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/cpu.h>
  14 #include <linux/cpufreq.h>
  15 #include <linux/cpu_cooling.h>
  16 #include <linux/device.h>
  17 #include <linux/energy_model.h>
  18 #include <linux/err.h>
  19 #include <linux/export.h>
  20 #include <linux/pm_opp.h>
  21 #include <linux/pm_qos.h>
  22 #include <linux/slab.h>
  23 #include <linux/thermal.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  * @last_load: load measured by the latest call to cpufreq_get_requested_power()
  54  * @cpufreq_state: integer value representing the current state of cpufreq
  55  *      cooling devices.
  56  * @max_level: maximum cooling level. One less than total number of valid
  57  *      cpufreq frequencies.
  58  * @em: Reference on the Energy Model of the device
  59  * @cdev: thermal_cooling_device pointer to keep track of the
  60  *      registered cooling device.
  61  * @policy: cpufreq policy.
  62  * @cooling_ops: cpufreq callbacks to thermal cooling device ops
  63  * @idle_time: idle time stats
  64  * @qos_req: PM QoS contraint to apply
  65  *
  66  * This structure is required for keeping information of each registered
  67  * cpufreq_cooling_device.
  68  */
  69 struct cpufreq_cooling_device {
  70         u32 last_load;
  71         unsigned int cpufreq_state;
  72         unsigned int max_level;
  73         struct em_perf_domain *em;
  74         struct cpufreq_policy *policy;
  75         struct thermal_cooling_device_ops cooling_ops;
  76 #ifndef CONFIG_SMP
  77         struct time_in_idle *idle_time;
  78 #endif
  79         struct freq_qos_request qos_req;
  80 };
  81
  82 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
  83 /**
  84  * get_level: Find the level for a particular frequency
  85  * @cpufreq_cdev: cpufreq_cdev for which the property is required
  86  * @freq: Frequency
  87  *
  88  * Return: level corresponding to the frequency.
  89  */
  90 static unsigned long get_level(struct cpufreq_cooling_device *cpufreq_cdev,
  91                                unsigned int freq)
  92 {
  93         int i;
  94
  95         for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
  96                 if (freq > cpufreq_cdev->em->table[i].frequency)
  97                         break;
  98         }
  99
 100         return cpufreq_cdev->max_level - i - 1;
 101 }
 102
 103 static u32 cpu_freq_to_power(struct cpufreq_cooling_device *cpufreq_cdev,
 104                              u32 freq)
 105 {
 106         int i;
 107
 108         for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
 109                 if (freq > cpufreq_cdev->em->table[i].frequency)
 110                         break;
 111         }
 112
 113         return cpufreq_cdev->em->table[i + 1].power;
 114 }
 115
 116 static u32 cpu_power_to_freq(struct cpufreq_cooling_device *cpufreq_cdev,
 117                              u32 power)
 118 {
 119         int i;
 120
 121         for (i = cpufreq_cdev->max_level; i > 0; i--) {
 122                 if (power >= cpufreq_cdev->em->table[i].power)
 123                         break;
 124         }
 125
 126         return cpufreq_cdev->em->table[i].frequency;
 127 }
 128
 129 /**
 130  * get_load() - get load for a cpu
 131  * @cpufreq_cdev: struct cpufreq_cooling_device for the cpu
 132  * @cpu: cpu number
 133  * @cpu_idx: index of the cpu in time_in_idle array
 134  *
 135  * Return: The average load of cpu @cpu in percentage since this
 136  * function was last called.
 137  */
 138 #ifdef CONFIG_SMP
 139 static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
 140                     int cpu_idx)
 141 {
 142         unsigned long max = arch_scale_cpu_capacity(cpu);
 143         unsigned long util;
 144
 145         util = sched_cpu_util(cpu, max);
 146         return (util * 100) / max;
 147 }
 148 #else /* !CONFIG_SMP */
 149 static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
 150                     int cpu_idx)
 151 {
 152         u32 load;
 153         u64 now, now_idle, delta_time, delta_idle;
 154         struct time_in_idle *idle_time = &cpufreq_cdev->idle_time[cpu_idx];
 155
 156         now_idle = get_cpu_idle_time(cpu, &now, 0);
 157         delta_idle = now_idle - idle_time->time;
 158         delta_time = now - idle_time->timestamp;
 159
 160         if (delta_time <= delta_idle)
 161                 load = 0;
 162         else
 163                 load = div64_u64(100 * (delta_time - delta_idle), delta_time);
 164
 165         idle_time->time = now_idle;
 166         idle_time->timestamp = now;
 167
 168         return load;
 169 }
 170 #endif /* CONFIG_SMP */
 171
 172 /**
 173  * get_dynamic_power() - calculate the dynamic power
 174  * @cpufreq_cdev:       &cpufreq_cooling_device for this cdev
 175  * @freq:       current frequency
 176  *
 177  * Return: the dynamic power consumed by the cpus described by
 178  * @cpufreq_cdev.
 179  */
 180 static u32 get_dynamic_power(struct cpufreq_cooling_device *cpufreq_cdev,
 181                              unsigned long freq)
 182 {
 183         u32 raw_cpu_power;
 184
 185         raw_cpu_power = cpu_freq_to_power(cpufreq_cdev, freq);
 186         return (raw_cpu_power * cpufreq_cdev->last_load) / 100;
 187 }
 188
 189 /**
 190  * cpufreq_get_requested_power() - get the current power
 191  * @cdev:       &thermal_cooling_device pointer
 192  * @power:      pointer in which to store the resulting power
 193  *
 194  * Calculate the current power consumption of the cpus in milliwatts
 195  * and store it in @power.  This function should actually calculate
 196  * the requested power, but it's hard to get the frequency that
 197  * cpufreq would have assigned if there were no thermal limits.
 198  * Instead, we calculate the current power on the assumption that the
 199  * immediate future will look like the immediate past.
 200  *
 201  * We use the current frequency and the average load since this
 202  * function was last called.  In reality, there could have been
 203  * multiple opps since this function was last called and that affects
 204  * the load calculation.  While it's not perfectly accurate, this
 205  * simplification is good enough and works.  REVISIT this, as more
 206  * complex code may be needed if experiments show that it's not
 207  * accurate enough.
 208  *
 209  * Return: 0 on success, -E* if getting the static power failed.
 210  */
 211 static int cpufreq_get_requested_power(struct thermal_cooling_device *cdev,
 212                                        u32 *power)
 213 {
 214         unsigned long freq;
 215         int i = 0, cpu;
 216         u32 total_load = 0;
 217         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 218         struct cpufreq_policy *policy = cpufreq_cdev->policy;
 219         u32 *load_cpu = NULL;
 220
 221         freq = cpufreq_quick_get(policy->cpu);
 222
 223         if (trace_thermal_power_cpu_get_power_enabled()) {
 224                 u32 ncpus = cpumask_weight(policy->related_cpus);
 225
 226                 load_cpu = kcalloc(ncpus, sizeof(*load_cpu), GFP_KERNEL);
 227         }
 228
 229         for_each_cpu(cpu, policy->related_cpus) {
 230                 u32 load;
 231
 232                 if (cpu_online(cpu))
 233                         load = get_load(cpufreq_cdev, cpu, i);
 234                 else
 235                         load = 0;
 236
 237                 total_load += load;
 238                 if (load_cpu)
 239                         load_cpu[i] = load;
 240
 241                 i++;
 242         }
 243
 244         cpufreq_cdev->last_load = total_load;
 245
 246         *power = get_dynamic_power(cpufreq_cdev, freq);
 247
 248         if (load_cpu) {
 249                 trace_thermal_power_cpu_get_power(policy->related_cpus, freq,
 250                                                   load_cpu, i, *power);
 251
 252                 kfree(load_cpu);
 253         }
 254
 255         return 0;
 256 }
 257
 258 /**
 259  * cpufreq_state2power() - convert a cpu cdev state to power consumed
 260  * @cdev:       &thermal_cooling_device pointer
 261  * @state:      cooling device state to be converted
 262  * @power:      pointer in which to store the resulting power
 263  *
 264  * Convert cooling device state @state into power consumption in
 265  * milliwatts assuming 100% load.  Store the calculated power in
 266  * @power.
 267  *
 268  * Return: 0 on success, -EINVAL if the cooling device state could not
 269  * be converted into a frequency or other -E* if there was an error
 270  * when calculating the static power.
 271  */
 272 static int cpufreq_state2power(struct thermal_cooling_device *cdev,
 273                                unsigned long state, u32 *power)
 274 {
 275         unsigned int freq, num_cpus, idx;
 276         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 277
 278         /* Request state should be less than max_level */
 279         if (state > cpufreq_cdev->max_level)
 280                 return -EINVAL;
 281
 282         num_cpus = cpumask_weight(cpufreq_cdev->policy->cpus);
 283
 284         idx = cpufreq_cdev->max_level - state;
 285         freq = cpufreq_cdev->em->table[idx].frequency;
 286         *power = cpu_freq_to_power(cpufreq_cdev, freq) * num_cpus;
 287
 288         return 0;
 289 }
 290
 291 /**
 292  * cpufreq_power2state() - convert power to a cooling device state
 293  * @cdev:       &thermal_cooling_device pointer
 294  * @power:      power in milliwatts to be converted
 295  * @state:      pointer in which to store the resulting state
 296  *
 297  * Calculate a cooling device state for the cpus described by @cdev
 298  * that would allow them to consume at most @power mW and store it in
 299  * @state.  Note that this calculation depends on external factors
 300  * such as the cpu load or the current static power.  Calling this
 301  * function with the same power as input can yield different cooling
 302  * device states depending on those external factors.
 303  *
 304  * Return: 0 on success, -ENODEV if no cpus are online or -EINVAL if
 305  * the calculated frequency could not be converted to a valid state.
 306  * The latter should not happen unless the frequencies available to
 307  * cpufreq have changed since the initialization of the cpu cooling
 308  * device.
 309  */
 310 static int cpufreq_power2state(struct thermal_cooling_device *cdev,
 311                                u32 power, unsigned long *state)
 312 {
 313         unsigned int target_freq;
 314         u32 last_load, normalised_power;
 315         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 316         struct cpufreq_policy *policy = cpufreq_cdev->policy;
 317
 318         last_load = cpufreq_cdev->last_load ?: 1;
 319         normalised_power = (power * 100) / last_load;
 320         target_freq = cpu_power_to_freq(cpufreq_cdev, normalised_power);
 321
 322         *state = get_level(cpufreq_cdev, target_freq);
 323         trace_thermal_power_cpu_limit(policy->related_cpus, target_freq, *state,
 324                                       power);
 325         return 0;
 326 }
 327
 328 static inline bool em_is_sane(struct cpufreq_cooling_device *cpufreq_cdev,
 329                               struct em_perf_domain *em) {
 330         struct cpufreq_policy *policy;
 331         unsigned int nr_levels;
 332
 333         if (!em || em_is_artificial(em))
 334                 return false;
 335
 336         policy = cpufreq_cdev->policy;
 337         if (!cpumask_equal(policy->related_cpus, em_span_cpus(em))) {
 338                 pr_err("The span of pd %*pbl is misaligned with cpufreq policy %*pbl\n",
 339                         cpumask_pr_args(em_span_cpus(em)),
 340                         cpumask_pr_args(policy->related_cpus));
 341                 return false;
 342         }
 343
 344         nr_levels = cpufreq_cdev->max_level + 1;
 345         if (em_pd_nr_perf_states(em) != nr_levels) {
 346                 pr_err("The number of performance states in pd %*pbl (%u) doesn't match the number of cooling levels (%u)\n",
 347                         cpumask_pr_args(em_span_cpus(em)),
 348                         em_pd_nr_perf_states(em), nr_levels);
 349                 return false;
 350         }
 351
 352         return true;
 353 }
 354 #endif /* CONFIG_THERMAL_GOV_POWER_ALLOCATOR */
 355
 356 #ifdef CONFIG_SMP
 357 static inline int allocate_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
 358 {
 359         return 0;
 360 }
 361
 362 static inline void free_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
 363 {
 364 }
 365 #else
 366 static int allocate_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
 367 {
 368         unsigned int num_cpus = cpumask_weight(cpufreq_cdev->policy->related_cpus);
 369
 370         cpufreq_cdev->idle_time = kcalloc(num_cpus,
 371                                           sizeof(*cpufreq_cdev->idle_time),
 372                                           GFP_KERNEL);
 373         if (!cpufreq_cdev->idle_time)
 374                 return -ENOMEM;
 375
 376         return 0;
 377 }
 378
 379 static void free_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
 380 {
 381         kfree(cpufreq_cdev->idle_time);
 382         cpufreq_cdev->idle_time = NULL;
 383 }
 384 #endif /* CONFIG_SMP */
 385
 386 static unsigned int get_state_freq(struct cpufreq_cooling_device *cpufreq_cdev,
 387                                    unsigned long state)
 388 {
 389         struct cpufreq_policy *policy;
 390         unsigned long idx;
 391
 392 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
 393         /* Use the Energy Model table if available */
 394         if (cpufreq_cdev->em) {
 395                 idx = cpufreq_cdev->max_level - state;
 396                 return cpufreq_cdev->em->table[idx].frequency;
 397         }
 398 #endif
 399
 400         /* Otherwise, fallback on the CPUFreq table */
 401         policy = cpufreq_cdev->policy;
 402         if (policy->freq_table_sorted == CPUFREQ_TABLE_SORTED_ASCENDING)
 403                 idx = cpufreq_cdev->max_level - state;
 404         else
 405                 idx = state;
 406
 407         return policy->freq_table[idx].frequency;
 408 }
 409
 410 /* cpufreq cooling device callback functions are defined below */
 411
 412 /**
 413  * cpufreq_get_max_state - callback function to get the max cooling state.
 414  * @cdev: thermal cooling device pointer.
 415  * @state: fill this variable with the max cooling state.
 416  *
 417  * Callback for the thermal cooling device to return the cpufreq
 418  * max cooling state.
 419  *
 420  * Return: 0 on success, an error code otherwise.
 421  */
 422 static int cpufreq_get_max_state(struct thermal_cooling_device *cdev,
 423                                  unsigned long *state)
 424 {
 425         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 426
 427         *state = cpufreq_cdev->max_level;
 428         return 0;
 429 }
 430
 431 /**
 432  * cpufreq_get_cur_state - callback function to get the current cooling state.
 433  * @cdev: thermal cooling device pointer.
 434  * @state: fill this variable with the current cooling state.
 435  *
 436  * Callback for the thermal cooling device to return the cpufreq
 437  * current cooling state.
 438  *
 439  * Return: 0 on success, an error code otherwise.
 440  */
 441 static int cpufreq_get_cur_state(struct thermal_cooling_device *cdev,
 442                                  unsigned long *state)
 443 {
 444         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 445
 446         *state = cpufreq_cdev->cpufreq_state;
 447
 448         return 0;
 449 }
 450
 451 /**
 452  * cpufreq_set_cur_state - callback function to set the current cooling state.
 453  * @cdev: thermal cooling device pointer.
 454  * @state: set this variable to the current cooling state.
 455  *
 456  * Callback for the thermal cooling device to change the cpufreq
 457  * current cooling state.
 458  *
 459  * Return: 0 on success, an error code otherwise.
 460  */
 461 static int cpufreq_set_cur_state(struct thermal_cooling_device *cdev,
 462                                  unsigned long state)
 463 {
 464         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 465         struct cpumask *cpus;
 466         unsigned int frequency;
 467         int ret;
 468
 469         /* Request state should be less than max_level */
 470         if (state > cpufreq_cdev->max_level)
 471                 return -EINVAL;
 472
 473         /* Check if the old cooling action is same as new cooling action */
 474         if (cpufreq_cdev->cpufreq_state == state)
 475                 return 0;
 476
 477         frequency = get_state_freq(cpufreq_cdev, state);
 478
 479         ret = freq_qos_update_request(&cpufreq_cdev->qos_req, frequency);
 480         if (ret >= 0) {
 481                 cpufreq_cdev->cpufreq_state = state;
 482                 cpus = cpufreq_cdev->policy->related_cpus;
 483                 arch_update_thermal_pressure(cpus, frequency);
 484                 ret = 0;
 485         }
 486
 487         return ret;
 488 }
 489
 490 /**
 491  * __cpufreq_cooling_register - helper function to create cpufreq cooling device
 492  * @np: a valid struct device_node to the cooling device device tree node
 493  * @policy: cpufreq policy
 494  * Normally this should be same as cpufreq policy->related_cpus.
 495  * @em: Energy Model of the cpufreq policy
 496  *
 497  * This interface function registers the cpufreq cooling device with the name
 498  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 499  * cooling devices. It also gives the opportunity to link the cooling device
 500  * with a device tree node, in order to bind it via the thermal DT code.
 501  *
 502  * Return: a valid struct thermal_cooling_device pointer on success,
 503  * on failure, it returns a corresponding ERR_PTR().
 504  */
 505 static struct thermal_cooling_device *
 506 __cpufreq_cooling_register(struct device_node *np,
 507                         struct cpufreq_policy *policy,
 508                         struct em_perf_domain *em)
 509 {
 510         struct thermal_cooling_device *cdev;
 511         struct cpufreq_cooling_device *cpufreq_cdev;
 512         unsigned int i;
 513         struct device *dev;
 514         int ret;
 515         struct thermal_cooling_device_ops *cooling_ops;
 516         char *name;
 517
 518         dev = get_cpu_device(policy->cpu);
 519         if (unlikely(!dev)) {
 520                 pr_warn("No cpu device for cpu %d\n", policy->cpu);
 521                 return ERR_PTR(-ENODEV);
 522         }
 523
 524         if (IS_ERR_OR_NULL(policy)) {
 525                 pr_err("%s: cpufreq policy isn't valid: %p\n", __func__, policy);
 526                 return ERR_PTR(-EINVAL);
 527         }
 528
 529         i = cpufreq_table_count_valid_entries(policy);
 530         if (!i) {
 531                 pr_debug("%s: CPUFreq table not found or has no valid entries\n",
 532                          __func__);
 533                 return ERR_PTR(-ENODEV);
 534         }
 535
 536         cpufreq_cdev = kzalloc(sizeof(*cpufreq_cdev), GFP_KERNEL);
 537         if (!cpufreq_cdev)
 538                 return ERR_PTR(-ENOMEM);
 539
 540         cpufreq_cdev->policy = policy;
 541
 542         ret = allocate_idle_time(cpufreq_cdev);
 543         if (ret) {
 544                 cdev = ERR_PTR(ret);
 545                 goto free_cdev;
 546         }
 547
 548         /* max_level is an index, not a counter */
 549         cpufreq_cdev->max_level = i - 1;
 550
 551         cooling_ops = &cpufreq_cdev->cooling_ops;
 552         cooling_ops->get_max_state = cpufreq_get_max_state;
 553         cooling_ops->get_cur_state = cpufreq_get_cur_state;
 554         cooling_ops->set_cur_state = cpufreq_set_cur_state;
 555
 556 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
 557         if (em_is_sane(cpufreq_cdev, em)) {
 558                 cpufreq_cdev->em = em;
 559                 cooling_ops->get_requested_power = cpufreq_get_requested_power;
 560                 cooling_ops->state2power = cpufreq_state2power;
 561                 cooling_ops->power2state = cpufreq_power2state;
 562         } else
 563 #endif
 564         if (policy->freq_table_sorted == CPUFREQ_TABLE_UNSORTED) {
 565                 pr_err("%s: unsorted frequency tables are not supported\n",
 566                        __func__);
 567                 cdev = ERR_PTR(-EINVAL);
 568                 goto free_idle_time;
 569         }
 570
 571         ret = freq_qos_add_request(&policy->constraints,
 572                                    &cpufreq_cdev->qos_req, FREQ_QOS_MAX,
 573                                    get_state_freq(cpufreq_cdev, 0));
 574         if (ret < 0) {
 575                 pr_err("%s: Failed to add freq constraint (%d)\n", __func__,
 576                        ret);
 577                 cdev = ERR_PTR(ret);
 578                 goto free_idle_time;
 579         }
 580
 581         cdev = ERR_PTR(-ENOMEM);
 582         name = kasprintf(GFP_KERNEL, "cpufreq-%s", dev_name(dev));
 583         if (!name)
 584                 goto remove_qos_req;
 585
 586         cdev = thermal_of_cooling_device_register(np, name, cpufreq_cdev,
 587                                                   cooling_ops);
 588         kfree(name);
 589
 590         if (IS_ERR(cdev))
 591                 goto remove_qos_req;
 592
 593         return cdev;
 594
 595 remove_qos_req:
 596         freq_qos_remove_request(&cpufreq_cdev->qos_req);
 597 free_idle_time:
 598         free_idle_time(cpufreq_cdev);
 599 free_cdev:
 600         kfree(cpufreq_cdev);
 601         return cdev;
 602 }
 603
 604 /**
 605  * cpufreq_cooling_register - function to create cpufreq cooling device.
 606  * @policy: cpufreq policy
 607  *
 608  * This interface function registers the cpufreq cooling device with the name
 609  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 610  * cooling devices.
 611  *
 612  * Return: a valid struct thermal_cooling_device pointer on success,
 613  * on failure, it returns a corresponding ERR_PTR().
 614  */
 615 struct thermal_cooling_device *
 616 cpufreq_cooling_register(struct cpufreq_policy *policy)
 617 {
 618         return __cpufreq_cooling_register(NULL, policy, NULL);
 619 }
 620 EXPORT_SYMBOL_GPL(cpufreq_cooling_register);
 621
 622 /**
 623  * of_cpufreq_cooling_register - function to create cpufreq cooling device.
 624  * @policy: cpufreq policy
 625  *
 626  * This interface function registers the cpufreq cooling device with the name
 627  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 628  * cooling devices. Using this API, the cpufreq cooling device will be
 629  * linked to the device tree node provided.
 630  *
 631  * Using this function, the cooling device will implement the power
 632  * extensions by using a simple cpu power model.  The cpus must have
 633  * registered their OPPs using the OPP library.
 634  *
 635  * It also takes into account, if property present in policy CPU node, the
 636  * static power consumed by the cpu.
 637  *
 638  * Return: a valid struct thermal_cooling_device pointer on success,
 639  * and NULL on failure.
 640  */
 641 struct thermal_cooling_device *
 642 of_cpufreq_cooling_register(struct cpufreq_policy *policy)
 643 {
 644         struct device_node *np = of_get_cpu_node(policy->cpu, NULL);
 645         struct thermal_cooling_device *cdev = NULL;
 646
 647         if (!np) {
 648                 pr_err("cpufreq_cooling: OF node not available for cpu%d\n",
 649                        policy->cpu);
 650                 return NULL;
 651         }
 652
 653         if (of_find_property(np, "#cooling-cells", NULL)) {
 654                 struct em_perf_domain *em = em_cpu_get(policy->cpu);
 655
 656                 cdev = __cpufreq_cooling_register(np, policy, em);
 657                 if (IS_ERR(cdev)) {
 658                         pr_err("cpufreq_cooling: cpu%d failed to register as cooling device: %ld\n",
 659                                policy->cpu, PTR_ERR(cdev));
 660                         cdev = NULL;
 661                 }
 662         }
 663
 664         of_node_put(np);
 665         return cdev;
 666 }
 667 EXPORT_SYMBOL_GPL(of_cpufreq_cooling_register);
 668
 669 /**
 670  * cpufreq_cooling_unregister - function to remove cpufreq cooling device.
 671  * @cdev: thermal cooling device pointer.
 672  *
 673  * This interface function unregisters the "thermal-cpufreq-%x" cooling device.
 674  */
 675 void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
 676 {
 677         struct cpufreq_cooling_device *cpufreq_cdev;
 678
 679         if (!cdev)
 680                 return;
 681
 682         cpufreq_cdev = cdev->devdata;
 683
 684         thermal_cooling_device_unregister(cdev);
 685         freq_qos_remove_request(&cpufreq_cdev->qos_req);
 686         free_idle_time(cpufreq_cdev);
 687         kfree(cpufreq_cdev);
 688 }
 689 EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister);