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"
28 #define int_to_frac(x) ((x) << FRAC_BITS)
29 #define frac_to_int(x) ((x) >> FRAC_BITS)
32 * mul_frac() - multiply two fixed-point numbers
33 * @x: first multiplicand
34 * @y: second multiplicand
36 * Return: the result of multiplying two fixed-point numbers. The
37 * result is also a fixed-point number.
39 static inline s64 mul_frac(s64 x, s64 y)
41 return (x * y) >> FRAC_BITS;
45 * div_frac() - divide two fixed-point numbers
49 * Return: the result of dividing two fixed-point numbers. The
50 * result is also a fixed-point number.
52 static inline s64 div_frac(s64 x, s64 y)
54 return div_s64(x << FRAC_BITS, y);
58 * struct power_allocator_params - parameters for the power allocator governor
59 * @err_integral: accumulated error in the PID controller.
60 * @prev_err: error in the previous iteration of the PID controller.
61 * Used to calculate the derivative term.
62 * @trip_switch_on: first passive trip point of the thermal zone. The
63 * governor switches on when this trip point is crossed.
64 * @trip_max_desired_temperature: last passive trip point of the thermal
65 * zone. The temperature we are
68 struct power_allocator_params {
72 int trip_max_desired_temperature;
76 * pid_controller() - PID controller
77 * @tz: thermal zone we are operating in
78 * @current_temp: the current temperature in millicelsius
79 * @control_temp: the target temperature in millicelsius
80 * @max_allocatable_power: maximum allocatable power for this thermal zone
82 * This PID controller increases the available power budget so that the
83 * temperature of the thermal zone gets as close as possible to
84 * @control_temp and limits the power if it exceeds it. k_po is the
85 * proportional term when we are overshooting, k_pu is the
86 * proportional term when we are undershooting. integral_cutoff is a
87 * threshold below which we stop accumulating the error. The
88 * accumulated error is only valid if the requested power will make
89 * the system warmer. If the system is mostly idle, there's no point
90 * in accumulating positive error.
92 * Return: The power budget for the next period.
94 static u32 pid_controller(struct thermal_zone_device *tz,
95 unsigned long current_temp,
96 unsigned long control_temp,
97 u32 max_allocatable_power)
99 s64 p, i, d, power_range;
100 s32 err, max_power_frac;
101 struct power_allocator_params *params = tz->governor_data;
103 max_power_frac = int_to_frac(max_allocatable_power);
105 err = ((s32)control_temp - (s32)current_temp);
106 err = int_to_frac(err);
108 /* Calculate the proportional term */
109 p = mul_frac(err < 0 ? tz->tzp->k_po : tz->tzp->k_pu, err);
112 * Calculate the integral term
114 * if the error is less than cut off allow integration (but
115 * the integral is limited to max power)
117 i = mul_frac(tz->tzp->k_i, params->err_integral);
119 if (err < int_to_frac(tz->tzp->integral_cutoff)) {
120 s64 i_next = i + mul_frac(tz->tzp->k_i, err);
122 if (abs64(i_next) < max_power_frac) {
124 params->err_integral += err;
129 * Calculate the derivative term
131 * We do err - prev_err, so with a positive k_d, a decreasing
132 * error (i.e. driving closer to the line) results in less
133 * power being applied, slowing down the controller)
135 d = mul_frac(tz->tzp->k_d, err - params->prev_err);
136 d = div_frac(d, tz->passive_delay);
137 params->prev_err = err;
139 power_range = p + i + d;
141 /* feed-forward the known sustainable dissipatable power */
142 power_range = tz->tzp->sustainable_power + frac_to_int(power_range);
144 power_range = clamp(power_range, (s64)0, (s64)max_allocatable_power);
146 trace_thermal_power_allocator_pid(tz, frac_to_int(err),
147 frac_to_int(params->err_integral),
148 frac_to_int(p), frac_to_int(i),
149 frac_to_int(d), power_range);
155 * divvy_up_power() - divvy the allocated power between the actors
156 * @req_power: each actor's requested power
157 * @max_power: each actor's maximum available power
158 * @num_actors: size of the @req_power, @max_power and @granted_power's array
159 * @total_req_power: sum of @req_power
160 * @power_range: total allocated power
161 * @granted_power: output array: each actor's granted power
162 * @extra_actor_power: an appropriately sized array to be used in the
163 * function as temporary storage of the extra power given
166 * This function divides the total allocated power (@power_range)
167 * fairly between the actors. It first tries to give each actor a
168 * share of the @power_range according to how much power it requested
169 * compared to the rest of the actors. For example, if only one actor
170 * requests power, then it receives all the @power_range. If
171 * three actors each requests 1mW, each receives a third of the
174 * If any actor received more than their maximum power, then that
175 * surplus is re-divvied among the actors based on how far they are
176 * from their respective maximums.
178 * Granted power for each actor is written to @granted_power, which
179 * should've been allocated by the calling function.
181 static void divvy_up_power(u32 *req_power, u32 *max_power, int num_actors,
182 u32 total_req_power, u32 power_range,
183 u32 *granted_power, u32 *extra_actor_power)
185 u32 extra_power, capped_extra_power;
189 * Prevent division by 0 if none of the actors request power.
191 if (!total_req_power)
194 capped_extra_power = 0;
196 for (i = 0; i < num_actors; i++) {
197 u64 req_range = req_power[i] * power_range;
199 granted_power[i] = DIV_ROUND_CLOSEST_ULL(req_range,
202 if (granted_power[i] > max_power[i]) {
203 extra_power += granted_power[i] - max_power[i];
204 granted_power[i] = max_power[i];
207 extra_actor_power[i] = max_power[i] - granted_power[i];
208 capped_extra_power += extra_actor_power[i];
215 * Re-divvy the reclaimed extra among actors based on
216 * how far they are from the max
218 extra_power = min(extra_power, capped_extra_power);
219 if (capped_extra_power > 0)
220 for (i = 0; i < num_actors; i++)
221 granted_power[i] += (extra_actor_power[i] *
222 extra_power) / capped_extra_power;
225 static int allocate_power(struct thermal_zone_device *tz,
226 unsigned long current_temp,
227 unsigned long control_temp)
229 struct thermal_instance *instance;
230 struct power_allocator_params *params = tz->governor_data;
231 u32 *req_power, *max_power, *granted_power, *extra_actor_power;
232 u32 total_req_power, max_allocatable_power;
233 u32 total_granted_power, power_range;
234 int i, num_actors, total_weight, ret = 0;
235 int trip_max_desired_temperature = params->trip_max_desired_temperature;
237 mutex_lock(&tz->lock);
241 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
242 if ((instance->trip == trip_max_desired_temperature) &&
243 cdev_is_power_actor(instance->cdev)) {
245 total_weight += instance->weight;
250 * We need to allocate three arrays of the same size:
251 * req_power, max_power and granted_power. They are going to
252 * be needed until this function returns. Allocate them all
253 * in one go to simplify the allocation and deallocation
256 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*max_power));
257 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*granted_power));
258 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*extra_actor_power));
259 req_power = devm_kcalloc(&tz->device, num_actors * 4,
260 sizeof(*req_power), GFP_KERNEL);
266 max_power = &req_power[num_actors];
267 granted_power = &req_power[2 * num_actors];
268 extra_actor_power = &req_power[3 * num_actors];
272 max_allocatable_power = 0;
274 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
276 struct thermal_cooling_device *cdev = instance->cdev;
278 if (instance->trip != trip_max_desired_temperature)
281 if (!cdev_is_power_actor(cdev))
284 if (cdev->ops->get_requested_power(cdev, tz, &req_power[i]))
288 weight = 1 << FRAC_BITS;
290 weight = instance->weight;
292 req_power[i] = frac_to_int(weight * req_power[i]);
294 if (power_actor_get_max_power(cdev, tz, &max_power[i]))
297 total_req_power += req_power[i];
298 max_allocatable_power += max_power[i];
303 power_range = pid_controller(tz, current_temp, control_temp,
304 max_allocatable_power);
306 divvy_up_power(req_power, max_power, num_actors, total_req_power,
307 power_range, granted_power, extra_actor_power);
309 total_granted_power = 0;
311 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
312 if (instance->trip != trip_max_desired_temperature)
315 if (!cdev_is_power_actor(instance->cdev))
318 power_actor_set_power(instance->cdev, instance,
320 total_granted_power += granted_power[i];
325 trace_thermal_power_allocator(tz, req_power, total_req_power,
326 granted_power, total_granted_power,
327 num_actors, power_range,
328 max_allocatable_power, current_temp,
329 (s32)control_temp - (s32)current_temp);
331 devm_kfree(&tz->device, req_power);
333 mutex_unlock(&tz->lock);
338 static int get_governor_trips(struct thermal_zone_device *tz,
339 struct power_allocator_params *params)
341 int i, ret, last_passive;
342 bool found_first_passive;
344 found_first_passive = false;
348 for (i = 0; i < tz->trips; i++) {
349 enum thermal_trip_type type;
351 ret = tz->ops->get_trip_type(tz, i, &type);
355 if (!found_first_passive) {
356 if (type == THERMAL_TRIP_PASSIVE) {
357 params->trip_switch_on = i;
358 found_first_passive = true;
360 } else if (type == THERMAL_TRIP_PASSIVE) {
367 if (last_passive != -1) {
368 params->trip_max_desired_temperature = last_passive;
377 static void reset_pid_controller(struct power_allocator_params *params)
379 params->err_integral = 0;
380 params->prev_err = 0;
383 static void allow_maximum_power(struct thermal_zone_device *tz)
385 struct thermal_instance *instance;
386 struct power_allocator_params *params = tz->governor_data;
388 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
389 if ((instance->trip != params->trip_max_desired_temperature) ||
390 (!cdev_is_power_actor(instance->cdev)))
393 instance->target = 0;
394 instance->cdev->updated = false;
395 thermal_cdev_update(instance->cdev);
400 * power_allocator_bind() - bind the power_allocator governor to a thermal zone
401 * @tz: thermal zone to bind it to
403 * Check that the thermal zone is valid for this governor, that is, it
404 * has two thermal trips. If so, initialize the PID controller
405 * parameters and bind it to the thermal zone.
407 * Return: 0 on success, -EINVAL if the trips were invalid or -ENOMEM
408 * if we ran out of memory.
410 static int power_allocator_bind(struct thermal_zone_device *tz)
413 struct power_allocator_params *params;
414 unsigned long switch_on_temp, control_temp;
415 u32 temperature_threshold;
417 if (!tz->tzp || !tz->tzp->sustainable_power) {
419 "power_allocator: missing sustainable_power\n");
423 params = devm_kzalloc(&tz->device, sizeof(*params), GFP_KERNEL);
427 ret = get_governor_trips(tz, params);
430 "thermal zone %s has wrong trip setup for power allocator\n",
435 ret = tz->ops->get_trip_temp(tz, params->trip_switch_on,
440 ret = tz->ops->get_trip_temp(tz, params->trip_max_desired_temperature,
445 temperature_threshold = control_temp - switch_on_temp;
447 tz->tzp->k_po = tz->tzp->k_po ?:
448 int_to_frac(tz->tzp->sustainable_power) / temperature_threshold;
449 tz->tzp->k_pu = tz->tzp->k_pu ?:
450 int_to_frac(2 * tz->tzp->sustainable_power) /
451 temperature_threshold;
452 tz->tzp->k_i = tz->tzp->k_i ?: int_to_frac(10) / 1000;
454 * The default for k_d and integral_cutoff is 0, so we can
455 * leave them as they are.
458 reset_pid_controller(params);
460 tz->governor_data = params;
465 devm_kfree(&tz->device, params);
469 static void power_allocator_unbind(struct thermal_zone_device *tz)
471 dev_dbg(&tz->device, "Unbinding from thermal zone %d\n", tz->id);
472 devm_kfree(&tz->device, tz->governor_data);
473 tz->governor_data = NULL;
476 static int power_allocator_throttle(struct thermal_zone_device *tz, int trip)
479 unsigned long switch_on_temp, control_temp, current_temp;
480 struct power_allocator_params *params = tz->governor_data;
483 * We get called for every trip point but we only need to do
484 * our calculations once
486 if (trip != params->trip_max_desired_temperature)
489 ret = thermal_zone_get_temp(tz, ¤t_temp);
491 dev_warn(&tz->device, "Failed to get temperature: %d\n", ret);
495 ret = tz->ops->get_trip_temp(tz, params->trip_switch_on,
498 dev_warn(&tz->device,
499 "Failed to get switch on temperature: %d\n", ret);
503 if (current_temp < switch_on_temp) {
505 reset_pid_controller(params);
506 allow_maximum_power(tz);
512 ret = tz->ops->get_trip_temp(tz, params->trip_max_desired_temperature,
515 dev_warn(&tz->device,
516 "Failed to get the maximum desired temperature: %d\n",
521 return allocate_power(tz, current_temp, control_temp);
524 static struct thermal_governor thermal_gov_power_allocator = {
525 .name = "power_allocator",
526 .bind_to_tz = power_allocator_bind,
527 .unbind_from_tz = power_allocator_unbind,
528 .throttle = power_allocator_throttle,
531 int thermal_gov_power_allocator_register(void)
533 return thermal_register_governor(&thermal_gov_power_allocator);
536 void thermal_gov_power_allocator_unregister(void)
538 thermal_unregister_governor(&thermal_gov_power_allocator);