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