1 // SPDX-License-Identifier: GPL-2.0
3 * linux/drivers/thermal/cpufreq_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/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 #include <linux/units.h>
26 #include <trace/events/thermal.h>
29 * Cooling state <-> CPUFreq frequency
31 * Cooling states are translated to frequencies throughout this driver and this
32 * is the relation between them.
34 * Highest cooling state corresponds to lowest possible frequency.
37 * level 0 --> 1st Max Freq
38 * level 1 --> 2nd Max Freq
43 * struct time_in_idle - Idle time stats
44 * @time: previous reading of the absolute time that this cpu was idle
45 * @timestamp: wall time of the last invocation of get_cpu_idle_time_us()
53 * struct cpufreq_cooling_device - data for cooling device with cpufreq
54 * @last_load: load measured by the latest call to cpufreq_get_requested_power()
55 * @cpufreq_state: integer value representing the current state of cpufreq
57 * @max_level: maximum cooling level. One less than total number of valid
58 * cpufreq frequencies.
59 * @em: Reference on the Energy Model of the device
60 * @cdev: thermal_cooling_device pointer to keep track of the
61 * registered cooling device.
62 * @policy: cpufreq policy.
63 * @idle_time: idle time stats
64 * @qos_req: PM QoS contraint to apply
66 * This structure is required for keeping information of each registered
67 * cpufreq_cooling_device.
69 struct cpufreq_cooling_device {
71 unsigned int cpufreq_state;
72 unsigned int max_level;
73 struct em_perf_domain *em;
74 struct cpufreq_policy *policy;
76 struct time_in_idle *idle_time;
78 struct freq_qos_request qos_req;
81 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
83 * get_level: Find the level for a particular frequency
84 * @cpufreq_cdev: cpufreq_cdev for which the property is required
87 * Return: level corresponding to the frequency.
89 static unsigned long get_level(struct cpufreq_cooling_device *cpufreq_cdev,
94 for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
95 if (freq > cpufreq_cdev->em->table[i].frequency)
99 return cpufreq_cdev->max_level - i - 1;
102 static u32 cpu_freq_to_power(struct cpufreq_cooling_device *cpufreq_cdev,
105 unsigned long power_mw;
108 for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
109 if (freq > cpufreq_cdev->em->table[i].frequency)
113 power_mw = cpufreq_cdev->em->table[i + 1].power;
114 power_mw /= MICROWATT_PER_MILLIWATT;
119 static u32 cpu_power_to_freq(struct cpufreq_cooling_device *cpufreq_cdev,
122 unsigned long em_power_mw;
125 for (i = cpufreq_cdev->max_level; i > 0; i--) {
126 /* Convert EM power to milli-Watts to make safe comparison */
127 em_power_mw = cpufreq_cdev->em->table[i].power;
128 em_power_mw /= MICROWATT_PER_MILLIWATT;
129 if (power >= em_power_mw)
133 return cpufreq_cdev->em->table[i].frequency;
137 * get_load() - get load for a cpu
138 * @cpufreq_cdev: struct cpufreq_cooling_device for the cpu
140 * @cpu_idx: index of the cpu in time_in_idle array
142 * Return: The average load of cpu @cpu in percentage since this
143 * function was last called.
146 static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
149 unsigned long util = sched_cpu_util(cpu);
151 return (util * 100) / arch_scale_cpu_capacity(cpu);
153 #else /* !CONFIG_SMP */
154 static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
158 u64 now, now_idle, delta_time, delta_idle;
159 struct time_in_idle *idle_time = &cpufreq_cdev->idle_time[cpu_idx];
161 now_idle = get_cpu_idle_time(cpu, &now, 0);
162 delta_idle = now_idle - idle_time->time;
163 delta_time = now - idle_time->timestamp;
165 if (delta_time <= delta_idle)
168 load = div64_u64(100 * (delta_time - delta_idle), delta_time);
170 idle_time->time = now_idle;
171 idle_time->timestamp = now;
175 #endif /* CONFIG_SMP */
178 * get_dynamic_power() - calculate the dynamic power
179 * @cpufreq_cdev: &cpufreq_cooling_device for this cdev
180 * @freq: current frequency
182 * Return: the dynamic power consumed by the cpus described by
185 static u32 get_dynamic_power(struct cpufreq_cooling_device *cpufreq_cdev,
190 raw_cpu_power = cpu_freq_to_power(cpufreq_cdev, freq);
191 return (raw_cpu_power * cpufreq_cdev->last_load) / 100;
195 * cpufreq_get_requested_power() - get the current power
196 * @cdev: &thermal_cooling_device pointer
197 * @power: pointer in which to store the resulting power
199 * Calculate the current power consumption of the cpus in milliwatts
200 * and store it in @power. This function should actually calculate
201 * the requested power, but it's hard to get the frequency that
202 * cpufreq would have assigned if there were no thermal limits.
203 * Instead, we calculate the current power on the assumption that the
204 * immediate future will look like the immediate past.
206 * We use the current frequency and the average load since this
207 * function was last called. In reality, there could have been
208 * multiple opps since this function was last called and that affects
209 * the load calculation. While it's not perfectly accurate, this
210 * simplification is good enough and works. REVISIT this, as more
211 * complex code may be needed if experiments show that it's not
214 * Return: 0 on success, -E* if getting the static power failed.
216 static int cpufreq_get_requested_power(struct thermal_cooling_device *cdev,
222 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
223 struct cpufreq_policy *policy = cpufreq_cdev->policy;
224 u32 *load_cpu = NULL;
226 freq = cpufreq_quick_get(policy->cpu);
228 if (trace_thermal_power_cpu_get_power_enabled()) {
229 u32 ncpus = cpumask_weight(policy->related_cpus);
231 load_cpu = kcalloc(ncpus, sizeof(*load_cpu), GFP_KERNEL);
234 for_each_cpu(cpu, policy->related_cpus) {
238 load = get_load(cpufreq_cdev, cpu, i);
249 cpufreq_cdev->last_load = total_load;
251 *power = get_dynamic_power(cpufreq_cdev, freq);
254 trace_thermal_power_cpu_get_power(policy->related_cpus, freq,
255 load_cpu, i, *power);
264 * cpufreq_state2power() - convert a cpu cdev state to power consumed
265 * @cdev: &thermal_cooling_device pointer
266 * @state: cooling device state to be converted
267 * @power: pointer in which to store the resulting power
269 * Convert cooling device state @state into power consumption in
270 * milliwatts assuming 100% load. Store the calculated power in
273 * Return: 0 on success, -EINVAL if the cooling device state could not
274 * be converted into a frequency or other -E* if there was an error
275 * when calculating the static power.
277 static int cpufreq_state2power(struct thermal_cooling_device *cdev,
278 unsigned long state, u32 *power)
280 unsigned int freq, num_cpus, idx;
281 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
283 /* Request state should be less than max_level */
284 if (state > cpufreq_cdev->max_level)
287 num_cpus = cpumask_weight(cpufreq_cdev->policy->cpus);
289 idx = cpufreq_cdev->max_level - state;
290 freq = cpufreq_cdev->em->table[idx].frequency;
291 *power = cpu_freq_to_power(cpufreq_cdev, freq) * num_cpus;
297 * cpufreq_power2state() - convert power to a cooling device state
298 * @cdev: &thermal_cooling_device pointer
299 * @power: power in milliwatts to be converted
300 * @state: pointer in which to store the resulting state
302 * Calculate a cooling device state for the cpus described by @cdev
303 * that would allow them to consume at most @power mW and store it in
304 * @state. Note that this calculation depends on external factors
305 * such as the cpu load or the current static power. Calling this
306 * function with the same power as input can yield different cooling
307 * device states depending on those external factors.
309 * Return: 0 on success, -ENODEV if no cpus are online or -EINVAL if
310 * the calculated frequency could not be converted to a valid state.
311 * The latter should not happen unless the frequencies available to
312 * cpufreq have changed since the initialization of the cpu cooling
315 static int cpufreq_power2state(struct thermal_cooling_device *cdev,
316 u32 power, unsigned long *state)
318 unsigned int target_freq;
319 u32 last_load, normalised_power;
320 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
321 struct cpufreq_policy *policy = cpufreq_cdev->policy;
323 last_load = cpufreq_cdev->last_load ?: 1;
324 normalised_power = (power * 100) / last_load;
325 target_freq = cpu_power_to_freq(cpufreq_cdev, normalised_power);
327 *state = get_level(cpufreq_cdev, target_freq);
328 trace_thermal_power_cpu_limit(policy->related_cpus, target_freq, *state,
333 static inline bool em_is_sane(struct cpufreq_cooling_device *cpufreq_cdev,
334 struct em_perf_domain *em) {
335 struct cpufreq_policy *policy;
336 unsigned int nr_levels;
338 if (!em || em_is_artificial(em))
341 policy = cpufreq_cdev->policy;
342 if (!cpumask_equal(policy->related_cpus, em_span_cpus(em))) {
343 pr_err("The span of pd %*pbl is misaligned with cpufreq policy %*pbl\n",
344 cpumask_pr_args(em_span_cpus(em)),
345 cpumask_pr_args(policy->related_cpus));
349 nr_levels = cpufreq_cdev->max_level + 1;
350 if (em_pd_nr_perf_states(em) != nr_levels) {
351 pr_err("The number of performance states in pd %*pbl (%u) doesn't match the number of cooling levels (%u)\n",
352 cpumask_pr_args(em_span_cpus(em)),
353 em_pd_nr_perf_states(em), nr_levels);
359 #endif /* CONFIG_THERMAL_GOV_POWER_ALLOCATOR */
362 static inline int allocate_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
367 static inline void free_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
371 static int allocate_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
373 unsigned int num_cpus = cpumask_weight(cpufreq_cdev->policy->related_cpus);
375 cpufreq_cdev->idle_time = kcalloc(num_cpus,
376 sizeof(*cpufreq_cdev->idle_time),
378 if (!cpufreq_cdev->idle_time)
384 static void free_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
386 kfree(cpufreq_cdev->idle_time);
387 cpufreq_cdev->idle_time = NULL;
389 #endif /* CONFIG_SMP */
391 static unsigned int get_state_freq(struct cpufreq_cooling_device *cpufreq_cdev,
394 struct cpufreq_policy *policy;
397 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
398 /* Use the Energy Model table if available */
399 if (cpufreq_cdev->em) {
400 idx = cpufreq_cdev->max_level - state;
401 return cpufreq_cdev->em->table[idx].frequency;
405 /* Otherwise, fallback on the CPUFreq table */
406 policy = cpufreq_cdev->policy;
407 if (policy->freq_table_sorted == CPUFREQ_TABLE_SORTED_ASCENDING)
408 idx = cpufreq_cdev->max_level - state;
412 return policy->freq_table[idx].frequency;
415 /* cpufreq cooling device callback functions are defined below */
418 * cpufreq_get_max_state - callback function to get the max cooling state.
419 * @cdev: thermal cooling device pointer.
420 * @state: fill this variable with the max cooling state.
422 * Callback for the thermal cooling device to return the cpufreq
425 * Return: 0 on success, an error code otherwise.
427 static int cpufreq_get_max_state(struct thermal_cooling_device *cdev,
428 unsigned long *state)
430 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
432 *state = cpufreq_cdev->max_level;
437 * cpufreq_get_cur_state - callback function to get the current cooling state.
438 * @cdev: thermal cooling device pointer.
439 * @state: fill this variable with the current cooling state.
441 * Callback for the thermal cooling device to return the cpufreq
442 * current cooling state.
444 * Return: 0 on success, an error code otherwise.
446 static int cpufreq_get_cur_state(struct thermal_cooling_device *cdev,
447 unsigned long *state)
449 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
451 *state = cpufreq_cdev->cpufreq_state;
457 * cpufreq_set_cur_state - callback function to set the current cooling state.
458 * @cdev: thermal cooling device pointer.
459 * @state: set this variable to the current cooling state.
461 * Callback for the thermal cooling device to change the cpufreq
462 * current cooling state.
464 * Return: 0 on success, an error code otherwise.
466 static int cpufreq_set_cur_state(struct thermal_cooling_device *cdev,
469 struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
470 struct cpumask *cpus;
471 unsigned int frequency;
474 /* Request state should be less than max_level */
475 if (state > cpufreq_cdev->max_level)
478 /* Check if the old cooling action is same as new cooling action */
479 if (cpufreq_cdev->cpufreq_state == state)
482 frequency = get_state_freq(cpufreq_cdev, state);
484 ret = freq_qos_update_request(&cpufreq_cdev->qos_req, frequency);
486 cpufreq_cdev->cpufreq_state = state;
487 cpus = cpufreq_cdev->policy->related_cpus;
488 arch_update_thermal_pressure(cpus, frequency);
495 /* Bind cpufreq callbacks to thermal cooling device ops */
497 static struct thermal_cooling_device_ops cpufreq_cooling_ops = {
498 .get_max_state = cpufreq_get_max_state,
499 .get_cur_state = cpufreq_get_cur_state,
500 .set_cur_state = cpufreq_set_cur_state,
504 * __cpufreq_cooling_register - helper function to create cpufreq cooling device
505 * @np: a valid struct device_node to the cooling device device tree node
506 * @policy: cpufreq policy
507 * Normally this should be same as cpufreq policy->related_cpus.
508 * @em: Energy Model of the cpufreq policy
510 * This interface function registers the cpufreq cooling device with the name
511 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
512 * cooling devices. It also gives the opportunity to link the cooling device
513 * with a device tree node, in order to bind it via the thermal DT code.
515 * Return: a valid struct thermal_cooling_device pointer on success,
516 * on failure, it returns a corresponding ERR_PTR().
518 static struct thermal_cooling_device *
519 __cpufreq_cooling_register(struct device_node *np,
520 struct cpufreq_policy *policy,
521 struct em_perf_domain *em)
523 struct thermal_cooling_device *cdev;
524 struct cpufreq_cooling_device *cpufreq_cdev;
528 struct thermal_cooling_device_ops *cooling_ops;
531 dev = get_cpu_device(policy->cpu);
532 if (unlikely(!dev)) {
533 pr_warn("No cpu device for cpu %d\n", policy->cpu);
534 return ERR_PTR(-ENODEV);
537 if (IS_ERR_OR_NULL(policy)) {
538 pr_err("%s: cpufreq policy isn't valid: %p\n", __func__, policy);
539 return ERR_PTR(-EINVAL);
542 i = cpufreq_table_count_valid_entries(policy);
544 pr_debug("%s: CPUFreq table not found or has no valid entries\n",
546 return ERR_PTR(-ENODEV);
549 cpufreq_cdev = kzalloc(sizeof(*cpufreq_cdev), GFP_KERNEL);
551 return ERR_PTR(-ENOMEM);
553 cpufreq_cdev->policy = policy;
555 ret = allocate_idle_time(cpufreq_cdev);
561 /* max_level is an index, not a counter */
562 cpufreq_cdev->max_level = i - 1;
564 cooling_ops = &cpufreq_cooling_ops;
566 #ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
567 if (em_is_sane(cpufreq_cdev, em)) {
568 cpufreq_cdev->em = em;
569 cooling_ops->get_requested_power = cpufreq_get_requested_power;
570 cooling_ops->state2power = cpufreq_state2power;
571 cooling_ops->power2state = cpufreq_power2state;
574 if (policy->freq_table_sorted == CPUFREQ_TABLE_UNSORTED) {
575 pr_err("%s: unsorted frequency tables are not supported\n",
577 cdev = ERR_PTR(-EINVAL);
581 ret = freq_qos_add_request(&policy->constraints,
582 &cpufreq_cdev->qos_req, FREQ_QOS_MAX,
583 get_state_freq(cpufreq_cdev, 0));
585 pr_err("%s: Failed to add freq constraint (%d)\n", __func__,
591 cdev = ERR_PTR(-ENOMEM);
592 name = kasprintf(GFP_KERNEL, "cpufreq-%s", dev_name(dev));
596 cdev = thermal_of_cooling_device_register(np, name, cpufreq_cdev,
606 freq_qos_remove_request(&cpufreq_cdev->qos_req);
608 free_idle_time(cpufreq_cdev);
615 * cpufreq_cooling_register - function to create cpufreq cooling device.
616 * @policy: cpufreq policy
618 * This interface function registers the cpufreq cooling device with the name
619 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
622 * Return: a valid struct thermal_cooling_device pointer on success,
623 * on failure, it returns a corresponding ERR_PTR().
625 struct thermal_cooling_device *
626 cpufreq_cooling_register(struct cpufreq_policy *policy)
628 return __cpufreq_cooling_register(NULL, policy, NULL);
630 EXPORT_SYMBOL_GPL(cpufreq_cooling_register);
633 * of_cpufreq_cooling_register - function to create cpufreq cooling device.
634 * @policy: cpufreq policy
636 * This interface function registers the cpufreq cooling device with the name
637 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
638 * cooling devices. Using this API, the cpufreq cooling device will be
639 * linked to the device tree node provided.
641 * Using this function, the cooling device will implement the power
642 * extensions by using a simple cpu power model. The cpus must have
643 * registered their OPPs using the OPP library.
645 * It also takes into account, if property present in policy CPU node, the
646 * static power consumed by the cpu.
648 * Return: a valid struct thermal_cooling_device pointer on success,
649 * and NULL on failure.
651 struct thermal_cooling_device *
652 of_cpufreq_cooling_register(struct cpufreq_policy *policy)
654 struct device_node *np = of_get_cpu_node(policy->cpu, NULL);
655 struct thermal_cooling_device *cdev = NULL;
658 pr_err("cpufreq_cooling: OF node not available for cpu%d\n",
663 if (of_find_property(np, "#cooling-cells", NULL)) {
664 struct em_perf_domain *em = em_cpu_get(policy->cpu);
666 cdev = __cpufreq_cooling_register(np, policy, em);
668 pr_err("cpufreq_cooling: cpu%d failed to register as cooling device: %ld\n",
669 policy->cpu, PTR_ERR(cdev));
677 EXPORT_SYMBOL_GPL(of_cpufreq_cooling_register);
680 * cpufreq_cooling_unregister - function to remove cpufreq cooling device.
681 * @cdev: thermal cooling device pointer.
683 * This interface function unregisters the "thermal-cpufreq-%x" cooling device.
685 void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
687 struct cpufreq_cooling_device *cpufreq_cdev;
692 cpufreq_cdev = cdev->devdata;
694 thermal_cooling_device_unregister(cdev);
695 freq_qos_remove_request(&cpufreq_cdev->qos_req);
696 free_idle_time(cpufreq_cdev);
699 EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister);