1 // SPDX-License-Identifier: GPL-2.0
3 * A power allocator to manage temperature
5 * Copyright (C) 2014 ARM Ltd.
9 #define pr_fmt(fmt) "Power allocator: " fmt
11 #include <linux/rculist.h>
12 #include <linux/slab.h>
13 #include <linux/thermal.h>
15 #define CREATE_TRACE_POINTS
16 #include <trace/events/thermal_power_allocator.h>
18 #include "thermal_core.h"
20 #define INVALID_TRIP -1
23 #define int_to_frac(x) ((x) << FRAC_BITS)
24 #define frac_to_int(x) ((x) >> FRAC_BITS)
27 * mul_frac() - multiply two fixed-point numbers
28 * @x: first multiplicand
29 * @y: second multiplicand
31 * Return: the result of multiplying two fixed-point numbers. The
32 * result is also a fixed-point number.
34 static inline s64 mul_frac(s64 x, s64 y)
36 return (x * y) >> FRAC_BITS;
40 * div_frac() - divide two fixed-point numbers
44 * Return: the result of dividing two fixed-point numbers. The
45 * result is also a fixed-point number.
47 static inline s64 div_frac(s64 x, s64 y)
49 return div_s64(x << FRAC_BITS, y);
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
66 * @sustainable_power: Sustainable power (heat) that this thermal zone can
69 struct power_allocator_params {
74 int trip_max_desired_temperature;
75 u32 sustainable_power;
79 * estimate_sustainable_power() - Estimate the sustainable power of a thermal zone
80 * @tz: thermal zone we are operating in
82 * For thermal zones that don't provide a sustainable_power in their
83 * thermal_zone_params, estimate one. Calculate it using the minimum
84 * power of all the cooling devices as that gives a valid value that
85 * can give some degree of functionality. For optimal performance of
86 * this governor, provide a sustainable_power in the thermal zone's
87 * thermal_zone_params.
89 static u32 estimate_sustainable_power(struct thermal_zone_device *tz)
91 u32 sustainable_power = 0;
92 struct thermal_instance *instance;
93 struct power_allocator_params *params = tz->governor_data;
95 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
96 struct thermal_cooling_device *cdev = instance->cdev;
99 if (instance->trip != params->trip_max_desired_temperature)
102 if (!cdev_is_power_actor(cdev))
105 if (cdev->ops->state2power(cdev, instance->upper, &min_power))
108 sustainable_power += min_power;
111 return sustainable_power;
115 * estimate_pid_constants() - Estimate the constants for the PID controller
116 * @tz: thermal zone for which to estimate the constants
117 * @sustainable_power: sustainable power for the thermal zone
118 * @trip_switch_on: trip point number for the switch on temperature
119 * @control_temp: target temperature for the power allocator governor
120 * @force: whether to force the update of the constants
122 * This function is used to update the estimation of the PID
123 * controller constants in struct thermal_zone_parameters.
125 * If @force is not set, the values in the thermal zone's parameters
126 * are preserved if they are not zero. If @force is set, the values
127 * in thermal zone's parameters are overwritten.
129 static void estimate_pid_constants(struct thermal_zone_device *tz,
130 u32 sustainable_power, int trip_switch_on,
131 int control_temp, bool force)
135 u32 temperature_threshold;
138 ret = tz->ops->get_trip_temp(tz, trip_switch_on, &switch_on_temp);
142 temperature_threshold = control_temp - switch_on_temp;
144 * estimate_pid_constants() tries to find appropriate default
145 * values for thermal zones that don't provide them. If a
146 * system integrator has configured a thermal zone with two
147 * passive trip points at the same temperature, that person
148 * hasn't put any effort to set up the thermal zone properly
151 if (!temperature_threshold)
154 if (!tz->tzp->k_po || force)
155 tz->tzp->k_po = int_to_frac(sustainable_power) /
156 temperature_threshold;
158 if (!tz->tzp->k_pu || force)
159 tz->tzp->k_pu = int_to_frac(2 * sustainable_power) /
160 temperature_threshold;
162 if (!tz->tzp->k_i || force) {
163 k_i = tz->tzp->k_pu / 10;
164 tz->tzp->k_i = k_i > 0 ? k_i : 1;
168 * The default for k_d and integral_cutoff is 0, so we can
169 * leave them as they are.
174 * get_sustainable_power() - Get the right sustainable power
175 * @tz: thermal zone for which to estimate the constants
176 * @params: parameters for the power allocator governor
177 * @control_temp: target temperature for the power allocator governor
179 * This function is used for getting the proper sustainable power value based
180 * on variables which might be updated by the user sysfs interface. If that
181 * happen the new value is going to be estimated and updated. It is also used
182 * after thermal zone binding, where the initial values where set to 0.
184 static u32 get_sustainable_power(struct thermal_zone_device *tz,
185 struct power_allocator_params *params,
188 u32 sustainable_power;
190 if (!tz->tzp->sustainable_power)
191 sustainable_power = estimate_sustainable_power(tz);
193 sustainable_power = tz->tzp->sustainable_power;
195 /* Check if it's init value 0 or there was update via sysfs */
196 if (sustainable_power != params->sustainable_power) {
197 estimate_pid_constants(tz, sustainable_power,
198 params->trip_switch_on, control_temp,
201 /* Do the estimation only once and make available in sysfs */
202 tz->tzp->sustainable_power = sustainable_power;
203 params->sustainable_power = sustainable_power;
206 return sustainable_power;
210 * pid_controller() - PID controller
211 * @tz: thermal zone we are operating in
212 * @control_temp: the target temperature in millicelsius
213 * @max_allocatable_power: maximum allocatable power for this thermal zone
215 * This PID controller increases the available power budget so that the
216 * temperature of the thermal zone gets as close as possible to
217 * @control_temp and limits the power if it exceeds it. k_po is the
218 * proportional term when we are overshooting, k_pu is the
219 * proportional term when we are undershooting. integral_cutoff is a
220 * threshold below which we stop accumulating the error. The
221 * accumulated error is only valid if the requested power will make
222 * the system warmer. If the system is mostly idle, there's no point
223 * in accumulating positive error.
225 * Return: The power budget for the next period.
227 static u32 pid_controller(struct thermal_zone_device *tz,
229 u32 max_allocatable_power)
231 s64 p, i, d, power_range;
232 s32 err, max_power_frac;
233 u32 sustainable_power;
234 struct power_allocator_params *params = tz->governor_data;
236 max_power_frac = int_to_frac(max_allocatable_power);
238 sustainable_power = get_sustainable_power(tz, params, control_temp);
240 err = control_temp - tz->temperature;
241 err = int_to_frac(err);
243 /* Calculate the proportional term */
244 p = mul_frac(err < 0 ? tz->tzp->k_po : tz->tzp->k_pu, err);
247 * Calculate the integral term
249 * if the error is less than cut off allow integration (but
250 * the integral is limited to max power)
252 i = mul_frac(tz->tzp->k_i, params->err_integral);
254 if (err < int_to_frac(tz->tzp->integral_cutoff)) {
255 s64 i_next = i + mul_frac(tz->tzp->k_i, err);
257 if (abs(i_next) < max_power_frac) {
259 params->err_integral += err;
264 * Calculate the derivative term
266 * We do err - prev_err, so with a positive k_d, a decreasing
267 * error (i.e. driving closer to the line) results in less
268 * power being applied, slowing down the controller)
270 d = mul_frac(tz->tzp->k_d, err - params->prev_err);
271 d = div_frac(d, tz->passive_delay);
272 params->prev_err = err;
274 power_range = p + i + d;
276 /* feed-forward the known sustainable dissipatable power */
277 power_range = sustainable_power + frac_to_int(power_range);
279 power_range = clamp(power_range, (s64)0, (s64)max_allocatable_power);
281 trace_thermal_power_allocator_pid(tz, frac_to_int(err),
282 frac_to_int(params->err_integral),
283 frac_to_int(p), frac_to_int(i),
284 frac_to_int(d), power_range);
290 * power_actor_set_power() - limit the maximum power a cooling device consumes
291 * @cdev: pointer to &thermal_cooling_device
292 * @instance: thermal instance to update
293 * @power: the power in milliwatts
295 * Set the cooling device to consume at most @power milliwatts. The limit is
296 * expected to be a cap at the maximum power consumption.
298 * Return: 0 on success, -EINVAL if the cooling device does not
299 * implement the power actor API or -E* for other failures.
302 power_actor_set_power(struct thermal_cooling_device *cdev,
303 struct thermal_instance *instance, u32 power)
308 ret = cdev->ops->power2state(cdev, power, &state);
312 instance->target = clamp_val(state, instance->lower, instance->upper);
313 mutex_lock(&cdev->lock);
314 cdev->updated = false;
315 mutex_unlock(&cdev->lock);
316 thermal_cdev_update(cdev);
322 * divvy_up_power() - divvy the allocated power between the actors
323 * @req_power: each actor's requested power
324 * @max_power: each actor's maximum available power
325 * @num_actors: size of the @req_power, @max_power and @granted_power's array
326 * @total_req_power: sum of @req_power
327 * @power_range: total allocated power
328 * @granted_power: output array: each actor's granted power
329 * @extra_actor_power: an appropriately sized array to be used in the
330 * function as temporary storage of the extra power given
333 * This function divides the total allocated power (@power_range)
334 * fairly between the actors. It first tries to give each actor a
335 * share of the @power_range according to how much power it requested
336 * compared to the rest of the actors. For example, if only one actor
337 * requests power, then it receives all the @power_range. If
338 * three actors each requests 1mW, each receives a third of the
341 * If any actor received more than their maximum power, then that
342 * surplus is re-divvied among the actors based on how far they are
343 * from their respective maximums.
345 * Granted power for each actor is written to @granted_power, which
346 * should've been allocated by the calling function.
348 static void divvy_up_power(u32 *req_power, u32 *max_power, int num_actors,
349 u32 total_req_power, u32 power_range,
350 u32 *granted_power, u32 *extra_actor_power)
352 u32 extra_power, capped_extra_power;
356 * Prevent division by 0 if none of the actors request power.
358 if (!total_req_power)
361 capped_extra_power = 0;
363 for (i = 0; i < num_actors; i++) {
364 u64 req_range = (u64)req_power[i] * power_range;
366 granted_power[i] = DIV_ROUND_CLOSEST_ULL(req_range,
369 if (granted_power[i] > max_power[i]) {
370 extra_power += granted_power[i] - max_power[i];
371 granted_power[i] = max_power[i];
374 extra_actor_power[i] = max_power[i] - granted_power[i];
375 capped_extra_power += extra_actor_power[i];
382 * Re-divvy the reclaimed extra among actors based on
383 * how far they are from the max
385 extra_power = min(extra_power, capped_extra_power);
386 if (capped_extra_power > 0)
387 for (i = 0; i < num_actors; i++)
388 granted_power[i] += (extra_actor_power[i] *
389 extra_power) / capped_extra_power;
392 static int allocate_power(struct thermal_zone_device *tz,
395 struct thermal_instance *instance;
396 struct power_allocator_params *params = tz->governor_data;
397 u32 *req_power, *max_power, *granted_power, *extra_actor_power;
398 u32 *weighted_req_power;
399 u32 total_req_power, max_allocatable_power, total_weighted_req_power;
400 u32 total_granted_power, power_range;
401 int i, num_actors, total_weight, ret = 0;
402 int trip_max_desired_temperature = params->trip_max_desired_temperature;
404 mutex_lock(&tz->lock);
408 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
409 if ((instance->trip == trip_max_desired_temperature) &&
410 cdev_is_power_actor(instance->cdev)) {
412 total_weight += instance->weight;
422 * We need to allocate five arrays of the same size:
423 * req_power, max_power, granted_power, extra_actor_power and
424 * weighted_req_power. They are going to be needed until this
425 * function returns. Allocate them all in one go to simplify
426 * the allocation and deallocation logic.
428 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*max_power));
429 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*granted_power));
430 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*extra_actor_power));
431 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*weighted_req_power));
432 req_power = kcalloc(num_actors * 5, sizeof(*req_power), GFP_KERNEL);
438 max_power = &req_power[num_actors];
439 granted_power = &req_power[2 * num_actors];
440 extra_actor_power = &req_power[3 * num_actors];
441 weighted_req_power = &req_power[4 * num_actors];
444 total_weighted_req_power = 0;
446 max_allocatable_power = 0;
448 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
450 struct thermal_cooling_device *cdev = instance->cdev;
452 if (instance->trip != trip_max_desired_temperature)
455 if (!cdev_is_power_actor(cdev))
458 if (cdev->ops->get_requested_power(cdev, &req_power[i]))
462 weight = 1 << FRAC_BITS;
464 weight = instance->weight;
466 weighted_req_power[i] = frac_to_int(weight * req_power[i]);
468 if (cdev->ops->state2power(cdev, instance->lower,
472 total_req_power += req_power[i];
473 max_allocatable_power += max_power[i];
474 total_weighted_req_power += weighted_req_power[i];
479 power_range = pid_controller(tz, control_temp, max_allocatable_power);
481 divvy_up_power(weighted_req_power, max_power, num_actors,
482 total_weighted_req_power, power_range, granted_power,
485 total_granted_power = 0;
487 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
488 if (instance->trip != trip_max_desired_temperature)
491 if (!cdev_is_power_actor(instance->cdev))
494 power_actor_set_power(instance->cdev, instance,
496 total_granted_power += granted_power[i];
501 trace_thermal_power_allocator(tz, req_power, total_req_power,
502 granted_power, total_granted_power,
503 num_actors, power_range,
504 max_allocatable_power, tz->temperature,
505 control_temp - tz->temperature);
509 mutex_unlock(&tz->lock);
515 * get_governor_trips() - get the number of the two trip points that are key for this governor
516 * @tz: thermal zone to operate on
517 * @params: pointer to private data for this governor
519 * The power allocator governor works optimally with two trips points:
520 * a "switch on" trip point and a "maximum desired temperature". These
521 * are defined as the first and last passive trip points.
523 * If there is only one trip point, then that's considered to be the
524 * "maximum desired temperature" trip point and the governor is always
525 * on. If there are no passive or active trip points, then the
526 * governor won't do anything. In fact, its throttle function
527 * won't be called at all.
529 static void get_governor_trips(struct thermal_zone_device *tz,
530 struct power_allocator_params *params)
532 int i, last_active, last_passive;
533 bool found_first_passive;
535 found_first_passive = false;
536 last_active = INVALID_TRIP;
537 last_passive = INVALID_TRIP;
539 for (i = 0; i < tz->trips; i++) {
540 enum thermal_trip_type type;
543 ret = tz->ops->get_trip_type(tz, i, &type);
545 dev_warn(&tz->device,
546 "Failed to get trip point %d type: %d\n", i,
551 if (type == THERMAL_TRIP_PASSIVE) {
552 if (!found_first_passive) {
553 params->trip_switch_on = i;
554 found_first_passive = true;
558 } else if (type == THERMAL_TRIP_ACTIVE) {
565 if (last_passive != INVALID_TRIP) {
566 params->trip_max_desired_temperature = last_passive;
567 } else if (found_first_passive) {
568 params->trip_max_desired_temperature = params->trip_switch_on;
569 params->trip_switch_on = INVALID_TRIP;
571 params->trip_switch_on = INVALID_TRIP;
572 params->trip_max_desired_temperature = last_active;
576 static void reset_pid_controller(struct power_allocator_params *params)
578 params->err_integral = 0;
579 params->prev_err = 0;
582 static void allow_maximum_power(struct thermal_zone_device *tz)
584 struct thermal_instance *instance;
585 struct power_allocator_params *params = tz->governor_data;
587 mutex_lock(&tz->lock);
588 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
589 if ((instance->trip != params->trip_max_desired_temperature) ||
590 (!cdev_is_power_actor(instance->cdev)))
593 instance->target = 0;
594 mutex_lock(&instance->cdev->lock);
595 instance->cdev->updated = false;
596 mutex_unlock(&instance->cdev->lock);
597 thermal_cdev_update(instance->cdev);
599 mutex_unlock(&tz->lock);
603 * power_allocator_bind() - bind the power_allocator governor to a thermal zone
604 * @tz: thermal zone to bind it to
606 * Initialize the PID controller parameters and bind it to the thermal
609 * Return: 0 on success, or -ENOMEM if we ran out of memory.
611 static int power_allocator_bind(struct thermal_zone_device *tz)
614 struct power_allocator_params *params;
617 params = kzalloc(sizeof(*params), GFP_KERNEL);
622 tz->tzp = kzalloc(sizeof(*tz->tzp), GFP_KERNEL);
628 params->allocated_tzp = true;
631 if (!tz->tzp->sustainable_power)
632 dev_warn(&tz->device, "power_allocator: sustainable_power will be estimated\n");
634 get_governor_trips(tz, params);
637 ret = tz->ops->get_trip_temp(tz,
638 params->trip_max_desired_temperature,
641 estimate_pid_constants(tz, tz->tzp->sustainable_power,
642 params->trip_switch_on,
643 control_temp, false);
646 reset_pid_controller(params);
648 tz->governor_data = params;
658 static void power_allocator_unbind(struct thermal_zone_device *tz)
660 struct power_allocator_params *params = tz->governor_data;
662 dev_dbg(&tz->device, "Unbinding from thermal zone %d\n", tz->id);
664 if (params->allocated_tzp) {
669 kfree(tz->governor_data);
670 tz->governor_data = NULL;
673 static int power_allocator_throttle(struct thermal_zone_device *tz, int trip)
676 int switch_on_temp, control_temp;
677 struct power_allocator_params *params = tz->governor_data;
680 * We get called for every trip point but we only need to do
681 * our calculations once
683 if (trip != params->trip_max_desired_temperature)
686 ret = tz->ops->get_trip_temp(tz, params->trip_switch_on,
688 if (!ret && (tz->temperature < switch_on_temp)) {
690 reset_pid_controller(params);
691 allow_maximum_power(tz);
697 ret = tz->ops->get_trip_temp(tz, params->trip_max_desired_temperature,
700 dev_warn(&tz->device,
701 "Failed to get the maximum desired temperature: %d\n",
706 return allocate_power(tz, control_temp);
709 static struct thermal_governor thermal_gov_power_allocator = {
710 .name = "power_allocator",
711 .bind_to_tz = power_allocator_bind,
712 .unbind_from_tz = power_allocator_unbind,
713 .throttle = power_allocator_throttle,
715 THERMAL_GOVERNOR_DECLARE(thermal_gov_power_allocator);