cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

gov_power_allocator.c (21719B)


      1// SPDX-License-Identifier: GPL-2.0
      2/*
      3 * A power allocator to manage temperature
      4 *
      5 * Copyright (C) 2014 ARM Ltd.
      6 *
      7 */
      8
      9#define pr_fmt(fmt) "Power allocator: " fmt
     10
     11#include <linux/rculist.h>
     12#include <linux/slab.h>
     13#include <linux/thermal.h>
     14
     15#define CREATE_TRACE_POINTS
     16#include <trace/events/thermal_power_allocator.h>
     17
     18#include "thermal_core.h"
     19
     20#define INVALID_TRIP -1
     21
     22#define FRAC_BITS 10
     23#define int_to_frac(x) ((x) << FRAC_BITS)
     24#define frac_to_int(x) ((x) >> FRAC_BITS)
     25
     26/**
     27 * mul_frac() - multiply two fixed-point numbers
     28 * @x:	first multiplicand
     29 * @y:	second multiplicand
     30 *
     31 * Return: the result of multiplying two fixed-point numbers.  The
     32 * result is also a fixed-point number.
     33 */
     34static inline s64 mul_frac(s64 x, s64 y)
     35{
     36	return (x * y) >> FRAC_BITS;
     37}
     38
     39/**
     40 * div_frac() - divide two fixed-point numbers
     41 * @x:	the dividend
     42 * @y:	the divisor
     43 *
     44 * Return: the result of dividing two fixed-point numbers.  The
     45 * result is also a fixed-point number.
     46 */
     47static inline s64 div_frac(s64 x, s64 y)
     48{
     49	return div_s64(x << FRAC_BITS, y);
     50}
     51
     52/**
     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
     65 *					controlling for.
     66 * @sustainable_power:	Sustainable power (heat) that this thermal zone can
     67 *			dissipate
     68 */
     69struct power_allocator_params {
     70	bool allocated_tzp;
     71	s64 err_integral;
     72	s32 prev_err;
     73	int trip_switch_on;
     74	int trip_max_desired_temperature;
     75	u32 sustainable_power;
     76};
     77
     78/**
     79 * estimate_sustainable_power() - Estimate the sustainable power of a thermal zone
     80 * @tz: thermal zone we are operating in
     81 *
     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.
     88 */
     89static u32 estimate_sustainable_power(struct thermal_zone_device *tz)
     90{
     91	u32 sustainable_power = 0;
     92	struct thermal_instance *instance;
     93	struct power_allocator_params *params = tz->governor_data;
     94
     95	list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
     96		struct thermal_cooling_device *cdev = instance->cdev;
     97		u32 min_power;
     98
     99		if (instance->trip != params->trip_max_desired_temperature)
    100			continue;
    101
    102		if (!cdev_is_power_actor(cdev))
    103			continue;
    104
    105		if (cdev->ops->state2power(cdev, instance->upper, &min_power))
    106			continue;
    107
    108		sustainable_power += min_power;
    109	}
    110
    111	return sustainable_power;
    112}
    113
    114/**
    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 *
    121 * This function is used to update the estimation of the PID
    122 * controller constants in struct thermal_zone_parameters.
    123 */
    124static void estimate_pid_constants(struct thermal_zone_device *tz,
    125				   u32 sustainable_power, int trip_switch_on,
    126				   int control_temp)
    127{
    128	int ret;
    129	int switch_on_temp;
    130	u32 temperature_threshold;
    131	s32 k_i;
    132
    133	ret = tz->ops->get_trip_temp(tz, trip_switch_on, &switch_on_temp);
    134	if (ret)
    135		switch_on_temp = 0;
    136
    137	temperature_threshold = control_temp - switch_on_temp;
    138	/*
    139	 * estimate_pid_constants() tries to find appropriate default
    140	 * values for thermal zones that don't provide them. If a
    141	 * system integrator has configured a thermal zone with two
    142	 * passive trip points at the same temperature, that person
    143	 * hasn't put any effort to set up the thermal zone properly
    144	 * so just give up.
    145	 */
    146	if (!temperature_threshold)
    147		return;
    148
    149	tz->tzp->k_po = int_to_frac(sustainable_power) /
    150		temperature_threshold;
    151
    152	tz->tzp->k_pu = int_to_frac(2 * sustainable_power) /
    153		temperature_threshold;
    154
    155	k_i = tz->tzp->k_pu / 10;
    156	tz->tzp->k_i = k_i > 0 ? k_i : 1;
    157
    158	/*
    159	 * The default for k_d and integral_cutoff is 0, so we can
    160	 * leave them as they are.
    161	 */
    162}
    163
    164/**
    165 * get_sustainable_power() - Get the right sustainable power
    166 * @tz:		thermal zone for which to estimate the constants
    167 * @params:	parameters for the power allocator governor
    168 * @control_temp:	target temperature for the power allocator governor
    169 *
    170 * This function is used for getting the proper sustainable power value based
    171 * on variables which might be updated by the user sysfs interface. If that
    172 * happen the new value is going to be estimated and updated. It is also used
    173 * after thermal zone binding, where the initial values where set to 0.
    174 */
    175static u32 get_sustainable_power(struct thermal_zone_device *tz,
    176				 struct power_allocator_params *params,
    177				 int control_temp)
    178{
    179	u32 sustainable_power;
    180
    181	if (!tz->tzp->sustainable_power)
    182		sustainable_power = estimate_sustainable_power(tz);
    183	else
    184		sustainable_power = tz->tzp->sustainable_power;
    185
    186	/* Check if it's init value 0 or there was update via sysfs */
    187	if (sustainable_power != params->sustainable_power) {
    188		estimate_pid_constants(tz, sustainable_power,
    189				       params->trip_switch_on, control_temp);
    190
    191		/* Do the estimation only once and make available in sysfs */
    192		tz->tzp->sustainable_power = sustainable_power;
    193		params->sustainable_power = sustainable_power;
    194	}
    195
    196	return sustainable_power;
    197}
    198
    199/**
    200 * pid_controller() - PID controller
    201 * @tz:	thermal zone we are operating in
    202 * @control_temp:	the target temperature in millicelsius
    203 * @max_allocatable_power:	maximum allocatable power for this thermal zone
    204 *
    205 * This PID controller increases the available power budget so that the
    206 * temperature of the thermal zone gets as close as possible to
    207 * @control_temp and limits the power if it exceeds it.  k_po is the
    208 * proportional term when we are overshooting, k_pu is the
    209 * proportional term when we are undershooting.  integral_cutoff is a
    210 * threshold below which we stop accumulating the error.  The
    211 * accumulated error is only valid if the requested power will make
    212 * the system warmer.  If the system is mostly idle, there's no point
    213 * in accumulating positive error.
    214 *
    215 * Return: The power budget for the next period.
    216 */
    217static u32 pid_controller(struct thermal_zone_device *tz,
    218			  int control_temp,
    219			  u32 max_allocatable_power)
    220{
    221	s64 p, i, d, power_range;
    222	s32 err, max_power_frac;
    223	u32 sustainable_power;
    224	struct power_allocator_params *params = tz->governor_data;
    225
    226	max_power_frac = int_to_frac(max_allocatable_power);
    227
    228	sustainable_power = get_sustainable_power(tz, params, control_temp);
    229
    230	err = control_temp - tz->temperature;
    231	err = int_to_frac(err);
    232
    233	/* Calculate the proportional term */
    234	p = mul_frac(err < 0 ? tz->tzp->k_po : tz->tzp->k_pu, err);
    235
    236	/*
    237	 * Calculate the integral term
    238	 *
    239	 * if the error is less than cut off allow integration (but
    240	 * the integral is limited to max power)
    241	 */
    242	i = mul_frac(tz->tzp->k_i, params->err_integral);
    243
    244	if (err < int_to_frac(tz->tzp->integral_cutoff)) {
    245		s64 i_next = i + mul_frac(tz->tzp->k_i, err);
    246
    247		if (abs(i_next) < max_power_frac) {
    248			i = i_next;
    249			params->err_integral += err;
    250		}
    251	}
    252
    253	/*
    254	 * Calculate the derivative term
    255	 *
    256	 * We do err - prev_err, so with a positive k_d, a decreasing
    257	 * error (i.e. driving closer to the line) results in less
    258	 * power being applied, slowing down the controller)
    259	 */
    260	d = mul_frac(tz->tzp->k_d, err - params->prev_err);
    261	d = div_frac(d, jiffies_to_msecs(tz->passive_delay_jiffies));
    262	params->prev_err = err;
    263
    264	power_range = p + i + d;
    265
    266	/* feed-forward the known sustainable dissipatable power */
    267	power_range = sustainable_power + frac_to_int(power_range);
    268
    269	power_range = clamp(power_range, (s64)0, (s64)max_allocatable_power);
    270
    271	trace_thermal_power_allocator_pid(tz, frac_to_int(err),
    272					  frac_to_int(params->err_integral),
    273					  frac_to_int(p), frac_to_int(i),
    274					  frac_to_int(d), power_range);
    275
    276	return power_range;
    277}
    278
    279/**
    280 * power_actor_set_power() - limit the maximum power a cooling device consumes
    281 * @cdev:	pointer to &thermal_cooling_device
    282 * @instance:	thermal instance to update
    283 * @power:	the power in milliwatts
    284 *
    285 * Set the cooling device to consume at most @power milliwatts. The limit is
    286 * expected to be a cap at the maximum power consumption.
    287 *
    288 * Return: 0 on success, -EINVAL if the cooling device does not
    289 * implement the power actor API or -E* for other failures.
    290 */
    291static int
    292power_actor_set_power(struct thermal_cooling_device *cdev,
    293		      struct thermal_instance *instance, u32 power)
    294{
    295	unsigned long state;
    296	int ret;
    297
    298	ret = cdev->ops->power2state(cdev, power, &state);
    299	if (ret)
    300		return ret;
    301
    302	instance->target = clamp_val(state, instance->lower, instance->upper);
    303	mutex_lock(&cdev->lock);
    304	__thermal_cdev_update(cdev);
    305	mutex_unlock(&cdev->lock);
    306
    307	return 0;
    308}
    309
    310/**
    311 * divvy_up_power() - divvy the allocated power between the actors
    312 * @req_power:	each actor's requested power
    313 * @max_power:	each actor's maximum available power
    314 * @num_actors:	size of the @req_power, @max_power and @granted_power's array
    315 * @total_req_power: sum of @req_power
    316 * @power_range:	total allocated power
    317 * @granted_power:	output array: each actor's granted power
    318 * @extra_actor_power:	an appropriately sized array to be used in the
    319 *			function as temporary storage of the extra power given
    320 *			to the actors
    321 *
    322 * This function divides the total allocated power (@power_range)
    323 * fairly between the actors.  It first tries to give each actor a
    324 * share of the @power_range according to how much power it requested
    325 * compared to the rest of the actors.  For example, if only one actor
    326 * requests power, then it receives all the @power_range.  If
    327 * three actors each requests 1mW, each receives a third of the
    328 * @power_range.
    329 *
    330 * If any actor received more than their maximum power, then that
    331 * surplus is re-divvied among the actors based on how far they are
    332 * from their respective maximums.
    333 *
    334 * Granted power for each actor is written to @granted_power, which
    335 * should've been allocated by the calling function.
    336 */
    337static void divvy_up_power(u32 *req_power, u32 *max_power, int num_actors,
    338			   u32 total_req_power, u32 power_range,
    339			   u32 *granted_power, u32 *extra_actor_power)
    340{
    341	u32 extra_power, capped_extra_power;
    342	int i;
    343
    344	/*
    345	 * Prevent division by 0 if none of the actors request power.
    346	 */
    347	if (!total_req_power)
    348		total_req_power = 1;
    349
    350	capped_extra_power = 0;
    351	extra_power = 0;
    352	for (i = 0; i < num_actors; i++) {
    353		u64 req_range = (u64)req_power[i] * power_range;
    354
    355		granted_power[i] = DIV_ROUND_CLOSEST_ULL(req_range,
    356							 total_req_power);
    357
    358		if (granted_power[i] > max_power[i]) {
    359			extra_power += granted_power[i] - max_power[i];
    360			granted_power[i] = max_power[i];
    361		}
    362
    363		extra_actor_power[i] = max_power[i] - granted_power[i];
    364		capped_extra_power += extra_actor_power[i];
    365	}
    366
    367	if (!extra_power)
    368		return;
    369
    370	/*
    371	 * Re-divvy the reclaimed extra among actors based on
    372	 * how far they are from the max
    373	 */
    374	extra_power = min(extra_power, capped_extra_power);
    375	if (capped_extra_power > 0)
    376		for (i = 0; i < num_actors; i++) {
    377			u64 extra_range = (u64)extra_actor_power[i] * extra_power;
    378			granted_power[i] += DIV_ROUND_CLOSEST_ULL(extra_range,
    379							 capped_extra_power);
    380		}
    381}
    382
    383static int allocate_power(struct thermal_zone_device *tz,
    384			  int control_temp)
    385{
    386	struct thermal_instance *instance;
    387	struct power_allocator_params *params = tz->governor_data;
    388	u32 *req_power, *max_power, *granted_power, *extra_actor_power;
    389	u32 *weighted_req_power;
    390	u32 total_req_power, max_allocatable_power, total_weighted_req_power;
    391	u32 total_granted_power, power_range;
    392	int i, num_actors, total_weight, ret = 0;
    393	int trip_max_desired_temperature = params->trip_max_desired_temperature;
    394
    395	mutex_lock(&tz->lock);
    396
    397	num_actors = 0;
    398	total_weight = 0;
    399	list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
    400		if ((instance->trip == trip_max_desired_temperature) &&
    401		    cdev_is_power_actor(instance->cdev)) {
    402			num_actors++;
    403			total_weight += instance->weight;
    404		}
    405	}
    406
    407	if (!num_actors) {
    408		ret = -ENODEV;
    409		goto unlock;
    410	}
    411
    412	/*
    413	 * We need to allocate five arrays of the same size:
    414	 * req_power, max_power, granted_power, extra_actor_power and
    415	 * weighted_req_power.  They are going to be needed until this
    416	 * function returns.  Allocate them all in one go to simplify
    417	 * the allocation and deallocation logic.
    418	 */
    419	BUILD_BUG_ON(sizeof(*req_power) != sizeof(*max_power));
    420	BUILD_BUG_ON(sizeof(*req_power) != sizeof(*granted_power));
    421	BUILD_BUG_ON(sizeof(*req_power) != sizeof(*extra_actor_power));
    422	BUILD_BUG_ON(sizeof(*req_power) != sizeof(*weighted_req_power));
    423	req_power = kcalloc(num_actors * 5, sizeof(*req_power), GFP_KERNEL);
    424	if (!req_power) {
    425		ret = -ENOMEM;
    426		goto unlock;
    427	}
    428
    429	max_power = &req_power[num_actors];
    430	granted_power = &req_power[2 * num_actors];
    431	extra_actor_power = &req_power[3 * num_actors];
    432	weighted_req_power = &req_power[4 * num_actors];
    433
    434	i = 0;
    435	total_weighted_req_power = 0;
    436	total_req_power = 0;
    437	max_allocatable_power = 0;
    438
    439	list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
    440		int weight;
    441		struct thermal_cooling_device *cdev = instance->cdev;
    442
    443		if (instance->trip != trip_max_desired_temperature)
    444			continue;
    445
    446		if (!cdev_is_power_actor(cdev))
    447			continue;
    448
    449		if (cdev->ops->get_requested_power(cdev, &req_power[i]))
    450			continue;
    451
    452		if (!total_weight)
    453			weight = 1 << FRAC_BITS;
    454		else
    455			weight = instance->weight;
    456
    457		weighted_req_power[i] = frac_to_int(weight * req_power[i]);
    458
    459		if (cdev->ops->state2power(cdev, instance->lower,
    460					   &max_power[i]))
    461			continue;
    462
    463		total_req_power += req_power[i];
    464		max_allocatable_power += max_power[i];
    465		total_weighted_req_power += weighted_req_power[i];
    466
    467		i++;
    468	}
    469
    470	power_range = pid_controller(tz, control_temp, max_allocatable_power);
    471
    472	divvy_up_power(weighted_req_power, max_power, num_actors,
    473		       total_weighted_req_power, power_range, granted_power,
    474		       extra_actor_power);
    475
    476	total_granted_power = 0;
    477	i = 0;
    478	list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
    479		if (instance->trip != trip_max_desired_temperature)
    480			continue;
    481
    482		if (!cdev_is_power_actor(instance->cdev))
    483			continue;
    484
    485		power_actor_set_power(instance->cdev, instance,
    486				      granted_power[i]);
    487		total_granted_power += granted_power[i];
    488
    489		i++;
    490	}
    491
    492	trace_thermal_power_allocator(tz, req_power, total_req_power,
    493				      granted_power, total_granted_power,
    494				      num_actors, power_range,
    495				      max_allocatable_power, tz->temperature,
    496				      control_temp - tz->temperature);
    497
    498	kfree(req_power);
    499unlock:
    500	mutex_unlock(&tz->lock);
    501
    502	return ret;
    503}
    504
    505/**
    506 * get_governor_trips() - get the number of the two trip points that are key for this governor
    507 * @tz:	thermal zone to operate on
    508 * @params:	pointer to private data for this governor
    509 *
    510 * The power allocator governor works optimally with two trips points:
    511 * a "switch on" trip point and a "maximum desired temperature".  These
    512 * are defined as the first and last passive trip points.
    513 *
    514 * If there is only one trip point, then that's considered to be the
    515 * "maximum desired temperature" trip point and the governor is always
    516 * on.  If there are no passive or active trip points, then the
    517 * governor won't do anything.  In fact, its throttle function
    518 * won't be called at all.
    519 */
    520static void get_governor_trips(struct thermal_zone_device *tz,
    521			       struct power_allocator_params *params)
    522{
    523	int i, last_active, last_passive;
    524	bool found_first_passive;
    525
    526	found_first_passive = false;
    527	last_active = INVALID_TRIP;
    528	last_passive = INVALID_TRIP;
    529
    530	for (i = 0; i < tz->trips; i++) {
    531		enum thermal_trip_type type;
    532		int ret;
    533
    534		ret = tz->ops->get_trip_type(tz, i, &type);
    535		if (ret) {
    536			dev_warn(&tz->device,
    537				 "Failed to get trip point %d type: %d\n", i,
    538				 ret);
    539			continue;
    540		}
    541
    542		if (type == THERMAL_TRIP_PASSIVE) {
    543			if (!found_first_passive) {
    544				params->trip_switch_on = i;
    545				found_first_passive = true;
    546			} else  {
    547				last_passive = i;
    548			}
    549		} else if (type == THERMAL_TRIP_ACTIVE) {
    550			last_active = i;
    551		} else {
    552			break;
    553		}
    554	}
    555
    556	if (last_passive != INVALID_TRIP) {
    557		params->trip_max_desired_temperature = last_passive;
    558	} else if (found_first_passive) {
    559		params->trip_max_desired_temperature = params->trip_switch_on;
    560		params->trip_switch_on = INVALID_TRIP;
    561	} else {
    562		params->trip_switch_on = INVALID_TRIP;
    563		params->trip_max_desired_temperature = last_active;
    564	}
    565}
    566
    567static void reset_pid_controller(struct power_allocator_params *params)
    568{
    569	params->err_integral = 0;
    570	params->prev_err = 0;
    571}
    572
    573static void allow_maximum_power(struct thermal_zone_device *tz, bool update)
    574{
    575	struct thermal_instance *instance;
    576	struct power_allocator_params *params = tz->governor_data;
    577	u32 req_power;
    578
    579	mutex_lock(&tz->lock);
    580	list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
    581		struct thermal_cooling_device *cdev = instance->cdev;
    582
    583		if ((instance->trip != params->trip_max_desired_temperature) ||
    584		    (!cdev_is_power_actor(instance->cdev)))
    585			continue;
    586
    587		instance->target = 0;
    588		mutex_lock(&instance->cdev->lock);
    589		/*
    590		 * Call for updating the cooling devices local stats and avoid
    591		 * periods of dozen of seconds when those have not been
    592		 * maintained.
    593		 */
    594		cdev->ops->get_requested_power(cdev, &req_power);
    595
    596		if (update)
    597			__thermal_cdev_update(instance->cdev);
    598
    599		mutex_unlock(&instance->cdev->lock);
    600	}
    601	mutex_unlock(&tz->lock);
    602}
    603
    604/**
    605 * check_power_actors() - Check all cooling devices and warn when they are
    606 *			not power actors
    607 * @tz:		thermal zone to operate on
    608 *
    609 * Check all cooling devices in the @tz and warn every time they are missing
    610 * power actor API. The warning should help to investigate the issue, which
    611 * could be e.g. lack of Energy Model for a given device.
    612 *
    613 * Return: 0 on success, -EINVAL if any cooling device does not implement
    614 * the power actor API.
    615 */
    616static int check_power_actors(struct thermal_zone_device *tz)
    617{
    618	struct thermal_instance *instance;
    619	int ret = 0;
    620
    621	list_for_each_entry(instance, &tz->thermal_instances, tz_node) {
    622		if (!cdev_is_power_actor(instance->cdev)) {
    623			dev_warn(&tz->device, "power_allocator: %s is not a power actor\n",
    624				 instance->cdev->type);
    625			ret = -EINVAL;
    626		}
    627	}
    628
    629	return ret;
    630}
    631
    632/**
    633 * power_allocator_bind() - bind the power_allocator governor to a thermal zone
    634 * @tz:	thermal zone to bind it to
    635 *
    636 * Initialize the PID controller parameters and bind it to the thermal
    637 * zone.
    638 *
    639 * Return: 0 on success, or -ENOMEM if we ran out of memory, or -EINVAL
    640 * when there are unsupported cooling devices in the @tz.
    641 */
    642static int power_allocator_bind(struct thermal_zone_device *tz)
    643{
    644	int ret;
    645	struct power_allocator_params *params;
    646	int control_temp;
    647
    648	ret = check_power_actors(tz);
    649	if (ret)
    650		return ret;
    651
    652	params = kzalloc(sizeof(*params), GFP_KERNEL);
    653	if (!params)
    654		return -ENOMEM;
    655
    656	if (!tz->tzp) {
    657		tz->tzp = kzalloc(sizeof(*tz->tzp), GFP_KERNEL);
    658		if (!tz->tzp) {
    659			ret = -ENOMEM;
    660			goto free_params;
    661		}
    662
    663		params->allocated_tzp = true;
    664	}
    665
    666	if (!tz->tzp->sustainable_power)
    667		dev_warn(&tz->device, "power_allocator: sustainable_power will be estimated\n");
    668
    669	get_governor_trips(tz, params);
    670
    671	if (tz->trips > 0) {
    672		ret = tz->ops->get_trip_temp(tz,
    673					params->trip_max_desired_temperature,
    674					&control_temp);
    675		if (!ret)
    676			estimate_pid_constants(tz, tz->tzp->sustainable_power,
    677					       params->trip_switch_on,
    678					       control_temp);
    679	}
    680
    681	reset_pid_controller(params);
    682
    683	tz->governor_data = params;
    684
    685	return 0;
    686
    687free_params:
    688	kfree(params);
    689
    690	return ret;
    691}
    692
    693static void power_allocator_unbind(struct thermal_zone_device *tz)
    694{
    695	struct power_allocator_params *params = tz->governor_data;
    696
    697	dev_dbg(&tz->device, "Unbinding from thermal zone %d\n", tz->id);
    698
    699	if (params->allocated_tzp) {
    700		kfree(tz->tzp);
    701		tz->tzp = NULL;
    702	}
    703
    704	kfree(tz->governor_data);
    705	tz->governor_data = NULL;
    706}
    707
    708static int power_allocator_throttle(struct thermal_zone_device *tz, int trip)
    709{
    710	int ret;
    711	int switch_on_temp, control_temp;
    712	struct power_allocator_params *params = tz->governor_data;
    713	bool update;
    714
    715	/*
    716	 * We get called for every trip point but we only need to do
    717	 * our calculations once
    718	 */
    719	if (trip != params->trip_max_desired_temperature)
    720		return 0;
    721
    722	ret = tz->ops->get_trip_temp(tz, params->trip_switch_on,
    723				     &switch_on_temp);
    724	if (!ret && (tz->temperature < switch_on_temp)) {
    725		update = (tz->last_temperature >= switch_on_temp);
    726		tz->passive = 0;
    727		reset_pid_controller(params);
    728		allow_maximum_power(tz, update);
    729		return 0;
    730	}
    731
    732	tz->passive = 1;
    733
    734	ret = tz->ops->get_trip_temp(tz, params->trip_max_desired_temperature,
    735				&control_temp);
    736	if (ret) {
    737		dev_warn(&tz->device,
    738			 "Failed to get the maximum desired temperature: %d\n",
    739			 ret);
    740		return ret;
    741	}
    742
    743	return allocate_power(tz, control_temp);
    744}
    745
    746static struct thermal_governor thermal_gov_power_allocator = {
    747	.name		= "power_allocator",
    748	.bind_to_tz	= power_allocator_bind,
    749	.unbind_from_tz	= power_allocator_unbind,
    750	.throttle	= power_allocator_throttle,
    751};
    752THERMAL_GOVERNOR_DECLARE(thermal_gov_power_allocator);