Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
  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#include <linux/units.h>
 25
 26#include <trace/events/thermal.h>
 27
 28/*
 29 * Cooling state <-> CPUFreq frequency
 30 *
 31 * Cooling states are translated to frequencies throughout this driver and this
 32 * is the relation between them.
 33 *
 34 * Highest cooling state corresponds to lowest possible frequency.
 35 *
 36 * i.e.
 37 *	level 0 --> 1st Max Freq
 38 *	level 1 --> 2nd Max Freq
 39 *	...
 40 */
 41
 42/**
 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()
 46 */
 47struct time_in_idle {
 48	u64 time;
 49	u64 timestamp;
 50};
 51
 52/**
 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
 56 *	cooling	devices.
 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 * @cooling_ops: cpufreq callbacks to thermal cooling device ops
 64 * @idle_time: idle time stats
 65 * @qos_req: PM QoS contraint to apply
 66 *
 67 * This structure is required for keeping information of each registered
 68 * cpufreq_cooling_device.
 69 */
 70struct cpufreq_cooling_device {
 71	u32 last_load;
 72	unsigned int cpufreq_state;
 73	unsigned int max_level;
 74	struct em_perf_domain *em;
 75	struct cpufreq_policy *policy;
 76	struct thermal_cooling_device_ops cooling_ops;
 77#ifndef CONFIG_SMP
 78	struct time_in_idle *idle_time;
 79#endif
 80	struct freq_qos_request qos_req;
 81};
 82
 83#ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
 84/**
 85 * get_level: Find the level for a particular frequency
 86 * @cpufreq_cdev: cpufreq_cdev for which the property is required
 87 * @freq: Frequency
 88 *
 89 * Return: level corresponding to the frequency.
 90 */
 91static unsigned long get_level(struct cpufreq_cooling_device *cpufreq_cdev,
 92			       unsigned int freq)
 93{
 94	int i;
 95
 96	for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
 97		if (freq > cpufreq_cdev->em->table[i].frequency)
 98			break;
 99	}
100
101	return cpufreq_cdev->max_level - i - 1;
102}
103
104static u32 cpu_freq_to_power(struct cpufreq_cooling_device *cpufreq_cdev,
105			     u32 freq)
106{
107	unsigned long power_mw;
108	int i;
109
110	for (i = cpufreq_cdev->max_level - 1; i >= 0; i--) {
111		if (freq > cpufreq_cdev->em->table[i].frequency)
112			break;
113	}
114
115	power_mw = cpufreq_cdev->em->table[i + 1].power;
116	power_mw /= MICROWATT_PER_MILLIWATT;
117
118	return power_mw;
119}
120
121static u32 cpu_power_to_freq(struct cpufreq_cooling_device *cpufreq_cdev,
122			     u32 power)
123{
124	unsigned long em_power_mw;
125	int i;
126
127	for (i = cpufreq_cdev->max_level; i > 0; i--) {
128		/* Convert EM power to milli-Watts to make safe comparison */
129		em_power_mw = cpufreq_cdev->em->table[i].power;
130		em_power_mw /= MICROWATT_PER_MILLIWATT;
131		if (power >= em_power_mw)
132			break;
133	}
134
135	return cpufreq_cdev->em->table[i].frequency;
136}
137
138/**
139 * get_load() - get load for a cpu
140 * @cpufreq_cdev: struct cpufreq_cooling_device for the cpu
141 * @cpu: cpu number
142 * @cpu_idx: index of the cpu in time_in_idle array
143 *
144 * Return: The average load of cpu @cpu in percentage since this
145 * function was last called.
146 */
147#ifdef CONFIG_SMP
148static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
149		    int cpu_idx)
150{
151	unsigned long util = sched_cpu_util(cpu);
152
153	return (util * 100) / arch_scale_cpu_capacity(cpu);
154}
155#else /* !CONFIG_SMP */
156static u32 get_load(struct cpufreq_cooling_device *cpufreq_cdev, int cpu,
157		    int cpu_idx)
158{
159	u32 load;
160	u64 now, now_idle, delta_time, delta_idle;
161	struct time_in_idle *idle_time = &cpufreq_cdev->idle_time[cpu_idx];
162
163	now_idle = get_cpu_idle_time(cpu, &now, 0);
164	delta_idle = now_idle - idle_time->time;
165	delta_time = now - idle_time->timestamp;
166
167	if (delta_time <= delta_idle)
168		load = 0;
169	else
170		load = div64_u64(100 * (delta_time - delta_idle), delta_time);
171
172	idle_time->time = now_idle;
173	idle_time->timestamp = now;
174
175	return load;
176}
177#endif /* CONFIG_SMP */
178
179/**
180 * get_dynamic_power() - calculate the dynamic power
181 * @cpufreq_cdev:	&cpufreq_cooling_device for this cdev
182 * @freq:	current frequency
183 *
184 * Return: the dynamic power consumed by the cpus described by
185 * @cpufreq_cdev.
186 */
187static u32 get_dynamic_power(struct cpufreq_cooling_device *cpufreq_cdev,
188			     unsigned long freq)
189{
190	u32 raw_cpu_power;
191
192	raw_cpu_power = cpu_freq_to_power(cpufreq_cdev, freq);
193	return (raw_cpu_power * cpufreq_cdev->last_load) / 100;
194}
195
196/**
197 * cpufreq_get_requested_power() - get the current power
198 * @cdev:	&thermal_cooling_device pointer
199 * @power:	pointer in which to store the resulting power
200 *
201 * Calculate the current power consumption of the cpus in milliwatts
202 * and store it in @power.  This function should actually calculate
203 * the requested power, but it's hard to get the frequency that
204 * cpufreq would have assigned if there were no thermal limits.
205 * Instead, we calculate the current power on the assumption that the
206 * immediate future will look like the immediate past.
207 *
208 * We use the current frequency and the average load since this
209 * function was last called.  In reality, there could have been
210 * multiple opps since this function was last called and that affects
211 * the load calculation.  While it's not perfectly accurate, this
212 * simplification is good enough and works.  REVISIT this, as more
213 * complex code may be needed if experiments show that it's not
214 * accurate enough.
215 *
216 * Return: 0 on success, this function doesn't fail.
217 */
218static int cpufreq_get_requested_power(struct thermal_cooling_device *cdev,
219				       u32 *power)
220{
221	unsigned long freq;
222	int i = 0, cpu;
223	u32 total_load = 0;
224	struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
225	struct cpufreq_policy *policy = cpufreq_cdev->policy;
226
227	freq = cpufreq_quick_get(policy->cpu);
228
229	for_each_cpu(cpu, policy->related_cpus) {
230		u32 load;
231
232		if (cpu_online(cpu))
233			load = get_load(cpufreq_cdev, cpu, i);
234		else
235			load = 0;
236
237		total_load += load;
238	}
239
240	cpufreq_cdev->last_load = total_load;
241
242	*power = get_dynamic_power(cpufreq_cdev, freq);
243
244	trace_thermal_power_cpu_get_power_simple(policy->cpu, *power);
245
246	return 0;
247}
248
249/**
250 * cpufreq_state2power() - convert a cpu cdev state to power consumed
251 * @cdev:	&thermal_cooling_device pointer
252 * @state:	cooling device state to be converted
253 * @power:	pointer in which to store the resulting power
254 *
255 * Convert cooling device state @state into power consumption in
256 * milliwatts assuming 100% load.  Store the calculated power in
257 * @power.
258 *
259 * Return: 0 on success, -EINVAL if the cooling device state is bigger
260 * than maximum allowed.
261 */
262static int cpufreq_state2power(struct thermal_cooling_device *cdev,
263			       unsigned long state, u32 *power)
264{
265	unsigned int freq, num_cpus, idx;
266	struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
267
268	/* Request state should be less than max_level */
269	if (state > cpufreq_cdev->max_level)
270		return -EINVAL;
271
272	num_cpus = cpumask_weight(cpufreq_cdev->policy->cpus);
273
274	idx = cpufreq_cdev->max_level - state;
275	freq = cpufreq_cdev->em->table[idx].frequency;
276	*power = cpu_freq_to_power(cpufreq_cdev, freq) * num_cpus;
277
278	return 0;
279}
280
281/**
282 * cpufreq_power2state() - convert power to a cooling device state
283 * @cdev:	&thermal_cooling_device pointer
284 * @power:	power in milliwatts to be converted
285 * @state:	pointer in which to store the resulting state
286 *
287 * Calculate a cooling device state for the cpus described by @cdev
288 * that would allow them to consume at most @power mW and store it in
289 * @state.  Note that this calculation depends on external factors
290 * such as the CPUs load.  Calling this function with the same power
291 * as input can yield different cooling device states depending on those
292 * external factors.
293 *
294 * Return: 0 on success, this function doesn't fail.
295 */
296static int cpufreq_power2state(struct thermal_cooling_device *cdev,
297			       u32 power, unsigned long *state)
298{
299	unsigned int target_freq;
300	u32 last_load, normalised_power;
301	struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
302	struct cpufreq_policy *policy = cpufreq_cdev->policy;
303
304	last_load = cpufreq_cdev->last_load ?: 1;
305	normalised_power = (power * 100) / last_load;
306	target_freq = cpu_power_to_freq(cpufreq_cdev, normalised_power);
307
308	*state = get_level(cpufreq_cdev, target_freq);
309	trace_thermal_power_cpu_limit(policy->related_cpus, target_freq, *state,
310				      power);
311	return 0;
312}
313
314static inline bool em_is_sane(struct cpufreq_cooling_device *cpufreq_cdev,
315			      struct em_perf_domain *em) {
316	struct cpufreq_policy *policy;
317	unsigned int nr_levels;
318
319	if (!em || em_is_artificial(em))
320		return false;
321
322	policy = cpufreq_cdev->policy;
323	if (!cpumask_equal(policy->related_cpus, em_span_cpus(em))) {
324		pr_err("The span of pd %*pbl is misaligned with cpufreq policy %*pbl\n",
325			cpumask_pr_args(em_span_cpus(em)),
326			cpumask_pr_args(policy->related_cpus));
327		return false;
328	}
329
330	nr_levels = cpufreq_cdev->max_level + 1;
331	if (em_pd_nr_perf_states(em) != nr_levels) {
332		pr_err("The number of performance states in pd %*pbl (%u) doesn't match the number of cooling levels (%u)\n",
333			cpumask_pr_args(em_span_cpus(em)),
334			em_pd_nr_perf_states(em), nr_levels);
335		return false;
336	}
337
338	return true;
339}
340#endif /* CONFIG_THERMAL_GOV_POWER_ALLOCATOR */
341
342#ifdef CONFIG_SMP
343static inline int allocate_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
344{
345	return 0;
346}
347
348static inline void free_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
349{
350}
351#else
352static int allocate_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
353{
354	unsigned int num_cpus = cpumask_weight(cpufreq_cdev->policy->related_cpus);
355
356	cpufreq_cdev->idle_time = kcalloc(num_cpus,
357					  sizeof(*cpufreq_cdev->idle_time),
358					  GFP_KERNEL);
359	if (!cpufreq_cdev->idle_time)
360		return -ENOMEM;
361
362	return 0;
363}
364
365static void free_idle_time(struct cpufreq_cooling_device *cpufreq_cdev)
366{
367	kfree(cpufreq_cdev->idle_time);
368	cpufreq_cdev->idle_time = NULL;
369}
370#endif /* CONFIG_SMP */
371
372static unsigned int get_state_freq(struct cpufreq_cooling_device *cpufreq_cdev,
373				   unsigned long state)
374{
375	struct cpufreq_policy *policy;
376	unsigned long idx;
377
378#ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
379	/* Use the Energy Model table if available */
380	if (cpufreq_cdev->em) {
381		idx = cpufreq_cdev->max_level - state;
382		return cpufreq_cdev->em->table[idx].frequency;
383	}
384#endif
385
386	/* Otherwise, fallback on the CPUFreq table */
387	policy = cpufreq_cdev->policy;
388	if (policy->freq_table_sorted == CPUFREQ_TABLE_SORTED_ASCENDING)
389		idx = cpufreq_cdev->max_level - state;
390	else
391		idx = state;
392
393	return policy->freq_table[idx].frequency;
394}
395
396/* cpufreq cooling device callback functions are defined below */
397
398/**
399 * cpufreq_get_max_state - callback function to get the max cooling state.
400 * @cdev: thermal cooling device pointer.
401 * @state: fill this variable with the max cooling state.
402 *
403 * Callback for the thermal cooling device to return the cpufreq
404 * max cooling state.
405 *
406 * Return: 0 on success, this function doesn't fail.
407 */
408static int cpufreq_get_max_state(struct thermal_cooling_device *cdev,
409				 unsigned long *state)
410{
411	struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
412
413	*state = cpufreq_cdev->max_level;
414	return 0;
415}
416
417/**
418 * cpufreq_get_cur_state - callback function to get the current cooling state.
419 * @cdev: thermal cooling device pointer.
420 * @state: fill this variable with the current cooling state.
421 *
422 * Callback for the thermal cooling device to return the cpufreq
423 * current cooling state.
424 *
425 * Return: 0 on success, this function doesn't fail.
426 */
427static int cpufreq_get_cur_state(struct thermal_cooling_device *cdev,
428				 unsigned long *state)
429{
430	struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
431
432	*state = cpufreq_cdev->cpufreq_state;
433
434	return 0;
435}
436
437/**
438 * cpufreq_set_cur_state - callback function to set the current cooling state.
439 * @cdev: thermal cooling device pointer.
440 * @state: set this variable to the current cooling state.
441 *
442 * Callback for the thermal cooling device to change the cpufreq
443 * current cooling state.
444 *
445 * Return: 0 on success, an error code otherwise.
446 */
447static int cpufreq_set_cur_state(struct thermal_cooling_device *cdev,
448				 unsigned long state)
449{
450	struct cpufreq_cooling_device *cpufreq_cdev = cdev->devdata;
451	struct cpumask *cpus;
452	unsigned int frequency;
453	int ret;
454
455	/* Request state should be less than max_level */
456	if (state > cpufreq_cdev->max_level)
457		return -EINVAL;
458
459	/* Check if the old cooling action is same as new cooling action */
460	if (cpufreq_cdev->cpufreq_state == state)
461		return 0;
462
463	frequency = get_state_freq(cpufreq_cdev, state);
464
465	ret = freq_qos_update_request(&cpufreq_cdev->qos_req, frequency);
466	if (ret >= 0) {
467		cpufreq_cdev->cpufreq_state = state;
468		cpus = cpufreq_cdev->policy->related_cpus;
469		arch_update_thermal_pressure(cpus, frequency);
470		ret = 0;
471	}
472
473	return ret;
474}
475
476/**
477 * __cpufreq_cooling_register - helper function to create cpufreq cooling device
478 * @np: a valid struct device_node to the cooling device tree node
479 * @policy: cpufreq policy
480 * Normally this should be same as cpufreq policy->related_cpus.
481 * @em: Energy Model of the cpufreq policy
482 *
483 * This interface function registers the cpufreq cooling device with the name
484 * "cpufreq-%s". This API can support multiple instances of cpufreq
485 * cooling devices. It also gives the opportunity to link the cooling device
486 * with a device tree node, in order to bind it via the thermal DT code.
487 *
488 * Return: a valid struct thermal_cooling_device pointer on success,
489 * on failure, it returns a corresponding ERR_PTR().
490 */
491static struct thermal_cooling_device *
492__cpufreq_cooling_register(struct device_node *np,
493			struct cpufreq_policy *policy,
494			struct em_perf_domain *em)
495{
496	struct thermal_cooling_device *cdev;
497	struct cpufreq_cooling_device *cpufreq_cdev;
498	unsigned int i;
499	struct device *dev;
500	int ret;
501	struct thermal_cooling_device_ops *cooling_ops;
502	char *name;
503
504	if (IS_ERR_OR_NULL(policy)) {
505		pr_err("%s: cpufreq policy isn't valid: %p\n", __func__, policy);
506		return ERR_PTR(-EINVAL);
507	}
508
509	dev = get_cpu_device(policy->cpu);
510	if (unlikely(!dev)) {
511		pr_warn("No cpu device for cpu %d\n", policy->cpu);
512		return ERR_PTR(-ENODEV);
513	}
514
515	i = cpufreq_table_count_valid_entries(policy);
516	if (!i) {
517		pr_debug("%s: CPUFreq table not found or has no valid entries\n",
518			 __func__);
519		return ERR_PTR(-ENODEV);
520	}
521
522	cpufreq_cdev = kzalloc(sizeof(*cpufreq_cdev), GFP_KERNEL);
523	if (!cpufreq_cdev)
524		return ERR_PTR(-ENOMEM);
525
526	cpufreq_cdev->policy = policy;
527
528	ret = allocate_idle_time(cpufreq_cdev);
529	if (ret) {
530		cdev = ERR_PTR(ret);
531		goto free_cdev;
532	}
533
534	/* max_level is an index, not a counter */
535	cpufreq_cdev->max_level = i - 1;
536
537	cooling_ops = &cpufreq_cdev->cooling_ops;
538	cooling_ops->get_max_state = cpufreq_get_max_state;
539	cooling_ops->get_cur_state = cpufreq_get_cur_state;
540	cooling_ops->set_cur_state = cpufreq_set_cur_state;
541
542#ifdef CONFIG_THERMAL_GOV_POWER_ALLOCATOR
543	if (em_is_sane(cpufreq_cdev, em)) {
544		cpufreq_cdev->em = em;
545		cooling_ops->get_requested_power = cpufreq_get_requested_power;
546		cooling_ops->state2power = cpufreq_state2power;
547		cooling_ops->power2state = cpufreq_power2state;
548	} else
549#endif
550	if (policy->freq_table_sorted == CPUFREQ_TABLE_UNSORTED) {
551		pr_err("%s: unsorted frequency tables are not supported\n",
552		       __func__);
553		cdev = ERR_PTR(-EINVAL);
554		goto free_idle_time;
555	}
556
557	ret = freq_qos_add_request(&policy->constraints,
558				   &cpufreq_cdev->qos_req, FREQ_QOS_MAX,
559				   get_state_freq(cpufreq_cdev, 0));
560	if (ret < 0) {
561		pr_err("%s: Failed to add freq constraint (%d)\n", __func__,
562		       ret);
563		cdev = ERR_PTR(ret);
564		goto free_idle_time;
565	}
566
567	cdev = ERR_PTR(-ENOMEM);
568	name = kasprintf(GFP_KERNEL, "cpufreq-%s", dev_name(dev));
569	if (!name)
570		goto remove_qos_req;
571
572	cdev = thermal_of_cooling_device_register(np, name, cpufreq_cdev,
573						  cooling_ops);
574	kfree(name);
575
576	if (IS_ERR(cdev))
577		goto remove_qos_req;
578
579	return cdev;
580
581remove_qos_req:
582	freq_qos_remove_request(&cpufreq_cdev->qos_req);
583free_idle_time:
584	free_idle_time(cpufreq_cdev);
585free_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 * "cpufreq-%s". This API can support multiple instances of cpufreq cooling
596 * devices.
597 *
598 * Return: a valid struct thermal_cooling_device pointer on success,
599 * on failure, it returns a corresponding ERR_PTR().
600 */
601struct thermal_cooling_device *
602cpufreq_cooling_register(struct cpufreq_policy *policy)
603{
604	return __cpufreq_cooling_register(NULL, policy, NULL);
605}
606EXPORT_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 * "cpufreq-%s". This API can support multiple instances of cpufreq cooling
614 * devices. Using this API, the cpufreq cooling device will be linked to the
615 * device tree node provided.
616 *
617 * Using this function, the cooling device will implement the power
618 * extensions by using the Energy Model (if present).  The cpus must have
619 * registered their OPPs using the OPP library.
620 *
621 * Return: a valid struct thermal_cooling_device pointer on success,
622 * and NULL on failure.
623 */
624struct thermal_cooling_device *
625of_cpufreq_cooling_register(struct cpufreq_policy *policy)
626{
627	struct device_node *np = of_get_cpu_node(policy->cpu, NULL);
628	struct thermal_cooling_device *cdev = NULL;
629
630	if (!np) {
631		pr_err("cpufreq_cooling: OF node not available for cpu%d\n",
632		       policy->cpu);
633		return NULL;
634	}
635
636	if (of_find_property(np, "#cooling-cells", NULL)) {
637		struct em_perf_domain *em = em_cpu_get(policy->cpu);
638
639		cdev = __cpufreq_cooling_register(np, policy, em);
640		if (IS_ERR(cdev)) {
641			pr_err("cpufreq_cooling: cpu%d failed to register as cooling device: %ld\n",
642			       policy->cpu, PTR_ERR(cdev));
643			cdev = NULL;
644		}
645	}
646
647	of_node_put(np);
648	return cdev;
649}
650EXPORT_SYMBOL_GPL(of_cpufreq_cooling_register);
651
652/**
653 * cpufreq_cooling_unregister - function to remove cpufreq cooling device.
654 * @cdev: thermal cooling device pointer.
655 *
656 * This interface function unregisters the "cpufreq-%x" cooling device.
657 */
658void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
659{
660	struct cpufreq_cooling_device *cpufreq_cdev;
661
662	if (!cdev)
663		return;
664
665	cpufreq_cdev = cdev->devdata;
666
667	thermal_cooling_device_unregister(cdev);
668	freq_qos_remove_request(&cpufreq_cdev->qos_req);
669	free_idle_time(cpufreq_cdev);
670	kfree(cpufreq_cdev);
671}
672EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister);