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