1 // SPDX-License-Identifier: GPL-2.0
3 * linux/drivers/thermal/cpu_cooling.c
5 * Copyright (C) 2012 Samsung Electronics Co., Ltd(http://www.samsung.com)
7 * Copyright (C) 2012-2018 Linaro Limited.
9 * Authors: Amit Daniel <amit.kachhap@linaro.org>
10 * Viresh Kumar <viresh.kumar@linaro.org>
13 #include <linux/module.h>
14 #include <linux/thermal.h>
15 #include <linux/cpufreq.h>
16 #include <linux/err.h>
17 #include <linux/idr.h>
18 #include <linux/pm_opp.h>
19 #include <linux/pm_qos.h>
20 #include <linux/slab.h>
21 #include <linux/cpu.h>
22 #include <linux/cpu_cooling.h>
24 #include <trace/events/thermal.h>
27 * Cooling state <-> CPUFreq frequency
29 * Cooling states are translated to frequencies throughout this driver and this
30 * is the relation between them.
32 * Highest cooling state corresponds to lowest possible frequency.
35 * level 0 --> 1st Max Freq
36 * level 1 --> 2nd Max Freq
41 * struct freq_table - frequency table along with power entries
42 * @frequency: frequency in KHz
45 * This structure is built when the cooling device registers and helps
46 * in translating frequency to power and vice versa.
54 * struct time_in_idle - Idle time stats
55 * @time: previous reading of the absolute time that this cpu was idle
56 * @timestamp: wall time of the last invocation of get_cpu_idle_time_us()
64 * struct cpufreq_cooling_device - data for cooling device with cpufreq
65 * @id: unique integer value corresponding to each cpufreq_cooling_device
67 * @last_load: load measured by the latest call to cpufreq_get_requested_power()
68 * @cpufreq_state: integer value representing the current state of cpufreq
70 * @max_level: maximum cooling level. One less than total number of valid
71 * cpufreq frequencies.
72 * @freq_table: Freq table in descending order of frequencies
73 * @cdev: thermal_cooling_device pointer to keep track of the
74 * registered cooling device.
75 * @policy: cpufreq policy.
76 * @node: list_head to link all cpufreq_cooling_device together.
77 * @idle_time: idle time stats
79 * This structure is required for keeping information of each registered
80 * cpufreq_cooling_device.
82 struct cpufreq_cooling_device {
85 unsigned int cpufreq_state;
86 unsigned int max_level;
87 struct freq_table *freq_table; /* In descending order */
88 struct cpufreq_policy *policy;
89 struct list_head node;
90 struct time_in_idle *idle_time;
91 struct freq_qos_request qos_req;
94 static DEFINE_IDA(cpufreq_ida);
95 static DEFINE_MUTEX(cooling_list_lock);
96 static LIST_HEAD(cpufreq_cdev_list);
98 /* Below code defines functions to be used for cpufreq as cooling device */
101 * get_level: Find the level for a particular frequency
102 * @cpufreq_cdev: cpufreq_cdev for which the property is required
105 * Return: level corresponding to the frequency.
107 static unsigned long get_level(struct cpufreq_cooling_device *cpufreq_cdev,
110 struct freq_table *freq_table = cpufreq_cdev->freq_table;
113 for (level = 1; level <= cpufreq_cdev->max_level; level++)
114 if (freq > freq_table[level].frequency)
121 * update_freq_table() - Update the freq table with power numbers
122 * @cpufreq_cdev: the cpufreq cooling device in which to update the table
123 * @capacitance: dynamic power coefficient for these cpus
125 * Update the freq table with power numbers. This table will be used in
126 * cpu_power_to_freq() and cpu_freq_to_power() to convert between power and
127 * frequency efficiently. Power is stored in mW, frequency in KHz. The
128 * resulting table is in descending order.
130 * Return: 0 on success, -EINVAL if there are no OPPs for any CPUs,
131 * or -ENOMEM if we run out of memory.
133 static int update_freq_table(struct cpufreq_cooling_device *cpufreq_cdev,
136 struct freq_table *freq_table = cpufreq_cdev->freq_table;
137 struct dev_pm_opp *opp;
138 struct device *dev = NULL;
139 int num_opps = 0, cpu = cpufreq_cdev->policy->cpu, i;
141 dev = get_cpu_device(cpu);
142 if (unlikely(!dev)) {
143 pr_warn("No cpu device for cpu %d\n", cpu);
147 num_opps = dev_pm_opp_get_opp_count(dev);
152 * The cpufreq table is also built from the OPP table and so the count
155 if (num_opps != cpufreq_cdev->max_level + 1) {
156 dev_warn(dev, "Number of OPPs not matching with max_levels\n");
160 for (i = 0; i <= cpufreq_cdev->max_level; i++) {
161 unsigned long freq = freq_table[i].frequency * 1000;
162 u32 freq_mhz = freq_table[i].frequency / 1000;
167 * Find ceil frequency as 'freq' may be slightly lower than OPP
168 * freq due to truncation while converting to kHz.
170 opp = dev_pm_opp_find_freq_ceil(dev, &freq);
172 dev_err(dev, "failed to get opp for %lu frequency\n",
177 voltage_mv = dev_pm_opp_get_voltage(opp) / 1000;
181 * Do the multiplication with MHz and millivolt so as
184 power = (u64)capacitance * freq_mhz * voltage_mv * voltage_mv;
185 do_div(power, 1000000000);
187 /* power is stored in mW */
188 freq_table[i].power = power;
194 static u32 cpu_freq_to_power(struct cpufreq_cooling_device *cpufreq_cdev,
198 struct freq_table *freq_table = cpufreq_cdev->freq_table;
200 for (i = 1; i <= cpufreq_cdev->max_level; i++)
201 if (freq > freq_table[i].frequency)
204 return freq_table[i - 1].power;
207 static u32 cpu_power_to_freq(struct cpufreq_cooling_device *cpufreq_cdev,
211 struct freq_table *freq_table = cpufreq_cdev->freq_table;
213 for (i = 0; i < cpufreq_cdev->max_level; i++)
214 if (power >= freq_table[i].power)
217 return freq_table[i].frequency;
221 * get_load() - get load for a cpu since last updated
222 * @cpufreq_cdev: &struct cpufreq_cooling_device for this cpu
224 * @cpu_idx: index of the cpu in time_in_idle*
226 * Return: The average load of cpu @cpu in percentage since this
227 * function was last called.
229 static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
233 u64 now, now_idle, delta_time, delta_idle;
234 struct time_in_idle *idle_time = &cpufreq_cdev->idle_time[cpu_idx];
236 now_idle = get_cpu_idle_time(cpu, &now, 0);
237 delta_idle = now_idle - idle_time->time;
238 delta_time = now - idle_time->timestamp;
240 if (delta_time <= delta_idle)
243 load = div64_u64(100 * (delta_time - delta_idle), delta_time);
245 idle_time->time = now_idle;
246 idle_time->timestamp = now;
252 * get_dynamic_power() - calculate the dynamic power
253 * @cpufreq_cdev: &cpufreq_cooling_device for this cdev
254 * @freq: current frequency
256 * Return: the dynamic power consumed by the cpus described by
259 static u32 get_dynamic_power(struct cpufreq_cooling_device *cpufreq_cdev,
264 raw_cpu_power = cpu_freq_to_power(cpufreq_cdev, freq);
265 return (raw_cpu_power * cpufreq_cdev->last_load) / 100;
268 /* cpufreq cooling device callback functions are defined below */
271 * cpufreq_get_max_state - callback function to get the max cooling state.
272 * @cdev: thermal cooling device pointer.
273 * @state: fill this variable with the max cooling state.
275 * Callback for the thermal cooling device to return the cpufreq
278 * Return: 0 on success, an error code otherwise.
280 static int cpufreq_get_max_state(struct thermal_cooling_device *cdev,
281 unsigned long *state)
283 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
285 *state = cpufreq_cdev->max_level;
290 * cpufreq_get_cur_state - callback function to get the current cooling state.
291 * @cdev: thermal cooling device pointer.
292 * @state: fill this variable with the current cooling state.
294 * Callback for the thermal cooling device to return the cpufreq
295 * current cooling state.
297 * Return: 0 on success, an error code otherwise.
299 static int cpufreq_get_cur_state(struct thermal_cooling_device *cdev,
300 unsigned long *state)
302 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
304 *state = cpufreq_cdev->cpufreq_state;
310 * cpufreq_set_cur_state - callback function to set the current cooling state.
311 * @cdev: thermal cooling device pointer.
312 * @state: set this variable to the current cooling state.
314 * Callback for the thermal cooling device to change the cpufreq
315 * current cooling state.
317 * Return: 0 on success, an error code otherwise.
319 static int cpufreq_set_cur_state(struct thermal_cooling_device *cdev,
322 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
325 /* Request state should be less than max_level */
326 if (WARN_ON(state > cpufreq_cdev->max_level))
329 /* Check if the old cooling action is same as new cooling action */
330 if (cpufreq_cdev->cpufreq_state == state)
333 ret = freq_qos_update_request(&cpufreq_cdev->qos_req,
334 cpufreq_cdev->freq_table[state].frequency);
336 cpufreq_cdev->cpufreq_state = state;
342 * cpufreq_get_requested_power() - get the current power
343 * @cdev: &thermal_cooling_device pointer
344 * @tz: a valid thermal zone device pointer
345 * @power: pointer in which to store the resulting power
347 * Calculate the current power consumption of the cpus in milliwatts
348 * and store it in @power. This function should actually calculate
349 * the requested power, but it's hard to get the frequency that
350 * cpufreq would have assigned if there were no thermal limits.
351 * Instead, we calculate the current power on the assumption that the
352 * immediate future will look like the immediate past.
354 * We use the current frequency and the average load since this
355 * function was last called. In reality, there could have been
356 * multiple opps since this function was last called and that affects
357 * the load calculation. While it's not perfectly accurate, this
358 * simplification is good enough and works. REVISIT this, as more
359 * complex code may be needed if experiments show that it's not
362 * Return: 0 on success, -E* if getting the static power failed.
364 static int cpufreq_get_requested_power(struct thermal_cooling_device *cdev,
365 struct thermal_zone_device *tz,
371 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
372 struct cpufreq_policy *policy = cpufreq_cdev->policy;
373 u32 *load_cpu = NULL;
375 freq = cpufreq_quick_get(policy->cpu);
377 if (trace_thermal_power_cpu_get_power_enabled()) {
378 u32 ncpus = cpumask_weight(policy->related_cpus);
380 load_cpu = kcalloc(ncpus, sizeof(*load_cpu), GFP_KERNEL);
383 for_each_cpu(cpu, policy->related_cpus) {
387 load = get_load(cpufreq_cdev, cpu, i);
398 cpufreq_cdev->last_load = total_load;
400 *power = get_dynamic_power(cpufreq_cdev, freq);
403 trace_thermal_power_cpu_get_power(policy->related_cpus, freq,
404 load_cpu, i, *power);
413 * cpufreq_state2power() - convert a cpu cdev state to power consumed
414 * @cdev: &thermal_cooling_device pointer
415 * @tz: a valid thermal zone device pointer
416 * @state: cooling device state to be converted
417 * @power: pointer in which to store the resulting power
419 * Convert cooling device state @state into power consumption in
420 * milliwatts assuming 100% load. Store the calculated power in
423 * Return: 0 on success, -EINVAL if the cooling device state could not
424 * be converted into a frequency or other -E* if there was an error
425 * when calculating the static power.
427 static int cpufreq_state2power(struct thermal_cooling_device *cdev,
428 struct thermal_zone_device *tz,
429 unsigned long state, u32 *power)
431 unsigned int freq, num_cpus;
432 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
434 /* Request state should be less than max_level */
435 if (WARN_ON(state > cpufreq_cdev->max_level))
438 num_cpus = cpumask_weight(cpufreq_cdev->policy->cpus);
440 freq = cpufreq_cdev->freq_table[state].frequency;
441 *power = cpu_freq_to_power(cpufreq_cdev, freq) * num_cpus;
447 * cpufreq_power2state() - convert power to a cooling device state
448 * @cdev: &thermal_cooling_device pointer
449 * @tz: a valid thermal zone device pointer
450 * @power: power in milliwatts to be converted
451 * @state: pointer in which to store the resulting state
453 * Calculate a cooling device state for the cpus described by @cdev
454 * that would allow them to consume at most @power mW and store it in
455 * @state. Note that this calculation depends on external factors
456 * such as the cpu load or the current static power. Calling this
457 * function with the same power as input can yield different cooling
458 * device states depending on those external factors.
460 * Return: 0 on success, -ENODEV if no cpus are online or -EINVAL if
461 * the calculated frequency could not be converted to a valid state.
462 * The latter should not happen unless the frequencies available to
463 * cpufreq have changed since the initialization of the cpu cooling
466 static int cpufreq_power2state(struct thermal_cooling_device *cdev,
467 struct thermal_zone_device *tz, u32 power,
468 unsigned long *state)
470 unsigned int target_freq;
471 u32 last_load, normalised_power;
472 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
473 struct cpufreq_policy *policy = cpufreq_cdev->policy;
475 last_load = cpufreq_cdev->last_load ?: 1;
476 normalised_power = (power * 100) / last_load;
477 target_freq = cpu_power_to_freq(cpufreq_cdev, normalised_power);
479 *state = get_level(cpufreq_cdev, target_freq);
480 trace_thermal_power_cpu_limit(policy->related_cpus, target_freq, *state,
485 /* Bind cpufreq callbacks to thermal cooling device ops */
487 static struct thermal_cooling_device_ops cpufreq_cooling_ops = {
488 .get_max_state = cpufreq_get_max_state,
489 .get_cur_state = cpufreq_get_cur_state,
490 .set_cur_state = cpufreq_set_cur_state,
493 static struct thermal_cooling_device_ops cpufreq_power_cooling_ops = {
494 .get_max_state = cpufreq_get_max_state,
495 .get_cur_state = cpufreq_get_cur_state,
496 .set_cur_state = cpufreq_set_cur_state,
497 .get_requested_power = cpufreq_get_requested_power,
498 .state2power = cpufreq_state2power,
499 .power2state = cpufreq_power2state,
502 static unsigned int find_next_max(struct cpufreq_frequency_table *table,
503 unsigned int prev_max)
505 struct cpufreq_frequency_table *pos;
506 unsigned int max = 0;
508 cpufreq_for_each_valid_entry(pos, table) {
509 if (pos->frequency > max && pos->frequency < prev_max)
510 max = pos->frequency;
517 * __cpufreq_cooling_register - helper function to create cpufreq cooling device
518 * @np: a valid struct device_node to the cooling device device tree node
519 * @policy: cpufreq policy
520 * Normally this should be same as cpufreq policy->related_cpus.
521 * @capacitance: dynamic power coefficient for these cpus
523 * This interface function registers the cpufreq cooling device with the name
524 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
525 * cooling devices. It also gives the opportunity to link the cooling device
526 * with a device tree node, in order to bind it via the thermal DT code.
528 * Return: a valid struct thermal_cooling_device pointer on success,
529 * on failure, it returns a corresponding ERR_PTR().
531 static struct thermal_cooling_device *
532 __cpufreq_cooling_register(struct device_node *np,
533 struct cpufreq_policy *policy, u32 capacitance)
535 struct thermal_cooling_device *cdev;
536 struct cpufreq_cooling_device *cpufreq_cdev;
537 char dev_name[THERMAL_NAME_LENGTH];
538 unsigned int freq, i, num_cpus;
541 struct thermal_cooling_device_ops *cooling_ops;
543 dev = get_cpu_device(policy->cpu);
544 if (unlikely(!dev)) {
545 pr_warn("No cpu device for cpu %d\n", policy->cpu);
546 return ERR_PTR(-ENODEV);
550 if (IS_ERR_OR_NULL(policy)) {
551 pr_err("%s: cpufreq policy isn't valid: %p\n", __func__, policy);
552 return ERR_PTR(-EINVAL);
555 i = cpufreq_table_count_valid_entries(policy);
557 pr_debug("%s: CPUFreq table not found or has no valid entries\n",
559 return ERR_PTR(-ENODEV);
562 cpufreq_cdev = kzalloc(sizeof(*cpufreq_cdev), GFP_KERNEL);
564 return ERR_PTR(-ENOMEM);
566 cpufreq_cdev->policy = policy;
567 num_cpus = cpumask_weight(policy->related_cpus);
568 cpufreq_cdev->idle_time = kcalloc(num_cpus,
569 sizeof(*cpufreq_cdev->idle_time),
571 if (!cpufreq_cdev->idle_time) {
572 cdev = ERR_PTR(-ENOMEM);
576 /* max_level is an index, not a counter */
577 cpufreq_cdev->max_level = i - 1;
579 cpufreq_cdev->freq_table = kmalloc_array(i,
580 sizeof(*cpufreq_cdev->freq_table),
582 if (!cpufreq_cdev->freq_table) {
583 cdev = ERR_PTR(-ENOMEM);
587 ret = ida_simple_get(&cpufreq_ida, 0, 0, GFP_KERNEL);
592 cpufreq_cdev->id = ret;
594 snprintf(dev_name, sizeof(dev_name), "thermal-cpufreq-%d",
597 /* Fill freq-table in descending order of frequencies */
598 for (i = 0, freq = -1; i <= cpufreq_cdev->max_level; i++) {
599 freq = find_next_max(policy->freq_table, freq);
600 cpufreq_cdev->freq_table[i].frequency = freq;
602 /* Warn for duplicate entries */
604 pr_warn("%s: table has duplicate entries\n", __func__);
606 pr_debug("%s: freq:%u KHz\n", __func__, freq);
610 ret = update_freq_table(cpufreq_cdev, capacitance);
616 cooling_ops = &cpufreq_power_cooling_ops;
618 cooling_ops = &cpufreq_cooling_ops;
621 ret = freq_qos_add_request(&policy->constraints,
622 &cpufreq_cdev->qos_req, FREQ_QOS_MAX,
623 cpufreq_cdev->freq_table[0].frequency);
625 pr_err("%s: Failed to add freq constraint (%d)\n", __func__,
631 cdev = thermal_of_cooling_device_register(np, dev_name, cpufreq_cdev,
636 mutex_lock(&cooling_list_lock);
637 list_add(&cpufreq_cdev->node, &cpufreq_cdev_list);
638 mutex_unlock(&cooling_list_lock);
643 freq_qos_remove_request(&cpufreq_cdev->qos_req);
645 ida_simple_remove(&cpufreq_ida, cpufreq_cdev->id);
647 kfree(cpufreq_cdev->freq_table);
649 kfree(cpufreq_cdev->idle_time);
656 * cpufreq_cooling_register - function to create cpufreq cooling device.
657 * @policy: cpufreq policy
659 * This interface function registers the cpufreq cooling device with the name
660 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
663 * Return: a valid struct thermal_cooling_device pointer on success,
664 * on failure, it returns a corresponding ERR_PTR().
666 struct thermal_cooling_device *
667 cpufreq_cooling_register(struct cpufreq_policy *policy)
669 return __cpufreq_cooling_register(NULL, policy, 0);
671 EXPORT_SYMBOL_GPL(cpufreq_cooling_register);
674 * of_cpufreq_cooling_register - function to create cpufreq cooling device.
675 * @policy: cpufreq policy
677 * This interface function registers the cpufreq cooling device with the name
678 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
679 * cooling devices. Using this API, the cpufreq cooling device will be
680 * linked to the device tree node provided.
682 * Using this function, the cooling device will implement the power
683 * extensions by using a simple cpu power model. The cpus must have
684 * registered their OPPs using the OPP library.
686 * It also takes into account, if property present in policy CPU node, the
687 * static power consumed by the cpu.
689 * Return: a valid struct thermal_cooling_device pointer on success,
690 * and NULL on failure.
692 struct thermal_cooling_device *
693 of_cpufreq_cooling_register(struct cpufreq_policy *policy)
695 struct device_node *np = of_get_cpu_node(policy->cpu, NULL);
696 struct thermal_cooling_device *cdev = NULL;
700 pr_err("cpu_cooling: OF node not available for cpu%d\n",
705 if (of_find_property(np, "#cooling-cells", NULL)) {
706 of_property_read_u32(np, "dynamic-power-coefficient",
709 cdev = __cpufreq_cooling_register(np, policy, capacitance);
711 pr_err("cpu_cooling: cpu%d failed to register as cooling device: %ld\n",
712 policy->cpu, PTR_ERR(cdev));
720 EXPORT_SYMBOL_GPL(of_cpufreq_cooling_register);
723 * cpufreq_cooling_unregister - function to remove cpufreq cooling device.
724 * @cdev: thermal cooling device pointer.
726 * This interface function unregisters the "thermal-cpufreq-%x" cooling device.
728 void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
730 struct cpufreq_cooling_device *cpufreq_cdev;
735 cpufreq_cdev = cdev->devdata;
737 mutex_lock(&cooling_list_lock);
738 list_del(&cpufreq_cdev->node);
739 mutex_unlock(&cooling_list_lock);
741 thermal_cooling_device_unregister(cdev);
742 freq_qos_remove_request(&cpufreq_cdev->qos_req);
743 ida_simple_remove(&cpufreq_ida, cpufreq_cdev->id);
744 kfree(cpufreq_cdev->idle_time);
745 kfree(cpufreq_cdev->freq_table);
748 EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister);