drivers/thermal/cpu_cooling.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/drivers/thermal/cpu_cooling.c
   4  *
   5  *  Copyright (C) 2012  Samsung Electronics Co., Ltd(http://www.samsung.com)
   6  *
   7  *  Copyright (C) 2012-2018 Linaro Limited.
   8  *
   9  *  Authors:    Amit Daniel <amit.kachhap@linaro.org>
  10  *              Viresh Kumar <viresh.kumar@linaro.org>
  11  *
  12  */
  13 #include <linux/module.h>
  14 #include <linux/thermal.h>
  15 #include <linux/cpufreq.h>
  16 #include <linux/err.h>
  17 #include <linux/idr.h>
  18 #include <linux/pm_opp.h>
  19 #include <linux/pm_qos.h>
  20 #include <linux/slab.h>
  21 #include <linux/cpu.h>
  22 #include <linux/cpu_cooling.h>
  23 #include <linux/energy_model.h>
  24
  25 #include <trace/events/thermal.h>
  26
  27 /*
  28  * Cooling state <-> CPUFreq frequency
  29  *
  30  * Cooling states are translated to frequencies throughout this driver and this
  31  * is the relation between them.
  32  *
  33  * Highest cooling state corresponds to lowest possible frequency.
  34  *
  35  * i.e.
  36  *      level 0 --> 1st Max Freq
  37  *      level 1 --> 2nd Max Freq
  38  *      ...
  39  */
  40
  41 /**
  42  * struct time_in_idle - Idle time stats
  43  * @time: previous reading of the absolute time that this cpu was idle
  44  * @timestamp: wall time of the last invocation of get_cpu_idle_time_us()
  45  */
  46 struct time_in_idle {
  47         u64 time;
  48         u64 timestamp;
  49 };
  50
  51 /**
  52  * struct cpufreq_cooling_device - data for cooling device with cpufreq
  53  * @id: unique integer value corresponding to each cpufreq_cooling_device
  54  *      registered.
  55  * @last_load: load measured by the latest call to cpufreq_get_requested_power()
  56  * @cpufreq_state: integer value representing the current state of cpufreq
  57  *      cooling devices.
  58  * @max_level: maximum cooling level. One less than total number of valid
  59  *      cpufreq frequencies.
  60  * @em: Reference on the Energy Model of the device
  61  * @cdev: thermal_cooling_device pointer to keep track of the
  62  *      registered cooling device.
  63  * @policy: cpufreq policy.
  64  * @node: list_head to link all cpufreq_cooling_device together.
  65  * @idle_time: idle time stats
  66  *
  67  * This structure is required for keeping information of each registered
  68  * cpufreq_cooling_device.
  69  */
  70 struct cpufreq_cooling_device {
  71         int id;
  72         u32 last_load;
  73         unsigned int cpufreq_state;
  74         unsigned int max_level;
  75         struct em_perf_domain *em;
  76         struct cpufreq_policy *policy;
  77         struct list_head node;
  78         struct time_in_idle *idle_time;
  79         struct freq_qos_request qos_req;
  80 };
  81
  82 static DEFINE_IDA(cpufreq_ida);
  83 static DEFINE_MUTEX(cooling_list_lock);
  84 static LIST_HEAD(cpufreq_cdev_list);
  85
  86 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
  87 /**
  88  * get_level: Find the level for a particular frequency
  89  * @cpufreq_cdev: cpufreq_cdev for which the property is required
  90  * @freq: Frequency
  91  *
  92  * Return: level corresponding to the frequency.
  93  */
  94 static unsigned long get_level(struct cpufreq_cooling_device *cpufreq_cdev,
  95                                unsigned int freq)
  96 {
  97         int i;
  98
  99         for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
 100                 if (freq > cpufreq_cdev->em->table[i].frequency)
 101                         break;
 102         }
 103
 104         return cpufreq_cdev->max_level - i - 1;
 105 }
 106
 107 static u32 cpu_freq_to_power(struct cpufreq_cooling_device *cpufreq_cdev,
 108                              u32 freq)
 109 {
 110         int i;
 111
 112         for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
 113                 if (freq > cpufreq_cdev->em->table[i].frequency)
 114                         break;
 115         }
 116
 117         return cpufreq_cdev->em->table[i + 1].power;
 118 }
 119
 120 static u32 cpu_power_to_freq(struct cpufreq_cooling_device *cpufreq_cdev,
 121                              u32 power)
 122 {
 123         int i;
 124
 125         for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
 126                 if (power > cpufreq_cdev->em->table[i].power)
 127                         break;
 128         }
 129
 130         return cpufreq_cdev->em->table[i + 1].frequency;
 131 }
 132
 133 /**
 134  * get_load() - get load for a cpu since last updated
 135  * @cpufreq_cdev:       &struct cpufreq_cooling_device for this cpu
 136  * @cpu:        cpu number
 137  * @cpu_idx:    index of the cpu in time_in_idle*
 138  *
 139  * Return: The average load of cpu @cpu in percentage since this
 140  * function was last called.
 141  */
 142 static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
 143                     int cpu_idx)
 144 {
 145         u32 load;
 146         u64 now, now_idle, delta_time, delta_idle;
 147         struct time_in_idle *idle_time = &cpufreq_cdev->idle_time[cpu_idx];
 148
 149         now_idle = get_cpu_idle_time(cpu, &now, 0);
 150         delta_idle = now_idle - idle_time->time;
 151         delta_time = now - idle_time->timestamp;
 152
 153         if (delta_time <= delta_idle)
 154                 load = 0;
 155         else
 156                 load = div64_u64(100 * (delta_time - delta_idle), delta_time);
 157
 158         idle_time->time = now_idle;
 159         idle_time->timestamp = now;
 160
 161         return load;
 162 }
 163
 164 /**
 165  * get_dynamic_power() - calculate the dynamic power
 166  * @cpufreq_cdev:       &cpufreq_cooling_device for this cdev
 167  * @freq:       current frequency
 168  *
 169  * Return: the dynamic power consumed by the cpus described by
 170  * @cpufreq_cdev.
 171  */
 172 static u32 get_dynamic_power(struct cpufreq_cooling_device *cpufreq_cdev,
 173                              unsigned long freq)
 174 {
 175         u32 raw_cpu_power;
 176
 177         raw_cpu_power = cpu_freq_to_power(cpufreq_cdev, freq);
 178         return (raw_cpu_power * cpufreq_cdev->last_load) / 100;
 179 }
 180
 181 /**
 182  * cpufreq_get_requested_power() - get the current power
 183  * @cdev:       &thermal_cooling_device pointer
 184  * @tz:         a valid thermal zone device pointer
 185  * @power:      pointer in which to store the resulting power
 186  *
 187  * Calculate the current power consumption of the cpus in milliwatts
 188  * and store it in @power.  This function should actually calculate
 189  * the requested power, but it's hard to get the frequency that
 190  * cpufreq would have assigned if there were no thermal limits.
 191  * Instead, we calculate the current power on the assumption that the
 192  * immediate future will look like the immediate past.
 193  *
 194  * We use the current frequency and the average load since this
 195  * function was last called.  In reality, there could have been
 196  * multiple opps since this function was last called and that affects
 197  * the load calculation.  While it's not perfectly accurate, this
 198  * simplification is good enough and works.  REVISIT this, as more
 199  * complex code may be needed if experiments show that it's not
 200  * accurate enough.
 201  *
 202  * Return: 0 on success, -E* if getting the static power failed.
 203  */
 204 static int cpufreq_get_requested_power(struct thermal_cooling_device *cdev,
 205                                        struct thermal_zone_device *tz,
 206                                        u32 *power)
 207 {
 208         unsigned long freq;
 209         int i = 0, cpu;
 210         u32 total_load = 0;
 211         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 212         struct cpufreq_policy *policy = cpufreq_cdev->policy;
 213         u32 *load_cpu = NULL;
 214
 215         freq = cpufreq_quick_get(policy->cpu);
 216
 217         if (trace_thermal_power_cpu_get_power_enabled()) {
 218                 u32 ncpus = cpumask_weight(policy->related_cpus);
 219
 220                 load_cpu = kcalloc(ncpus, sizeof(*load_cpu), GFP_KERNEL);
 221         }
 222
 223         for_each_cpu(cpu, policy->related_cpus) {
 224                 u32 load;
 225
 226                 if (cpu_online(cpu))
 227                         load = get_load(cpufreq_cdev, cpu, i);
 228                 else
 229                         load = 0;
 230
 231                 total_load += load;
 232                 if (load_cpu)
 233                         load_cpu[i] = load;
 234
 235                 i++;
 236         }
 237
 238         cpufreq_cdev->last_load = total_load;
 239
 240         *power = get_dynamic_power(cpufreq_cdev, freq);
 241
 242         if (load_cpu) {
 243                 trace_thermal_power_cpu_get_power(policy->related_cpus, freq,
 244                                                   load_cpu, i, *power);
 245
 246                 kfree(load_cpu);
 247         }
 248
 249         return 0;
 250 }
 251
 252 /**
 253  * cpufreq_state2power() - convert a cpu cdev state to power consumed
 254  * @cdev:       &thermal_cooling_device pointer
 255  * @tz:         a valid thermal zone device pointer
 256  * @state:      cooling device state to be converted
 257  * @power:      pointer in which to store the resulting power
 258  *
 259  * Convert cooling device state @state into power consumption in
 260  * milliwatts assuming 100% load.  Store the calculated power in
 261  * @power.
 262  *
 263  * Return: 0 on success, -EINVAL if the cooling device state could not
 264  * be converted into a frequency or other -E* if there was an error
 265  * when calculating the static power.
 266  */
 267 static int cpufreq_state2power(struct thermal_cooling_device *cdev,
 268                                struct thermal_zone_device *tz,
 269                                unsigned long state, u32 *power)
 270 {
 271         unsigned int freq, num_cpus, idx;
 272         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 273
 274         /* Request state should be less than max_level */
 275         if (WARN_ON(state > cpufreq_cdev->max_level))
 276                 return -EINVAL;
 277
 278         num_cpus = cpumask_weight(cpufreq_cdev->policy->cpus);
 279
 280         idx = cpufreq_cdev->max_level - state;
 281         freq = cpufreq_cdev->em->table[idx].frequency;
 282         *power = cpu_freq_to_power(cpufreq_cdev, freq) * num_cpus;
 283
 284         return 0;
 285 }
 286
 287 /**
 288  * cpufreq_power2state() - convert power to a cooling device state
 289  * @cdev:       &thermal_cooling_device pointer
 290  * @tz:         a valid thermal zone device pointer
 291  * @power:      power in milliwatts to be converted
 292  * @state:      pointer in which to store the resulting state
 293  *
 294  * Calculate a cooling device state for the cpus described by @cdev
 295  * that would allow them to consume at most @power mW and store it in
 296  * @state.  Note that this calculation depends on external factors
 297  * such as the cpu load or the current static power.  Calling this
 298  * function with the same power as input can yield different cooling
 299  * device states depending on those external factors.
 300  *
 301  * Return: 0 on success, -ENODEV if no cpus are online or -EINVAL if
 302  * the calculated frequency could not be converted to a valid state.
 303  * The latter should not happen unless the frequencies available to
 304  * cpufreq have changed since the initialization of the cpu cooling
 305  * device.
 306  */
 307 static int cpufreq_power2state(struct thermal_cooling_device *cdev,
 308                                struct thermal_zone_device *tz, u32 power,
 309                                unsigned long *state)
 310 {
 311         unsigned int target_freq;
 312         u32 last_load, normalised_power;
 313         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 314         struct cpufreq_policy *policy = cpufreq_cdev->policy;
 315
 316         last_load = cpufreq_cdev->last_load ?: 1;
 317         normalised_power = (power * 100) / last_load;
 318         target_freq = cpu_power_to_freq(cpufreq_cdev, normalised_power);
 319
 320         *state = get_level(cpufreq_cdev, target_freq);
 321         trace_thermal_power_cpu_limit(policy->related_cpus, target_freq, *state,
 322                                       power);
 323         return 0;
 324 }
 325
 326 static inline bool em_is_sane(struct cpufreq_cooling_device *cpufreq_cdev,
 327                               struct em_perf_domain *em) {
 328         struct cpufreq_policy *policy;
 329         unsigned int nr_levels;
 330
 331         if (!em)
 332                 return false;
 333
 334         policy = cpufreq_cdev->policy;
 335         if (!cpumask_equal(policy->related_cpus, to_cpumask(em->cpus))) {
 336                 pr_err("The span of pd %*pbl is misaligned with cpufreq policy %*pbl\n",
 337                         cpumask_pr_args(to_cpumask(em->cpus)),
 338                         cpumask_pr_args(policy->related_cpus));
 339                 return false;
 340         }
 341
 342         nr_levels = cpufreq_cdev->max_level + 1;
 343         if (em->nr_cap_states != nr_levels) {
 344                 pr_err("The number of cap states in pd %*pbl (%u) doesn't match the number of cooling levels (%u)\n",
 345                         cpumask_pr_args(to_cpumask(em->cpus)),
 346                         em->nr_cap_states, nr_levels);
 347                 return false;
 348         }
 349
 350         return true;
 351 }
 352 #endif /* CONFIG_THERMAL_GOV_POWER_ALLOCATOR */
 353
 354 static unsigned int get_state_freq(struct cpufreq_cooling_device *cpufreq_cdev,
 355                                    unsigned long state)
 356 {
 357         struct cpufreq_policy *policy;
 358         unsigned long idx;
 359
 360 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
 361         /* Use the Energy Model table if available */
 362         if (cpufreq_cdev->em) {
 363                 idx = cpufreq_cdev->max_level - state;
 364                 return cpufreq_cdev->em->table[idx].frequency;
 365         }
 366 #endif
 367
 368         /* Otherwise, fallback on the CPUFreq table */
 369         policy = cpufreq_cdev->policy;
 370         if (policy->freq_table_sorted == CPUFREQ_TABLE_SORTED_ASCENDING)
 371                 idx = cpufreq_cdev->max_level - state;
 372         else
 373                 idx = state;
 374
 375         return policy->freq_table[idx].frequency;
 376 }
 377
 378 /* cpufreq cooling device callback functions are defined below */
 379
 380 /**
 381  * cpufreq_get_max_state - callback function to get the max cooling state.
 382  * @cdev: thermal cooling device pointer.
 383  * @state: fill this variable with the max cooling state.
 384  *
 385  * Callback for the thermal cooling device to return the cpufreq
 386  * max cooling state.
 387  *
 388  * Return: 0 on success, an error code otherwise.
 389  */
 390 static int cpufreq_get_max_state(struct thermal_cooling_device *cdev,
 391                                  unsigned long *state)
 392 {
 393         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 394
 395         *state = cpufreq_cdev->max_level;
 396         return 0;
 397 }
 398
 399 /**
 400  * cpufreq_get_cur_state - callback function to get the current cooling state.
 401  * @cdev: thermal cooling device pointer.
 402  * @state: fill this variable with the current cooling state.
 403  *
 404  * Callback for the thermal cooling device to return the cpufreq
 405  * current cooling state.
 406  *
 407  * Return: 0 on success, an error code otherwise.
 408  */
 409 static int cpufreq_get_cur_state(struct thermal_cooling_device *cdev,
 410                                  unsigned long *state)
 411 {
 412         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 413
 414         *state = cpufreq_cdev->cpufreq_state;
 415
 416         return 0;
 417 }
 418
 419 /**
 420  * cpufreq_set_cur_state - callback function to set the current cooling state.
 421  * @cdev: thermal cooling device pointer.
 422  * @state: set this variable to the current cooling state.
 423  *
 424  * Callback for the thermal cooling device to change the cpufreq
 425  * current cooling state.
 426  *
 427  * Return: 0 on success, an error code otherwise.
 428  */
 429 static int cpufreq_set_cur_state(struct thermal_cooling_device *cdev,
 430                                  unsigned long state)
 431 {
 432         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 433
 434         /* Request state should be less than max_level */
 435         if (WARN_ON(state > cpufreq_cdev->max_level))
 436                 return -EINVAL;
 437
 438         /* Check if the old cooling action is same as new cooling action */
 439         if (cpufreq_cdev->cpufreq_state == state)
 440                 return 0;
 441
 442         cpufreq_cdev->cpufreq_state = state;
 443
 444         return freq_qos_update_request(&cpufreq_cdev->qos_req,
 445                                 get_state_freq(cpufreq_cdev, state));
 446 }
 447
 448 /* Bind cpufreq callbacks to thermal cooling device ops */
 449
 450 static struct thermal_cooling_device_ops cpufreq_cooling_ops = {
 451         .get_max_state          = cpufreq_get_max_state,
 452         .get_cur_state          = cpufreq_get_cur_state,
 453         .set_cur_state          = cpufreq_set_cur_state,
 454 };
 455
 456 /**
 457  * __cpufreq_cooling_register - helper function to create cpufreq cooling device
 458  * @np: a valid struct device_node to the cooling device device tree node
 459  * @policy: cpufreq policy
 460  * Normally this should be same as cpufreq policy->related_cpus.
 461  * @em: Energy Model of the cpufreq policy
 462  *
 463  * This interface function registers the cpufreq cooling device with the name
 464  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 465  * cooling devices. It also gives the opportunity to link the cooling device
 466  * with a device tree node, in order to bind it via the thermal DT code.
 467  *
 468  * Return: a valid struct thermal_cooling_device pointer on success,
 469  * on failure, it returns a corresponding ERR_PTR().
 470  */
 471 static struct thermal_cooling_device *
 472 __cpufreq_cooling_register(struct device_node *np,
 473                         struct cpufreq_policy *policy,
 474                         struct em_perf_domain *em)
 475 {
 476         struct thermal_cooling_device *cdev;
 477         struct cpufreq_cooling_device *cpufreq_cdev;
 478         char dev_name[THERMAL_NAME_LENGTH];
 479         unsigned int i, num_cpus;
 480         struct device *dev;
 481         int ret;
 482         struct thermal_cooling_device_ops *cooling_ops;
 483
 484         dev = get_cpu_device(policy->cpu);
 485         if (unlikely(!dev)) {
 486                 pr_warn("No cpu device for cpu %d\n", policy->cpu);
 487                 return ERR_PTR(-ENODEV);
 488         }
 489
 490
 491         if (IS_ERR_OR_NULL(policy)) {
 492                 pr_err("%s: cpufreq policy isn't valid: %p\n", __func__, policy);
 493                 return ERR_PTR(-EINVAL);
 494         }
 495
 496         i = cpufreq_table_count_valid_entries(policy);
 497         if (!i) {
 498                 pr_debug("%s: CPUFreq table not found or has no valid entries\n",
 499                          __func__);
 500                 return ERR_PTR(-ENODEV);
 501         }
 502
 503         cpufreq_cdev = kzalloc(sizeof(*cpufreq_cdev), GFP_KERNEL);
 504         if (!cpufreq_cdev)
 505                 return ERR_PTR(-ENOMEM);
 506
 507         cpufreq_cdev->policy = policy;
 508         num_cpus = cpumask_weight(policy->related_cpus);
 509         cpufreq_cdev->idle_time = kcalloc(num_cpus,
 510                                          sizeof(*cpufreq_cdev->idle_time),
 511                                          GFP_KERNEL);
 512         if (!cpufreq_cdev->idle_time) {
 513                 cdev = ERR_PTR(-ENOMEM);
 514                 goto free_cdev;
 515         }
 516
 517         /* max_level is an index, not a counter */
 518         cpufreq_cdev->max_level = i - 1;
 519
 520         ret = ida_simple_get(&cpufreq_ida, 0, 0, GFP_KERNEL);
 521         if (ret < 0) {
 522                 cdev = ERR_PTR(ret);
 523                 goto free_idle_time;
 524         }
 525         cpufreq_cdev->id = ret;
 526
 527         snprintf(dev_name, sizeof(dev_name), "thermal-cpufreq-%d",
 528                  cpufreq_cdev->id);
 529
 530         cooling_ops = &cpufreq_cooling_ops;
 531
 532 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
 533         if (em_is_sane(cpufreq_cdev, em)) {
 534                 cpufreq_cdev->em = em;
 535                 cooling_ops->get_requested_power = cpufreq_get_requested_power;
 536                 cooling_ops->state2power = cpufreq_state2power;
 537                 cooling_ops->power2state = cpufreq_power2state;
 538         } else
 539 #endif
 540         if (policy->freq_table_sorted == CPUFREQ_TABLE_UNSORTED) {
 541                 pr_err("%s: unsorted frequency tables are not supported\n",
 542                        __func__);
 543                 cdev = ERR_PTR(-EINVAL);
 544                 goto remove_ida;
 545         }
 546
 547         ret = freq_qos_add_request(&policy->constraints,
 548                                    &cpufreq_cdev->qos_req, FREQ_QOS_MAX,
 549                                    get_state_freq(cpufreq_cdev, 0));
 550         if (ret < 0) {
 551                 pr_err("%s: Failed to add freq constraint (%d)\n", __func__,
 552                        ret);
 553                 cdev = ERR_PTR(ret);
 554                 goto remove_ida;
 555         }
 556
 557         cdev = thermal_of_cooling_device_register(np, dev_name, cpufreq_cdev,
 558                                                   cooling_ops);
 559         if (IS_ERR(cdev))
 560                 goto remove_qos_req;
 561
 562         mutex_lock(&cooling_list_lock);
 563         list_add(&cpufreq_cdev->node, &cpufreq_cdev_list);
 564         mutex_unlock(&cooling_list_lock);
 565
 566         return cdev;
 567
 568 remove_qos_req:
 569         freq_qos_remove_request(&cpufreq_cdev->qos_req);
 570 remove_ida:
 571         ida_simple_remove(&cpufreq_ida, cpufreq_cdev->id);
 572 free_idle_time:
 573         kfree(cpufreq_cdev->idle_time);
 574 free_cdev:
 575         kfree(cpufreq_cdev);
 576         return cdev;
 577 }
 578
 579 /**
 580  * cpufreq_cooling_register - function to create cpufreq cooling device.
 581  * @policy: cpufreq policy
 582  *
 583  * This interface function registers the cpufreq cooling device with the name
 584  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 585  * cooling devices.
 586  *
 587  * Return: a valid struct thermal_cooling_device pointer on success,
 588  * on failure, it returns a corresponding ERR_PTR().
 589  */
 590 struct thermal_cooling_device *
 591 cpufreq_cooling_register(struct cpufreq_policy *policy)
 592 {
 593         return __cpufreq_cooling_register(NULL, policy, NULL);
 594 }
 595 EXPORT_SYMBOL_GPL(cpufreq_cooling_register);
 596
 597 /**
 598  * of_cpufreq_cooling_register - function to create cpufreq cooling device.
 599  * @policy: cpufreq policy
 600  *
 601  * This interface function registers the cpufreq cooling device with the name
 602  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 603  * cooling devices. Using this API, the cpufreq cooling device will be
 604  * linked to the device tree node provided.
 605  *
 606  * Using this function, the cooling device will implement the power
 607  * extensions by using a simple cpu power model.  The cpus must have
 608  * registered their OPPs using the OPP library.
 609  *
 610  * It also takes into account, if property present in policy CPU node, the
 611  * static power consumed by the cpu.
 612  *
 613  * Return: a valid struct thermal_cooling_device pointer on success,
 614  * and NULL on failure.
 615  */
 616 struct thermal_cooling_device *
 617 of_cpufreq_cooling_register(struct cpufreq_policy *policy)
 618 {
 619         struct device_node *np = of_get_cpu_node(policy->cpu, NULL);
 620         struct thermal_cooling_device *cdev = NULL;
 621
 622         if (!np) {
 623                 pr_err("cpu_cooling: OF node not available for cpu%d\n",
 624                        policy->cpu);
 625                 return NULL;
 626         }
 627
 628         if (of_find_property(np, "#cooling-cells", NULL)) {
 629                 struct em_perf_domain *em = em_cpu_get(policy->cpu);
 630
 631                 cdev = __cpufreq_cooling_register(np, policy, em);
 632                 if (IS_ERR(cdev)) {
 633                         pr_err("cpu_cooling: cpu%d failed to register as cooling device: %ld\n",
 634                                policy->cpu, PTR_ERR(cdev));
 635                         cdev = NULL;
 636                 }
 637         }
 638
 639         of_node_put(np);
 640         return cdev;
 641 }
 642 EXPORT_SYMBOL_GPL(of_cpufreq_cooling_register);
 643
 644 /**
 645  * cpufreq_cooling_unregister - function to remove cpufreq cooling device.
 646  * @cdev: thermal cooling device pointer.
 647  *
 648  * This interface function unregisters the "thermal-cpufreq-%x" cooling device.
 649  */
 650 void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
 651 {
 652         struct cpufreq_cooling_device *cpufreq_cdev;
 653
 654         if (!cdev)
 655                 return;
 656
 657         cpufreq_cdev = cdev->devdata;
 658
 659         mutex_lock(&cooling_list_lock);
 660         list_del(&cpufreq_cdev->node);
 661         mutex_unlock(&cooling_list_lock);
 662
 663         thermal_cooling_device_unregister(cdev);
 664         freq_qos_remove_request(&cpufreq_cdev->qos_req);
 665         ida_simple_remove(&cpufreq_ida, cpufreq_cdev->id);
 666         kfree(cpufreq_cdev->idle_time);
 667         kfree(cpufreq_cdev);
 668 }
 669 EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister);