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 max = arch_scale_cpu_capacity(cpu);
 141         unsigned long util;
 142
 143         util = sched_cpu_util(cpu, max);
 144         return (util * 100) / max;
 145 }
 146 #else /* !CONFIG_SMP */
 147 static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
 148                     int cpu_idx)
 149 {
 150         u32 load;
 151         u64 now, now_idle, delta_time, delta_idle;
 152         struct time_in_idle *idle_time = &cpufreq_cdev->idle_time[cpu_idx];
 153
 154         now_idle = get_cpu_idle_time(cpu, &now, 0);
 155         delta_idle = now_idle - idle_time->time;
 156         delta_time = now - idle_time->timestamp;
 157
 158         if (delta_time <= delta_idle)
 159                 load = 0;
 160         else
 161                 load = div64_u64(100 * (delta_time - delta_idle), delta_time);
 162
 163         idle_time->time = now_idle;
 164         idle_time->timestamp = now;
 165
 166         return load;
 167 }
 168 #endif /* CONFIG_SMP */
 169
 170 /**
 171  * get_dynamic_power() - calculate the dynamic power
 172  * @cpufreq_cdev:       &cpufreq_cooling_device for this cdev
 173  * @freq:       current frequency
 174  *
 175  * Return: the dynamic power consumed by the cpus described by
 176  * @cpufreq_cdev.
 177  */
 178 static u32 get_dynamic_power(struct cpufreq_cooling_device *cpufreq_cdev,
 179                              unsigned long freq)
 180 {
 181         u32 raw_cpu_power;
 182
 183         raw_cpu_power = cpu_freq_to_power(cpufreq_cdev, freq);
 184         return (raw_cpu_power * cpufreq_cdev->last_load) / 100;
 185 }
 186
 187 /**
 188  * cpufreq_get_requested_power() - get the current power
 189  * @cdev:       &thermal_cooling_device pointer
 190  * @power:      pointer in which to store the resulting power
 191  *
 192  * Calculate the current power consumption of the cpus in milliwatts
 193  * and store it in @power.  This function should actually calculate
 194  * the requested power, but it's hard to get the frequency that
 195  * cpufreq would have assigned if there were no thermal limits.
 196  * Instead, we calculate the current power on the assumption that the
 197  * immediate future will look like the immediate past.
 198  *
 199  * We use the current frequency and the average load since this
 200  * function was last called.  In reality, there could have been
 201  * multiple opps since this function was last called and that affects
 202  * the load calculation.  While it's not perfectly accurate, this
 203  * simplification is good enough and works.  REVISIT this, as more
 204  * complex code may be needed if experiments show that it's not
 205  * accurate enough.
 206  *
 207  * Return: 0 on success, -E* if getting the static power failed.
 208  */
 209 static int cpufreq_get_requested_power(struct thermal_cooling_device *cdev,
 210                                        u32 *power)
 211 {
 212         unsigned long freq;
 213         int i = 0, cpu;
 214         u32 total_load = 0;
 215         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 216         struct cpufreq_policy *policy = cpufreq_cdev->policy;
 217         u32 *load_cpu = NULL;
 218
 219         freq = cpufreq_quick_get(policy->cpu);
 220
 221         if (trace_thermal_power_cpu_get_power_enabled()) {
 222                 u32 ncpus = cpumask_weight(policy->related_cpus);
 223
 224                 load_cpu = kcalloc(ncpus, sizeof(*load_cpu), GFP_KERNEL);
 225         }
 226
 227         for_each_cpu(cpu, policy->related_cpus) {
 228                 u32 load;
 229
 230                 if (cpu_online(cpu))
 231                         load = get_load(cpufreq_cdev, cpu, i);
 232                 else
 233                         load = 0;
 234
 235                 total_load += load;
 236                 if (load_cpu)
 237                         load_cpu[i] = load;
 238
 239                 i++;
 240         }
 241
 242         cpufreq_cdev->last_load = total_load;
 243
 244         *power = get_dynamic_power(cpufreq_cdev, freq);
 245
 246         if (load_cpu) {
 247                 trace_thermal_power_cpu_get_power(policy->related_cpus, freq,
 248                                                   load_cpu, i, *power);
 249
 250                 kfree(load_cpu);
 251         }
 252
 253         return 0;
 254 }
 255
 256 /**
 257  * cpufreq_state2power() - convert a cpu cdev state to power consumed
 258  * @cdev:       &thermal_cooling_device pointer
 259  * @state:      cooling device state to be converted
 260  * @power:      pointer in which to store the resulting power
 261  *
 262  * Convert cooling device state @state into power consumption in
 263  * milliwatts assuming 100% load.  Store the calculated power in
 264  * @power.
 265  *
 266  * Return: 0 on success, -EINVAL if the cooling device state could not
 267  * be converted into a frequency or other -E* if there was an error
 268  * when calculating the static power.
 269  */
 270 static int cpufreq_state2power(struct thermal_cooling_device *cdev,
 271                                unsigned long state, u32 *power)
 272 {
 273         unsigned int freq, num_cpus, idx;
 274         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 275
 276         /* Request state should be less than max_level */
 277         if (state > cpufreq_cdev->max_level)
 278                 return -EINVAL;
 279
 280         num_cpus = cpumask_weight(cpufreq_cdev->policy->cpus);
 281
 282         idx = cpufreq_cdev->max_level - state;
 283         freq = cpufreq_cdev->em->table[idx].frequency;
 284         *power = cpu_freq_to_power(cpufreq_cdev, freq) * num_cpus;
 285
 286         return 0;
 287 }
 288
 289 /**
 290  * cpufreq_power2state() - convert power to a cooling device state
 291  * @cdev:       &thermal_cooling_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                                u32 power, 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, em_span_cpus(em))) {
 336                 pr_err("The span of pd %*pbl is misaligned with cpufreq policy %*pbl\n",
 337                         cpumask_pr_args(em_span_cpus(em)),
 338                         cpumask_pr_args(policy->related_cpus));
 339                 return false;
 340         }
 341
 342         nr_levels = cpufreq_cdev->max_level + 1;
 343         if (em_pd_nr_perf_states(em) != nr_levels) {
 344                 pr_err("The number of performance states in pd %*pbl (%u) doesn't match the number of cooling levels (%u)\n",
 345                         cpumask_pr_args(em_span_cpus(em)),
 346                         em_pd_nr_perf_states(em), nr_levels);
 347                 return false;
 348         }
 349
 350         return true;
 351 }
 352 #endif /* CONFIG_THERMAL_GOV_POWER_ALLOCATOR */
 353
 354 #ifdef CONFIG_SMP
 355 static inline int allocate_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
 356 {
 357         return 0;
 358 }
 359
 360 static inline void free_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
 361 {
 362 }
 363 #else
 364 static int allocate_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
 365 {
 366         unsigned int num_cpus = cpumask_weight(cpufreq_cdev->policy->related_cpus);
 367
 368         cpufreq_cdev->idle_time = kcalloc(num_cpus,
 369                                           sizeof(*cpufreq_cdev->idle_time),
 370                                           GFP_KERNEL);
 371         if (!cpufreq_cdev->idle_time)
 372                 return -ENOMEM;
 373
 374         return 0;
 375 }
 376
 377 static void free_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
 378 {
 379         kfree(cpufreq_cdev->idle_time);
 380         cpufreq_cdev->idle_time = NULL;
 381 }
 382 #endif /* CONFIG_SMP */
 383
 384 static unsigned int get_state_freq(struct cpufreq_cooling_device *cpufreq_cdev,
 385                                    unsigned long state)
 386 {
 387         struct cpufreq_policy *policy;
 388         unsigned long idx;
 389
 390 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
 391         /* Use the Energy Model table if available */
 392         if (cpufreq_cdev->em) {
 393                 idx = cpufreq_cdev->max_level - state;
 394                 return cpufreq_cdev->em->table[idx].frequency;
 395         }
 396 #endif
 397
 398         /* Otherwise, fallback on the CPUFreq table */
 399         policy = cpufreq_cdev->policy;
 400         if (policy->freq_table_sorted == CPUFREQ_TABLE_SORTED_ASCENDING)
 401                 idx = cpufreq_cdev->max_level - state;
 402         else
 403                 idx = state;
 404
 405         return policy->freq_table[idx].frequency;
 406 }
 407
 408 /* cpufreq cooling device callback functions are defined below */
 409
 410 /**
 411  * cpufreq_get_max_state - callback function to get the max cooling state.
 412  * @cdev: thermal cooling device pointer.
 413  * @state: fill this variable with the max cooling state.
 414  *
 415  * Callback for the thermal cooling device to return the cpufreq
 416  * max cooling state.
 417  *
 418  * Return: 0 on success, an error code otherwise.
 419  */
 420 static int cpufreq_get_max_state(struct thermal_cooling_device *cdev,
 421                                  unsigned long *state)
 422 {
 423         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 424
 425         *state = cpufreq_cdev->max_level;
 426         return 0;
 427 }
 428
 429 /**
 430  * cpufreq_get_cur_state - callback function to get the current cooling state.
 431  * @cdev: thermal cooling device pointer.
 432  * @state: fill this variable with the current cooling state.
 433  *
 434  * Callback for the thermal cooling device to return the cpufreq
 435  * current cooling state.
 436  *
 437  * Return: 0 on success, an error code otherwise.
 438  */
 439 static int cpufreq_get_cur_state(struct thermal_cooling_device *cdev,
 440                                  unsigned long *state)
 441 {
 442         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 443
 444         *state = cpufreq_cdev->cpufreq_state;
 445
 446         return 0;
 447 }
 448
 449 /**
 450  * cpufreq_set_cur_state - callback function to set the current cooling state.
 451  * @cdev: thermal cooling device pointer.
 452  * @state: set this variable to the current cooling state.
 453  *
 454  * Callback for the thermal cooling device to change the cpufreq
 455  * current cooling state.
 456  *
 457  * Return: 0 on success, an error code otherwise.
 458  */
 459 static int cpufreq_set_cur_state(struct thermal_cooling_device *cdev,
 460                                  unsigned long state)
 461 {
 462         struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
 463         struct cpumask *cpus;
 464         unsigned int frequency;
 465         unsigned long max_capacity, capacity;
 466         int ret;
 467
 468         /* Request state should be less than max_level */
 469         if (state > cpufreq_cdev->max_level)
 470                 return -EINVAL;
 471
 472         /* Check if the old cooling action is same as new cooling action */
 473         if (cpufreq_cdev->cpufreq_state == state)
 474                 return 0;
 475
 476         frequency = get_state_freq(cpufreq_cdev, state);
 477
 478         ret = freq_qos_update_request(&cpufreq_cdev->qos_req, frequency);
 479         if (ret >= 0) {
 480                 cpufreq_cdev->cpufreq_state = state;
 481                 cpus = cpufreq_cdev->policy->related_cpus;
 482                 max_capacity = arch_scale_cpu_capacity(cpumask_first(cpus));
 483                 capacity = frequency * max_capacity;
 484                 capacity /= cpufreq_cdev->policy->cpuinfo.max_freq;
 485                 arch_set_thermal_pressure(cpus, max_capacity - capacity);
 486                 ret = 0;
 487         }
 488
 489         return ret;
 490 }
 491
 492 /* Bind cpufreq callbacks to thermal cooling device ops */
 493
 494 static struct thermal_cooling_device_ops cpufreq_cooling_ops = {
 495         .get_max_state          = cpufreq_get_max_state,
 496         .get_cur_state          = cpufreq_get_cur_state,
 497         .set_cur_state          = cpufreq_set_cur_state,
 498 };
 499
 500 /**
 501  * __cpufreq_cooling_register - helper function to create cpufreq cooling device
 502  * @np: a valid struct device_node to the cooling device device tree node
 503  * @policy: cpufreq policy
 504  * Normally this should be same as cpufreq policy->related_cpus.
 505  * @em: Energy Model of the cpufreq policy
 506  *
 507  * This interface function registers the cpufreq cooling device with the name
 508  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 509  * cooling devices. It also gives the opportunity to link the cooling device
 510  * with a device tree node, in order to bind it via the thermal DT code.
 511  *
 512  * Return: a valid struct thermal_cooling_device pointer on success,
 513  * on failure, it returns a corresponding ERR_PTR().
 514  */
 515 static struct thermal_cooling_device *
 516 __cpufreq_cooling_register(struct device_node *np,
 517                         struct cpufreq_policy *policy,
 518                         struct em_perf_domain *em)
 519 {
 520         struct thermal_cooling_device *cdev;
 521         struct cpufreq_cooling_device *cpufreq_cdev;
 522         unsigned int i;
 523         struct device *dev;
 524         int ret;
 525         struct thermal_cooling_device_ops *cooling_ops;
 526         char *name;
 527
 528         dev = get_cpu_device(policy->cpu);
 529         if (unlikely(!dev)) {
 530                 pr_warn("No cpu device for cpu %d\n", policy->cpu);
 531                 return ERR_PTR(-ENODEV);
 532         }
 533
 534         if (IS_ERR_OR_NULL(policy)) {
 535                 pr_err("%s: cpufreq policy isn't valid: %p\n", __func__, policy);
 536                 return ERR_PTR(-EINVAL);
 537         }
 538
 539         i = cpufreq_table_count_valid_entries(policy);
 540         if (!i) {
 541                 pr_debug("%s: CPUFreq table not found or has no valid entries\n",
 542                          __func__);
 543                 return ERR_PTR(-ENODEV);
 544         }
 545
 546         cpufreq_cdev = kzalloc(sizeof(*cpufreq_cdev), GFP_KERNEL);
 547         if (!cpufreq_cdev)
 548                 return ERR_PTR(-ENOMEM);
 549
 550         cpufreq_cdev->policy = policy;
 551
 552         ret = allocate_idle_time(cpufreq_cdev);
 553         if (ret) {
 554                 cdev = ERR_PTR(ret);
 555                 goto free_cdev;
 556         }
 557
 558         /* max_level is an index, not a counter */
 559         cpufreq_cdev->max_level = i - 1;
 560
 561         cooling_ops = &cpufreq_cooling_ops;
 562
 563 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
 564         if (em_is_sane(cpufreq_cdev, em)) {
 565                 cpufreq_cdev->em = em;
 566                 cooling_ops->get_requested_power = cpufreq_get_requested_power;
 567                 cooling_ops->state2power = cpufreq_state2power;
 568                 cooling_ops->power2state = cpufreq_power2state;
 569         } else
 570 #endif
 571         if (policy->freq_table_sorted == CPUFREQ_TABLE_UNSORTED) {
 572                 pr_err("%s: unsorted frequency tables are not supported\n",
 573                        __func__);
 574                 cdev = ERR_PTR(-EINVAL);
 575                 goto free_idle_time;
 576         }
 577
 578         ret = freq_qos_add_request(&policy->constraints,
 579                                    &cpufreq_cdev->qos_req, FREQ_QOS_MAX,
 580                                    get_state_freq(cpufreq_cdev, 0));
 581         if (ret < 0) {
 582                 pr_err("%s: Failed to add freq constraint (%d)\n", __func__,
 583                        ret);
 584                 cdev = ERR_PTR(ret);
 585                 goto free_idle_time;
 586         }
 587
 588         cdev = ERR_PTR(-ENOMEM);
 589         name = kasprintf(GFP_KERNEL, "cpufreq-%s", dev_name(dev));
 590         if (!name)
 591                 goto remove_qos_req;
 592
 593         cdev = thermal_of_cooling_device_register(np, name, cpufreq_cdev,
 594                                                   cooling_ops);
 595         kfree(name);
 596
 597         if (IS_ERR(cdev))
 598                 goto remove_qos_req;
 599
 600         return cdev;
 601
 602 remove_qos_req:
 603         freq_qos_remove_request(&cpufreq_cdev->qos_req);
 604 free_idle_time:
 605         free_idle_time(cpufreq_cdev);
 606 free_cdev:
 607         kfree(cpufreq_cdev);
 608         return cdev;
 609 }
 610
 611 /**
 612  * cpufreq_cooling_register - function to create cpufreq cooling device.
 613  * @policy: cpufreq policy
 614  *
 615  * This interface function registers the cpufreq cooling device with the name
 616  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 617  * cooling devices.
 618  *
 619  * Return: a valid struct thermal_cooling_device pointer on success,
 620  * on failure, it returns a corresponding ERR_PTR().
 621  */
 622 struct thermal_cooling_device *
 623 cpufreq_cooling_register(struct cpufreq_policy *policy)
 624 {
 625         return __cpufreq_cooling_register(NULL, policy, NULL);
 626 }
 627 EXPORT_SYMBOL_GPL(cpufreq_cooling_register);
 628
 629 /**
 630  * of_cpufreq_cooling_register - function to create cpufreq cooling device.
 631  * @policy: cpufreq policy
 632  *
 633  * This interface function registers the cpufreq cooling device with the name
 634  * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
 635  * cooling devices. Using this API, the cpufreq cooling device will be
 636  * linked to the device tree node provided.
 637  *
 638  * Using this function, the cooling device will implement the power
 639  * extensions by using a simple cpu power model.  The cpus must have
 640  * registered their OPPs using the OPP library.
 641  *
 642  * It also takes into account, if property present in policy CPU node, the
 643  * static power consumed by the cpu.
 644  *
 645  * Return: a valid struct thermal_cooling_device pointer on success,
 646  * and NULL on failure.
 647  */
 648 struct thermal_cooling_device *
 649 of_cpufreq_cooling_register(struct cpufreq_policy *policy)
 650 {
 651         struct device_node *np = of_get_cpu_node(policy->cpu, NULL);
 652         struct thermal_cooling_device *cdev = NULL;
 653
 654         if (!np) {
 655                 pr_err("cpufreq_cooling: OF node not available for cpu%d\n",
 656                        policy->cpu);
 657                 return NULL;
 658         }
 659
 660         if (of_find_property(np, "#cooling-cells", NULL)) {
 661                 struct em_perf_domain *em = em_cpu_get(policy->cpu);
 662
 663                 cdev = __cpufreq_cooling_register(np, policy, em);
 664                 if (IS_ERR(cdev)) {
 665                         pr_err("cpufreq_cooling: cpu%d failed to register as cooling device: %ld\n",
 666                                policy->cpu, PTR_ERR(cdev));
 667                         cdev = NULL;
 668                 }
 669         }
 670
 671         of_node_put(np);
 672         return cdev;
 673 }
 674 EXPORT_SYMBOL_GPL(of_cpufreq_cooling_register);
 675
 676 /**
 677  * cpufreq_cooling_unregister - function to remove cpufreq cooling device.
 678  * @cdev: thermal cooling device pointer.
 679  *
 680  * This interface function unregisters the "thermal-cpufreq-%x" cooling device.
 681  */
 682 void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
 683 {
 684         struct cpufreq_cooling_device *cpufreq_cdev;
 685
 686         if (!cdev)
 687                 return;
 688
 689         cpufreq_cdev = cdev->devdata;
 690
 691         thermal_cooling_device_unregister(cdev);
 692         freq_qos_remove_request(&cpufreq_cdev->qos_req);
 693         free_idle_time(cpufreq_cdev);
 694         kfree(cpufreq_cdev);
 695 }
 696 EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister);