kernel/sched/idle.c

   1 /*
   2  * Generic entry point for the idle threads
   3  */
   4 #include <linux/sched.h>
   5 #include <linux/cpu.h>
   6 #include <linux/cpuidle.h>
   7 #include <linux/cpuhotplug.h>
   8 #include <linux/tick.h>
   9 #include <linux/mm.h>
  10 #include <linux/stackprotector.h>
  11 #include <linux/suspend.h>
  12
  13 #include <asm/tlb.h>
  14
  15 #include <trace/events/power.h>
  16
  17 #include "sched.h"
  18
  19 /* Linker adds these: start and end of __cpuidle functions */
  20 extern char __cpuidle_text_start[], __cpuidle_text_end[];
  21
  22 /**
  23  * sched_idle_set_state - Record idle state for the current CPU.
  24  * @idle_state: State to record.
  25  */
  26 void sched_idle_set_state(struct cpuidle_state *idle_state)
  27 {
  28         idle_set_state(this_rq(), idle_state);
  29 }
  30
  31 static int __read_mostly cpu_idle_force_poll;
  32
  33 void cpu_idle_poll_ctrl(bool enable)
  34 {
  35         if (enable) {
  36                 cpu_idle_force_poll++;
  37         } else {
  38                 cpu_idle_force_poll--;
  39                 WARN_ON_ONCE(cpu_idle_force_poll < 0);
  40         }
  41 }
  42
  43 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
  44 static int __init cpu_idle_poll_setup(char *__unused)
  45 {
  46         cpu_idle_force_poll = 1;
  47         return 1;
  48 }
  49 __setup("nohlt", cpu_idle_poll_setup);
  50
  51 static int __init cpu_idle_nopoll_setup(char *__unused)
  52 {
  53         cpu_idle_force_poll = 0;
  54         return 1;
  55 }
  56 __setup("hlt", cpu_idle_nopoll_setup);
  57 #endif
  58
  59 static noinline int __cpuidle cpu_idle_poll(void)
  60 {
  61         rcu_idle_enter();
  62         trace_cpu_idle_rcuidle(0, smp_processor_id());
  63         local_irq_enable();
  64         stop_critical_timings();
  65         while (!tif_need_resched() &&
  66                 (cpu_idle_force_poll || tick_check_broadcast_expired()))
  67                 cpu_relax();
  68         start_critical_timings();
  69         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
  70         rcu_idle_exit();
  71         return 1;
  72 }
  73
  74 /* Weak implementations for optional arch specific functions */
  75 void __weak arch_cpu_idle_prepare(void) { }
  76 void __weak arch_cpu_idle_enter(void) { }
  77 void __weak arch_cpu_idle_exit(void) { }
  78 void __weak arch_cpu_idle_dead(void) { }
  79 void __weak arch_cpu_idle(void)
  80 {
  81         cpu_idle_force_poll = 1;
  82         local_irq_enable();
  83 }
  84
  85 /**
  86  * default_idle_call - Default CPU idle routine.
  87  *
  88  * To use when the cpuidle framework cannot be used.
  89  */
  90 void __cpuidle default_idle_call(void)
  91 {
  92         if (current_clr_polling_and_test()) {
  93                 local_irq_enable();
  94         } else {
  95                 stop_critical_timings();
  96                 arch_cpu_idle();
  97                 start_critical_timings();
  98         }
  99 }
 100
 101 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 102                       int next_state)
 103 {
 104         /*
 105          * The idle task must be scheduled, it is pointless to go to idle, just
 106          * update no idle residency and return.
 107          */
 108         if (current_clr_polling_and_test()) {
 109                 dev->last_residency = 0;
 110                 local_irq_enable();
 111                 return -EBUSY;
 112         }
 113
 114         /*
 115          * Enter the idle state previously returned by the governor decision.
 116          * This function will block until an interrupt occurs and will take
 117          * care of re-enabling the local interrupts
 118          */
 119         return cpuidle_enter(drv, dev, next_state);
 120 }
 121
 122 /**
 123  * cpuidle_idle_call - the main idle function
 124  *
 125  * NOTE: no locks or semaphores should be used here
 126  *
 127  * On archs that support TIF_POLLING_NRFLAG, is called with polling
 128  * set, and it returns with polling set.  If it ever stops polling, it
 129  * must clear the polling bit.
 130  */
 131 static void cpuidle_idle_call(void)
 132 {
 133         struct cpuidle_device *dev = cpuidle_get_device();
 134         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 135         int next_state, entered_state;
 136
 137         /*
 138          * Check if the idle task must be rescheduled. If it is the
 139          * case, exit the function after re-enabling the local irq.
 140          */
 141         if (need_resched()) {
 142                 local_irq_enable();
 143                 return;
 144         }
 145
 146         /*
 147          * Tell the RCU framework we are entering an idle section,
 148          * so no more rcu read side critical sections and one more
 149          * step to the grace period
 150          */
 151         rcu_idle_enter();
 152
 153         if (cpuidle_not_available(drv, dev)) {
 154                 default_idle_call();
 155                 goto exit_idle;
 156         }
 157
 158         /*
 159          * Suspend-to-idle ("freeze") is a system state in which all user space
 160          * has been frozen, all I/O devices have been suspended and the only
 161          * activity happens here and in iterrupts (if any).  In that case bypass
 162          * the cpuidle governor and go stratight for the deepest idle state
 163          * available.  Possibly also suspend the local tick and the entire
 164          * timekeeping to prevent timer interrupts from kicking us out of idle
 165          * until a proper wakeup interrupt happens.
 166          */
 167
 168         if (idle_should_freeze() || dev->use_deepest_state) {
 169                 if (idle_should_freeze()) {
 170                         entered_state = cpuidle_enter_freeze(drv, dev);
 171                         if (entered_state > 0) {
 172                                 local_irq_enable();
 173                                 goto exit_idle;
 174                         }
 175                 }
 176
 177                 next_state = cpuidle_find_deepest_state(drv, dev);
 178                 call_cpuidle(drv, dev, next_state);
 179         } else {
 180                 /*
 181                  * Ask the cpuidle framework to choose a convenient idle state.
 182                  */
 183                 next_state = cpuidle_select(drv, dev);
 184                 entered_state = call_cpuidle(drv, dev, next_state);
 185                 /*
 186                  * Give the governor an opportunity to reflect on the outcome
 187                  */
 188                 cpuidle_reflect(dev, entered_state);
 189         }
 190
 191 exit_idle:
 192         __current_set_polling();
 193
 194         /*
 195          * It is up to the idle functions to reenable local interrupts
 196          */
 197         if (WARN_ON_ONCE(irqs_disabled()))
 198                 local_irq_enable();
 199
 200         rcu_idle_exit();
 201 }
 202
 203 /*
 204  * Generic idle loop implementation
 205  *
 206  * Called with polling cleared.
 207  */
 208 static void do_idle(void)
 209 {
 210         /*
 211          * If the arch has a polling bit, we maintain an invariant:
 212          *
 213          * Our polling bit is clear if we're not scheduled (i.e. if rq->curr !=
 214          * rq->idle). This means that, if rq->idle has the polling bit set,
 215          * then setting need_resched is guaranteed to cause the CPU to
 216          * reschedule.
 217          */
 218
 219         __current_set_polling();
 220         tick_nohz_idle_enter();
 221
 222         while (!need_resched()) {
 223                 check_pgt_cache();
 224                 rmb();
 225
 226                 if (cpu_is_offline(smp_processor_id())) {
 227                         cpuhp_report_idle_dead();
 228                         arch_cpu_idle_dead();
 229                 }
 230
 231                 local_irq_disable();
 232                 arch_cpu_idle_enter();
 233
 234                 /*
 235                  * In poll mode we reenable interrupts and spin. Also if we
 236                  * detected in the wakeup from idle path that the tick
 237                  * broadcast device expired for us, we don't want to go deep
 238                  * idle as we know that the IPI is going to arrive right away.
 239                  */
 240                 if (cpu_idle_force_poll || tick_check_broadcast_expired())
 241                         cpu_idle_poll();
 242                 else
 243                         cpuidle_idle_call();
 244                 arch_cpu_idle_exit();
 245         }
 246
 247         /*
 248          * Since we fell out of the loop above, we know TIF_NEED_RESCHED must
 249          * be set, propagate it into PREEMPT_NEED_RESCHED.
 250          *
 251          * This is required because for polling idle loops we will not have had
 252          * an IPI to fold the state for us.
 253          */
 254         preempt_set_need_resched();
 255         tick_nohz_idle_exit();
 256         __current_clr_polling();
 257
 258         /*
 259          * We promise to call sched_ttwu_pending() and reschedule if
 260          * need_resched() is set while polling is set. That means that clearing
 261          * polling needs to be visible before doing these things.
 262          */
 263         smp_mb__after_atomic();
 264
 265         sched_ttwu_pending();
 266         schedule_preempt_disabled();
 267 }
 268
 269 bool cpu_in_idle(unsigned long pc)
 270 {
 271         return pc >= (unsigned long)__cpuidle_text_start &&
 272                 pc < (unsigned long)__cpuidle_text_end;
 273 }
 274
 275 struct idle_timer {
 276         struct hrtimer timer;
 277         int done;
 278 };
 279
 280 static enum hrtimer_restart idle_inject_timer_fn(struct hrtimer *timer)
 281 {
 282         struct idle_timer *it = container_of(timer, struct idle_timer, timer);
 283
 284         WRITE_ONCE(it->done, 1);
 285         set_tsk_need_resched(current);
 286
 287         return HRTIMER_NORESTART;
 288 }
 289
 290 void play_idle(unsigned long duration_ms)
 291 {
 292         struct idle_timer it;
 293
 294         /*
 295          * Only FIFO tasks can disable the tick since they don't need the forced
 296          * preemption.
 297          */
 298         WARN_ON_ONCE(current->policy != SCHED_FIFO);
 299         WARN_ON_ONCE(current->nr_cpus_allowed != 1);
 300         WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
 301         WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
 302         WARN_ON_ONCE(!duration_ms);
 303
 304         rcu_sleep_check();
 305         preempt_disable();
 306         current->flags |= PF_IDLE;
 307         cpuidle_use_deepest_state(true);
 308
 309         it.done = 0;
 310         hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 311         it.timer.function = idle_inject_timer_fn;
 312         hrtimer_start(&it.timer, ms_to_ktime(duration_ms), HRTIMER_MODE_REL_PINNED);
 313
 314         while (!READ_ONCE(it.done))
 315                 do_idle();
 316
 317         cpuidle_use_deepest_state(false);
 318         current->flags &= ~PF_IDLE;
 319
 320         preempt_fold_need_resched();
 321         preempt_enable();
 322 }
 323 EXPORT_SYMBOL_GPL(play_idle);
 324
 325 void cpu_startup_entry(enum cpuhp_state state)
 326 {
 327         /*
 328          * This #ifdef needs to die, but it's too late in the cycle to
 329          * make this generic (arm and sh have never invoked the canary
 330          * init for the non boot cpus!). Will be fixed in 3.11
 331          */
 332 #ifdef CONFIG_X86
 333         /*
 334          * If we're the non-boot CPU, nothing set the stack canary up
 335          * for us. The boot CPU already has it initialized but no harm
 336          * in doing it again. This is a good place for updating it, as
 337          * we wont ever return from this function (so the invalid
 338          * canaries already on the stack wont ever trigger).
 339          */
 340         boot_init_stack_canary();
 341 #endif
 342         arch_cpu_idle_prepare();
 343         cpuhp_online_idle(state);
 344         while (1)
 345                 do_idle();
 346 }