Loading...
Note: File does not exist in v6.13.7.
1/*
2 * linux/drivers/thermal/cpu_cooling.c
3 *
4 * Copyright (C) 2012 Samsung Electronics Co., Ltd(http://www.samsung.com)
5 * Copyright (C) 2012 Amit Daniel <amit.kachhap@linaro.org>
6 *
7 * Copyright (C) 2014 Viresh Kumar <viresh.kumar@linaro.org>
8 *
9 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; version 2 of the License.
13 *
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write to the Free Software Foundation, Inc.,
21 * 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
22 *
23 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
24 */
25#include <linux/module.h>
26#include <linux/thermal.h>
27#include <linux/cpufreq.h>
28#include <linux/err.h>
29#include <linux/pm_opp.h>
30#include <linux/slab.h>
31#include <linux/cpu.h>
32#include <linux/cpu_cooling.h>
33
34#include <trace/events/thermal.h>
35
36/*
37 * Cooling state <-> CPUFreq frequency
38 *
39 * Cooling states are translated to frequencies throughout this driver and this
40 * is the relation between them.
41 *
42 * Highest cooling state corresponds to lowest possible frequency.
43 *
44 * i.e.
45 * level 0 --> 1st Max Freq
46 * level 1 --> 2nd Max Freq
47 * ...
48 */
49
50/**
51 * struct power_table - frequency to power conversion
52 * @frequency: frequency in KHz
53 * @power: power in mW
54 *
55 * This structure is built when the cooling device registers and helps
56 * in translating frequency to power and viceversa.
57 */
58struct power_table {
59 u32 frequency;
60 u32 power;
61};
62
63/**
64 * struct cpufreq_cooling_device - data for cooling device with cpufreq
65 * @id: unique integer value corresponding to each cpufreq_cooling_device
66 * registered.
67 * @cool_dev: thermal_cooling_device pointer to keep track of the
68 * registered cooling device.
69 * @cpufreq_state: integer value representing the current state of cpufreq
70 * cooling devices.
71 * @clipped_freq: integer value representing the absolute value of the clipped
72 * frequency.
73 * @max_level: maximum cooling level. One less than total number of valid
74 * cpufreq frequencies.
75 * @allowed_cpus: all the cpus involved for this cpufreq_cooling_device.
76 * @node: list_head to link all cpufreq_cooling_device together.
77 * @last_load: load measured by the latest call to cpufreq_get_requested_power()
78 * @time_in_idle: previous reading of the absolute time that this cpu was idle
79 * @time_in_idle_timestamp: wall time of the last invocation of
80 * get_cpu_idle_time_us()
81 * @dyn_power_table: array of struct power_table for frequency to power
82 * conversion, sorted in ascending order.
83 * @dyn_power_table_entries: number of entries in the @dyn_power_table array
84 * @cpu_dev: the first cpu_device from @allowed_cpus that has OPPs registered
85 * @plat_get_static_power: callback to calculate the static power
86 *
87 * This structure is required for keeping information of each registered
88 * cpufreq_cooling_device.
89 */
90struct cpufreq_cooling_device {
91 int id;
92 struct thermal_cooling_device *cool_dev;
93 unsigned int cpufreq_state;
94 unsigned int clipped_freq;
95 unsigned int max_level;
96 unsigned int *freq_table; /* In descending order */
97 struct cpumask allowed_cpus;
98 struct list_head node;
99 u32 last_load;
100 u64 *time_in_idle;
101 u64 *time_in_idle_timestamp;
102 struct power_table *dyn_power_table;
103 int dyn_power_table_entries;
104 struct device *cpu_dev;
105 get_static_t plat_get_static_power;
106};
107static DEFINE_IDR(cpufreq_idr);
108static DEFINE_MUTEX(cooling_cpufreq_lock);
109
110static unsigned int cpufreq_dev_count;
111
112static DEFINE_MUTEX(cooling_list_lock);
113static LIST_HEAD(cpufreq_dev_list);
114
115/**
116 * get_idr - function to get a unique id.
117 * @idr: struct idr * handle used to create a id.
118 * @id: int * value generated by this function.
119 *
120 * This function will populate @id with an unique
121 * id, using the idr API.
122 *
123 * Return: 0 on success, an error code on failure.
124 */
125static int get_idr(struct idr *idr, int *id)
126{
127 int ret;
128
129 mutex_lock(&cooling_cpufreq_lock);
130 ret = idr_alloc(idr, NULL, 0, 0, GFP_KERNEL);
131 mutex_unlock(&cooling_cpufreq_lock);
132 if (unlikely(ret < 0))
133 return ret;
134 *id = ret;
135
136 return 0;
137}
138
139/**
140 * release_idr - function to free the unique id.
141 * @idr: struct idr * handle used for creating the id.
142 * @id: int value representing the unique id.
143 */
144static void release_idr(struct idr *idr, int id)
145{
146 mutex_lock(&cooling_cpufreq_lock);
147 idr_remove(idr, id);
148 mutex_unlock(&cooling_cpufreq_lock);
149}
150
151/* Below code defines functions to be used for cpufreq as cooling device */
152
153/**
154 * get_level: Find the level for a particular frequency
155 * @cpufreq_dev: cpufreq_dev for which the property is required
156 * @freq: Frequency
157 *
158 * Return: level on success, THERMAL_CSTATE_INVALID on error.
159 */
160static unsigned long get_level(struct cpufreq_cooling_device *cpufreq_dev,
161 unsigned int freq)
162{
163 unsigned long level;
164
165 for (level = 0; level <= cpufreq_dev->max_level; level++) {
166 if (freq == cpufreq_dev->freq_table[level])
167 return level;
168
169 if (freq > cpufreq_dev->freq_table[level])
170 break;
171 }
172
173 return THERMAL_CSTATE_INVALID;
174}
175
176/**
177 * cpufreq_cooling_get_level - for a given cpu, return the cooling level.
178 * @cpu: cpu for which the level is required
179 * @freq: the frequency of interest
180 *
181 * This function will match the cooling level corresponding to the
182 * requested @freq and return it.
183 *
184 * Return: The matched cooling level on success or THERMAL_CSTATE_INVALID
185 * otherwise.
186 */
187unsigned long cpufreq_cooling_get_level(unsigned int cpu, unsigned int freq)
188{
189 struct cpufreq_cooling_device *cpufreq_dev;
190
191 mutex_lock(&cooling_list_lock);
192 list_for_each_entry(cpufreq_dev, &cpufreq_dev_list, node) {
193 if (cpumask_test_cpu(cpu, &cpufreq_dev->allowed_cpus)) {
194 mutex_unlock(&cooling_list_lock);
195 return get_level(cpufreq_dev, freq);
196 }
197 }
198 mutex_unlock(&cooling_list_lock);
199
200 pr_err("%s: cpu:%d not part of any cooling device\n", __func__, cpu);
201 return THERMAL_CSTATE_INVALID;
202}
203EXPORT_SYMBOL_GPL(cpufreq_cooling_get_level);
204
205/**
206 * cpufreq_thermal_notifier - notifier callback for cpufreq policy change.
207 * @nb: struct notifier_block * with callback info.
208 * @event: value showing cpufreq event for which this function invoked.
209 * @data: callback-specific data
210 *
211 * Callback to hijack the notification on cpufreq policy transition.
212 * Every time there is a change in policy, we will intercept and
213 * update the cpufreq policy with thermal constraints.
214 *
215 * Return: 0 (success)
216 */
217static int cpufreq_thermal_notifier(struct notifier_block *nb,
218 unsigned long event, void *data)
219{
220 struct cpufreq_policy *policy = data;
221 unsigned long clipped_freq;
222 struct cpufreq_cooling_device *cpufreq_dev;
223
224 if (event != CPUFREQ_ADJUST)
225 return NOTIFY_DONE;
226
227 mutex_lock(&cooling_list_lock);
228 list_for_each_entry(cpufreq_dev, &cpufreq_dev_list, node) {
229 if (!cpumask_test_cpu(policy->cpu, &cpufreq_dev->allowed_cpus))
230 continue;
231
232 /*
233 * policy->max is the maximum allowed frequency defined by user
234 * and clipped_freq is the maximum that thermal constraints
235 * allow.
236 *
237 * If clipped_freq is lower than policy->max, then we need to
238 * readjust policy->max.
239 *
240 * But, if clipped_freq is greater than policy->max, we don't
241 * need to do anything.
242 */
243 clipped_freq = cpufreq_dev->clipped_freq;
244
245 if (policy->max > clipped_freq)
246 cpufreq_verify_within_limits(policy, 0, clipped_freq);
247 break;
248 }
249 mutex_unlock(&cooling_list_lock);
250
251 return NOTIFY_OK;
252}
253
254/**
255 * build_dyn_power_table() - create a dynamic power to frequency table
256 * @cpufreq_device: the cpufreq cooling device in which to store the table
257 * @capacitance: dynamic power coefficient for these cpus
258 *
259 * Build a dynamic power to frequency table for this cpu and store it
260 * in @cpufreq_device. This table will be used in cpu_power_to_freq() and
261 * cpu_freq_to_power() to convert between power and frequency
262 * efficiently. Power is stored in mW, frequency in KHz. The
263 * resulting table is in ascending order.
264 *
265 * Return: 0 on success, -EINVAL if there are no OPPs for any CPUs,
266 * -ENOMEM if we run out of memory or -EAGAIN if an OPP was
267 * added/enabled while the function was executing.
268 */
269static int build_dyn_power_table(struct cpufreq_cooling_device *cpufreq_device,
270 u32 capacitance)
271{
272 struct power_table *power_table;
273 struct dev_pm_opp *opp;
274 struct device *dev = NULL;
275 int num_opps = 0, cpu, i, ret = 0;
276 unsigned long freq;
277
278 for_each_cpu(cpu, &cpufreq_device->allowed_cpus) {
279 dev = get_cpu_device(cpu);
280 if (!dev) {
281 dev_warn(&cpufreq_device->cool_dev->device,
282 "No cpu device for cpu %d\n", cpu);
283 continue;
284 }
285
286 num_opps = dev_pm_opp_get_opp_count(dev);
287 if (num_opps > 0)
288 break;
289 else if (num_opps < 0)
290 return num_opps;
291 }
292
293 if (num_opps == 0)
294 return -EINVAL;
295
296 power_table = kcalloc(num_opps, sizeof(*power_table), GFP_KERNEL);
297 if (!power_table)
298 return -ENOMEM;
299
300 rcu_read_lock();
301
302 for (freq = 0, i = 0;
303 opp = dev_pm_opp_find_freq_ceil(dev, &freq), !IS_ERR(opp);
304 freq++, i++) {
305 u32 freq_mhz, voltage_mv;
306 u64 power;
307
308 if (i >= num_opps) {
309 rcu_read_unlock();
310 ret = -EAGAIN;
311 goto free_power_table;
312 }
313
314 freq_mhz = freq / 1000000;
315 voltage_mv = dev_pm_opp_get_voltage(opp) / 1000;
316
317 /*
318 * Do the multiplication with MHz and millivolt so as
319 * to not overflow.
320 */
321 power = (u64)capacitance * freq_mhz * voltage_mv * voltage_mv;
322 do_div(power, 1000000000);
323
324 /* frequency is stored in power_table in KHz */
325 power_table[i].frequency = freq / 1000;
326
327 /* power is stored in mW */
328 power_table[i].power = power;
329 }
330
331 rcu_read_unlock();
332
333 if (i != num_opps) {
334 ret = PTR_ERR(opp);
335 goto free_power_table;
336 }
337
338 cpufreq_device->cpu_dev = dev;
339 cpufreq_device->dyn_power_table = power_table;
340 cpufreq_device->dyn_power_table_entries = i;
341
342 return 0;
343
344free_power_table:
345 kfree(power_table);
346
347 return ret;
348}
349
350static u32 cpu_freq_to_power(struct cpufreq_cooling_device *cpufreq_device,
351 u32 freq)
352{
353 int i;
354 struct power_table *pt = cpufreq_device->dyn_power_table;
355
356 for (i = 1; i < cpufreq_device->dyn_power_table_entries; i++)
357 if (freq < pt[i].frequency)
358 break;
359
360 return pt[i - 1].power;
361}
362
363static u32 cpu_power_to_freq(struct cpufreq_cooling_device *cpufreq_device,
364 u32 power)
365{
366 int i;
367 struct power_table *pt = cpufreq_device->dyn_power_table;
368
369 for (i = 1; i < cpufreq_device->dyn_power_table_entries; i++)
370 if (power < pt[i].power)
371 break;
372
373 return pt[i - 1].frequency;
374}
375
376/**
377 * get_load() - get load for a cpu since last updated
378 * @cpufreq_device: &struct cpufreq_cooling_device for this cpu
379 * @cpu: cpu number
380 * @cpu_idx: index of the cpu in cpufreq_device->allowed_cpus
381 *
382 * Return: The average load of cpu @cpu in percentage since this
383 * function was last called.
384 */
385static u32 get_load(struct cpufreq_cooling_device *cpufreq_device, int cpu,
386 int cpu_idx)
387{
388 u32 load;
389 u64 now, now_idle, delta_time, delta_idle;
390
391 now_idle = get_cpu_idle_time(cpu, &now, 0);
392 delta_idle = now_idle - cpufreq_device->time_in_idle[cpu_idx];
393 delta_time = now - cpufreq_device->time_in_idle_timestamp[cpu_idx];
394
395 if (delta_time <= delta_idle)
396 load = 0;
397 else
398 load = div64_u64(100 * (delta_time - delta_idle), delta_time);
399
400 cpufreq_device->time_in_idle[cpu_idx] = now_idle;
401 cpufreq_device->time_in_idle_timestamp[cpu_idx] = now;
402
403 return load;
404}
405
406/**
407 * get_static_power() - calculate the static power consumed by the cpus
408 * @cpufreq_device: struct &cpufreq_cooling_device for this cpu cdev
409 * @tz: thermal zone device in which we're operating
410 * @freq: frequency in KHz
411 * @power: pointer in which to store the calculated static power
412 *
413 * Calculate the static power consumed by the cpus described by
414 * @cpu_actor running at frequency @freq. This function relies on a
415 * platform specific function that should have been provided when the
416 * actor was registered. If it wasn't, the static power is assumed to
417 * be negligible. The calculated static power is stored in @power.
418 *
419 * Return: 0 on success, -E* on failure.
420 */
421static int get_static_power(struct cpufreq_cooling_device *cpufreq_device,
422 struct thermal_zone_device *tz, unsigned long freq,
423 u32 *power)
424{
425 struct dev_pm_opp *opp;
426 unsigned long voltage;
427 struct cpumask *cpumask = &cpufreq_device->allowed_cpus;
428 unsigned long freq_hz = freq * 1000;
429
430 if (!cpufreq_device->plat_get_static_power ||
431 !cpufreq_device->cpu_dev) {
432 *power = 0;
433 return 0;
434 }
435
436 rcu_read_lock();
437
438 opp = dev_pm_opp_find_freq_exact(cpufreq_device->cpu_dev, freq_hz,
439 true);
440 voltage = dev_pm_opp_get_voltage(opp);
441
442 rcu_read_unlock();
443
444 if (voltage == 0) {
445 dev_warn_ratelimited(cpufreq_device->cpu_dev,
446 "Failed to get voltage for frequency %lu: %ld\n",
447 freq_hz, IS_ERR(opp) ? PTR_ERR(opp) : 0);
448 return -EINVAL;
449 }
450
451 return cpufreq_device->plat_get_static_power(cpumask, tz->passive_delay,
452 voltage, power);
453}
454
455/**
456 * get_dynamic_power() - calculate the dynamic power
457 * @cpufreq_device: &cpufreq_cooling_device for this cdev
458 * @freq: current frequency
459 *
460 * Return: the dynamic power consumed by the cpus described by
461 * @cpufreq_device.
462 */
463static u32 get_dynamic_power(struct cpufreq_cooling_device *cpufreq_device,
464 unsigned long freq)
465{
466 u32 raw_cpu_power;
467
468 raw_cpu_power = cpu_freq_to_power(cpufreq_device, freq);
469 return (raw_cpu_power * cpufreq_device->last_load) / 100;
470}
471
472/* cpufreq cooling device callback functions are defined below */
473
474/**
475 * cpufreq_get_max_state - callback function to get the max cooling state.
476 * @cdev: thermal cooling device pointer.
477 * @state: fill this variable with the max cooling state.
478 *
479 * Callback for the thermal cooling device to return the cpufreq
480 * max cooling state.
481 *
482 * Return: 0 on success, an error code otherwise.
483 */
484static int cpufreq_get_max_state(struct thermal_cooling_device *cdev,
485 unsigned long *state)
486{
487 struct cpufreq_cooling_device *cpufreq_device = cdev->devdata;
488
489 *state = cpufreq_device->max_level;
490 return 0;
491}
492
493/**
494 * cpufreq_get_cur_state - callback function to get the current cooling state.
495 * @cdev: thermal cooling device pointer.
496 * @state: fill this variable with the current cooling state.
497 *
498 * Callback for the thermal cooling device to return the cpufreq
499 * current cooling state.
500 *
501 * Return: 0 on success, an error code otherwise.
502 */
503static int cpufreq_get_cur_state(struct thermal_cooling_device *cdev,
504 unsigned long *state)
505{
506 struct cpufreq_cooling_device *cpufreq_device = cdev->devdata;
507
508 *state = cpufreq_device->cpufreq_state;
509
510 return 0;
511}
512
513/**
514 * cpufreq_set_cur_state - callback function to set the current cooling state.
515 * @cdev: thermal cooling device pointer.
516 * @state: set this variable to the current cooling state.
517 *
518 * Callback for the thermal cooling device to change the cpufreq
519 * current cooling state.
520 *
521 * Return: 0 on success, an error code otherwise.
522 */
523static int cpufreq_set_cur_state(struct thermal_cooling_device *cdev,
524 unsigned long state)
525{
526 struct cpufreq_cooling_device *cpufreq_device = cdev->devdata;
527 unsigned int cpu = cpumask_any(&cpufreq_device->allowed_cpus);
528 unsigned int clip_freq;
529
530 /* Request state should be less than max_level */
531 if (WARN_ON(state > cpufreq_device->max_level))
532 return -EINVAL;
533
534 /* Check if the old cooling action is same as new cooling action */
535 if (cpufreq_device->cpufreq_state == state)
536 return 0;
537
538 clip_freq = cpufreq_device->freq_table[state];
539 cpufreq_device->cpufreq_state = state;
540 cpufreq_device->clipped_freq = clip_freq;
541
542 cpufreq_update_policy(cpu);
543
544 return 0;
545}
546
547/**
548 * cpufreq_get_requested_power() - get the current power
549 * @cdev: &thermal_cooling_device pointer
550 * @tz: a valid thermal zone device pointer
551 * @power: pointer in which to store the resulting power
552 *
553 * Calculate the current power consumption of the cpus in milliwatts
554 * and store it in @power. This function should actually calculate
555 * the requested power, but it's hard to get the frequency that
556 * cpufreq would have assigned if there were no thermal limits.
557 * Instead, we calculate the current power on the assumption that the
558 * immediate future will look like the immediate past.
559 *
560 * We use the current frequency and the average load since this
561 * function was last called. In reality, there could have been
562 * multiple opps since this function was last called and that affects
563 * the load calculation. While it's not perfectly accurate, this
564 * simplification is good enough and works. REVISIT this, as more
565 * complex code may be needed if experiments show that it's not
566 * accurate enough.
567 *
568 * Return: 0 on success, -E* if getting the static power failed.
569 */
570static int cpufreq_get_requested_power(struct thermal_cooling_device *cdev,
571 struct thermal_zone_device *tz,
572 u32 *power)
573{
574 unsigned long freq;
575 int i = 0, cpu, ret;
576 u32 static_power, dynamic_power, total_load = 0;
577 struct cpufreq_cooling_device *cpufreq_device = cdev->devdata;
578 u32 *load_cpu = NULL;
579
580 cpu = cpumask_any_and(&cpufreq_device->allowed_cpus, cpu_online_mask);
581
582 /*
583 * All the CPUs are offline, thus the requested power by
584 * the cdev is 0
585 */
586 if (cpu >= nr_cpu_ids) {
587 *power = 0;
588 return 0;
589 }
590
591 freq = cpufreq_quick_get(cpu);
592
593 if (trace_thermal_power_cpu_get_power_enabled()) {
594 u32 ncpus = cpumask_weight(&cpufreq_device->allowed_cpus);
595
596 load_cpu = kcalloc(ncpus, sizeof(*load_cpu), GFP_KERNEL);
597 }
598
599 for_each_cpu(cpu, &cpufreq_device->allowed_cpus) {
600 u32 load;
601
602 if (cpu_online(cpu))
603 load = get_load(cpufreq_device, cpu, i);
604 else
605 load = 0;
606
607 total_load += load;
608 if (trace_thermal_power_cpu_limit_enabled() && load_cpu)
609 load_cpu[i] = load;
610
611 i++;
612 }
613
614 cpufreq_device->last_load = total_load;
615
616 dynamic_power = get_dynamic_power(cpufreq_device, freq);
617 ret = get_static_power(cpufreq_device, tz, freq, &static_power);
618 if (ret) {
619 kfree(load_cpu);
620 return ret;
621 }
622
623 if (load_cpu) {
624 trace_thermal_power_cpu_get_power(
625 &cpufreq_device->allowed_cpus,
626 freq, load_cpu, i, dynamic_power, static_power);
627
628 kfree(load_cpu);
629 }
630
631 *power = static_power + dynamic_power;
632 return 0;
633}
634
635/**
636 * cpufreq_state2power() - convert a cpu cdev state to power consumed
637 * @cdev: &thermal_cooling_device pointer
638 * @tz: a valid thermal zone device pointer
639 * @state: cooling device state to be converted
640 * @power: pointer in which to store the resulting power
641 *
642 * Convert cooling device state @state into power consumption in
643 * milliwatts assuming 100% load. Store the calculated power in
644 * @power.
645 *
646 * Return: 0 on success, -EINVAL if the cooling device state could not
647 * be converted into a frequency or other -E* if there was an error
648 * when calculating the static power.
649 */
650static int cpufreq_state2power(struct thermal_cooling_device *cdev,
651 struct thermal_zone_device *tz,
652 unsigned long state, u32 *power)
653{
654 unsigned int freq, num_cpus;
655 cpumask_t cpumask;
656 u32 static_power, dynamic_power;
657 int ret;
658 struct cpufreq_cooling_device *cpufreq_device = cdev->devdata;
659
660 cpumask_and(&cpumask, &cpufreq_device->allowed_cpus, cpu_online_mask);
661 num_cpus = cpumask_weight(&cpumask);
662
663 /* None of our cpus are online, so no power */
664 if (num_cpus == 0) {
665 *power = 0;
666 return 0;
667 }
668
669 freq = cpufreq_device->freq_table[state];
670 if (!freq)
671 return -EINVAL;
672
673 dynamic_power = cpu_freq_to_power(cpufreq_device, freq) * num_cpus;
674 ret = get_static_power(cpufreq_device, tz, freq, &static_power);
675 if (ret)
676 return ret;
677
678 *power = static_power + dynamic_power;
679 return 0;
680}
681
682/**
683 * cpufreq_power2state() - convert power to a cooling device state
684 * @cdev: &thermal_cooling_device pointer
685 * @tz: a valid thermal zone device pointer
686 * @power: power in milliwatts to be converted
687 * @state: pointer in which to store the resulting state
688 *
689 * Calculate a cooling device state for the cpus described by @cdev
690 * that would allow them to consume at most @power mW and store it in
691 * @state. Note that this calculation depends on external factors
692 * such as the cpu load or the current static power. Calling this
693 * function with the same power as input can yield different cooling
694 * device states depending on those external factors.
695 *
696 * Return: 0 on success, -ENODEV if no cpus are online or -EINVAL if
697 * the calculated frequency could not be converted to a valid state.
698 * The latter should not happen unless the frequencies available to
699 * cpufreq have changed since the initialization of the cpu cooling
700 * device.
701 */
702static int cpufreq_power2state(struct thermal_cooling_device *cdev,
703 struct thermal_zone_device *tz, u32 power,
704 unsigned long *state)
705{
706 unsigned int cpu, cur_freq, target_freq;
707 int ret;
708 s32 dyn_power;
709 u32 last_load, normalised_power, static_power;
710 struct cpufreq_cooling_device *cpufreq_device = cdev->devdata;
711
712 cpu = cpumask_any_and(&cpufreq_device->allowed_cpus, cpu_online_mask);
713
714 /* None of our cpus are online */
715 if (cpu >= nr_cpu_ids)
716 return -ENODEV;
717
718 cur_freq = cpufreq_quick_get(cpu);
719 ret = get_static_power(cpufreq_device, tz, cur_freq, &static_power);
720 if (ret)
721 return ret;
722
723 dyn_power = power - static_power;
724 dyn_power = dyn_power > 0 ? dyn_power : 0;
725 last_load = cpufreq_device->last_load ?: 1;
726 normalised_power = (dyn_power * 100) / last_load;
727 target_freq = cpu_power_to_freq(cpufreq_device, normalised_power);
728
729 *state = cpufreq_cooling_get_level(cpu, target_freq);
730 if (*state == THERMAL_CSTATE_INVALID) {
731 dev_warn_ratelimited(&cdev->device,
732 "Failed to convert %dKHz for cpu %d into a cdev state\n",
733 target_freq, cpu);
734 return -EINVAL;
735 }
736
737 trace_thermal_power_cpu_limit(&cpufreq_device->allowed_cpus,
738 target_freq, *state, power);
739 return 0;
740}
741
742/* Bind cpufreq callbacks to thermal cooling device ops */
743
744static struct thermal_cooling_device_ops cpufreq_cooling_ops = {
745 .get_max_state = cpufreq_get_max_state,
746 .get_cur_state = cpufreq_get_cur_state,
747 .set_cur_state = cpufreq_set_cur_state,
748};
749
750static struct thermal_cooling_device_ops cpufreq_power_cooling_ops = {
751 .get_max_state = cpufreq_get_max_state,
752 .get_cur_state = cpufreq_get_cur_state,
753 .set_cur_state = cpufreq_set_cur_state,
754 .get_requested_power = cpufreq_get_requested_power,
755 .state2power = cpufreq_state2power,
756 .power2state = cpufreq_power2state,
757};
758
759/* Notifier for cpufreq policy change */
760static struct notifier_block thermal_cpufreq_notifier_block = {
761 .notifier_call = cpufreq_thermal_notifier,
762};
763
764static unsigned int find_next_max(struct cpufreq_frequency_table *table,
765 unsigned int prev_max)
766{
767 struct cpufreq_frequency_table *pos;
768 unsigned int max = 0;
769
770 cpufreq_for_each_valid_entry(pos, table) {
771 if (pos->frequency > max && pos->frequency < prev_max)
772 max = pos->frequency;
773 }
774
775 return max;
776}
777
778/**
779 * __cpufreq_cooling_register - helper function to create cpufreq cooling device
780 * @np: a valid struct device_node to the cooling device device tree node
781 * @clip_cpus: cpumask of cpus where the frequency constraints will happen.
782 * Normally this should be same as cpufreq policy->related_cpus.
783 * @capacitance: dynamic power coefficient for these cpus
784 * @plat_static_func: function to calculate the static power consumed by these
785 * cpus (optional)
786 *
787 * This interface function registers the cpufreq cooling device with the name
788 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
789 * cooling devices. It also gives the opportunity to link the cooling device
790 * with a device tree node, in order to bind it via the thermal DT code.
791 *
792 * Return: a valid struct thermal_cooling_device pointer on success,
793 * on failure, it returns a corresponding ERR_PTR().
794 */
795static struct thermal_cooling_device *
796__cpufreq_cooling_register(struct device_node *np,
797 const struct cpumask *clip_cpus, u32 capacitance,
798 get_static_t plat_static_func)
799{
800 struct cpufreq_policy *policy;
801 struct thermal_cooling_device *cool_dev;
802 struct cpufreq_cooling_device *cpufreq_dev;
803 char dev_name[THERMAL_NAME_LENGTH];
804 struct cpufreq_frequency_table *pos, *table;
805 struct cpumask temp_mask;
806 unsigned int freq, i, num_cpus;
807 int ret;
808 struct thermal_cooling_device_ops *cooling_ops;
809
810 cpumask_and(&temp_mask, clip_cpus, cpu_online_mask);
811 policy = cpufreq_cpu_get(cpumask_first(&temp_mask));
812 if (!policy) {
813 pr_debug("%s: CPUFreq policy not found\n", __func__);
814 return ERR_PTR(-EPROBE_DEFER);
815 }
816
817 table = policy->freq_table;
818 if (!table) {
819 pr_debug("%s: CPUFreq table not found\n", __func__);
820 cool_dev = ERR_PTR(-ENODEV);
821 goto put_policy;
822 }
823
824 cpufreq_dev = kzalloc(sizeof(*cpufreq_dev), GFP_KERNEL);
825 if (!cpufreq_dev) {
826 cool_dev = ERR_PTR(-ENOMEM);
827 goto put_policy;
828 }
829
830 num_cpus = cpumask_weight(clip_cpus);
831 cpufreq_dev->time_in_idle = kcalloc(num_cpus,
832 sizeof(*cpufreq_dev->time_in_idle),
833 GFP_KERNEL);
834 if (!cpufreq_dev->time_in_idle) {
835 cool_dev = ERR_PTR(-ENOMEM);
836 goto free_cdev;
837 }
838
839 cpufreq_dev->time_in_idle_timestamp =
840 kcalloc(num_cpus, sizeof(*cpufreq_dev->time_in_idle_timestamp),
841 GFP_KERNEL);
842 if (!cpufreq_dev->time_in_idle_timestamp) {
843 cool_dev = ERR_PTR(-ENOMEM);
844 goto free_time_in_idle;
845 }
846
847 /* Find max levels */
848 cpufreq_for_each_valid_entry(pos, table)
849 cpufreq_dev->max_level++;
850
851 cpufreq_dev->freq_table = kmalloc(sizeof(*cpufreq_dev->freq_table) *
852 cpufreq_dev->max_level, GFP_KERNEL);
853 if (!cpufreq_dev->freq_table) {
854 cool_dev = ERR_PTR(-ENOMEM);
855 goto free_time_in_idle_timestamp;
856 }
857
858 /* max_level is an index, not a counter */
859 cpufreq_dev->max_level--;
860
861 cpumask_copy(&cpufreq_dev->allowed_cpus, clip_cpus);
862
863 if (capacitance) {
864 cpufreq_dev->plat_get_static_power = plat_static_func;
865
866 ret = build_dyn_power_table(cpufreq_dev, capacitance);
867 if (ret) {
868 cool_dev = ERR_PTR(ret);
869 goto free_table;
870 }
871
872 cooling_ops = &cpufreq_power_cooling_ops;
873 } else {
874 cooling_ops = &cpufreq_cooling_ops;
875 }
876
877 ret = get_idr(&cpufreq_idr, &cpufreq_dev->id);
878 if (ret) {
879 cool_dev = ERR_PTR(ret);
880 goto free_power_table;
881 }
882
883 /* Fill freq-table in descending order of frequencies */
884 for (i = 0, freq = -1; i <= cpufreq_dev->max_level; i++) {
885 freq = find_next_max(table, freq);
886 cpufreq_dev->freq_table[i] = freq;
887
888 /* Warn for duplicate entries */
889 if (!freq)
890 pr_warn("%s: table has duplicate entries\n", __func__);
891 else
892 pr_debug("%s: freq:%u KHz\n", __func__, freq);
893 }
894
895 snprintf(dev_name, sizeof(dev_name), "thermal-cpufreq-%d",
896 cpufreq_dev->id);
897
898 cool_dev = thermal_of_cooling_device_register(np, dev_name, cpufreq_dev,
899 cooling_ops);
900 if (IS_ERR(cool_dev))
901 goto remove_idr;
902
903 cpufreq_dev->clipped_freq = cpufreq_dev->freq_table[0];
904 cpufreq_dev->cool_dev = cool_dev;
905
906 mutex_lock(&cooling_cpufreq_lock);
907
908 mutex_lock(&cooling_list_lock);
909 list_add(&cpufreq_dev->node, &cpufreq_dev_list);
910 mutex_unlock(&cooling_list_lock);
911
912 /* Register the notifier for first cpufreq cooling device */
913 if (!cpufreq_dev_count++)
914 cpufreq_register_notifier(&thermal_cpufreq_notifier_block,
915 CPUFREQ_POLICY_NOTIFIER);
916 mutex_unlock(&cooling_cpufreq_lock);
917
918 goto put_policy;
919
920remove_idr:
921 release_idr(&cpufreq_idr, cpufreq_dev->id);
922free_power_table:
923 kfree(cpufreq_dev->dyn_power_table);
924free_table:
925 kfree(cpufreq_dev->freq_table);
926free_time_in_idle_timestamp:
927 kfree(cpufreq_dev->time_in_idle_timestamp);
928free_time_in_idle:
929 kfree(cpufreq_dev->time_in_idle);
930free_cdev:
931 kfree(cpufreq_dev);
932put_policy:
933 cpufreq_cpu_put(policy);
934
935 return cool_dev;
936}
937
938/**
939 * cpufreq_cooling_register - function to create cpufreq cooling device.
940 * @clip_cpus: cpumask of cpus where the frequency constraints will happen.
941 *
942 * This interface function registers the cpufreq cooling device with the name
943 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
944 * cooling devices.
945 *
946 * Return: a valid struct thermal_cooling_device pointer on success,
947 * on failure, it returns a corresponding ERR_PTR().
948 */
949struct thermal_cooling_device *
950cpufreq_cooling_register(const struct cpumask *clip_cpus)
951{
952 return __cpufreq_cooling_register(NULL, clip_cpus, 0, NULL);
953}
954EXPORT_SYMBOL_GPL(cpufreq_cooling_register);
955
956/**
957 * of_cpufreq_cooling_register - function to create cpufreq cooling device.
958 * @np: a valid struct device_node to the cooling device device tree node
959 * @clip_cpus: cpumask of cpus where the frequency constraints will happen.
960 *
961 * This interface function registers the cpufreq cooling device with the name
962 * "thermal-cpufreq-%x". This api can support multiple instances of cpufreq
963 * cooling devices. Using this API, the cpufreq cooling device will be
964 * linked to the device tree node provided.
965 *
966 * Return: a valid struct thermal_cooling_device pointer on success,
967 * on failure, it returns a corresponding ERR_PTR().
968 */
969struct thermal_cooling_device *
970of_cpufreq_cooling_register(struct device_node *np,
971 const struct cpumask *clip_cpus)
972{
973 if (!np)
974 return ERR_PTR(-EINVAL);
975
976 return __cpufreq_cooling_register(np, clip_cpus, 0, NULL);
977}
978EXPORT_SYMBOL_GPL(of_cpufreq_cooling_register);
979
980/**
981 * cpufreq_power_cooling_register() - create cpufreq cooling device with power extensions
982 * @clip_cpus: cpumask of cpus where the frequency constraints will happen
983 * @capacitance: dynamic power coefficient for these cpus
984 * @plat_static_func: function to calculate the static power consumed by these
985 * cpus (optional)
986 *
987 * This interface function registers the cpufreq cooling device with
988 * the name "thermal-cpufreq-%x". This api can support multiple
989 * instances of cpufreq cooling devices. Using this function, the
990 * cooling device will implement the power extensions by using a
991 * simple cpu power model. The cpus must have registered their OPPs
992 * using the OPP library.
993 *
994 * An optional @plat_static_func may be provided to calculate the
995 * static power consumed by these cpus. If the platform's static
996 * power consumption is unknown or negligible, make it NULL.
997 *
998 * Return: a valid struct thermal_cooling_device pointer on success,
999 * on failure, it returns a corresponding ERR_PTR().
1000 */
1001struct thermal_cooling_device *
1002cpufreq_power_cooling_register(const struct cpumask *clip_cpus, u32 capacitance,
1003 get_static_t plat_static_func)
1004{
1005 return __cpufreq_cooling_register(NULL, clip_cpus, capacitance,
1006 plat_static_func);
1007}
1008EXPORT_SYMBOL(cpufreq_power_cooling_register);
1009
1010/**
1011 * of_cpufreq_power_cooling_register() - create cpufreq cooling device with power extensions
1012 * @np: a valid struct device_node to the cooling device device tree node
1013 * @clip_cpus: cpumask of cpus where the frequency constraints will happen
1014 * @capacitance: dynamic power coefficient for these cpus
1015 * @plat_static_func: function to calculate the static power consumed by these
1016 * cpus (optional)
1017 *
1018 * This interface function registers the cpufreq cooling device with
1019 * the name "thermal-cpufreq-%x". This api can support multiple
1020 * instances of cpufreq cooling devices. Using this API, the cpufreq
1021 * cooling device will be linked to the device tree node provided.
1022 * Using this function, the cooling device will implement the power
1023 * extensions by using a simple cpu power model. The cpus must have
1024 * registered their OPPs using the OPP library.
1025 *
1026 * An optional @plat_static_func may be provided to calculate the
1027 * static power consumed by these cpus. If the platform's static
1028 * power consumption is unknown or negligible, make it NULL.
1029 *
1030 * Return: a valid struct thermal_cooling_device pointer on success,
1031 * on failure, it returns a corresponding ERR_PTR().
1032 */
1033struct thermal_cooling_device *
1034of_cpufreq_power_cooling_register(struct device_node *np,
1035 const struct cpumask *clip_cpus,
1036 u32 capacitance,
1037 get_static_t plat_static_func)
1038{
1039 if (!np)
1040 return ERR_PTR(-EINVAL);
1041
1042 return __cpufreq_cooling_register(np, clip_cpus, capacitance,
1043 plat_static_func);
1044}
1045EXPORT_SYMBOL(of_cpufreq_power_cooling_register);
1046
1047/**
1048 * cpufreq_cooling_unregister - function to remove cpufreq cooling device.
1049 * @cdev: thermal cooling device pointer.
1050 *
1051 * This interface function unregisters the "thermal-cpufreq-%x" cooling device.
1052 */
1053void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
1054{
1055 struct cpufreq_cooling_device *cpufreq_dev;
1056
1057 if (!cdev)
1058 return;
1059
1060 cpufreq_dev = cdev->devdata;
1061
1062 /* Unregister the notifier for the last cpufreq cooling device */
1063 mutex_lock(&cooling_cpufreq_lock);
1064 if (!--cpufreq_dev_count)
1065 cpufreq_unregister_notifier(&thermal_cpufreq_notifier_block,
1066 CPUFREQ_POLICY_NOTIFIER);
1067
1068 mutex_lock(&cooling_list_lock);
1069 list_del(&cpufreq_dev->node);
1070 mutex_unlock(&cooling_list_lock);
1071
1072 mutex_unlock(&cooling_cpufreq_lock);
1073
1074 thermal_cooling_device_unregister(cpufreq_dev->cool_dev);
1075 release_idr(&cpufreq_idr, cpufreq_dev->id);
1076 kfree(cpufreq_dev->dyn_power_table);
1077 kfree(cpufreq_dev->time_in_idle_timestamp);
1078 kfree(cpufreq_dev->time_in_idle);
1079 kfree(cpufreq_dev->freq_table);
1080 kfree(cpufreq_dev);
1081}
1082EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister);