kernel/time/tick-sched.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
tick_get_tick_sched
tick_do_update_jiffies64
tick_init_jiffy_update
tick_sched_do_timer
tick_sched_handle
check_tick_dependency
can_stop_full_tick
nohz_full_kick_func
tick_nohz_full_kick
tick_nohz_full_kick_cpu
tick_nohz_full_kick_all
tick_nohz_dep_set_all
tick_nohz_dep_set
tick_nohz_dep_clear
tick_nohz_dep_set_cpu
tick_nohz_dep_clear_cpu
tick_nohz_dep_set_task
tick_nohz_dep_clear_task
tick_nohz_dep_set_signal
tick_nohz_dep_clear_signal
__tick_nohz_task_switch
tick_nohz_full_setup
tick_nohz_cpu_down
tick_nohz_init
setup_tick_nohz
tick_nohz_tick_stopped
tick_nohz_tick_stopped_cpu
tick_nohz_update_jiffies
update_ts_time_stats
tick_nohz_stop_idle
tick_nohz_start_idle
get_cpu_idle_time_us
get_cpu_iowait_time_us
tick_nohz_restart
local_timer_softirq_pending
tick_nohz_next_event
tick_nohz_stop_tick
tick_nohz_retain_tick
tick_nohz_stop_sched_tick
tick_nohz_restart_sched_tick
tick_nohz_full_update_tick
can_stop_idle_tick
__tick_nohz_idle_stop_tick
tick_nohz_idle_stop_tick
tick_nohz_idle_retain_tick
tick_nohz_idle_enter
tick_nohz_irq_exit
tick_nohz_idle_got_tick
tick_nohz_get_next_hrtimer
tick_nohz_get_sleep_length
tick_nohz_get_idle_calls_cpu
tick_nohz_get_idle_calls
tick_nohz_account_idle_ticks
__tick_nohz_idle_restart_tick
tick_nohz_idle_restart_tick
tick_nohz_idle_exit
tick_nohz_handler
tick_nohz_activate
tick_nohz_switch_to_nohz
tick_nohz_irq_enter
tick_nohz_switch_to_nohz
tick_nohz_irq_enter
tick_nohz_activate
tick_irq_enter
tick_sched_timer
skew_tick
tick_setup_sched_timer
tick_cancel_sched_timer
tick_clock_notify
tick_oneshot_notify
tick_check_oneshot_change
   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
   4  *  Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
   5  *  Copyright(C) 2006-2007  Timesys Corp., Thomas Gleixner
   6  *
   7  *  No idle tick implementation for low and high resolution timers
   8  *
   9  *  Started by: Thomas Gleixner and Ingo Molnar
  10  */
  11 #include <linux/cpu.h>
  12 #include <linux/err.h>
  13 #include <linux/hrtimer.h>
  14 #include <linux/interrupt.h>
  15 #include <linux/kernel_stat.h>
  16 #include <linux/percpu.h>
  17 #include <linux/nmi.h>
  18 #include <linux/profile.h>
  19 #include <linux/sched/signal.h>
  20 #include <linux/sched/clock.h>
  21 #include <linux/sched/stat.h>
  22 #include <linux/sched/nohz.h>
  23 #include <linux/module.h>
  24 #include <linux/irq_work.h>
  25 #include <linux/posix-timers.h>
  26 #include <linux/context_tracking.h>
  27 #include <linux/mm.h>
  28 
  29 #include <asm/irq_regs.h>
  30 
  31 #include "tick-internal.h"
  32 
  33 #include <trace/events/timer.h>
  34 
  35 /*
  36  * Per-CPU nohz control structure
  37  */
  38 static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched);
  39 
  40 struct tick_sched *tick_get_tick_sched(int cpu)
  41 {
  42         return &per_cpu(tick_cpu_sched, cpu);
  43 }
  44 
  45 #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS)
  46 /*
  47  * The time, when the last jiffy update happened. Protected by jiffies_lock.
  48  */
  49 static ktime_t last_jiffies_update;
  50 
  51 /*
  52  * Must be called with interrupts disabled !
  53  */
  54 static void tick_do_update_jiffies64(ktime_t now)
  55 {
  56         unsigned long ticks = 0;
  57         ktime_t delta;
  58 
  59         /*
  60          * Do a quick check without holding jiffies_lock:
  61          * The READ_ONCE() pairs with two updates done later in this function.
  62          */
  63         delta = ktime_sub(now, READ_ONCE(last_jiffies_update));
  64         if (delta < tick_period)
  65                 return;
  66 
  67         /* Reevaluate with jiffies_lock held */
  68         write_seqlock(&jiffies_lock);
  69 
  70         delta = ktime_sub(now, last_jiffies_update);
  71         if (delta >= tick_period) {
  72 
  73                 delta = ktime_sub(delta, tick_period);
  74                 /* Pairs with the lockless read in this function. */
  75                 WRITE_ONCE(last_jiffies_update,
  76                            ktime_add(last_jiffies_update, tick_period));
  77 
  78                 /* Slow path for long timeouts */
  79                 if (unlikely(delta >= tick_period)) {
  80                         s64 incr = ktime_to_ns(tick_period);
  81 
  82                         ticks = ktime_divns(delta, incr);
  83 
  84                         /* Pairs with the lockless read in this function. */
  85                         WRITE_ONCE(last_jiffies_update,
  86                                    ktime_add_ns(last_jiffies_update,
  87                                                 incr * ticks));
  88                 }
  89                 do_timer(++ticks);
  90 
  91                 /* Keep the tick_next_period variable up to date */
  92                 tick_next_period = ktime_add(last_jiffies_update, tick_period);
  93         } else {
  94                 write_sequnlock(&jiffies_lock);
  95                 return;
  96         }
  97         write_sequnlock(&jiffies_lock);
  98         update_wall_time();
  99 }
 100 
 101 /*
 102  * Initialize and return retrieve the jiffies update.
 103  */
 104 static ktime_t tick_init_jiffy_update(void)
 105 {
 106         ktime_t period;
 107 
 108         write_seqlock(&jiffies_lock);
 109         /* Did we start the jiffies update yet ? */
 110         if (last_jiffies_update == 0)
 111                 last_jiffies_update = tick_next_period;
 112         period = last_jiffies_update;
 113         write_sequnlock(&jiffies_lock);
 114         return period;
 115 }
 116 
 117 static void tick_sched_do_timer(struct tick_sched *ts, ktime_t now)
 118 {
 119         int cpu = smp_processor_id();
 120 
 121 #ifdef CONFIG_NO_HZ_COMMON
 122         /*
 123          * Check if the do_timer duty was dropped. We don't care about
 124          * concurrency: This happens only when the CPU in charge went
 125          * into a long sleep. If two CPUs happen to assign themselves to
 126          * this duty, then the jiffies update is still serialized by
 127          * jiffies_lock.
 128          *
 129          * If nohz_full is enabled, this should not happen because the
 130          * tick_do_timer_cpu never relinquishes.
 131          */
 132         if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) {
 133 #ifdef CONFIG_NO_HZ_FULL
 134                 WARN_ON(tick_nohz_full_running);
 135 #endif
 136                 tick_do_timer_cpu = cpu;
 137         }
 138 #endif
 139 
 140         /* Check, if the jiffies need an update */
 141         if (tick_do_timer_cpu == cpu)
 142                 tick_do_update_jiffies64(now);
 143 
 144         if (ts->inidle)
 145                 ts->got_idle_tick = 1;
 146 }
 147 
 148 static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs)
 149 {
 150 #ifdef CONFIG_NO_HZ_COMMON
 151         /*
 152          * When we are idle and the tick is stopped, we have to touch
 153          * the watchdog as we might not schedule for a really long
 154          * time. This happens on complete idle SMP systems while
 155          * waiting on the login prompt. We also increment the "start of
 156          * idle" jiffy stamp so the idle accounting adjustment we do
 157          * when we go busy again does not account too much ticks.
 158          */
 159         if (ts->tick_stopped) {
 160                 touch_softlockup_watchdog_sched();
 161                 if (is_idle_task(current))
 162                         ts->idle_jiffies++;
 163                 /*
 164                  * In case the current tick fired too early past its expected
 165                  * expiration, make sure we don't bypass the next clock reprogramming
 166                  * to the same deadline.
 167                  */
 168                 ts->next_tick = 0;
 169         }
 170 #endif
 171         update_process_times(user_mode(regs));
 172         profile_tick(CPU_PROFILING);
 173 }
 174 #endif
 175 
 176 #ifdef CONFIG_NO_HZ_FULL
 177 cpumask_var_t tick_nohz_full_mask;
 178 bool tick_nohz_full_running;
 179 static atomic_t tick_dep_mask;
 180 
 181 static bool check_tick_dependency(atomic_t *dep)
 182 {
 183         int val = atomic_read(dep);
 184 
 185         if (val & TICK_DEP_MASK_POSIX_TIMER) {
 186                 trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER);
 187                 return true;
 188         }
 189 
 190         if (val & TICK_DEP_MASK_PERF_EVENTS) {
 191                 trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS);
 192                 return true;
 193         }
 194 
 195         if (val & TICK_DEP_MASK_SCHED) {
 196                 trace_tick_stop(0, TICK_DEP_MASK_SCHED);
 197                 return true;
 198         }
 199 
 200         if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) {
 201                 trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE);
 202                 return true;
 203         }
 204 
 205         return false;
 206 }
 207 
 208 static bool can_stop_full_tick(int cpu, struct tick_sched *ts)
 209 {
 210         lockdep_assert_irqs_disabled();
 211 
 212         if (unlikely(!cpu_online(cpu)))
 213                 return false;
 214 
 215         if (check_tick_dependency(&tick_dep_mask))
 216                 return false;
 217 
 218         if (check_tick_dependency(&ts->tick_dep_mask))
 219                 return false;
 220 
 221         if (check_tick_dependency(&current->tick_dep_mask))
 222                 return false;
 223 
 224         if (check_tick_dependency(&current->signal->tick_dep_mask))
 225                 return false;
 226 
 227         return true;
 228 }
 229 
 230 static void nohz_full_kick_func(struct irq_work *work)
 231 {
 232         /* Empty, the tick restart happens on tick_nohz_irq_exit() */
 233 }
 234 
 235 static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = {
 236         .func = nohz_full_kick_func,
 237 };
 238 
 239 /*
 240  * Kick this CPU if it's full dynticks in order to force it to
 241  * re-evaluate its dependency on the tick and restart it if necessary.
 242  * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(),
 243  * is NMI safe.
 244  */
 245 static void tick_nohz_full_kick(void)
 246 {
 247         if (!tick_nohz_full_cpu(smp_processor_id()))
 248                 return;
 249 
 250         irq_work_queue(this_cpu_ptr(&nohz_full_kick_work));
 251 }
 252 
 253 /*
 254  * Kick the CPU if it's full dynticks in order to force it to
 255  * re-evaluate its dependency on the tick and restart it if necessary.
 256  */
 257 void tick_nohz_full_kick_cpu(int cpu)
 258 {
 259         if (!tick_nohz_full_cpu(cpu))
 260                 return;
 261 
 262         irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu);
 263 }
 264 
 265 /*
 266  * Kick all full dynticks CPUs in order to force these to re-evaluate
 267  * their dependency on the tick and restart it if necessary.
 268  */
 269 static void tick_nohz_full_kick_all(void)
 270 {
 271         int cpu;
 272 
 273         if (!tick_nohz_full_running)
 274                 return;
 275 
 276         preempt_disable();
 277         for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask)
 278                 tick_nohz_full_kick_cpu(cpu);
 279         preempt_enable();
 280 }
 281 
 282 static void tick_nohz_dep_set_all(atomic_t *dep,
 283                                   enum tick_dep_bits bit)
 284 {
 285         int prev;
 286 
 287         prev = atomic_fetch_or(BIT(bit), dep);
 288         if (!prev)
 289                 tick_nohz_full_kick_all();
 290 }
 291 
 292 /*
 293  * Set a global tick dependency. Used by perf events that rely on freq and
 294  * by unstable clock.
 295  */
 296 void tick_nohz_dep_set(enum tick_dep_bits bit)
 297 {
 298         tick_nohz_dep_set_all(&tick_dep_mask, bit);
 299 }
 300 
 301 void tick_nohz_dep_clear(enum tick_dep_bits bit)
 302 {
 303         atomic_andnot(BIT(bit), &tick_dep_mask);
 304 }
 305 
 306 /*
 307  * Set per-CPU tick dependency. Used by scheduler and perf events in order to
 308  * manage events throttling.
 309  */
 310 void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit)
 311 {
 312         int prev;
 313         struct tick_sched *ts;
 314 
 315         ts = per_cpu_ptr(&tick_cpu_sched, cpu);
 316 
 317         prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask);
 318         if (!prev) {
 319                 preempt_disable();
 320                 /* Perf needs local kick that is NMI safe */
 321                 if (cpu == smp_processor_id()) {
 322                         tick_nohz_full_kick();
 323                 } else {
 324                         /* Remote irq work not NMI-safe */
 325                         if (!WARN_ON_ONCE(in_nmi()))
 326                                 tick_nohz_full_kick_cpu(cpu);
 327                 }
 328                 preempt_enable();
 329         }
 330 }
 331 
 332 void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit)
 333 {
 334         struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu);
 335 
 336         atomic_andnot(BIT(bit), &ts->tick_dep_mask);
 337 }
 338 
 339 /*
 340  * Set a per-task tick dependency. Posix CPU timers need this in order to elapse
 341  * per task timers.
 342  */
 343 void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit)
 344 {
 345         /*
 346          * We could optimize this with just kicking the target running the task
 347          * if that noise matters for nohz full users.
 348          */
 349         tick_nohz_dep_set_all(&tsk->tick_dep_mask, bit);
 350 }
 351 
 352 void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit)
 353 {
 354         atomic_andnot(BIT(bit), &tsk->tick_dep_mask);
 355 }
 356 
 357 /*
 358  * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse
 359  * per process timers.
 360  */
 361 void tick_nohz_dep_set_signal(struct signal_struct *sig, enum tick_dep_bits bit)
 362 {
 363         tick_nohz_dep_set_all(&sig->tick_dep_mask, bit);
 364 }
 365 
 366 void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit)
 367 {
 368         atomic_andnot(BIT(bit), &sig->tick_dep_mask);
 369 }
 370 
 371 /*
 372  * Re-evaluate the need for the tick as we switch the current task.
 373  * It might need the tick due to per task/process properties:
 374  * perf events, posix CPU timers, ...
 375  */
 376 void __tick_nohz_task_switch(void)
 377 {
 378         unsigned long flags;
 379         struct tick_sched *ts;
 380 
 381         local_irq_save(flags);
 382 
 383         if (!tick_nohz_full_cpu(smp_processor_id()))
 384                 goto out;
 385 
 386         ts = this_cpu_ptr(&tick_cpu_sched);
 387 
 388         if (ts->tick_stopped) {
 389                 if (atomic_read(&current->tick_dep_mask) ||
 390                     atomic_read(&current->signal->tick_dep_mask))
 391                         tick_nohz_full_kick();
 392         }
 393 out:
 394         local_irq_restore(flags);
 395 }
 396 
 397 /* Get the boot-time nohz CPU list from the kernel parameters. */
 398 void __init tick_nohz_full_setup(cpumask_var_t cpumask)
 399 {
 400         alloc_bootmem_cpumask_var(&tick_nohz_full_mask);
 401         cpumask_copy(tick_nohz_full_mask, cpumask);
 402         tick_nohz_full_running = true;
 403 }
 404 
 405 static int tick_nohz_cpu_down(unsigned int cpu)
 406 {
 407         /*
 408          * The tick_do_timer_cpu CPU handles housekeeping duty (unbound
 409          * timers, workqueues, timekeeping, ...) on behalf of full dynticks
 410          * CPUs. It must remain online when nohz full is enabled.
 411          */
 412         if (tick_nohz_full_running && tick_do_timer_cpu == cpu)
 413                 return -EBUSY;
 414         return 0;
 415 }
 416 
 417 void __init tick_nohz_init(void)
 418 {
 419         int cpu, ret;
 420 
 421         if (!tick_nohz_full_running)
 422                 return;
 423 
 424         /*
 425          * Full dynticks uses irq work to drive the tick rescheduling on safe
 426          * locking contexts. But then we need irq work to raise its own
 427          * interrupts to avoid circular dependency on the tick
 428          */
 429         if (!arch_irq_work_has_interrupt()) {
 430                 pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n");
 431                 cpumask_clear(tick_nohz_full_mask);
 432                 tick_nohz_full_running = false;
 433                 return;
 434         }
 435 
 436         if (IS_ENABLED(CONFIG_PM_SLEEP_SMP) &&
 437                         !IS_ENABLED(CONFIG_PM_SLEEP_SMP_NONZERO_CPU)) {
 438                 cpu = smp_processor_id();
 439 
 440                 if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) {
 441                         pr_warn("NO_HZ: Clearing %d from nohz_full range "
 442                                 "for timekeeping\n", cpu);
 443                         cpumask_clear_cpu(cpu, tick_nohz_full_mask);
 444                 }
 445         }
 446 
 447         for_each_cpu(cpu, tick_nohz_full_mask)
 448                 context_tracking_cpu_set(cpu);
 449 
 450         ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
 451                                         "kernel/nohz:predown", NULL,
 452                                         tick_nohz_cpu_down);
 453         WARN_ON(ret < 0);
 454         pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n",
 455                 cpumask_pr_args(tick_nohz_full_mask));
 456 }
 457 #endif
 458 
 459 /*
 460  * NOHZ - aka dynamic tick functionality
 461  */
 462 #ifdef CONFIG_NO_HZ_COMMON
 463 /*
 464  * NO HZ enabled ?
 465  */
 466 bool tick_nohz_enabled __read_mostly  = true;
 467 unsigned long tick_nohz_active  __read_mostly;
 468 /*
 469  * Enable / Disable tickless mode
 470  */
 471 static int __init setup_tick_nohz(char *str)
 472 {
 473         return (kstrtobool(str, &tick_nohz_enabled) == 0);
 474 }
 475 
 476 __setup("nohz=", setup_tick_nohz);
 477 
 478 bool tick_nohz_tick_stopped(void)
 479 {
 480         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 481 
 482         return ts->tick_stopped;
 483 }
 484 
 485 bool tick_nohz_tick_stopped_cpu(int cpu)
 486 {
 487         struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu);
 488 
 489         return ts->tick_stopped;
 490 }
 491 
 492 /**
 493  * tick_nohz_update_jiffies - update jiffies when idle was interrupted
 494  *
 495  * Called from interrupt entry when the CPU was idle
 496  *
 497  * In case the sched_tick was stopped on this CPU, we have to check if jiffies
 498  * must be updated. Otherwise an interrupt handler could use a stale jiffy
 499  * value. We do this unconditionally on any CPU, as we don't know whether the
 500  * CPU, which has the update task assigned is in a long sleep.
 501  */
 502 static void tick_nohz_update_jiffies(ktime_t now)
 503 {
 504         unsigned long flags;
 505 
 506         __this_cpu_write(tick_cpu_sched.idle_waketime, now);
 507 
 508         local_irq_save(flags);
 509         tick_do_update_jiffies64(now);
 510         local_irq_restore(flags);
 511 
 512         touch_softlockup_watchdog_sched();
 513 }
 514 
 515 /*
 516  * Updates the per-CPU time idle statistics counters
 517  */
 518 static void
 519 update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time)
 520 {
 521         ktime_t delta;
 522 
 523         if (ts->idle_active) {
 524                 delta = ktime_sub(now, ts->idle_entrytime);
 525                 if (nr_iowait_cpu(cpu) > 0)
 526                         ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta);
 527                 else
 528                         ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
 529                 ts->idle_entrytime = now;
 530         }
 531 
 532         if (last_update_time)
 533                 *last_update_time = ktime_to_us(now);
 534 
 535 }
 536 
 537 static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now)
 538 {
 539         update_ts_time_stats(smp_processor_id(), ts, now, NULL);
 540         ts->idle_active = 0;
 541 
 542         sched_clock_idle_wakeup_event();
 543 }
 544 
 545 static void tick_nohz_start_idle(struct tick_sched *ts)
 546 {
 547         ts->idle_entrytime = ktime_get();
 548         ts->idle_active = 1;
 549         sched_clock_idle_sleep_event();
 550 }
 551 
 552 /**
 553  * get_cpu_idle_time_us - get the total idle time of a CPU
 554  * @cpu: CPU number to query
 555  * @last_update_time: variable to store update time in. Do not update
 556  * counters if NULL.
 557  *
 558  * Return the cumulative idle time (since boot) for a given
 559  * CPU, in microseconds.
 560  *
 561  * This time is measured via accounting rather than sampling,
 562  * and is as accurate as ktime_get() is.
 563  *
 564  * This function returns -1 if NOHZ is not enabled.
 565  */
 566 u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
 567 {
 568         struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
 569         ktime_t now, idle;
 570 
 571         if (!tick_nohz_active)
 572                 return -1;
 573 
 574         now = ktime_get();
 575         if (last_update_time) {
 576                 update_ts_time_stats(cpu, ts, now, last_update_time);
 577                 idle = ts->idle_sleeptime;
 578         } else {
 579                 if (ts->idle_active && !nr_iowait_cpu(cpu)) {
 580                         ktime_t delta = ktime_sub(now, ts->idle_entrytime);
 581 
 582                         idle = ktime_add(ts->idle_sleeptime, delta);
 583                 } else {
 584                         idle = ts->idle_sleeptime;
 585                 }
 586         }
 587 
 588         return ktime_to_us(idle);
 589 
 590 }
 591 EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
 592 
 593 /**
 594  * get_cpu_iowait_time_us - get the total iowait time of a CPU
 595  * @cpu: CPU number to query
 596  * @last_update_time: variable to store update time in. Do not update
 597  * counters if NULL.
 598  *
 599  * Return the cumulative iowait time (since boot) for a given
 600  * CPU, in microseconds.
 601  *
 602  * This time is measured via accounting rather than sampling,
 603  * and is as accurate as ktime_get() is.
 604  *
 605  * This function returns -1 if NOHZ is not enabled.
 606  */
 607 u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
 608 {
 609         struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
 610         ktime_t now, iowait;
 611 
 612         if (!tick_nohz_active)
 613                 return -1;
 614 
 615         now = ktime_get();
 616         if (last_update_time) {
 617                 update_ts_time_stats(cpu, ts, now, last_update_time);
 618                 iowait = ts->iowait_sleeptime;
 619         } else {
 620                 if (ts->idle_active && nr_iowait_cpu(cpu) > 0) {
 621                         ktime_t delta = ktime_sub(now, ts->idle_entrytime);
 622 
 623                         iowait = ktime_add(ts->iowait_sleeptime, delta);
 624                 } else {
 625                         iowait = ts->iowait_sleeptime;
 626                 }
 627         }
 628 
 629         return ktime_to_us(iowait);
 630 }
 631 EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
 632 
 633 static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
 634 {
 635         hrtimer_cancel(&ts->sched_timer);
 636         hrtimer_set_expires(&ts->sched_timer, ts->last_tick);
 637 
 638         /* Forward the time to expire in the future */
 639         hrtimer_forward(&ts->sched_timer, now, tick_period);
 640 
 641         if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
 642                 hrtimer_start_expires(&ts->sched_timer,
 643                                       HRTIMER_MODE_ABS_PINNED_HARD);
 644         } else {
 645                 tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1);
 646         }
 647 
 648         /*
 649          * Reset to make sure next tick stop doesn't get fooled by past
 650          * cached clock deadline.
 651          */
 652         ts->next_tick = 0;
 653 }
 654 
 655 static inline bool local_timer_softirq_pending(void)
 656 {
 657         return local_softirq_pending() & BIT(TIMER_SOFTIRQ);
 658 }
 659 
 660 static ktime_t tick_nohz_next_event(struct tick_sched *ts, int cpu)
 661 {
 662         u64 basemono, next_tick, next_tmr, next_rcu, delta, expires;
 663         unsigned long basejiff;
 664         unsigned int seq;
 665 
 666         /* Read jiffies and the time when jiffies were updated last */
 667         do {
 668                 seq = read_seqbegin(&jiffies_lock);
 669                 basemono = last_jiffies_update;
 670                 basejiff = jiffies;
 671         } while (read_seqretry(&jiffies_lock, seq));
 672         ts->last_jiffies = basejiff;
 673         ts->timer_expires_base = basemono;
 674 
 675         /*
 676          * Keep the periodic tick, when RCU, architecture or irq_work
 677          * requests it.
 678          * Aside of that check whether the local timer softirq is
 679          * pending. If so its a bad idea to call get_next_timer_interrupt()
 680          * because there is an already expired timer, so it will request
 681          * immeditate expiry, which rearms the hardware timer with a
 682          * minimal delta which brings us back to this place
 683          * immediately. Lather, rinse and repeat...
 684          */
 685         if (rcu_needs_cpu(basemono, &next_rcu) || arch_needs_cpu() ||
 686             irq_work_needs_cpu() || local_timer_softirq_pending()) {
 687                 next_tick = basemono + TICK_NSEC;
 688         } else {
 689                 /*
 690                  * Get the next pending timer. If high resolution
 691                  * timers are enabled this only takes the timer wheel
 692                  * timers into account. If high resolution timers are
 693                  * disabled this also looks at the next expiring
 694                  * hrtimer.
 695                  */
 696                 next_tmr = get_next_timer_interrupt(basejiff, basemono);
 697                 ts->next_timer = next_tmr;
 698                 /* Take the next rcu event into account */
 699                 next_tick = next_rcu < next_tmr ? next_rcu : next_tmr;
 700         }
 701 
 702         /*
 703          * If the tick is due in the next period, keep it ticking or
 704          * force prod the timer.
 705          */
 706         delta = next_tick - basemono;
 707         if (delta <= (u64)TICK_NSEC) {
 708                 /*
 709                  * Tell the timer code that the base is not idle, i.e. undo
 710                  * the effect of get_next_timer_interrupt():
 711                  */
 712                 timer_clear_idle();
 713                 /*
 714                  * We've not stopped the tick yet, and there's a timer in the
 715                  * next period, so no point in stopping it either, bail.
 716                  */
 717                 if (!ts->tick_stopped) {
 718                         ts->timer_expires = 0;
 719                         goto out;
 720                 }
 721         }
 722 
 723         /*
 724          * If this CPU is the one which had the do_timer() duty last, we limit
 725          * the sleep time to the timekeeping max_deferment value.
 726          * Otherwise we can sleep as long as we want.
 727          */
 728         delta = timekeeping_max_deferment();
 729         if (cpu != tick_do_timer_cpu &&
 730             (tick_do_timer_cpu != TICK_DO_TIMER_NONE || !ts->do_timer_last))
 731                 delta = KTIME_MAX;
 732 
 733         /* Calculate the next expiry time */
 734         if (delta < (KTIME_MAX - basemono))
 735                 expires = basemono + delta;
 736         else
 737                 expires = KTIME_MAX;
 738 
 739         ts->timer_expires = min_t(u64, expires, next_tick);
 740 
 741 out:
 742         return ts->timer_expires;
 743 }
 744 
 745 static void tick_nohz_stop_tick(struct tick_sched *ts, int cpu)
 746 {
 747         struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev);
 748         u64 basemono = ts->timer_expires_base;
 749         u64 expires = ts->timer_expires;
 750         ktime_t tick = expires;
 751 
 752         /* Make sure we won't be trying to stop it twice in a row. */
 753         ts->timer_expires_base = 0;
 754 
 755         /*
 756          * If this CPU is the one which updates jiffies, then give up
 757          * the assignment and let it be taken by the CPU which runs
 758          * the tick timer next, which might be this CPU as well. If we
 759          * don't drop this here the jiffies might be stale and
 760          * do_timer() never invoked. Keep track of the fact that it
 761          * was the one which had the do_timer() duty last.
 762          */
 763         if (cpu == tick_do_timer_cpu) {
 764                 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
 765                 ts->do_timer_last = 1;
 766         } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) {
 767                 ts->do_timer_last = 0;
 768         }
 769 
 770         /* Skip reprogram of event if its not changed */
 771         if (ts->tick_stopped && (expires == ts->next_tick)) {
 772                 /* Sanity check: make sure clockevent is actually programmed */
 773                 if (tick == KTIME_MAX || ts->next_tick == hrtimer_get_expires(&ts->sched_timer))
 774                         return;
 775 
 776                 WARN_ON_ONCE(1);
 777                 printk_once("basemono: %llu ts->next_tick: %llu dev->next_event: %llu timer->active: %d timer->expires: %llu\n",
 778                             basemono, ts->next_tick, dev->next_event,
 779                             hrtimer_active(&ts->sched_timer), hrtimer_get_expires(&ts->sched_timer));
 780         }
 781 
 782         /*
 783          * nohz_stop_sched_tick can be called several times before
 784          * the nohz_restart_sched_tick is called. This happens when
 785          * interrupts arrive which do not cause a reschedule. In the
 786          * first call we save the current tick time, so we can restart
 787          * the scheduler tick in nohz_restart_sched_tick.
 788          */
 789         if (!ts->tick_stopped) {
 790                 calc_load_nohz_start();
 791                 quiet_vmstat();
 792 
 793                 ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
 794                 ts->tick_stopped = 1;
 795                 trace_tick_stop(1, TICK_DEP_MASK_NONE);
 796         }
 797 
 798         ts->next_tick = tick;
 799 
 800         /*
 801          * If the expiration time == KTIME_MAX, then we simply stop
 802          * the tick timer.
 803          */
 804         if (unlikely(expires == KTIME_MAX)) {
 805                 if (ts->nohz_mode == NOHZ_MODE_HIGHRES)
 806                         hrtimer_cancel(&ts->sched_timer);
 807                 return;
 808         }
 809 
 810         if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
 811                 hrtimer_start(&ts->sched_timer, tick,
 812                               HRTIMER_MODE_ABS_PINNED_HARD);
 813         } else {
 814                 hrtimer_set_expires(&ts->sched_timer, tick);
 815                 tick_program_event(tick, 1);
 816         }
 817 }
 818 
 819 static void tick_nohz_retain_tick(struct tick_sched *ts)
 820 {
 821         ts->timer_expires_base = 0;
 822 }
 823 
 824 #ifdef CONFIG_NO_HZ_FULL
 825 static void tick_nohz_stop_sched_tick(struct tick_sched *ts, int cpu)
 826 {
 827         if (tick_nohz_next_event(ts, cpu))
 828                 tick_nohz_stop_tick(ts, cpu);
 829         else
 830                 tick_nohz_retain_tick(ts);
 831 }
 832 #endif /* CONFIG_NO_HZ_FULL */
 833 
 834 static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now)
 835 {
 836         /* Update jiffies first */
 837         tick_do_update_jiffies64(now);
 838 
 839         /*
 840          * Clear the timer idle flag, so we avoid IPIs on remote queueing and
 841          * the clock forward checks in the enqueue path:
 842          */
 843         timer_clear_idle();
 844 
 845         calc_load_nohz_stop();
 846         touch_softlockup_watchdog_sched();
 847         /*
 848          * Cancel the scheduled timer and restore the tick
 849          */
 850         ts->tick_stopped  = 0;
 851         ts->idle_exittime = now;
 852 
 853         tick_nohz_restart(ts, now);
 854 }
 855 
 856 static void tick_nohz_full_update_tick(struct tick_sched *ts)
 857 {
 858 #ifdef CONFIG_NO_HZ_FULL
 859         int cpu = smp_processor_id();
 860 
 861         if (!tick_nohz_full_cpu(cpu))
 862                 return;
 863 
 864         if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE)
 865                 return;
 866 
 867         if (can_stop_full_tick(cpu, ts))
 868                 tick_nohz_stop_sched_tick(ts, cpu);
 869         else if (ts->tick_stopped)
 870                 tick_nohz_restart_sched_tick(ts, ktime_get());
 871 #endif
 872 }
 873 
 874 static bool can_stop_idle_tick(int cpu, struct tick_sched *ts)
 875 {
 876         /*
 877          * If this CPU is offline and it is the one which updates
 878          * jiffies, then give up the assignment and let it be taken by
 879          * the CPU which runs the tick timer next. If we don't drop
 880          * this here the jiffies might be stale and do_timer() never
 881          * invoked.
 882          */
 883         if (unlikely(!cpu_online(cpu))) {
 884                 if (cpu == tick_do_timer_cpu)
 885                         tick_do_timer_cpu = TICK_DO_TIMER_NONE;
 886                 /*
 887                  * Make sure the CPU doesn't get fooled by obsolete tick
 888                  * deadline if it comes back online later.
 889                  */
 890                 ts->next_tick = 0;
 891                 return false;
 892         }
 893 
 894         if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE))
 895                 return false;
 896 
 897         if (need_resched())
 898                 return false;
 899 
 900         if (unlikely(local_softirq_pending())) {
 901                 static int ratelimit;
 902 
 903                 if (ratelimit < 10 &&
 904                     (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) {
 905                         pr_warn("NOHZ: local_softirq_pending %02x\n",
 906                                 (unsigned int) local_softirq_pending());
 907                         ratelimit++;
 908                 }
 909                 return false;
 910         }
 911 
 912         if (tick_nohz_full_enabled()) {
 913                 /*
 914                  * Keep the tick alive to guarantee timekeeping progression
 915                  * if there are full dynticks CPUs around
 916                  */
 917                 if (tick_do_timer_cpu == cpu)
 918                         return false;
 919                 /*
 920                  * Boot safety: make sure the timekeeping duty has been
 921                  * assigned before entering dyntick-idle mode,
 922                  * tick_do_timer_cpu is TICK_DO_TIMER_BOOT
 923                  */
 924                 if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_BOOT))
 925                         return false;
 926 
 927                 /* Should not happen for nohz-full */
 928                 if (WARN_ON_ONCE(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
 929                         return false;
 930         }
 931 
 932         return true;
 933 }
 934 
 935 static void __tick_nohz_idle_stop_tick(struct tick_sched *ts)
 936 {
 937         ktime_t expires;
 938         int cpu = smp_processor_id();
 939 
 940         /*
 941          * If tick_nohz_get_sleep_length() ran tick_nohz_next_event(), the
 942          * tick timer expiration time is known already.
 943          */
 944         if (ts->timer_expires_base)
 945                 expires = ts->timer_expires;
 946         else if (can_stop_idle_tick(cpu, ts))
 947                 expires = tick_nohz_next_event(ts, cpu);
 948         else
 949                 return;
 950 
 951         ts->idle_calls++;
 952 
 953         if (expires > 0LL) {
 954                 int was_stopped = ts->tick_stopped;
 955 
 956                 tick_nohz_stop_tick(ts, cpu);
 957 
 958                 ts->idle_sleeps++;
 959                 ts->idle_expires = expires;
 960 
 961                 if (!was_stopped && ts->tick_stopped) {
 962                         ts->idle_jiffies = ts->last_jiffies;
 963                         nohz_balance_enter_idle(cpu);
 964                 }
 965         } else {
 966                 tick_nohz_retain_tick(ts);
 967         }
 968 }
 969 
 970 /**
 971  * tick_nohz_idle_stop_tick - stop the idle tick from the idle task
 972  *
 973  * When the next event is more than a tick into the future, stop the idle tick
 974  */
 975 void tick_nohz_idle_stop_tick(void)
 976 {
 977         __tick_nohz_idle_stop_tick(this_cpu_ptr(&tick_cpu_sched));
 978 }
 979 
 980 void tick_nohz_idle_retain_tick(void)
 981 {
 982         tick_nohz_retain_tick(this_cpu_ptr(&tick_cpu_sched));
 983         /*
 984          * Undo the effect of get_next_timer_interrupt() called from
 985          * tick_nohz_next_event().
 986          */
 987         timer_clear_idle();
 988 }
 989 
 990 /**
 991  * tick_nohz_idle_enter - prepare for entering idle on the current CPU
 992  *
 993  * Called when we start the idle loop.
 994  */
 995 void tick_nohz_idle_enter(void)
 996 {
 997         struct tick_sched *ts;
 998 
 999         lockdep_assert_irqs_enabled();
1000 
1001         local_irq_disable();
1002 
1003         ts = this_cpu_ptr(&tick_cpu_sched);
1004 
1005         WARN_ON_ONCE(ts->timer_expires_base);
1006 
1007         ts->inidle = 1;
1008         tick_nohz_start_idle(ts);
1009 
1010         local_irq_enable();
1011 }
1012 
1013 /**
1014  * tick_nohz_irq_exit - update next tick event from interrupt exit
1015  *
1016  * When an interrupt fires while we are idle and it doesn't cause
1017  * a reschedule, it may still add, modify or delete a timer, enqueue
1018  * an RCU callback, etc...
1019  * So we need to re-calculate and reprogram the next tick event.
1020  */
1021 void tick_nohz_irq_exit(void)
1022 {
1023         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1024 
1025         if (ts->inidle)
1026                 tick_nohz_start_idle(ts);
1027         else
1028                 tick_nohz_full_update_tick(ts);
1029 }
1030 
1031 /**
1032  * tick_nohz_idle_got_tick - Check whether or not the tick handler has run
1033  */
1034 bool tick_nohz_idle_got_tick(void)
1035 {
1036         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1037 
1038         if (ts->got_idle_tick) {
1039                 ts->got_idle_tick = 0;
1040                 return true;
1041         }
1042         return false;
1043 }
1044 
1045 /**
1046  * tick_nohz_get_next_hrtimer - return the next expiration time for the hrtimer
1047  * or the tick, whatever that expires first. Note that, if the tick has been
1048  * stopped, it returns the next hrtimer.
1049  *
1050  * Called from power state control code with interrupts disabled
1051  */
1052 ktime_t tick_nohz_get_next_hrtimer(void)
1053 {
1054         return __this_cpu_read(tick_cpu_device.evtdev)->next_event;
1055 }
1056 
1057 /**
1058  * tick_nohz_get_sleep_length - return the expected length of the current sleep
1059  * @delta_next: duration until the next event if the tick cannot be stopped
1060  *
1061  * Called from power state control code with interrupts disabled
1062  */
1063 ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next)
1064 {
1065         struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev);
1066         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1067         int cpu = smp_processor_id();
1068         /*
1069          * The idle entry time is expected to be a sufficient approximation of
1070          * the current time at this point.
1071          */
1072         ktime_t now = ts->idle_entrytime;
1073         ktime_t next_event;
1074 
1075         WARN_ON_ONCE(!ts->inidle);
1076 
1077         *delta_next = ktime_sub(dev->next_event, now);
1078 
1079         if (!can_stop_idle_tick(cpu, ts))
1080                 return *delta_next;
1081 
1082         next_event = tick_nohz_next_event(ts, cpu);
1083         if (!next_event)
1084                 return *delta_next;
1085 
1086         /*
1087          * If the next highres timer to expire is earlier than next_event, the
1088          * idle governor needs to know that.
1089          */
1090         next_event = min_t(u64, next_event,
1091                            hrtimer_next_event_without(&ts->sched_timer));
1092 
1093         return ktime_sub(next_event, now);
1094 }
1095 
1096 /**
1097  * tick_nohz_get_idle_calls_cpu - return the current idle calls counter value
1098  * for a particular CPU.
1099  *
1100  * Called from the schedutil frequency scaling governor in scheduler context.
1101  */
1102 unsigned long tick_nohz_get_idle_calls_cpu(int cpu)
1103 {
1104         struct tick_sched *ts = tick_get_tick_sched(cpu);
1105 
1106         return ts->idle_calls;
1107 }
1108 
1109 /**
1110  * tick_nohz_get_idle_calls - return the current idle calls counter value
1111  *
1112  * Called from the schedutil frequency scaling governor in scheduler context.
1113  */
1114 unsigned long tick_nohz_get_idle_calls(void)
1115 {
1116         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1117 
1118         return ts->idle_calls;
1119 }
1120 
1121 static void tick_nohz_account_idle_ticks(struct tick_sched *ts)
1122 {
1123 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1124         unsigned long ticks;
1125 
1126         if (vtime_accounting_cpu_enabled())
1127                 return;
1128         /*
1129          * We stopped the tick in idle. Update process times would miss the
1130          * time we slept as update_process_times does only a 1 tick
1131          * accounting. Enforce that this is accounted to idle !
1132          */
1133         ticks = jiffies - ts->idle_jiffies;
1134         /*
1135          * We might be one off. Do not randomly account a huge number of ticks!
1136          */
1137         if (ticks && ticks < LONG_MAX)
1138                 account_idle_ticks(ticks);
1139 #endif
1140 }
1141 
1142 static void __tick_nohz_idle_restart_tick(struct tick_sched *ts, ktime_t now)
1143 {
1144         tick_nohz_restart_sched_tick(ts, now);
1145         tick_nohz_account_idle_ticks(ts);
1146 }
1147 
1148 void tick_nohz_idle_restart_tick(void)
1149 {
1150         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1151 
1152         if (ts->tick_stopped)
1153                 __tick_nohz_idle_restart_tick(ts, ktime_get());
1154 }
1155 
1156 /**
1157  * tick_nohz_idle_exit - restart the idle tick from the idle task
1158  *
1159  * Restart the idle tick when the CPU is woken up from idle
1160  * This also exit the RCU extended quiescent state. The CPU
1161  * can use RCU again after this function is called.
1162  */
1163 void tick_nohz_idle_exit(void)
1164 {
1165         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1166         bool idle_active, tick_stopped;
1167         ktime_t now;
1168 
1169         local_irq_disable();
1170 
1171         WARN_ON_ONCE(!ts->inidle);
1172         WARN_ON_ONCE(ts->timer_expires_base);
1173 
1174         ts->inidle = 0;
1175         idle_active = ts->idle_active;
1176         tick_stopped = ts->tick_stopped;
1177 
1178         if (idle_active || tick_stopped)
1179                 now = ktime_get();
1180 
1181         if (idle_active)
1182                 tick_nohz_stop_idle(ts, now);
1183 
1184         if (tick_stopped)
1185                 __tick_nohz_idle_restart_tick(ts, now);
1186 
1187         local_irq_enable();
1188 }
1189 
1190 /*
1191  * The nohz low res interrupt handler
1192  */
1193 static void tick_nohz_handler(struct clock_event_device *dev)
1194 {
1195         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1196         struct pt_regs *regs = get_irq_regs();
1197         ktime_t now = ktime_get();
1198 
1199         dev->next_event = KTIME_MAX;
1200 
1201         tick_sched_do_timer(ts, now);
1202         tick_sched_handle(ts, regs);
1203 
1204         /* No need to reprogram if we are running tickless  */
1205         if (unlikely(ts->tick_stopped))
1206                 return;
1207 
1208         hrtimer_forward(&ts->sched_timer, now, tick_period);
1209         tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1);
1210 }
1211 
1212 static inline void tick_nohz_activate(struct tick_sched *ts, int mode)
1213 {
1214         if (!tick_nohz_enabled)
1215                 return;
1216         ts->nohz_mode = mode;
1217         /* One update is enough */
1218         if (!test_and_set_bit(0, &tick_nohz_active))
1219                 timers_update_nohz();
1220 }
1221 
1222 /**
1223  * tick_nohz_switch_to_nohz - switch to nohz mode
1224  */
1225 static void tick_nohz_switch_to_nohz(void)
1226 {
1227         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1228         ktime_t next;
1229 
1230         if (!tick_nohz_enabled)
1231                 return;
1232 
1233         if (tick_switch_to_oneshot(tick_nohz_handler))
1234                 return;
1235 
1236         /*
1237          * Recycle the hrtimer in ts, so we can share the
1238          * hrtimer_forward with the highres code.
1239          */
1240         hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD);
1241         /* Get the next period */
1242         next = tick_init_jiffy_update();
1243 
1244         hrtimer_set_expires(&ts->sched_timer, next);
1245         hrtimer_forward_now(&ts->sched_timer, tick_period);
1246         tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1);
1247         tick_nohz_activate(ts, NOHZ_MODE_LOWRES);
1248 }
1249 
1250 static inline void tick_nohz_irq_enter(void)
1251 {
1252         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1253         ktime_t now;
1254 
1255         if (!ts->idle_active && !ts->tick_stopped)
1256                 return;
1257         now = ktime_get();
1258         if (ts->idle_active)
1259                 tick_nohz_stop_idle(ts, now);
1260         if (ts->tick_stopped)
1261                 tick_nohz_update_jiffies(now);
1262 }
1263 
1264 #else
1265 
1266 static inline void tick_nohz_switch_to_nohz(void) { }
1267 static inline void tick_nohz_irq_enter(void) { }
1268 static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { }
1269 
1270 #endif /* CONFIG_NO_HZ_COMMON */
1271 
1272 /*
1273  * Called from irq_enter to notify about the possible interruption of idle()
1274  */
1275 void tick_irq_enter(void)
1276 {
1277         tick_check_oneshot_broadcast_this_cpu();
1278         tick_nohz_irq_enter();
1279 }
1280 
1281 /*
1282  * High resolution timer specific code
1283  */
1284 #ifdef CONFIG_HIGH_RES_TIMERS
1285 /*
1286  * We rearm the timer until we get disabled by the idle code.
1287  * Called with interrupts disabled.
1288  */
1289 static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer)
1290 {
1291         struct tick_sched *ts =
1292                 container_of(timer, struct tick_sched, sched_timer);
1293         struct pt_regs *regs = get_irq_regs();
1294         ktime_t now = ktime_get();
1295 
1296         tick_sched_do_timer(ts, now);
1297 
1298         /*
1299          * Do not call, when we are not in irq context and have
1300          * no valid regs pointer
1301          */
1302         if (regs)
1303                 tick_sched_handle(ts, regs);
1304         else
1305                 ts->next_tick = 0;
1306 
1307         /* No need to reprogram if we are in idle or full dynticks mode */
1308         if (unlikely(ts->tick_stopped))
1309                 return HRTIMER_NORESTART;
1310 
1311         hrtimer_forward(timer, now, tick_period);
1312 
1313         return HRTIMER_RESTART;
1314 }
1315 
1316 static int sched_skew_tick;
1317 
1318 static int __init skew_tick(char *str)
1319 {
1320         get_option(&str, &sched_skew_tick);
1321 
1322         return 0;
1323 }
1324 early_param("skew_tick", skew_tick);
1325 
1326 /**
1327  * tick_setup_sched_timer - setup the tick emulation timer
1328  */
1329 void tick_setup_sched_timer(void)
1330 {
1331         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1332         ktime_t now = ktime_get();
1333 
1334         /*
1335          * Emulate tick processing via per-CPU hrtimers:
1336          */
1337         hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD);
1338         ts->sched_timer.function = tick_sched_timer;
1339 
1340         /* Get the next period (per-CPU) */
1341         hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update());
1342 
1343         /* Offset the tick to avert jiffies_lock contention. */
1344         if (sched_skew_tick) {
1345                 u64 offset = ktime_to_ns(tick_period) >> 1;
1346                 do_div(offset, num_possible_cpus());
1347                 offset *= smp_processor_id();
1348                 hrtimer_add_expires_ns(&ts->sched_timer, offset);
1349         }
1350 
1351         hrtimer_forward(&ts->sched_timer, now, tick_period);
1352         hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED_HARD);
1353         tick_nohz_activate(ts, NOHZ_MODE_HIGHRES);
1354 }
1355 #endif /* HIGH_RES_TIMERS */
1356 
1357 #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS
1358 void tick_cancel_sched_timer(int cpu)
1359 {
1360         struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
1361 
1362 # ifdef CONFIG_HIGH_RES_TIMERS
1363         if (ts->sched_timer.base)
1364                 hrtimer_cancel(&ts->sched_timer);
1365 # endif
1366 
1367         memset(ts, 0, sizeof(*ts));
1368 }
1369 #endif
1370 
1371 /**
1372  * Async notification about clocksource changes
1373  */
1374 void tick_clock_notify(void)
1375 {
1376         int cpu;
1377 
1378         for_each_possible_cpu(cpu)
1379                 set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks);
1380 }
1381 
1382 /*
1383  * Async notification about clock event changes
1384  */
1385 void tick_oneshot_notify(void)
1386 {
1387         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1388 
1389         set_bit(0, &ts->check_clocks);
1390 }
1391 
1392 /**
1393  * Check, if a change happened, which makes oneshot possible.
1394  *
1395  * Called cyclic from the hrtimer softirq (driven by the timer
1396  * softirq) allow_nohz signals, that we can switch into low-res nohz
1397  * mode, because high resolution timers are disabled (either compile
1398  * or runtime). Called with interrupts disabled.
1399  */
1400 int tick_check_oneshot_change(int allow_nohz)
1401 {
1402         struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1403 
1404         if (!test_and_clear_bit(0, &ts->check_clocks))
1405                 return 0;
1406 
1407         if (ts->nohz_mode != NOHZ_MODE_INACTIVE)
1408                 return 0;
1409 
1410         if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available())
1411                 return 0;
1412 
1413         if (!allow_nohz)
1414                 return 1;
1415 
1416         tick_nohz_switch_to_nohz();
1417         return 0;
1418 }
/* [<][>][^][v][top][bottom][index][help] */
root/kernel/time/tick-sched.c

DEFINITIONS
