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/slab.h>
12 #include <linux/thermal.h>
14 #define CREATE_TRACE_POINTS
15 #include "thermal_trace_ipa.h"
17 #include "thermal_core.h"
19 #define INVALID_TRIP -1
22 #define int_to_frac(x) ((x) << FRAC_BITS)
23 #define frac_to_int(x) ((x) >> FRAC_BITS)
26 * mul_frac() - multiply two fixed-point numbers
27 * @x: first multiplicand
28 * @y: second multiplicand
30 * Return: the result of multiplying two fixed-point numbers. The
31 * result is also a fixed-point number.
33 static inline s64 mul_frac(s64 x, s64 y)
35 return (x * y) >> FRAC_BITS;
39 * div_frac() - divide two fixed-point numbers
43 * Return: the result of dividing two fixed-point numbers. The
44 * result is also a fixed-point number.
46 static inline s64 div_frac(s64 x, s64 y)
48 return div_s64(x << FRAC_BITS, y);
52 * struct power_allocator_params - parameters for the power allocator governor
53 * @allocated_tzp: whether we have allocated tzp for this thermal zone and
54 * it needs to be freed on unbind
55 * @err_integral: accumulated error in the PID controller.
56 * @prev_err: error in the previous iteration of the PID controller.
57 * Used to calculate the derivative term.
58 * @trip_switch_on: first passive trip point of the thermal zone. The
59 * governor switches on when this trip point is crossed.
60 * If the thermal zone only has one passive trip point,
61 * @trip_switch_on should be INVALID_TRIP.
62 * @trip_max_desired_temperature: last passive trip point of the thermal
63 * zone. The temperature we are
65 * @sustainable_power: Sustainable power (heat) that this thermal zone can
68 struct power_allocator_params {
73 int trip_max_desired_temperature;
74 u32 sustainable_power;
78 * estimate_sustainable_power() - Estimate the sustainable power of a thermal zone
79 * @tz: thermal zone we are operating in
81 * For thermal zones that don't provide a sustainable_power in their
82 * thermal_zone_params, estimate one. Calculate it using the minimum
83 * power of all the cooling devices as that gives a valid value that
84 * can give some degree of functionality. For optimal performance of
85 * this governor, provide a sustainable_power in the thermal zone's
86 * thermal_zone_params.
88 static u32 estimate_sustainable_power(struct thermal_zone_device *tz)
90 u32 sustainable_power = 0;
91 struct thermal_instance *instance;
92 struct power_allocator_params *params = tz->governor_data;
94 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
95 struct thermal_cooling_device *cdev = instance->cdev;
98 if (instance->trip != params->trip_max_desired_temperature)
101 if (!cdev_is_power_actor(cdev))
104 if (cdev->ops->state2power(cdev, instance->upper, &min_power))
107 sustainable_power += min_power;
110 return sustainable_power;
114 * estimate_pid_constants() - Estimate the constants for the PID controller
115 * @tz: thermal zone for which to estimate the constants
116 * @sustainable_power: sustainable power for the thermal zone
117 * @trip_switch_on: trip point number for the switch on temperature
118 * @control_temp: target temperature for the power allocator governor
120 * This function is used to update the estimation of the PID
121 * controller constants in struct thermal_zone_parameters.
123 static void estimate_pid_constants(struct thermal_zone_device *tz,
124 u32 sustainable_power, int trip_switch_on,
127 struct thermal_trip trip;
128 u32 temperature_threshold = control_temp;
132 ret = __thermal_zone_get_trip(tz, trip_switch_on, &trip);
134 temperature_threshold -= trip.temperature;
137 * estimate_pid_constants() tries to find appropriate default
138 * values for thermal zones that don't provide them. If a
139 * system integrator has configured a thermal zone with two
140 * passive trip points at the same temperature, that person
141 * hasn't put any effort to set up the thermal zone properly
144 if (!temperature_threshold)
147 tz->tzp->k_po = int_to_frac(sustainable_power) /
148 temperature_threshold;
150 tz->tzp->k_pu = int_to_frac(2 * sustainable_power) /
151 temperature_threshold;
153 k_i = tz->tzp->k_pu / 10;
154 tz->tzp->k_i = k_i > 0 ? k_i : 1;
157 * The default for k_d and integral_cutoff is 0, so we can
158 * leave them as they are.
163 * get_sustainable_power() - Get the right sustainable power
164 * @tz: thermal zone for which to estimate the constants
165 * @params: parameters for the power allocator governor
166 * @control_temp: target temperature for the power allocator governor
168 * This function is used for getting the proper sustainable power value based
169 * on variables which might be updated by the user sysfs interface. If that
170 * happen the new value is going to be estimated and updated. It is also used
171 * after thermal zone binding, where the initial values where set to 0.
173 static u32 get_sustainable_power(struct thermal_zone_device *tz,
174 struct power_allocator_params *params,
177 u32 sustainable_power;
179 if (!tz->tzp->sustainable_power)
180 sustainable_power = estimate_sustainable_power(tz);
182 sustainable_power = tz->tzp->sustainable_power;
184 /* Check if it's init value 0 or there was update via sysfs */
185 if (sustainable_power != params->sustainable_power) {
186 estimate_pid_constants(tz, sustainable_power,
187 params->trip_switch_on, control_temp);
189 /* Do the estimation only once and make available in sysfs */
190 tz->tzp->sustainable_power = sustainable_power;
191 params->sustainable_power = sustainable_power;
194 return sustainable_power;
198 * pid_controller() - PID controller
199 * @tz: thermal zone we are operating in
200 * @control_temp: the target temperature in millicelsius
201 * @max_allocatable_power: maximum allocatable power for this thermal zone
203 * This PID controller increases the available power budget so that the
204 * temperature of the thermal zone gets as close as possible to
205 * @control_temp and limits the power if it exceeds it. k_po is the
206 * proportional term when we are overshooting, k_pu is the
207 * proportional term when we are undershooting. integral_cutoff is a
208 * threshold below which we stop accumulating the error. The
209 * accumulated error is only valid if the requested power will make
210 * the system warmer. If the system is mostly idle, there's no point
211 * in accumulating positive error.
213 * Return: The power budget for the next period.
215 static u32 pid_controller(struct thermal_zone_device *tz,
217 u32 max_allocatable_power)
219 s64 p, i, d, power_range;
220 s32 err, max_power_frac;
221 u32 sustainable_power;
222 struct power_allocator_params *params = tz->governor_data;
224 max_power_frac = int_to_frac(max_allocatable_power);
226 sustainable_power = get_sustainable_power(tz, params, control_temp);
228 err = control_temp - tz->temperature;
229 err = int_to_frac(err);
231 /* Calculate the proportional term */
232 p = mul_frac(err < 0 ? tz->tzp->k_po : tz->tzp->k_pu, err);
235 * Calculate the integral term
237 * if the error is less than cut off allow integration (but
238 * the integral is limited to max power)
240 i = mul_frac(tz->tzp->k_i, params->err_integral);
242 if (err < int_to_frac(tz->tzp->integral_cutoff)) {
243 s64 i_next = i + mul_frac(tz->tzp->k_i, err);
245 if (abs(i_next) < max_power_frac) {
247 params->err_integral += err;
252 * Calculate the derivative term
254 * We do err - prev_err, so with a positive k_d, a decreasing
255 * error (i.e. driving closer to the line) results in less
256 * power being applied, slowing down the controller)
258 d = mul_frac(tz->tzp->k_d, err - params->prev_err);
259 d = div_frac(d, jiffies_to_msecs(tz->passive_delay_jiffies));
260 params->prev_err = err;
262 power_range = p + i + d;
264 /* feed-forward the known sustainable dissipatable power */
265 power_range = sustainable_power + frac_to_int(power_range);
267 power_range = clamp(power_range, (s64)0, (s64)max_allocatable_power);
269 trace_thermal_power_allocator_pid(tz, frac_to_int(err),
270 frac_to_int(params->err_integral),
271 frac_to_int(p), frac_to_int(i),
272 frac_to_int(d), power_range);
278 * power_actor_set_power() - limit the maximum power a cooling device consumes
279 * @cdev: pointer to &thermal_cooling_device
280 * @instance: thermal instance to update
281 * @power: the power in milliwatts
283 * Set the cooling device to consume at most @power milliwatts. The limit is
284 * expected to be a cap at the maximum power consumption.
286 * Return: 0 on success, -EINVAL if the cooling device does not
287 * implement the power actor API or -E* for other failures.
290 power_actor_set_power(struct thermal_cooling_device *cdev,
291 struct thermal_instance *instance, u32 power)
296 ret = cdev->ops->power2state(cdev, power, &state);
300 instance->target = clamp_val(state, instance->lower, instance->upper);
301 mutex_lock(&cdev->lock);
302 __thermal_cdev_update(cdev);
303 mutex_unlock(&cdev->lock);
309 * divvy_up_power() - divvy the allocated power between the actors
310 * @req_power: each actor's requested power
311 * @max_power: each actor's maximum available power
312 * @num_actors: size of the @req_power, @max_power and @granted_power's array
313 * @total_req_power: sum of @req_power
314 * @power_range: total allocated power
315 * @granted_power: output array: each actor's granted power
316 * @extra_actor_power: an appropriately sized array to be used in the
317 * function as temporary storage of the extra power given
320 * This function divides the total allocated power (@power_range)
321 * fairly between the actors. It first tries to give each actor a
322 * share of the @power_range according to how much power it requested
323 * compared to the rest of the actors. For example, if only one actor
324 * requests power, then it receives all the @power_range. If
325 * three actors each requests 1mW, each receives a third of the
328 * If any actor received more than their maximum power, then that
329 * surplus is re-divvied among the actors based on how far they are
330 * from their respective maximums.
332 * Granted power for each actor is written to @granted_power, which
333 * should've been allocated by the calling function.
335 static void divvy_up_power(u32 *req_power, u32 *max_power, int num_actors,
336 u32 total_req_power, u32 power_range,
337 u32 *granted_power, u32 *extra_actor_power)
339 u32 extra_power, capped_extra_power;
343 * Prevent division by 0 if none of the actors request power.
345 if (!total_req_power)
348 capped_extra_power = 0;
350 for (i = 0; i < num_actors; i++) {
351 u64 req_range = (u64)req_power[i] * power_range;
353 granted_power[i] = DIV_ROUND_CLOSEST_ULL(req_range,
356 if (granted_power[i] > max_power[i]) {
357 extra_power += granted_power[i] - max_power[i];
358 granted_power[i] = max_power[i];
361 extra_actor_power[i] = max_power[i] - granted_power[i];
362 capped_extra_power += extra_actor_power[i];
369 * Re-divvy the reclaimed extra among actors based on
370 * how far they are from the max
372 extra_power = min(extra_power, capped_extra_power);
373 if (capped_extra_power > 0)
374 for (i = 0; i < num_actors; i++) {
375 u64 extra_range = (u64)extra_actor_power[i] * extra_power;
376 granted_power[i] += DIV_ROUND_CLOSEST_ULL(extra_range,
381 static int allocate_power(struct thermal_zone_device *tz,
384 struct thermal_instance *instance;
385 struct power_allocator_params *params = tz->governor_data;
386 u32 *req_power, *max_power, *granted_power, *extra_actor_power;
387 u32 *weighted_req_power;
388 u32 total_req_power, max_allocatable_power, total_weighted_req_power;
389 u32 total_granted_power, power_range;
390 int i, num_actors, total_weight, ret = 0;
391 int trip_max_desired_temperature = params->trip_max_desired_temperature;
395 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
396 if ((instance->trip == trip_max_desired_temperature) &&
397 cdev_is_power_actor(instance->cdev)) {
399 total_weight += instance->weight;
407 * We need to allocate five arrays of the same size:
408 * req_power, max_power, granted_power, extra_actor_power and
409 * weighted_req_power. They are going to be needed until this
410 * function returns. Allocate them all in one go to simplify
411 * the allocation and deallocation logic.
413 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*max_power));
414 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*granted_power));
415 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*extra_actor_power));
416 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*weighted_req_power));
417 req_power = kcalloc(num_actors * 5, sizeof(*req_power), GFP_KERNEL);
421 max_power = &req_power[num_actors];
422 granted_power = &req_power[2 * num_actors];
423 extra_actor_power = &req_power[3 * num_actors];
424 weighted_req_power = &req_power[4 * num_actors];
427 total_weighted_req_power = 0;
429 max_allocatable_power = 0;
431 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
433 struct thermal_cooling_device *cdev = instance->cdev;
435 if (instance->trip != trip_max_desired_temperature)
438 if (!cdev_is_power_actor(cdev))
441 if (cdev->ops->get_requested_power(cdev, &req_power[i]))
445 weight = 1 << FRAC_BITS;
447 weight = instance->weight;
449 weighted_req_power[i] = frac_to_int(weight * req_power[i]);
451 if (cdev->ops->state2power(cdev, instance->lower,
455 total_req_power += req_power[i];
456 max_allocatable_power += max_power[i];
457 total_weighted_req_power += weighted_req_power[i];
462 power_range = pid_controller(tz, control_temp, max_allocatable_power);
464 divvy_up_power(weighted_req_power, max_power, num_actors,
465 total_weighted_req_power, power_range, granted_power,
468 total_granted_power = 0;
470 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
471 if (instance->trip != trip_max_desired_temperature)
474 if (!cdev_is_power_actor(instance->cdev))
477 power_actor_set_power(instance->cdev, instance,
479 total_granted_power += granted_power[i];
484 trace_thermal_power_allocator(tz, req_power, total_req_power,
485 granted_power, total_granted_power,
486 num_actors, power_range,
487 max_allocatable_power, tz->temperature,
488 control_temp - tz->temperature);
496 * get_governor_trips() - get the number of the two trip points that are key for this governor
497 * @tz: thermal zone to operate on
498 * @params: pointer to private data for this governor
500 * The power allocator governor works optimally with two trips points:
501 * a "switch on" trip point and a "maximum desired temperature". These
502 * are defined as the first and last passive trip points.
504 * If there is only one trip point, then that's considered to be the
505 * "maximum desired temperature" trip point and the governor is always
506 * on. If there are no passive or active trip points, then the
507 * governor won't do anything. In fact, its throttle function
508 * won't be called at all.
510 static void get_governor_trips(struct thermal_zone_device *tz,
511 struct power_allocator_params *params)
513 int i, last_active, last_passive;
514 bool found_first_passive;
516 found_first_passive = false;
517 last_active = INVALID_TRIP;
518 last_passive = INVALID_TRIP;
520 for (i = 0; i < tz->num_trips; i++) {
521 struct thermal_trip trip;
524 ret = __thermal_zone_get_trip(tz, i, &trip);
526 dev_warn(&tz->device,
527 "Failed to get trip point %d type: %d\n", i,
532 if (trip.type == THERMAL_TRIP_PASSIVE) {
533 if (!found_first_passive) {
534 params->trip_switch_on = i;
535 found_first_passive = true;
539 } else if (trip.type == THERMAL_TRIP_ACTIVE) {
546 if (last_passive != INVALID_TRIP) {
547 params->trip_max_desired_temperature = last_passive;
548 } else if (found_first_passive) {
549 params->trip_max_desired_temperature = params->trip_switch_on;
550 params->trip_switch_on = INVALID_TRIP;
552 params->trip_switch_on = INVALID_TRIP;
553 params->trip_max_desired_temperature = last_active;
557 static void reset_pid_controller(struct power_allocator_params *params)
559 params->err_integral = 0;
560 params->prev_err = 0;
563 static void allow_maximum_power(struct thermal_zone_device *tz, bool update)
565 struct thermal_instance *instance;
566 struct power_allocator_params *params = tz->governor_data;
569 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
570 struct thermal_cooling_device *cdev = instance->cdev;
572 if ((instance->trip != params->trip_max_desired_temperature) ||
573 (!cdev_is_power_actor(instance->cdev)))
576 instance->target = 0;
577 mutex_lock(&instance->cdev->lock);
579 * Call for updating the cooling devices local stats and avoid
580 * periods of dozen of seconds when those have not been
583 cdev->ops->get_requested_power(cdev, &req_power);
586 __thermal_cdev_update(instance->cdev);
588 mutex_unlock(&instance->cdev->lock);
593 * check_power_actors() - Check all cooling devices and warn when they are
595 * @tz: thermal zone to operate on
597 * Check all cooling devices in the @tz and warn every time they are missing
598 * power actor API. The warning should help to investigate the issue, which
599 * could be e.g. lack of Energy Model for a given device.
601 * Return: 0 on success, -EINVAL if any cooling device does not implement
602 * the power actor API.
604 static int check_power_actors(struct thermal_zone_device *tz)
606 struct thermal_instance *instance;
609 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
610 if (!cdev_is_power_actor(instance->cdev)) {
611 dev_warn(&tz->device, "power_allocator: %s is not a power actor\n",
612 instance->cdev->type);
621 * power_allocator_bind() - bind the power_allocator governor to a thermal zone
622 * @tz: thermal zone to bind it to
624 * Initialize the PID controller parameters and bind it to the thermal
627 * Return: 0 on success, or -ENOMEM if we ran out of memory, or -EINVAL
628 * when there are unsupported cooling devices in the @tz.
630 static int power_allocator_bind(struct thermal_zone_device *tz)
633 struct power_allocator_params *params;
634 struct thermal_trip trip;
636 ret = check_power_actors(tz);
640 params = kzalloc(sizeof(*params), GFP_KERNEL);
645 tz->tzp = kzalloc(sizeof(*tz->tzp), GFP_KERNEL);
651 params->allocated_tzp = true;
654 if (!tz->tzp->sustainable_power)
655 dev_warn(&tz->device, "power_allocator: sustainable_power will be estimated\n");
657 get_governor_trips(tz, params);
659 if (tz->num_trips > 0) {
660 ret = __thermal_zone_get_trip(tz, params->trip_max_desired_temperature,
663 estimate_pid_constants(tz, tz->tzp->sustainable_power,
664 params->trip_switch_on,
668 reset_pid_controller(params);
670 tz->governor_data = params;
680 static void power_allocator_unbind(struct thermal_zone_device *tz)
682 struct power_allocator_params *params = tz->governor_data;
684 dev_dbg(&tz->device, "Unbinding from thermal zone %d\n", tz->id);
686 if (params->allocated_tzp) {
691 kfree(tz->governor_data);
692 tz->governor_data = NULL;
695 static int power_allocator_throttle(struct thermal_zone_device *tz, int trip_id)
697 struct power_allocator_params *params = tz->governor_data;
698 struct thermal_trip trip;
702 lockdep_assert_held(&tz->lock);
705 * We get called for every trip point but we only need to do
706 * our calculations once
708 if (trip_id != params->trip_max_desired_temperature)
711 ret = __thermal_zone_get_trip(tz, params->trip_switch_on, &trip);
712 if (!ret && (tz->temperature < trip.temperature)) {
713 update = (tz->last_temperature >= trip.temperature);
715 reset_pid_controller(params);
716 allow_maximum_power(tz, update);
722 ret = __thermal_zone_get_trip(tz, params->trip_max_desired_temperature, &trip);
724 dev_warn(&tz->device, "Failed to get the maximum desired temperature: %d\n",
729 return allocate_power(tz, trip.temperature);
732 static struct thermal_governor thermal_gov_power_allocator = {
733 .name = "power_allocator",
734 .bind_to_tz = power_allocator_bind,
735 .unbind_from_tz = power_allocator_unbind,
736 .throttle = power_allocator_throttle,
738 THERMAL_GOVERNOR_DECLARE(thermal_gov_power_allocator);