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 25 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices); 26 27 DEFINE_MUTEX(cpuidle_lock); 28 LIST_HEAD(cpuidle_detected_devices); 29 30 static int enabled_devices; 31 static int off __read_mostly; 32 static int initialized __read_mostly; 33 34 int cpuidle_disabled(void) 35 { 36 return off; 37 } 38 void disable_cpuidle(void) 39 { 40 off = 1; 41 } 42 43 static int __cpuidle_register_device(struct cpuidle_device *dev); 44 45 static inline int cpuidle_enter(struct cpuidle_device *dev, 46 struct cpuidle_driver *drv, int index) 47 { 48 struct cpuidle_state *target_state = &drv->states[index]; 49 return target_state->enter(dev, drv, index); 50 } 51 52 static inline int cpuidle_enter_tk(struct cpuidle_device *dev, 53 struct cpuidle_driver *drv, int index) 54 { 55 return cpuidle_wrap_enter(dev, drv, index, cpuidle_enter); 56 } 57 58 typedef int (*cpuidle_enter_t)(struct cpuidle_device *dev, 59 struct cpuidle_driver *drv, int index); 60 61 static cpuidle_enter_t cpuidle_enter_ops; 62 63 /** 64 * cpuidle_play_dead - cpu off-lining 65 * 66 * Returns in case of an error or no driver 67 */ 68 int cpuidle_play_dead(void) 69 { 70 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices); 71 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); 72 int i; 73 74 if (!drv) 75 return -ENODEV; 76 77 /* Find lowest-power state that supports long-term idle */ 78 for (i = drv->state_count - 1; i >= CPUIDLE_DRIVER_STATE_START; i--) 79 if (drv->states[i].enter_dead) 80 return drv->states[i].enter_dead(dev, i); 81 82 return -ENODEV; 83 } 84 85 /** 86 * cpuidle_enter_state - enter the state and update stats 87 * @dev: cpuidle device for this cpu 88 * @drv: cpuidle driver for this cpu 89 * @next_state: index into drv->states of the state to enter 90 */ 91 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv, 92 int next_state) 93 { 94 int entered_state; 95 96 entered_state = cpuidle_enter_ops(dev, drv, next_state); 97 98 if (entered_state >= 0) { 99 /* Update cpuidle counters */ 100 /* This can be moved to within driver enter routine 101 * but that results in multiple copies of same code. 102 */ 103 dev->states_usage[entered_state].time += dev->last_residency; 104 dev->states_usage[entered_state].usage++; 105 } else { 106 dev->last_residency = 0; 107 } 108 109 return entered_state; 110 } 111 112 /** 113 * cpuidle_idle_call - the main idle loop 114 * 115 * NOTE: no locks or semaphores should be used here 116 * return non-zero on failure 117 */ 118 int cpuidle_idle_call(void) 119 { 120 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices); 121 struct cpuidle_driver *drv; 122 int next_state, entered_state; 123 124 if (off) 125 return -ENODEV; 126 127 if (!initialized) 128 return -ENODEV; 129 130 /* check if the device is ready */ 131 if (!dev || !dev->enabled) 132 return -EBUSY; 133 134 drv = cpuidle_get_cpu_driver(dev); 135 136 /* ask the governor for the next state */ 137 next_state = cpuidle_curr_governor->select(drv, dev); 138 if (need_resched()) { 139 dev->last_residency = 0; 140 /* give the governor an opportunity to reflect on the outcome */ 141 if (cpuidle_curr_governor->reflect) 142 cpuidle_curr_governor->reflect(dev, next_state); 143 local_irq_enable(); 144 return 0; 145 } 146 147 trace_cpu_idle_rcuidle(next_state, dev->cpu); 148 149 if (cpuidle_state_is_coupled(dev, drv, next_state)) 150 entered_state = cpuidle_enter_state_coupled(dev, drv, 151 next_state); 152 else 153 entered_state = cpuidle_enter_state(dev, drv, next_state); 154 155 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu); 156 157 /* give the governor an opportunity to reflect on the outcome */ 158 if (cpuidle_curr_governor->reflect) 159 cpuidle_curr_governor->reflect(dev, entered_state); 160 161 return 0; 162 } 163 164 /** 165 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler 166 */ 167 void cpuidle_install_idle_handler(void) 168 { 169 if (enabled_devices) { 170 /* Make sure all changes finished before we switch to new idle */ 171 smp_wmb(); 172 initialized = 1; 173 } 174 } 175 176 /** 177 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler 178 */ 179 void cpuidle_uninstall_idle_handler(void) 180 { 181 if (enabled_devices) { 182 initialized = 0; 183 kick_all_cpus_sync(); 184 } 185 } 186 187 /** 188 * cpuidle_pause_and_lock - temporarily disables CPUIDLE 189 */ 190 void cpuidle_pause_and_lock(void) 191 { 192 mutex_lock(&cpuidle_lock); 193 cpuidle_uninstall_idle_handler(); 194 } 195 196 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock); 197 198 /** 199 * cpuidle_resume_and_unlock - resumes CPUIDLE operation 200 */ 201 void cpuidle_resume_and_unlock(void) 202 { 203 cpuidle_install_idle_handler(); 204 mutex_unlock(&cpuidle_lock); 205 } 206 207 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock); 208 209 /* Currently used in suspend/resume path to suspend cpuidle */ 210 void cpuidle_pause(void) 211 { 212 mutex_lock(&cpuidle_lock); 213 cpuidle_uninstall_idle_handler(); 214 mutex_unlock(&cpuidle_lock); 215 } 216 217 /* Currently used in suspend/resume path to resume cpuidle */ 218 void cpuidle_resume(void) 219 { 220 mutex_lock(&cpuidle_lock); 221 cpuidle_install_idle_handler(); 222 mutex_unlock(&cpuidle_lock); 223 } 224 225 /** 226 * cpuidle_wrap_enter - performs timekeeping and irqen around enter function 227 * @dev: pointer to a valid cpuidle_device object 228 * @drv: pointer to a valid cpuidle_driver object 229 * @index: index of the target cpuidle state. 230 */ 231 int cpuidle_wrap_enter(struct cpuidle_device *dev, 232 struct cpuidle_driver *drv, int index, 233 int (*enter)(struct cpuidle_device *dev, 234 struct cpuidle_driver *drv, int index)) 235 { 236 ktime_t time_start, time_end; 237 s64 diff; 238 239 time_start = ktime_get(); 240 241 index = enter(dev, drv, index); 242 243 time_end = ktime_get(); 244 245 local_irq_enable(); 246 247 diff = ktime_to_us(ktime_sub(time_end, time_start)); 248 if (diff > INT_MAX) 249 diff = INT_MAX; 250 251 dev->last_residency = (int) diff; 252 253 return index; 254 } 255 256 #ifdef CONFIG_ARCH_HAS_CPU_RELAX 257 static int poll_idle(struct cpuidle_device *dev, 258 struct cpuidle_driver *drv, int index) 259 { 260 ktime_t t1, t2; 261 s64 diff; 262 263 t1 = ktime_get(); 264 local_irq_enable(); 265 while (!need_resched()) 266 cpu_relax(); 267 268 t2 = ktime_get(); 269 diff = ktime_to_us(ktime_sub(t2, t1)); 270 if (diff > INT_MAX) 271 diff = INT_MAX; 272 273 dev->last_residency = (int) diff; 274 275 return index; 276 } 277 278 static void poll_idle_init(struct cpuidle_driver *drv) 279 { 280 struct cpuidle_state *state = &drv->states[0]; 281 282 snprintf(state->name, CPUIDLE_NAME_LEN, "POLL"); 283 snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE"); 284 state->exit_latency = 0; 285 state->target_residency = 0; 286 state->power_usage = -1; 287 state->flags = 0; 288 state->enter = poll_idle; 289 state->disabled = false; 290 } 291 #else 292 static void poll_idle_init(struct cpuidle_driver *drv) {} 293 #endif /* CONFIG_ARCH_HAS_CPU_RELAX */ 294 295 /** 296 * cpuidle_enable_device - enables idle PM for a CPU 297 * @dev: the CPU 298 * 299 * This function must be called between cpuidle_pause_and_lock and 300 * cpuidle_resume_and_unlock when used externally. 301 */ 302 int cpuidle_enable_device(struct cpuidle_device *dev) 303 { 304 int ret, i; 305 struct cpuidle_driver *drv; 306 307 if (!dev) 308 return -EINVAL; 309 310 if (dev->enabled) 311 return 0; 312 313 drv = cpuidle_get_cpu_driver(dev); 314 315 if (!drv || !cpuidle_curr_governor) 316 return -EIO; 317 318 if (!dev->state_count) 319 dev->state_count = drv->state_count; 320 321 if (dev->registered == 0) { 322 ret = __cpuidle_register_device(dev); 323 if (ret) 324 return ret; 325 } 326 327 cpuidle_enter_ops = drv->en_core_tk_irqen ? 328 cpuidle_enter_tk : cpuidle_enter; 329 330 poll_idle_init(drv); 331 332 ret = cpuidle_add_device_sysfs(dev); 333 if (ret) 334 return ret; 335 336 if (cpuidle_curr_governor->enable && 337 (ret = cpuidle_curr_governor->enable(drv, dev))) 338 goto fail_sysfs; 339 340 for (i = 0; i < dev->state_count; i++) { 341 dev->states_usage[i].usage = 0; 342 dev->states_usage[i].time = 0; 343 } 344 dev->last_residency = 0; 345 346 smp_wmb(); 347 348 dev->enabled = 1; 349 350 enabled_devices++; 351 return 0; 352 353 fail_sysfs: 354 cpuidle_remove_device_sysfs(dev); 355 356 return ret; 357 } 358 359 EXPORT_SYMBOL_GPL(cpuidle_enable_device); 360 361 /** 362 * cpuidle_disable_device - disables idle PM for a CPU 363 * @dev: the CPU 364 * 365 * This function must be called between cpuidle_pause_and_lock and 366 * cpuidle_resume_and_unlock when used externally. 367 */ 368 void cpuidle_disable_device(struct cpuidle_device *dev) 369 { 370 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); 371 372 if (!dev || !dev->enabled) 373 return; 374 375 if (!drv || !cpuidle_curr_governor) 376 return; 377 378 dev->enabled = 0; 379 380 if (cpuidle_curr_governor->disable) 381 cpuidle_curr_governor->disable(drv, dev); 382 383 cpuidle_remove_device_sysfs(dev); 384 enabled_devices--; 385 } 386 387 EXPORT_SYMBOL_GPL(cpuidle_disable_device); 388 389 /** 390 * __cpuidle_register_device - internal register function called before register 391 * and enable routines 392 * @dev: the cpu 393 * 394 * cpuidle_lock mutex must be held before this is called 395 */ 396 static int __cpuidle_register_device(struct cpuidle_device *dev) 397 { 398 int ret; 399 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); 400 401 if (!try_module_get(drv->owner)) 402 return -EINVAL; 403 404 per_cpu(cpuidle_devices, dev->cpu) = dev; 405 list_add(&dev->device_list, &cpuidle_detected_devices); 406 ret = cpuidle_add_sysfs(dev); 407 if (ret) 408 goto err_sysfs; 409 410 ret = cpuidle_coupled_register_device(dev); 411 if (ret) 412 goto err_coupled; 413 414 dev->registered = 1; 415 return 0; 416 417 err_coupled: 418 cpuidle_remove_sysfs(dev); 419 err_sysfs: 420 list_del(&dev->device_list); 421 per_cpu(cpuidle_devices, dev->cpu) = NULL; 422 module_put(drv->owner); 423 return ret; 424 } 425 426 /** 427 * cpuidle_register_device - registers a CPU's idle PM feature 428 * @dev: the cpu 429 */ 430 int cpuidle_register_device(struct cpuidle_device *dev) 431 { 432 int ret; 433 434 if (!dev) 435 return -EINVAL; 436 437 mutex_lock(&cpuidle_lock); 438 439 if ((ret = __cpuidle_register_device(dev))) { 440 mutex_unlock(&cpuidle_lock); 441 return ret; 442 } 443 444 cpuidle_enable_device(dev); 445 cpuidle_install_idle_handler(); 446 447 mutex_unlock(&cpuidle_lock); 448 449 return 0; 450 451 } 452 453 EXPORT_SYMBOL_GPL(cpuidle_register_device); 454 455 /** 456 * cpuidle_unregister_device - unregisters a CPU's idle PM feature 457 * @dev: the cpu 458 */ 459 void cpuidle_unregister_device(struct cpuidle_device *dev) 460 { 461 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); 462 463 if (dev->registered == 0) 464 return; 465 466 cpuidle_pause_and_lock(); 467 468 cpuidle_disable_device(dev); 469 470 cpuidle_remove_sysfs(dev); 471 list_del(&dev->device_list); 472 per_cpu(cpuidle_devices, dev->cpu) = NULL; 473 474 cpuidle_coupled_unregister_device(dev); 475 476 cpuidle_resume_and_unlock(); 477 478 module_put(drv->owner); 479 } 480 481 EXPORT_SYMBOL_GPL(cpuidle_unregister_device); 482 483 #ifdef CONFIG_SMP 484 485 static void smp_callback(void *v) 486 { 487 /* we already woke the CPU up, nothing more to do */ 488 } 489 490 /* 491 * This function gets called when a part of the kernel has a new latency 492 * requirement. This means we need to get all processors out of their C-state, 493 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that 494 * wakes them all right up. 495 */ 496 static int cpuidle_latency_notify(struct notifier_block *b, 497 unsigned long l, void *v) 498 { 499 smp_call_function(smp_callback, NULL, 1); 500 return NOTIFY_OK; 501 } 502 503 static struct notifier_block cpuidle_latency_notifier = { 504 .notifier_call = cpuidle_latency_notify, 505 }; 506 507 static inline void latency_notifier_init(struct notifier_block *n) 508 { 509 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n); 510 } 511 512 #else /* CONFIG_SMP */ 513 514 #define latency_notifier_init(x) do { } while (0) 515 516 #endif /* CONFIG_SMP */ 517 518 /** 519 * cpuidle_init - core initializer 520 */ 521 static int __init cpuidle_init(void) 522 { 523 int ret; 524 525 if (cpuidle_disabled()) 526 return -ENODEV; 527 528 ret = cpuidle_add_interface(cpu_subsys.dev_root); 529 if (ret) 530 return ret; 531 532 latency_notifier_init(&cpuidle_latency_notifier); 533 534 return 0; 535 } 536 537 module_param(off, int, 0444); 538 core_initcall(cpuidle_init); 539