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         bool reflect;
  86
  87         /*
  88          * Check if the idle task must be rescheduled. If it is the
  89          * case, exit the function after re-enabling the local irq.
  90          */
  91         if (need_resched()) {
  92                 local_irq_enable();
  93                 return;
  94         }
  95
  96         /*
  97          * During the idle period, stop measuring the disabled irqs
  98          * critical sections latencies
  99          */
 100         stop_critical_timings();
 101
 102         /*
 103          * Tell the RCU framework we are entering an idle section,
 104          * so no more rcu read side critical sections and one more
 105          * step to the grace period
 106          */
 107         rcu_idle_enter();
 108
 109         if (cpuidle_not_available(drv, dev))
 110                 goto use_default;
 111
 112         /*
 113          * Suspend-to-idle ("freeze") is a system state in which all user space
 114          * has been frozen, all I/O devices have been suspended and the only
 115          * activity happens here and in iterrupts (if any).  In that case bypass
 116          * the cpuidle governor and go stratight for the deepest idle state
 117          * available.  Possibly also suspend the local tick and the entire
 118          * timekeeping to prevent timer interrupts from kicking us out of idle
 119          * until a proper wakeup interrupt happens.
 120          */
 121         if (idle_should_freeze()) {
 122                 entered_state = cpuidle_enter_freeze(drv, dev);
 123                 if (entered_state >= 0) {
 124                         local_irq_enable();
 125                         goto exit_idle;
 126                 }
 127
 128                 reflect = false;
 129                 next_state = cpuidle_find_deepest_state(drv, dev);
 130         } else {
 131                 reflect = true;
 132                 /*
 133                  * Ask the cpuidle framework to choose a convenient idle state.
 134                  */
 135                 next_state = cpuidle_select(drv, dev);
 136         }
 137         /* Fall back to the default arch idle method on errors. */
 138         if (next_state < 0)
 139                 goto use_default;
 140
 141         /*
 142          * The idle task must be scheduled, it is pointless to
 143          * go to idle, just update no idle residency and get
 144          * out of this function
 145          */
 146         if (current_clr_polling_and_test()) {
 147                 dev->last_residency = 0;
 148                 entered_state = next_state;
 149                 local_irq_enable();
 150                 goto exit_idle;
 151         }
 152
 153         broadcast = drv->states[next_state].flags & CPUIDLE_FLAG_TIMER_STOP;
 154
 155         /*
 156          * Tell the time framework to switch to a broadcast timer
 157          * because our local timer will be shutdown. If a local timer
 158          * is used from another cpu as a broadcast timer, this call may
 159          * fail if it is not available
 160          */
 161         if (broadcast &&
 162             clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu))
 163                 goto use_default;
 164
 165         /* Take note of the planned idle state. */
 166         idle_set_state(this_rq(), &drv->states[next_state]);
 167
 168         /*
 169          * Enter the idle state previously returned by the governor decision.
 170          * This function will block until an interrupt occurs and will take
 171          * care of re-enabling the local interrupts
 172          */
 173         entered_state = cpuidle_enter(drv, dev, next_state);
 174
 175         /* The cpu is no longer idle or about to enter idle. */
 176         idle_set_state(this_rq(), NULL);
 177
 178         if (broadcast)
 179                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu);
 180
 181         /*
 182          * Give the governor an opportunity to reflect on the outcome
 183          */
 184         if (reflect)
 185                 cpuidle_reflect(dev, entered_state);
 186
 187 exit_idle:
 188         __current_set_polling();
 189
 190         /*
 191          * It is up to the idle functions to reenable local interrupts
 192          */
 193         if (WARN_ON_ONCE(irqs_disabled()))
 194                 local_irq_enable();
 195
 196         rcu_idle_exit();
 197         start_critical_timings();
 198         return;
 199
 200 use_default:
 201         /*
 202          * We can't use the cpuidle framework, let's use the default
 203          * idle routine.
 204          */
 205         if (current_clr_polling_and_test())
 206                 local_irq_enable();
 207         else
 208                 arch_cpu_idle();
 209
 210         goto exit_idle;
 211 }
 212
 213 /*
 214  * Generic idle loop implementation
 215  *
 216  * Called with polling cleared.
 217  */
 218 static void cpu_idle_loop(void)
 219 {
 220         while (1) {
 221                 /*
 222                  * If the arch has a polling bit, we maintain an invariant:
 223                  *
 224                  * Our polling bit is clear if we're not scheduled (i.e. if
 225                  * rq->curr != rq->idle).  This means that, if rq->idle has
 226                  * the polling bit set, then setting need_resched is
 227                  * guaranteed to cause the cpu to reschedule.
 228                  */
 229
 230                 __current_set_polling();
 231                 tick_nohz_idle_enter();
 232
 233                 while (!need_resched()) {
 234                         check_pgt_cache();
 235                         rmb();
 236
 237                         if (cpu_is_offline(smp_processor_id()))
 238                                 arch_cpu_idle_dead();
 239
 240                         local_irq_disable();
 241                         arch_cpu_idle_enter();
 242
 243                         /*
 244                          * In poll mode we reenable interrupts and spin.
 245                          *
 246                          * Also if we detected in the wakeup from idle
 247                          * path that the tick broadcast device expired
 248                          * for us, we don't want to go deep idle as we
 249                          * know that the IPI is going to arrive right
 250                          * away
 251                          */
 252                         if (cpu_idle_force_poll || tick_check_broadcast_expired())
 253                                 cpu_idle_poll();
 254                         else
 255                                 cpuidle_idle_call();
 256
 257                         arch_cpu_idle_exit();
 258                 }
 259
 260                 /*
 261                  * Since we fell out of the loop above, we know
 262                  * TIF_NEED_RESCHED must be set, propagate it into
 263                  * PREEMPT_NEED_RESCHED.
 264                  *
 265                  * This is required because for polling idle loops we will
 266                  * not have had an IPI to fold the state for us.
 267                  */
 268                 preempt_set_need_resched();
 269                 tick_nohz_idle_exit();
 270                 __current_clr_polling();
 271
 272                 /*
 273                  * We promise to call sched_ttwu_pending and reschedule
 274                  * if need_resched is set while polling is set.  That
 275                  * means that clearing polling needs to be visible
 276                  * before doing these things.
 277                  */
 278                 smp_mb__after_atomic();
 279
 280                 sched_ttwu_pending();
 281                 schedule_preempt_disabled();
 282         }
 283 }
 284
 285 void cpu_startup_entry(enum cpuhp_state state)
 286 {
 287         /*
 288          * This #ifdef needs to die, but it's too late in the cycle to
 289          * make this generic (arm and sh have never invoked the canary
 290          * init for the non boot cpus!). Will be fixed in 3.11
 291          */
 292 #ifdef CONFIG_X86
 293         /*
 294          * If we're the non-boot CPU, nothing set the stack canary up
 295          * for us. The boot CPU already has it initialized but no harm
 296          * in doing it again. This is a good place for updating it, as
 297          * we wont ever return from this function (so the invalid
 298          * canaries already on the stack wont ever trigger).
 299          */
 300         boot_init_stack_canary();
 301 #endif
 302         arch_cpu_idle_prepare();
 303         cpu_idle_loop();
 304 }