thermal_core.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  thermal.c - Generic Thermal Management Sysfs support.
   4 *
   5 *  Copyright (C) 2008 Intel Corp
   6 *  Copyright (C) 2008 Zhang Rui <rui.zhang@intel.com>
   7 *  Copyright (C) 2008 Sujith Thomas <sujith.thomas@intel.com>
   8 */
   9
  10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12#include <linux/device.h>
  13#include <linux/err.h>
  14#include <linux/export.h>
  15#include <linux/slab.h>
  16#include <linux/kdev_t.h>
  17#include <linux/idr.h>
  18#include <linux/thermal.h>
  19#include <linux/reboot.h>
  20#include <linux/string.h>
  21#include <linux/of.h>
  22#include <linux/suspend.h>
  23
  24#define CREATE_TRACE_POINTS
  25#include "thermal_trace.h"
  26
  27#include "thermal_core.h"
  28#include "thermal_hwmon.h"
  29
  30static DEFINE_IDA(thermal_tz_ida);
  31static DEFINE_IDA(thermal_cdev_ida);
  32
  33static LIST_HEAD(thermal_tz_list);
  34static LIST_HEAD(thermal_cdev_list);
  35static LIST_HEAD(thermal_governor_list);
  36
  37static DEFINE_MUTEX(thermal_list_lock);
  38static DEFINE_MUTEX(thermal_governor_lock);
  39
  40static struct thermal_governor *def_governor;
  41
  42static bool thermal_pm_suspended;
  43
  44static struct workqueue_struct *thermal_wq __ro_after_init;
  45
  46/*
  47 * Governor section: set of functions to handle thermal governors
  48 *
  49 * Functions to help in the life cycle of thermal governors within
  50 * the thermal core and by the thermal governor code.
  51 */
  52
  53static struct thermal_governor *__find_governor(const char *name)
  54{
  55	struct thermal_governor *pos;
  56
  57	if (!name || !name[0])
  58		return def_governor;
  59
  60	list_for_each_entry(pos, &thermal_governor_list, governor_list)
  61		if (!strncasecmp(name, pos->name, THERMAL_NAME_LENGTH))
  62			return pos;
  63
  64	return NULL;
  65}
  66
  67/**
  68 * bind_previous_governor() - bind the previous governor of the thermal zone
  69 * @tz:		a valid pointer to a struct thermal_zone_device
  70 * @failed_gov_name:	the name of the governor that failed to register
  71 *
  72 * Register the previous governor of the thermal zone after a new
  73 * governor has failed to be bound.
  74 */
  75static void bind_previous_governor(struct thermal_zone_device *tz,
  76				   const char *failed_gov_name)
  77{
  78	if (tz->governor && tz->governor->bind_to_tz) {
  79		if (tz->governor->bind_to_tz(tz)) {
  80			dev_err(&tz->device,
  81				"governor %s failed to bind and the previous one (%s) failed to bind again, thermal zone %s has no governor\n",
  82				failed_gov_name, tz->governor->name, tz->type);
  83			tz->governor = NULL;
  84		}
  85	}
  86}
  87
  88/**
  89 * thermal_set_governor() - Switch to another governor
  90 * @tz:		a valid pointer to a struct thermal_zone_device
  91 * @new_gov:	pointer to the new governor
  92 *
  93 * Change the governor of thermal zone @tz.
  94 *
  95 * Return: 0 on success, an error if the new governor's bind_to_tz() failed.
  96 */
  97static int thermal_set_governor(struct thermal_zone_device *tz,
  98				struct thermal_governor *new_gov)
  99{
 100	int ret = 0;
 101
 102	if (tz->governor && tz->governor->unbind_from_tz)
 103		tz->governor->unbind_from_tz(tz);
 104
 105	if (new_gov && new_gov->bind_to_tz) {
 106		ret = new_gov->bind_to_tz(tz);
 107		if (ret) {
 108			bind_previous_governor(tz, new_gov->name);
 109
 110			return ret;
 111		}
 112	}
 113
 114	tz->governor = new_gov;
 115
 116	return ret;
 117}
 118
 119int thermal_register_governor(struct thermal_governor *governor)
 120{
 121	int err;
 122	const char *name;
 123	struct thermal_zone_device *pos;
 124
 125	if (!governor)
 126		return -EINVAL;
 127
 128	guard(mutex)(&thermal_governor_lock);
 129
 130	err = -EBUSY;
 131	if (!__find_governor(governor->name)) {
 132		bool match_default;
 133
 134		err = 0;
 135		list_add(&governor->governor_list, &thermal_governor_list);
 136		match_default = !strncmp(governor->name,
 137					 DEFAULT_THERMAL_GOVERNOR,
 138					 THERMAL_NAME_LENGTH);
 139
 140		if (!def_governor && match_default)
 141			def_governor = governor;
 142	}
 143
 144	guard(mutex)(&thermal_list_lock);
 145
 146	list_for_each_entry(pos, &thermal_tz_list, node) {
 147		/*
 148		 * only thermal zones with specified tz->tzp->governor_name
 149		 * may run with tz->govenor unset
 150		 */
 151		if (pos->governor)
 152			continue;
 153
 154		name = pos->tzp->governor_name;
 155
 156		if (!strncasecmp(name, governor->name, THERMAL_NAME_LENGTH)) {
 157			int ret;
 158
 159			ret = thermal_set_governor(pos, governor);
 160			if (ret)
 161				dev_err(&pos->device,
 162					"Failed to set governor %s for thermal zone %s: %d\n",
 163					governor->name, pos->type, ret);
 164		}
 165	}
 166
 167	return err;
 168}
 169
 170void thermal_unregister_governor(struct thermal_governor *governor)
 171{
 172	struct thermal_zone_device *pos;
 173
 174	if (!governor)
 175		return;
 176
 177	guard(mutex)(&thermal_governor_lock);
 178
 179	if (!__find_governor(governor->name))
 180		return;
 181
 182	list_del(&governor->governor_list);
 183
 184	guard(mutex)(&thermal_list_lock);
 185
 186	list_for_each_entry(pos, &thermal_tz_list, node) {
 187		if (!strncasecmp(pos->governor->name, governor->name,
 188				 THERMAL_NAME_LENGTH))
 189			thermal_set_governor(pos, NULL);
 190	}
 191}
 192
 193int thermal_zone_device_set_policy(struct thermal_zone_device *tz,
 194				   char *policy)
 195{
 196	struct thermal_governor *gov;
 197	int ret = -EINVAL;
 198
 199	guard(mutex)(&thermal_governor_lock);
 200	guard(thermal_zone)(tz);
 201
 202	gov = __find_governor(strim(policy));
 203	if (gov)
 204		ret = thermal_set_governor(tz, gov);
 205
 206	thermal_notify_tz_gov_change(tz, policy);
 207
 208	return ret;
 209}
 210
 211int thermal_build_list_of_policies(char *buf)
 212{
 213	struct thermal_governor *pos;
 214	ssize_t count = 0;
 215
 216	guard(mutex)(&thermal_governor_lock);
 217
 218	list_for_each_entry(pos, &thermal_governor_list, governor_list) {
 219		count += sysfs_emit_at(buf, count, "%s ", pos->name);
 220	}
 221	count += sysfs_emit_at(buf, count, "\n");
 222
 223	return count;
 224}
 225
 226static void __init thermal_unregister_governors(void)
 227{
 228	struct thermal_governor **governor;
 229
 230	for_each_governor_table(governor)
 231		thermal_unregister_governor(*governor);
 232}
 233
 234static int __init thermal_register_governors(void)
 235{
 236	int ret = 0;
 237	struct thermal_governor **governor;
 238
 239	for_each_governor_table(governor) {
 240		ret = thermal_register_governor(*governor);
 241		if (ret) {
 242			pr_err("Failed to register governor: '%s'",
 243			       (*governor)->name);
 244			break;
 245		}
 246
 247		pr_info("Registered thermal governor '%s'",
 248			(*governor)->name);
 249	}
 250
 251	if (ret) {
 252		struct thermal_governor **gov;
 253
 254		for_each_governor_table(gov) {
 255			if (gov == governor)
 256				break;
 257			thermal_unregister_governor(*gov);
 258		}
 259	}
 260
 261	return ret;
 262}
 263
 264static int __thermal_zone_device_set_mode(struct thermal_zone_device *tz,
 265					  enum thermal_device_mode mode)
 266{
 267	if (tz->ops.change_mode) {
 268		int ret;
 269
 270		ret = tz->ops.change_mode(tz, mode);
 271		if (ret)
 272			return ret;
 273	}
 274
 275	tz->mode = mode;
 276
 277	return 0;
 278}
 279
 280static void thermal_zone_broken_disable(struct thermal_zone_device *tz)
 281{
 282	struct thermal_trip_desc *td;
 283
 284	dev_err(&tz->device, "Unable to get temperature, disabling!\n");
 285	/*
 286	 * This function only runs for enabled thermal zones, so no need to
 287	 * check for the current mode.
 288	 */
 289	__thermal_zone_device_set_mode(tz, THERMAL_DEVICE_DISABLED);
 290	thermal_notify_tz_disable(tz);
 291
 292	for_each_trip_desc(tz, td) {
 293		if (td->trip.type == THERMAL_TRIP_CRITICAL &&
 294		    td->trip.temperature > THERMAL_TEMP_INVALID) {
 295			dev_crit(&tz->device,
 296				 "Disabled thermal zone with critical trip point\n");
 297			return;
 298		}
 299	}
 300}
 301
 302/*
 303 * Zone update section: main control loop applied to each zone while monitoring
 304 * in polling mode. The monitoring is done using a workqueue.
 305 * Same update may be done on a zone by calling thermal_zone_device_update().
 306 *
 307 * An update means:
 308 * - Non-critical trips will invoke the governor responsible for that zone;
 309 * - Hot trips will produce a notification to userspace;
 310 * - Critical trip point will cause a system shutdown.
 311 */
 312static void thermal_zone_device_set_polling(struct thermal_zone_device *tz,
 313					    unsigned long delay)
 314{
 315	if (delay > HZ)
 316		delay = round_jiffies_relative(delay);
 317
 318	mod_delayed_work(thermal_wq, &tz->poll_queue, delay);
 319}
 320
 321static void thermal_zone_recheck(struct thermal_zone_device *tz, int error)
 322{
 323	if (error == -EAGAIN) {
 324		thermal_zone_device_set_polling(tz, THERMAL_RECHECK_DELAY);
 325		return;
 326	}
 327
 328	/*
 329	 * Print the message once to reduce log noise.  It will be followed by
 330	 * another one if the temperature cannot be determined after multiple
 331	 * attempts.
 332	 */
 333	if (tz->recheck_delay_jiffies == THERMAL_RECHECK_DELAY)
 334		dev_info(&tz->device, "Temperature check failed (%d)\n", error);
 335
 336	thermal_zone_device_set_polling(tz, tz->recheck_delay_jiffies);
 337
 338	tz->recheck_delay_jiffies += max(tz->recheck_delay_jiffies >> 1, 1ULL);
 339	if (tz->recheck_delay_jiffies > THERMAL_MAX_RECHECK_DELAY) {
 340		thermal_zone_broken_disable(tz);
 341		/*
 342		 * Restore the original recheck delay value to allow the thermal
 343		 * zone to try to recover when it is reenabled by user space.
 344		 */
 345		tz->recheck_delay_jiffies = THERMAL_RECHECK_DELAY;
 346	}
 347}
 348
 349static void monitor_thermal_zone(struct thermal_zone_device *tz)
 350{
 351	if (tz->passive > 0 && tz->passive_delay_jiffies)
 352		thermal_zone_device_set_polling(tz, tz->passive_delay_jiffies);
 353	else if (tz->polling_delay_jiffies)
 354		thermal_zone_device_set_polling(tz, tz->polling_delay_jiffies);
 355}
 356
 357static struct thermal_governor *thermal_get_tz_governor(struct thermal_zone_device *tz)
 358{
 359	if (tz->governor)
 360		return tz->governor;
 361
 362	return def_governor;
 363}
 364
 365void thermal_governor_update_tz(struct thermal_zone_device *tz,
 366				enum thermal_notify_event reason)
 367{
 368	if (!tz->governor || !tz->governor->update_tz)
 369		return;
 370
 371	tz->governor->update_tz(tz, reason);
 372}
 373
 374static void thermal_zone_device_halt(struct thermal_zone_device *tz,
 375				     enum hw_protection_action action)
 376{
 377	/*
 378	 * poweroff_delay_ms must be a carefully profiled positive value.
 379	 * Its a must for forced_emergency_poweroff_work to be scheduled.
 380	 */
 381	int poweroff_delay_ms = CONFIG_THERMAL_EMERGENCY_POWEROFF_DELAY_MS;
 382	const char *msg = "Temperature too high";
 383
 384	dev_emerg(&tz->device, "%s: critical temperature reached\n", tz->type);
 385
 386	__hw_protection_trigger(msg, poweroff_delay_ms, action);
 387}
 388
 389void thermal_zone_device_critical(struct thermal_zone_device *tz)
 390{
 391	thermal_zone_device_halt(tz, HWPROT_ACT_DEFAULT);
 392}
 393EXPORT_SYMBOL(thermal_zone_device_critical);
 394
 395void thermal_zone_device_critical_shutdown(struct thermal_zone_device *tz)
 396{
 397	thermal_zone_device_halt(tz, HWPROT_ACT_SHUTDOWN);
 398}
 399
 400void thermal_zone_device_critical_reboot(struct thermal_zone_device *tz)
 401{
 402	thermal_zone_device_halt(tz, HWPROT_ACT_REBOOT);
 403}
 404
 405static void handle_critical_trips(struct thermal_zone_device *tz,
 406				  const struct thermal_trip *trip)
 407{
 408	trace_thermal_zone_trip(tz, thermal_zone_trip_id(tz, trip), trip->type);
 409
 410	if (trip->type == THERMAL_TRIP_CRITICAL)
 411		tz->ops.critical(tz);
 412	else if (tz->ops.hot)
 413		tz->ops.hot(tz);
 414}
 415
 416static void move_trip_to_sorted_list(struct thermal_trip_desc *td,
 417				     struct list_head *list)
 418{
 419	struct thermal_trip_desc *entry;
 420
 421	/*
 422	 * Delete upfront and then add to make relocation within the same list
 423	 * work.
 424	 */
 425	list_del(&td->list_node);
 426
 427	/* Assume that the new entry is likely to be the last one. */
 428	list_for_each_entry_reverse(entry, list, list_node) {
 429		if (entry->threshold <= td->threshold) {
 430			list_add(&td->list_node, &entry->list_node);
 431			return;
 432		}
 433	}
 434	list_add(&td->list_node, list);
 435}
 436
 437static void move_to_trips_high(struct thermal_zone_device *tz,
 438			       struct thermal_trip_desc *td)
 439{
 440	td->threshold = td->trip.temperature;
 441	move_trip_to_sorted_list(td, &tz->trips_high);
 442}
 443
 444static void move_to_trips_reached(struct thermal_zone_device *tz,
 445				  struct thermal_trip_desc *td)
 446{
 447	td->threshold = td->trip.temperature - td->trip.hysteresis;
 448	move_trip_to_sorted_list(td, &tz->trips_reached);
 449}
 450
 451static void move_to_trips_invalid(struct thermal_zone_device *tz,
 452				  struct thermal_trip_desc *td)
 453{
 454	td->threshold = INT_MAX;
 455	list_move(&td->list_node, &tz->trips_invalid);
 456}
 457
 458static void thermal_governor_trip_crossed(struct thermal_governor *governor,
 459					  struct thermal_zone_device *tz,
 460					  const struct thermal_trip *trip,
 461					  bool upward)
 462{
 463	if (trip->type == THERMAL_TRIP_HOT || trip->type == THERMAL_TRIP_CRITICAL)
 464		return;
 465
 466	if (governor->trip_crossed)
 467		governor->trip_crossed(tz, trip, upward);
 468}
 469
 470static void thermal_trip_crossed(struct thermal_zone_device *tz,
 471				 struct thermal_trip_desc *td,
 472				 struct thermal_governor *governor,
 473				 bool upward)
 474{
 475	const struct thermal_trip *trip = &td->trip;
 476
 477	if (upward) {
 478		if (trip->type == THERMAL_TRIP_PASSIVE)
 479			tz->passive++;
 480		else if (trip->type == THERMAL_TRIP_CRITICAL ||
 481			 trip->type == THERMAL_TRIP_HOT)
 482			handle_critical_trips(tz, trip);
 483
 484		thermal_notify_tz_trip_up(tz, trip);
 485		thermal_debug_tz_trip_up(tz, trip);
 486	} else {
 487		if (trip->type == THERMAL_TRIP_PASSIVE) {
 488			tz->passive--;
 489			WARN_ON(tz->passive < 0);
 490		}
 491		thermal_notify_tz_trip_down(tz, trip);
 492		thermal_debug_tz_trip_down(tz, trip);
 493	}
 494	thermal_governor_trip_crossed(governor, tz, trip, upward);
 495}
 496
 497void thermal_zone_set_trip_hyst(struct thermal_zone_device *tz,
 498				struct thermal_trip *trip, int hyst)
 499{
 500	struct thermal_trip_desc *td = trip_to_trip_desc(trip);
 501
 502	WRITE_ONCE(trip->hysteresis, hyst);
 503	thermal_notify_tz_trip_change(tz, trip);
 504	/*
 505	 * If the zone temperature is above or at the trip temperature, the trip
 506	 * is in the trips_reached list and its threshold is equal to its low
 507	 * temperature.  It needs to stay in that list, but its threshold needs
 508	 * to be updated and the list ordering may need to be restored.
 509	 */
 510	if (tz->temperature >= td->threshold)
 511		move_to_trips_reached(tz, td);
 512}
 513
 514void thermal_zone_set_trip_temp(struct thermal_zone_device *tz,
 515				struct thermal_trip *trip, int temp)
 516{
 517	struct thermal_trip_desc *td = trip_to_trip_desc(trip);
 518	int old_temp = trip->temperature;
 519
 520	if (old_temp == temp)
 521		return;
 522
 523	WRITE_ONCE(trip->temperature, temp);
 524	thermal_notify_tz_trip_change(tz, trip);
 525
 526	if (old_temp == THERMAL_TEMP_INVALID) {
 527		/*
 528		 * The trip was invalid before the change, so move it to the
 529		 * trips_high list regardless of the new temperature value
 530		 * because there is no mitigation under way for it.  If a
 531		 * mitigation needs to be started, the trip will be moved to the
 532		 * trips_reached list later.
 533		 */
 534		move_to_trips_high(tz, td);
 535		return;
 536	}
 537
 538	if (temp == THERMAL_TEMP_INVALID) {
 539		/*
 540		 * If the trip is in the trips_reached list, mitigation is under
 541		 * way for it and it needs to be stopped because the trip is
 542		 * effectively going away.
 543		 */
 544		if (tz->temperature >= td->threshold)
 545			thermal_trip_crossed(tz, td, thermal_get_tz_governor(tz), false);
 546
 547		move_to_trips_invalid(tz, td);
 548		return;
 549	}
 550
 551	/*
 552	 * The trip stays on its current list, but its threshold needs to be
 553	 * updated due to the temperature change and the list ordering may need
 554	 * to be restored.
 555	 */
 556	if (tz->temperature >= td->threshold)
 557		move_to_trips_reached(tz, td);
 558	else
 559		move_to_trips_high(tz, td);
 560}
 561EXPORT_SYMBOL_GPL(thermal_zone_set_trip_temp);
 562
 563static void thermal_zone_handle_trips(struct thermal_zone_device *tz,
 564				      struct thermal_governor *governor,
 565				      int *low, int *high)
 566{
 567	struct thermal_trip_desc *td, *next;
 568	LIST_HEAD(way_down_list);
 569
 570	/* Check the trips that were below or at the zone temperature. */
 571	list_for_each_entry_safe_reverse(td, next, &tz->trips_reached, list_node) {
 572		if (td->threshold <= tz->temperature)
 573			break;
 574
 575		thermal_trip_crossed(tz, td, governor, false);
 576		/*
 577		 * The current trips_high list needs to be processed before
 578		 * adding new entries to it, so put them on a temporary list.
 579		 */
 580		list_move(&td->list_node, &way_down_list);
 581	}
 582	/* Check the trips that were previously above the zone temperature. */
 583	list_for_each_entry_safe(td, next, &tz->trips_high, list_node) {
 584		if (td->threshold > tz->temperature)
 585			break;
 586
 587		thermal_trip_crossed(tz, td, governor, true);
 588		move_to_trips_reached(tz, td);
 589	}
 590	/* Move all of the trips from the temporary list to trips_high. */
 591	list_for_each_entry_safe(td, next, &way_down_list, list_node)
 592		move_to_trips_high(tz, td);
 593
 594	if (!list_empty(&tz->trips_reached)) {
 595		td = list_last_entry(&tz->trips_reached,
 596				     struct thermal_trip_desc, list_node);
 597		/*
 598		 * Set the "low" value below the current trip threshold in case
 599		 * the zone temperature is at that threshold and stays there,
 600		 * which would trigger a new interrupt immediately in vain.
 601		 */
 602		*low = td->threshold - 1;
 603	}
 604	if (!list_empty(&tz->trips_high)) {
 605		td = list_first_entry(&tz->trips_high,
 606				      struct thermal_trip_desc, list_node);
 607		*high = td->threshold;
 608	}
 609}
 610
 611void __thermal_zone_device_update(struct thermal_zone_device *tz,
 612				  enum thermal_notify_event event)
 613{
 614	struct thermal_governor *governor = thermal_get_tz_governor(tz);
 615	int low = -INT_MAX, high = INT_MAX;
 616	int temp, ret;
 617
 618	if (tz->state != TZ_STATE_READY || tz->mode != THERMAL_DEVICE_ENABLED)
 619		return;
 620
 621	ret = __thermal_zone_get_temp(tz, &temp);
 622	if (ret) {
 623		thermal_zone_recheck(tz, ret);
 624		return;
 625	} else if (temp <= THERMAL_TEMP_INVALID) {
 626		/*
 627		 * Special case: No valid temperature value is available, but
 628		 * the zone owner does not want the core to do anything about
 629		 * it.  Continue regular zone polling if needed, so that this
 630		 * function can be called again, but skip everything else.
 631		 */
 632		goto monitor;
 633	}
 634
 635	tz->recheck_delay_jiffies = THERMAL_RECHECK_DELAY;
 636
 637	tz->last_temperature = tz->temperature;
 638	tz->temperature = temp;
 639
 640	trace_thermal_temperature(tz);
 641
 642	thermal_genl_sampling_temp(tz->id, temp);
 643
 644	tz->notify_event = event;
 645
 646	thermal_zone_handle_trips(tz, governor, &low, &high);
 647
 648	thermal_thresholds_handle(tz, &low, &high);
 649
 650	thermal_zone_set_trips(tz, low, high);
 651
 652	if (governor->manage)
 653		governor->manage(tz);
 654
 655	thermal_debug_update_trip_stats(tz);
 656
 657monitor:
 658	monitor_thermal_zone(tz);
 659}
 660
 661static int thermal_zone_device_set_mode(struct thermal_zone_device *tz,
 662					enum thermal_device_mode mode)
 663{
 664	int ret;
 665
 666	guard(thermal_zone)(tz);
 667
 668	/* do nothing if mode isn't changing */
 669	if (mode == tz->mode)
 670		return 0;
 671
 672	ret = __thermal_zone_device_set_mode(tz, mode);
 673	if (ret)
 674		return ret;
 675
 676	__thermal_zone_device_update(tz, THERMAL_EVENT_UNSPECIFIED);
 677
 678	if (mode == THERMAL_DEVICE_ENABLED)
 679		thermal_notify_tz_enable(tz);
 680	else
 681		thermal_notify_tz_disable(tz);
 682
 683	return 0;
 684}
 685
 686int thermal_zone_device_enable(struct thermal_zone_device *tz)
 687{
 688	return thermal_zone_device_set_mode(tz, THERMAL_DEVICE_ENABLED);
 689}
 690EXPORT_SYMBOL_GPL(thermal_zone_device_enable);
 691
 692int thermal_zone_device_disable(struct thermal_zone_device *tz)
 693{
 694	return thermal_zone_device_set_mode(tz, THERMAL_DEVICE_DISABLED);
 695}
 696EXPORT_SYMBOL_GPL(thermal_zone_device_disable);
 697
 698static bool thermal_zone_is_present(struct thermal_zone_device *tz)
 699{
 700	return !list_empty(&tz->node);
 701}
 702
 703void thermal_zone_device_update(struct thermal_zone_device *tz,
 704				enum thermal_notify_event event)
 705{
 706	guard(thermal_zone)(tz);
 707
 708	if (thermal_zone_is_present(tz))
 709		__thermal_zone_device_update(tz, event);
 710}
 711EXPORT_SYMBOL_GPL(thermal_zone_device_update);
 712
 713int for_each_thermal_governor(int (*cb)(struct thermal_governor *, void *),
 714			      void *data)
 715{
 716	struct thermal_governor *gov;
 717
 718	guard(mutex)(&thermal_governor_lock);
 719
 720	list_for_each_entry(gov, &thermal_governor_list, governor_list) {
 721		int ret;
 722
 723		ret = cb(gov, data);
 724		if (ret)
 725			return ret;
 726	}
 727
 728	return 0;
 729}
 730
 731int for_each_thermal_cooling_device(int (*cb)(struct thermal_cooling_device *,
 732					      void *), void *data)
 733{
 734	struct thermal_cooling_device *cdev;
 735
 736	guard(mutex)(&thermal_list_lock);
 737
 738	list_for_each_entry(cdev, &thermal_cdev_list, node) {
 739		int ret;
 740
 741		ret = cb(cdev, data);
 742		if (ret)
 743			return ret;
 744	}
 745
 746	return 0;
 747}
 748
 749int for_each_thermal_zone(int (*cb)(struct thermal_zone_device *, void *),
 750			  void *data)
 751{
 752	struct thermal_zone_device *tz;
 753
 754	guard(mutex)(&thermal_list_lock);
 755
 756	list_for_each_entry(tz, &thermal_tz_list, node) {
 757		int ret;
 758
 759		ret = cb(tz, data);
 760		if (ret)
 761			return ret;
 762	}
 763
 764	return 0;
 765}
 766
 767struct thermal_zone_device *thermal_zone_get_by_id(int id)
 768{
 769	struct thermal_zone_device *tz;
 770
 771	guard(mutex)(&thermal_list_lock);
 772
 773	list_for_each_entry(tz, &thermal_tz_list, node) {
 774		if (tz->id == id) {
 775			get_device(&tz->device);
 776			return tz;
 777		}
 778	}
 779
 780	return NULL;
 781}
 782
 783/*
 784 * Device management section: cooling devices, zones devices, and binding
 785 *
 786 * Set of functions provided by the thermal core for:
 787 * - cooling devices lifecycle: registration, unregistration,
 788 *				binding, and unbinding.
 789 * - thermal zone devices lifecycle: registration, unregistration,
 790 *				     binding, and unbinding.
 791 */
 792
 793static int thermal_instance_add(struct thermal_instance *new_instance,
 794				struct thermal_cooling_device *cdev,
 795				struct thermal_trip_desc *td)
 796{
 797	struct thermal_instance *instance;
 798
 799	list_for_each_entry(instance, &td->thermal_instances, trip_node) {
 800		if (instance->cdev == cdev)
 801			return -EEXIST;
 802	}
 803
 804	list_add_tail(&new_instance->trip_node, &td->thermal_instances);
 805
 806	guard(cooling_dev)(cdev);
 807
 808	list_add_tail(&new_instance->cdev_node, &cdev->thermal_instances);
 809
 810	return 0;
 811}
 812
 813/**
 814 * thermal_bind_cdev_to_trip - bind a cooling device to a thermal zone
 815 * @tz:		pointer to struct thermal_zone_device
 816 * @td:		descriptor of the trip point to bind @cdev to
 817 * @cdev:	pointer to struct thermal_cooling_device
 818 * @cool_spec:	cooling specification for the trip point and @cdev
 819 *
 820 * This interface function bind a thermal cooling device to the certain trip
 821 * point of a thermal zone device.
 822 * This function is usually called in the thermal zone device .bind callback.
 823 *
 824 * Return: 0 on success, the proper error value otherwise.
 825 */
 826static int thermal_bind_cdev_to_trip(struct thermal_zone_device *tz,
 827				     struct thermal_trip_desc *td,
 828				     struct thermal_cooling_device *cdev,
 829				     struct cooling_spec *cool_spec)
 830{
 831	struct thermal_instance *dev;
 832	bool upper_no_limit;
 833	int result;
 834
 835	/* lower default 0, upper default max_state */
 836	if (cool_spec->lower == THERMAL_NO_LIMIT)
 837		cool_spec->lower = 0;
 838
 839	if (cool_spec->upper == THERMAL_NO_LIMIT) {
 840		cool_spec->upper = cdev->max_state;
 841		upper_no_limit = true;
 842	} else {
 843		upper_no_limit = false;
 844	}
 845
 846	if (cool_spec->lower > cool_spec->upper || cool_spec->upper > cdev->max_state)
 847		return -EINVAL;
 848
 849	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 850	if (!dev)
 851		return -ENOMEM;
 852
 853	dev->cdev = cdev;
 854	dev->trip = &td->trip;
 855	dev->upper = cool_spec->upper;
 856	dev->upper_no_limit = upper_no_limit;
 857	dev->lower = cool_spec->lower;
 858	dev->target = THERMAL_NO_TARGET;
 859	dev->weight = cool_spec->weight;
 860
 861	result = ida_alloc(&tz->ida, GFP_KERNEL);
 862	if (result < 0)
 863		goto free_mem;
 864
 865	dev->id = result;
 866	sprintf(dev->name, "cdev%d", dev->id);
 867	result =
 868	    sysfs_create_link(&tz->device.kobj, &cdev->device.kobj, dev->name);
 869	if (result)
 870		goto release_ida;
 871
 872	snprintf(dev->attr_name, sizeof(dev->attr_name), "cdev%d_trip_point",
 873		 dev->id);
 874	sysfs_attr_init(&dev->attr.attr);
 875	dev->attr.attr.name = dev->attr_name;
 876	dev->attr.attr.mode = 0444;
 877	dev->attr.show = trip_point_show;
 878	result = device_create_file(&tz->device, &dev->attr);
 879	if (result)
 880		goto remove_symbol_link;
 881
 882	snprintf(dev->weight_attr_name, sizeof(dev->weight_attr_name),
 883		 "cdev%d_weight", dev->id);
 884	sysfs_attr_init(&dev->weight_attr.attr);
 885	dev->weight_attr.attr.name = dev->weight_attr_name;
 886	dev->weight_attr.attr.mode = S_IWUSR | S_IRUGO;
 887	dev->weight_attr.show = weight_show;
 888	dev->weight_attr.store = weight_store;
 889	result = device_create_file(&tz->device, &dev->weight_attr);
 890	if (result)
 891		goto remove_trip_file;
 892
 893	result = thermal_instance_add(dev, cdev, td);
 894	if (result)
 895		goto remove_weight_file;
 896
 897	thermal_governor_update_tz(tz, THERMAL_TZ_BIND_CDEV);
 898
 899	return 0;
 900
 901remove_weight_file:
 902	device_remove_file(&tz->device, &dev->weight_attr);
 903remove_trip_file:
 904	device_remove_file(&tz->device, &dev->attr);
 905remove_symbol_link:
 906	sysfs_remove_link(&tz->device.kobj, dev->name);
 907release_ida:
 908	ida_free(&tz->ida, dev->id);
 909free_mem:
 910	kfree(dev);
 911	return result;
 912}
 913
 914static void thermal_instance_delete(struct thermal_instance *instance)
 915{
 916	list_del(&instance->trip_node);
 917
 918	guard(cooling_dev)(instance->cdev);
 919
 920	list_del(&instance->cdev_node);
 921}
 922
 923/**
 924 * thermal_unbind_cdev_from_trip - unbind a cooling device from a thermal zone.
 925 * @tz:		pointer to a struct thermal_zone_device.
 926 * @td:		descriptor of the trip point to unbind @cdev from
 927 * @cdev:	pointer to a struct thermal_cooling_device.
 928 *
 929 * This interface function unbind a thermal cooling device from the certain
 930 * trip point of a thermal zone device.
 931 * This function is usually called in the thermal zone device .unbind callback.
 932 */
 933static void thermal_unbind_cdev_from_trip(struct thermal_zone_device *tz,
 934					  struct thermal_trip_desc *td,
 935					  struct thermal_cooling_device *cdev)
 936{
 937	struct thermal_instance *pos, *next;
 938
 939	list_for_each_entry_safe(pos, next, &td->thermal_instances, trip_node) {
 940		if (pos->cdev == cdev) {
 941			thermal_instance_delete(pos);
 942			goto unbind;
 943		}
 944	}
 945
 946	return;
 947
 948unbind:
 949	thermal_governor_update_tz(tz, THERMAL_TZ_UNBIND_CDEV);
 950
 951	device_remove_file(&tz->device, &pos->weight_attr);
 952	device_remove_file(&tz->device, &pos->attr);
 953	sysfs_remove_link(&tz->device.kobj, pos->name);
 954	ida_free(&tz->ida, pos->id);
 955	kfree(pos);
 956}
 957
 958static void thermal_release(struct device *dev)
 959{
 960	struct thermal_zone_device *tz;
 961	struct thermal_cooling_device *cdev;
 962
 963	if (!strncmp(dev_name(dev), "thermal_zone",
 964		     sizeof("thermal_zone") - 1)) {
 965		tz = to_thermal_zone(dev);
 966		thermal_zone_destroy_device_groups(tz);
 967		mutex_destroy(&tz->lock);
 968		complete(&tz->removal);
 969	} else if (!strncmp(dev_name(dev), "cooling_device",
 970			    sizeof("cooling_device") - 1)) {
 971		cdev = to_cooling_device(dev);
 972		thermal_cooling_device_destroy_sysfs(cdev);
 973		kfree_const(cdev->type);
 974		ida_free(&thermal_cdev_ida, cdev->id);
 975		kfree(cdev);
 976	}
 977}
 978
 979static struct class *thermal_class;
 980
 981static inline
 982void print_bind_err_msg(struct thermal_zone_device *tz,
 983			const struct thermal_trip_desc *td,
 984			struct thermal_cooling_device *cdev, int ret)
 985{
 986	dev_err(&tz->device, "binding cdev %s to trip %d failed: %d\n",
 987		cdev->type, thermal_zone_trip_id(tz, &td->trip), ret);
 988}
 989
 990static bool __thermal_zone_cdev_bind(struct thermal_zone_device *tz,
 991				     struct thermal_cooling_device *cdev)
 992{
 993	struct thermal_trip_desc *td;
 994	bool update_tz = false;
 995
 996	if (!tz->ops.should_bind)
 997		return false;
 998
 999	for_each_trip_desc(tz, td) {
1000		struct cooling_spec c = {
1001			.upper = THERMAL_NO_LIMIT,
1002			.lower = THERMAL_NO_LIMIT,
1003			.weight = THERMAL_WEIGHT_DEFAULT
1004		};
1005		int ret;
1006
1007		if (!tz->ops.should_bind(tz, &td->trip, cdev, &c))
1008			continue;
1009
1010		ret = thermal_bind_cdev_to_trip(tz, td, cdev, &c);
1011		if (ret) {
1012			print_bind_err_msg(tz, td, cdev, ret);
1013			continue;
1014		}
1015
1016		update_tz = true;
1017	}
1018
1019	return update_tz;
1020}
1021
1022static void thermal_zone_cdev_bind(struct thermal_zone_device *tz,
1023				   struct thermal_cooling_device *cdev)
1024{
1025	guard(thermal_zone)(tz);
1026
1027	if (__thermal_zone_cdev_bind(tz, cdev))
1028		__thermal_zone_device_update(tz, THERMAL_EVENT_UNSPECIFIED);
1029}
1030
1031static void thermal_cooling_device_init_complete(struct thermal_cooling_device *cdev)
1032{
1033	struct thermal_zone_device *tz;
1034
1035	guard(mutex)(&thermal_list_lock);
1036
1037	list_add(&cdev->node, &thermal_cdev_list);
1038
1039	list_for_each_entry(tz, &thermal_tz_list, node)
1040		thermal_zone_cdev_bind(tz, cdev);
1041}
1042
1043/**
1044 * __thermal_cooling_device_register() - register a new thermal cooling device
1045 * @np:		a pointer to a device tree node.
1046 * @type:	the thermal cooling device type.
1047 * @devdata:	device private data.
1048 * @ops:	standard thermal cooling devices callbacks.
1049 *
1050 * This interface function adds a new thermal cooling device (fan/processor/...)
1051 * to /sys/class/thermal/ folder as cooling_device[0-*]. It tries to bind itself
1052 * to all the thermal zone devices registered at the same time.
1053 * It also gives the opportunity to link the cooling device to a device tree
1054 * node, so that it can be bound to a thermal zone created out of device tree.
1055 *
1056 * Return: a pointer to the created struct thermal_cooling_device or an
1057 * ERR_PTR. Caller must check return value with IS_ERR*() helpers.
1058 */
1059static struct thermal_cooling_device *
1060__thermal_cooling_device_register(struct device_node *np,
1061				  const char *type, void *devdata,
1062				  const struct thermal_cooling_device_ops *ops)
1063{
1064	struct thermal_cooling_device *cdev;
1065	unsigned long current_state;
1066	int id, ret;
1067
1068	if (!ops || !ops->get_max_state || !ops->get_cur_state ||
1069	    !ops->set_cur_state)
1070		return ERR_PTR(-EINVAL);
1071
1072	if (!thermal_class)
1073		return ERR_PTR(-ENODEV);
1074
1075	cdev = kzalloc(sizeof(*cdev), GFP_KERNEL);
1076	if (!cdev)
1077		return ERR_PTR(-ENOMEM);
1078
1079	ret = ida_alloc(&thermal_cdev_ida, GFP_KERNEL);
1080	if (ret < 0)
1081		goto out_kfree_cdev;
1082	cdev->id = ret;
1083	id = ret;
1084
1085	cdev->type = kstrdup_const(type ? type : "", GFP_KERNEL);
1086	if (!cdev->type) {
1087		ret = -ENOMEM;
1088		goto out_ida_remove;
1089	}
1090
1091	mutex_init(&cdev->lock);
1092	INIT_LIST_HEAD(&cdev->thermal_instances);
1093	cdev->np = np;
1094	cdev->ops = ops;
1095	cdev->updated = false;
1096	cdev->device.class = thermal_class;
1097	cdev->devdata = devdata;
1098
1099	ret = cdev->ops->get_max_state(cdev, &cdev->max_state);
1100	if (ret)
1101		goto out_cdev_type;
1102
1103	/*
1104	 * The cooling device's current state is only needed for debug
1105	 * initialization below, so a failure to get it does not cause
1106	 * the entire cooling device initialization to fail.  However,
1107	 * the debug will not work for the device if its initial state
1108	 * cannot be determined and drivers are responsible for ensuring
1109	 * that this will not happen.
1110	 */
1111	ret = cdev->ops->get_cur_state(cdev, &current_state);
1112	if (ret)
1113		current_state = ULONG_MAX;
1114
1115	thermal_cooling_device_setup_sysfs(cdev);
1116
1117	ret = dev_set_name(&cdev->device, "cooling_device%d", cdev->id);
1118	if (ret)
1119		goto out_cooling_dev;
1120
1121	ret = device_register(&cdev->device);
1122	if (ret) {
1123		/* thermal_release() handles rest of the cleanup */
1124		put_device(&cdev->device);
1125		return ERR_PTR(ret);
1126	}
1127
1128	if (current_state <= cdev->max_state)
1129		thermal_debug_cdev_add(cdev, current_state);
1130
1131	thermal_cooling_device_init_complete(cdev);
1132
1133	return cdev;
1134
1135out_cooling_dev:
1136	thermal_cooling_device_destroy_sysfs(cdev);
1137out_cdev_type:
1138	kfree_const(cdev->type);
1139out_ida_remove:
1140	ida_free(&thermal_cdev_ida, id);
1141out_kfree_cdev:
1142	kfree(cdev);
1143	return ERR_PTR(ret);
1144}
1145
1146/**
1147 * thermal_cooling_device_register() - register a new thermal cooling device
1148 * @type:	the thermal cooling device type.
1149 * @devdata:	device private data.
1150 * @ops:		standard thermal cooling devices callbacks.
1151 *
1152 * This interface function adds a new thermal cooling device (fan/processor/...)
1153 * to /sys/class/thermal/ folder as cooling_device[0-*]. It tries to bind itself
1154 * to all the thermal zone devices registered at the same time.
1155 *
1156 * Return: a pointer to the created struct thermal_cooling_device or an
1157 * ERR_PTR. Caller must check return value with IS_ERR*() helpers.
1158 */
1159struct thermal_cooling_device *
1160thermal_cooling_device_register(const char *type, void *devdata,
1161				const struct thermal_cooling_device_ops *ops)
1162{
1163	return __thermal_cooling_device_register(NULL, type, devdata, ops);
1164}
1165EXPORT_SYMBOL_GPL(thermal_cooling_device_register);
1166
1167/**
1168 * thermal_of_cooling_device_register() - register an OF thermal cooling device
1169 * @np:		a pointer to a device tree node.
1170 * @type:	the thermal cooling device type.
1171 * @devdata:	device private data.
1172 * @ops:		standard thermal cooling devices callbacks.
1173 *
1174 * This function will register a cooling device with device tree node reference.
1175 * This interface function adds a new thermal cooling device (fan/processor/...)
1176 * to /sys/class/thermal/ folder as cooling_device[0-*]. It tries to bind itself
1177 * to all the thermal zone devices registered at the same time.
1178 *
1179 * Return: a pointer to the created struct thermal_cooling_device or an
1180 * ERR_PTR. Caller must check return value with IS_ERR*() helpers.
1181 */
1182struct thermal_cooling_device *
1183thermal_of_cooling_device_register(struct device_node *np,
1184				   const char *type, void *devdata,
1185				   const struct thermal_cooling_device_ops *ops)
1186{
1187	return __thermal_cooling_device_register(np, type, devdata, ops);
1188}
1189EXPORT_SYMBOL_GPL(thermal_of_cooling_device_register);
1190
1191static void thermal_cooling_device_release(struct device *dev, void *res)
1192{
1193	thermal_cooling_device_unregister(
1194				*(struct thermal_cooling_device **)res);
1195}
1196
1197/**
1198 * devm_thermal_of_cooling_device_register() - register an OF thermal cooling
1199 *					       device
1200 * @dev:	a valid struct device pointer of a sensor device.
1201 * @np:		a pointer to a device tree node.
1202 * @type:	the thermal cooling device type.
1203 * @devdata:	device private data.
1204 * @ops:	standard thermal cooling devices callbacks.
1205 *
1206 * This function will register a cooling device with device tree node reference.
1207 * This interface function adds a new thermal cooling device (fan/processor/...)
1208 * to /sys/class/thermal/ folder as cooling_device[0-*]. It tries to bind itself
1209 * to all the thermal zone devices registered at the same time.
1210 *
1211 * Return: a pointer to the created struct thermal_cooling_device or an
1212 * ERR_PTR. Caller must check return value with IS_ERR*() helpers.
1213 */
1214struct thermal_cooling_device *
1215devm_thermal_of_cooling_device_register(struct device *dev,
1216				struct device_node *np,
1217				const char *type, void *devdata,
1218				const struct thermal_cooling_device_ops *ops)
1219{
1220	struct thermal_cooling_device **ptr, *tcd;
1221
1222	ptr = devres_alloc(thermal_cooling_device_release, sizeof(*ptr),
1223			   GFP_KERNEL);
1224	if (!ptr)
1225		return ERR_PTR(-ENOMEM);
1226
1227	tcd = __thermal_cooling_device_register(np, type, devdata, ops);
1228	if (IS_ERR(tcd)) {
1229		devres_free(ptr);
1230		return tcd;
1231	}
1232
1233	*ptr = tcd;
1234	devres_add(dev, ptr);
1235
1236	return tcd;
1237}
1238EXPORT_SYMBOL_GPL(devm_thermal_of_cooling_device_register);
1239
1240static bool thermal_cooling_device_present(struct thermal_cooling_device *cdev)
1241{
1242	struct thermal_cooling_device *pos = NULL;
1243
1244	list_for_each_entry(pos, &thermal_cdev_list, node) {
1245		if (pos == cdev)
1246			return true;
1247	}
1248
1249	return false;
1250}
1251
1252/**
1253 * thermal_cooling_device_update - Update a cooling device object
1254 * @cdev: Target cooling device.
1255 *
1256 * Update @cdev to reflect a change of the underlying hardware or platform.
1257 *
1258 * Must be called when the maximum cooling state of @cdev becomes invalid and so
1259 * its .get_max_state() callback needs to be run to produce the new maximum
1260 * cooling state value.
1261 */
1262void thermal_cooling_device_update(struct thermal_cooling_device *cdev)
1263{
1264	struct thermal_instance *ti;
1265	unsigned long state;
1266
1267	if (IS_ERR_OR_NULL(cdev))
1268		return;
1269
1270	/*
1271	 * Hold thermal_list_lock throughout the update to prevent the device
1272	 * from going away while being updated.
1273	 */
1274	guard(mutex)(&thermal_list_lock);
1275
1276	if (!thermal_cooling_device_present(cdev))
1277		return;
1278
1279	/*
1280	 * Update under the cdev lock to prevent the state from being set beyond
1281	 * the new limit concurrently.
1282	 */
1283	guard(cooling_dev)(cdev);
1284
1285	if (cdev->ops->get_max_state(cdev, &cdev->max_state))
1286		return;
1287
1288	thermal_cooling_device_stats_reinit(cdev);
1289
1290	list_for_each_entry(ti, &cdev->thermal_instances, cdev_node) {
1291		if (ti->upper == cdev->max_state)
1292			continue;
1293
1294		if (ti->upper < cdev->max_state) {
1295			if (ti->upper_no_limit)
1296				ti->upper = cdev->max_state;
1297
1298			continue;
1299		}
1300
1301		ti->upper = cdev->max_state;
1302		if (ti->lower > ti->upper)
1303			ti->lower = ti->upper;
1304
1305		if (ti->target == THERMAL_NO_TARGET)
1306			continue;
1307
1308		if (ti->target > ti->upper)
1309			ti->target = ti->upper;
1310	}
1311
1312	if (cdev->ops->get_cur_state(cdev, &state) || state > cdev->max_state)
1313		return;
1314
1315	thermal_cooling_device_stats_update(cdev, state);
1316}
1317EXPORT_SYMBOL_GPL(thermal_cooling_device_update);
1318
1319static void __thermal_zone_cdev_unbind(struct thermal_zone_device *tz,
1320				       struct thermal_cooling_device *cdev)
1321{
1322	struct thermal_trip_desc *td;
1323
1324	for_each_trip_desc(tz, td)
1325		thermal_unbind_cdev_from_trip(tz, td, cdev);
1326}
1327
1328static void thermal_zone_cdev_unbind(struct thermal_zone_device *tz,
1329				     struct thermal_cooling_device *cdev)
1330{
1331	guard(thermal_zone)(tz);
1332
1333	__thermal_zone_cdev_unbind(tz, cdev);
1334}
1335
1336static bool thermal_cooling_device_exit(struct thermal_cooling_device *cdev)
1337{
1338	struct thermal_zone_device *tz;
1339
1340	guard(mutex)(&thermal_list_lock);
1341
1342	if (!thermal_cooling_device_present(cdev))
1343		return false;
1344
1345	list_del(&cdev->node);
1346
1347	list_for_each_entry(tz, &thermal_tz_list, node)
1348		thermal_zone_cdev_unbind(tz, cdev);
1349
1350	return true;
1351}
1352
1353/**
1354 * thermal_cooling_device_unregister() - removes a thermal cooling device
1355 * @cdev: Thermal cooling device to remove.
1356 */
1357void thermal_cooling_device_unregister(struct thermal_cooling_device *cdev)
1358{
1359	if (!cdev)
1360		return;
1361
1362	thermal_debug_cdev_remove(cdev);
1363
1364	if (thermal_cooling_device_exit(cdev))
1365		device_unregister(&cdev->device);
1366}
1367EXPORT_SYMBOL_GPL(thermal_cooling_device_unregister);
1368
1369int thermal_zone_get_crit_temp(struct thermal_zone_device *tz, int *temp)
1370{
1371	const struct thermal_trip_desc *td;
1372	int ret = -EINVAL;
1373
1374	if (tz->ops.get_crit_temp)
1375		return tz->ops.get_crit_temp(tz, temp);
1376
1377	guard(thermal_zone)(tz);
1378
1379	for_each_trip_desc(tz, td) {
1380		const struct thermal_trip *trip = &td->trip;
1381
1382		if (trip->type == THERMAL_TRIP_CRITICAL) {
1383			*temp = trip->temperature;
1384			ret = 0;
1385			break;
1386		}
1387	}
1388
1389	return ret;
1390}
1391EXPORT_SYMBOL_GPL(thermal_zone_get_crit_temp);
1392
1393static void thermal_zone_device_check(struct work_struct *work)
1394{
1395	struct thermal_zone_device *tz = container_of(work, struct
1396						      thermal_zone_device,
1397						      poll_queue.work);
1398	thermal_zone_device_update(tz, THERMAL_EVENT_UNSPECIFIED);
1399}
1400
1401static void thermal_zone_device_init(struct thermal_zone_device *tz)
1402{
1403	struct thermal_trip_desc *td, *next;
1404
1405	INIT_DELAYED_WORK(&tz->poll_queue, thermal_zone_device_check);
1406
1407	tz->temperature = THERMAL_TEMP_INIT;
1408	tz->passive = 0;
1409	tz->prev_low_trip = -INT_MAX;
1410	tz->prev_high_trip = INT_MAX;
1411	for_each_trip_desc(tz, td) {
1412		struct thermal_instance *instance;
1413
1414		list_for_each_entry(instance, &td->thermal_instances, trip_node)
1415			instance->initialized = false;
1416	}
1417	/*
1418	 * At this point, all valid trips need to be moved to trips_high so that
1419	 * mitigation can be started if the zone temperature is above them.
1420	 */
1421	list_for_each_entry_safe(td, next, &tz->trips_invalid, list_node) {
1422		if (td->trip.temperature != THERMAL_TEMP_INVALID)
1423			move_to_trips_high(tz, td);
1424	}
1425	/* The trips_reached list may not be empty during system resume. */
1426	list_for_each_entry_safe(td, next, &tz->trips_reached, list_node) {
1427		if (td->trip.temperature == THERMAL_TEMP_INVALID)
1428			move_to_trips_invalid(tz, td);
1429		else
1430			move_to_trips_high(tz, td);
1431	}
1432}
1433
1434static int thermal_zone_init_governor(struct thermal_zone_device *tz)
1435{
1436	struct thermal_governor *governor;
1437
1438	guard(mutex)(&thermal_governor_lock);
1439
1440	if (tz->tzp)
1441		governor = __find_governor(tz->tzp->governor_name);
1442	else
1443		governor = def_governor;
1444
1445	return thermal_set_governor(tz, governor);
1446}
1447
1448static void thermal_zone_init_complete(struct thermal_zone_device *tz)
1449{
1450	struct thermal_cooling_device *cdev;
1451
1452	guard(mutex)(&thermal_list_lock);
1453
1454	list_add_tail(&tz->node, &thermal_tz_list);
1455
1456	guard(thermal_zone)(tz);
1457
1458	/* Bind cooling devices for this zone. */
1459	list_for_each_entry(cdev, &thermal_cdev_list, node)
1460		__thermal_zone_cdev_bind(tz, cdev);
1461
1462	tz->state &= ~TZ_STATE_FLAG_INIT;
1463	/*
1464	 * If system suspend or resume is in progress at this point, the
1465	 * new thermal zone needs to be marked as suspended because
1466	 * thermal_pm_notify() has run already.
1467	 */
1468	if (thermal_pm_suspended)
1469		tz->state |= TZ_STATE_FLAG_SUSPENDED;
1470
1471	__thermal_zone_device_update(tz, THERMAL_EVENT_UNSPECIFIED);
1472}
1473
1474/**
1475 * thermal_zone_device_register_with_trips() - register a new thermal zone device
1476 * @type:	the thermal zone device type
1477 * @trips:	a pointer to an array of thermal trips
1478 * @num_trips:	the number of trip points the thermal zone support
1479 * @devdata:	private device data
1480 * @ops:	standard thermal zone device callbacks
1481 * @tzp:	thermal zone platform parameters
1482 * @passive_delay: number of milliseconds to wait between polls when
1483 *		   performing passive cooling
1484 * @polling_delay: number of milliseconds to wait between polls when checking
1485 *		   whether trip points have been crossed (0 for interrupt
1486 *		   driven systems)
1487 *
1488 * This interface function adds a new thermal zone device (sensor) to
1489 * /sys/class/thermal folder as thermal_zone[0-*]. It tries to bind all the
1490 * thermal cooling devices registered at the same time.
1491 * thermal_zone_device_unregister() must be called when the device is no
1492 * longer needed. The passive cooling depends on the .get_trend() return value.
1493 *
1494 * Return: a pointer to the created struct thermal_zone_device or an
1495 * in case of error, an ERR_PTR. Caller must check return value with
1496 * IS_ERR*() helpers.
1497 */
1498struct thermal_zone_device *
1499thermal_zone_device_register_with_trips(const char *type,
1500					const struct thermal_trip *trips,
1501					int num_trips, void *devdata,
1502					const struct thermal_zone_device_ops *ops,
1503					const struct thermal_zone_params *tzp,
1504					unsigned int passive_delay,
1505					unsigned int polling_delay)
1506{
1507	const struct thermal_trip *trip = trips;
1508	struct thermal_zone_device *tz;
1509	struct thermal_trip_desc *td;
1510	int id;
1511	int result;
1512
1513	if (!type || strlen(type) == 0) {
1514		pr_err("No thermal zone type defined\n");
1515		return ERR_PTR(-EINVAL);
1516	}
1517
1518	if (strlen(type) >= THERMAL_NAME_LENGTH) {
1519		pr_err("Thermal zone name (%s) too long, should be under %d chars\n",
1520		       type, THERMAL_NAME_LENGTH);
1521		return ERR_PTR(-EINVAL);
1522	}
1523
1524	if (num_trips < 0) {
1525		pr_err("Incorrect number of thermal trips\n");
1526		return ERR_PTR(-EINVAL);
1527	}
1528
1529	if (!ops || !ops->get_temp) {
1530		pr_err("Thermal zone device ops not defined or invalid\n");
1531		return ERR_PTR(-EINVAL);
1532	}
1533
1534	if (num_trips > 0 && !trips)
1535		return ERR_PTR(-EINVAL);
1536
1537	if (polling_delay && passive_delay > polling_delay)
1538		return ERR_PTR(-EINVAL);
1539
1540	if (!thermal_class)
1541		return ERR_PTR(-ENODEV);
1542
1543	tz = kzalloc(struct_size(tz, trips, num_trips), GFP_KERNEL);
1544	if (!tz)
1545		return ERR_PTR(-ENOMEM);
1546
1547	if (tzp) {
1548		tz->tzp = kmemdup(tzp, sizeof(*tzp), GFP_KERNEL);
1549		if (!tz->tzp) {
1550			result = -ENOMEM;
1551			goto free_tz;
1552		}
1553	}
1554
1555	INIT_LIST_HEAD(&tz->node);
1556	INIT_LIST_HEAD(&tz->trips_high);
1557	INIT_LIST_HEAD(&tz->trips_reached);
1558	INIT_LIST_HEAD(&tz->trips_invalid);
1559	ida_init(&tz->ida);
1560	mutex_init(&tz->lock);
1561	init_completion(&tz->removal);
1562	init_completion(&tz->resume);
1563	id = ida_alloc(&thermal_tz_ida, GFP_KERNEL);
1564	if (id < 0) {
1565		result = id;
1566		goto free_tzp;
1567	}
1568
1569	tz->id = id;
1570	strscpy(tz->type, type, sizeof(tz->type));
1571
1572	tz->ops = *ops;
1573	if (!tz->ops.critical)
1574		tz->ops.critical = thermal_zone_device_critical;
1575
1576	tz->device.class = thermal_class;
1577	tz->devdata = devdata;
1578	tz->num_trips = num_trips;
1579	for_each_trip_desc(tz, td) {
1580		td->trip = *trip++;
1581		INIT_LIST_HEAD(&td->thermal_instances);
1582		INIT_LIST_HEAD(&td->list_node);
1583		/*
1584		 * Mark all thresholds as invalid to start with even though
1585		 * this only matters for the trips that start as invalid and
1586		 * become valid later.
1587		 */
1588		move_to_trips_invalid(tz, td);
1589	}
1590
1591	tz->polling_delay_jiffies = msecs_to_jiffies(polling_delay);
1592	tz->passive_delay_jiffies = msecs_to_jiffies(passive_delay);
1593	tz->recheck_delay_jiffies = THERMAL_RECHECK_DELAY;
1594
1595	tz->state = TZ_STATE_FLAG_INIT;
1596
1597	result = dev_set_name(&tz->device, "thermal_zone%d", tz->id);
1598	if (result)
1599		goto remove_id;
1600
1601	thermal_zone_device_init(tz);
1602
1603	result = thermal_zone_init_governor(tz);
1604	if (result)
1605		goto remove_id;
1606
1607	/* sys I/F */
1608	/* Add nodes that are always present via .groups */
1609	result = thermal_zone_create_device_groups(tz);
1610	if (result)
1611		goto remove_id;
1612
1613	result = device_register(&tz->device);
1614	if (result)
1615		goto release_device;
1616
1617	if (!tz->tzp || !tz->tzp->no_hwmon) {
1618		result = thermal_add_hwmon_sysfs(tz);
1619		if (result)
1620			goto unregister;
1621	}
1622
1623	result = thermal_thresholds_init(tz);
1624	if (result)
1625		goto remove_hwmon;
1626
1627	thermal_zone_init_complete(tz);
1628
1629	thermal_notify_tz_create(tz);
1630
1631	thermal_debug_tz_add(tz);
1632
1633	return tz;
1634
1635remove_hwmon:
1636	thermal_remove_hwmon_sysfs(tz);
1637unregister:
1638	device_del(&tz->device);
1639release_device:
1640	put_device(&tz->device);
1641	wait_for_completion(&tz->removal);
1642remove_id:
1643	ida_free(&thermal_tz_ida, id);
1644free_tzp:
1645	kfree(tz->tzp);
1646free_tz:
1647	kfree(tz);
1648	return ERR_PTR(result);
1649}
1650EXPORT_SYMBOL_GPL(thermal_zone_device_register_with_trips);
1651
1652struct thermal_zone_device *thermal_tripless_zone_device_register(
1653					const char *type,
1654					void *devdata,
1655					const struct thermal_zone_device_ops *ops,
1656					const struct thermal_zone_params *tzp)
1657{
1658	return thermal_zone_device_register_with_trips(type, NULL, 0, devdata,
1659						       ops, tzp, 0, 0);
1660}
1661EXPORT_SYMBOL_GPL(thermal_tripless_zone_device_register);
1662
1663void *thermal_zone_device_priv(struct thermal_zone_device *tzd)
1664{
1665	return tzd->devdata;
1666}
1667EXPORT_SYMBOL_GPL(thermal_zone_device_priv);
1668
1669const char *thermal_zone_device_type(struct thermal_zone_device *tzd)
1670{
1671	return tzd->type;
1672}
1673EXPORT_SYMBOL_GPL(thermal_zone_device_type);
1674
1675int thermal_zone_device_id(struct thermal_zone_device *tzd)
1676{
1677	return tzd->id;
1678}
1679EXPORT_SYMBOL_GPL(thermal_zone_device_id);
1680
1681struct device *thermal_zone_device(struct thermal_zone_device *tzd)
1682{
1683	return &tzd->device;
1684}
1685EXPORT_SYMBOL_GPL(thermal_zone_device);
1686
1687static bool thermal_zone_exit(struct thermal_zone_device *tz)
1688{
1689	struct thermal_cooling_device *cdev;
1690
1691	guard(mutex)(&thermal_list_lock);
1692
1693	if (list_empty(&tz->node))
1694		return false;
1695
1696	guard(thermal_zone)(tz);
1697
1698	tz->state |= TZ_STATE_FLAG_EXIT;
1699	list_del_init(&tz->node);
1700
1701	/* Unbind all cdevs associated with this thermal zone. */
1702	list_for_each_entry(cdev, &thermal_cdev_list, node)
1703		__thermal_zone_cdev_unbind(tz, cdev);
1704
1705	return true;
1706}
1707
1708/**
1709 * thermal_zone_device_unregister - removes the registered thermal zone device
1710 * @tz: the thermal zone device to remove
1711 */
1712void thermal_zone_device_unregister(struct thermal_zone_device *tz)
1713{
1714	if (!tz)
1715		return;
1716
1717	thermal_debug_tz_remove(tz);
1718
1719	if (!thermal_zone_exit(tz))
1720		return;
1721
1722	cancel_delayed_work_sync(&tz->poll_queue);
1723
1724	thermal_set_governor(tz, NULL);
1725
1726	thermal_thresholds_exit(tz);
1727	thermal_remove_hwmon_sysfs(tz);
1728	ida_free(&thermal_tz_ida, tz->id);
1729	ida_destroy(&tz->ida);
1730
1731	device_del(&tz->device);
1732	put_device(&tz->device);
1733
1734	thermal_notify_tz_delete(tz);
1735
1736	wait_for_completion(&tz->removal);
1737	kfree(tz->tzp);
1738	kfree(tz);
1739}
1740EXPORT_SYMBOL_GPL(thermal_zone_device_unregister);
1741
1742/**
1743 * thermal_zone_get_zone_by_name() - search for a zone and returns its ref
1744 * @name: thermal zone name to fetch the temperature
1745 *
1746 * When only one zone is found with the passed name, returns a reference to it.
1747 *
1748 * Return: On success returns a reference to an unique thermal zone with
1749 * matching name equals to @name, an ERR_PTR otherwise (-EINVAL for invalid
1750 * paramenters, -ENODEV for not found and -EEXIST for multiple matches).
1751 */
1752struct thermal_zone_device *thermal_zone_get_zone_by_name(const char *name)
1753{
1754	struct thermal_zone_device *pos = NULL, *ref = ERR_PTR(-EINVAL);
1755	unsigned int found = 0;
1756
1757	if (!name)
1758		return ERR_PTR(-EINVAL);
1759
1760	guard(mutex)(&thermal_list_lock);
1761
1762	list_for_each_entry(pos, &thermal_tz_list, node)
1763		if (!strncasecmp(name, pos->type, THERMAL_NAME_LENGTH)) {
1764			found++;
1765			ref = pos;
1766		}
1767
1768	if (!found)
1769		return ERR_PTR(-ENODEV);
1770
1771	/* Success only when one zone is found. */
1772	if (found > 1)
1773		return ERR_PTR(-EEXIST);
1774
1775	return ref;
1776}
1777EXPORT_SYMBOL_GPL(thermal_zone_get_zone_by_name);
1778
1779static void thermal_zone_device_resume(struct work_struct *work)
1780{
1781	struct thermal_zone_device *tz;
1782
1783	tz = container_of(work, struct thermal_zone_device, poll_queue.work);
1784
1785	guard(thermal_zone)(tz);
1786
1787	/* If the thermal zone is going away, there's nothing to do. */
1788	if (tz->state & TZ_STATE_FLAG_EXIT)
1789		return;
1790
1791	tz->state &= ~(TZ_STATE_FLAG_SUSPENDED | TZ_STATE_FLAG_RESUMING);
1792
1793	thermal_debug_tz_resume(tz);
1794	thermal_zone_device_init(tz);
1795	thermal_governor_update_tz(tz, THERMAL_TZ_RESUME);
1796	__thermal_zone_device_update(tz, THERMAL_TZ_RESUME);
1797
1798	complete(&tz->resume);
1799}
1800
1801static void thermal_zone_pm_prepare(struct thermal_zone_device *tz)
1802{
1803	guard(thermal_zone)(tz);
1804
1805	if (tz->state & TZ_STATE_FLAG_RESUMING) {
1806		/*
1807		 * thermal_zone_device_resume() queued up for this zone has not
1808		 * acquired the lock yet, so release it to let the function run
1809		 * and wait util it has done the work.
1810		 */
1811		scoped_guard(thermal_zone_reverse, tz) {
1812			wait_for_completion(&tz->resume);
1813		}
1814	}
1815
1816	tz->state |= TZ_STATE_FLAG_SUSPENDED;
1817
1818	/* Prevent new work from getting to the workqueue subsequently. */
1819	cancel_delayed_work(&tz->poll_queue);
1820}
1821
1822static void thermal_pm_notify_prepare(void)
1823{
1824	struct thermal_zone_device *tz;
1825
1826	guard(mutex)(&thermal_list_lock);
1827
1828	thermal_pm_suspended = true;
1829
1830	list_for_each_entry(tz, &thermal_tz_list, node)
1831		thermal_zone_pm_prepare(tz);
1832}
1833
1834static void thermal_zone_pm_complete(struct thermal_zone_device *tz)
1835{
1836	guard(thermal_zone)(tz);
1837
1838	reinit_completion(&tz->resume);
1839	tz->state |= TZ_STATE_FLAG_RESUMING;
1840
1841	/*
1842	 * Replace the work function with the resume one, which will restore the
1843	 * original work function and schedule the polling work if needed.
1844	 */
1845	INIT_DELAYED_WORK(&tz->poll_queue, thermal_zone_device_resume);
1846	/* Queue up the work without a delay. */
1847	mod_delayed_work(thermal_wq, &tz->poll_queue, 0);
1848}
1849
1850static void thermal_pm_notify_complete(void)
1851{
1852	struct thermal_zone_device *tz;
1853
1854	guard(mutex)(&thermal_list_lock);
1855
1856	thermal_pm_suspended = false;
1857
1858	list_for_each_entry(tz, &thermal_tz_list, node)
1859		thermal_zone_pm_complete(tz);
1860}
1861
1862static int thermal_pm_notify(struct notifier_block *nb,
1863			     unsigned long mode, void *_unused)
1864{
1865	switch (mode) {
1866	case PM_HIBERNATION_PREPARE:
1867	case PM_RESTORE_PREPARE:
1868	case PM_SUSPEND_PREPARE:
1869		thermal_pm_notify_prepare();
1870		/*
1871		 * Allow any leftover thermal work items already on the
1872		 * worqueue to complete so they don't get in the way later.
1873		 */
1874		flush_workqueue(thermal_wq);
1875		break;
1876	case PM_POST_HIBERNATION:
1877	case PM_POST_RESTORE:
1878	case PM_POST_SUSPEND:
1879		thermal_pm_notify_complete();
1880		break;
1881	default:
1882		break;
1883	}
1884	return 0;
1885}
1886
1887static struct notifier_block thermal_pm_nb = {
1888	.notifier_call = thermal_pm_notify,
1889	/*
1890	 * Run at the lowest priority to avoid interference between the thermal
1891	 * zone resume work items spawned by thermal_pm_notify() and the other
1892	 * PM notifiers.
1893	 */
1894	.priority = INT_MIN,
1895};
1896
1897static int __init thermal_init(void)
1898{
1899	int result;
1900
1901	thermal_debug_init();
1902
1903	result = thermal_netlink_init();
1904	if (result)
1905		goto error;
1906
1907	thermal_wq = alloc_workqueue("thermal_events",
1908				      WQ_FREEZABLE | WQ_POWER_EFFICIENT | WQ_PERCPU, 0);
1909	if (!thermal_wq) {
1910		result = -ENOMEM;
1911		goto unregister_netlink;
1912	}
1913
1914	result = thermal_register_governors();
1915	if (result)
1916		goto destroy_workqueue;
1917
1918	thermal_class = kzalloc(sizeof(*thermal_class), GFP_KERNEL);
1919	if (!thermal_class) {
1920		result = -ENOMEM;
1921		goto unregister_governors;
1922	}
1923
1924	thermal_class->name = "thermal";
1925	thermal_class->dev_release = thermal_release;
1926
1927	result = class_register(thermal_class);
1928	if (result) {
1929		kfree(thermal_class);
1930		thermal_class = NULL;
1931		goto unregister_governors;
1932	}
1933
1934	result = register_pm_notifier(&thermal_pm_nb);
1935	if (result)
1936		pr_warn("Thermal: Can not register suspend notifier, return %d\n",
1937			result);
1938
1939	return 0;
1940
1941unregister_governors:
1942	thermal_unregister_governors();
1943destroy_workqueue:
1944	destroy_workqueue(thermal_wq);
1945unregister_netlink:
1946	thermal_netlink_exit();
1947error:
1948	mutex_destroy(&thermal_list_lock);
1949	mutex_destroy(&thermal_governor_lock);
1950	return result;
1951}
1952postcore_initcall(thermal_init);