1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * A power allocator to manage temperature
4 *
5 * Copyright (C) 2014 ARM Ltd.
6 *
7 */
8
9 #define pr_fmt(fmt) "Power allocator: " fmt
10
11 #include <linux/rculist.h>
12 #include <linux/slab.h>
13 #include <linux/thermal.h>
14
15 #define CREATE_TRACE_POINTS
16 #include <trace/events/thermal_power_allocator.h>
17
18 #include "thermal_core.h"
19
20 #define INVALID_TRIP -1
21
22 #define FRAC_BITS 10
23 #define int_to_frac(x) ((x) << FRAC_BITS)
24 #define frac_to_int(x) ((x) >> FRAC_BITS)
25
26 /**
27 * mul_frac() - multiply two fixed-point numbers
28 * @x: first multiplicand
29 * @y: second multiplicand
30 *
31 * Return: the result of multiplying two fixed-point numbers. The
32 * result is also a fixed-point number.
33 */
mul_frac(s64 x,s64 y)34 static inline s64 mul_frac(s64 x, s64 y)
35 {
36 return (x * y) >> FRAC_BITS;
37 }
38
39 /**
40 * div_frac() - divide two fixed-point numbers
41 * @x: the dividend
42 * @y: the divisor
43 *
44 * Return: the result of dividing two fixed-point numbers. The
45 * result is also a fixed-point number.
46 */
div_frac(s64 x,s64 y)47 static inline s64 div_frac(s64 x, s64 y)
48 {
49 return div_s64(x << FRAC_BITS, y);
50 }
51
52 /**
53 * struct power_allocator_params - parameters for the power allocator governor
54 * @allocated_tzp: whether we have allocated tzp for this thermal zone and
55 * it needs to be freed on unbind
56 * @err_integral: accumulated error in the PID controller.
57 * @prev_err: error in the previous iteration of the PID controller.
58 * Used to calculate the derivative term.
59 * @trip_switch_on: first passive trip point of the thermal zone. The
60 * governor switches on when this trip point is crossed.
61 * If the thermal zone only has one passive trip point,
62 * @trip_switch_on should be INVALID_TRIP.
63 * @trip_max_desired_temperature: last passive trip point of the thermal
64 * zone. The temperature we are
65 * controlling for.
66 */
67 struct power_allocator_params {
68 bool allocated_tzp;
69 s64 err_integral;
70 s32 prev_err;
71 int trip_switch_on;
72 int trip_max_desired_temperature;
73 };
74
75 /**
76 * estimate_sustainable_power() - Estimate the sustainable power of a thermal zone
77 * @tz: thermal zone we are operating in
78 *
79 * For thermal zones that don't provide a sustainable_power in their
80 * thermal_zone_params, estimate one. Calculate it using the minimum
81 * power of all the cooling devices as that gives a valid value that
82 * can give some degree of functionality. For optimal performance of
83 * this governor, provide a sustainable_power in the thermal zone's
84 * thermal_zone_params.
85 */
estimate_sustainable_power(struct thermal_zone_device * tz)86 static u32 estimate_sustainable_power(struct thermal_zone_device *tz)
87 {
88 u32 sustainable_power = 0;
89 struct thermal_instance *instance;
90 struct power_allocator_params *params = tz->governor_data;
91
92 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
93 struct thermal_cooling_device *cdev = instance->cdev;
94 u32 min_power;
95
96 if (instance->trip != params->trip_max_desired_temperature)
97 continue;
98
99 if (power_actor_get_min_power(cdev, &min_power))
100 continue;
101
102 sustainable_power += min_power;
103 }
104
105 return sustainable_power;
106 }
107
108 /**
109 * estimate_pid_constants() - Estimate the constants for the PID controller
110 * @tz: thermal zone for which to estimate the constants
111 * @sustainable_power: sustainable power for the thermal zone
112 * @trip_switch_on: trip point number for the switch on temperature
113 * @control_temp: target temperature for the power allocator governor
114 * @force: whether to force the update of the constants
115 *
116 * This function is used to update the estimation of the PID
117 * controller constants in struct thermal_zone_parameters.
118 * Sustainable power is provided in case it was estimated. The
119 * estimated sustainable_power should not be stored in the
120 * thermal_zone_parameters so it has to be passed explicitly to this
121 * function.
122 *
123 * If @force is not set, the values in the thermal zone's parameters
124 * are preserved if they are not zero. If @force is set, the values
125 * in thermal zone's parameters are overwritten.
126 */
estimate_pid_constants(struct thermal_zone_device * tz,u32 sustainable_power,int trip_switch_on,int control_temp,bool force)127 static void estimate_pid_constants(struct thermal_zone_device *tz,
128 u32 sustainable_power, int trip_switch_on,
129 int control_temp, bool force)
130 {
131 int ret;
132 int switch_on_temp;
133 u32 temperature_threshold;
134
135 ret = tz->ops->get_trip_temp(tz, trip_switch_on, &switch_on_temp);
136 if (ret)
137 switch_on_temp = 0;
138
139 temperature_threshold = control_temp - switch_on_temp;
140 /*
141 * estimate_pid_constants() tries to find appropriate default
142 * values for thermal zones that don't provide them. If a
143 * system integrator has configured a thermal zone with two
144 * passive trip points at the same temperature, that person
145 * hasn't put any effort to set up the thermal zone properly
146 * so just give up.
147 */
148 if (!temperature_threshold)
149 return;
150
151 if (!tz->tzp->k_po || force)
152 tz->tzp->k_po = int_to_frac(sustainable_power) /
153 temperature_threshold;
154
155 if (!tz->tzp->k_pu || force)
156 tz->tzp->k_pu = int_to_frac(2 * sustainable_power) /
157 temperature_threshold;
158
159 if (!tz->tzp->k_i || force)
160 tz->tzp->k_i = int_to_frac(10) / 1000;
161 /*
162 * The default for k_d and integral_cutoff is 0, so we can
163 * leave them as they are.
164 */
165 }
166
167 /**
168 * pid_controller() - PID controller
169 * @tz: thermal zone we are operating in
170 * @control_temp: the target temperature in millicelsius
171 * @max_allocatable_power: maximum allocatable power for this thermal zone
172 *
173 * This PID controller increases the available power budget so that the
174 * temperature of the thermal zone gets as close as possible to
175 * @control_temp and limits the power if it exceeds it. k_po is the
176 * proportional term when we are overshooting, k_pu is the
177 * proportional term when we are undershooting. integral_cutoff is a
178 * threshold below which we stop accumulating the error. The
179 * accumulated error is only valid if the requested power will make
180 * the system warmer. If the system is mostly idle, there's no point
181 * in accumulating positive error.
182 *
183 * Return: The power budget for the next period.
184 */
pid_controller(struct thermal_zone_device * tz,int control_temp,u32 max_allocatable_power)185 static u32 pid_controller(struct thermal_zone_device *tz,
186 int control_temp,
187 u32 max_allocatable_power)
188 {
189 s64 p, i, d, power_range;
190 s32 err, max_power_frac;
191 u32 sustainable_power;
192 struct power_allocator_params *params = tz->governor_data;
193
194 max_power_frac = int_to_frac(max_allocatable_power);
195
196 if (tz->tzp->sustainable_power) {
197 sustainable_power = tz->tzp->sustainable_power;
198 } else {
199 sustainable_power = estimate_sustainable_power(tz);
200 estimate_pid_constants(tz, sustainable_power,
201 params->trip_switch_on, control_temp,
202 true);
203 }
204
205 err = control_temp - tz->temperature;
206 err = int_to_frac(err);
207
208 /* Calculate the proportional term */
209 p = mul_frac(err < 0 ? tz->tzp->k_po : tz->tzp->k_pu, err);
210
211 /*
212 * Calculate the integral term
213 *
214 * if the error is less than cut off allow integration (but
215 * the integral is limited to max power)
216 */
217 i = mul_frac(tz->tzp->k_i, params->err_integral);
218
219 if (err < int_to_frac(tz->tzp->integral_cutoff)) {
220 s64 i_next = i + mul_frac(tz->tzp->k_i, err);
221
222 if (abs(i_next) < max_power_frac) {
223 i = i_next;
224 params->err_integral += err;
225 }
226 }
227
228 /*
229 * Calculate the derivative term
230 *
231 * We do err - prev_err, so with a positive k_d, a decreasing
232 * error (i.e. driving closer to the line) results in less
233 * power being applied, slowing down the controller)
234 */
235 d = mul_frac(tz->tzp->k_d, err - params->prev_err);
236 d = div_frac(d, tz->passive_delay);
237 params->prev_err = err;
238
239 power_range = p + i + d;
240
241 /* feed-forward the known sustainable dissipatable power */
242 power_range = sustainable_power + frac_to_int(power_range);
243
244 power_range = clamp(power_range, (s64)0, (s64)max_allocatable_power);
245
246 trace_thermal_power_allocator_pid(tz, frac_to_int(err),
247 frac_to_int(params->err_integral),
248 frac_to_int(p), frac_to_int(i),
249 frac_to_int(d), power_range);
250
251 return power_range;
252 }
253
254 /**
255 * divvy_up_power() - divvy the allocated power between the actors
256 * @req_power: each actor's requested power
257 * @max_power: each actor's maximum available power
258 * @num_actors: size of the @req_power, @max_power and @granted_power's array
259 * @total_req_power: sum of @req_power
260 * @power_range: total allocated power
261 * @granted_power: output array: each actor's granted power
262 * @extra_actor_power: an appropriately sized array to be used in the
263 * function as temporary storage of the extra power given
264 * to the actors
265 *
266 * This function divides the total allocated power (@power_range)
267 * fairly between the actors. It first tries to give each actor a
268 * share of the @power_range according to how much power it requested
269 * compared to the rest of the actors. For example, if only one actor
270 * requests power, then it receives all the @power_range. If
271 * three actors each requests 1mW, each receives a third of the
272 * @power_range.
273 *
274 * If any actor received more than their maximum power, then that
275 * surplus is re-divvied among the actors based on how far they are
276 * from their respective maximums.
277 *
278 * Granted power for each actor is written to @granted_power, which
279 * should've been allocated by the calling function.
280 */
divvy_up_power(u32 * req_power,u32 * max_power,int num_actors,u32 total_req_power,u32 power_range,u32 * granted_power,u32 * extra_actor_power)281 static void divvy_up_power(u32 *req_power, u32 *max_power, int num_actors,
282 u32 total_req_power, u32 power_range,
283 u32 *granted_power, u32 *extra_actor_power)
284 {
285 u32 extra_power, capped_extra_power;
286 int i;
287
288 /*
289 * Prevent division by 0 if none of the actors request power.
290 */
291 if (!total_req_power)
292 total_req_power = 1;
293
294 capped_extra_power = 0;
295 extra_power = 0;
296 for (i = 0; i < num_actors; i++) {
297 u64 req_range = (u64)req_power[i] * power_range;
298
299 granted_power[i] = DIV_ROUND_CLOSEST_ULL(req_range,
300 total_req_power);
301
302 if (granted_power[i] > max_power[i]) {
303 extra_power += granted_power[i] - max_power[i];
304 granted_power[i] = max_power[i];
305 }
306
307 extra_actor_power[i] = max_power[i] - granted_power[i];
308 capped_extra_power += extra_actor_power[i];
309 }
310
311 if (!extra_power)
312 return;
313
314 /*
315 * Re-divvy the reclaimed extra among actors based on
316 * how far they are from the max
317 */
318 extra_power = min(extra_power, capped_extra_power);
319 if (capped_extra_power > 0)
320 for (i = 0; i < num_actors; i++)
321 granted_power[i] += (extra_actor_power[i] *
322 extra_power) / capped_extra_power;
323 }
324
allocate_power(struct thermal_zone_device * tz,int control_temp)325 static int allocate_power(struct thermal_zone_device *tz,
326 int control_temp)
327 {
328 struct thermal_instance *instance;
329 struct power_allocator_params *params = tz->governor_data;
330 u32 *req_power, *max_power, *granted_power, *extra_actor_power;
331 u32 *weighted_req_power;
332 u32 total_req_power, max_allocatable_power, total_weighted_req_power;
333 u32 total_granted_power, power_range;
334 int i, num_actors, total_weight, ret = 0;
335 int trip_max_desired_temperature = params->trip_max_desired_temperature;
336
337 mutex_lock(&tz->lock);
338
339 num_actors = 0;
340 total_weight = 0;
341 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
342 if ((instance->trip == trip_max_desired_temperature) &&
343 cdev_is_power_actor(instance->cdev)) {
344 num_actors++;
345 total_weight += instance->weight;
346 }
347 }
348
349 if (!num_actors) {
350 ret = -ENODEV;
351 goto unlock;
352 }
353
354 /*
355 * We need to allocate five arrays of the same size:
356 * req_power, max_power, granted_power, extra_actor_power and
357 * weighted_req_power. They are going to be needed until this
358 * function returns. Allocate them all in one go to simplify
359 * the allocation and deallocation logic.
360 */
361 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*max_power));
362 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*granted_power));
363 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*extra_actor_power));
364 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*weighted_req_power));
365 req_power = kcalloc(num_actors * 5, sizeof(*req_power), GFP_KERNEL);
366 if (!req_power) {
367 ret = -ENOMEM;
368 goto unlock;
369 }
370
371 max_power = &req_power[num_actors];
372 granted_power = &req_power[2 * num_actors];
373 extra_actor_power = &req_power[3 * num_actors];
374 weighted_req_power = &req_power[4 * num_actors];
375
376 i = 0;
377 total_weighted_req_power = 0;
378 total_req_power = 0;
379 max_allocatable_power = 0;
380
381 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
382 int weight;
383 struct thermal_cooling_device *cdev = instance->cdev;
384
385 if (instance->trip != trip_max_desired_temperature)
386 continue;
387
388 if (!cdev_is_power_actor(cdev))
389 continue;
390
391 if (cdev->ops->get_requested_power(cdev, &req_power[i]))
392 continue;
393
394 if (!total_weight)
395 weight = 1 << FRAC_BITS;
396 else
397 weight = instance->weight;
398
399 weighted_req_power[i] = frac_to_int(weight * req_power[i]);
400
401 if (power_actor_get_max_power(cdev, &max_power[i]))
402 continue;
403
404 total_req_power += req_power[i];
405 max_allocatable_power += max_power[i];
406 total_weighted_req_power += weighted_req_power[i];
407
408 i++;
409 }
410
411 power_range = pid_controller(tz, control_temp, max_allocatable_power);
412
413 divvy_up_power(weighted_req_power, max_power, num_actors,
414 total_weighted_req_power, power_range, granted_power,
415 extra_actor_power);
416
417 total_granted_power = 0;
418 i = 0;
419 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
420 if (instance->trip != trip_max_desired_temperature)
421 continue;
422
423 if (!cdev_is_power_actor(instance->cdev))
424 continue;
425
426 power_actor_set_power(instance->cdev, instance,
427 granted_power[i]);
428 total_granted_power += granted_power[i];
429
430 i++;
431 }
432
433 trace_thermal_power_allocator(tz, req_power, total_req_power,
434 granted_power, total_granted_power,
435 num_actors, power_range,
436 max_allocatable_power, tz->temperature,
437 control_temp - tz->temperature);
438
439 kfree(req_power);
440 unlock:
441 mutex_unlock(&tz->lock);
442
443 return ret;
444 }
445
446 /**
447 * get_governor_trips() - get the number of the two trip points that are key for this governor
448 * @tz: thermal zone to operate on
449 * @params: pointer to private data for this governor
450 *
451 * The power allocator governor works optimally with two trips points:
452 * a "switch on" trip point and a "maximum desired temperature". These
453 * are defined as the first and last passive trip points.
454 *
455 * If there is only one trip point, then that's considered to be the
456 * "maximum desired temperature" trip point and the governor is always
457 * on. If there are no passive or active trip points, then the
458 * governor won't do anything. In fact, its throttle function
459 * won't be called at all.
460 */
get_governor_trips(struct thermal_zone_device * tz,struct power_allocator_params * params)461 static void get_governor_trips(struct thermal_zone_device *tz,
462 struct power_allocator_params *params)
463 {
464 int i, last_active, last_passive;
465 bool found_first_passive;
466
467 found_first_passive = false;
468 last_active = INVALID_TRIP;
469 last_passive = INVALID_TRIP;
470
471 for (i = 0; i < tz->trips; i++) {
472 enum thermal_trip_type type;
473 int ret;
474
475 ret = tz->ops->get_trip_type(tz, i, &type);
476 if (ret) {
477 dev_warn(&tz->device,
478 "Failed to get trip point %d type: %d\n", i,
479 ret);
480 continue;
481 }
482
483 if (type == THERMAL_TRIP_PASSIVE) {
484 if (!found_first_passive) {
485 params->trip_switch_on = i;
486 found_first_passive = true;
487 } else {
488 last_passive = i;
489 }
490 } else if (type == THERMAL_TRIP_ACTIVE) {
491 last_active = i;
492 } else {
493 break;
494 }
495 }
496
497 if (last_passive != INVALID_TRIP) {
498 params->trip_max_desired_temperature = last_passive;
499 } else if (found_first_passive) {
500 params->trip_max_desired_temperature = params->trip_switch_on;
501 params->trip_switch_on = INVALID_TRIP;
502 } else {
503 params->trip_switch_on = INVALID_TRIP;
504 params->trip_max_desired_temperature = last_active;
505 }
506 }
507
reset_pid_controller(struct power_allocator_params * params)508 static void reset_pid_controller(struct power_allocator_params *params)
509 {
510 params->err_integral = 0;
511 params->prev_err = 0;
512 }
513
allow_maximum_power(struct thermal_zone_device * tz)514 static void allow_maximum_power(struct thermal_zone_device *tz)
515 {
516 struct thermal_instance *instance;
517 struct power_allocator_params *params = tz->governor_data;
518
519 mutex_lock(&tz->lock);
520 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
521 if ((instance->trip != params->trip_max_desired_temperature) ||
522 (!cdev_is_power_actor(instance->cdev)))
523 continue;
524
525 instance->target = 0;
526 mutex_lock(&instance->cdev->lock);
527 instance->cdev->updated = false;
528 mutex_unlock(&instance->cdev->lock);
529 thermal_cdev_update(instance->cdev);
530 }
531 mutex_unlock(&tz->lock);
532 }
533
534 /**
535 * power_allocator_bind() - bind the power_allocator governor to a thermal zone
536 * @tz: thermal zone to bind it to
537 *
538 * Initialize the PID controller parameters and bind it to the thermal
539 * zone.
540 *
541 * Return: 0 on success, or -ENOMEM if we ran out of memory.
542 */
power_allocator_bind(struct thermal_zone_device * tz)543 static int power_allocator_bind(struct thermal_zone_device *tz)
544 {
545 int ret;
546 struct power_allocator_params *params;
547 int control_temp;
548
549 params = kzalloc(sizeof(*params), GFP_KERNEL);
550 if (!params)
551 return -ENOMEM;
552
553 if (!tz->tzp) {
554 tz->tzp = kzalloc(sizeof(*tz->tzp), GFP_KERNEL);
555 if (!tz->tzp) {
556 ret = -ENOMEM;
557 goto free_params;
558 }
559
560 params->allocated_tzp = true;
561 }
562
563 if (!tz->tzp->sustainable_power)
564 dev_warn(&tz->device, "power_allocator: sustainable_power will be estimated\n");
565
566 get_governor_trips(tz, params);
567
568 if (tz->trips > 0) {
569 ret = tz->ops->get_trip_temp(tz,
570 params->trip_max_desired_temperature,
571 &control_temp);
572 if (!ret)
573 estimate_pid_constants(tz, tz->tzp->sustainable_power,
574 params->trip_switch_on,
575 control_temp, false);
576 }
577
578 reset_pid_controller(params);
579
580 tz->governor_data = params;
581
582 return 0;
583
584 free_params:
585 kfree(params);
586
587 return ret;
588 }
589
power_allocator_unbind(struct thermal_zone_device * tz)590 static void power_allocator_unbind(struct thermal_zone_device *tz)
591 {
592 struct power_allocator_params *params = tz->governor_data;
593
594 dev_dbg(&tz->device, "Unbinding from thermal zone %d\n", tz->id);
595
596 if (params->allocated_tzp) {
597 kfree(tz->tzp);
598 tz->tzp = NULL;
599 }
600
601 kfree(tz->governor_data);
602 tz->governor_data = NULL;
603 }
604
power_allocator_throttle(struct thermal_zone_device * tz,int trip)605 static int power_allocator_throttle(struct thermal_zone_device *tz, int trip)
606 {
607 int ret;
608 int switch_on_temp, control_temp;
609 struct power_allocator_params *params = tz->governor_data;
610
611 /*
612 * We get called for every trip point but we only need to do
613 * our calculations once
614 */
615 if (trip != params->trip_max_desired_temperature)
616 return 0;
617
618 ret = tz->ops->get_trip_temp(tz, params->trip_switch_on,
619 &switch_on_temp);
620 if (!ret && (tz->temperature < switch_on_temp)) {
621 tz->passive = 0;
622 reset_pid_controller(params);
623 allow_maximum_power(tz);
624 return 0;
625 }
626
627 tz->passive = 1;
628
629 ret = tz->ops->get_trip_temp(tz, params->trip_max_desired_temperature,
630 &control_temp);
631 if (ret) {
632 dev_warn(&tz->device,
633 "Failed to get the maximum desired temperature: %d\n",
634 ret);
635 return ret;
636 }
637
638 return allocate_power(tz, control_temp);
639 }
640
641 static struct thermal_governor thermal_gov_power_allocator = {
642 .name = "power_allocator",
643 .bind_to_tz = power_allocator_bind,
644 .unbind_from_tz = power_allocator_unbind,
645 .throttle = power_allocator_throttle,
646 };
647 THERMAL_GOVERNOR_DECLARE(thermal_gov_power_allocator);
648