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                 arch_cpu_idle();
  90 }
  91
  92 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
  93                       int next_state)
  94 {
  95         /* Fall back to the default arch idle method on errors. */
  96         if (next_state < 0) {
  97                 default_idle_call();
  98                 return next_state;
  99         }
 100
 101         /*
 102          * The idle task must be scheduled, it is pointless to go to idle, just
 103          * update no idle residency and return.
 104          */
 105         if (current_clr_polling_and_test()) {
 106                 dev->last_residency = 0;
 107                 local_irq_enable();
 108                 return -EBUSY;
 109         }
 110
 111         /*
 112          * Enter the idle state previously returned by the governor decision.
 113          * This function will block until an interrupt occurs and will take
 114          * care of re-enabling the local interrupts
 115          */
 116         return cpuidle_enter(drv, dev, next_state);
 117 }
 118
 119 /**
 120  * cpuidle_idle_call - the main idle function
 121  *
 122  * NOTE: no locks or semaphores should be used here
 123  *
 124  * On archs that support TIF_POLLING_NRFLAG, is called with polling
 125  * set, and it returns with polling set.  If it ever stops polling, it
 126  * must clear the polling bit.
 127  */
 128 static void cpuidle_idle_call(void)
 129 {
 130         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
 131         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 132         int next_state, entered_state;
 133
 134         /*
 135          * Check if the idle task must be rescheduled. If it is the
 136          * case, exit the function after re-enabling the local irq.
 137          */
 138         if (need_resched()) {
 139                 local_irq_enable();
 140                 return;
 141         }
 142
 143         /*
 144          * During the idle period, stop measuring the disabled irqs
 145          * critical sections latencies
 146          */
 147         stop_critical_timings();
 148
 149         /*
 150          * Tell the RCU framework we are entering an idle section,
 151          * so no more rcu read side critical sections and one more
 152          * step to the grace period
 153          */
 154         rcu_idle_enter();
 155
 156         if (cpuidle_not_available(drv, dev)) {
 157                 default_idle_call();
 158                 goto exit_idle;
 159         }
 160
 161         /*
 162          * Suspend-to-idle ("freeze") is a system state in which all user space
 163          * has been frozen, all I/O devices have been suspended and the only
 164          * activity happens here and in iterrupts (if any).  In that case bypass
 165          * the cpuidle governor and go stratight for the deepest idle state
 166          * available.  Possibly also suspend the local tick and the entire
 167          * timekeeping to prevent timer interrupts from kicking us out of idle
 168          * until a proper wakeup interrupt happens.
 169          */
 170         if (idle_should_freeze()) {
 171                 entered_state = cpuidle_enter_freeze(drv, dev);
 172                 if (entered_state >= 0) {
 173                         local_irq_enable();
 174                         goto exit_idle;
 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         start_critical_timings();
 202 }
 203
 204 DEFINE_PER_CPU(bool, cpu_dead_idle);
 205
 206 /*
 207  * Generic idle loop implementation
 208  *
 209  * Called with polling cleared.
 210  */
 211 static void cpu_idle_loop(void)
 212 {
 213         while (1) {
 214                 /*
 215                  * If the arch has a polling bit, we maintain an invariant:
 216                  *
 217                  * Our polling bit is clear if we're not scheduled (i.e. if
 218                  * rq->curr != rq->idle).  This means that, if rq->idle has
 219                  * the polling bit set, then setting need_resched is
 220                  * guaranteed to cause the cpu to reschedule.
 221                  */
 222
 223                 __current_set_polling();
 224                 tick_nohz_idle_enter();
 225
 226                 while (!need_resched()) {
 227                         check_pgt_cache();
 228                         rmb();
 229
 230                         if (cpu_is_offline(smp_processor_id())) {
 231                                 rcu_cpu_notify(NULL, CPU_DYING_IDLE,
 232                                                (void *)(long)smp_processor_id());
 233                                 smp_mb(); /* all activity before dead. */
 234                                 this_cpu_write(cpu_dead_idle, true);
 235                                 arch_cpu_idle_dead();
 236                         }
 237
 238                         local_irq_disable();
 239                         arch_cpu_idle_enter();
 240
 241                         /*
 242                          * In poll mode we reenable interrupts and spin.
 243                          *
 244                          * Also if we detected in the wakeup from idle
 245                          * path that the tick broadcast device expired
 246                          * for us, we don't want to go deep idle as we
 247                          * know that the IPI is going to arrive right
 248                          * away
 249                          */
 250                         if (cpu_idle_force_poll || tick_check_broadcast_expired())
 251                                 cpu_idle_poll();
 252                         else
 253                                 cpuidle_idle_call();
 254
 255                         arch_cpu_idle_exit();
 256                 }
 257
 258                 /*
 259                  * Since we fell out of the loop above, we know
 260                  * TIF_NEED_RESCHED must be set, propagate it into
 261                  * PREEMPT_NEED_RESCHED.
 262                  *
 263                  * This is required because for polling idle loops we will
 264                  * not have had an IPI to fold the state for us.
 265                  */
 266                 preempt_set_need_resched();
 267                 tick_nohz_idle_exit();
 268                 __current_clr_polling();
 269
 270                 /*
 271                  * We promise to call sched_ttwu_pending and reschedule
 272                  * if need_resched is set while polling is set.  That
 273                  * means that clearing polling needs to be visible
 274                  * before doing these things.
 275                  */
 276                 smp_mb__after_atomic();
 277
 278                 sched_ttwu_pending();
 279                 schedule_preempt_disabled();
 280         }
 281 }
 282
 283 void cpu_startup_entry(enum cpuhp_state state)
 284 {
 285         /*
 286          * This #ifdef needs to die, but it's too late in the cycle to
 287          * make this generic (arm and sh have never invoked the canary
 288          * init for the non boot cpus!). Will be fixed in 3.11
 289          */
 290 #ifdef CONFIG_X86
 291         /*
 292          * If we're the non-boot CPU, nothing set the stack canary up
 293          * for us. The boot CPU already has it initialized but no harm
 294          * in doing it again. This is a good place for updating it, as
 295          * we wont ever return from this function (so the invalid
 296          * canaries already on the stack wont ever trigger).
 297          */
 298         boot_init_stack_canary();
 299 #endif
 300         arch_cpu_idle_prepare();
 301         cpu_idle_loop();
 302 }