_scaling_governor - show the current policy for the specified CPU
394  */
395 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
396 {
397         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
398                 return sprintf(buf, "powersave\n");
399         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
400                 return sprintf(buf, "performance\n");
401         else if (policy->governor)
402                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
403         return -EINVAL;
404 }
405
406
407 /**
408  * store_scaling_governor - store policy for the specified CPU
409  */
410 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
411                                        const char *buf, size_t count)
412 {
413         unsigned int ret = -EINVAL;
414         char    str_governor[16];
415         struct cpufreq_policy new_policy;
416
417         ret = cpufreq_get_policy(&new_policy, policy->cpu);
418         if (ret)
419                 return ret;
420
421         ret = sscanf (buf, "%15s", str_governor);
422         if (ret != 1)
423                 return -EINVAL;
424
425         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
426                 return -EINVAL;
427
428         lock_cpu_hotplug();
429
430         /* Do not use cpufreq_set_policy here or the user_policy.max
431            will be wrongly overridden */
432         mutex_lock(&policy->lock);
433         ret = __cpufreq_set_policy(policy, &new_policy);
434
435         policy->user_policy.policy = policy->policy;
436         policy->user_policy.governor = policy->governor;
437         mutex_unlock(&policy->lock);
438
439         unlock_cpu_hotplug();
440
441         return ret ? ret : count;
442 }
443
444 /**
445  * show_scaling_driver - show the cpufreq driver currently loaded
446  */
447 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
448 {
449         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
450 }
451
452 /**
453  * show_scaling_available_governors - show the available CPUfreq governors
454  */
455 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
456                                 char *buf)
457 {
458         ssize_t i = 0;
459         struct cpufreq_governor *t;
460
461         if (!cpufreq_driver->target) {
462                 i += sprintf(buf, "performance powersave");
463                 goto out;
464         }
465
466         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
467                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
468                         goto out;
469                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
470         }
471 out:
472         i += sprintf(&buf[i], "\n");
473         return i;
474 }
475 /**
476  * show_affected_cpus - show the CPUs affected by each transition
477  */
478 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
479 {
480         ssize_t i = 0;
481         unsigned int cpu;
482
483         for_each_cpu_mask(cpu, policy->cpus) {
484                 if (i)
485                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
486                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
487                 if (i >= (PAGE_SIZE - 5))
488                     break;
489         }
490         i += sprintf(&buf[i], "\n");
491         return i;
492 }
493
494
495 #define define_one_ro(_name) \
496 static struct freq_attr _name = \
497 __ATTR(_name, 0444, show_##_name, NULL)
498
499 #define define_one_ro0400(_name) \
500 static struct freq_attr _name = \
501 __ATTR(_name, 0400, show_##_name, NULL)
502
503 #define define_one_rw(_name) \
504 static struct freq_attr _name = \
505 __ATTR(_name, 0644, show_##_name, store_##_name)
506
507 define_one_ro0400(cpuinfo_cur_freq);
508 define_one_ro(cpuinfo_min_freq);
509 define_one_ro(cpuinfo_max_freq);
510 define_one_ro(scaling_available_governors);
511 define_one_ro(scaling_driver);
512 define_one_ro(scaling_cur_freq);
513 define_one_ro(affected_cpus);
514 define_one_rw(scaling_min_freq);
515 define_one_rw(scaling_max_freq);
516 define_one_rw(scaling_governor);
517
518 static struct attribute * default_attrs[] = {
519         &cpuinfo_min_freq.attr,
520         &cpuinfo_max_freq.attr,
521         &scaling_min_freq.attr,
522         &scaling_max_freq.attr,
523         &affected_cpus.attr,
524         &scaling_governor.attr,
525         &scaling_driver.attr,
526         &scaling_available_governors.attr,
527         NULL
528 };
529
530 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
531 #define to_attr(a) container_of(a,struct freq_attr,attr)
532
533 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
534 {
535         struct cpufreq_policy * policy = to_policy(kobj);
536         struct freq_attr * fattr = to_attr(attr);
537         ssize_t ret;
538         policy = cpufreq_cpu_get(policy->cpu);
539         if (!policy)
540                 return -EINVAL;
541         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
542         cpufreq_cpu_put(policy);
543         return ret;
544 }
545
546 static ssize_t store(struct kobject * kobj, struct attribute * attr,
547                      const char * buf, size_t count)
548 {
549         struct cpufreq_policy * policy = to_policy(kobj);
550         struct freq_attr * fattr = to_attr(attr);
551         ssize_t ret;
552         policy = cpufreq_cpu_get(policy->cpu);
553         if (!policy)
554                 return -EINVAL;
555         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
556         cpufreq_cpu_put(policy);
557         return ret;
558 }
559
560 static void cpufreq_sysfs_release(struct kobject * kobj)
561 {
562         struct cpufreq_policy * policy = to_policy(kobj);
563         dprintk("last reference is dropped\n");
564         complete(&policy->kobj_unregister);
565 }
566
567 static struct sysfs_ops sysfs_ops = {
568         .show   = show,
569         .store  = store,
570 };
571
572 static struct kobj_type ktype_cpufreq = {
573         .sysfs_ops      = &sysfs_ops,
574         .default_attrs  = default_attrs,
575         .release        = cpufreq_sysfs_release,
576 };
577
578
579 /**
580  * cpufreq_add_dev - add a CPU device
581  *
582  * Adds the cpufreq interface for a CPU device.
583  */
584 static int cpufreq_add_dev (struct sys_device * sys_dev)
585 {
586         unsigned int cpu = sys_dev->id;
587         int ret = 0;
588         struct cpufreq_policy new_policy;
589         struct cpufreq_policy *policy;
590         struct freq_attr **drv_attr;
591         struct sys_device *cpu_sys_dev;
592         unsigned long flags;
593         unsigned int j;
594 #ifdef CONFIG_SMP
595         struct cpufreq_policy *managed_policy;
596 #endif
597
598         if (cpu_is_offline(cpu))
599                 return 0;
600
601         cpufreq_debug_disable_ratelimit();
602         dprintk("adding CPU %u\n", cpu);
603
604 #ifdef CONFIG_SMP
605         /* check whether a different CPU already registered this
606          * CPU because it is in the same boat. */
607         policy = cpufreq_cpu_get(cpu);
608         if (unlikely(policy)) {
609                 cpufreq_cpu_put(policy);
610                 cpufreq_debug_enable_ratelimit();
611                 return 0;
612         }
613 #endif
614
615         if (!try_module_get(cpufreq_driver->owner)) {
616                 ret = -EINVAL;
617                 goto module_out;
618         }
619
620         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
621         if (!policy) {
622                 ret = -ENOMEM;
623                 goto nomem_out;
624         }
625
626         policy->cpu = cpu;
627         policy->cpus = cpumask_of_cpu(cpu);
628
629         mutex_init(&policy->lock);
630         mutex_lock(&policy->lock);
631         init_completion(&policy->kobj_unregister);
632         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
633
634         /* call driver. From then on the cpufreq must be able
635          * to accept all calls to ->verify and ->setpolicy for this CPU
636          */
637         ret = cpufreq_driver->init(policy);
638         if (ret) {
639                 dprintk("initialization failed\n");
640                 mutex_unlock(&policy->lock);
641                 goto err_out;
642         }
643
644 #ifdef CONFIG_SMP
645         for_each_cpu_mask(j, policy->cpus) {
646                 if (cpu == j)
647                         continue;
648
649                 /* check for existing affected CPUs.  They may not be aware
650                  * of it due to CPU Hotplug.
651                  */
652                 managed_policy = cpufreq_cpu_get(j);
653                 if (unlikely(managed_policy)) {
654                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
655                         managed_policy->cpus = policy->cpus;
656                         cpufreq_cpu_data[cpu] = managed_policy;
657                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
658
659                         dprintk("CPU already managed, adding link\n");
660                         sysfs_create_link(&sys_dev->kobj,
661                                           &managed_policy->kobj, "cpufreq");
662
663                         cpufreq_debug_enable_ratelimit();
664                         mutex_unlock(&policy->lock);
665                         ret = 0;
666                         goto err_out_driver_exit; /* call driver->exit() */
667                 }
668         }
669 #endif
670         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
671
672         /* prepare interface data */
673         policy->kobj.parent = &sys_dev->kobj;
674         policy->kobj.ktype = &ktype_cpufreq;
675         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
676
677         ret = kobject_register(&policy->kobj);
678         if (ret) {
679                 mutex_unlock(&policy->lock);
680                 goto err_out_driver_exit;
681         }
682         /* set up files for this cpu device */
683         drv_attr = cpufreq_driver->attr;
684         while ((drv_attr) && (*drv_attr)) {
685                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
686                 drv_attr++;
687         }
688         if (cpufreq_driver->get)
689                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
690         if (cpufreq_driver->target)
691                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
692
693         spin_lock_irqsave(&cpufreq_driver_lock, flags);
694         for_each_cpu_mask(j, policy->cpus)
695                 cpufreq_cpu_data[j] = policy;
696         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
697
698         /* symlink affected CPUs */
699         for_each_cpu_mask(j, policy->cpus) {
700                 if (j == cpu)
701                         continue;
702                 if (!cpu_online(j))
703                         continue;
704
705                 dprintk("CPU %u already managed, adding link\n", j);
706                 cpufreq_cpu_get(cpu);
707                 cpu_sys_dev = get_cpu_sysdev(j);
708                 sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
709                                   "cpufreq");
710         }
711
712         policy->governor = NULL; /* to assure that the starting sequence is
713                                   * run in cpufreq_set_policy */
714         mutex_unlock(&policy->lock);
715
716         /* set default policy */
717         ret = cpufreq_set_policy(&new_policy);
718         if (ret) {
719                 dprintk("setting policy failed\n");
720                 goto err_out_unregister;
721         }
722
723         module_put(cpufreq_driver->owner);
724         dprintk("initialization complete\n");
725         cpufreq_debug_enable_ratelimit();
726
727         return 0;
728
729
730 err_out_unregister:
731         spin_lock_irqsave(&cpufreq_driver_lock, flags);
732         for_each_cpu_mask(j, policy->cpus)
733                 cpufreq_cpu_data[j] = NULL;
734         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
735
736         kobject_unregister(&policy->kobj);
737         wait_for_completion(&policy->kobj_unregister);
738
739 err_out_driver_exit:
740         if (cpufreq_driver->exit)
741                 cpufreq_driver->exit(policy);
742
743 err_out:
744         kfree(policy);
745
746 nomem_out:
747         module_put(cpufreq_driver->owner);
748 module_out:
749         cpufreq_debug_enable_ratelimit();
750         return ret;
751 }
752
753
754 /**
755  * cpufreq_remove_dev - remove a CPU device
756  *
757  * Removes the cpufreq interface for a CPU device.
758  */
759 static int cpufreq_remove_dev (struct sys_device * sys_dev)
760 {
761         unsigned int cpu = sys_dev->id;
762         unsigned long flags;
763         struct cpufreq_policy *data;
764 #ifdef CONFIG_SMP
765         struct sys_device *cpu_sys_dev;
766         unsigned int j;
767 #endif
768
769         cpufreq_debug_disable_ratelimit();
770         dprintk("unregistering CPU %u\n", cpu);
771
772         spin_lock_irqsave(&cpufreq_driver_lock, flags);
773         data = cpufreq_cpu_data[cpu];
774
775         if (!data) {
776                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
777                 cpufreq_debug_enable_ratelimit();
778                 return -EINVAL;
779         }
780         cpufreq_cpu_data[cpu] = NULL;
781
782
783 #ifdef CONFIG_SMP
784         /* if this isn't the CPU which is the parent of the kobj, we
785          * only need to unlink, put and exit
786          */
787         if (unlikely(cpu != data->cpu)) {
788                 dprintk("removing link\n");
789                 cpu_clear(cpu, data->cpus);
790                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
791                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
792                 cpufreq_cpu_put(data);
793                 cpufreq_debug_enable_ratelimit();
794                 return 0;
795         }
796 #endif
797
798
799         if (!kobject_get(&data->kobj)) {
800                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
801                 cpufreq_debug_enable_ratelimit();
802                 return -EFAULT;
803         }
804
805 #ifdef CONFIG_SMP
806         /* if we have other CPUs still registered, we need to unlink them,
807          * or else wait_for_completion below will lock up. Clean the
808          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
809          * links afterwards.
810          */
811         if (unlikely(cpus_weight(data->cpus) > 1)) {
812                 for_each_cpu_mask(j, data->cpus) {
813                         if (j == cpu)
814                                 continue;
815                         cpufreq_cpu_data[j] = NULL;
816                 }
817         }
818
819         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
820
821         if (unlikely(cpus_weight(data->cpus) > 1)) {
822                 for_each_cpu_mask(j, data->cpus) {
823                         if (j == cpu)
824                                 continue;
825                         dprintk("removing link for cpu %u\n", j);
826                         cpu_sys_dev = get_cpu_sysdev(j);
827                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
828                         cpufreq_cpu_put(data);
829                 }
830         }
831 #else
832         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
833 #endif
834
835         mutex_lock(&data->lock);
836         if (cpufreq_driver->target)
837                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
838         mutex_unlock(&data->lock);
839
840         kobject_unregister(&data->kobj);
841
842         kobject_put(&data->kobj);
843
844         /* we need to make sure that the underlying kobj is actually
845          * not referenced anymore by anybody before we proceed with
846          * unloading.
847          */
848         dprintk("waiting for dropping of refcount\n");
849         wait_for_completion(&data->kobj_unregister);
850         dprintk("wait complete\n");
851
852         if (cpufreq_driver->exit)
853                 cpufreq_driver->exit(data);
854
855         kfree(data);
856
857         cpufreq_debug_enable_ratelimit();
858         return 0;
859 }
860
861
862 static void handle_update(void *data)
863 {
864         unsigned int cpu = (unsigned int)(long)data;
865         dprintk("handle_update for cpu %u called\n", cpu);
866         cpufreq_update_policy(cpu);
867 }
868
869 /**
870  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
871  *      @cpu: cpu number
872  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
873  *      @new_freq: CPU frequency the CPU actually runs at
874  *
875  *      We adjust to current frequency first, and need to clean up later. So either call
876  *      to cpufreq_update_policy() or schedule handle_update()).
877  */
878 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
879 {
880         struct cpufreq_freqs freqs;
881
882         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
883                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
884
885         freqs.cpu = cpu;
886         freqs.old = old_freq;
887         freqs.new = new_freq;
888         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
889         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
890 }
891
892
893 /**
894  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
895  * @cpu: CPU number
896  *
897  * This is the last known freq, without actually getting it from the driver.
898  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
899  */
900 unsigned int cpufreq_quick_get(unsigned int cpu)
901 {
902         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
903         unsigned int ret = 0;
904
905         if (policy) {
906                 mutex_lock(&policy->lock);
907                 ret = policy->cur;
908                 mutex_unlock(&policy->lock);
909                 cpufreq_cpu_put(policy);
910         }
911
912         return (ret);
913 }
914 EXPORT_SYMBOL(cpufreq_quick_get);
915
916
917 /**
918  * cpufreq_get - get the current CPU frequency (in kHz)
919  * @cpu: CPU number
920  *
921  * Get the CPU current (static) CPU frequency
922  */
923 unsigned int cpufreq_get(unsigned int cpu)
924 {
925         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
926         unsigned int ret = 0;
927
928         if (!policy)
929                 return 0;
930
931         if (!cpufreq_driver->get)
932                 goto out;
933
934         mutex_lock(&policy->lock);
935
936         ret = cpufreq_driver->get(cpu);
937
938         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
939                 /* verify no discrepancy between actual and saved value exists */
940                 if (unlikely(ret != policy->cur)) {
941                         cpufreq_out_of_sync(cpu, policy->cur, ret);
942                         schedule_work(&policy->update);
943                 }
944         }
945
946         mutex_unlock(&policy->lock);
947
948 out:
949         cpufreq_cpu_put(policy);
950
951         return (ret);
952 }
953 EXPORT_SYMBOL(cpufreq_get);
954
955
956 /**
957  *      cpufreq_suspend - let the low level driver prepare for suspend
958  */
959
960 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
961 {
962         int cpu = sysdev->id;
963         unsigned int ret = 0;
964         unsigned int cur_freq = 0;
965         struct cpufreq_policy *cpu_policy;
966
967         dprintk("resuming cpu %u\n", cpu);
968
969         if (!cpu_online(cpu))
970                 return 0;
971
972         /* we may be lax here as interrupts are off. Nonetheless
973          * we need to grab the correct cpu policy, as to check
974          * whether we really run on this CPU.
975          */
976
977         cpu_policy = cpufreq_cpu_get(cpu);
978         if (!cpu_policy)
979                 return -EINVAL;
980
981         /* only handle each CPU group once */
982         if (unlikely(cpu_policy->cpu != cpu)) {
983                 cpufreq_cpu_put(cpu_policy);
984                 return 0;
985         }
986
987         if (cpufreq_driver->suspend) {
988                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
989                 if (ret) {
990                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
991                                         "step on CPU %u\n", cpu_policy->cpu);
992                         cpufreq_cpu_put(cpu_policy);
993                         return ret;
994                 }
995         }
996
997
998         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
999                 goto out;
1000
1001         if (cpufreq_driver->get)
1002                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1003
1004         if (!cur_freq || !cpu_policy->cur) {
1005                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1006                        "frequency is what timing core thinks it is.\n");
1007                 goto out;
1008         }
1009
1010         if (unlikely(cur_freq != cpu_policy->cur)) {
1011                 struct cpufreq_freqs freqs;
1012
1013                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1014                         dprintk("Warning: CPU frequency is %u, "
1015                                "cpufreq assumed %u kHz.\n",
1016                                cur_freq, cpu_policy->cur);
1017
1018                 freqs.cpu = cpu;
1019                 freqs.old = cpu_policy->cur;
1020                 freqs.new = cur_freq;
1021
1022                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
1023                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1024                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1025
1026                 cpu_policy->cur = cur_freq;
1027         }
1028
1029 out:
1030         cpufreq_cpu_put(cpu_policy);
1031         return 0;
1032 }
1033
1034 /**
1035  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1036  *
1037  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1038  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1039  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1040  *          restored.
1041  */
1042 static int cpufreq_resume(struct sys_device * sysdev)
1043 {
1044         int cpu = sysdev->id;
1045         unsigned int ret = 0;
1046         struct cpufreq_policy *cpu_policy;
1047
1048         dprintk("resuming cpu %u\n", cpu);
1049
1050         if (!cpu_online(cpu))
1051                 return 0;
1052
1053         /* we may be lax here as interrupts are off. Nonetheless
1054          * we need to grab the correct cpu policy, as to check
1055          * whether we really run on this CPU.
1056          */
1057
1058         cpu_policy = cpufreq_cpu_get(cpu);
1059         if (!cpu_policy)
1060                 return -EINVAL;
1061
1062         /* only handle each CPU group once */
1063         if (unlikely(cpu_policy->cpu != cpu)) {
1064                 cpufreq_cpu_put(cpu_policy);
1065                 return 0;
1066         }
1067
1068         if (cpufreq_driver->resume) {
1069                 ret = cpufreq_driver->resume(cpu_policy);
1070                 if (ret) {
1071                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1072                                         "step on CPU %u\n", cpu_policy->cpu);
1073                         cpufreq_cpu_put(cpu_policy);
1074                         return ret;
1075                 }
1076         }
1077
1078         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1079                 unsigned int cur_freq = 0;
1080
1081                 if (cpufreq_driver->get)
1082                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1083
1084                 if (!cur_freq || !cpu_policy->cur) {
1085                         printk(KERN_ERR "cpufreq: resume failed to assert "
1086                                         "current frequency is what timing core "
1087                                         "thinks it is.\n");
1088                         goto out;
1089                 }
1090
1091                 if (unlikely(cur_freq != cpu_policy->cur)) {
1092                         struct cpufreq_freqs freqs;
1093
1094                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1095                                 dprintk("Warning: CPU frequency"
1096                                        "is %u, cpufreq assumed %u kHz.\n",
1097                                        cur_freq, cpu_policy->cur);
1098
1099                         freqs.cpu = cpu;
1100                         freqs.old = cpu_policy->cur;
1101                         freqs.new = cur_freq;
1102
1103                         blocking_notifier_call_chain(
1104                                         &cpufreq_transition_notifier_list,
1105                                         CPUFREQ_RESUMECHANGE, &freqs);
1106                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1107
1108                         cpu_policy->cur = cur_freq;
1109                 }
1110         }
1111
1112 out:
1113         schedule_work(&cpu_policy->update);
1114         cpufreq_cpu_put(cpu_policy);
1115         return ret;
1116 }
1117
1118 static struct sysdev_driver cpufreq_sysdev_driver = {
1119         .add            = cpufreq_add_dev,
1120         .remove         = cpufreq_remove_dev,
1121         .suspend        = cpufreq_suspend,
1122         .resume         = cpufreq_resume,
1123 };
1124
1125
1126 /*********************************************************************
1127  *                     NOTIFIER LISTS INTERFACE                      *
1128  *********************************************************************/
1129
1130 /**
1131  *      cpufreq_register_notifier - register a driver with cpufreq
1132  *      @nb: notifier function to register
1133  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1134  *
1135  *      Add a driver to one of two lists: either a list of drivers that
1136  *      are notified about clock rate changes (once before and once after
1137  *      the transition), or a list of drivers that are notified about
1138  *      changes in cpufreq policy.
1139  *
1140  *      This function may sleep, and has the same return conditions as
1141  *      blocking_notifier_chain_register.
1142  */
1143 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1144 {
1145         int ret;
1146
1147         switch (list) {
1148         case CPUFREQ_TRANSITION_NOTIFIER:
1149                 ret = blocking_notifier_chain_register(
1150                                 &cpufreq_transition_notifier_list, nb);
1151                 break;
1152         case CPUFREQ_POLICY_NOTIFIER:
1153                 ret = blocking_notifier_chain_register(
1154                                 &cpufreq_policy_notifier_list, nb);
1155                 break;
1156         default:
1157                 ret = -EINVAL;
1158         }
1159
1160         return ret;
1161 }
1162 EXPORT_SYMBOL(cpufreq_register_notifier);
1163
1164
1165 /**
1166  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1167  *      @nb: notifier block to be unregistered
1168  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1169  *
1170  *      Remove a driver from the CPU frequency notifier list.
1171  *
1172  *      This function may sleep, and has the same return conditions as
1173  *      blocking_notifier_chain_unregister.
1174  */
1175 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1176 {
1177         int ret;
1178
1179         switch (list) {
1180         case CPUFREQ_TRANSITION_NOTIFIER:
1181                 ret = blocking_notifier_chain_unregister(
1182                                 &cpufreq_transition_notifier_list, nb);
1183                 break;
1184         case CPUFREQ_POLICY_NOTIFIER:
1185                 ret = blocking_notifier_chain_unregister(
1186                                 &cpufreq_policy_notifier_list, nb);
1187                 break;
1188         default:
1189                 ret = -EINVAL;
1190         }
1191
1192         return ret;
1193 }
1194 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1195
1196
1197 /*********************************************************************
1198  *                              GOVERNORS                            *
1199  *********************************************************************/
1200
1201
1202 /* Must be called with lock_cpu_hotplug held */
1203 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1204                             unsigned int target_freq,
1205                             unsigned int relation)
1206 {
1207         int retval = -EINVAL;
1208
1209         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1210                 target_freq, relation);
1211         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1212                 retval = cpufreq_driver->target(policy, target_freq, relation);
1213
1214         return retval;
1215 }
1216 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1217
1218 int cpufreq_driver_target(struct cpufreq_policy *policy,
1219                           unsigned int target_freq,
1220                           unsigned int relation)
1221 {
1222         int ret;
1223
1224         policy = cpufreq_cpu_get(policy->cpu);
1225         if (!policy)
1226                 return -EINVAL;
1227
1228         lock_cpu_hotplug();
1229         mutex_lock(&policy->lock);
1230
1231         ret = __cpufreq_driver_target(policy, target_freq, relation);
1232
1233         mutex_unlock(&policy->lock);
1234         unlock_cpu_hotplug();
1235
1236         cpufreq_cpu_put(policy);
1237         return ret;
1238 }
1239 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1240
1241 /*
1242  * Locking: Must be called with the lock_cpu_hotplug() lock held
1243  * when "event" is CPUFREQ_GOV_LIMITS
1244  */
1245
1246 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1247 {
1248         int ret;
1249
1250         if (!try_module_get(policy->governor->owner))
1251                 return -EINVAL;
1252
1253         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1254         ret = policy->governor->governor(policy, event);
1255
1256         /* we keep one module reference alive for each CPU governed by this CPU */
1257         if ((event != CPUFREQ_GOV_START) || ret)
1258                 module_put(policy->governor->owner);
1259         if ((event == CPUFREQ_GOV_STOP) && !ret)
1260                 module_put(policy->governor->owner);
1261
1262         return ret;
1263 }
1264
1265
1266 int cpufreq_register_governor(struct cpufreq_governor *governor)
1267 {
1268         struct cpufreq_governor *t;
1269
1270         if (!governor)
1271                 return -EINVAL;
1272
1273         mutex_lock(&cpufreq_governor_mutex);
1274
1275         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1276                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1277                         mutex_unlock(&cpufreq_governor_mutex);
1278                         return -EBUSY;
1279                 }
1280         }
1281         list_add(&governor->governor_list, &cpufreq_governor_list);
1282
1283         mutex_unlock(&cpufreq_governor_mutex);
1284         return 0;
1285 }
1286 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1287
1288
1289 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1290 {
1291         if (!governor)
1292                 return;
1293
1294         mutex_lock(&cpufreq_governor_mutex);
1295         list_del(&governor->governor_list);
1296         mutex_unlock(&cpufreq_governor_mutex);
1297         return;
1298 }
1299 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1300
1301
1302
1303 /*********************************************************************
1304  *                          POLICY INTERFACE                         *
1305  *********************************************************************/
1306
1307 /**
1308  * cpufreq_get_policy - get the current cpufreq_policy
1309  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1310  *
1311  * Reads the current cpufreq policy.
1312  */
1313 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1314 {
1315         struct cpufreq_policy *cpu_policy;
1316         if (!policy)
1317                 return -EINVAL;
1318
1319         cpu_policy = cpufreq_cpu_get(cpu);
1320         if (!cpu_policy)
1321                 return -EINVAL;
1322
1323         mutex_lock(&cpu_policy->lock);
1324         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1325         mutex_unlock(&cpu_policy->lock);
1326
1327         cpufreq_cpu_put(cpu_policy);
1328         return 0;
1329 }
1330 EXPORT_SYMBOL(cpufreq_get_policy);
1331
1332
1333 /*
1334  * Locking: Must be called with the lock_cpu_hotplug() lock held
1335  */
1336 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1337 {
1338         int ret = 0;
1339
1340         cpufreq_debug_disable_ratelimit();
1341         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1342                 policy->min, policy->max);
1343
1344         memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
1345
1346         /* verify the cpu speed can be set within this limit */
1347         ret = cpufreq_driver->verify(policy);
1348         if (ret)
1349                 goto error_out;
1350
1351         /* adjust if necessary - all reasons */
1352         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1353                         CPUFREQ_ADJUST, policy);
1354
1355         /* adjust if necessary - hardware incompatibility*/
1356         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1357                         CPUFREQ_INCOMPATIBLE, policy);
1358
1359         /* verify the cpu speed can be set within this limit,
1360            which might be different to the first one */
1361         ret = cpufreq_driver->verify(policy);
1362         if (ret)
1363                 goto error_out;
1364
1365         /* notification of the new policy */
1366         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1367                         CPUFREQ_NOTIFY, policy);
1368
1369         data->min = policy->min;
1370         data->max = policy->max;
1371
1372         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1373
1374         if (cpufreq_driver->setpolicy) {
1375                 data->policy = policy->policy;
1376                 dprintk("setting range\n");
1377                 ret = cpufreq_driver->setpolicy(policy);
1378         } else {
1379                 if (policy->governor != data->governor) {
1380                         /* save old, working values */
1381                         struct cpufreq_governor *old_gov = data->governor;
1382
1383                         dprintk("governor switch\n");
1384
1385                         /* end old governor */
1386                         if (data->governor)
1387                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1388
1389                         /* start new governor */
1390                         data->governor = policy->governor;
1391                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1392                                 /* new governor failed, so re-start old one */
1393                                 dprintk("starting governor %s failed\n", data->governor->name);
1394                                 if (old_gov) {
1395                                         data->governor = old_gov;
1396                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1397                                 }
1398                                 ret = -EINVAL;
1399                                 goto error_out;
1400                         }
1401                         /* might be a policy change, too, so fall through */
1402                 }
1403                 dprintk("governor: change or update limits\n");
1404                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1405         }
1406
1407 error_out:
1408         cpufreq_debug_enable_ratelimit();
1409         return ret;
1410 }
1411
1412 /**
1413  *      cpufreq_set_policy - set a new CPUFreq policy
1414  *      @policy: policy to be set.
1415  *
1416  *      Sets a new CPU frequency and voltage scaling policy.
1417  */
1418 int cpufreq_set_policy(struct cpufreq_policy *policy)
1419 {
1420         int ret = 0;
1421         struct cpufreq_policy *data;
1422
1423         if (!policy)
1424                 return -EINVAL;
1425
1426         data = cpufreq_cpu_get(policy->cpu);
1427         if (!data)
1428                 return -EINVAL;
1429
1430         lock_cpu_hotplug();
1431
1432         /* lock this CPU */
1433         mutex_lock(&data->lock);
1434
1435         ret = __cpufreq_set_policy(data, policy);
1436         data->user_policy.min = data->min;
1437         data->user_policy.max = data->max;
1438         data->user_policy.policy = data->policy;
1439         data->user_policy.governor = data->governor;
1440
1441         mutex_unlock(&data->lock);
1442
1443         unlock_cpu_hotplug();
1444         cpufreq_cpu_put(data);
1445
1446         return ret;
1447 }
1448 EXPORT_SYMBOL(cpufreq_set_policy);
1449
1450
1451 /**
1452  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1453  *      @cpu: CPU which shall be re-evaluated
1454  *
1455  *      Usefull for policy notifiers which have different necessities
1456  *      at different times.
1457  */
1458 int cpufreq_update_policy(unsigned int cpu)
1459 {
1460         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1461         struct cpufreq_policy policy;
1462         int ret = 0;
1463
1464         if (!data)
1465                 return -ENODEV;
1466
1467         lock_cpu_hotplug();
1468         mutex_lock(&data->lock);
1469
1470         dprintk("updating policy for CPU %u\n", cpu);
1471         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1472         policy.min = data->user_policy.min;
1473         policy.max = data->user_policy.max;
1474         policy.policy = data->user_policy.policy;
1475         policy.governor = data->user_policy.governor;
1476
1477         /* BIOS might change freq behind our back
1478           -> ask driver for current freq and notify governors about a change */
1479         if (cpufreq_driver->get) {
1480                 policy.cur = cpufreq_driver->get(cpu);
1481                 if (!data->cur) {
1482                         dprintk("Driver did not initialize current freq");
1483                         data->cur = policy.cur;
1484                 } else {
1485                         if (data->cur != policy.cur)
1486                                 cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1487                 }
1488         }
1489
1490         ret = __cpufreq_set_policy(data, &policy);
1491
1492         mutex_unlock(&data->lock);
1493         unlock_cpu_hotplug();
1494         cpufreq_cpu_put(data);
1495         return ret;
1496 }
1497 EXPORT_SYMBOL(cpufreq_update_policy);
1498
1499 #ifdef CONFIG_HOTPLUG_CPU
1500 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1501                                         unsigned long action, void *hcpu)
1502 {
1503         unsigned int cpu = (unsigned long)hcpu;
1504         struct cpufreq_policy *policy;
1505         struct sys_device *sys_dev;
1506
1507         sys_dev = get_cpu_sysdev(cpu);
1508
1509         if (sys_dev) {
1510                 switch (action) {
1511                 case CPU_ONLINE:
1512                         cpufreq_add_dev(sys_dev);
1513                         break;
1514                 case CPU_DOWN_PREPARE:
1515                         /*
1516                          * We attempt to put this cpu in lowest frequency
1517                          * possible before going down. This will permit
1518                          * hardware-managed P-State to switch other related
1519                          * threads to min or higher speeds if possible.
1520                          */
1521                         policy = cpufreq_cpu_data[cpu];
1522                         if (policy) {
1523                                 cpufreq_driver_target(policy, policy->min,
1524                                                 CPUFREQ_RELATION_H);
1525                         }
1526                         break;
1527                 case CPU_DEAD:
1528                         cpufreq_remove_dev(sys_dev);
1529                         break;
1530                 }
1531         }
1532         return NOTIFY_OK;
1533 }
1534
1535 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1536 {
1537     .notifier_call = cpufreq_cpu_callback,
1538 };
1539 #endif /* CONFIG_HOTPLUG_CPU */
1540
1541 /*********************************************************************
1542  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1543  *********************************************************************/
1544
1545 /**
1546  * cpufreq_register_driver - register a CPU Frequency driver
1547  * @driver_data: A struct cpufreq_driver containing the values#
1548  * submitted by the CPU Frequency driver.
1549  *
1550  *   Registers a CPU Frequency driver to this core code. This code
1551  * returns zero on success, -EBUSY when another driver got here first
1552  * (and isn't unregistered in the meantime).
1553  *
1554  */
1555 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1556 {
1557         unsigned long flags;
1558         int ret;
1559
1560         if (!driver_data || !driver_data->verify || !driver_data->init ||
1561             ((!driver_data->setpolicy) && (!driver_data->target)))
1562                 return -EINVAL;
1563
1564         dprintk("trying to register driver %s\n", driver_data->name);
1565
1566         if (driver_data->setpolicy)
1567                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1568
1569         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1570         if (cpufreq_driver) {
1571                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1572                 return -EBUSY;
1573         }
1574         cpufreq_driver = driver_data;
1575         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1576
1577         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1578
1579         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1580                 int i;
1581                 ret = -ENODEV;
1582
1583                 /* check for at least one working CPU */
1584                 for (i=0; i<NR_CPUS; i++)
1585                         if (cpufreq_cpu_data[i])
1586                                 ret = 0;
1587
1588                 /* if all ->init() calls failed, unregister */
1589                 if (ret) {
1590                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1591                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1592
1593                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1594                         cpufreq_driver = NULL;
1595                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1596                 }
1597         }
1598
1599         if (!ret) {
1600                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1601                 dprintk("driver %s up and running\n", driver_data->name);
1602                 cpufreq_debug_enable_ratelimit();
1603         }
1604
1605         return (ret);
1606 }
1607 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1608
1609
1610 /**
1611  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1612  *
1613  *    Unregister the current CPUFreq driver. Only call this if you have
1614  * the right to do so, i.e. if you have succeeded in initialising before!
1615  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1616  * currently not initialised.
1617  */
1618 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1619 {
1620         unsigned long flags;
1621
1622         cpufreq_debug_disable_ratelimit();
1623
1624         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1625                 cpufreq_debug_enable_ratelimit();
1626                 return -EINVAL;
1627         }
1628
1629         dprintk("unregistering driver %s\n", driver->name);
1630
1631         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1632         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1633
1634         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1635         cpufreq_driver = NULL;
1636         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1637
1638         return 0;
1639 }
1640 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);