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/cpuhotplug.h>
   8 #include <linux/tick.h>
   9 #include <linux/mm.h>
  10 #include <linux/stackprotector.h>
  11 #include <linux/suspend.h>
  12
  13 #include <asm/tlb.h>
  14
  15 #include <trace/events/power.h>
  16
  17 #include "sched.h"
  18
  19 /**
  20  * sched_idle_set_state - Record idle state for the current CPU.
  21  * @idle_state: State to record.
  22  */
  23 void sched_idle_set_state(struct cpuidle_state *idle_state)
  24 {
  25         idle_set_state(this_rq(), idle_state);
  26 }
  27
  28 static int __read_mostly cpu_idle_force_poll;
  29
  30 void cpu_idle_poll_ctrl(bool enable)
  31 {
  32         if (enable) {
  33                 cpu_idle_force_poll++;
  34         } else {
  35                 cpu_idle_force_poll--;
  36                 WARN_ON_ONCE(cpu_idle_force_poll < 0);
  37         }
  38 }
  39
  40 #ifdef CONFIG_GENERIC_IDLE_POLL_SETUP
  41 static int __init cpu_idle_poll_setup(char *__unused)
  42 {
  43         cpu_idle_force_poll = 1;
  44         return 1;
  45 }
  46 __setup("nohlt", cpu_idle_poll_setup);
  47
  48 static int __init cpu_idle_nopoll_setup(char *__unused)
  49 {
  50         cpu_idle_force_poll = 0;
  51         return 1;
  52 }
  53 __setup("hlt", cpu_idle_nopoll_setup);
  54 #endif
  55
  56 static inline int cpu_idle_poll(void)
  57 {
  58         rcu_idle_enter();
  59         trace_cpu_idle_rcuidle(0, smp_processor_id());
  60         local_irq_enable();
  61         stop_critical_timings();
  62         while (!tif_need_resched() &&
  63                 (cpu_idle_force_poll || tick_check_broadcast_expired()))
  64                 cpu_relax();
  65         start_critical_timings();
  66         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
  67         rcu_idle_exit();
  68         return 1;
  69 }
  70
  71 /* Weak implementations for optional arch specific functions */
  72 void __weak arch_cpu_idle_prepare(void) { }
  73 void __weak arch_cpu_idle_enter(void) { }
  74 void __weak arch_cpu_idle_exit(void) { }
  75 void __weak arch_cpu_idle_dead(void) { }
  76 void __weak arch_cpu_idle(void)
  77 {
  78         cpu_idle_force_poll = 1;
  79         local_irq_enable();
  80 }
  81
  82 /**
  83  * default_idle_call - Default CPU idle routine.
  84  *
  85  * To use when the cpuidle framework cannot be used.
  86  */
  87 void default_idle_call(void)
  88 {
  89         if (current_clr_polling_and_test()) {
  90                 local_irq_enable();
  91         } else {
  92                 stop_critical_timings();
  93                 arch_cpu_idle();
  94                 start_critical_timings();
  95         }
  96 }
  97
  98 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
  99                       int next_state)
 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 = cpuidle_get_device();
 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          * Tell the RCU framework we are entering an idle section,
 145          * so no more rcu read side critical sections and one more
 146          * step to the grace period
 147          */
 148         rcu_idle_enter();
 149
 150         if (cpuidle_not_available(drv, dev)) {
 151                 default_idle_call();
 152                 goto exit_idle;
 153         }
 154
 155         /*
 156          * Suspend-to-idle ("freeze") is a system state in which all user space
 157          * has been frozen, all I/O devices have been suspended and the only
 158          * activity happens here and in iterrupts (if any).  In that case bypass
 159          * the cpuidle governor and go stratight for the deepest idle state
 160          * available.  Possibly also suspend the local tick and the entire
 161          * timekeeping to prevent timer interrupts from kicking us out of idle
 162          * until a proper wakeup interrupt happens.
 163          */
 164         if (idle_should_freeze()) {
 165                 entered_state = cpuidle_enter_freeze(drv, dev);
 166                 if (entered_state > 0) {
 167                         local_irq_enable();
 168                         goto exit_idle;
 169                 }
 170
 171                 next_state = cpuidle_find_deepest_state(drv, dev);
 172                 call_cpuidle(drv, dev, next_state);
 173         } else {
 174                 /*
 175                  * Ask the cpuidle framework to choose a convenient idle state.
 176                  */
 177                 next_state = cpuidle_select(drv, dev);
 178                 entered_state = call_cpuidle(drv, dev, next_state);
 179                 /*
 180                  * Give the governor an opportunity to reflect on the outcome
 181                  */
 182                 cpuidle_reflect(dev, entered_state);
 183         }
 184
 185 exit_idle:
 186         __current_set_polling();
 187
 188         /*
 189          * It is up to the idle functions to reenable local interrupts
 190          */
 191         if (WARN_ON_ONCE(irqs_disabled()))
 192                 local_irq_enable();
 193
 194         rcu_idle_exit();
 195 }
 196
 197 /*
 198  * Generic idle loop implementation
 199  *
 200  * Called with polling cleared.
 201  */
 202 static void cpu_idle_loop(void)
 203 {
 204         int cpu = smp_processor_id();
 205
 206         while (1) {
 207                 /*
 208                  * If the arch has a polling bit, we maintain an invariant:
 209                  *
 210                  * Our polling bit is clear if we're not scheduled (i.e. if
 211                  * rq->curr != rq->idle).  This means that, if rq->idle has
 212                  * the polling bit set, then setting need_resched is
 213                  * guaranteed to cause the cpu to reschedule.
 214                  */
 215
 216                 __current_set_polling();
 217                 quiet_vmstat();
 218                 tick_nohz_idle_enter();
 219
 220                 while (!need_resched()) {
 221                         check_pgt_cache();
 222                         rmb();
 223
 224                         if (cpu_is_offline(cpu)) {
 225                                 cpuhp_report_idle_dead();
 226                                 arch_cpu_idle_dead();
 227                         }
 228
 229                         local_irq_disable();
 230                         arch_cpu_idle_enter();
 231
 232                         /*
 233                          * In poll mode we reenable interrupts and spin.
 234                          *
 235                          * Also if we detected in the wakeup from idle
 236                          * path that the tick broadcast device expired
 237                          * for us, we don't want to go deep idle as we
 238                          * know that the IPI is going to arrive right
 239                          * away
 240                          */
 241                         if (cpu_idle_force_poll || tick_check_broadcast_expired())
 242                                 cpu_idle_poll();
 243                         else
 244                                 cpuidle_idle_call();
 245
 246                         arch_cpu_idle_exit();
 247                 }
 248
 249                 /*
 250                  * Since we fell out of the loop above, we know
 251                  * TIF_NEED_RESCHED must be set, propagate it into
 252                  * PREEMPT_NEED_RESCHED.
 253                  *
 254                  * This is required because for polling idle loops we will
 255                  * not have had an IPI to fold the state for us.
 256                  */
 257                 preempt_set_need_resched();
 258                 tick_nohz_idle_exit();
 259                 __current_clr_polling();
 260
 261                 /*
 262                  * We promise to call sched_ttwu_pending and reschedule
 263                  * if need_resched is set while polling is set.  That
 264                  * means that clearing polling needs to be visible
 265                  * before doing these things.
 266                  */
 267                 smp_mb__after_atomic();
 268
 269                 sched_ttwu_pending();
 270                 schedule_preempt_disabled();
 271         }
 272 }
 273
 274 void cpu_startup_entry(enum cpuhp_state state)
 275 {
 276         /*
 277          * This #ifdef needs to die, but it's too late in the cycle to
 278          * make this generic (arm and sh have never invoked the canary
 279          * init for the non boot cpus!). Will be fixed in 3.11
 280          */
 281 #ifdef CONFIG_X86
 282         /*
 283          * If we're the non-boot CPU, nothing set the stack canary up
 284          * for us. The boot CPU already has it initialized but no harm
 285          * in doing it again. This is a good place for updating it, as
 286          * we wont ever return from this function (so the invalid
 287          * canaries already on the stack wont ever trigger).
 288          */
 289         boot_init_stack_canary();
 290 #endif
 291         arch_cpu_idle_prepare();
 292         cpuhp_online_idle(state);
 293         cpu_idle_loop();
 294 }