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/tick.h>
   8 #include <linux/mm.h>
   9 #include <linux/stackprotector.h>
  10 #include <linux/suspend.h>
  11
  12 #include <asm/tlb.h>
  13
  14 #include <trace/events/power.h>
  15
  16 #include "sched.h"
  17
  18 /**
  19  * sched_idle_set_state - Record idle state for the current CPU.
  20  * @idle_state: State to record.
  21  */
  22 void sched_idle_set_state(struct cpuidle_state *idle_state)
  23 {
  24         idle_set_state(this_rq(), idle_state);
  25 }
  26
  27 static int __read_mostly cpu_idle_force_poll;
  28
  29 void cpu_idle_poll_ctrl(bool enable)
  30 {
  31         if (enable) {
  32                 cpu_idle_force_poll++;
  33         } else {
  34                 cpu_idle_force_poll--;
  35                 WARN_ON_ONCE(cpu_idle_force_poll < 0);
  36         }
  37 }
  38
  39 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
  40 static int __init cpu_idle_poll_setup(char *__unused)
  41 {
  42         cpu_idle_force_poll = 1;
  43         return 1;
  44 }
  45 __setup("nohlt", cpu_idle_poll_setup);
  46
  47 static int __init cpu_idle_nopoll_setup(char *__unused)
  48 {
  49         cpu_idle_force_poll = 0;
  50         return 1;
  51 }
  52 __setup("hlt", cpu_idle_nopoll_setup);
  53 #endif
  54
  55 static inline int cpu_idle_poll(void)
  56 {
  57         rcu_idle_enter();
  58         trace_cpu_idle_rcuidle(0, smp_processor_id());
  59         local_irq_enable();
  60         while (!tif_need_resched() &&
  61                 (cpu_idle_force_poll || tick_check_broadcast_expired()))
  62                 cpu_relax();
  63         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
  64         rcu_idle_exit();
  65         return 1;
  66 }
  67
  68 /* Weak implementations for optional arch specific functions */
  69 void __weak arch_cpu_idle_prepare(void) { }
  70 void __weak arch_cpu_idle_enter(void) { }
  71 void __weak arch_cpu_idle_exit(void) { }
  72 void __weak arch_cpu_idle_dead(void) { }
  73 void __weak arch_cpu_idle(void)
  74 {
  75         cpu_idle_force_poll = 1;
  76         local_irq_enable();
  77 }
  78
  79 /**
  80  * default_idle_call - Default CPU idle routine.
  81  *
  82  * To use when the cpuidle framework cannot be used.
  83  */
  84 void default_idle_call(void)
  85 {
  86         if (current_clr_polling_and_test()) {
  87                 local_irq_enable();
  88         } else {
  89                 stop_critical_timings();
  90                 arch_cpu_idle();
  91                 start_critical_timings();
  92         }
  93 }
  94
  95 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
  96                       int next_state)
  97 {
  98         /* Fall back to the default arch idle method on errors. */
  99         if (next_state < 0) {
 100                 default_idle_call();
 101                 return next_state;
 102         }
 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 = __this_cpu_read(cpuidle_devices);
 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         if (idle_should_freeze()) {
 168                 entered_state = cpuidle_enter_freeze(drv, dev);
 169                 if (entered_state >= 0) {
 170                         local_irq_enable();
 171                         goto exit_idle;
 172                 }
 173
 174                 next_state = cpuidle_find_deepest_state(drv, dev);
 175                 call_cpuidle(drv, dev, next_state);
 176         } else {
 177                 /*
 178                  * Ask the cpuidle framework to choose a convenient idle state.
 179                  */
 180                 next_state = cpuidle_select(drv, dev);
 181                 entered_state = call_cpuidle(drv, dev, next_state);
 182                 /*
 183                  * Give the governor an opportunity to reflect on the outcome
 184                  */
 185                 cpuidle_reflect(dev, entered_state);
 186         }
 187
 188 exit_idle:
 189         __current_set_polling();
 190
 191         /*
 192          * It is up to the idle functions to reenable local interrupts
 193          */
 194         if (WARN_ON_ONCE(irqs_disabled()))
 195                 local_irq_enable();
 196
 197         rcu_idle_exit();
 198 }
 199
 200 DEFINE_PER_CPU(bool, cpu_dead_idle);
 201
 202 /*
 203  * Generic idle loop implementation
 204  *
 205  * Called with polling cleared.
 206  */
 207 static void cpu_idle_loop(void)
 208 {
 209         while (1) {
 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
 214                  * rq->curr != rq->idle).  This means that, if rq->idle has
 215                  * the polling bit set, then setting need_resched is
 216                  * guaranteed to cause the cpu to 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                                 rcu_cpu_notify(NULL, CPU_DYING_IDLE,
 228                                                (void *)(long)smp_processor_id());
 229                                 smp_mb(); /* all activity before dead. */
 230                                 this_cpu_write(cpu_dead_idle, true);
 231                                 arch_cpu_idle_dead();
 232                         }
 233
 234                         local_irq_disable();
 235                         arch_cpu_idle_enter();
 236
 237                         /*
 238                          * In poll mode we reenable interrupts and spin.
 239                          *
 240                          * Also if we detected in the wakeup from idle
 241                          * path that the tick broadcast device expired
 242                          * for us, we don't want to go deep idle as we
 243                          * know that the IPI is going to arrive right
 244                          * away
 245                          */
 246                         if (cpu_idle_force_poll || tick_check_broadcast_expired())
 247                                 cpu_idle_poll();
 248                         else
 249                                 cpuidle_idle_call();
 250
 251                         arch_cpu_idle_exit();
 252                 }
 253
 254                 /*
 255                  * Since we fell out of the loop above, we know
 256                  * TIF_NEED_RESCHED must be set, propagate it into
 257                  * PREEMPT_NEED_RESCHED.
 258                  *
 259                  * This is required because for polling idle loops we will
 260                  * not have had an IPI to fold the state for us.
 261                  */
 262                 preempt_set_need_resched();
 263                 tick_nohz_idle_exit();
 264                 __current_clr_polling();
 265
 266                 /*
 267                  * We promise to call sched_ttwu_pending and reschedule
 268                  * if need_resched is set while polling is set.  That
 269                  * means that clearing polling needs to be visible
 270                  * before doing these things.
 271                  */
 272                 smp_mb__after_atomic();
 273
 274                 sched_ttwu_pending();
 275                 schedule_preempt_disabled();
 276         }
 277 }
 278
 279 void cpu_startup_entry(enum cpuhp_state state)
 280 {
 281         /*
 282          * This #ifdef needs to die, but it's too late in the cycle to
 283          * make this generic (arm and sh have never invoked the canary
 284          * init for the non boot cpus!). Will be fixed in 3.11
 285          */
 286 #ifdef CONFIG_X86
 287         /*
 288          * If we're the non-boot CPU, nothing set the stack canary up
 289          * for us. The boot CPU already has it initialized but no harm
 290          * in doing it again. This is a good place for updating it, as
 291          * we wont ever return from this function (so the invalid
 292          * canaries already on the stack wont ever trigger).
 293          */
 294         boot_init_stack_canary();
 295 #endif
 296         arch_cpu_idle_prepare();
 297         cpu_idle_loop();
 298 }