| 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, dead_state = -1; |
| 73 | int power_usage = -1; |
| 74 | |
| 75 | if (!drv) |
| 76 | return -ENODEV; |
| 77 | |
| 78 | /* Find lowest-power state that supports long-term idle */ |
| 79 | for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) { |
| 80 | struct cpuidle_state *s = &drv->states[i]; |
| 81 | |
| 82 | if (s->power_usage < power_usage && s->enter_dead) { |
| 83 | power_usage = s->power_usage; |
| 84 | dead_state = i; |
| 85 | } |
| 86 | } |
| 87 | |
| 88 | if (dead_state != -1) |
| 89 | return drv->states[dead_state].enter_dead(dev, dead_state); |
| 90 | |
| 91 | return -ENODEV; |
| 92 | } |
| 93 | |
| 94 | /** |
| 95 | * cpuidle_enter_state - enter the state and update stats |
| 96 | * @dev: cpuidle device for this cpu |
| 97 | * @drv: cpuidle driver for this cpu |
| 98 | * @next_state: index into drv->states of the state to enter |
| 99 | */ |
| 100 | int cpuidle_enter_state(struct cpuidle_device *dev, struct cpuidle_driver *drv, |
| 101 | int next_state) |
| 102 | { |
| 103 | int entered_state; |
| 104 | |
| 105 | entered_state = cpuidle_enter_ops(dev, drv, next_state); |
| 106 | |
| 107 | if (entered_state >= 0) { |
| 108 | /* Update cpuidle counters */ |
| 109 | /* This can be moved to within driver enter routine |
| 110 | * but that results in multiple copies of same code. |
| 111 | */ |
| 112 | dev->states_usage[entered_state].time += |
| 113 | (unsigned long long)dev->last_residency; |
| 114 | dev->states_usage[entered_state].usage++; |
| 115 | } else { |
| 116 | dev->last_residency = 0; |
| 117 | } |
| 118 | |
| 119 | return entered_state; |
| 120 | } |
| 121 | |
| 122 | /** |
| 123 | * cpuidle_idle_call - the main idle loop |
| 124 | * |
| 125 | * NOTE: no locks or semaphores should be used here |
| 126 | * return non-zero on failure |
| 127 | */ |
| 128 | int cpuidle_idle_call(void) |
| 129 | { |
| 130 | struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices); |
| 131 | struct cpuidle_driver *drv; |
| 132 | int next_state, entered_state; |
| 133 | |
| 134 | if (off) |
| 135 | return -ENODEV; |
| 136 | |
| 137 | if (!initialized) |
| 138 | return -ENODEV; |
| 139 | |
| 140 | /* check if the device is ready */ |
| 141 | if (!dev || !dev->enabled) |
| 142 | return -EBUSY; |
| 143 | |
| 144 | drv = cpuidle_get_cpu_driver(dev); |
| 145 | |
| 146 | /* ask the governor for the next state */ |
| 147 | next_state = cpuidle_curr_governor->select(drv, dev); |
| 148 | if (need_resched()) { |
| 149 | dev->last_residency = 0; |
| 150 | /* give the governor an opportunity to reflect on the outcome */ |
| 151 | if (cpuidle_curr_governor->reflect) |
| 152 | cpuidle_curr_governor->reflect(dev, next_state); |
| 153 | local_irq_enable(); |
| 154 | return 0; |
| 155 | } |
| 156 | |
| 157 | trace_power_start_rcuidle(POWER_CSTATE, next_state, dev->cpu); |
| 158 | trace_cpu_idle_rcuidle(next_state, dev->cpu); |
| 159 | |
| 160 | if (cpuidle_state_is_coupled(dev, drv, next_state)) |
| 161 | entered_state = cpuidle_enter_state_coupled(dev, drv, |
| 162 | next_state); |
| 163 | else |
| 164 | entered_state = cpuidle_enter_state(dev, drv, next_state); |
| 165 | |
| 166 | trace_power_end_rcuidle(dev->cpu); |
| 167 | trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu); |
| 168 | |
| 169 | /* give the governor an opportunity to reflect on the outcome */ |
| 170 | if (cpuidle_curr_governor->reflect) |
| 171 | cpuidle_curr_governor->reflect(dev, entered_state); |
| 172 | |
| 173 | return 0; |
| 174 | } |
| 175 | |
| 176 | /** |
| 177 | * cpuidle_install_idle_handler - installs the cpuidle idle loop handler |
| 178 | */ |
| 179 | void cpuidle_install_idle_handler(void) |
| 180 | { |
| 181 | if (enabled_devices) { |
| 182 | /* Make sure all changes finished before we switch to new idle */ |
| 183 | smp_wmb(); |
| 184 | initialized = 1; |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | /** |
| 189 | * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler |
| 190 | */ |
| 191 | void cpuidle_uninstall_idle_handler(void) |
| 192 | { |
| 193 | if (enabled_devices) { |
| 194 | initialized = 0; |
| 195 | kick_all_cpus_sync(); |
| 196 | } |
| 197 | } |
| 198 | |
| 199 | /** |
| 200 | * cpuidle_pause_and_lock - temporarily disables CPUIDLE |
| 201 | */ |
| 202 | void cpuidle_pause_and_lock(void) |
| 203 | { |
| 204 | mutex_lock(&cpuidle_lock); |
| 205 | cpuidle_uninstall_idle_handler(); |
| 206 | } |
| 207 | |
| 208 | EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock); |
| 209 | |
| 210 | /** |
| 211 | * cpuidle_resume_and_unlock - resumes CPUIDLE operation |
| 212 | */ |
| 213 | void cpuidle_resume_and_unlock(void) |
| 214 | { |
| 215 | cpuidle_install_idle_handler(); |
| 216 | mutex_unlock(&cpuidle_lock); |
| 217 | } |
| 218 | |
| 219 | EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock); |
| 220 | |
| 221 | /* Currently used in suspend/resume path to suspend cpuidle */ |
| 222 | void cpuidle_pause(void) |
| 223 | { |
| 224 | mutex_lock(&cpuidle_lock); |
| 225 | cpuidle_uninstall_idle_handler(); |
| 226 | mutex_unlock(&cpuidle_lock); |
| 227 | } |
| 228 | |
| 229 | /* Currently used in suspend/resume path to resume cpuidle */ |
| 230 | void cpuidle_resume(void) |
| 231 | { |
| 232 | mutex_lock(&cpuidle_lock); |
| 233 | cpuidle_install_idle_handler(); |
| 234 | mutex_unlock(&cpuidle_lock); |
| 235 | } |
| 236 | |
| 237 | /** |
| 238 | * cpuidle_wrap_enter - performs timekeeping and irqen around enter function |
| 239 | * @dev: pointer to a valid cpuidle_device object |
| 240 | * @drv: pointer to a valid cpuidle_driver object |
| 241 | * @index: index of the target cpuidle state. |
| 242 | */ |
| 243 | int cpuidle_wrap_enter(struct cpuidle_device *dev, |
| 244 | struct cpuidle_driver *drv, int index, |
| 245 | int (*enter)(struct cpuidle_device *dev, |
| 246 | struct cpuidle_driver *drv, int index)) |
| 247 | { |
| 248 | ktime_t time_start, time_end; |
| 249 | s64 diff; |
| 250 | |
| 251 | time_start = ktime_get(); |
| 252 | |
| 253 | index = enter(dev, drv, index); |
| 254 | |
| 255 | time_end = ktime_get(); |
| 256 | |
| 257 | local_irq_enable(); |
| 258 | |
| 259 | diff = ktime_to_us(ktime_sub(time_end, time_start)); |
| 260 | if (diff > INT_MAX) |
| 261 | diff = INT_MAX; |
| 262 | |
| 263 | dev->last_residency = (int) diff; |
| 264 | |
| 265 | return index; |
| 266 | } |
| 267 | |
| 268 | #ifdef CONFIG_ARCH_HAS_CPU_RELAX |
| 269 | static int poll_idle(struct cpuidle_device *dev, |
| 270 | struct cpuidle_driver *drv, int index) |
| 271 | { |
| 272 | ktime_t t1, t2; |
| 273 | s64 diff; |
| 274 | |
| 275 | t1 = ktime_get(); |
| 276 | local_irq_enable(); |
| 277 | while (!need_resched()) |
| 278 | cpu_relax(); |
| 279 | |
| 280 | t2 = ktime_get(); |
| 281 | diff = ktime_to_us(ktime_sub(t2, t1)); |
| 282 | if (diff > INT_MAX) |
| 283 | diff = INT_MAX; |
| 284 | |
| 285 | dev->last_residency = (int) diff; |
| 286 | |
| 287 | return index; |
| 288 | } |
| 289 | |
| 290 | static void poll_idle_init(struct cpuidle_driver *drv) |
| 291 | { |
| 292 | struct cpuidle_state *state = &drv->states[0]; |
| 293 | |
| 294 | snprintf(state->name, CPUIDLE_NAME_LEN, "POLL"); |
| 295 | snprintf(state->desc, CPUIDLE_DESC_LEN, "CPUIDLE CORE POLL IDLE"); |
| 296 | state->exit_latency = 0; |
| 297 | state->target_residency = 0; |
| 298 | state->power_usage = -1; |
| 299 | state->flags = 0; |
| 300 | state->enter = poll_idle; |
| 301 | state->disabled = false; |
| 302 | } |
| 303 | #else |
| 304 | static void poll_idle_init(struct cpuidle_driver *drv) {} |
| 305 | #endif /* CONFIG_ARCH_HAS_CPU_RELAX */ |
| 306 | |
| 307 | /** |
| 308 | * cpuidle_enable_device - enables idle PM for a CPU |
| 309 | * @dev: the CPU |
| 310 | * |
| 311 | * This function must be called between cpuidle_pause_and_lock and |
| 312 | * cpuidle_resume_and_unlock when used externally. |
| 313 | */ |
| 314 | int cpuidle_enable_device(struct cpuidle_device *dev) |
| 315 | { |
| 316 | int ret, i; |
| 317 | struct cpuidle_driver *drv; |
| 318 | |
| 319 | if (!dev) |
| 320 | return -EINVAL; |
| 321 | |
| 322 | if (dev->enabled) |
| 323 | return 0; |
| 324 | |
| 325 | drv = cpuidle_get_cpu_driver(dev); |
| 326 | |
| 327 | if (!drv || !cpuidle_curr_governor) |
| 328 | return -EIO; |
| 329 | |
| 330 | if (!dev->state_count) |
| 331 | dev->state_count = drv->state_count; |
| 332 | |
| 333 | if (dev->registered == 0) { |
| 334 | ret = __cpuidle_register_device(dev); |
| 335 | if (ret) |
| 336 | return ret; |
| 337 | } |
| 338 | |
| 339 | cpuidle_enter_ops = drv->en_core_tk_irqen ? |
| 340 | cpuidle_enter_tk : cpuidle_enter; |
| 341 | |
| 342 | poll_idle_init(drv); |
| 343 | |
| 344 | ret = cpuidle_add_device_sysfs(dev); |
| 345 | if (ret) |
| 346 | return ret; |
| 347 | |
| 348 | if (cpuidle_curr_governor->enable && |
| 349 | (ret = cpuidle_curr_governor->enable(drv, dev))) |
| 350 | goto fail_sysfs; |
| 351 | |
| 352 | for (i = 0; i < dev->state_count; i++) { |
| 353 | dev->states_usage[i].usage = 0; |
| 354 | dev->states_usage[i].time = 0; |
| 355 | } |
| 356 | dev->last_residency = 0; |
| 357 | |
| 358 | smp_wmb(); |
| 359 | |
| 360 | dev->enabled = 1; |
| 361 | |
| 362 | enabled_devices++; |
| 363 | return 0; |
| 364 | |
| 365 | fail_sysfs: |
| 366 | cpuidle_remove_device_sysfs(dev); |
| 367 | |
| 368 | return ret; |
| 369 | } |
| 370 | |
| 371 | EXPORT_SYMBOL_GPL(cpuidle_enable_device); |
| 372 | |
| 373 | /** |
| 374 | * cpuidle_disable_device - disables idle PM for a CPU |
| 375 | * @dev: the CPU |
| 376 | * |
| 377 | * This function must be called between cpuidle_pause_and_lock and |
| 378 | * cpuidle_resume_and_unlock when used externally. |
| 379 | */ |
| 380 | void cpuidle_disable_device(struct cpuidle_device *dev) |
| 381 | { |
| 382 | struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); |
| 383 | |
| 384 | if (!dev || !dev->enabled) |
| 385 | return; |
| 386 | |
| 387 | if (!drv || !cpuidle_curr_governor) |
| 388 | return; |
| 389 | |
| 390 | dev->enabled = 0; |
| 391 | |
| 392 | if (cpuidle_curr_governor->disable) |
| 393 | cpuidle_curr_governor->disable(drv, dev); |
| 394 | |
| 395 | cpuidle_remove_device_sysfs(dev); |
| 396 | enabled_devices--; |
| 397 | } |
| 398 | |
| 399 | EXPORT_SYMBOL_GPL(cpuidle_disable_device); |
| 400 | |
| 401 | /** |
| 402 | * __cpuidle_register_device - internal register function called before register |
| 403 | * and enable routines |
| 404 | * @dev: the cpu |
| 405 | * |
| 406 | * cpuidle_lock mutex must be held before this is called |
| 407 | */ |
| 408 | static int __cpuidle_register_device(struct cpuidle_device *dev) |
| 409 | { |
| 410 | int ret; |
| 411 | struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); |
| 412 | |
| 413 | if (!try_module_get(drv->owner)) |
| 414 | return -EINVAL; |
| 415 | |
| 416 | per_cpu(cpuidle_devices, dev->cpu) = dev; |
| 417 | list_add(&dev->device_list, &cpuidle_detected_devices); |
| 418 | ret = cpuidle_add_sysfs(dev); |
| 419 | if (ret) |
| 420 | goto err_sysfs; |
| 421 | |
| 422 | ret = cpuidle_coupled_register_device(dev); |
| 423 | if (ret) |
| 424 | goto err_coupled; |
| 425 | |
| 426 | dev->registered = 1; |
| 427 | return 0; |
| 428 | |
| 429 | err_coupled: |
| 430 | cpuidle_remove_sysfs(dev); |
| 431 | err_sysfs: |
| 432 | list_del(&dev->device_list); |
| 433 | per_cpu(cpuidle_devices, dev->cpu) = NULL; |
| 434 | module_put(drv->owner); |
| 435 | return ret; |
| 436 | } |
| 437 | |
| 438 | /** |
| 439 | * cpuidle_register_device - registers a CPU's idle PM feature |
| 440 | * @dev: the cpu |
| 441 | */ |
| 442 | int cpuidle_register_device(struct cpuidle_device *dev) |
| 443 | { |
| 444 | int ret; |
| 445 | |
| 446 | if (!dev) |
| 447 | return -EINVAL; |
| 448 | |
| 449 | mutex_lock(&cpuidle_lock); |
| 450 | |
| 451 | if ((ret = __cpuidle_register_device(dev))) { |
| 452 | mutex_unlock(&cpuidle_lock); |
| 453 | return ret; |
| 454 | } |
| 455 | |
| 456 | cpuidle_enable_device(dev); |
| 457 | cpuidle_install_idle_handler(); |
| 458 | |
| 459 | mutex_unlock(&cpuidle_lock); |
| 460 | |
| 461 | return 0; |
| 462 | |
| 463 | } |
| 464 | |
| 465 | EXPORT_SYMBOL_GPL(cpuidle_register_device); |
| 466 | |
| 467 | /** |
| 468 | * cpuidle_unregister_device - unregisters a CPU's idle PM feature |
| 469 | * @dev: the cpu |
| 470 | */ |
| 471 | void cpuidle_unregister_device(struct cpuidle_device *dev) |
| 472 | { |
| 473 | struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); |
| 474 | |
| 475 | if (dev->registered == 0) |
| 476 | return; |
| 477 | |
| 478 | cpuidle_pause_and_lock(); |
| 479 | |
| 480 | cpuidle_disable_device(dev); |
| 481 | |
| 482 | cpuidle_remove_sysfs(dev); |
| 483 | list_del(&dev->device_list); |
| 484 | per_cpu(cpuidle_devices, dev->cpu) = NULL; |
| 485 | |
| 486 | cpuidle_coupled_unregister_device(dev); |
| 487 | |
| 488 | cpuidle_resume_and_unlock(); |
| 489 | |
| 490 | module_put(drv->owner); |
| 491 | } |
| 492 | |
| 493 | EXPORT_SYMBOL_GPL(cpuidle_unregister_device); |
| 494 | |
| 495 | #ifdef CONFIG_SMP |
| 496 | |
| 497 | static void smp_callback(void *v) |
| 498 | { |
| 499 | /* we already woke the CPU up, nothing more to do */ |
| 500 | } |
| 501 | |
| 502 | /* |
| 503 | * This function gets called when a part of the kernel has a new latency |
| 504 | * requirement. This means we need to get all processors out of their C-state, |
| 505 | * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that |
| 506 | * wakes them all right up. |
| 507 | */ |
| 508 | static int cpuidle_latency_notify(struct notifier_block *b, |
| 509 | unsigned long l, void *v) |
| 510 | { |
| 511 | smp_call_function(smp_callback, NULL, 1); |
| 512 | return NOTIFY_OK; |
| 513 | } |
| 514 | |
| 515 | static struct notifier_block cpuidle_latency_notifier = { |
| 516 | .notifier_call = cpuidle_latency_notify, |
| 517 | }; |
| 518 | |
| 519 | static inline void latency_notifier_init(struct notifier_block *n) |
| 520 | { |
| 521 | pm_qos_add_notifier(PM_QOS_CPU_DMA_LATENCY, n); |
| 522 | } |
| 523 | |
| 524 | #else /* CONFIG_SMP */ |
| 525 | |
| 526 | #define latency_notifier_init(x) do { } while (0) |
| 527 | |
| 528 | #endif /* CONFIG_SMP */ |
| 529 | |
| 530 | /** |
| 531 | * cpuidle_init - core initializer |
| 532 | */ |
| 533 | static int __init cpuidle_init(void) |
| 534 | { |
| 535 | int ret; |
| 536 | |
| 537 | if (cpuidle_disabled()) |
| 538 | return -ENODEV; |
| 539 | |
| 540 | ret = cpuidle_add_interface(cpu_subsys.dev_root); |
| 541 | if (ret) |
| 542 | return ret; |
| 543 | |
| 544 | latency_notifier_init(&cpuidle_latency_notifier); |
| 545 | |
| 546 | return 0; |
| 547 | } |
| 548 | |
| 549 | module_param(off, int, 0444); |
| 550 | core_initcall(cpuidle_init); |