cpuidle.c revision fc850f39ea54c760ce438a601cfea8ab80c4898e
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 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 <trace/events/power.h> 22 23#include "cpuidle.h" 24 25DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices); 26 27DEFINE_MUTEX(cpuidle_lock); 28LIST_HEAD(cpuidle_detected_devices); 29 30static int enabled_devices; 31static int off __read_mostly; 32static int initialized __read_mostly; 33 34int cpuidle_disabled(void) 35{ 36 return off; 37} 38void disable_cpuidle(void) 39{ 40 off = 1; 41} 42 43#if defined(CONFIG_ARCH_HAS_CPU_IDLE_WAIT) 44static void cpuidle_kick_cpus(void) 45{ 46 cpu_idle_wait(); 47} 48#elif defined(CONFIG_SMP) 49# error "Arch needs cpu_idle_wait() equivalent here" 50#else /* !CONFIG_ARCH_HAS_CPU_IDLE_WAIT && !CONFIG_SMP */ 51static void cpuidle_kick_cpus(void) {} 52#endif 53 54static int __cpuidle_register_device(struct cpuidle_device *dev); 55 56static inline int cpuidle_enter(struct cpuidle_device *dev, 57 struct cpuidle_driver *drv, int index) 58{ 59 struct cpuidle_state *target_state = &drv->states[index]; 60 return target_state->enter(dev, drv, index); 61} 62 63static inline int cpuidle_enter_tk(struct cpuidle_device *dev, 64 struct cpuidle_driver *drv, int index) 65{ 66 return cpuidle_wrap_enter(dev, drv, index, cpuidle_enter); 67} 68 69typedef int (*cpuidle_enter_t)(struct cpuidle_device *dev, 70 struct cpuidle_driver *drv, int index); 71 72static cpuidle_enter_t cpuidle_enter_ops; 73 74/** 75 * cpuidle_idle_call - the main idle loop 76 * 77 * NOTE: no locks or semaphores should be used here 78 * return non-zero on failure 79 */ 80int cpuidle_idle_call(void) 81{ 82 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices); 83 struct cpuidle_driver *drv = cpuidle_get_driver(); 84 int next_state, entered_state; 85 86 if (off) 87 return -ENODEV; 88 89 if (!initialized) 90 return -ENODEV; 91 92 /* check if the device is ready */ 93 if (!dev || !dev->enabled) 94 return -EBUSY; 95 96#if 0 97 /* shows regressions, re-enable for 2.6.29 */ 98 /* 99 * run any timers that can be run now, at this point 100 * before calculating the idle duration etc. 101 */ 102 hrtimer_peek_ahead_timers(); 103#endif 104 105 /* ask the governor for the next state */ 106 next_state = cpuidle_curr_governor->select(drv, dev); 107 if (need_resched()) { 108 local_irq_enable(); 109 return 0; 110 } 111 112 trace_power_start(POWER_CSTATE, next_state, dev->cpu); 113 trace_cpu_idle(next_state, dev->cpu); 114 115 entered_state = cpuidle_enter_ops(dev, drv, next_state); 116 117 trace_power_end(dev->cpu); 118 trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu); 119 120 if (entered_state >= 0) { 121 /* Update cpuidle counters */ 122 /* This can be moved to within driver enter routine 123 * but that results in multiple copies of same code. 124 */ 125 dev->states_usage[entered_state].time += 126 (unsigned long long)dev->last_residency; 127 dev->states_usage[entered_state].usage++; 128 } else { 129 dev->last_residency = 0; 130 } 131 132 /* give the governor an opportunity to reflect on the outcome */ 133 if (cpuidle_curr_governor->reflect) 134 cpuidle_curr_governor->reflect(dev, entered_state); 135 136 return 0; 137} 138 139/** 140 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler 141 */ 142void cpuidle_install_idle_handler(void) 143{ 144 if (enabled_devices) { 145 /* Make sure all changes finished before we switch to new idle */ 146 smp_wmb(); 147 initialized = 1; 148 } 149} 150 151/** 152 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler 153 */ 154void cpuidle_uninstall_idle_handler(void) 155{ 156 if (enabled_devices) { 157 initialized = 0; 158 cpuidle_kick_cpus(); 159 } 160} 161 162/** 163 * cpuidle_pause_and_lock - temporarily disables CPUIDLE 164 */ 165void cpuidle_pause_and_lock(void) 166{ 167 mutex_lock(&cpuidle_lock); 168 cpuidle_uninstall_idle_handler(); 169} 170 171EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock); 172 173/** 174 * cpuidle_resume_and_unlock - resumes CPUIDLE operation 175 */ 176void cpuidle_resume_and_unlock(void) 177{ 178 cpuidle_install_idle_handler(); 179 mutex_unlock(&cpuidle_lock); 180} 181 182EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock); 183 184/** 185 * cpuidle_wrap_enter - performs timekeeping and irqen around enter function 186 * @dev: pointer to a valid cpuidle_device object 187 * @drv: pointer to a valid cpuidle_driver object 188 * @index: index of the target cpuidle state. 189 */ 190int cpuidle_wrap_enter(struct cpuidle_device *dev, 191 struct cpuidle_driver *drv, int index, 192 int (*enter)(struct cpuidle_device *dev, 193 struct cpuidle_driver *drv, int index)) 194{ 195 ktime_t time_start, time_end; 196 s64 diff; 197 198 time_start = ktime_get(); 199 200 index = enter(dev, drv, index); 201 202 time_end = ktime_get(); 203 204 local_irq_enable(); 205 206 diff = ktime_to_us(ktime_sub(time_end, time_start)); 207 if (diff > INT_MAX) 208 diff = INT_MAX; 209 210 dev->last_residency = (int) diff; 211 212 return index; 213} 214 215#ifdef CONFIG_ARCH_HAS_CPU_RELAX 216static int poll_idle(struct cpuidle_device *dev, 217 struct cpuidle_driver *drv, int index) 218{ 219 ktime_t t1, t2; 220 s64 diff; 221 222 t1 = ktime_get(); 223 local_irq_enable(); 224 while (!need_resched()) 225 cpu_relax(); 226 227 t2 = ktime_get(); 228 diff = ktime_to_us(ktime_sub(t2, t1)); 229 if (diff > INT_MAX) 230 diff = INT_MAX; 231 232 dev->last_residency = (int) diff; 233 234 return index; 235} 236 237static void poll_idle_init(struct cpuidle_driver *drv) 238{ 239 struct cpuidle_state *state = &drv->states[0]; 240 241 snprintf(state->name, CPUIDLE_NAME_LEN, "POLL"); 242 snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE"); 243 state->exit_latency = 0; 244 state->target_residency = 0; 245 state->power_usage = -1; 246 state->flags = 0; 247 state->enter = poll_idle; 248 state->disable = 0; 249} 250#else 251static void poll_idle_init(struct cpuidle_driver *drv) {} 252#endif /* CONFIG_ARCH_HAS_CPU_RELAX */ 253 254/** 255 * cpuidle_enable_device - enables idle PM for a CPU 256 * @dev: the CPU 257 * 258 * This function must be called between cpuidle_pause_and_lock and 259 * cpuidle_resume_and_unlock when used externally. 260 */ 261int cpuidle_enable_device(struct cpuidle_device *dev) 262{ 263 int ret, i; 264 struct cpuidle_driver *drv = cpuidle_get_driver(); 265 266 if (dev->enabled) 267 return 0; 268 if (!drv || !cpuidle_curr_governor) 269 return -EIO; 270 if (!dev->state_count) 271 dev->state_count = drv->state_count; 272 273 if (dev->registered == 0) { 274 ret = __cpuidle_register_device(dev); 275 if (ret) 276 return ret; 277 } 278 279 cpuidle_enter_ops = drv->en_core_tk_irqen ? 280 cpuidle_enter_tk : cpuidle_enter; 281 282 poll_idle_init(drv); 283 284 if ((ret = cpuidle_add_state_sysfs(dev))) 285 return ret; 286 287 if (cpuidle_curr_governor->enable && 288 (ret = cpuidle_curr_governor->enable(drv, dev))) 289 goto fail_sysfs; 290 291 for (i = 0; i < dev->state_count; i++) { 292 dev->states_usage[i].usage = 0; 293 dev->states_usage[i].time = 0; 294 } 295 dev->last_residency = 0; 296 297 smp_wmb(); 298 299 dev->enabled = 1; 300 301 enabled_devices++; 302 return 0; 303 304fail_sysfs: 305 cpuidle_remove_state_sysfs(dev); 306 307 return ret; 308} 309 310EXPORT_SYMBOL_GPL(cpuidle_enable_device); 311 312/** 313 * cpuidle_disable_device - disables idle PM for a CPU 314 * @dev: the CPU 315 * 316 * This function must be called between cpuidle_pause_and_lock and 317 * cpuidle_resume_and_unlock when used externally. 318 */ 319void cpuidle_disable_device(struct cpuidle_device *dev) 320{ 321 if (!dev->enabled) 322 return; 323 if (!cpuidle_get_driver() || !cpuidle_curr_governor) 324 return; 325 326 dev->enabled = 0; 327 328 if (cpuidle_curr_governor->disable) 329 cpuidle_curr_governor->disable(cpuidle_get_driver(), dev); 330 331 cpuidle_remove_state_sysfs(dev); 332 enabled_devices--; 333} 334 335EXPORT_SYMBOL_GPL(cpuidle_disable_device); 336 337/** 338 * __cpuidle_register_device - internal register function called before register 339 * and enable routines 340 * @dev: the cpu 341 * 342 * cpuidle_lock mutex must be held before this is called 343 */ 344static int __cpuidle_register_device(struct cpuidle_device *dev) 345{ 346 int ret; 347 struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu); 348 struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver(); 349 350 if (!dev) 351 return -EINVAL; 352 if (!try_module_get(cpuidle_driver->owner)) 353 return -EINVAL; 354 355 init_completion(&dev->kobj_unregister); 356 357 per_cpu(cpuidle_devices, dev->cpu) = dev; 358 list_add(&dev->device_list, &cpuidle_detected_devices); 359 if ((ret = cpuidle_add_sysfs(cpu_dev))) { 360 module_put(cpuidle_driver->owner); 361 return ret; 362 } 363 364 dev->registered = 1; 365 return 0; 366} 367 368/** 369 * cpuidle_register_device - registers a CPU's idle PM feature 370 * @dev: the cpu 371 */ 372int cpuidle_register_device(struct cpuidle_device *dev) 373{ 374 int ret; 375 376 mutex_lock(&cpuidle_lock); 377 378 if ((ret = __cpuidle_register_device(dev))) { 379 mutex_unlock(&cpuidle_lock); 380 return ret; 381 } 382 383 cpuidle_enable_device(dev); 384 cpuidle_install_idle_handler(); 385 386 mutex_unlock(&cpuidle_lock); 387 388 return 0; 389 390} 391 392EXPORT_SYMBOL_GPL(cpuidle_register_device); 393 394/** 395 * cpuidle_unregister_device - unregisters a CPU's idle PM feature 396 * @dev: the cpu 397 */ 398void cpuidle_unregister_device(struct cpuidle_device *dev) 399{ 400 struct device *cpu_dev = get_cpu_device((unsigned long)dev->cpu); 401 struct cpuidle_driver *cpuidle_driver = cpuidle_get_driver(); 402 403 if (dev->registered == 0) 404 return; 405 406 cpuidle_pause_and_lock(); 407 408 cpuidle_disable_device(dev); 409 410 cpuidle_remove_sysfs(cpu_dev); 411 list_del(&dev->device_list); 412 wait_for_completion(&dev->kobj_unregister); 413 per_cpu(cpuidle_devices, dev->cpu) = NULL; 414 415 cpuidle_resume_and_unlock(); 416 417 module_put(cpuidle_driver->owner); 418} 419 420EXPORT_SYMBOL_GPL(cpuidle_unregister_device); 421 422#ifdef CONFIG_SMP 423 424static void smp_callback(void *v) 425{ 426 /* we already woke the CPU up, nothing more to do */ 427} 428 429/* 430 * This function gets called when a part of the kernel has a new latency 431 * requirement. This means we need to get all processors out of their C-state, 432 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that 433 * wakes them all right up. 434 */ 435static int cpuidle_latency_notify(struct notifier_block *b, 436 unsigned long l, void *v) 437{ 438 smp_call_function(smp_callback, NULL, 1); 439 return NOTIFY_OK; 440} 441 442static struct notifier_block cpuidle_latency_notifier = { 443 .notifier_call = cpuidle_latency_notify, 444}; 445 446static inline void latency_notifier_init(struct notifier_block *n) 447{ 448 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n); 449} 450 451#else /* CONFIG_SMP */ 452 453#define latency_notifier_init(x) do { } while (0) 454 455#endif /* CONFIG_SMP */ 456 457/** 458 * cpuidle_init - core initializer 459 */ 460static int __init cpuidle_init(void) 461{ 462 int ret; 463 464 if (cpuidle_disabled()) 465 return -ENODEV; 466 467 ret = cpuidle_add_interface(cpu_subsys.dev_root); 468 if (ret) 469 return ret; 470 471 latency_notifier_init(&cpuidle_latency_notifier); 472 473 return 0; 474} 475 476module_param(off, int, 0444); 477core_initcall(cpuidle_init); 478