SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/cpufreq/cpufreq.c

   1 /*
   2  *  linux/drivers/cpufreq/cpufreq.c
   3  *
   4  *  Copyright (C) 2001 Russell King
   5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
   6  *
   7  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
   8  *      Added handling for CPU hotplug
   9  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
  10  *      Fix handling for CPU hotplug -- affected CPUs
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17
  18 #include <linux/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/notifier.h>
  22 #include <linux/cpufreq.h>
  23 #include <linux/delay.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/device.h>
  27 #include <linux/slab.h>
  28 #include <linux/cpu.h>
  29 #include <linux/completion.h>
  30 #include <linux/mutex.h>
  31
  32 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
  33                                                 "cpufreq-core", msg)
  34
  35 /**
  36  * The "cpufreq driver" - the arch- or hardware-dependent low
  37  * level driver of CPUFreq support, and its spinlock. This lock
  38  * also protects the cpufreq_cpu_data array.
  39  */
  40 static struct cpufreq_driver *cpufreq_driver;
  41 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
  42 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(struct work_struct *work);
  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 struct srcu_notifier_head cpufreq_transition_notifier_list;
  57
  58 static int __init init_cpufreq_transition_notifier_list(void)
  59 {
  60         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
  61         return 0;
  62 }
  63 pure_initcall(init_cpufreq_transition_notifier_list);
  64
  65 static LIST_HEAD(cpufreq_governor_list);
  66 static DEFINE_MUTEX (cpufreq_governor_mutex);
  67
  68 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
  69 {
  70         struct cpufreq_policy *data;
  71         unsigned long flags;
  72
  73         if (cpu >= NR_CPUS)
  74                 goto err_out;
  75
  76         /* get the cpufreq driver */
  77         spin_lock_irqsave(&cpufreq_driver_lock, flags);
  78
  79         if (!cpufreq_driver)
  80                 goto err_out_unlock;
  81
  82         if (!try_module_get(cpufreq_driver->owner))
  83                 goto err_out_unlock;
  84
  85
  86         /* get the CPU */
  87         data = cpufreq_cpu_data[cpu];
  88
  89         if (!data)
  90                 goto err_out_put_module;
  91
  92         if (!kobject_get(&data->kobj))
  93                 goto err_out_put_module;
  94
  95         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  96         return data;
  97
  98 err_out_put_module:
  99         module_put(cpufreq_driver->owner);
 100 err_out_unlock:
 101         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 102 err_out:
 103         return NULL;
 104 }
 105 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 106
 107
 108 void cpufreq_cpu_put(struct cpufreq_policy *data)
 109 {
 110         kobject_put(&data->kobj);
 111         module_put(cpufreq_driver->owner);
 112 }
 113 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 114
 115
 116 /*********************************************************************
 117  *                     UNIFIED DEBUG HELPERS                         *
 118  *********************************************************************/
 119 #ifdef CONFIG_CPU_FREQ_DEBUG
 120
 121 /* what part(s) of the CPUfreq subsystem are debugged? */
 122 static unsigned int debug;
 123
 124 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 125  * set or modify this value.
 126  */
 127 static unsigned int debug_ratelimit = 1;
 128
 129 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 130  * loading of a cpufreq driver, temporarily disabled when a new policy
 131  * is set, and disabled upon cpufreq driver removal
 132  */
 133 static unsigned int disable_ratelimit = 1;
 134 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 135
 136 static void cpufreq_debug_enable_ratelimit(void)
 137 {
 138         unsigned long flags;
 139
 140         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 141         if (disable_ratelimit)
 142                 disable_ratelimit--;
 143         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 144 }
 145
 146 static void cpufreq_debug_disable_ratelimit(void)
 147 {
 148         unsigned long flags;
 149
 150         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 151         disable_ratelimit++;
 152         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 153 }
 154
 155 void cpufreq_debug_printk(unsigned int type, const char *prefix,
 156                                                         const char *fmt, ...)
 157 {
 158         char s[256];
 159         va_list args;
 160         unsigned int len;
 161         unsigned long flags;
 162
 163         WARN_ON(!prefix);
 164         if (type & debug) {
 165                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 166                 if (!disable_ratelimit && debug_ratelimit
 167                                         && !printk_ratelimit()) {
 168                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 169                         return;
 170                 }
 171                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 172
 173                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 174
 175                 va_start(args, fmt);
 176                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 177                 va_end(args);
 178
 179                 printk(s);
 180
 181                 WARN_ON(len < 5);
 182         }
 183 }
 184 EXPORT_SYMBOL(cpufreq_debug_printk);
 185
 186
 187 module_param(debug, uint, 0644);
 188 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core,"
 189                         " 2 to debug drivers, and 4 to debug governors.");
 190
 191 module_param(debug_ratelimit, uint, 0644);
 192 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging:"
 193                                         " set to 0 to disable ratelimiting.");
 194
 195 #else /* !CONFIG_CPU_FREQ_DEBUG */
 196
 197 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 198 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 199
 200 #endif /* CONFIG_CPU_FREQ_DEBUG */
 201
 202
 203 /*********************************************************************
 204  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 205  *********************************************************************/
 206
 207 /**
 208  * adjust_jiffies - adjust the system "loops_per_jiffy"
 209  *
 210  * This function alters the system "loops_per_jiffy" for the clock
 211  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 212  * systems as each CPU might be scaled differently. So, use the arch
 213  * per-CPU loops_per_jiffy value wherever possible.
 214  */
 215 #ifndef CONFIG_SMP
 216 static unsigned long l_p_j_ref;
 217 static unsigned int  l_p_j_ref_freq;
 218
 219 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 220 {
 221         if (ci->flags & CPUFREQ_CONST_LOOPS)
 222                 return;
 223
 224         if (!l_p_j_ref_freq) {
 225                 l_p_j_ref = loops_per_jiffy;
 226                 l_p_j_ref_freq = ci->old;
 227                 dprintk("saving %lu as reference value for loops_per_jiffy;"
 228                         "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 229         }
 230         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 231             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 232             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 233                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
 234                                                                 ci->new);
 235                 dprintk("scaling loops_per_jiffy to %lu"
 236                         "for frequency %u kHz\n", loops_per_jiffy, ci->new);
 237         }
 238 }
 239 #else
 240 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 241 {
 242         return;
 243 }
 244 #endif
 245
 246
 247 /**
 248  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 249  * on frequency transition.
 250  *
 251  * This function calls the transition notifiers and the "adjust_jiffies"
 252  * function. It is called twice on all CPU frequency changes that have
 253  * external effects.
 254  */
 255 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 256 {
 257         struct cpufreq_policy *policy;
 258
 259         BUG_ON(irqs_disabled());
 260
 261         freqs->flags = cpufreq_driver->flags;
 262         dprintk("notification %u of frequency transition to %u kHz\n",
 263                 state, freqs->new);
 264
 265         policy = cpufreq_cpu_data[freqs->cpu];
 266         switch (state) {
 267
 268         case CPUFREQ_PRECHANGE:
 269                 /* detect if the driver reported a value as "old frequency"
 270                  * which is not equal to what the cpufreq core thinks is
 271                  * "old frequency".
 272                  */
 273                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 274                         if ((policy) && (policy->cpu == freqs->cpu) &&
 275                             (policy->cur) && (policy->cur != freqs->old)) {
 276                                 dprintk("Warning: CPU frequency is"
 277                                         " %u, cpufreq assumed %u kHz.\n",
 278                                         freqs->old, policy->cur);
 279                                 freqs->old = policy->cur;
 280                         }
 281                 }
 282                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 283                                 CPUFREQ_PRECHANGE, freqs);
 284                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 285                 break;
 286
 287         case CPUFREQ_POSTCHANGE:
 288                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 289                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 290                                 CPUFREQ_POSTCHANGE, freqs);
 291                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 292                         policy->cur = freqs->new;
 293                 break;
 294         }
 295 }
 296 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 297
 298
 299
 300 /*********************************************************************
 301  *                          SYSFS INTERFACE                          *
 302  *********************************************************************/
 303
 304 static struct cpufreq_governor *__find_governor(const char *str_governor)
 305 {
 306         struct cpufreq_governor *t;
 307
 308         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
 309                 if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN))
 310                         return t;
 311
 312         return NULL;
 313 }
 314
 315 /**
 316  * cpufreq_parse_governor - parse a governor string
 317  */
 318 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 319                                 struct cpufreq_governor **governor)
 320 {
 321         int err = -EINVAL;
 322
 323         if (!cpufreq_driver)
 324                 goto out;
 325
 326         if (cpufreq_driver->setpolicy) {
 327                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 328                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 329                         err = 0;
 330                 } else if (!strnicmp(str_governor, "powersave",
 331                                                 CPUFREQ_NAME_LEN)) {
 332                         *policy = CPUFREQ_POLICY_POWERSAVE;
 333                         err = 0;
 334                 }
 335         } else if (cpufreq_driver->target) {
 336                 struct cpufreq_governor *t;
 337
 338                 mutex_lock(&cpufreq_governor_mutex);
 339
 340                 t = __find_governor(str_governor);
 341
 342                 if (t == NULL) {
 343                         char *name = kasprintf(GFP_KERNEL, "cpufreq_%s",
 344                                                                 str_governor);
 345
 346                         if (name) {
 347                                 int ret;
 348
 349                                 mutex_unlock(&cpufreq_governor_mutex);
 350                                 ret = request_module(name);
 351                                 mutex_lock(&cpufreq_governor_mutex);
 352
 353                                 if (ret == 0)
 354                                         t = __find_governor(str_governor);
 355                         }
 356
 357                         kfree(name);
 358                 }
 359
 360                 if (t != NULL) {
 361                         *governor = t;
 362                         err = 0;
 363                 }
 364
 365                 mutex_unlock(&cpufreq_governor_mutex);
 366         }
 367   out:
 368         return err;
 369 }
 370
 371
 372 /* drivers/base/cpu.c */
 373 extern struct sysdev_class cpu_sysdev_class;
 374
 375
 376 /**
 377  * cpufreq_per_cpu_attr_read() / show_##file_name() -
 378  * print out cpufreq information
 379  *
 380  * Write out information from cpufreq_driver->policy[cpu]; object must be
 381  * "unsigned int".
 382  */
 383
 384 #define show_one(file_name, object)                     \
 385 static ssize_t show_##file_name                         \
 386 (struct cpufreq_policy * policy, char *buf)             \
 387 {                                                       \
 388         return sprintf (buf, "%u\n", policy->object);   \
 389 }
 390
 391 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 392 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 393 show_one(scaling_min_freq, min);
 394 show_one(scaling_max_freq, max);
 395 show_one(scaling_cur_freq, cur);
 396
 397 static int __cpufreq_set_policy(struct cpufreq_policy *data,
 398                                 struct cpufreq_policy *policy);
 399
 400 /**
 401  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 402  */
 403 #define store_one(file_name, object)                    \
 404 static ssize_t store_##file_name                                        \
 405 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 406 {                                                                       \
 407         unsigned int ret = -EINVAL;                                     \
 408         struct cpufreq_policy new_policy;                               \
 409                                                                         \
 410         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 411         if (ret)                                                        \
 412                 return -EINVAL;                                         \
 413                                                                         \
 414         ret = sscanf (buf, "%u", &new_policy.object);                   \
 415         if (ret != 1)                                                   \
 416                 return -EINVAL;                                         \
 417                                                                         \
 418         mutex_lock(&policy->lock);                                      \
 419         ret = __cpufreq_set_policy(policy, &new_policy);                \
 420         policy->user_policy.object = policy->object;                    \
 421         mutex_unlock(&policy->lock);                                    \
 422                                                                         \
 423         return ret ? ret : count;                                       \
 424 }
 425
 426 store_one(scaling_min_freq,min);
 427 store_one(scaling_max_freq,max);
 428
 429 /**
 430  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 431  */
 432 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy,
 433                                                         char *buf)
 434 {
 435         unsigned int cur_freq = cpufreq_get(policy->cpu);
 436         if (!cur_freq)
 437                 return sprintf(buf, "<unknown>");
 438         return sprintf(buf, "%u\n", cur_freq);
 439 }
 440
 441
 442 /**
 443  * show_scaling_governor - show the current policy for the specified CPU
 444  */
 445 static ssize_t show_scaling_governor (struct cpufreq_policy * policy,
 446                                                         char *buf)
 447 {
 448         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 449                 return sprintf(buf, "powersave\n");
 450         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 451                 return sprintf(buf, "performance\n");
 452         else if (policy->governor)
 453                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 454         return -EINVAL;
 455 }
 456
 457
 458 /**
 459  * store_scaling_governor - store policy for the specified CPU
 460  */
 461 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 462                                        const char *buf, size_t count)
 463 {
 464         unsigned int ret = -EINVAL;
 465         char    str_governor[16];
 466         struct cpufreq_policy new_policy;
 467
 468         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 469         if (ret)
 470                 return ret;
 471
 472         ret = sscanf (buf, "%15s", str_governor);
 473         if (ret != 1)
 474                 return -EINVAL;
 475
 476         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
 477                                                 &new_policy.governor))
 478                 return -EINVAL;
 479
 480         /* Do not use cpufreq_set_policy here or the user_policy.max
 481            will be wrongly overridden */
 482         mutex_lock(&policy->lock);
 483         ret = __cpufreq_set_policy(policy, &new_policy);
 484
 485         policy->user_policy.policy = policy->policy;
 486         policy->user_policy.governor = policy->governor;
 487         mutex_unlock(&policy->lock);
 488
 489         if (ret)
 490                 return ret;
 491         else
 492                 return count;
 493 }
 494
 495 /**
 496  * show_scaling_driver - show the cpufreq driver currently loaded
 497  */
 498 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 499 {
 500         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 501 }
 502
 503 /**
 504  * show_scaling_available_governors - show the available CPUfreq governors
 505  */
 506 static ssize_t show_scaling_available_governors (struct cpufreq_policy *policy,
 507                                 char *buf)
 508 {
 509         ssize_t i = 0;
 510         struct cpufreq_governor *t;
 511
 512         if (!cpufreq_driver->target) {
 513                 i += sprintf(buf, "performance powersave");
 514                 goto out;
 515         }
 516
 517         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 518                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 519                         goto out;
 520                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 521         }
 522 out:
 523         i += sprintf(&buf[i], "\n");
 524         return i;
 525 }
 526 /**
 527  * show_affected_cpus - show the CPUs affected by each transition
 528  */
 529 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 530 {
 531         ssize_t i = 0;
 532         unsigned int cpu;
 533
 534         for_each_cpu_mask(cpu, policy->cpus) {
 535                 if (i)
 536                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 537                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 538                 if (i >= (PAGE_SIZE - 5))
 539                     break;
 540         }
 541         i += sprintf(&buf[i], "\n");
 542         return i;
 543 }
 544
 545
 546 #define define_one_ro(_name) \
 547 static struct freq_attr _name = \
 548 __ATTR(_name, 0444, show_##_name, NULL)
 549
 550 #define define_one_ro0400(_name) \
 551 static struct freq_attr _name = \
 552 __ATTR(_name, 0400, show_##_name, NULL)
 553
 554 #define define_one_rw(_name) \
 555 static struct freq_attr _name = \
 556 __ATTR(_name, 0644, show_##_name, store_##_name)
 557
 558 define_one_ro0400(cpuinfo_cur_freq);
 559 define_one_ro(cpuinfo_min_freq);
 560 define_one_ro(cpuinfo_max_freq);
 561 define_one_ro(scaling_available_governors);
 562 define_one_ro(scaling_driver);
 563 define_one_ro(scaling_cur_freq);
 564 define_one_ro(affected_cpus);
 565 define_one_rw(scaling_min_freq);
 566 define_one_rw(scaling_max_freq);
 567 define_one_rw(scaling_governor);
 568
 569 static struct attribute * default_attrs[] = {
 570         &cpuinfo_min_freq.attr,
 571         &cpuinfo_max_freq.attr,
 572         &scaling_min_freq.attr,
 573         &scaling_max_freq.attr,
 574         &affected_cpus.attr,
 575         &scaling_governor.attr,
 576         &scaling_driver.attr,
 577         &scaling_available_governors.attr,
 578         NULL
 579 };
 580
 581 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 582 #define to_attr(a) container_of(a,struct freq_attr,attr)
 583
 584 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 585 {
 586         struct cpufreq_policy * policy = to_policy(kobj);
 587         struct freq_attr * fattr = to_attr(attr);
 588         ssize_t ret;
 589         policy = cpufreq_cpu_get(policy->cpu);
 590         if (!policy)
 591                 return -EINVAL;
 592         if (fattr->show)
 593                 ret = fattr->show(policy, buf);
 594         else
 595                 ret = -EIO;
 596
 597         cpufreq_cpu_put(policy);
 598         return ret;
 599 }
 600
 601 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 602                      const char * buf, size_t count)
 603 {
 604         struct cpufreq_policy * policy = to_policy(kobj);
 605         struct freq_attr * fattr = to_attr(attr);
 606         ssize_t ret;
 607         policy = cpufreq_cpu_get(policy->cpu);
 608         if (!policy)
 609                 return -EINVAL;
 610         if (fattr->store)
 611                 ret = fattr->store(policy, buf, count);
 612         else
 613                 ret = -EIO;
 614
 615         cpufreq_cpu_put(policy);
 616         return ret;
 617 }
 618
 619 static void cpufreq_sysfs_release(struct kobject * kobj)
 620 {
 621         struct cpufreq_policy * policy = to_policy(kobj);
 622         dprintk("last reference is dropped\n");
 623         complete(&policy->kobj_unregister);
 624 }
 625
 626 static struct sysfs_ops sysfs_ops = {
 627         .show   = show,
 628         .store  = store,
 629 };
 630
 631 static struct kobj_type ktype_cpufreq = {
 632         .sysfs_ops      = &sysfs_ops,
 633         .default_attrs  = default_attrs,
 634         .release        = cpufreq_sysfs_release,
 635 };
 636
 637
 638 /**
 639  * cpufreq_add_dev - add a CPU device
 640  *
 641  * Adds the cpufreq interface for a CPU device.
 642  */
 643 static int cpufreq_add_dev (struct sys_device * sys_dev)
 644 {
 645         unsigned int cpu = sys_dev->id;
 646         int ret = 0;
 647         struct cpufreq_policy new_policy;
 648         struct cpufreq_policy *policy;
 649         struct freq_attr **drv_attr;
 650         struct sys_device *cpu_sys_dev;
 651         unsigned long flags;
 652         unsigned int j;
 653 #ifdef CONFIG_SMP
 654         struct cpufreq_policy *managed_policy;
 655 #endif
 656
 657         if (cpu_is_offline(cpu))
 658                 return 0;
 659
 660         cpufreq_debug_disable_ratelimit();
 661         dprintk("adding CPU %u\n", cpu);
 662
 663 #ifdef CONFIG_SMP
 664         /* check whether a different CPU already registered this
 665          * CPU because it is in the same boat. */
 666         policy = cpufreq_cpu_get(cpu);
 667         if (unlikely(policy)) {
 668                 cpufreq_cpu_put(policy);
 669                 cpufreq_debug_enable_ratelimit();
 670                 return 0;
 671         }
 672 #endif
 673
 674         if (!try_module_get(cpufreq_driver->owner)) {
 675                 ret = -EINVAL;
 676                 goto module_out;
 677         }
 678
 679         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 680         if (!policy) {
 681                 ret = -ENOMEM;
 682                 goto nomem_out;
 683         }
 684
 685         policy->cpu = cpu;
 686         policy->cpus = cpumask_of_cpu(cpu);
 687
 688         mutex_init(&policy->lock);
 689         mutex_lock(&policy->lock);
 690         init_completion(&policy->kobj_unregister);
 691         INIT_WORK(&policy->update, handle_update);
 692
 693         /* call driver. From then on the cpufreq must be able
 694          * to accept all calls to ->verify and ->setpolicy for this CPU
 695          */
 696         ret = cpufreq_driver->init(policy);
 697         if (ret) {
 698                 dprintk("initialization failed\n");
 699                 mutex_unlock(&policy->lock);
 700                 goto err_out;
 701         }
 702
 703 #ifdef CONFIG_SMP
 704         for_each_cpu_mask(j, policy->cpus) {
 705                 if (cpu == j)
 706                         continue;
 707
 708                 /* check for existing affected CPUs.  They may not be aware
 709                  * of it due to CPU Hotplug.
 710                  */
 711                 managed_policy = cpufreq_cpu_get(j);
 712                 if (unlikely(managed_policy)) {
 713                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 714                         managed_policy->cpus = policy->cpus;
 715                         cpufreq_cpu_data[cpu] = managed_policy;
 716                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 717
 718                         dprintk("CPU already managed, adding link\n");
 719                         ret = sysfs_create_link(&sys_dev->kobj,
 720                                                 &managed_policy->kobj,
 721                                                 "cpufreq");
 722                         if (ret) {
 723                                 mutex_unlock(&policy->lock);
 724                                 goto err_out_driver_exit;
 725                         }
 726
 727                         cpufreq_debug_enable_ratelimit();
 728                         mutex_unlock(&policy->lock);
 729                         ret = 0;
 730                         goto err_out_driver_exit; /* call driver->exit() */
 731                 }
 732         }
 733 #endif
 734         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 735
 736         /* prepare interface data */
 737         policy->kobj.parent = &sys_dev->kobj;
 738         policy->kobj.ktype = &ktype_cpufreq;
 739         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 740
 741         ret = kobject_register(&policy->kobj);
 742         if (ret) {
 743                 mutex_unlock(&policy->lock);
 744                 goto err_out_driver_exit;
 745         }
 746         /* set up files for this cpu device */
 747         drv_attr = cpufreq_driver->attr;
 748         while ((drv_attr) && (*drv_attr)) {
 749                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 750                 drv_attr++;
 751         }
 752         if (cpufreq_driver->get)
 753                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 754         if (cpufreq_driver->target)
 755                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 756
 757         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 758         for_each_cpu_mask(j, policy->cpus)
 759                 cpufreq_cpu_data[j] = policy;
 760         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 761
 762         /* symlink affected CPUs */
 763         for_each_cpu_mask(j, policy->cpus) {
 764                 if (j == cpu)
 765                         continue;
 766                 if (!cpu_online(j))
 767                         continue;
 768
 769                 dprintk("CPU %u already managed, adding link\n", j);
 770                 cpufreq_cpu_get(cpu);
 771                 cpu_sys_dev = get_cpu_sysdev(j);
 772                 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 773                                         "cpufreq");
 774                 if (ret) {
 775                         mutex_unlock(&policy->lock);
 776                         goto err_out_unregister;
 777                 }
 778         }
 779
 780         policy->governor = NULL; /* to assure that the starting sequence is
 781                                   * run in cpufreq_set_policy */
 782         mutex_unlock(&policy->lock);
 783
 784         /* set default policy */
 785         ret = cpufreq_set_policy(&new_policy);
 786         if (ret) {
 787                 dprintk("setting policy failed\n");
 788                 goto err_out_unregister;
 789         }
 790
 791         module_put(cpufreq_driver->owner);
 792         dprintk("initialization complete\n");
 793         cpufreq_debug_enable_ratelimit();
 794
 795         return 0;
 796
 797
 798 err_out_unregister:
 799         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 800         for_each_cpu_mask(j, policy->cpus)
 801                 cpufreq_cpu_data[j] = NULL;
 802         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 803
 804         kobject_unregister(&policy->kobj);
 805         wait_for_completion(&policy->kobj_unregister);
 806
 807 err_out_driver_exit:
 808         if (cpufreq_driver->exit)
 809                 cpufreq_driver->exit(policy);
 810
 811 err_out:
 812         kfree(policy);
 813
 814 nomem_out:
 815         module_put(cpufreq_driver->owner);
 816 module_out:
 817         cpufreq_debug_enable_ratelimit();
 818         return ret;
 819 }
 820
 821
 822 /**
 823  * cpufreq_remove_dev - remove a CPU device
 824  *
 825  * Removes the cpufreq interface for a CPU device.
 826  */
 827 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 828 {
 829         unsigned int cpu = sys_dev->id;
 830         unsigned long flags;
 831         struct cpufreq_policy *data;
 832 #ifdef CONFIG_SMP
 833         struct sys_device *cpu_sys_dev;
 834         unsigned int j;
 835 #endif
 836
 837         cpufreq_debug_disable_ratelimit();
 838         dprintk("unregistering CPU %u\n", cpu);
 839
 840         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 841         data = cpufreq_cpu_data[cpu];
 842
 843         if (!data) {
 844                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 845                 cpufreq_debug_enable_ratelimit();
 846                 return -EINVAL;
 847         }
 848         cpufreq_cpu_data[cpu] = NULL;
 849
 850
 851 #ifdef CONFIG_SMP
 852         /* if this isn't the CPU which is the parent of the kobj, we
 853          * only need to unlink, put and exit
 854          */
 855         if (unlikely(cpu != data->cpu)) {
 856                 dprintk("removing link\n");
 857                 cpu_clear(cpu, data->cpus);
 858                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 859                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 860                 cpufreq_cpu_put(data);
 861                 cpufreq_debug_enable_ratelimit();
 862                 return 0;
 863         }
 864 #endif
 865
 866
 867         if (!kobject_get(&data->kobj)) {
 868                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 869                 cpufreq_debug_enable_ratelimit();
 870                 return -EFAULT;
 871         }
 872
 873 #ifdef CONFIG_SMP
 874         /* if we have other CPUs still registered, we need to unlink them,
 875          * or else wait_for_completion below will lock up. Clean the
 876          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 877          * links afterwards.
 878          */
 879         if (unlikely(cpus_weight(data->cpus) > 1)) {
 880                 for_each_cpu_mask(j, data->cpus) {
 881                         if (j == cpu)
 882                                 continue;
 883                         cpufreq_cpu_data[j] = NULL;
 884                 }
 885         }
 886
 887         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 888
 889         if (unlikely(cpus_weight(data->cpus) > 1)) {
 890                 for_each_cpu_mask(j, data->cpus) {
 891                         if (j == cpu)
 892                                 continue;
 893                         dprintk("removing link for cpu %u\n", j);
 894                         cpu_sys_dev = get_cpu_sysdev(j);
 895                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 896                         cpufreq_cpu_put(data);
 897                 }
 898         }
 899 #else
 900         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 901 #endif
 902
 903         mutex_lock(&data->lock);
 904         if (cpufreq_driver->target)
 905                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 906         mutex_unlock(&data->lock);
 907
 908         kobject_unregister(&data->kobj);
 909
 910         kobject_put(&data->kobj);
 911
 912         /* we need to make sure that the underlying kobj is actually
 913          * not referenced anymore by anybody before we proceed with
 914          * unloading.
 915          */
 916         dprintk("waiting for dropping of refcount\n");
 917         wait_for_completion(&data->kobj_unregister);
 918         dprintk("wait complete\n");
 919
 920         if (cpufreq_driver->exit)
 921                 cpufreq_driver->exit(data);
 922
 923         kfree(data);
 924
 925         cpufreq_debug_enable_ratelimit();
 926         return 0;
 927 }
 928
 929
 930 static void handle_update(struct work_struct *work)
 931 {
 932         struct cpufreq_policy *policy =
 933                 container_of(work, struct cpufreq_policy, update);
 934         unsigned int cpu = policy->cpu;
 935         dprintk("handle_update for cpu %u called\n", cpu);
 936         cpufreq_update_policy(cpu);
 937 }
 938
 939 /**
 940  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 941  *      @cpu: cpu number
 942  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 943  *      @new_freq: CPU frequency the CPU actually runs at
 944  *
 945  *      We adjust to current frequency first, and need to clean up later. So either call
 946  *      to cpufreq_update_policy() or schedule handle_update()).
 947  */
 948 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
 949                                 unsigned int new_freq)
 950 {
 951         struct cpufreq_freqs freqs;
 952
 953         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
 954                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 955
 956         freqs.cpu = cpu;
 957         freqs.old = old_freq;
 958         freqs.new = new_freq;
 959         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 960         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 961 }
 962
 963
 964 /**
 965  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
 966  * @cpu: CPU number
 967  *
 968  * This is the last known freq, without actually getting it from the driver.
 969  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 970  */
 971 unsigned int cpufreq_quick_get(unsigned int cpu)
 972 {
 973         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 974         unsigned int ret_freq = 0;
 975
 976         if (policy) {
 977                 mutex_lock(&policy->lock);
 978                 ret_freq = policy->cur;
 979                 mutex_unlock(&policy->lock);
 980                 cpufreq_cpu_put(policy);
 981         }
 982
 983         return (ret_freq);
 984 }
 985 EXPORT_SYMBOL(cpufreq_quick_get);
 986
 987
 988 /**
 989  * cpufreq_get - get the current CPU frequency (in kHz)
 990  * @cpu: CPU number
 991  *
 992  * Get the CPU current (static) CPU frequency
 993  */
 994 unsigned int cpufreq_get(unsigned int cpu)
 995 {
 996         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 997         unsigned int ret_freq = 0;
 998
 999         if (!policy)
1000                 return 0;
1001
1002         if (!cpufreq_driver->get)
1003                 goto out;
1004
1005         mutex_lock(&policy->lock);
1006
1007         ret_freq = cpufreq_driver->get(cpu);
1008
1009         if (ret_freq && policy->cur &&
1010                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1011                 /* verify no discrepancy between actual and
1012                                         saved value exists */
1013                 if (unlikely(ret_freq != policy->cur)) {
1014                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1015                         schedule_work(&policy->update);
1016                 }
1017         }
1018
1019         mutex_unlock(&policy->lock);
1020
1021 out:
1022         cpufreq_cpu_put(policy);
1023
1024         return (ret_freq);
1025 }
1026 EXPORT_SYMBOL(cpufreq_get);
1027
1028
1029 /**
1030  *      cpufreq_suspend - let the low level driver prepare for suspend
1031  */
1032
1033 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
1034 {
1035         int cpu = sysdev->id;
1036         int ret = 0;
1037         unsigned int cur_freq = 0;
1038         struct cpufreq_policy *cpu_policy;
1039
1040         dprintk("suspending cpu %u\n", cpu);
1041
1042         if (!cpu_online(cpu))
1043                 return 0;
1044
1045         /* we may be lax here as interrupts are off. Nonetheless
1046          * we need to grab the correct cpu policy, as to check
1047          * whether we really run on this CPU.
1048          */
1049
1050         cpu_policy = cpufreq_cpu_get(cpu);
1051         if (!cpu_policy)
1052                 return -EINVAL;
1053
1054         /* only handle each CPU group once */
1055         if (unlikely(cpu_policy->cpu != cpu)) {
1056                 cpufreq_cpu_put(cpu_policy);
1057                 return 0;
1058         }
1059
1060         if (cpufreq_driver->suspend) {
1061                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1062                 if (ret) {
1063                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1064                                         "step on CPU %u\n", cpu_policy->cpu);
1065                         cpufreq_cpu_put(cpu_policy);
1066                         return ret;
1067                 }
1068         }
1069
1070
1071         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1072                 goto out;
1073
1074         if (cpufreq_driver->get)
1075                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1076
1077         if (!cur_freq || !cpu_policy->cur) {
1078                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1079                        "frequency is what timing core thinks it is.\n");
1080                 goto out;
1081         }
1082
1083         if (unlikely(cur_freq != cpu_policy->cur)) {
1084                 struct cpufreq_freqs freqs;
1085
1086                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1087                         dprintk("Warning: CPU frequency is %u, "
1088                                "cpufreq assumed %u kHz.\n",
1089                                cur_freq, cpu_policy->cur);
1090
1091                 freqs.cpu = cpu;
1092                 freqs.old = cpu_policy->cur;
1093                 freqs.new = cur_freq;
1094
1095                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
1096                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1097                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1098
1099                 cpu_policy->cur = cur_freq;
1100         }
1101
1102 out:
1103         cpufreq_cpu_put(cpu_policy);
1104         return 0;
1105 }
1106
1107 /**
1108  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1109  *
1110  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1111  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1112  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1113  *          restored.
1114  */
1115 static int cpufreq_resume(struct sys_device * sysdev)
1116 {
1117         int cpu = sysdev->id;
1118         int ret = 0;
1119         struct cpufreq_policy *cpu_policy;
1120
1121         dprintk("resuming cpu %u\n", cpu);
1122
1123         if (!cpu_online(cpu))
1124                 return 0;
1125
1126         /* we may be lax here as interrupts are off. Nonetheless
1127          * we need to grab the correct cpu policy, as to check
1128          * whether we really run on this CPU.
1129          */
1130
1131         cpu_policy = cpufreq_cpu_get(cpu);
1132         if (!cpu_policy)
1133                 return -EINVAL;
1134
1135         /* only handle each CPU group once */
1136         if (unlikely(cpu_policy->cpu != cpu)) {
1137                 cpufreq_cpu_put(cpu_policy);
1138                 return 0;
1139         }
1140
1141         if (cpufreq_driver->resume) {
1142                 ret = cpufreq_driver->resume(cpu_policy);
1143                 if (ret) {
1144                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1145                                         "step on CPU %u\n", cpu_policy->cpu);
1146                         cpufreq_cpu_put(cpu_policy);
1147                         return ret;
1148                 }
1149         }
1150
1151         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1152                 unsigned int cur_freq = 0;
1153
1154                 if (cpufreq_driver->get)
1155                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1156
1157                 if (!cur_freq || !cpu_policy->cur) {
1158                         printk(KERN_ERR "cpufreq: resume failed to assert "
1159                                         "current frequency is what timing core "
1160                                         "thinks it is.\n");
1161                         goto out;
1162                 }
1163
1164                 if (unlikely(cur_freq != cpu_policy->cur)) {
1165                         struct cpufreq_freqs freqs;
1166
1167                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1168                                 dprintk("Warning: CPU frequency"
1169                                        "is %u, cpufreq assumed %u kHz.\n",
1170                                        cur_freq, cpu_policy->cur);
1171
1172                         freqs.cpu = cpu;
1173                         freqs.old = cpu_policy->cur;
1174                         freqs.new = cur_freq;
1175
1176                         srcu_notifier_call_chain(
1177                                         &cpufreq_transition_notifier_list,
1178                                         CPUFREQ_RESUMECHANGE, &freqs);
1179                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1180
1181                         cpu_policy->cur = cur_freq;
1182                 }
1183         }
1184
1185 out:
1186         schedule_work(&cpu_policy->update);
1187         cpufreq_cpu_put(cpu_policy);
1188         return ret;
1189 }
1190
1191 static struct sysdev_driver cpufreq_sysdev_driver = {
1192         .add            = cpufreq_add_dev,
1193         .remove         = cpufreq_remove_dev,
1194         .suspend        = cpufreq_suspend,
1195         .resume         = cpufreq_resume,
1196 };
1197
1198
1199 /*********************************************************************
1200  *                     NOTIFIER LISTS INTERFACE                      *
1201  *********************************************************************/
1202
1203 /**
1204  *      cpufreq_register_notifier - register a driver with cpufreq
1205  *      @nb: notifier function to register
1206  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1207  *
1208  *      Add a driver to one of two lists: either a list of drivers that
1209  *      are notified about clock rate changes (once before and once after
1210  *      the transition), or a list of drivers that are notified about
1211  *      changes in cpufreq policy.
1212  *
1213  *      This function may sleep, and has the same return conditions as
1214  *      blocking_notifier_chain_register.
1215  */
1216 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1217 {
1218         int ret;
1219
1220         switch (list) {
1221         case CPUFREQ_TRANSITION_NOTIFIER:
1222                 ret = srcu_notifier_chain_register(
1223                                 &cpufreq_transition_notifier_list, nb);
1224                 break;
1225         case CPUFREQ_POLICY_NOTIFIER:
1226                 ret = blocking_notifier_chain_register(
1227                                 &cpufreq_policy_notifier_list, nb);
1228                 break;
1229         default:
1230                 ret = -EINVAL;
1231         }
1232
1233         return ret;
1234 }
1235 EXPORT_SYMBOL(cpufreq_register_notifier);
1236
1237
1238 /**
1239  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1240  *      @nb: notifier block to be unregistered
1241  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1242  *
1243  *      Remove a driver from the CPU frequency notifier list.
1244  *
1245  *      This function may sleep, and has the same return conditions as
1246  *      blocking_notifier_chain_unregister.
1247  */
1248 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1249 {
1250         int ret;
1251
1252         switch (list) {
1253         case CPUFREQ_TRANSITION_NOTIFIER:
1254                 ret = srcu_notifier_chain_unregister(
1255                                 &cpufreq_transition_notifier_list, nb);
1256                 break;
1257         case CPUFREQ_POLICY_NOTIFIER:
1258                 ret = blocking_notifier_chain_unregister(
1259                                 &cpufreq_policy_notifier_list, nb);
1260                 break;
1261         default:
1262                 ret = -EINVAL;
1263         }
1264
1265         return ret;
1266 }
1267 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1268
1269
1270 /*********************************************************************
1271  *                              GOVERNORS                            *
1272  *********************************************************************/
1273
1274
1275 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1276                             unsigned int target_freq,
1277                             unsigned int relation)
1278 {
1279         int retval = -EINVAL;
1280
1281         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1282                 target_freq, relation);
1283         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1284                 retval = cpufreq_driver->target(policy, target_freq, relation);
1285
1286         return retval;
1287 }
1288 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1289
1290 int cpufreq_driver_target(struct cpufreq_policy *policy,
1291                           unsigned int target_freq,
1292                           unsigned int relation)
1293 {
1294         int ret;
1295
1296         policy = cpufreq_cpu_get(policy->cpu);
1297         if (!policy)
1298                 return -EINVAL;
1299
1300         mutex_lock(&policy->lock);
1301
1302         ret = __cpufreq_driver_target(policy, target_freq, relation);
1303
1304         mutex_unlock(&policy->lock);
1305
1306         cpufreq_cpu_put(policy);
1307         return ret;
1308 }
1309 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1310
1311 int cpufreq_driver_getavg(struct cpufreq_policy *policy)
1312 {
1313         int ret = 0;
1314
1315         policy = cpufreq_cpu_get(policy->cpu);
1316         if (!policy)
1317                 return -EINVAL;
1318
1319         mutex_lock(&policy->lock);
1320
1321         if (cpu_online(policy->cpu) && cpufreq_driver->getavg)
1322                 ret = cpufreq_driver->getavg(policy->cpu);
1323
1324         mutex_unlock(&policy->lock);
1325
1326         cpufreq_cpu_put(policy);
1327         return ret;
1328 }
1329 EXPORT_SYMBOL_GPL(cpufreq_driver_getavg);
1330
1331 /*
1332  * when "event" is CPUFREQ_GOV_LIMITS
1333  */
1334
1335 static int __cpufreq_governor(struct cpufreq_policy *policy,
1336                                         unsigned int event)
1337 {
1338         int ret;
1339
1340         if (!try_module_get(policy->governor->owner))
1341                 return -EINVAL;
1342
1343         dprintk("__cpufreq_governor for CPU %u, event %u\n",
1344                                                 policy->cpu, event);
1345         ret = policy->governor->governor(policy, event);
1346
1347         /* we keep one module reference alive for
1348                         each CPU governed by this CPU */
1349         if ((event != CPUFREQ_GOV_START) || ret)
1350                 module_put(policy->governor->owner);
1351         if ((event == CPUFREQ_GOV_STOP) && !ret)
1352                 module_put(policy->governor->owner);
1353
1354         return ret;
1355 }
1356
1357
1358 int cpufreq_register_governor(struct cpufreq_governor *governor)
1359 {
1360         int err;
1361
1362         if (!governor)
1363                 return -EINVAL;
1364
1365         mutex_lock(&cpufreq_governor_mutex);
1366
1367         err = -EBUSY;
1368         if (__find_governor(governor->name) == NULL) {
1369                 err = 0;
1370                 list_add(&governor->governor_list, &cpufreq_governor_list);
1371         }
1372
1373         mutex_unlock(&cpufreq_governor_mutex);
1374         return err;
1375 }
1376 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1377
1378
1379 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1380 {
1381         if (!governor)
1382                 return;
1383
1384         mutex_lock(&cpufreq_governor_mutex);
1385         list_del(&governor->governor_list);
1386         mutex_unlock(&cpufreq_governor_mutex);
1387         return;
1388 }
1389 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1390
1391
1392
1393 /*********************************************************************
1394  *                          POLICY INTERFACE                         *
1395  *********************************************************************/
1396
1397 /**
1398  * cpufreq_get_policy - get the current cpufreq_policy
1399  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1400  *
1401  * Reads the current cpufreq policy.
1402  */
1403 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1404 {
1405         struct cpufreq_policy *cpu_policy;
1406         if (!policy)
1407                 return -EINVAL;
1408
1409         cpu_policy = cpufreq_cpu_get(cpu);
1410         if (!cpu_policy)
1411                 return -EINVAL;
1412
1413         mutex_lock(&cpu_policy->lock);
1414         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1415         mutex_unlock(&cpu_policy->lock);
1416
1417         cpufreq_cpu_put(cpu_policy);
1418         return 0;
1419 }
1420 EXPORT_SYMBOL(cpufreq_get_policy);
1421
1422
1423 /*
1424  * data   : current policy.
1425  * policy : policy to be set.
1426  */
1427 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1428                                 struct cpufreq_policy *policy)
1429 {
1430         int ret = 0;
1431
1432         cpufreq_debug_disable_ratelimit();
1433         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1434                 policy->min, policy->max);
1435
1436         memcpy(&policy->cpuinfo, &data->cpuinfo,
1437                                 sizeof(struct cpufreq_cpuinfo));
1438
1439         if (policy->min > data->min && policy->min > policy->max) {
1440                 ret = -EINVAL;
1441                 goto error_out;
1442         }
1443
1444         /* verify the cpu speed can be set within this limit */
1445         ret = cpufreq_driver->verify(policy);
1446         if (ret)
1447                 goto error_out;
1448
1449         /* adjust if necessary - all reasons */
1450         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1451                         CPUFREQ_ADJUST, policy);
1452
1453         /* adjust if necessary - hardware incompatibility*/
1454         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1455                         CPUFREQ_INCOMPATIBLE, policy);
1456
1457         /* verify the cpu speed can be set within this limit,
1458            which might be different to the first one */
1459         ret = cpufreq_driver->verify(policy);
1460         if (ret)
1461                 goto error_out;
1462
1463         /* notification of the new policy */
1464         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1465                         CPUFREQ_NOTIFY, policy);
1466
1467         data->min = policy->min;
1468         data->max = policy->max;
1469
1470         dprintk("new min and max freqs are %u - %u kHz\n",
1471                                         data->min, data->max);
1472
1473         if (cpufreq_driver->setpolicy) {
1474                 data->policy = policy->policy;
1475                 dprintk("setting range\n");
1476                 ret = cpufreq_driver->setpolicy(policy);
1477         } else {
1478                 if (policy->governor != data->governor) {
1479                         /* save old, working values */
1480                         struct cpufreq_governor *old_gov = data->governor;
1481
1482                         dprintk("governor switch\n");
1483
1484                         /* end old governor */
1485                         if (data->governor)
1486                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1487
1488                         /* start new governor */
1489                         data->governor = policy->governor;
1490                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1491                                 /* new governor failed, so re-start old one */
1492                                 dprintk("starting governor %s failed\n",
1493                                                         data->governor->name);
1494                                 if (old_gov) {
1495                                         data->governor = old_gov;
1496                                         __cpufreq_governor(data,
1497                                                            CPUFREQ_GOV_START);
1498                                 }
1499                                 ret = -EINVAL;
1500                                 goto error_out;
1501                         }
1502                         /* might be a policy change, too, so fall through */
1503                 }
1504                 dprintk("governor: change or update limits\n");
1505                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1506         }
1507
1508 error_out:
1509         cpufreq_debug_enable_ratelimit();
1510         return ret;
1511 }
1512
1513 /**
1514  *      cpufreq_set_policy - set a new CPUFreq policy
1515  *      @policy: policy to be set.
1516  *
1517  *      Sets a new CPU frequency and voltage scaling policy.
1518  */
1519 int cpufreq_set_policy(struct cpufreq_policy *policy)
1520 {
1521         int ret = 0;
1522         struct cpufreq_policy *data;
1523
1524         if (!policy)
1525                 return -EINVAL;
1526
1527         data = cpufreq_cpu_get(policy->cpu);
1528         if (!data)
1529                 return -EINVAL;
1530
1531         /* lock this CPU */
1532         mutex_lock(&data->lock);
1533
1534         ret = __cpufreq_set_policy(data, policy);
1535         data->user_policy.min = data->min;
1536         data->user_policy.max = data->max;
1537         data->user_policy.policy = data->policy;
1538         data->user_policy.governor = data->governor;
1539
1540         mutex_unlock(&data->lock);
1541
1542         cpufreq_cpu_put(data);
1543
1544         return ret;
1545 }
1546 EXPORT_SYMBOL(cpufreq_set_policy);
1547
1548
1549 /**
1550  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1551  *      @cpu: CPU which shall be re-evaluated
1552  *
1553  *      Usefull for policy notifiers which have different necessities
1554  *      at different times.
1555  */
1556 int cpufreq_update_policy(unsigned int cpu)
1557 {
1558         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1559         struct cpufreq_policy policy;
1560         int ret = 0;
1561
1562         if (!data)
1563                 return -ENODEV;
1564
1565         mutex_lock(&data->lock);
1566
1567         dprintk("updating policy for CPU %u\n", cpu);
1568         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1569         policy.min = data->user_policy.min;
1570         policy.max = data->user_policy.max;
1571         policy.policy = data->user_policy.policy;
1572         policy.governor = data->user_policy.governor;
1573
1574         /* BIOS might change freq behind our back
1575           -> ask driver for current freq and notify governors about a change */
1576         if (cpufreq_driver->get) {
1577                 policy.cur = cpufreq_driver->get(cpu);
1578                 if (!data->cur) {
1579                         dprintk("Driver did not initialize current freq");
1580                         data->cur = policy.cur;
1581                 } else {
1582                         if (data->cur != policy.cur)
1583                                 cpufreq_out_of_sync(cpu, data->cur,
1584                                                                 policy.cur);
1585                 }
1586         }
1587
1588         ret = __cpufreq_set_policy(data, &policy);
1589
1590         mutex_unlock(&data->lock);
1591         cpufreq_cpu_put(data);
1592         return ret;
1593 }
1594 EXPORT_SYMBOL(cpufreq_update_policy);
1595
1596 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1597                                         unsigned long action, void *hcpu)
1598 {
1599         unsigned int cpu = (unsigned long)hcpu;
1600         struct cpufreq_policy *policy;
1601         struct sys_device *sys_dev;
1602
1603         sys_dev = get_cpu_sysdev(cpu);
1604
1605         if (sys_dev) {
1606                 switch (action) {
1607                 case CPU_ONLINE:
1608                         cpufreq_add_dev(sys_dev);
1609                         break;
1610                 case CPU_DOWN_PREPARE:
1611                         /*
1612                          * We attempt to put this cpu in lowest frequency
1613                          * possible before going down. This will permit
1614                          * hardware-managed P-State to switch other related
1615                          * threads to min or higher speeds if possible.
1616                          */
1617                         policy = cpufreq_cpu_data[cpu];
1618                         if (policy) {
1619                                 cpufreq_driver_target(policy, policy->min,
1620                                                 CPUFREQ_RELATION_H);
1621                         }
1622                         break;
1623                 case CPU_DEAD:
1624                         cpufreq_remove_dev(sys_dev);
1625                         break;
1626                 }
1627         }
1628         return NOTIFY_OK;
1629 }
1630
1631 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1632 {
1633     .notifier_call = cpufreq_cpu_callback,
1634 };
1635
1636 /*********************************************************************
1637  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1638  *********************************************************************/
1639
1640 /**
1641  * cpufreq_register_driver - register a CPU Frequency driver
1642  * @driver_data: A struct cpufreq_driver containing the values#
1643  * submitted by the CPU Frequency driver.
1644  *
1645  *   Registers a CPU Frequency driver to this core code. This code
1646  * returns zero on success, -EBUSY when another driver got here first
1647  * (and isn't unregistered in the meantime).
1648  *
1649  */
1650 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1651 {
1652         unsigned long flags;
1653         int ret;
1654
1655         if (!driver_data || !driver_data->verify || !driver_data->init ||
1656             ((!driver_data->setpolicy) && (!driver_data->target)))
1657                 return -EINVAL;
1658
1659         dprintk("trying to register driver %s\n", driver_data->name);
1660
1661         if (driver_data->setpolicy)
1662                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1663
1664         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1665         if (cpufreq_driver) {
1666                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1667                 return -EBUSY;
1668         }
1669         cpufreq_driver = driver_data;
1670         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1671
1672         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1673
1674         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1675                 int i;
1676                 ret = -ENODEV;
1677
1678                 /* check for at least one working CPU */
1679                 for (i=0; i<NR_CPUS; i++)
1680                         if (cpufreq_cpu_data[i])
1681                                 ret = 0;
1682
1683                 /* if all ->init() calls failed, unregister */
1684                 if (ret) {
1685                         dprintk("no CPU initialized for driver %s\n",
1686                                                         driver_data->name);
1687                         sysdev_driver_unregister(&cpu_sysdev_class,
1688                                                 &cpufreq_sysdev_driver);
1689
1690                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1691                         cpufreq_driver = NULL;
1692                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1693                 }
1694         }
1695
1696         if (!ret) {
1697                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1698                 dprintk("driver %s up and running\n", driver_data->name);
1699                 cpufreq_debug_enable_ratelimit();
1700         }
1701
1702         return (ret);
1703 }
1704 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1705
1706
1707 /**
1708  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1709  *
1710  *    Unregister the current CPUFreq driver. Only call this if you have
1711  * the right to do so, i.e. if you have succeeded in initialising before!
1712  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1713  * currently not initialised.
1714  */
1715 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1716 {
1717         unsigned long flags;
1718
1719         cpufreq_debug_disable_ratelimit();
1720
1721         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1722                 cpufreq_debug_enable_ratelimit();
1723                 return -EINVAL;
1724         }
1725
1726         dprintk("unregistering driver %s\n", driver->name);
1727
1728         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1729         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1730
1731         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1732         cpufreq_driver = NULL;
1733         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1734
1735         return 0;
1736 }
1737 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);