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
  11 #include <asm/tlb.h>
  12
  13 #include <trace/events/power.h>
  14
  15 static int __read_mostly cpu_idle_force_poll;
  16
  17 void cpu_idle_poll_ctrl(bool enable)
  18 {
  19         if (enable) {
  20                 cpu_idle_force_poll++;
  21         } else {
  22                 cpu_idle_force_poll--;
  23                 WARN_ON_ONCE(cpu_idle_force_poll < 0);
  24         }
  25 }
  26
  27 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
  28 static int __init cpu_idle_poll_setup(char *__unused)
  29 {
  30         cpu_idle_force_poll = 1;
  31         return 1;
  32 }
  33 __setup("nohlt", cpu_idle_poll_setup);
  34
  35 static int __init cpu_idle_nopoll_setup(char *__unused)
  36 {
  37         cpu_idle_force_poll = 0;
  38         return 1;
  39 }
  40 __setup("hlt", cpu_idle_nopoll_setup);
  41 #endif
  42
  43 static inline int cpu_idle_poll(void)
  44 {
  45         rcu_idle_enter();
  46         trace_cpu_idle_rcuidle(0, smp_processor_id());
  47         local_irq_enable();
  48         while (!tif_need_resched())
  49                 cpu_relax();
  50         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
  51         rcu_idle_exit();
  52         return 1;
  53 }
  54
  55 /* Weak implementations for optional arch specific functions */
  56 void __weak arch_cpu_idle_prepare(void) { }
  57 void __weak arch_cpu_idle_enter(void) { }
  58 void __weak arch_cpu_idle_exit(void) { }
  59 void __weak arch_cpu_idle_dead(void) { }
  60 void __weak arch_cpu_idle(void)
  61 {
  62         cpu_idle_force_poll = 1;
  63         local_irq_enable();
  64 }
  65
  66 /**
  67  * cpuidle_idle_call - the main idle function
  68  *
  69  * NOTE: no locks or semaphores should be used here
  70  */
  71 static void cpuidle_idle_call(void)
  72 {
  73         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
  74         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  75         int next_state, entered_state;
  76         bool broadcast;
  77
  78         /*
  79          * Check if the idle task must be rescheduled. If it is the
  80          * case, exit the function after re-enabling the local irq.
  81          */
  82         if (need_resched()) {
  83                 local_irq_enable();
  84                 return;
  85         }
  86
  87         /*
  88          * During the idle period, stop measuring the disabled irqs
  89          * critical sections latencies
  90          */
  91         stop_critical_timings();
  92
  93         /*
  94          * Tell the RCU framework we are entering an idle section,
  95          * so no more rcu read side critical sections and one more
  96          * step to the grace period
  97          */
  98         rcu_idle_enter();
  99
 100         /*
 101          * Ask the cpuidle framework to choose a convenient idle state.
 102          * Fall back to the default arch idle method on errors.
 103          */
 104         next_state = cpuidle_select(drv, dev);
 105         if (next_state < 0) {
 106 use_default:
 107                 /*
 108                  * We can't use the cpuidle framework, let's use the default
 109                  * idle routine.
 110                  */
 111                 if (current_clr_polling_and_test())
 112                         local_irq_enable();
 113                 else
 114                         arch_cpu_idle();
 115
 116                 goto exit_idle;
 117         }
 118
 119
 120         /*
 121          * The idle task must be scheduled, it is pointless to
 122          * go to idle, just update no idle residency and get
 123          * out of this function
 124          */
 125         if (current_clr_polling_and_test()) {
 126                 dev->last_residency = 0;
 127                 entered_state = next_state;
 128                 local_irq_enable();
 129                 goto exit_idle;
 130         }
 131
 132         broadcast = !!(drv->states[next_state].flags & CPUIDLE_FLAG_TIMER_STOP);
 133
 134         /*
 135          * Tell the time framework to switch to a broadcast timer
 136          * because our local timer will be shutdown. If a local timer
 137          * is used from another cpu as a broadcast timer, this call may
 138          * fail if it is not available
 139          */
 140         if (broadcast &&
 141             clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu))
 142                 goto use_default;
 143
 144         trace_cpu_idle_rcuidle(next_state, dev->cpu);
 145
 146         /*
 147          * Enter the idle state previously returned by the governor decision.
 148          * This function will block until an interrupt occurs and will take
 149          * care of re-enabling the local interrupts
 150          */
 151         entered_state = cpuidle_enter(drv, dev, next_state);
 152
 153         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, dev->cpu);
 154
 155         if (broadcast)
 156                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu);
 157
 158         /*
 159          * Give the governor an opportunity to reflect on the outcome
 160          */
 161         cpuidle_reflect(dev, entered_state);
 162
 163 exit_idle:
 164         __current_set_polling();
 165
 166         /*
 167          * It is up to the idle functions to reenable local interrupts
 168          */
 169         if (WARN_ON_ONCE(irqs_disabled()))
 170                 local_irq_enable();
 171
 172         rcu_idle_exit();
 173         start_critical_timings();
 174 }
 175
 176 /*
 177  * Generic idle loop implementation
 178  */
 179 static void cpu_idle_loop(void)
 180 {
 181         while (1) {
 182                 tick_nohz_idle_enter();
 183
 184                 while (!need_resched()) {
 185                         check_pgt_cache();
 186                         rmb();
 187
 188                         if (cpu_is_offline(smp_processor_id()))
 189                                 arch_cpu_idle_dead();
 190
 191                         local_irq_disable();
 192                         arch_cpu_idle_enter();
 193
 194                         /*
 195                          * In poll mode we reenable interrupts and spin.
 196                          *
 197                          * Also if we detected in the wakeup from idle
 198                          * path that the tick broadcast device expired
 199                          * for us, we don't want to go deep idle as we
 200                          * know that the IPI is going to arrive right
 201                          * away
 202                          */
 203                         if (cpu_idle_force_poll || tick_check_broadcast_expired())
 204                                 cpu_idle_poll();
 205                         else
 206                                 cpuidle_idle_call();
 207
 208                         arch_cpu_idle_exit();
 209                 }
 210
 211                 /*
 212                  * Since we fell out of the loop above, we know
 213                  * TIF_NEED_RESCHED must be set, propagate it into
 214                  * PREEMPT_NEED_RESCHED.
 215                  *
 216                  * This is required because for polling idle loops we will
 217                  * not have had an IPI to fold the state for us.
 218                  */
 219                 preempt_set_need_resched();
 220                 tick_nohz_idle_exit();
 221                 schedule_preempt_disabled();
 222         }
 223 }
 224
 225 void cpu_startup_entry(enum cpuhp_state state)
 226 {
 227         /*
 228          * This #ifdef needs to die, but it's too late in the cycle to
 229          * make this generic (arm and sh have never invoked the canary
 230          * init for the non boot cpus!). Will be fixed in 3.11
 231          */
 232 #ifdef CONFIG_X86
 233         /*
 234          * If we're the non-boot CPU, nothing set the stack canary up
 235          * for us. The boot CPU already has it initialized but no harm
 236          * in doing it again. This is a good place for updating it, as
 237          * we wont ever return from this function (so the invalid
 238          * canaries already on the stack wont ever trigger).
 239          */
 240         boot_init_stack_canary();
 241 #endif
 242         __current_set_polling();
 243         arch_cpu_idle_prepare();
 244         cpu_idle_loop();
 245 }