cpuidle.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664
  1. /*
  2. * cpuidle.c - core cpuidle infrastructure
  3. *
  4. * (C) 2006-2007 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
  5. * Shaohua Li <shaohua.li@intel.com>
  6. * Adam Belay <abelay@novell.com>
  7. *
  8. * This code is licenced under the GPL.
  9. */
  10. #include <linux/clockchips.h>
  11. #include <linux/kernel.h>
  12. #include <linux/mutex.h>
  13. #include <linux/sched.h>
  14. #include <linux/notifier.h>
  15. #include <linux/pm_qos.h>
  16. #include <linux/cpu.h>
  17. #include <linux/cpuidle.h>
  18. #include <linux/ktime.h>
  19. #include <linux/hrtimer.h>
  20. #include <linux/module.h>
  21. #include <linux/suspend.h>
  22. #include <linux/tick.h>
  23. #include <trace/events/power.h>
  24. #include "cpuidle.h"
  25. DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices);
  26. DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev);
  27. DEFINE_MUTEX(cpuidle_lock);
  28. LIST_HEAD(cpuidle_detected_devices);
  29. static int enabled_devices;
  30. static int off __read_mostly;
  31. static int initialized __read_mostly;
  32. int cpuidle_disabled(void)
  33. {
  34. return off;
  35. }
  36. void disable_cpuidle(void)
  37. {
  38. off = 1;
  39. }
  40. bool cpuidle_not_available(struct cpuidle_driver *drv,
  41. struct cpuidle_device *dev)
  42. {
  43. return off || !initialized || !drv || !dev || !dev->enabled;
  44. }
  45. /**
  46. * cpuidle_play_dead - cpu off-lining
  47. *
  48. * Returns in case of an error or no driver
  49. */
  50. int cpuidle_play_dead(void)
  51. {
  52. struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
  53. struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  54. int i;
  55. if (!drv)
  56. return -ENODEV;
  57. /* Find lowest-power state that supports long-term idle */
  58. for (i = drv->state_count - 1; i >= 0; i--)
  59. if (drv->states[i].enter_dead)
  60. return drv->states[i].enter_dead(dev, i);
  61. return -ENODEV;
  62. }
  63. static int find_deepest_state(struct cpuidle_driver *drv,
  64. struct cpuidle_device *dev,
  65. unsigned int max_latency,
  66. unsigned int forbidden_flags,
  67. bool freeze)
  68. {
  69. unsigned int latency_req = 0;
  70. int i, ret = 0;
  71. for (i = 1; i < drv->state_count; i++) {
  72. struct cpuidle_state *s = &drv->states[i];
  73. struct cpuidle_state_usage *su = &dev->states_usage[i];
  74. if (s->disabled || su->disable || s->exit_latency <= latency_req
  75. || s->exit_latency > max_latency
  76. || (s->flags & forbidden_flags)
  77. || (freeze && !s->enter_freeze))
  78. continue;
  79. latency_req = s->exit_latency;
  80. ret = i;
  81. }
  82. return ret;
  83. }
  84. #ifdef CONFIG_SUSPEND
  85. /**
  86. * cpuidle_find_deepest_state - Find the deepest available idle state.
  87. * @drv: cpuidle driver for the given CPU.
  88. * @dev: cpuidle device for the given CPU.
  89. */
  90. int cpuidle_find_deepest_state(struct cpuidle_driver *drv,
  91. struct cpuidle_device *dev)
  92. {
  93. return find_deepest_state(drv, dev, UINT_MAX, 0, false);
  94. }
  95. static void enter_freeze_proper(struct cpuidle_driver *drv,
  96. struct cpuidle_device *dev, int index)
  97. {
  98. /*
  99. * trace_suspend_resume() called by tick_freeze() for the last CPU
  100. * executing it contains RCU usage regarded as invalid in the idle
  101. * context, so tell RCU about that.
  102. */
  103. RCU_NONIDLE(tick_freeze());
  104. /*
  105. * The state used here cannot be a "coupled" one, because the "coupled"
  106. * cpuidle mechanism enables interrupts and doing that with timekeeping
  107. * suspended is generally unsafe.
  108. */
  109. stop_critical_timings();
  110. drv->states[index].enter_freeze(dev, drv, index);
  111. WARN_ON(!irqs_disabled());
  112. /*
  113. * timekeeping_resume() that will be called by tick_unfreeze() for the
  114. * first CPU executing it calls functions containing RCU read-side
  115. * critical sections, so tell RCU about that.
  116. */
  117. RCU_NONIDLE(tick_unfreeze());
  118. start_critical_timings();
  119. }
  120. /**
  121. * cpuidle_enter_freeze - Enter an idle state suitable for suspend-to-idle.
  122. * @drv: cpuidle driver for the given CPU.
  123. * @dev: cpuidle device for the given CPU.
  124. *
  125. * If there are states with the ->enter_freeze callback, find the deepest of
  126. * them and enter it with frozen tick.
  127. */
  128. int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
  129. {
  130. int index;
  131. /*
  132. * Find the deepest state with ->enter_freeze present, which guarantees
  133. * that interrupts won't be enabled when it exits and allows the tick to
  134. * be frozen safely.
  135. */
  136. index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
  137. if (index > 0)
  138. enter_freeze_proper(drv, dev, index);
  139. return index;
  140. }
  141. #endif /* CONFIG_SUSPEND */
  142. /**
  143. * cpuidle_enter_state - enter the state and update stats
  144. * @dev: cpuidle device for this cpu
  145. * @drv: cpuidle driver for this cpu
  146. * @index: index into the states table in @drv of the state to enter
  147. */
  148. int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv,
  149. int index)
  150. {
  151. int entered_state;
  152. struct cpuidle_state *target_state = &drv->states[index];
  153. bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP);
  154. ktime_t time_start, time_end;
  155. s64 diff;
  156. /*
  157. * Tell the time framework to switch to a broadcast timer because our
  158. * local timer will be shut down. If a local timer is used from another
  159. * CPU as a broadcast timer, this call may fail if it is not available.
  160. */
  161. if (broadcast && tick_broadcast_enter()) {
  162. index = find_deepest_state(drv, dev, target_state->exit_latency,
  163. CPUIDLE_FLAG_TIMER_STOP, false);
  164. if (index < 0) {
  165. default_idle_call();
  166. return -EBUSY;
  167. }
  168. target_state = &drv->states[index];
  169. }
  170. /* Take note of the planned idle state. */
  171. sched_idle_set_state(target_state);
  172. trace_cpu_idle_rcuidle(index, dev->cpu);
  173. time_start = ns_to_ktime(local_clock());
  174. stop_critical_timings();
  175. entered_state = target_state->enter(dev, drv, index);
  176. start_critical_timings();
  177. time_end = ns_to_ktime(local_clock());
  178. trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
  179. /* The cpu is no longer idle or about to enter idle. */
  180. sched_idle_set_state(NULL);
  181. if (broadcast) {
  182. if (WARN_ON_ONCE(!irqs_disabled()))
  183. local_irq_disable();
  184. tick_broadcast_exit();
  185. }
  186. if (!cpuidle_state_is_coupled(drv, index))
  187. local_irq_enable();
  188. diff = ktime_us_delta(time_end, time_start);
  189. if (diff > INT_MAX)
  190. diff = INT_MAX;
  191. dev->last_residency = (int) diff;
  192. if (entered_state >= 0) {
  193. /* Update cpuidle counters */
  194. /* This can be moved to within driver enter routine
  195. * but that results in multiple copies of same code.
  196. */
  197. dev->states_usage[entered_state].time += dev->last_residency;
  198. dev->states_usage[entered_state].usage++;
  199. } else {
  200. dev->last_residency = 0;
  201. }
  202. return entered_state;
  203. }
  204. /**
  205. * cpuidle_select - ask the cpuidle framework to choose an idle state
  206. *
  207. * @drv: the cpuidle driver
  208. * @dev: the cpuidle device
  209. *
  210. * Returns the index of the idle state. The return value must not be negative.
  211. */
  212. int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
  213. {
  214. return cpuidle_curr_governor->select(drv, dev);
  215. }
  216. /**
  217. * cpuidle_enter - enter into the specified idle state
  218. *
  219. * @drv: the cpuidle driver tied with the cpu
  220. * @dev: the cpuidle device
  221. * @index: the index in the idle state table
  222. *
  223. * Returns the index in the idle state, < 0 in case of error.
  224. * The error code depends on the backend driver
  225. */
  226. int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev,
  227. int index)
  228. {
  229. if (cpuidle_state_is_coupled(drv, index))
  230. return cpuidle_enter_state_coupled(dev, drv, index);
  231. return cpuidle_enter_state(dev, drv, index);
  232. }
  233. /**
  234. * cpuidle_reflect - tell the underlying governor what was the state
  235. * we were in
  236. *
  237. * @dev : the cpuidle device
  238. * @index: the index in the idle state table
  239. *
  240. */
  241. void cpuidle_reflect(struct cpuidle_device *dev, int index)
  242. {
  243. if (cpuidle_curr_governor->reflect && index >= 0)
  244. cpuidle_curr_governor->reflect(dev, index);
  245. }
  246. /**
  247. * cpuidle_install_idle_handler - installs the cpuidle idle loop handler
  248. */
  249. void cpuidle_install_idle_handler(void)
  250. {
  251. if (enabled_devices) {
  252. /* Make sure all changes finished before we switch to new idle */
  253. smp_wmb();
  254. initialized = 1;
  255. }
  256. }
  257. /**
  258. * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler
  259. */
  260. void cpuidle_uninstall_idle_handler(void)
  261. {
  262. if (enabled_devices) {
  263. initialized = 0;
  264. wake_up_all_idle_cpus();
  265. }
  266. /*
  267. * Make sure external observers (such as the scheduler)
  268. * are done looking at pointed idle states.
  269. */
  270. synchronize_rcu();
  271. }
  272. /**
  273. * cpuidle_pause_and_lock - temporarily disables CPUIDLE
  274. */
  275. void cpuidle_pause_and_lock(void)
  276. {
  277. mutex_lock(&cpuidle_lock);
  278. cpuidle_uninstall_idle_handler();
  279. }
  280. EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock);
  281. /**
  282. * cpuidle_resume_and_unlock - resumes CPUIDLE operation
  283. */
  284. void cpuidle_resume_and_unlock(void)
  285. {
  286. cpuidle_install_idle_handler();
  287. mutex_unlock(&cpuidle_lock);
  288. }
  289. EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock);
  290. /* Currently used in suspend/resume path to suspend cpuidle */
  291. void cpuidle_pause(void)
  292. {
  293. mutex_lock(&cpuidle_lock);
  294. cpuidle_uninstall_idle_handler();
  295. mutex_unlock(&cpuidle_lock);
  296. }
  297. /* Currently used in suspend/resume path to resume cpuidle */
  298. void cpuidle_resume(void)
  299. {
  300. mutex_lock(&cpuidle_lock);
  301. cpuidle_install_idle_handler();
  302. mutex_unlock(&cpuidle_lock);
  303. }
  304. /**
  305. * cpuidle_enable_device - enables idle PM for a CPU
  306. * @dev: the CPU
  307. *
  308. * This function must be called between cpuidle_pause_and_lock and
  309. * cpuidle_resume_and_unlock when used externally.
  310. */
  311. int cpuidle_enable_device(struct cpuidle_device *dev)
  312. {
  313. int ret;
  314. struct cpuidle_driver *drv;
  315. if (!dev)
  316. return -EINVAL;
  317. if (dev->enabled)
  318. return 0;
  319. drv = cpuidle_get_cpu_driver(dev);
  320. if (!drv || !cpuidle_curr_governor)
  321. return -EIO;
  322. if (!dev->registered)
  323. return -EINVAL;
  324. ret = cpuidle_add_device_sysfs(dev);
  325. if (ret)
  326. return ret;
  327. if (cpuidle_curr_governor->enable &&
  328. (ret = cpuidle_curr_governor->enable(drv, dev)))
  329. goto fail_sysfs;
  330. smp_wmb();
  331. dev->enabled = 1;
  332. enabled_devices++;
  333. return 0;
  334. fail_sysfs:
  335. cpuidle_remove_device_sysfs(dev);
  336. return ret;
  337. }
  338. EXPORT_SYMBOL_GPL(cpuidle_enable_device);
  339. /**
  340. * cpuidle_disable_device - disables idle PM for a CPU
  341. * @dev: the CPU
  342. *
  343. * This function must be called between cpuidle_pause_and_lock and
  344. * cpuidle_resume_and_unlock when used externally.
  345. */
  346. void cpuidle_disable_device(struct cpuidle_device *dev)
  347. {
  348. struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  349. if (!dev || !dev->enabled)
  350. return;
  351. if (!drv || !cpuidle_curr_governor)
  352. return;
  353. dev->enabled = 0;
  354. if (cpuidle_curr_governor->disable)
  355. cpuidle_curr_governor->disable(drv, dev);
  356. cpuidle_remove_device_sysfs(dev);
  357. enabled_devices--;
  358. }
  359. EXPORT_SYMBOL_GPL(cpuidle_disable_device);
  360. static void __cpuidle_unregister_device(struct cpuidle_device *dev)
  361. {
  362. struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  363. list_del(&dev->device_list);
  364. per_cpu(cpuidle_devices, dev->cpu) = NULL;
  365. module_put(drv->owner);
  366. dev->registered = 0;
  367. }
  368. static void __cpuidle_device_init(struct cpuidle_device *dev)
  369. {
  370. memset(dev->states_usage, 0, sizeof(dev->states_usage));
  371. dev->last_residency = 0;
  372. }
  373. /**
  374. * __cpuidle_register_device - internal register function called before register
  375. * and enable routines
  376. * @dev: the cpu
  377. *
  378. * cpuidle_lock mutex must be held before this is called
  379. */
  380. static int __cpuidle_register_device(struct cpuidle_device *dev)
  381. {
  382. int ret;
  383. struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  384. if (!try_module_get(drv->owner))
  385. return -EINVAL;
  386. per_cpu(cpuidle_devices, dev->cpu) = dev;
  387. list_add(&dev->device_list, &cpuidle_detected_devices);
  388. ret = cpuidle_coupled_register_device(dev);
  389. if (ret)
  390. __cpuidle_unregister_device(dev);
  391. else
  392. dev->registered = 1;
  393. return ret;
  394. }
  395. /**
  396. * cpuidle_register_device - registers a CPU's idle PM feature
  397. * @dev: the cpu
  398. */
  399. int cpuidle_register_device(struct cpuidle_device *dev)
  400. {
  401. int ret = -EBUSY;
  402. if (!dev)
  403. return -EINVAL;
  404. mutex_lock(&cpuidle_lock);
  405. if (dev->registered)
  406. goto out_unlock;
  407. __cpuidle_device_init(dev);
  408. ret = __cpuidle_register_device(dev);
  409. if (ret)
  410. goto out_unlock;
  411. ret = cpuidle_add_sysfs(dev);
  412. if (ret)
  413. goto out_unregister;
  414. ret = cpuidle_enable_device(dev);
  415. if (ret)
  416. goto out_sysfs;
  417. cpuidle_install_idle_handler();
  418. out_unlock:
  419. mutex_unlock(&cpuidle_lock);
  420. return ret;
  421. out_sysfs:
  422. cpuidle_remove_sysfs(dev);
  423. out_unregister:
  424. __cpuidle_unregister_device(dev);
  425. goto out_unlock;
  426. }
  427. EXPORT_SYMBOL_GPL(cpuidle_register_device);
  428. /**
  429. * cpuidle_unregister_device - unregisters a CPU's idle PM feature
  430. * @dev: the cpu
  431. */
  432. void cpuidle_unregister_device(struct cpuidle_device *dev)
  433. {
  434. if (!dev || dev->registered == 0)
  435. return;
  436. cpuidle_pause_and_lock();
  437. cpuidle_disable_device(dev);
  438. cpuidle_remove_sysfs(dev);
  439. __cpuidle_unregister_device(dev);
  440. cpuidle_coupled_unregister_device(dev);
  441. cpuidle_resume_and_unlock();
  442. }
  443. EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
  444. /**
  445. * cpuidle_unregister: unregister a driver and the devices. This function
  446. * can be used only if the driver has been previously registered through
  447. * the cpuidle_register function.
  448. *
  449. * @drv: a valid pointer to a struct cpuidle_driver
  450. */
  451. void cpuidle_unregister(struct cpuidle_driver *drv)
  452. {
  453. int cpu;
  454. struct cpuidle_device *device;
  455. for_each_cpu(cpu, drv->cpumask) {
  456. device = &per_cpu(cpuidle_dev, cpu);
  457. cpuidle_unregister_device(device);
  458. }
  459. cpuidle_unregister_driver(drv);
  460. }
  461. EXPORT_SYMBOL_GPL(cpuidle_unregister);
  462. /**
  463. * cpuidle_register: registers the driver and the cpu devices with the
  464. * coupled_cpus passed as parameter. This function is used for all common
  465. * initialization pattern there are in the arch specific drivers. The
  466. * devices is globally defined in this file.
  467. *
  468. * @drv : a valid pointer to a struct cpuidle_driver
  469. * @coupled_cpus: a cpumask for the coupled states
  470. *
  471. * Returns 0 on success, < 0 otherwise
  472. */
  473. int cpuidle_register(struct cpuidle_driver *drv,
  474. const struct cpumask *const coupled_cpus)
  475. {
  476. int ret, cpu;
  477. struct cpuidle_device *device;
  478. ret = cpuidle_register_driver(drv);
  479. if (ret) {
  480. pr_err("failed to register cpuidle driver\n");
  481. return ret;
  482. }
  483. for_each_cpu(cpu, drv->cpumask) {
  484. device = &per_cpu(cpuidle_dev, cpu);
  485. device->cpu = cpu;
  486. #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED
  487. /*
  488. * On multiplatform for ARM, the coupled idle states could be
  489. * enabled in the kernel even if the cpuidle driver does not
  490. * use it. Note, coupled_cpus is a struct copy.
  491. */
  492. if (coupled_cpus)
  493. device->coupled_cpus = *coupled_cpus;
  494. #endif
  495. ret = cpuidle_register_device(device);
  496. if (!ret)
  497. continue;
  498. pr_err("Failed to register cpuidle device for cpu%d\n", cpu);
  499. cpuidle_unregister(drv);
  500. break;
  501. }
  502. return ret;
  503. }
  504. EXPORT_SYMBOL_GPL(cpuidle_register);
  505. #ifdef CONFIG_SMP
  506. /*
  507. * This function gets called when a part of the kernel has a new latency
  508. * requirement. This means we need to get all processors out of their C-state,
  509. * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
  510. * wakes them all right up.
  511. */
  512. static int cpuidle_latency_notify(struct notifier_block *b,
  513. unsigned long l, void *v)
  514. {
  515. wake_up_all_idle_cpus();
  516. return NOTIFY_OK;
  517. }
  518. static struct notifier_block cpuidle_latency_notifier = {
  519. .notifier_call = cpuidle_latency_notify,
  520. };
  521. static inline void latency_notifier_init(struct notifier_block *n)
  522. {
  523. pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n);
  524. }
  525. #else /* CONFIG_SMP */
  526. #define latency_notifier_init(x) do { } while (0)
  527. #endif /* CONFIG_SMP */
  528. /**
  529. * cpuidle_init - core initializer
  530. */
  531. static int __init cpuidle_init(void)
  532. {
  533. int ret;
  534. if (cpuidle_disabled())
  535. return -ENODEV;
  536. ret = cpuidle_add_interface(cpu_subsys.dev_root);
  537. if (ret)
  538. return ret;
  539. latency_notifier_init(&cpuidle_latency_notifier);
  540. return 0;
  541. }
  542. module_param(off, int, 0444);
  543. core_initcall(cpuidle_init);