| 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/percpu-defs.h" | 
|---|
| 12 | #include <linux/clockchips.h> | 
|---|
| 13 | #include <linux/kernel.h> | 
|---|
| 14 | #include <linux/mutex.h> | 
|---|
| 15 | #include <linux/sched.h> | 
|---|
| 16 | #include <linux/sched/clock.h> | 
|---|
| 17 | #include <linux/sched/idle.h> | 
|---|
| 18 | #include <linux/notifier.h> | 
|---|
| 19 | #include <linux/pm_qos.h> | 
|---|
| 20 | #include <linux/cpu.h> | 
|---|
| 21 | #include <linux/cpuidle.h> | 
|---|
| 22 | #include <linux/ktime.h> | 
|---|
| 23 | #include <linux/hrtimer.h> | 
|---|
| 24 | #include <linux/module.h> | 
|---|
| 25 | #include <linux/suspend.h> | 
|---|
| 26 | #include <linux/tick.h> | 
|---|
| 27 | #include <linux/mmu_context.h> | 
|---|
| 28 | #include <linux/context_tracking.h> | 
|---|
| 29 | #include <trace/events/power.h> | 
|---|
| 30 |  | 
|---|
| 31 | #include "cpuidle.h" | 
|---|
| 32 |  | 
|---|
| 33 | DEFINE_PER_CPU(struct cpuidle_device *, cpuidle_devices); | 
|---|
| 34 | DEFINE_PER_CPU(struct cpuidle_device, cpuidle_dev); | 
|---|
| 35 |  | 
|---|
| 36 | DEFINE_MUTEX(cpuidle_lock); | 
|---|
| 37 | LIST_HEAD(cpuidle_detected_devices); | 
|---|
| 38 |  | 
|---|
| 39 | static int enabled_devices; | 
|---|
| 40 | static int off __read_mostly; | 
|---|
| 41 | static int initialized __read_mostly; | 
|---|
| 42 |  | 
|---|
| 43 | int cpuidle_disabled(void) | 
|---|
| 44 | { | 
|---|
| 45 | return off; | 
|---|
| 46 | } | 
|---|
| 47 | void disable_cpuidle(void) | 
|---|
| 48 | { | 
|---|
| 49 | off = 1; | 
|---|
| 50 | } | 
|---|
| 51 |  | 
|---|
| 52 | bool cpuidle_not_available(struct cpuidle_driver *drv, | 
|---|
| 53 | struct cpuidle_device *dev) | 
|---|
| 54 | { | 
|---|
| 55 | return off || !initialized || !drv || !dev || !dev->enabled; | 
|---|
| 56 | } | 
|---|
| 57 |  | 
|---|
| 58 | /** | 
|---|
| 59 | * cpuidle_play_dead - cpu off-lining | 
|---|
| 60 | * | 
|---|
| 61 | * Returns in case of an error or no driver | 
|---|
| 62 | */ | 
|---|
| 63 | int cpuidle_play_dead(void) | 
|---|
| 64 | { | 
|---|
| 65 | struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices); | 
|---|
| 66 | struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); | 
|---|
| 67 | int i; | 
|---|
| 68 |  | 
|---|
| 69 | if (!drv) | 
|---|
| 70 | return -ENODEV; | 
|---|
| 71 |  | 
|---|
| 72 | for (i = drv->state_count - 1; i >= 0; i--) { | 
|---|
| 73 | if (drv->states[i].enter_dead) | 
|---|
| 74 | drv->states[i].enter_dead(dev, i); | 
|---|
| 75 | } | 
|---|
| 76 |  | 
|---|
| 77 | /* | 
|---|
| 78 | * If :enter_dead() is successful, it will never return, so reaching | 
|---|
| 79 | * here means that all of them failed above or were not present. | 
|---|
| 80 | */ | 
|---|
| 81 | return -ENODEV; | 
|---|
| 82 | } | 
|---|
| 83 |  | 
|---|
| 84 | static int find_deepest_state(struct cpuidle_driver *drv, | 
|---|
| 85 | struct cpuidle_device *dev, | 
|---|
| 86 | u64 max_latency_ns, | 
|---|
| 87 | unsigned int forbidden_flags, | 
|---|
| 88 | bool s2idle) | 
|---|
| 89 | { | 
|---|
| 90 | u64 latency_req = 0; | 
|---|
| 91 | int i, ret = 0; | 
|---|
| 92 |  | 
|---|
| 93 | for (i = 1; i < drv->state_count; i++) { | 
|---|
| 94 | struct cpuidle_state *s = &drv->states[i]; | 
|---|
| 95 |  | 
|---|
| 96 | if (dev->states_usage[i].disable || | 
|---|
| 97 | s->exit_latency_ns <= latency_req || | 
|---|
| 98 | s->exit_latency_ns > max_latency_ns || | 
|---|
| 99 | (s->flags & forbidden_flags) || | 
|---|
| 100 | (s2idle && !s->enter_s2idle)) | 
|---|
| 101 | continue; | 
|---|
| 102 |  | 
|---|
| 103 | latency_req = s->exit_latency_ns; | 
|---|
| 104 | ret = i; | 
|---|
| 105 | } | 
|---|
| 106 | return ret; | 
|---|
| 107 | } | 
|---|
| 108 |  | 
|---|
| 109 | /** | 
|---|
| 110 | * cpuidle_use_deepest_state - Set/unset governor override mode. | 
|---|
| 111 | * @latency_limit_ns: Idle state exit latency limit (or no override if 0). | 
|---|
| 112 | * | 
|---|
| 113 | * If @latency_limit_ns is nonzero, set the current CPU to use the deepest idle | 
|---|
| 114 | * state with exit latency within @latency_limit_ns (override governors going | 
|---|
| 115 | * forward), or do not override governors if it is zero. | 
|---|
| 116 | */ | 
|---|
| 117 | void cpuidle_use_deepest_state(u64 latency_limit_ns) | 
|---|
| 118 | { | 
|---|
| 119 | struct cpuidle_device *dev; | 
|---|
| 120 |  | 
|---|
| 121 | preempt_disable(); | 
|---|
| 122 | dev = cpuidle_get_device(); | 
|---|
| 123 | if (dev) | 
|---|
| 124 | dev->forced_idle_latency_limit_ns = latency_limit_ns; | 
|---|
| 125 | preempt_enable(); | 
|---|
| 126 | } | 
|---|
| 127 |  | 
|---|
| 128 | /** | 
|---|
| 129 | * cpuidle_find_deepest_state - Find the deepest available idle state. | 
|---|
| 130 | * @drv: cpuidle driver for the given CPU. | 
|---|
| 131 | * @dev: cpuidle device for the given CPU. | 
|---|
| 132 | * @latency_limit_ns: Idle state exit latency limit | 
|---|
| 133 | * | 
|---|
| 134 | * Return: the index of the deepest available idle state. | 
|---|
| 135 | */ | 
|---|
| 136 | int cpuidle_find_deepest_state(struct cpuidle_driver *drv, | 
|---|
| 137 | struct cpuidle_device *dev, | 
|---|
| 138 | u64 latency_limit_ns) | 
|---|
| 139 | { | 
|---|
| 140 | return find_deepest_state(drv, dev, max_latency_ns: latency_limit_ns, forbidden_flags: 0, s2idle: false); | 
|---|
| 141 | } | 
|---|
| 142 |  | 
|---|
| 143 | #ifdef CONFIG_SUSPEND | 
|---|
| 144 | static noinstr void enter_s2idle_proper(struct cpuidle_driver *drv, | 
|---|
| 145 | struct cpuidle_device *dev, int index) | 
|---|
| 146 | { | 
|---|
| 147 | struct cpuidle_state *target_state = &drv->states[index]; | 
|---|
| 148 | ktime_t time_start, time_end; | 
|---|
| 149 |  | 
|---|
| 150 | instrumentation_begin(); | 
|---|
| 151 |  | 
|---|
| 152 | time_start = ns_to_ktime(ns: local_clock_noinstr()); | 
|---|
| 153 |  | 
|---|
| 154 | tick_freeze(); | 
|---|
| 155 | /* | 
|---|
| 156 | * The state used here cannot be a "coupled" one, because the "coupled" | 
|---|
| 157 | * cpuidle mechanism enables interrupts and doing that with timekeeping | 
|---|
| 158 | * suspended is generally unsafe. | 
|---|
| 159 | */ | 
|---|
| 160 | stop_critical_timings(); | 
|---|
| 161 | if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) { | 
|---|
| 162 | ct_cpuidle_enter(); | 
|---|
| 163 | /* Annotate away the indirect call */ | 
|---|
| 164 | instrumentation_begin(); | 
|---|
| 165 | } | 
|---|
| 166 | target_state->enter_s2idle(dev, drv, index); | 
|---|
| 167 | if (WARN_ON_ONCE(!irqs_disabled())) | 
|---|
| 168 | raw_local_irq_disable(); | 
|---|
| 169 | if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) { | 
|---|
| 170 | instrumentation_end(); | 
|---|
| 171 | ct_cpuidle_exit(); | 
|---|
| 172 | } | 
|---|
| 173 | tick_unfreeze(); | 
|---|
| 174 | start_critical_timings(); | 
|---|
| 175 |  | 
|---|
| 176 | time_end = ns_to_ktime(ns: local_clock_noinstr()); | 
|---|
| 177 |  | 
|---|
| 178 | dev->states_usage[index].s2idle_time += ktime_us_delta(later: time_end, earlier: time_start); | 
|---|
| 179 | dev->states_usage[index].s2idle_usage++; | 
|---|
| 180 | instrumentation_end(); | 
|---|
| 181 | } | 
|---|
| 182 |  | 
|---|
| 183 | /** | 
|---|
| 184 | * cpuidle_enter_s2idle - Enter an idle state suitable for suspend-to-idle. | 
|---|
| 185 | * @drv: cpuidle driver for the given CPU. | 
|---|
| 186 | * @dev: cpuidle device for the given CPU. | 
|---|
| 187 | * | 
|---|
| 188 | * If there are states with the ->enter_s2idle callback, find the deepest of | 
|---|
| 189 | * them and enter it with frozen tick. | 
|---|
| 190 | */ | 
|---|
| 191 | int cpuidle_enter_s2idle(struct cpuidle_driver *drv, struct cpuidle_device *dev) | 
|---|
| 192 | { | 
|---|
| 193 | int index; | 
|---|
| 194 |  | 
|---|
| 195 | /* | 
|---|
| 196 | * Find the deepest state with ->enter_s2idle present, which guarantees | 
|---|
| 197 | * that interrupts won't be enabled when it exits and allows the tick to | 
|---|
| 198 | * be frozen safely. | 
|---|
| 199 | */ | 
|---|
| 200 | index = find_deepest_state(drv, dev, U64_MAX, forbidden_flags: 0, s2idle: true); | 
|---|
| 201 | if (index > 0) { | 
|---|
| 202 | enter_s2idle_proper(drv, dev, index); | 
|---|
| 203 | local_irq_enable(); | 
|---|
| 204 | } | 
|---|
| 205 | return index; | 
|---|
| 206 | } | 
|---|
| 207 | #endif /* CONFIG_SUSPEND */ | 
|---|
| 208 |  | 
|---|
| 209 | /** | 
|---|
| 210 | * cpuidle_enter_state - enter the state and update stats | 
|---|
| 211 | * @dev: cpuidle device for this cpu | 
|---|
| 212 | * @drv: cpuidle driver for this cpu | 
|---|
| 213 | * @index: index into the states table in @drv of the state to enter | 
|---|
| 214 | */ | 
|---|
| 215 | noinstr int cpuidle_enter_state(struct cpuidle_device *dev, | 
|---|
| 216 | struct cpuidle_driver *drv, | 
|---|
| 217 | int index) | 
|---|
| 218 | { | 
|---|
| 219 | int entered_state; | 
|---|
| 220 |  | 
|---|
| 221 | struct cpuidle_state *target_state = &drv->states[index]; | 
|---|
| 222 | bool broadcast = !!(target_state->flags & CPUIDLE_FLAG_TIMER_STOP); | 
|---|
| 223 | ktime_t time_start, time_end; | 
|---|
| 224 |  | 
|---|
| 225 | instrumentation_begin(); | 
|---|
| 226 |  | 
|---|
| 227 | /* | 
|---|
| 228 | * Tell the time framework to switch to a broadcast timer because our | 
|---|
| 229 | * local timer will be shut down.  If a local timer is used from another | 
|---|
| 230 | * CPU as a broadcast timer, this call may fail if it is not available. | 
|---|
| 231 | */ | 
|---|
| 232 | if (broadcast && tick_broadcast_enter()) { | 
|---|
| 233 | index = find_deepest_state(drv, dev, max_latency_ns: target_state->exit_latency_ns, | 
|---|
| 234 | CPUIDLE_FLAG_TIMER_STOP, s2idle: false); | 
|---|
| 235 |  | 
|---|
| 236 | target_state = &drv->states[index]; | 
|---|
| 237 | broadcast = false; | 
|---|
| 238 | } | 
|---|
| 239 |  | 
|---|
| 240 | if (target_state->flags & CPUIDLE_FLAG_TLB_FLUSHED) | 
|---|
| 241 | leave_mm(); | 
|---|
| 242 |  | 
|---|
| 243 | /* Take note of the planned idle state. */ | 
|---|
| 244 | sched_idle_set_state(idle_state: target_state); | 
|---|
| 245 |  | 
|---|
| 246 | trace_cpu_idle(state: index, cpu_id: dev->cpu); | 
|---|
| 247 | time_start = ns_to_ktime(ns: local_clock_noinstr()); | 
|---|
| 248 |  | 
|---|
| 249 | stop_critical_timings(); | 
|---|
| 250 | if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) { | 
|---|
| 251 | ct_cpuidle_enter(); | 
|---|
| 252 | /* Annotate away the indirect call */ | 
|---|
| 253 | instrumentation_begin(); | 
|---|
| 254 | } | 
|---|
| 255 |  | 
|---|
| 256 | /* | 
|---|
| 257 | * NOTE!! | 
|---|
| 258 | * | 
|---|
| 259 | * For cpuidle_state::enter() methods that do *NOT* set | 
|---|
| 260 | * CPUIDLE_FLAG_RCU_IDLE RCU will be disabled here and these functions | 
|---|
| 261 | * must be marked either noinstr or __cpuidle. | 
|---|
| 262 | * | 
|---|
| 263 | * For cpuidle_state::enter() methods that *DO* set | 
|---|
| 264 | * CPUIDLE_FLAG_RCU_IDLE this isn't required, but they must mark the | 
|---|
| 265 | * function calling ct_cpuidle_enter() as noinstr/__cpuidle and all | 
|---|
| 266 | * functions called within the RCU-idle region. | 
|---|
| 267 | */ | 
|---|
| 268 | entered_state = target_state->enter(dev, drv, index); | 
|---|
| 269 |  | 
|---|
| 270 | if (WARN_ONCE(!irqs_disabled(), "%ps leaked IRQ state", target_state->enter)) | 
|---|
| 271 | raw_local_irq_disable(); | 
|---|
| 272 |  | 
|---|
| 273 | if (!(target_state->flags & CPUIDLE_FLAG_RCU_IDLE)) { | 
|---|
| 274 | instrumentation_end(); | 
|---|
| 275 | ct_cpuidle_exit(); | 
|---|
| 276 | } | 
|---|
| 277 | start_critical_timings(); | 
|---|
| 278 |  | 
|---|
| 279 | sched_clock_idle_wakeup_event(); | 
|---|
| 280 | time_end = ns_to_ktime(ns: local_clock_noinstr()); | 
|---|
| 281 | trace_cpu_idle(PWR_EVENT_EXIT, cpu_id: dev->cpu); | 
|---|
| 282 |  | 
|---|
| 283 | /* The cpu is no longer idle or about to enter idle. */ | 
|---|
| 284 | sched_idle_set_state(NULL); | 
|---|
| 285 |  | 
|---|
| 286 | if (broadcast) | 
|---|
| 287 | tick_broadcast_exit(); | 
|---|
| 288 |  | 
|---|
| 289 | if (!cpuidle_state_is_coupled(drv, state: index)) | 
|---|
| 290 | local_irq_enable(); | 
|---|
| 291 |  | 
|---|
| 292 | if (entered_state >= 0) { | 
|---|
| 293 | s64 diff, delay = drv->states[entered_state].exit_latency_ns; | 
|---|
| 294 | int i; | 
|---|
| 295 |  | 
|---|
| 296 | /* | 
|---|
| 297 | * Update cpuidle counters | 
|---|
| 298 | * This can be moved to within driver enter routine, | 
|---|
| 299 | * but that results in multiple copies of same code. | 
|---|
| 300 | */ | 
|---|
| 301 | diff = ktime_sub(time_end, time_start); | 
|---|
| 302 |  | 
|---|
| 303 | dev->last_residency_ns = diff; | 
|---|
| 304 | dev->states_usage[entered_state].time_ns += diff; | 
|---|
| 305 | dev->states_usage[entered_state].usage++; | 
|---|
| 306 |  | 
|---|
| 307 | if (diff < drv->states[entered_state].target_residency_ns) { | 
|---|
| 308 | for (i = entered_state - 1; i >= 0; i--) { | 
|---|
| 309 | if (dev->states_usage[i].disable) | 
|---|
| 310 | continue; | 
|---|
| 311 |  | 
|---|
| 312 | /* Shallower states are enabled, so update. */ | 
|---|
| 313 | dev->states_usage[entered_state].above++; | 
|---|
| 314 | trace_cpu_idle_miss(cpu_id: dev->cpu, state: entered_state, below: false); | 
|---|
| 315 | break; | 
|---|
| 316 | } | 
|---|
| 317 | } else if (diff > delay) { | 
|---|
| 318 | for (i = entered_state + 1; i < drv->state_count; i++) { | 
|---|
| 319 | if (dev->states_usage[i].disable) | 
|---|
| 320 | continue; | 
|---|
| 321 |  | 
|---|
| 322 | /* | 
|---|
| 323 | * Update if a deeper state would have been a | 
|---|
| 324 | * better match for the observed idle duration. | 
|---|
| 325 | */ | 
|---|
| 326 | if (diff - delay >= drv->states[i].target_residency_ns) { | 
|---|
| 327 | dev->states_usage[entered_state].below++; | 
|---|
| 328 | trace_cpu_idle_miss(cpu_id: dev->cpu, state: entered_state, below: true); | 
|---|
| 329 | } | 
|---|
| 330 |  | 
|---|
| 331 | break; | 
|---|
| 332 | } | 
|---|
| 333 | } | 
|---|
| 334 | } else { | 
|---|
| 335 | dev->last_residency_ns = 0; | 
|---|
| 336 | dev->states_usage[index].rejected++; | 
|---|
| 337 | } | 
|---|
| 338 |  | 
|---|
| 339 | instrumentation_end(); | 
|---|
| 340 |  | 
|---|
| 341 | return entered_state; | 
|---|
| 342 | } | 
|---|
| 343 |  | 
|---|
| 344 | /** | 
|---|
| 345 | * cpuidle_select - ask the cpuidle framework to choose an idle state | 
|---|
| 346 | * | 
|---|
| 347 | * @drv: the cpuidle driver | 
|---|
| 348 | * @dev: the cpuidle device | 
|---|
| 349 | * @stop_tick: indication on whether or not to stop the tick | 
|---|
| 350 | * | 
|---|
| 351 | * Returns the index of the idle state.  The return value must not be negative. | 
|---|
| 352 | * | 
|---|
| 353 | * The memory location pointed to by @stop_tick is expected to be written the | 
|---|
| 354 | * 'false' boolean value if the scheduler tick should not be stopped before | 
|---|
| 355 | * entering the returned state. | 
|---|
| 356 | */ | 
|---|
| 357 | int cpuidle_select(struct cpuidle_driver *drv, struct cpuidle_device *dev, | 
|---|
| 358 | bool *stop_tick) | 
|---|
| 359 | { | 
|---|
| 360 | return cpuidle_curr_governor->select(drv, dev, stop_tick); | 
|---|
| 361 | } | 
|---|
| 362 |  | 
|---|
| 363 | /** | 
|---|
| 364 | * cpuidle_enter - enter into the specified idle state | 
|---|
| 365 | * | 
|---|
| 366 | * @drv:   the cpuidle driver tied with the cpu | 
|---|
| 367 | * @dev:   the cpuidle device | 
|---|
| 368 | * @index: the index in the idle state table | 
|---|
| 369 | * | 
|---|
| 370 | * Returns the index in the idle state, < 0 in case of error. | 
|---|
| 371 | * The error code depends on the backend driver | 
|---|
| 372 | */ | 
|---|
| 373 | int cpuidle_enter(struct cpuidle_driver *drv, struct cpuidle_device *dev, | 
|---|
| 374 | int index) | 
|---|
| 375 | { | 
|---|
| 376 | int ret = 0; | 
|---|
| 377 |  | 
|---|
| 378 | /* | 
|---|
| 379 | * Store the next hrtimer, which becomes either next tick or the next | 
|---|
| 380 | * timer event, whatever expires first. Additionally, to make this data | 
|---|
| 381 | * useful for consumers outside cpuidle, we rely on that the governor's | 
|---|
| 382 | * ->select() callback have decided, whether to stop the tick or not. | 
|---|
| 383 | */ | 
|---|
| 384 | WRITE_ONCE(dev->next_hrtimer, tick_nohz_get_next_hrtimer()); | 
|---|
| 385 |  | 
|---|
| 386 | if (cpuidle_state_is_coupled(drv, state: index)) | 
|---|
| 387 | ret = cpuidle_enter_state_coupled(dev, drv, next_state: index); | 
|---|
| 388 | else | 
|---|
| 389 | ret = cpuidle_enter_state(dev, drv, index); | 
|---|
| 390 |  | 
|---|
| 391 | WRITE_ONCE(dev->next_hrtimer, 0); | 
|---|
| 392 | return ret; | 
|---|
| 393 | } | 
|---|
| 394 |  | 
|---|
| 395 | /** | 
|---|
| 396 | * cpuidle_reflect - tell the underlying governor what was the state | 
|---|
| 397 | * we were in | 
|---|
| 398 | * | 
|---|
| 399 | * @dev  : the cpuidle device | 
|---|
| 400 | * @index: the index in the idle state table | 
|---|
| 401 | * | 
|---|
| 402 | */ | 
|---|
| 403 | void cpuidle_reflect(struct cpuidle_device *dev, int index) | 
|---|
| 404 | { | 
|---|
| 405 | if (cpuidle_curr_governor->reflect && index >= 0) | 
|---|
| 406 | cpuidle_curr_governor->reflect(dev, index); | 
|---|
| 407 | } | 
|---|
| 408 |  | 
|---|
| 409 | /* | 
|---|
| 410 | * Min polling interval of 10usec is a guess. It is assuming that | 
|---|
| 411 | * for most users, the time for a single ping-pong workload like | 
|---|
| 412 | * perf bench pipe would generally complete within 10usec but | 
|---|
| 413 | * this is hardware dependent. Actual time can be estimated with | 
|---|
| 414 | * | 
|---|
| 415 | * perf bench sched pipe -l 10000 | 
|---|
| 416 | * | 
|---|
| 417 | * Run multiple times to avoid cpufreq effects. | 
|---|
| 418 | */ | 
|---|
| 419 | #define CPUIDLE_POLL_MIN 10000 | 
|---|
| 420 | #define CPUIDLE_POLL_MAX (TICK_NSEC / 16) | 
|---|
| 421 |  | 
|---|
| 422 | /** | 
|---|
| 423 | * cpuidle_poll_time - return amount of time to poll for, | 
|---|
| 424 | * governors can override dev->poll_limit_ns if necessary | 
|---|
| 425 | * | 
|---|
| 426 | * @drv:   the cpuidle driver tied with the cpu | 
|---|
| 427 | * @dev:   the cpuidle device | 
|---|
| 428 | * | 
|---|
| 429 | */ | 
|---|
| 430 | __cpuidle u64 cpuidle_poll_time(struct cpuidle_driver *drv, | 
|---|
| 431 | struct cpuidle_device *dev) | 
|---|
| 432 | { | 
|---|
| 433 | int i; | 
|---|
| 434 | u64 limit_ns; | 
|---|
| 435 |  | 
|---|
| 436 | BUILD_BUG_ON(CPUIDLE_POLL_MIN > CPUIDLE_POLL_MAX); | 
|---|
| 437 |  | 
|---|
| 438 | if (dev->poll_limit_ns) | 
|---|
| 439 | return dev->poll_limit_ns; | 
|---|
| 440 |  | 
|---|
| 441 | limit_ns = CPUIDLE_POLL_MAX; | 
|---|
| 442 | for (i = 1; i < drv->state_count; i++) { | 
|---|
| 443 | u64 state_limit; | 
|---|
| 444 |  | 
|---|
| 445 | if (dev->states_usage[i].disable) | 
|---|
| 446 | continue; | 
|---|
| 447 |  | 
|---|
| 448 | state_limit = drv->states[i].target_residency_ns; | 
|---|
| 449 | if (state_limit < CPUIDLE_POLL_MIN) | 
|---|
| 450 | continue; | 
|---|
| 451 |  | 
|---|
| 452 | limit_ns = min_t(u64, state_limit, CPUIDLE_POLL_MAX); | 
|---|
| 453 | break; | 
|---|
| 454 | } | 
|---|
| 455 |  | 
|---|
| 456 | dev->poll_limit_ns = limit_ns; | 
|---|
| 457 |  | 
|---|
| 458 | return dev->poll_limit_ns; | 
|---|
| 459 | } | 
|---|
| 460 |  | 
|---|
| 461 | /** | 
|---|
| 462 | * cpuidle_install_idle_handler - installs the cpuidle idle loop handler | 
|---|
| 463 | */ | 
|---|
| 464 | void cpuidle_install_idle_handler(void) | 
|---|
| 465 | { | 
|---|
| 466 | if (enabled_devices) { | 
|---|
| 467 | /* Make sure all changes finished before we switch to new idle */ | 
|---|
| 468 | smp_wmb(); | 
|---|
| 469 | initialized = 1; | 
|---|
| 470 | } | 
|---|
| 471 | } | 
|---|
| 472 |  | 
|---|
| 473 | /** | 
|---|
| 474 | * cpuidle_uninstall_idle_handler - uninstalls the cpuidle idle loop handler | 
|---|
| 475 | */ | 
|---|
| 476 | void cpuidle_uninstall_idle_handler(void) | 
|---|
| 477 | { | 
|---|
| 478 | if (enabled_devices) { | 
|---|
| 479 | initialized = 0; | 
|---|
| 480 | wake_up_all_idle_cpus(); | 
|---|
| 481 | } | 
|---|
| 482 |  | 
|---|
| 483 | /* | 
|---|
| 484 | * Make sure external observers (such as the scheduler) | 
|---|
| 485 | * are done looking at pointed idle states. | 
|---|
| 486 | */ | 
|---|
| 487 | synchronize_rcu(); | 
|---|
| 488 | } | 
|---|
| 489 |  | 
|---|
| 490 | /** | 
|---|
| 491 | * cpuidle_pause_and_lock - temporarily disables CPUIDLE | 
|---|
| 492 | */ | 
|---|
| 493 | void cpuidle_pause_and_lock(void) | 
|---|
| 494 | { | 
|---|
| 495 | mutex_lock(lock: &cpuidle_lock); | 
|---|
| 496 | cpuidle_uninstall_idle_handler(); | 
|---|
| 497 | } | 
|---|
| 498 |  | 
|---|
| 499 | EXPORT_SYMBOL_GPL(cpuidle_pause_and_lock); | 
|---|
| 500 |  | 
|---|
| 501 | /** | 
|---|
| 502 | * cpuidle_resume_and_unlock - resumes CPUIDLE operation | 
|---|
| 503 | */ | 
|---|
| 504 | void cpuidle_resume_and_unlock(void) | 
|---|
| 505 | { | 
|---|
| 506 | cpuidle_install_idle_handler(); | 
|---|
| 507 | mutex_unlock(lock: &cpuidle_lock); | 
|---|
| 508 | } | 
|---|
| 509 |  | 
|---|
| 510 | EXPORT_SYMBOL_GPL(cpuidle_resume_and_unlock); | 
|---|
| 511 |  | 
|---|
| 512 | /* Currently used in suspend/resume path to suspend cpuidle */ | 
|---|
| 513 | void cpuidle_pause(void) | 
|---|
| 514 | { | 
|---|
| 515 | mutex_lock(lock: &cpuidle_lock); | 
|---|
| 516 | cpuidle_uninstall_idle_handler(); | 
|---|
| 517 | mutex_unlock(lock: &cpuidle_lock); | 
|---|
| 518 | } | 
|---|
| 519 |  | 
|---|
| 520 | /* Currently used in suspend/resume path to resume cpuidle */ | 
|---|
| 521 | void cpuidle_resume(void) | 
|---|
| 522 | { | 
|---|
| 523 | mutex_lock(lock: &cpuidle_lock); | 
|---|
| 524 | cpuidle_install_idle_handler(); | 
|---|
| 525 | mutex_unlock(lock: &cpuidle_lock); | 
|---|
| 526 | } | 
|---|
| 527 |  | 
|---|
| 528 | /** | 
|---|
| 529 | * cpuidle_enable_device - enables idle PM for a CPU | 
|---|
| 530 | * @dev: the CPU | 
|---|
| 531 | * | 
|---|
| 532 | * This function must be called between cpuidle_pause_and_lock and | 
|---|
| 533 | * cpuidle_resume_and_unlock when used externally. | 
|---|
| 534 | */ | 
|---|
| 535 | int cpuidle_enable_device(struct cpuidle_device *dev) | 
|---|
| 536 | { | 
|---|
| 537 | int ret; | 
|---|
| 538 | struct cpuidle_driver *drv; | 
|---|
| 539 |  | 
|---|
| 540 | if (!dev) | 
|---|
| 541 | return -EINVAL; | 
|---|
| 542 |  | 
|---|
| 543 | if (dev->enabled) | 
|---|
| 544 | return 0; | 
|---|
| 545 |  | 
|---|
| 546 | if (!cpuidle_curr_governor) | 
|---|
| 547 | return -EIO; | 
|---|
| 548 |  | 
|---|
| 549 | drv = cpuidle_get_cpu_driver(dev); | 
|---|
| 550 |  | 
|---|
| 551 | if (!drv) | 
|---|
| 552 | return -EIO; | 
|---|
| 553 |  | 
|---|
| 554 | if (!dev->registered) | 
|---|
| 555 | return -EINVAL; | 
|---|
| 556 |  | 
|---|
| 557 | ret = cpuidle_add_device_sysfs(device: dev); | 
|---|
| 558 | if (ret) | 
|---|
| 559 | return ret; | 
|---|
| 560 |  | 
|---|
| 561 | if (cpuidle_curr_governor->enable) { | 
|---|
| 562 | ret = cpuidle_curr_governor->enable(drv, dev); | 
|---|
| 563 | if (ret) | 
|---|
| 564 | goto fail_sysfs; | 
|---|
| 565 | } | 
|---|
| 566 |  | 
|---|
| 567 | smp_wmb(); | 
|---|
| 568 |  | 
|---|
| 569 | dev->enabled = 1; | 
|---|
| 570 |  | 
|---|
| 571 | enabled_devices++; | 
|---|
| 572 | return 0; | 
|---|
| 573 |  | 
|---|
| 574 | fail_sysfs: | 
|---|
| 575 | cpuidle_remove_device_sysfs(device: dev); | 
|---|
| 576 |  | 
|---|
| 577 | return ret; | 
|---|
| 578 | } | 
|---|
| 579 |  | 
|---|
| 580 | EXPORT_SYMBOL_GPL(cpuidle_enable_device); | 
|---|
| 581 |  | 
|---|
| 582 | /** | 
|---|
| 583 | * cpuidle_disable_device - disables idle PM for a CPU | 
|---|
| 584 | * @dev: the CPU | 
|---|
| 585 | * | 
|---|
| 586 | * This function must be called between cpuidle_pause_and_lock and | 
|---|
| 587 | * cpuidle_resume_and_unlock when used externally. | 
|---|
| 588 | */ | 
|---|
| 589 | void cpuidle_disable_device(struct cpuidle_device *dev) | 
|---|
| 590 | { | 
|---|
| 591 | struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); | 
|---|
| 592 |  | 
|---|
| 593 | if (!dev || !dev->enabled) | 
|---|
| 594 | return; | 
|---|
| 595 |  | 
|---|
| 596 | if (!drv || !cpuidle_curr_governor) | 
|---|
| 597 | return; | 
|---|
| 598 |  | 
|---|
| 599 | dev->enabled = 0; | 
|---|
| 600 |  | 
|---|
| 601 | if (cpuidle_curr_governor->disable) | 
|---|
| 602 | cpuidle_curr_governor->disable(drv, dev); | 
|---|
| 603 |  | 
|---|
| 604 | cpuidle_remove_device_sysfs(device: dev); | 
|---|
| 605 | enabled_devices--; | 
|---|
| 606 | } | 
|---|
| 607 |  | 
|---|
| 608 | EXPORT_SYMBOL_GPL(cpuidle_disable_device); | 
|---|
| 609 |  | 
|---|
| 610 | static void __cpuidle_unregister_device(struct cpuidle_device *dev) | 
|---|
| 611 | { | 
|---|
| 612 | struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); | 
|---|
| 613 |  | 
|---|
| 614 | list_del(entry: &dev->device_list); | 
|---|
| 615 | per_cpu(cpuidle_devices, dev->cpu) = NULL; | 
|---|
| 616 | module_put(module: drv->owner); | 
|---|
| 617 |  | 
|---|
| 618 | dev->registered = 0; | 
|---|
| 619 | } | 
|---|
| 620 |  | 
|---|
| 621 | static void __cpuidle_device_init(struct cpuidle_device *dev) | 
|---|
| 622 | { | 
|---|
| 623 | memset(s: dev->states_usage, c: 0, n: sizeof(dev->states_usage)); | 
|---|
| 624 | dev->last_residency_ns = 0; | 
|---|
| 625 | dev->next_hrtimer = 0; | 
|---|
| 626 | } | 
|---|
| 627 |  | 
|---|
| 628 | /** | 
|---|
| 629 | * __cpuidle_register_device - internal register function called before register | 
|---|
| 630 | * and enable routines | 
|---|
| 631 | * @dev: the cpu | 
|---|
| 632 | * | 
|---|
| 633 | * cpuidle_lock mutex must be held before this is called | 
|---|
| 634 | */ | 
|---|
| 635 | static int __cpuidle_register_device(struct cpuidle_device *dev) | 
|---|
| 636 | { | 
|---|
| 637 | struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); | 
|---|
| 638 | unsigned int cpu = dev->cpu; | 
|---|
| 639 | int i, ret; | 
|---|
| 640 |  | 
|---|
| 641 | if (per_cpu(cpuidle_devices, cpu)) { | 
|---|
| 642 | pr_info( "CPU%d: cpuidle device already registered\n", cpu); | 
|---|
| 643 | return -EEXIST; | 
|---|
| 644 | } | 
|---|
| 645 |  | 
|---|
| 646 | if (!try_module_get(module: drv->owner)) | 
|---|
| 647 | return -EINVAL; | 
|---|
| 648 |  | 
|---|
| 649 | for (i = 0; i < drv->state_count; i++) { | 
|---|
| 650 | if (drv->states[i].flags & CPUIDLE_FLAG_UNUSABLE) | 
|---|
| 651 | dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER; | 
|---|
| 652 |  | 
|---|
| 653 | if (drv->states[i].flags & CPUIDLE_FLAG_OFF) | 
|---|
| 654 | dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_USER; | 
|---|
| 655 | } | 
|---|
| 656 |  | 
|---|
| 657 | per_cpu(cpuidle_devices, cpu) = dev; | 
|---|
| 658 | list_add(new: &dev->device_list, head: &cpuidle_detected_devices); | 
|---|
| 659 |  | 
|---|
| 660 | ret = cpuidle_coupled_register_device(dev); | 
|---|
| 661 | if (ret) | 
|---|
| 662 | __cpuidle_unregister_device(dev); | 
|---|
| 663 | else | 
|---|
| 664 | dev->registered = 1; | 
|---|
| 665 |  | 
|---|
| 666 | return ret; | 
|---|
| 667 | } | 
|---|
| 668 |  | 
|---|
| 669 | /** | 
|---|
| 670 | * cpuidle_register_device - registers a CPU's idle PM feature | 
|---|
| 671 | * @dev: the cpu | 
|---|
| 672 | */ | 
|---|
| 673 | int cpuidle_register_device(struct cpuidle_device *dev) | 
|---|
| 674 | { | 
|---|
| 675 | int ret = -EBUSY; | 
|---|
| 676 |  | 
|---|
| 677 | if (!dev) | 
|---|
| 678 | return -EINVAL; | 
|---|
| 679 |  | 
|---|
| 680 | mutex_lock(lock: &cpuidle_lock); | 
|---|
| 681 |  | 
|---|
| 682 | if (dev->registered) | 
|---|
| 683 | goto out_unlock; | 
|---|
| 684 |  | 
|---|
| 685 | __cpuidle_device_init(dev); | 
|---|
| 686 |  | 
|---|
| 687 | ret = __cpuidle_register_device(dev); | 
|---|
| 688 | if (ret) | 
|---|
| 689 | goto out_unlock; | 
|---|
| 690 |  | 
|---|
| 691 | ret = cpuidle_add_sysfs(dev); | 
|---|
| 692 | if (ret) | 
|---|
| 693 | goto out_unregister; | 
|---|
| 694 |  | 
|---|
| 695 | ret = cpuidle_enable_device(dev); | 
|---|
| 696 | if (ret) | 
|---|
| 697 | goto out_sysfs; | 
|---|
| 698 |  | 
|---|
| 699 | cpuidle_install_idle_handler(); | 
|---|
| 700 |  | 
|---|
| 701 | out_unlock: | 
|---|
| 702 | mutex_unlock(lock: &cpuidle_lock); | 
|---|
| 703 |  | 
|---|
| 704 | return ret; | 
|---|
| 705 |  | 
|---|
| 706 | out_sysfs: | 
|---|
| 707 | cpuidle_remove_sysfs(dev); | 
|---|
| 708 | out_unregister: | 
|---|
| 709 | __cpuidle_unregister_device(dev); | 
|---|
| 710 | goto out_unlock; | 
|---|
| 711 | } | 
|---|
| 712 |  | 
|---|
| 713 | EXPORT_SYMBOL_GPL(cpuidle_register_device); | 
|---|
| 714 |  | 
|---|
| 715 | /** | 
|---|
| 716 | * cpuidle_unregister_device - unregisters a CPU's idle PM feature | 
|---|
| 717 | * @dev: the cpu | 
|---|
| 718 | */ | 
|---|
| 719 | void cpuidle_unregister_device(struct cpuidle_device *dev) | 
|---|
| 720 | { | 
|---|
| 721 | if (!dev || dev->registered == 0) | 
|---|
| 722 | return; | 
|---|
| 723 |  | 
|---|
| 724 | cpuidle_pause_and_lock(); | 
|---|
| 725 |  | 
|---|
| 726 | cpuidle_disable_device(dev); | 
|---|
| 727 |  | 
|---|
| 728 | cpuidle_remove_sysfs(dev); | 
|---|
| 729 |  | 
|---|
| 730 | __cpuidle_unregister_device(dev); | 
|---|
| 731 |  | 
|---|
| 732 | cpuidle_coupled_unregister_device(dev); | 
|---|
| 733 |  | 
|---|
| 734 | cpuidle_resume_and_unlock(); | 
|---|
| 735 | } | 
|---|
| 736 |  | 
|---|
| 737 | EXPORT_SYMBOL_GPL(cpuidle_unregister_device); | 
|---|
| 738 |  | 
|---|
| 739 | /** | 
|---|
| 740 | * cpuidle_unregister: unregister a driver and the devices. This function | 
|---|
| 741 | * can be used only if the driver has been previously registered through | 
|---|
| 742 | * the cpuidle_register function. | 
|---|
| 743 | * | 
|---|
| 744 | * @drv: a valid pointer to a struct cpuidle_driver | 
|---|
| 745 | */ | 
|---|
| 746 | void cpuidle_unregister(struct cpuidle_driver *drv) | 
|---|
| 747 | { | 
|---|
| 748 | int cpu; | 
|---|
| 749 | struct cpuidle_device *device; | 
|---|
| 750 |  | 
|---|
| 751 | for_each_cpu(cpu, drv->cpumask) { | 
|---|
| 752 | device = &per_cpu(cpuidle_dev, cpu); | 
|---|
| 753 | cpuidle_unregister_device(device); | 
|---|
| 754 | } | 
|---|
| 755 |  | 
|---|
| 756 | cpuidle_unregister_driver(drv); | 
|---|
| 757 | } | 
|---|
| 758 | EXPORT_SYMBOL_GPL(cpuidle_unregister); | 
|---|
| 759 |  | 
|---|
| 760 | /** | 
|---|
| 761 | * cpuidle_register: registers the driver and the cpu devices with the | 
|---|
| 762 | * coupled_cpus passed as parameter. This function is used for all common | 
|---|
| 763 | * initialization pattern there are in the arch specific drivers. The | 
|---|
| 764 | * devices is globally defined in this file. | 
|---|
| 765 | * | 
|---|
| 766 | * @drv         : a valid pointer to a struct cpuidle_driver | 
|---|
| 767 | * @coupled_cpus: a cpumask for the coupled states | 
|---|
| 768 | * | 
|---|
| 769 | * Returns 0 on success, < 0 otherwise | 
|---|
| 770 | */ | 
|---|
| 771 | int cpuidle_register(struct cpuidle_driver *drv, | 
|---|
| 772 | const struct cpumask *const coupled_cpus) | 
|---|
| 773 | { | 
|---|
| 774 | int ret, cpu; | 
|---|
| 775 | struct cpuidle_device *device; | 
|---|
| 776 |  | 
|---|
| 777 | ret = cpuidle_register_driver(drv); | 
|---|
| 778 | if (ret) { | 
|---|
| 779 | pr_err( "failed to register cpuidle driver\n"); | 
|---|
| 780 | return ret; | 
|---|
| 781 | } | 
|---|
| 782 |  | 
|---|
| 783 | for_each_cpu(cpu, drv->cpumask) { | 
|---|
| 784 | device = &per_cpu(cpuidle_dev, cpu); | 
|---|
| 785 | device->cpu = cpu; | 
|---|
| 786 |  | 
|---|
| 787 | #ifdef CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED | 
|---|
| 788 | /* | 
|---|
| 789 | * On multiplatform for ARM, the coupled idle states could be | 
|---|
| 790 | * enabled in the kernel even if the cpuidle driver does not | 
|---|
| 791 | * use it. Note, coupled_cpus is a struct copy. | 
|---|
| 792 | */ | 
|---|
| 793 | if (coupled_cpus) | 
|---|
| 794 | device->coupled_cpus = *coupled_cpus; | 
|---|
| 795 | #endif | 
|---|
| 796 | ret = cpuidle_register_device(device); | 
|---|
| 797 | if (!ret) | 
|---|
| 798 | continue; | 
|---|
| 799 |  | 
|---|
| 800 | pr_err( "Failed to register cpuidle device for cpu%d\n", cpu); | 
|---|
| 801 |  | 
|---|
| 802 | cpuidle_unregister(drv); | 
|---|
| 803 | break; | 
|---|
| 804 | } | 
|---|
| 805 |  | 
|---|
| 806 | return ret; | 
|---|
| 807 | } | 
|---|
| 808 | EXPORT_SYMBOL_GPL(cpuidle_register); | 
|---|
| 809 |  | 
|---|
| 810 | /** | 
|---|
| 811 | * cpuidle_init - core initializer | 
|---|
| 812 | */ | 
|---|
| 813 | static int __init cpuidle_init(void) | 
|---|
| 814 | { | 
|---|
| 815 | if (cpuidle_disabled()) | 
|---|
| 816 | return -ENODEV; | 
|---|
| 817 |  | 
|---|
| 818 | return cpuidle_add_interface(); | 
|---|
| 819 | } | 
|---|
| 820 |  | 
|---|
| 821 | module_param(off, int, 0444); | 
|---|
| 822 | module_param_string(governor, param_governor, CPUIDLE_NAME_LEN, 0444); | 
|---|
| 823 | core_initcall(cpuidle_init); | 
|---|
| 824 |  | 
|---|