drivers/thermal/cpufreq_cooling.c

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