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 static int __read_mostly cpu_idle_force_poll;
  19
  20 void cpu_idle_poll_ctrl(bool enable)
  21 {
  22         if (enable) {
  23                 cpu_idle_force_poll++;
  24         } else {
  25                 cpu_idle_force_poll--;
  26                 WARN_ON_ONCE(cpu_idle_force_poll < 0);
  27         }
  28 }
  29
  30 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
  31 static int __init cpu_idle_poll_setup(char *__unused)
  32 {
  33         cpu_idle_force_poll = 1;
  34         return 1;
  35 }
  36 __setup("nohlt", cpu_idle_poll_setup);
  37
  38 static int __init cpu_idle_nopoll_setup(char *__unused)
  39 {
  40         cpu_idle_force_poll = 0;
  41         return 1;
  42 }
  43 __setup("hlt", cpu_idle_nopoll_setup);
  44 #endif
  45
  46 static inline int cpu_idle_poll(void)
  47 {
  48         rcu_idle_enter();
  49         trace_cpu_idle_rcuidle(0, smp_processor_id());
  50         local_irq_enable();
  51         while (!tif_need_resched() &&
  52                 (cpu_idle_force_poll || tick_check_broadcast_expired()))
  53                 cpu_relax();
  54         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
  55         rcu_idle_exit();
  56         return 1;
  57 }
  58
  59 /* Weak implementations for optional arch specific functions */
  60 void __weak arch_cpu_idle_prepare(void) { }
  61 void __weak arch_cpu_idle_enter(void) { }
  62 void __weak arch_cpu_idle_exit(void) { }
  63 void __weak arch_cpu_idle_dead(void) { }
  64 void __weak arch_cpu_idle(void)
  65 {
  66         cpu_idle_force_poll = 1;
  67         local_irq_enable();
  68 }
  69
  70 /**
  71  * cpuidle_idle_call - the main idle function
  72  *
  73  * NOTE: no locks or semaphores should be used here
  74  *
  75  * On archs that support TIF_POLLING_NRFLAG, is called with polling
  76  * set, and it returns with polling set.  If it ever stops polling, it
  77  * must clear the polling bit.
  78  */
  79 static void cpuidle_idle_call(void)
  80 {
  81         struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
  82         struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
  83         int next_state, entered_state;
  84         unsigned int broadcast;
  85
  86         /*
  87          * Check if the idle task must be rescheduled. If it is the
  88          * case, exit the function after re-enabling the local irq.
  89          */
  90         if (need_resched()) {
  91                 local_irq_enable();
  92                 return;
  93         }
  94
  95         /*
  96          * During the idle period, stop measuring the disabled irqs
  97          * critical sections latencies
  98          */
  99         stop_critical_timings();
 100
 101         /*
 102          * Tell the RCU framework we are entering an idle section,
 103          * so no more rcu read side critical sections and one more
 104          * step to the grace period
 105          */
 106         rcu_idle_enter();
 107
 108         /*
 109          * Suspend-to-idle ("freeze") is a system state in which all user space
 110          * has been frozen, all I/O devices have been suspended and the only
 111          * activity happens here and in iterrupts (if any).  In that case bypass
 112          * the cpuidle governor and go stratight for the deepest idle state
 113          * available.  Possibly also suspend the local tick and the entire
 114          * timekeeping to prevent timer interrupts from kicking us out of idle
 115          * until a proper wakeup interrupt happens.
 116          */
 117         if (idle_should_freeze()) {
 118                 cpuidle_enter_freeze();
 119                 local_irq_enable();
 120                 goto exit_idle;
 121         }
 122
 123         /*
 124          * Ask the cpuidle framework to choose a convenient idle state.
 125          * Fall back to the default arch idle method on errors.
 126          */
 127         next_state = cpuidle_select(drv, dev);
 128         if (next_state < 0) {
 129 use_default:
 130                 /*
 131                  * We can't use the cpuidle framework, let's use the default
 132                  * idle routine.
 133                  */
 134                 if (current_clr_polling_and_test())
 135                         local_irq_enable();
 136                 else
 137                         arch_cpu_idle();
 138
 139                 goto exit_idle;
 140         }
 141
 142
 143         /*
 144          * The idle task must be scheduled, it is pointless to
 145          * go to idle, just update no idle residency and get
 146          * out of this function
 147          */
 148         if (current_clr_polling_and_test()) {
 149                 dev->last_residency = 0;
 150                 entered_state = next_state;
 151                 local_irq_enable();
 152                 goto exit_idle;
 153         }
 154
 155         broadcast = drv->states[next_state].flags & CPUIDLE_FLAG_TIMER_STOP;
 156
 157         /*
 158          * Tell the time framework to switch to a broadcast timer
 159          * because our local timer will be shutdown. If a local timer
 160          * is used from another cpu as a broadcast timer, this call may
 161          * fail if it is not available
 162          */
 163         if (broadcast &&
 164             clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu))
 165                 goto use_default;
 166
 167         /* Take note of the planned idle state. */
 168         idle_set_state(this_rq(), &drv->states[next_state]);
 169
 170         /*
 171          * Enter the idle state previously returned by the governor decision.
 172          * This function will block until an interrupt occurs and will take
 173          * care of re-enabling the local interrupts
 174          */
 175         entered_state = cpuidle_enter(drv, dev, next_state);
 176
 177         /* The cpu is no longer idle or about to enter idle. */
 178         idle_set_state(this_rq(), NULL);
 179
 180         if (broadcast)
 181                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu);
 182
 183         /*
 184          * Give the governor an opportunity to reflect on the outcome
 185          */
 186         cpuidle_reflect(dev, entered_state);
 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         start_critical_timings();
 199 }
 200
 201 /*
 202  * Generic idle loop implementation
 203  *
 204  * Called with polling cleared.
 205  */
 206 static void cpu_idle_loop(void)
 207 {
 208         while (1) {
 209                 /*
 210                  * If the arch has a polling bit, we maintain an invariant:
 211                  *
 212                  * Our polling bit is clear if we're not scheduled (i.e. if
 213                  * rq->curr != rq->idle).  This means that, if rq->idle has
 214                  * the polling bit set, then setting need_resched is
 215                  * guaranteed to cause the cpu to reschedule.
 216                  */
 217
 218                 __current_set_polling();
 219                 tick_nohz_idle_enter();
 220
 221                 while (!need_resched()) {
 222                         check_pgt_cache();
 223                         rmb();
 224
 225                         if (cpu_is_offline(smp_processor_id()))
 226                                 arch_cpu_idle_dead();
 227
 228                         local_irq_disable();
 229                         arch_cpu_idle_enter();
 230
 231                         /*
 232                          * In poll mode we reenable interrupts and spin.
 233                          *
 234                          * Also if we detected in the wakeup from idle
 235                          * path that the tick broadcast device expired
 236                          * for us, we don't want to go deep idle as we
 237                          * know that the IPI is going to arrive right
 238                          * away
 239                          */
 240                         if (cpu_idle_force_poll || tick_check_broadcast_expired())
 241                                 cpu_idle_poll();
 242                         else
 243                                 cpuidle_idle_call();
 244
 245                         arch_cpu_idle_exit();
 246                 }
 247
 248                 /*
 249                  * Since we fell out of the loop above, we know
 250                  * TIF_NEED_RESCHED must be set, propagate it into
 251                  * PREEMPT_NEED_RESCHED.
 252                  *
 253                  * This is required because for polling idle loops we will
 254                  * not have had an IPI to fold the state for us.
 255                  */
 256                 preempt_set_need_resched();
 257                 tick_nohz_idle_exit();
 258                 __current_clr_polling();
 259
 260                 /*
 261                  * We promise to call sched_ttwu_pending and reschedule
 262                  * if need_resched is set while polling is set.  That
 263                  * means that clearing polling needs to be visible
 264                  * before doing these things.
 265                  */
 266                 smp_mb__after_atomic();
 267
 268                 sched_ttwu_pending();
 269                 schedule_preempt_disabled();
 270         }
 271 }
 272
 273 void cpu_startup_entry(enum cpuhp_state state)
 274 {
 275         /*
 276          * This #ifdef needs to die, but it's too late in the cycle to
 277          * make this generic (arm and sh have never invoked the canary
 278          * init for the non boot cpus!). Will be fixed in 3.11
 279          */
 280 #ifdef CONFIG_X86
 281         /*
 282          * If we're the non-boot CPU, nothing set the stack canary up
 283          * for us. The boot CPU already has it initialized but no harm
 284          * in doing it again. This is a good place for updating it, as
 285          * we wont ever return from this function (so the invalid
 286          * canaries already on the stack wont ever trigger).
 287          */
 288         boot_init_stack_canary();
 289 #endif
 290         arch_cpu_idle_prepare();
 291         cpu_idle_loop();
 292 }