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/clockchips.h> 12 #include <linux/kernel.h> 13 #include <linux/mutex.h> 14 #include <linux/sched.h> 15 #include <linux/notifier.h> 16 #include <linux/pm_qos.h> 17 #include <linux/cpu.h> 18 #include <linux/cpuidle.h> 19 #include <linux/ktime.h> 20 #include <linux/hrtimer.h> 21 #include <linux/module.h> 22 #include <linux/suspend.h> 23 #include <linux/tick.h> 24 #include <trace/events/power.h> 25 26 #include "cpuidle.h" 27 28 DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices); 29 DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev); 30 31 DEFINE_MUTEX(cpuidle_lock); 32 LIST_HEAD(cpuidle_detected_devices); 33 34 static int enabled_devices; 35 static int off __read_mostly; 36 static int initialized __read_mostly; 37 38 int cpuidle_disabled(void) 39 { 40 return off; 41 } 42 void disable_cpuidle(void) 43 { 44 off = 1; 45 } 46 47 bool cpuidle_not_available(struct cpuidle_driver *drv, 48 struct cpuidle_device *dev) 49 { 50 return off || !initialized || !drv || !dev || !dev->enabled; 51 } 52 53 /** 54 * cpuidle_play_dead - cpu off-lining 55 * 56 * Returns in case of an error or no driver 57 */ 58 int cpuidle_play_dead(void) 59 { 60 struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices); 61 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); 62 int i; 63 64 if (!drv) 65 return -ENODEV; 66 67 /* Find lowest-power state that supports long-term idle */ 68 for (i = drv->state_count - 1; i >= CPUIDLE_DRIVER_STATE_START; i--) 69 if (drv->states[i].enter_dead) 70 return drv->states[i].enter_dead(dev, i); 71 72 return -ENODEV; 73 } 74 75 static int find_deepest_state(struct cpuidle_driver *drv, 76 struct cpuidle_device *dev, bool freeze) 77 { 78 unsigned int latency_req = 0; 79 int i, ret = freeze ? -1 : CPUIDLE_DRIVER_STATE_START - 1; 80 81 for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) { 82 struct cpuidle_state *s = &drv->states[i]; 83 struct cpuidle_state_usage *su = &dev->states_usage[i]; 84 85 if (s->disabled || su->disable || s->exit_latency <= latency_req 86 || (freeze && !s->enter_freeze)) 87 continue; 88 89 latency_req = s->exit_latency; 90 ret = i; 91 } 92 return ret; 93 } 94 95 /** 96 * cpuidle_find_deepest_state - Find the deepest available idle state. 97 * @drv: cpuidle driver for the given CPU. 98 * @dev: cpuidle device for the given CPU. 99 */ 100 int cpuidle_find_deepest_state(struct cpuidle_driver *drv, 101 struct cpuidle_device *dev) 102 { 103 return find_deepest_state(drv, dev, false); 104 } 105 106 static void enter_freeze_proper(struct cpuidle_driver *drv, 107 struct cpuidle_device *dev, int index) 108 { 109 tick_freeze(); 110 /* 111 * The state used here cannot be a "coupled" one, because the "coupled" 112 * cpuidle mechanism enables interrupts and doing that with timekeeping 113 * suspended is generally unsafe. 114 */ 115 drv->states[index].enter_freeze(dev, drv, index); 116 WARN_ON(!irqs_disabled()); 117 /* 118 * timekeeping_resume() that will be called by tick_unfreeze() for the 119 * last CPU executing it calls functions containing RCU read-side 120 * critical sections, so tell RCU about that. 121 */ 122 RCU_NONIDLE(tick_unfreeze()); 123 } 124 125 /** 126 * cpuidle_enter_freeze - Enter an idle state suitable for suspend-to-idle. 127 * @drv: cpuidle driver for the given CPU. 128 * @dev: cpuidle device for the given CPU. 129 * 130 * If there are states with the ->enter_freeze callback, find the deepest of 131 * them and enter it with frozen tick. 132 */ 133 int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev) 134 { 135 int index; 136 137 /* 138 * Find the deepest state with ->enter_freeze present, which guarantees 139 * that interrupts won't be enabled when it exits and allows the tick to 140 * be frozen safely. 141 */ 142 index = find_deepest_state(drv, dev, true); 143 if (index >= 0) 144 enter_freeze_proper(drv, dev, index); 145 146 return index; 147 } 148 149 /** 150 * cpuidle_enter_state - enter the state and update stats 151 * @dev: cpuidle device for this cpu 152 * @drv: cpuidle driver for this cpu 153 * @next_state: index into drv->states of the state to enter 154 */ 155 int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv, 156 int index) 157 { 158 int entered_state; 159 160 struct cpuidle_state *target_state = &drv->states[index]; 161 bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP); 162 ktime_t time_start, time_end; 163 s64 diff; 164 165 /* 166 * Tell the time framework to switch to a broadcast timer because our 167 * local timer will be shut down. If a local timer is used from another 168 * CPU as a broadcast timer, this call may fail if it is not available. 169 */ 170 if (broadcast && tick_broadcast_enter()) 171 return -EBUSY; 172 173 trace_cpu_idle_rcuidle(index, dev->cpu); 174 time_start = ktime_get(); 175 176 entered_state = target_state->enter(dev, drv, index); 177 178 time_end = ktime_get(); 179 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu); 180 181 if (broadcast) { 182 if (WARN_ON_ONCE(!irqs_disabled())) 183 local_irq_disable(); 184 185 tick_broadcast_exit(); 186 } 187 188 if (!cpuidle_state_is_coupled(dev, drv, entered_state)) 189 local_irq_enable(); 190 191 diff = ktime_to_us(ktime_sub(time_end, time_start)); 192 if (diff > INT_MAX) 193 diff = INT_MAX; 194 195 dev->last_residency = (int) diff; 196 197 if (entered_state >= 0) { 198 /* Update cpuidle counters */ 199 /* This can be moved to within driver enter routine 200 * but that results in multiple copies of same code. 201 */ 202 dev->states_usage[entered_state].time += dev->last_residency; 203 dev->states_usage[entered_state].usage++; 204 } else { 205 dev->last_residency = 0; 206 } 207 208 return entered_state; 209 } 210 211 /** 212 * cpuidle_select - ask the cpuidle framework to choose an idle state 213 * 214 * @drv: the cpuidle driver 215 * @dev: the cpuidle device 216 * 217 * Returns the index of the idle state. 218 */ 219 int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev) 220 { 221 return cpuidle_curr_governor->select(drv, dev); 222 } 223 224 /** 225 * cpuidle_enter - enter into the specified idle state 226 * 227 * @drv: the cpuidle driver tied with the cpu 228 * @dev: the cpuidle device 229 * @index: the index in the idle state table 230 * 231 * Returns the index in the idle state, < 0 in case of error. 232 * The error code depends on the backend driver 233 */ 234 int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev, 235 int index) 236 { 237 if (cpuidle_state_is_coupled(dev, drv, index)) 238 return cpuidle_enter_state_coupled(dev, drv, index); 239 return cpuidle_enter_state(dev, drv, index); 240 } 241 242 /** 243 * cpuidle_reflect - tell the underlying governor what was the state 244 * we were in 245 * 246 * @dev : the cpuidle device 247 * @index: the index in the idle state table 248 * 249 */ 250 void cpuidle_reflect(struct cpuidle_device *dev, int index) 251 { 252 if (cpuidle_curr_governor->reflect) 253 cpuidle_curr_governor->reflect(dev, index); 254 } 255 256 /** 257 * cpuidle_install_idle_handler - installs the cpuidle idle loop handler 258 */ 259 void cpuidle_install_idle_handler(void) 260 { 261 if (enabled_devices) { 262 /* Make sure all changes finished before we switch to new idle */ 263 smp_wmb(); 264 initialized = 1; 265 } 266 } 267 268 /** 269 * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler 270 */ 271 void cpuidle_uninstall_idle_handler(void) 272 { 273 if (enabled_devices) { 274 initialized = 0; 275 wake_up_all_idle_cpus(); 276 } 277 278 /* 279 * Make sure external observers (such as the scheduler) 280 * are done looking at pointed idle states. 281 */ 282 synchronize_rcu(); 283 } 284 285 /** 286 * cpuidle_pause_and_lock - temporarily disables CPUIDLE 287 */ 288 void cpuidle_pause_and_lock(void) 289 { 290 mutex_lock(&cpuidle_lock); 291 cpuidle_uninstall_idle_handler(); 292 } 293 294 EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock); 295 296 /** 297 * cpuidle_resume_and_unlock - resumes CPUIDLE operation 298 */ 299 void cpuidle_resume_and_unlock(void) 300 { 301 cpuidle_install_idle_handler(); 302 mutex_unlock(&cpuidle_lock); 303 } 304 305 EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock); 306 307 /* Currently used in suspend/resume path to suspend cpuidle */ 308 void cpuidle_pause(void) 309 { 310 mutex_lock(&cpuidle_lock); 311 cpuidle_uninstall_idle_handler(); 312 mutex_unlock(&cpuidle_lock); 313 } 314 315 /* Currently used in suspend/resume path to resume cpuidle */ 316 void cpuidle_resume(void) 317 { 318 mutex_lock(&cpuidle_lock); 319 cpuidle_install_idle_handler(); 320 mutex_unlock(&cpuidle_lock); 321 } 322 323 /** 324 * cpuidle_enable_device - enables idle PM for a CPU 325 * @dev: the CPU 326 * 327 * This function must be called between cpuidle_pause_and_lock and 328 * cpuidle_resume_and_unlock when used externally. 329 */ 330 int cpuidle_enable_device(struct cpuidle_device *dev) 331 { 332 int ret; 333 struct cpuidle_driver *drv; 334 335 if (!dev) 336 return -EINVAL; 337 338 if (dev->enabled) 339 return 0; 340 341 drv = cpuidle_get_cpu_driver(dev); 342 343 if (!drv || !cpuidle_curr_governor) 344 return -EIO; 345 346 if (!dev->registered) 347 return -EINVAL; 348 349 ret = cpuidle_add_device_sysfs(dev); 350 if (ret) 351 return ret; 352 353 if (cpuidle_curr_governor->enable && 354 (ret = cpuidle_curr_governor->enable(drv, dev))) 355 goto fail_sysfs; 356 357 smp_wmb(); 358 359 dev->enabled = 1; 360 361 enabled_devices++; 362 return 0; 363 364 fail_sysfs: 365 cpuidle_remove_device_sysfs(dev); 366 367 return ret; 368 } 369 370 EXPORT_SYMBOL_GPL(cpuidle_enable_device); 371 372 /** 373 * cpuidle_disable_device - disables idle PM for a CPU 374 * @dev: the CPU 375 * 376 * This function must be called between cpuidle_pause_and_lock and 377 * cpuidle_resume_and_unlock when used externally. 378 */ 379 void cpuidle_disable_device(struct cpuidle_device *dev) 380 { 381 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); 382 383 if (!dev || !dev->enabled) 384 return; 385 386 if (!drv || !cpuidle_curr_governor) 387 return; 388 389 dev->enabled = 0; 390 391 if (cpuidle_curr_governor->disable) 392 cpuidle_curr_governor->disable(drv, dev); 393 394 cpuidle_remove_device_sysfs(dev); 395 enabled_devices--; 396 } 397 398 EXPORT_SYMBOL_GPL(cpuidle_disable_device); 399 400 static void __cpuidle_unregister_device(struct cpuidle_device *dev) 401 { 402 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); 403 404 list_del(&dev->device_list); 405 per_cpu(cpuidle_devices, dev->cpu) = NULL; 406 module_put(drv->owner); 407 } 408 409 static void __cpuidle_device_init(struct cpuidle_device *dev) 410 { 411 memset(dev->states_usage, 0, sizeof(dev->states_usage)); 412 dev->last_residency = 0; 413 } 414 415 /** 416 * __cpuidle_register_device - internal register function called before register 417 * and enable routines 418 * @dev: the cpu 419 * 420 * cpuidle_lock mutex must be held before this is called 421 */ 422 static int __cpuidle_register_device(struct cpuidle_device *dev) 423 { 424 int ret; 425 struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); 426 427 if (!try_module_get(drv->owner)) 428 return -EINVAL; 429 430 per_cpu(cpuidle_devices, dev->cpu) = dev; 431 list_add(&dev->device_list, &cpuidle_detected_devices); 432 433 ret = cpuidle_coupled_register_device(dev); 434 if (ret) 435 __cpuidle_unregister_device(dev); 436 else 437 dev->registered = 1; 438 439 return ret; 440 } 441 442 /** 443 * cpuidle_register_device - registers a CPU's idle PM feature 444 * @dev: the cpu 445 */ 446 int cpuidle_register_device(struct cpuidle_device *dev) 447 { 448 int ret = -EBUSY; 449 450 if (!dev) 451 return -EINVAL; 452 453 mutex_lock(&cpuidle_lock); 454 455 if (dev->registered) 456 goto out_unlock; 457 458 __cpuidle_device_init(dev); 459 460 ret = __cpuidle_register_device(dev); 461 if (ret) 462 goto out_unlock; 463 464 ret = cpuidle_add_sysfs(dev); 465 if (ret) 466 goto out_unregister; 467 468 ret = cpuidle_enable_device(dev); 469 if (ret) 470 goto out_sysfs; 471 472 cpuidle_install_idle_handler(); 473 474 out_unlock: 475 mutex_unlock(&cpuidle_lock); 476 477 return ret; 478 479 out_sysfs: 480 cpuidle_remove_sysfs(dev); 481 out_unregister: 482 __cpuidle_unregister_device(dev); 483 goto out_unlock; 484 } 485 486 EXPORT_SYMBOL_GPL(cpuidle_register_device); 487 488 /** 489 * cpuidle_unregister_device - unregisters a CPU's idle PM feature 490 * @dev: the cpu 491 */ 492 void cpuidle_unregister_device(struct cpuidle_device *dev) 493 { 494 if (!dev || dev->registered == 0) 495 return; 496 497 cpuidle_pause_and_lock(); 498 499 cpuidle_disable_device(dev); 500 501 cpuidle_remove_sysfs(dev); 502 503 __cpuidle_unregister_device(dev); 504 505 cpuidle_coupled_unregister_device(dev); 506 507 cpuidle_resume_and_unlock(); 508 } 509 510 EXPORT_SYMBOL_GPL(cpuidle_unregister_device); 511 512 /** 513 * cpuidle_unregister: unregister a driver and the devices. This function 514 * can be used only if the driver has been previously registered through 515 * the cpuidle_register function. 516 * 517 * @drv: a valid pointer to a struct cpuidle_driver 518 */ 519 void cpuidle_unregister(struct cpuidle_driver *drv) 520 { 521 int cpu; 522 struct cpuidle_device *device; 523 524 for_each_cpu(cpu, drv->cpumask) { 525 device = &per_cpu(cpuidle_dev, cpu); 526 cpuidle_unregister_device(device); 527 } 528 529 cpuidle_unregister_driver(drv); 530 } 531 EXPORT_SYMBOL_GPL(cpuidle_unregister); 532 533 /** 534 * cpuidle_register: registers the driver and the cpu devices with the 535 * coupled_cpus passed as parameter. This function is used for all common 536 * initialization pattern there are in the arch specific drivers. The 537 * devices is globally defined in this file. 538 * 539 * @drv : a valid pointer to a struct cpuidle_driver 540 * @coupled_cpus: a cpumask for the coupled states 541 * 542 * Returns 0 on success, < 0 otherwise 543 */ 544 int cpuidle_register(struct cpuidle_driver *drv, 545 const struct cpumask *const coupled_cpus) 546 { 547 int ret, cpu; 548 struct cpuidle_device *device; 549 550 ret = cpuidle_register_driver(drv); 551 if (ret) { 552 pr_err("failed to register cpuidle driver\n"); 553 return ret; 554 } 555 556 for_each_cpu(cpu, drv->cpumask) { 557 device = &per_cpu(cpuidle_dev, cpu); 558 device->cpu = cpu; 559 560 #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED 561 /* 562 * On multiplatform for ARM, the coupled idle states could be 563 * enabled in the kernel even if the cpuidle driver does not 564 * use it. Note, coupled_cpus is a struct copy. 565 */ 566 if (coupled_cpus) 567 device->coupled_cpus = *coupled_cpus; 568 #endif 569 ret = cpuidle_register_device(device); 570 if (!ret) 571 continue; 572 573 pr_err("Failed to register cpuidle device for cpu%d\n", cpu); 574 575 cpuidle_unregister(drv); 576 break; 577 } 578 579 return ret; 580 } 581 EXPORT_SYMBOL_GPL(cpuidle_register); 582 583 #ifdef CONFIG_SMP 584 585 /* 586 * This function gets called when a part of the kernel has a new latency 587 * requirement. This means we need to get all processors out of their C-state, 588 * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that 589 * wakes them all right up. 590 */ 591 static int cpuidle_latency_notify(struct notifier_block *b, 592 unsigned long l, void *v) 593 { 594 wake_up_all_idle_cpus(); 595 return NOTIFY_OK; 596 } 597 598 static struct notifier_block cpuidle_latency_notifier = { 599 .notifier_call = cpuidle_latency_notify, 600 }; 601 602 static inline void latency_notifier_init(struct notifier_block *n) 603 { 604 pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n); 605 } 606 607 #else /* CONFIG_SMP */ 608 609 #define latency_notifier_init(x) do { } while (0) 610 611 #endif /* CONFIG_SMP */ 612 613 /** 614 * cpuidle_init - core initializer 615 */ 616 static int __init cpuidle_init(void) 617 { 618 int ret; 619 620 if (cpuidle_disabled()) 621 return -ENODEV; 622 623 ret = cpuidle_add_interface(cpu_subsys.dev_root); 624 if (ret) 625 return ret; 626 627 latency_notifier_init(&cpuidle_latency_notifier); 628 629 return 0; 630 } 631 632 module_param(off, int, 0444); 633 core_initcall(cpuidle_init); 634