2 * A power allocator to manage temperature
4 * Copyright (C) 2014 ARM Ltd.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
10 * This program is distributed "as is" WITHOUT ANY WARRANTY of any
11 * kind, whether express or implied; without even the implied warranty
12 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
16 #define pr_fmt(fmt) "Power allocator: " fmt
18 #include <linux/rculist.h>
19 #include <linux/slab.h>
20 #include <linux/thermal.h>
22 #define CREATE_TRACE_POINTS
23 #include <trace/events/thermal_power_allocator.h>
25 #include "thermal_core.h"
27 #define INVALID_TRIP -1
30 #define int_to_frac(x) ((x) << FRAC_BITS)
31 #define frac_to_int(x) ((x) >> FRAC_BITS)
34 * mul_frac() - multiply two fixed-point numbers
35 * @x: first multiplicand
36 * @y: second multiplicand
38 * Return: the result of multiplying two fixed-point numbers. The
39 * result is also a fixed-point number.
41 static inline s64 mul_frac(s64 x, s64 y)
43 return (x * y) >> FRAC_BITS;
47 * div_frac() - divide two fixed-point numbers
51 * Return: the result of dividing two fixed-point numbers. The
52 * result is also a fixed-point number.
54 static inline s64 div_frac(s64 x, s64 y)
56 return div_s64(x << FRAC_BITS, y);
60 * struct power_allocator_params - parameters for the power allocator governor
61 * @allocated_tzp: whether we have allocated tzp for this thermal zone and
62 * it needs to be freed on unbind
63 * @err_integral: accumulated error in the PID controller.
64 * @prev_err: error in the previous iteration of the PID controller.
65 * Used to calculate the derivative term.
66 * @trip_switch_on: first passive trip point of the thermal zone. The
67 * governor switches on when this trip point is crossed.
68 * If the thermal zone only has one passive trip point,
69 * @trip_switch_on should be INVALID_TRIP.
70 * @trip_max_desired_temperature: last passive trip point of the thermal
71 * zone. The temperature we are
74 struct power_allocator_params {
79 int trip_max_desired_temperature;
83 * estimate_sustainable_power() - Estimate the sustainable power of a thermal zone
84 * @tz: thermal zone we are operating in
86 * For thermal zones that don't provide a sustainable_power in their
87 * thermal_zone_params, estimate one. Calculate it using the minimum
88 * power of all the cooling devices as that gives a valid value that
89 * can give some degree of functionality. For optimal performance of
90 * this governor, provide a sustainable_power in the thermal zone's
91 * thermal_zone_params.
93 static u32 estimate_sustainable_power(struct thermal_zone_device *tz)
95 u32 sustainable_power = 0;
96 struct thermal_instance *instance;
97 struct power_allocator_params *params = tz->governor_data;
99 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
100 struct thermal_cooling_device *cdev = instance->cdev;
103 if (instance->trip != params->trip_max_desired_temperature)
106 if (power_actor_get_min_power(cdev, tz, &min_power))
109 sustainable_power += min_power;
112 return sustainable_power;
116 * estimate_pid_constants() - Estimate the constants for the PID controller
117 * @tz: thermal zone for which to estimate the constants
118 * @sustainable_power: sustainable power for the thermal zone
119 * @trip_switch_on: trip point number for the switch on temperature
120 * @control_temp: target temperature for the power allocator governor
121 * @force: whether to force the update of the constants
123 * This function is used to update the estimation of the PID
124 * controller constants in struct thermal_zone_parameters.
125 * Sustainable power is provided in case it was estimated. The
126 * estimated sustainable_power should not be stored in the
127 * thermal_zone_parameters so it has to be passed explicitly to this
130 * If @force is not set, the values in the thermal zone's parameters
131 * are preserved if they are not zero. If @force is set, the values
132 * in thermal zone's parameters are overwritten.
134 static void estimate_pid_constants(struct thermal_zone_device *tz,
135 u32 sustainable_power, int trip_switch_on,
136 int control_temp, bool force)
140 u32 temperature_threshold;
142 ret = tz->ops->get_trip_temp(tz, trip_switch_on, &switch_on_temp);
146 temperature_threshold = control_temp - switch_on_temp;
148 if (!tz->tzp->k_po || force)
149 tz->tzp->k_po = int_to_frac(sustainable_power) /
150 temperature_threshold;
152 if (!tz->tzp->k_pu || force)
153 tz->tzp->k_pu = int_to_frac(2 * sustainable_power) /
154 temperature_threshold;
156 if (!tz->tzp->k_i || force)
157 tz->tzp->k_i = int_to_frac(10) / 1000;
159 * The default for k_d and integral_cutoff is 0, so we can
160 * leave them as they are.
165 * pid_controller() - PID controller
166 * @tz: thermal zone we are operating in
167 * @current_temp: the current temperature in millicelsius
168 * @control_temp: the target temperature in millicelsius
169 * @max_allocatable_power: maximum allocatable power for this thermal zone
171 * This PID controller increases the available power budget so that the
172 * temperature of the thermal zone gets as close as possible to
173 * @control_temp and limits the power if it exceeds it. k_po is the
174 * proportional term when we are overshooting, k_pu is the
175 * proportional term when we are undershooting. integral_cutoff is a
176 * threshold below which we stop accumulating the error. The
177 * accumulated error is only valid if the requested power will make
178 * the system warmer. If the system is mostly idle, there's no point
179 * in accumulating positive error.
181 * Return: The power budget for the next period.
183 static u32 pid_controller(struct thermal_zone_device *tz,
186 u32 max_allocatable_power)
188 s64 p, i, d, power_range;
189 s32 err, max_power_frac;
190 u32 sustainable_power;
191 struct power_allocator_params *params = tz->governor_data;
193 max_power_frac = int_to_frac(max_allocatable_power);
195 if (tz->tzp->sustainable_power) {
196 sustainable_power = tz->tzp->sustainable_power;
198 sustainable_power = estimate_sustainable_power(tz);
199 estimate_pid_constants(tz, sustainable_power,
200 params->trip_switch_on, control_temp,
204 err = control_temp - current_temp;
205 err = int_to_frac(err);
207 /* Calculate the proportional term */
208 p = mul_frac(err < 0 ? tz->tzp->k_po : tz->tzp->k_pu, err);
211 * Calculate the integral term
213 * if the error is less than cut off allow integration (but
214 * the integral is limited to max power)
216 i = mul_frac(tz->tzp->k_i, params->err_integral);
218 if (err < int_to_frac(tz->tzp->integral_cutoff)) {
219 s64 i_next = i + mul_frac(tz->tzp->k_i, err);
221 if (abs64(i_next) < max_power_frac) {
223 params->err_integral += err;
228 * Calculate the derivative term
230 * We do err - prev_err, so with a positive k_d, a decreasing
231 * error (i.e. driving closer to the line) results in less
232 * power being applied, slowing down the controller)
234 d = mul_frac(tz->tzp->k_d, err - params->prev_err);
235 d = div_frac(d, tz->passive_delay);
236 params->prev_err = err;
238 power_range = p + i + d;
240 /* feed-forward the known sustainable dissipatable power */
241 power_range = sustainable_power + frac_to_int(power_range);
243 power_range = clamp(power_range, (s64)0, (s64)max_allocatable_power);
245 trace_thermal_power_allocator_pid(tz, frac_to_int(err),
246 frac_to_int(params->err_integral),
247 frac_to_int(p), frac_to_int(i),
248 frac_to_int(d), power_range);
254 * divvy_up_power() - divvy the allocated power between the actors
255 * @req_power: each actor's requested power
256 * @max_power: each actor's maximum available power
257 * @num_actors: size of the @req_power, @max_power and @granted_power's array
258 * @total_req_power: sum of @req_power
259 * @power_range: total allocated power
260 * @granted_power: output array: each actor's granted power
261 * @extra_actor_power: an appropriately sized array to be used in the
262 * function as temporary storage of the extra power given
265 * This function divides the total allocated power (@power_range)
266 * fairly between the actors. It first tries to give each actor a
267 * share of the @power_range according to how much power it requested
268 * compared to the rest of the actors. For example, if only one actor
269 * requests power, then it receives all the @power_range. If
270 * three actors each requests 1mW, each receives a third of the
273 * If any actor received more than their maximum power, then that
274 * surplus is re-divvied among the actors based on how far they are
275 * from their respective maximums.
277 * Granted power for each actor is written to @granted_power, which
278 * should've been allocated by the calling function.
280 static void divvy_up_power(u32 *req_power, u32 *max_power, int num_actors,
281 u32 total_req_power, u32 power_range,
282 u32 *granted_power, u32 *extra_actor_power)
284 u32 extra_power, capped_extra_power;
288 * Prevent division by 0 if none of the actors request power.
290 if (!total_req_power)
293 capped_extra_power = 0;
295 for (i = 0; i < num_actors; i++) {
296 u64 req_range = req_power[i] * power_range;
298 granted_power[i] = DIV_ROUND_CLOSEST_ULL(req_range,
301 if (granted_power[i] > max_power[i]) {
302 extra_power += granted_power[i] - max_power[i];
303 granted_power[i] = max_power[i];
306 extra_actor_power[i] = max_power[i] - granted_power[i];
307 capped_extra_power += extra_actor_power[i];
314 * Re-divvy the reclaimed extra among actors based on
315 * how far they are from the max
317 extra_power = min(extra_power, capped_extra_power);
318 if (capped_extra_power > 0)
319 for (i = 0; i < num_actors; i++)
320 granted_power[i] += (extra_actor_power[i] *
321 extra_power) / capped_extra_power;
324 static int allocate_power(struct thermal_zone_device *tz,
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;
337 mutex_lock(&tz->lock);
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)) {
345 total_weight += instance->weight;
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.
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);
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];
377 total_weighted_req_power = 0;
379 max_allocatable_power = 0;
381 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
383 struct thermal_cooling_device *cdev = instance->cdev;
385 if (instance->trip != trip_max_desired_temperature)
388 if (!cdev_is_power_actor(cdev))
391 if (cdev->ops->get_requested_power(cdev, tz, &req_power[i]))
395 weight = 1 << FRAC_BITS;
397 weight = instance->weight;
399 weighted_req_power[i] = frac_to_int(weight * req_power[i]);
401 if (power_actor_get_max_power(cdev, tz, &max_power[i]))
404 total_req_power += req_power[i];
405 max_allocatable_power += max_power[i];
406 total_weighted_req_power += weighted_req_power[i];
411 power_range = pid_controller(tz, current_temp, control_temp,
412 max_allocatable_power);
414 divvy_up_power(weighted_req_power, max_power, num_actors,
415 total_weighted_req_power, power_range, granted_power,
418 total_granted_power = 0;
420 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
421 if (instance->trip != trip_max_desired_temperature)
424 if (!cdev_is_power_actor(instance->cdev))
427 power_actor_set_power(instance->cdev, instance,
429 total_granted_power += granted_power[i];
434 trace_thermal_power_allocator(tz, req_power, total_req_power,
435 granted_power, total_granted_power,
436 num_actors, power_range,
437 max_allocatable_power, current_temp,
438 control_temp - current_temp);
442 mutex_unlock(&tz->lock);
448 * get_governor_trips() - get the number of the two trip points that are key for this governor
449 * @tz: thermal zone to operate on
450 * @params: pointer to private data for this governor
452 * The power allocator governor works optimally with two trips points:
453 * a "switch on" trip point and a "maximum desired temperature". These
454 * are defined as the first and last passive trip points.
456 * If there is only one trip point, then that's considered to be the
457 * "maximum desired temperature" trip point and the governor is always
458 * on. If there are no passive or active trip points, then the
459 * governor won't do anything. In fact, its throttle function
460 * won't be called at all.
462 static void get_governor_trips(struct thermal_zone_device *tz,
463 struct power_allocator_params *params)
465 int i, last_active, last_passive;
466 bool found_first_passive;
468 found_first_passive = false;
469 last_active = INVALID_TRIP;
470 last_passive = INVALID_TRIP;
472 for (i = 0; i < tz->trips; i++) {
473 enum thermal_trip_type type;
476 ret = tz->ops->get_trip_type(tz, i, &type);
478 dev_warn(&tz->device,
479 "Failed to get trip point %d type: %d\n", i,
484 if (type == THERMAL_TRIP_PASSIVE) {
485 if (!found_first_passive) {
486 params->trip_switch_on = i;
487 found_first_passive = true;
491 } else if (type == THERMAL_TRIP_ACTIVE) {
498 if (last_passive != INVALID_TRIP) {
499 params->trip_max_desired_temperature = last_passive;
500 } else if (found_first_passive) {
501 params->trip_max_desired_temperature = params->trip_switch_on;
502 params->trip_switch_on = INVALID_TRIP;
504 params->trip_switch_on = INVALID_TRIP;
505 params->trip_max_desired_temperature = last_active;
509 static void reset_pid_controller(struct power_allocator_params *params)
511 params->err_integral = 0;
512 params->prev_err = 0;
515 static void allow_maximum_power(struct thermal_zone_device *tz)
517 struct thermal_instance *instance;
518 struct power_allocator_params *params = tz->governor_data;
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)))
525 instance->target = 0;
526 instance->cdev->updated = false;
527 thermal_cdev_update(instance->cdev);
532 * power_allocator_bind() - bind the power_allocator governor to a thermal zone
533 * @tz: thermal zone to bind it to
535 * Initialize the PID controller parameters and bind it to the thermal
538 * Return: 0 on success, or -ENOMEM if we ran out of memory.
540 static int power_allocator_bind(struct thermal_zone_device *tz)
543 struct power_allocator_params *params;
546 params = kzalloc(sizeof(*params), GFP_KERNEL);
551 tz->tzp = kzalloc(sizeof(*tz->tzp), GFP_KERNEL);
557 params->allocated_tzp = true;
560 if (!tz->tzp->sustainable_power)
561 dev_warn(&tz->device, "power_allocator: sustainable_power will be estimated\n");
563 get_governor_trips(tz, params);
566 ret = tz->ops->get_trip_temp(tz,
567 params->trip_max_desired_temperature,
570 estimate_pid_constants(tz, tz->tzp->sustainable_power,
571 params->trip_switch_on,
572 control_temp, false);
575 reset_pid_controller(params);
577 tz->governor_data = params;
587 static void power_allocator_unbind(struct thermal_zone_device *tz)
589 struct power_allocator_params *params = tz->governor_data;
591 dev_dbg(&tz->device, "Unbinding from thermal zone %d\n", tz->id);
593 if (params->allocated_tzp) {
598 kfree(tz->governor_data);
599 tz->governor_data = NULL;
602 static int power_allocator_throttle(struct thermal_zone_device *tz, int trip)
605 int switch_on_temp, control_temp, current_temp;
606 struct power_allocator_params *params = tz->governor_data;
609 * We get called for every trip point but we only need to do
610 * our calculations once
612 if (trip != params->trip_max_desired_temperature)
615 ret = thermal_zone_get_temp(tz, ¤t_temp);
617 dev_warn(&tz->device, "Failed to get temperature: %d\n", ret);
621 ret = tz->ops->get_trip_temp(tz, params->trip_switch_on,
623 if (!ret && (current_temp < switch_on_temp)) {
625 reset_pid_controller(params);
626 allow_maximum_power(tz);
632 ret = tz->ops->get_trip_temp(tz, params->trip_max_desired_temperature,
635 dev_warn(&tz->device,
636 "Failed to get the maximum desired temperature: %d\n",
641 return allocate_power(tz, current_temp, control_temp);
644 static struct thermal_governor thermal_gov_power_allocator = {
645 .name = "power_allocator",
646 .bind_to_tz = power_allocator_bind,
647 .unbind_from_tz = power_allocator_unbind,
648 .throttle = power_allocator_throttle,
651 int thermal_gov_power_allocator_register(void)
653 return thermal_register_governor(&thermal_gov_power_allocator);
656 void thermal_gov_power_allocator_unregister(void)
658 thermal_unregister_governor(&thermal_gov_power_allocator);