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"
20 #define int_to_frac(x) ((x) << FRAC_BITS)
21 #define frac_to_int(x) ((x) >> FRAC_BITS)
24 * mul_frac() - multiply two fixed-point numbers
25 * @x: first multiplicand
26 * @y: second multiplicand
28 * Return: the result of multiplying two fixed-point numbers. The
29 * result is also a fixed-point number.
31 static inline s64 mul_frac(s64 x, s64 y)
33 return (x * y) >> FRAC_BITS;
37 * div_frac() - divide two fixed-point numbers
41 * Return: the result of dividing two fixed-point numbers. The
42 * result is also a fixed-point number.
44 static inline s64 div_frac(s64 x, s64 y)
46 return div_s64(x << FRAC_BITS, y);
50 * struct power_actor - internal power information for power actor
51 * @req_power: requested power value (not weighted)
52 * @max_power: max allocatable power for this actor
53 * @granted_power: granted power for this actor
54 * @extra_actor_power: extra power that this actor can receive
55 * @weighted_req_power: weighted requested power as input to IPA
61 u32 extra_actor_power;
62 u32 weighted_req_power;
66 * struct power_allocator_params - parameters for the power allocator governor
67 * @allocated_tzp: whether we have allocated tzp for this thermal zone and
68 * it needs to be freed on unbind
69 * @update_cdevs: whether or not update cdevs on the next run
70 * @err_integral: accumulated error in the PID controller.
71 * @prev_err: error in the previous iteration of the PID controller.
72 * Used to calculate the derivative term.
73 * @sustainable_power: Sustainable power (heat) that this thermal zone can
75 * @trip_switch_on: first passive trip point of the thermal zone. The
76 * governor switches on when this trip point is crossed.
77 * If the thermal zone only has one passive trip point,
78 * @trip_switch_on should be NULL.
79 * @trip_max: last passive trip point of the thermal zone. The
80 * temperature we are controlling for.
81 * @total_weight: Sum of all thermal instances weights
82 * @num_actors: number of cooling devices supporting IPA callbacks
83 * @buffer_size: internal buffer size, to avoid runtime re-calculation
84 * @power: buffer for all power actors internal power information
86 struct power_allocator_params {
91 u32 sustainable_power;
92 const struct thermal_trip *trip_switch_on;
93 const struct thermal_trip *trip_max;
95 unsigned int num_actors;
96 unsigned int buffer_size;
97 struct power_actor *power;
100 static bool power_actor_is_valid(struct power_allocator_params *params,
101 struct thermal_instance *instance)
103 return (instance->trip == params->trip_max &&
104 cdev_is_power_actor(instance->cdev));
108 * estimate_sustainable_power() - Estimate the sustainable power of a thermal zone
109 * @tz: thermal zone we are operating in
111 * For thermal zones that don't provide a sustainable_power in their
112 * thermal_zone_params, estimate one. Calculate it using the minimum
113 * power of all the cooling devices as that gives a valid value that
114 * can give some degree of functionality. For optimal performance of
115 * this governor, provide a sustainable_power in the thermal zone's
116 * thermal_zone_params.
118 static u32 estimate_sustainable_power(struct thermal_zone_device *tz)
120 struct power_allocator_params *params = tz->governor_data;
121 struct thermal_cooling_device *cdev;
122 struct thermal_instance *instance;
123 u32 sustainable_power = 0;
126 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
127 if (!power_actor_is_valid(params, instance))
130 cdev = instance->cdev;
131 if (cdev->ops->state2power(cdev, instance->upper, &min_power))
134 sustainable_power += min_power;
137 return sustainable_power;
141 * estimate_pid_constants() - Estimate the constants for the PID controller
142 * @tz: thermal zone for which to estimate the constants
143 * @sustainable_power: sustainable power for the thermal zone
144 * @trip_switch_on: trip point for the switch on temperature
145 * @control_temp: target temperature for the power allocator governor
147 * This function is used to update the estimation of the PID
148 * controller constants in struct thermal_zone_parameters.
150 static void estimate_pid_constants(struct thermal_zone_device *tz,
151 u32 sustainable_power,
152 const struct thermal_trip *trip_switch_on,
155 u32 temperature_threshold = control_temp;
159 temperature_threshold -= trip_switch_on->temperature;
162 * estimate_pid_constants() tries to find appropriate default
163 * values for thermal zones that don't provide them. If a
164 * system integrator has configured a thermal zone with two
165 * passive trip points at the same temperature, that person
166 * hasn't put any effort to set up the thermal zone properly
169 if (!temperature_threshold)
172 tz->tzp->k_po = int_to_frac(sustainable_power) /
173 temperature_threshold;
175 tz->tzp->k_pu = int_to_frac(2 * sustainable_power) /
176 temperature_threshold;
178 k_i = tz->tzp->k_pu / 10;
179 tz->tzp->k_i = k_i > 0 ? k_i : 1;
182 * The default for k_d and integral_cutoff is 0, so we can
183 * leave them as they are.
188 * get_sustainable_power() - Get the right sustainable power
189 * @tz: thermal zone for which to estimate the constants
190 * @params: parameters for the power allocator governor
191 * @control_temp: target temperature for the power allocator governor
193 * This function is used for getting the proper sustainable power value based
194 * on variables which might be updated by the user sysfs interface. If that
195 * happen the new value is going to be estimated and updated. It is also used
196 * after thermal zone binding, where the initial values where set to 0.
198 static u32 get_sustainable_power(struct thermal_zone_device *tz,
199 struct power_allocator_params *params,
202 u32 sustainable_power;
204 if (!tz->tzp->sustainable_power)
205 sustainable_power = estimate_sustainable_power(tz);
207 sustainable_power = tz->tzp->sustainable_power;
209 /* Check if it's init value 0 or there was update via sysfs */
210 if (sustainable_power != params->sustainable_power) {
211 estimate_pid_constants(tz, sustainable_power,
212 params->trip_switch_on, control_temp);
214 /* Do the estimation only once and make available in sysfs */
215 tz->tzp->sustainable_power = sustainable_power;
216 params->sustainable_power = sustainable_power;
219 return sustainable_power;
223 * pid_controller() - PID controller
224 * @tz: thermal zone we are operating in
225 * @control_temp: the target temperature in millicelsius
226 * @max_allocatable_power: maximum allocatable power for this thermal zone
228 * This PID controller increases the available power budget so that the
229 * temperature of the thermal zone gets as close as possible to
230 * @control_temp and limits the power if it exceeds it. k_po is the
231 * proportional term when we are overshooting, k_pu is the
232 * proportional term when we are undershooting. integral_cutoff is a
233 * threshold below which we stop accumulating the error. The
234 * accumulated error is only valid if the requested power will make
235 * the system warmer. If the system is mostly idle, there's no point
236 * in accumulating positive error.
238 * Return: The power budget for the next period.
240 static u32 pid_controller(struct thermal_zone_device *tz,
242 u32 max_allocatable_power)
244 struct power_allocator_params *params = tz->governor_data;
245 s64 p, i, d, power_range;
246 s32 err, max_power_frac;
247 u32 sustainable_power;
249 max_power_frac = int_to_frac(max_allocatable_power);
251 sustainable_power = get_sustainable_power(tz, params, control_temp);
253 err = control_temp - tz->temperature;
254 err = int_to_frac(err);
256 /* Calculate the proportional term */
257 p = mul_frac(err < 0 ? tz->tzp->k_po : tz->tzp->k_pu, err);
260 * Calculate the integral term
262 * if the error is less than cut off allow integration (but
263 * the integral is limited to max power)
265 i = mul_frac(tz->tzp->k_i, params->err_integral);
267 if (err < int_to_frac(tz->tzp->integral_cutoff)) {
268 s64 i_next = i + mul_frac(tz->tzp->k_i, err);
270 if (abs(i_next) < max_power_frac) {
272 params->err_integral += err;
277 * Calculate the derivative term
279 * We do err - prev_err, so with a positive k_d, a decreasing
280 * error (i.e. driving closer to the line) results in less
281 * power being applied, slowing down the controller)
283 d = mul_frac(tz->tzp->k_d, err - params->prev_err);
284 d = div_frac(d, jiffies_to_msecs(tz->passive_delay_jiffies));
285 params->prev_err = err;
287 power_range = p + i + d;
289 /* feed-forward the known sustainable dissipatable power */
290 power_range = sustainable_power + frac_to_int(power_range);
292 power_range = clamp(power_range, (s64)0, (s64)max_allocatable_power);
294 trace_thermal_power_allocator_pid(tz, frac_to_int(err),
295 frac_to_int(params->err_integral),
296 frac_to_int(p), frac_to_int(i),
297 frac_to_int(d), power_range);
303 * power_actor_set_power() - limit the maximum power a cooling device consumes
304 * @cdev: pointer to &thermal_cooling_device
305 * @instance: thermal instance to update
306 * @power: the power in milliwatts
308 * Set the cooling device to consume at most @power milliwatts. The limit is
309 * expected to be a cap at the maximum power consumption.
311 * Return: 0 on success, -EINVAL if the cooling device does not
312 * implement the power actor API or -E* for other failures.
315 power_actor_set_power(struct thermal_cooling_device *cdev,
316 struct thermal_instance *instance, u32 power)
321 ret = cdev->ops->power2state(cdev, power, &state);
325 instance->target = clamp_val(state, instance->lower, instance->upper);
326 mutex_lock(&cdev->lock);
327 __thermal_cdev_update(cdev);
328 mutex_unlock(&cdev->lock);
334 * divvy_up_power() - divvy the allocated power between the actors
335 * @power: buffer for all power actors internal power information
336 * @num_actors: number of power actors in this thermal zone
337 * @total_req_power: sum of all weighted requested power for all actors
338 * @power_range: total allocated power
340 * This function divides the total allocated power (@power_range)
341 * fairly between the actors. It first tries to give each actor a
342 * share of the @power_range according to how much power it requested
343 * compared to the rest of the actors. For example, if only one actor
344 * requests power, then it receives all the @power_range. If
345 * three actors each requests 1mW, each receives a third of the
348 * If any actor received more than their maximum power, then that
349 * surplus is re-divvied among the actors based on how far they are
350 * from their respective maximums.
352 static void divvy_up_power(struct power_actor *power, int num_actors,
353 u32 total_req_power, u32 power_range)
355 u32 capped_extra_power = 0;
360 * Prevent division by 0 if none of the actors request power.
362 if (!total_req_power)
365 for (i = 0; i < num_actors; i++) {
366 struct power_actor *pa = &power[i];
367 u64 req_range = (u64)pa->req_power * power_range;
369 pa->granted_power = DIV_ROUND_CLOSEST_ULL(req_range,
372 if (pa->granted_power > pa->max_power) {
373 extra_power += pa->granted_power - pa->max_power;
374 pa->granted_power = pa->max_power;
377 pa->extra_actor_power = pa->max_power - pa->granted_power;
378 capped_extra_power += pa->extra_actor_power;
381 if (!extra_power || !capped_extra_power)
385 * Re-divvy the reclaimed extra among actors based on
386 * how far they are from the max
388 extra_power = min(extra_power, capped_extra_power);
390 for (i = 0; i < num_actors; i++) {
391 struct power_actor *pa = &power[i];
392 u64 extra_range = pa->extra_actor_power;
394 extra_range *= extra_power;
395 pa->granted_power += DIV_ROUND_CLOSEST_ULL(extra_range,
400 static void allocate_power(struct thermal_zone_device *tz, int control_temp)
402 struct power_allocator_params *params = tz->governor_data;
403 unsigned int num_actors = params->num_actors;
404 struct power_actor *power = params->power;
405 struct thermal_cooling_device *cdev;
406 struct thermal_instance *instance;
407 u32 total_weighted_req_power = 0;
408 u32 max_allocatable_power = 0;
409 u32 total_granted_power = 0;
410 u32 total_req_power = 0;
411 u32 power_range, weight;
417 /* Clean all buffers for new power estimations */
418 memset(power, 0, params->buffer_size);
420 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
421 struct power_actor *pa = &power[i];
423 if (!power_actor_is_valid(params, instance))
426 cdev = instance->cdev;
428 ret = cdev->ops->get_requested_power(cdev, &pa->req_power);
432 if (!params->total_weight)
433 weight = 1 << FRAC_BITS;
435 weight = instance->weight;
437 pa->weighted_req_power = frac_to_int(weight * pa->req_power);
439 ret = cdev->ops->state2power(cdev, instance->lower,
444 total_req_power += pa->req_power;
445 max_allocatable_power += pa->max_power;
446 total_weighted_req_power += pa->weighted_req_power;
451 power_range = pid_controller(tz, control_temp, max_allocatable_power);
453 divvy_up_power(power, num_actors, total_weighted_req_power,
457 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
458 struct power_actor *pa = &power[i];
460 if (!power_actor_is_valid(params, instance))
463 power_actor_set_power(instance->cdev, instance,
465 total_granted_power += pa->granted_power;
467 trace_thermal_power_actor(tz, i, pa->req_power,
472 trace_thermal_power_allocator(tz, total_req_power, total_granted_power,
473 num_actors, power_range,
474 max_allocatable_power, tz->temperature,
475 control_temp - tz->temperature);
479 * get_governor_trips() - get the two trip points that are key for this governor
480 * @tz: thermal zone to operate on
481 * @params: pointer to private data for this governor
483 * The power allocator governor works optimally with two trips points:
484 * a "switch on" trip point and a "maximum desired temperature". These
485 * are defined as the first and last passive trip points.
487 * If there is only one trip point, then that's considered to be the
488 * "maximum desired temperature" trip point and the governor is always
489 * on. If there are no passive or active trip points, then the
490 * governor won't do anything. In fact, its throttle function
491 * won't be called at all.
493 static void get_governor_trips(struct thermal_zone_device *tz,
494 struct power_allocator_params *params)
496 const struct thermal_trip *first_passive = NULL;
497 const struct thermal_trip *last_passive = NULL;
498 const struct thermal_trip *last_active = NULL;
499 const struct thermal_trip_desc *td;
501 for_each_trip_desc(tz, td) {
502 const struct thermal_trip *trip = &td->trip;
504 switch (trip->type) {
505 case THERMAL_TRIP_PASSIVE:
506 if (!first_passive) {
507 first_passive = trip;
512 case THERMAL_TRIP_ACTIVE:
521 params->trip_switch_on = first_passive;
522 params->trip_max = last_passive;
523 } else if (first_passive) {
524 params->trip_switch_on = NULL;
525 params->trip_max = first_passive;
527 params->trip_switch_on = NULL;
528 params->trip_max = last_active;
532 static void reset_pid_controller(struct power_allocator_params *params)
534 params->err_integral = 0;
535 params->prev_err = 0;
538 static void allow_maximum_power(struct thermal_zone_device *tz)
540 struct power_allocator_params *params = tz->governor_data;
541 struct thermal_cooling_device *cdev;
542 struct thermal_instance *instance;
545 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
546 if (!power_actor_is_valid(params, instance))
549 cdev = instance->cdev;
551 instance->target = 0;
552 mutex_lock(&cdev->lock);
554 * Call for updating the cooling devices local stats and avoid
555 * periods of dozen of seconds when those have not been
558 cdev->ops->get_requested_power(cdev, &req_power);
560 if (params->update_cdevs)
561 __thermal_cdev_update(cdev);
563 mutex_unlock(&cdev->lock);
568 * check_power_actors() - Check all cooling devices and warn when they are
570 * @tz: thermal zone to operate on
571 * @params: power allocator private data
573 * Check all cooling devices in the @tz and warn every time they are missing
574 * power actor API. The warning should help to investigate the issue, which
575 * could be e.g. lack of Energy Model for a given device.
577 * If all of the cooling devices currently attached to @tz implement the power
578 * actor API, return the number of them (which may be 0, because some cooling
579 * devices may be attached later). Otherwise, return -EINVAL.
581 static int check_power_actors(struct thermal_zone_device *tz,
582 struct power_allocator_params *params)
584 struct thermal_instance *instance;
587 list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
588 if (instance->trip != params->trip_max)
591 if (!cdev_is_power_actor(instance->cdev)) {
592 dev_warn(&tz->device, "power_allocator: %s is not a power actor\n",
593 instance->cdev->type);
602 static int allocate_actors_buffer(struct power_allocator_params *params,
607 kfree(params->power);
609 /* There might be no cooling devices yet. */
615 params->power = kcalloc(num_actors, sizeof(struct power_actor),
617 if (!params->power) {
622 params->num_actors = num_actors;
623 params->buffer_size = num_actors * sizeof(struct power_actor);
628 params->num_actors = 0;
629 params->buffer_size = 0;
630 params->power = NULL;
634 static void power_allocator_update_tz(struct thermal_zone_device *tz,
635 enum thermal_notify_event reason)
637 struct power_allocator_params *params = tz->governor_data;
638 struct thermal_instance *instance;
642 case THERMAL_TZ_BIND_CDEV:
643 case THERMAL_TZ_UNBIND_CDEV:
644 list_for_each_entry(instance, &tz->thermal_instances, tz_node)
645 if (power_actor_is_valid(params, instance))
648 if (num_actors == params->num_actors)
651 allocate_actors_buffer(params, num_actors);
653 case THERMAL_INSTANCE_WEIGHT_CHANGED:
654 params->total_weight = 0;
655 list_for_each_entry(instance, &tz->thermal_instances, tz_node)
656 if (power_actor_is_valid(params, instance))
657 params->total_weight += instance->weight;
665 * power_allocator_bind() - bind the power_allocator governor to a thermal zone
666 * @tz: thermal zone to bind it to
668 * Initialize the PID controller parameters and bind it to the thermal
671 * Return: 0 on success, or -ENOMEM if we ran out of memory, or -EINVAL
672 * when there are unsupported cooling devices in the @tz.
674 static int power_allocator_bind(struct thermal_zone_device *tz)
676 struct power_allocator_params *params;
679 params = kzalloc(sizeof(*params), GFP_KERNEL);
683 get_governor_trips(tz, params);
685 ret = check_power_actors(tz, params);
687 dev_warn(&tz->device, "power_allocator: binding failed\n");
692 ret = allocate_actors_buffer(params, ret);
694 dev_warn(&tz->device, "power_allocator: allocation failed\n");
700 tz->tzp = kzalloc(sizeof(*tz->tzp), GFP_KERNEL);
706 params->allocated_tzp = true;
709 if (!tz->tzp->sustainable_power)
710 dev_warn(&tz->device, "power_allocator: sustainable_power will be estimated\n");
712 params->sustainable_power = tz->tzp->sustainable_power;
714 if (params->trip_max)
715 estimate_pid_constants(tz, tz->tzp->sustainable_power,
716 params->trip_switch_on,
717 params->trip_max->temperature);
719 reset_pid_controller(params);
721 tz->governor_data = params;
726 kfree(params->power);
732 static void power_allocator_unbind(struct thermal_zone_device *tz)
734 struct power_allocator_params *params = tz->governor_data;
736 dev_dbg(&tz->device, "Unbinding from thermal zone %d\n", tz->id);
738 if (params->allocated_tzp) {
743 kfree(params->power);
744 kfree(tz->governor_data);
745 tz->governor_data = NULL;
748 static void power_allocator_manage(struct thermal_zone_device *tz)
750 struct power_allocator_params *params = tz->governor_data;
751 const struct thermal_trip *trip = params->trip_switch_on;
753 lockdep_assert_held(&tz->lock);
755 if (trip && tz->temperature < trip->temperature) {
756 reset_pid_controller(params);
757 allow_maximum_power(tz);
758 params->update_cdevs = false;
762 if (!params->trip_max)
765 allocate_power(tz, params->trip_max->temperature);
766 params->update_cdevs = true;
769 static struct thermal_governor thermal_gov_power_allocator = {
770 .name = "power_allocator",
771 .bind_to_tz = power_allocator_bind,
772 .unbind_from_tz = power_allocator_unbind,
773 .manage = power_allocator_manage,
774 .update_tz = power_allocator_update_tz,
776 THERMAL_GOVERNOR_DECLARE(thermal_gov_power_allocator);