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 */ 34 static 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 */ 47 static 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 */ 69 struct 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 */ 89 static 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 */ 124 static 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 */ 175 static 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 */ 217 static 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, tz->passive_delay); 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 */ 291 static int 292 power_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 cdev->updated = false; 305 mutex_unlock(&cdev->lock); 306 thermal_cdev_update(cdev); 307 308 return 0; 309 } 310 311 /** 312 * divvy_up_power() - divvy the allocated power between the actors 313 * @req_power: each actor's requested power 314 * @max_power: each actor's maximum available power 315 * @num_actors: size of the @req_power, @max_power and @granted_power's array 316 * @total_req_power: sum of @req_power 317 * @power_range: total allocated power 318 * @granted_power: output array: each actor's granted power 319 * @extra_actor_power: an appropriately sized array to be used in the 320 * function as temporary storage of the extra power given 321 * to the actors 322 * 323 * This function divides the total allocated power (@power_range) 324 * fairly between the actors. It first tries to give each actor a 325 * share of the @power_range according to how much power it requested 326 * compared to the rest of the actors. For example, if only one actor 327 * requests power, then it receives all the @power_range. If 328 * three actors each requests 1mW, each receives a third of the 329 * @power_range. 330 * 331 * If any actor received more than their maximum power, then that 332 * surplus is re-divvied among the actors based on how far they are 333 * from their respective maximums. 334 * 335 * Granted power for each actor is written to @granted_power, which 336 * should've been allocated by the calling function. 337 */ 338 static void divvy_up_power(u32 *req_power, u32 *max_power, int num_actors, 339 u32 total_req_power, u32 power_range, 340 u32 *granted_power, u32 *extra_actor_power) 341 { 342 u32 extra_power, capped_extra_power; 343 int i; 344 345 /* 346 * Prevent division by 0 if none of the actors request power. 347 */ 348 if (!total_req_power) 349 total_req_power = 1; 350 351 capped_extra_power = 0; 352 extra_power = 0; 353 for (i = 0; i < num_actors; i++) { 354 u64 req_range = (u64)req_power[i] * power_range; 355 356 granted_power[i] = DIV_ROUND_CLOSEST_ULL(req_range, 357 total_req_power); 358 359 if (granted_power[i] > max_power[i]) { 360 extra_power += granted_power[i] - max_power[i]; 361 granted_power[i] = max_power[i]; 362 } 363 364 extra_actor_power[i] = max_power[i] - granted_power[i]; 365 capped_extra_power += extra_actor_power[i]; 366 } 367 368 if (!extra_power) 369 return; 370 371 /* 372 * Re-divvy the reclaimed extra among actors based on 373 * how far they are from the max 374 */ 375 extra_power = min(extra_power, capped_extra_power); 376 if (capped_extra_power > 0) 377 for (i = 0; i < num_actors; i++) 378 granted_power[i] += (extra_actor_power[i] * 379 extra_power) / capped_extra_power; 380 } 381 382 static int allocate_power(struct thermal_zone_device *tz, 383 int control_temp) 384 { 385 struct thermal_instance *instance; 386 struct power_allocator_params *params = tz->governor_data; 387 u32 *req_power, *max_power, *granted_power, *extra_actor_power; 388 u32 *weighted_req_power; 389 u32 total_req_power, max_allocatable_power, total_weighted_req_power; 390 u32 total_granted_power, power_range; 391 int i, num_actors, total_weight, ret = 0; 392 int trip_max_desired_temperature = params->trip_max_desired_temperature; 393 394 mutex_lock(&tz->lock); 395 396 num_actors = 0; 397 total_weight = 0; 398 list_for_each_entry(instance, &tz->thermal_instances, tz_node) { 399 if ((instance->trip == trip_max_desired_temperature) && 400 cdev_is_power_actor(instance->cdev)) { 401 num_actors++; 402 total_weight += instance->weight; 403 } 404 } 405 406 if (!num_actors) { 407 ret = -ENODEV; 408 goto unlock; 409 } 410 411 /* 412 * We need to allocate five arrays of the same size: 413 * req_power, max_power, granted_power, extra_actor_power and 414 * weighted_req_power. They are going to be needed until this 415 * function returns. Allocate them all in one go to simplify 416 * the allocation and deallocation logic. 417 */ 418 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*max_power)); 419 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*granted_power)); 420 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*extra_actor_power)); 421 BUILD_BUG_ON(sizeof(*req_power) != sizeof(*weighted_req_power)); 422 req_power = kcalloc(num_actors * 5, sizeof(*req_power), GFP_KERNEL); 423 if (!req_power) { 424 ret = -ENOMEM; 425 goto unlock; 426 } 427 428 max_power = &req_power[num_actors]; 429 granted_power = &req_power[2 * num_actors]; 430 extra_actor_power = &req_power[3 * num_actors]; 431 weighted_req_power = &req_power[4 * num_actors]; 432 433 i = 0; 434 total_weighted_req_power = 0; 435 total_req_power = 0; 436 max_allocatable_power = 0; 437 438 list_for_each_entry(instance, &tz->thermal_instances, tz_node) { 439 int weight; 440 struct thermal_cooling_device *cdev = instance->cdev; 441 442 if (instance->trip != trip_max_desired_temperature) 443 continue; 444 445 if (!cdev_is_power_actor(cdev)) 446 continue; 447 448 if (cdev->ops->get_requested_power(cdev, &req_power[i])) 449 continue; 450 451 if (!total_weight) 452 weight = 1 << FRAC_BITS; 453 else 454 weight = instance->weight; 455 456 weighted_req_power[i] = frac_to_int(weight * req_power[i]); 457 458 if (cdev->ops->state2power(cdev, instance->lower, 459 &max_power[i])) 460 continue; 461 462 total_req_power += req_power[i]; 463 max_allocatable_power += max_power[i]; 464 total_weighted_req_power += weighted_req_power[i]; 465 466 i++; 467 } 468 469 power_range = pid_controller(tz, control_temp, max_allocatable_power); 470 471 divvy_up_power(weighted_req_power, max_power, num_actors, 472 total_weighted_req_power, power_range, granted_power, 473 extra_actor_power); 474 475 total_granted_power = 0; 476 i = 0; 477 list_for_each_entry(instance, &tz->thermal_instances, tz_node) { 478 if (instance->trip != trip_max_desired_temperature) 479 continue; 480 481 if (!cdev_is_power_actor(instance->cdev)) 482 continue; 483 484 power_actor_set_power(instance->cdev, instance, 485 granted_power[i]); 486 total_granted_power += granted_power[i]; 487 488 i++; 489 } 490 491 trace_thermal_power_allocator(tz, req_power, total_req_power, 492 granted_power, total_granted_power, 493 num_actors, power_range, 494 max_allocatable_power, tz->temperature, 495 control_temp - tz->temperature); 496 497 kfree(req_power); 498 unlock: 499 mutex_unlock(&tz->lock); 500 501 return ret; 502 } 503 504 /** 505 * get_governor_trips() - get the number of the two trip points that are key for this governor 506 * @tz: thermal zone to operate on 507 * @params: pointer to private data for this governor 508 * 509 * The power allocator governor works optimally with two trips points: 510 * a "switch on" trip point and a "maximum desired temperature". These 511 * are defined as the first and last passive trip points. 512 * 513 * If there is only one trip point, then that's considered to be the 514 * "maximum desired temperature" trip point and the governor is always 515 * on. If there are no passive or active trip points, then the 516 * governor won't do anything. In fact, its throttle function 517 * won't be called at all. 518 */ 519 static void get_governor_trips(struct thermal_zone_device *tz, 520 struct power_allocator_params *params) 521 { 522 int i, last_active, last_passive; 523 bool found_first_passive; 524 525 found_first_passive = false; 526 last_active = INVALID_TRIP; 527 last_passive = INVALID_TRIP; 528 529 for (i = 0; i < tz->trips; i++) { 530 enum thermal_trip_type type; 531 int ret; 532 533 ret = tz->ops->get_trip_type(tz, i, &type); 534 if (ret) { 535 dev_warn(&tz->device, 536 "Failed to get trip point %d type: %d\n", i, 537 ret); 538 continue; 539 } 540 541 if (type == THERMAL_TRIP_PASSIVE) { 542 if (!found_first_passive) { 543 params->trip_switch_on = i; 544 found_first_passive = true; 545 } else { 546 last_passive = i; 547 } 548 } else if (type == THERMAL_TRIP_ACTIVE) { 549 last_active = i; 550 } else { 551 break; 552 } 553 } 554 555 if (last_passive != INVALID_TRIP) { 556 params->trip_max_desired_temperature = last_passive; 557 } else if (found_first_passive) { 558 params->trip_max_desired_temperature = params->trip_switch_on; 559 params->trip_switch_on = INVALID_TRIP; 560 } else { 561 params->trip_switch_on = INVALID_TRIP; 562 params->trip_max_desired_temperature = last_active; 563 } 564 } 565 566 static void reset_pid_controller(struct power_allocator_params *params) 567 { 568 params->err_integral = 0; 569 params->prev_err = 0; 570 } 571 572 static void allow_maximum_power(struct thermal_zone_device *tz) 573 { 574 struct thermal_instance *instance; 575 struct power_allocator_params *params = tz->governor_data; 576 577 mutex_lock(&tz->lock); 578 list_for_each_entry(instance, &tz->thermal_instances, tz_node) { 579 if ((instance->trip != params->trip_max_desired_temperature) || 580 (!cdev_is_power_actor(instance->cdev))) 581 continue; 582 583 instance->target = 0; 584 mutex_lock(&instance->cdev->lock); 585 instance->cdev->updated = false; 586 mutex_unlock(&instance->cdev->lock); 587 thermal_cdev_update(instance->cdev); 588 } 589 mutex_unlock(&tz->lock); 590 } 591 592 /** 593 * power_allocator_bind() - bind the power_allocator governor to a thermal zone 594 * @tz: thermal zone to bind it to 595 * 596 * Initialize the PID controller parameters and bind it to the thermal 597 * zone. 598 * 599 * Return: 0 on success, or -ENOMEM if we ran out of memory. 600 */ 601 static int power_allocator_bind(struct thermal_zone_device *tz) 602 { 603 int ret; 604 struct power_allocator_params *params; 605 int control_temp; 606 607 params = kzalloc(sizeof(*params), GFP_KERNEL); 608 if (!params) 609 return -ENOMEM; 610 611 if (!tz->tzp) { 612 tz->tzp = kzalloc(sizeof(*tz->tzp), GFP_KERNEL); 613 if (!tz->tzp) { 614 ret = -ENOMEM; 615 goto free_params; 616 } 617 618 params->allocated_tzp = true; 619 } 620 621 if (!tz->tzp->sustainable_power) 622 dev_warn(&tz->device, "power_allocator: sustainable_power will be estimated\n"); 623 624 get_governor_trips(tz, params); 625 626 if (tz->trips > 0) { 627 ret = tz->ops->get_trip_temp(tz, 628 params->trip_max_desired_temperature, 629 &control_temp); 630 if (!ret) 631 estimate_pid_constants(tz, tz->tzp->sustainable_power, 632 params->trip_switch_on, 633 control_temp); 634 } 635 636 reset_pid_controller(params); 637 638 tz->governor_data = params; 639 640 return 0; 641 642 free_params: 643 kfree(params); 644 645 return ret; 646 } 647 648 static void power_allocator_unbind(struct thermal_zone_device *tz) 649 { 650 struct power_allocator_params *params = tz->governor_data; 651 652 dev_dbg(&tz->device, "Unbinding from thermal zone %d\n", tz->id); 653 654 if (params->allocated_tzp) { 655 kfree(tz->tzp); 656 tz->tzp = NULL; 657 } 658 659 kfree(tz->governor_data); 660 tz->governor_data = NULL; 661 } 662 663 static int power_allocator_throttle(struct thermal_zone_device *tz, int trip) 664 { 665 int ret; 666 int switch_on_temp, control_temp; 667 struct power_allocator_params *params = tz->governor_data; 668 669 /* 670 * We get called for every trip point but we only need to do 671 * our calculations once 672 */ 673 if (trip != params->trip_max_desired_temperature) 674 return 0; 675 676 ret = tz->ops->get_trip_temp(tz, params->trip_switch_on, 677 &switch_on_temp); 678 if (!ret && (tz->temperature < switch_on_temp)) { 679 tz->passive = 0; 680 reset_pid_controller(params); 681 allow_maximum_power(tz); 682 return 0; 683 } 684 685 tz->passive = 1; 686 687 ret = tz->ops->get_trip_temp(tz, params->trip_max_desired_temperature, 688 &control_temp); 689 if (ret) { 690 dev_warn(&tz->device, 691 "Failed to get the maximum desired temperature: %d\n", 692 ret); 693 return ret; 694 } 695 696 return allocate_power(tz, control_temp); 697 } 698 699 static struct thermal_governor thermal_gov_power_allocator = { 700 .name = "power_allocator", 701 .bind_to_tz = power_allocator_bind, 702 .unbind_from_tz = power_allocator_unbind, 703 .throttle = power_allocator_throttle, 704 }; 705 THERMAL_GOVERNOR_DECLARE(thermal_gov_power_allocator); 706