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         bool reflect;
  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         if (cpuidle_not_available(drv, dev))
 109                 goto use_default;
 110
 111         /*
 112          * Suspend-to-idle ("freeze") is a system state in which all user space
 113          * has been frozen, all I/O devices have been suspended and the only
 114          * activity happens here and in iterrupts (if any).  In that case bypass
 115          * the cpuidle governor and go stratight for the deepest idle state
 116          * available.  Possibly also suspend the local tick and the entire
 117          * timekeeping to prevent timer interrupts from kicking us out of idle
 118          * until a proper wakeup interrupt happens.
 119          */
 120         if (idle_should_freeze()) {
 121                 entered_state = cpuidle_enter_freeze(drv, dev);
 122                 if (entered_state >= 0) {
 123                         local_irq_enable();
 124                         goto exit_idle;
 125                 }
 126
 127                 reflect = false;
 128                 next_state = cpuidle_find_deepest_state(drv, dev);
 129         } else {
 130                 reflect = true;
 131                 /*
 132                  * Ask the cpuidle framework to choose a convenient idle state.
 133                  */
 134                 next_state = cpuidle_select(drv, dev);
 135         }
 136         /* Fall back to the default arch idle method on errors. */
 137         if (next_state < 0)
 138                 goto use_default;
 139
 140         /*
 141          * The idle task must be scheduled, it is pointless to
 142          * go to idle, just update no idle residency and get
 143          * out of this function
 144          */
 145         if (current_clr_polling_and_test()) {
 146                 dev->last_residency = 0;
 147                 entered_state = next_state;
 148                 local_irq_enable();
 149                 goto exit_idle;
 150         }
 151
 152         /* Take note of the planned idle state. */
 153         idle_set_state(this_rq(), &drv->states[next_state]);
 154
 155         /*
 156          * Enter the idle state previously returned by the governor decision.
 157          * This function will block until an interrupt occurs and will take
 158          * care of re-enabling the local interrupts
 159          */
 160         entered_state = cpuidle_enter(drv, dev, next_state);
 161
 162         /* The cpu is no longer idle or about to enter idle. */
 163         idle_set_state(this_rq(), NULL);
 164
 165         if (entered_state == -EBUSY)
 166                 goto use_default;
 167
 168         /*
 169          * Give the governor an opportunity to reflect on the outcome
 170          */
 171         if (reflect)
 172                 cpuidle_reflect(dev, entered_state);
 173
 174 exit_idle:
 175         __current_set_polling();
 176
 177         /*
 178          * It is up to the idle functions to reenable local interrupts
 179          */
 180         if (WARN_ON_ONCE(irqs_disabled()))
 181                 local_irq_enable();
 182
 183         rcu_idle_exit();
 184         start_critical_timings();
 185         return;
 186
 187 use_default:
 188         /*
 189          * We can't use the cpuidle framework, let's use the default
 190          * idle routine.
 191          */
 192         if (current_clr_polling_and_test())
 193                 local_irq_enable();
 194         else
 195                 arch_cpu_idle();
 196
 197         goto exit_idle;
 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 }