xnu-4570.20.62.tar.gz

[apple/xnu.git] / osfmk / kern / priority.c

diff --git a/osfmk/kern/priority.c b/osfmk/kern/priority.c
index 0b64ac8aa13de300d24cc8ae9f2f225732f21516..d61fb3584b78ef3cce6068177173eca5b6edc9dd 100644 (file)
--- a/osfmk/kern/priority.c
+++ b/osfmk/kern/priority.c
@@ -75,11 +75,18 @@
 #include <kern/processor.h>
 #include <kern/ledger.h>
 #include <machine/machparam.h>
 #include <kern/processor.h>
 #include <kern/ledger.h>
 #include <machine/machparam.h>

+#include <kern/machine.h>

 
 #ifdef CONFIG_MACH_APPROXIMATE_TIME
 #include <machine/commpage.h>  /* for commpage_update_mach_approximate_time */
 #endif
 
 
 #ifdef CONFIG_MACH_APPROXIMATE_TIME
 #include <machine/commpage.h>  /* for commpage_update_mach_approximate_time */
 #endif
 

+#if MONOTONIC
+#include <kern/monotonic.h>
+#endif /* MONOTONIC */
+
+static void sched_update_thread_bucket(thread_t thread);
+

 /*
  *     thread_quantum_expire:
  *
 /*
  *     thread_quantum_expire:
  *


@@ -97,8 +104,13 @@ thread_quantum_expire(
        thread_t                        thread = p1;
        ast_t                           preempt;
        uint64_t                        ctime;
        thread_t                        thread = p1;
        ast_t                           preempt;
        uint64_t                        ctime;

+       int                                     urgency;
+       uint64_t                        ignore1, ignore2;

 
        assert(processor == current_processor());
 
        assert(processor == current_processor());

+       assert(thread == current_thread());
+
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_QUANTUM_EXPIRED) | DBG_FUNC_START, 0, 0, 0, 0, 0);

 
        SCHED_STATS_QUANTUM_TIMER_EXPIRATION(processor);
 
 
        SCHED_STATS_QUANTUM_TIMER_EXPIRATION(processor);
 


@@ -108,30 +120,37 @@ thread_quantum_expire(
         * Because this balance adjustment could potentially attempt to wake this very
         * thread, we must credit the ledger before taking the thread lock. The ledger
         * pointers are only manipulated by the thread itself at the ast boundary.
         * Because this balance adjustment could potentially attempt to wake this very
         * thread, we must credit the ledger before taking the thread lock. The ledger
         * pointers are only manipulated by the thread itself at the ast boundary.

+        *
+        * TODO: This fails to account for the time between when the timer was armed and when it fired.
+        * It should be based on the system_timer and running a thread_timer_event operation here.

         */
        ledger_credit(thread->t_ledger, task_ledgers.cpu_time, thread->quantum_remaining);
        ledger_credit(thread->t_threadledger, thread_ledgers.cpu_time, thread->quantum_remaining);
         */
        ledger_credit(thread->t_ledger, task_ledgers.cpu_time, thread->quantum_remaining);
        ledger_credit(thread->t_threadledger, thread_ledgers.cpu_time, thread->quantum_remaining);

-#ifdef CONFIG_BANK

        if (thread->t_bankledger) {
                ledger_credit(thread->t_bankledger, bank_ledgers.cpu_time,
                                (thread->quantum_remaining - thread->t_deduct_bank_ledger_time));
        }
        thread->t_deduct_bank_ledger_time = 0;
        if (thread->t_bankledger) {
                ledger_credit(thread->t_bankledger, bank_ledgers.cpu_time,
                                (thread->quantum_remaining - thread->t_deduct_bank_ledger_time));
        }
        thread->t_deduct_bank_ledger_time = 0;

-#endif

 
        ctime = mach_absolute_time();
 
 
        ctime = mach_absolute_time();
 

+#ifdef CONFIG_MACH_APPROXIMATE_TIME
+       commpage_update_mach_approximate_time(ctime);
+#endif
+
+#if MONOTONIC
+       mt_sched_update(thread);
+#endif /* MONOTONIC */
+

        thread_lock(thread);
 
        /*
         * We've run up until our quantum expiration, and will (potentially)
         * continue without re-entering the scheduler, so update this now.
         */
        thread_lock(thread);
 
        /*
         * We've run up until our quantum expiration, and will (potentially)
         * continue without re-entering the scheduler, so update this now.
         */

+       processor->last_dispatch = ctime;

        thread->last_run_time = ctime;
 
        thread->last_run_time = ctime;
 

-#ifdef CONFIG_MACH_APPROXIMATE_TIME
-       commpage_update_mach_approximate_time(ctime);
-#endif

        /*
         *      Check for fail-safe trip.
         */
        /*
         *      Check for fail-safe trip.
         */


@@ -160,16 +179,15 @@ thread_quantum_expire(
        else
                SCHED(lightweight_update_priority)(thread);
 
        else
                SCHED(lightweight_update_priority)(thread);
 

-       SCHED(quantum_expire)(thread);
-       
-       processor->current_pri = thread->sched_pri;
-       processor->current_thmode = thread->sched_mode;
+       if (thread->sched_mode != TH_MODE_REALTIME)
+               SCHED(quantum_expire)(thread);
+
+       processor_state_update_from_thread(processor, thread);

 
        /*
         *      This quantum is up, give this thread another.
         */
 
        /*
         *      This quantum is up, give this thread another.
         */

-       if (first_timeslice(processor))
-               processor->timeslice--;
+       processor->first_timeslice = FALSE;

 
        thread_quantum_init(thread);
 
 
        thread_quantum_init(thread);
 


@@ -189,21 +207,51 @@ thread_quantum_expire(
                                         PROCESSOR_DATA(processor, thread_timer));
        }
 
                                         PROCESSOR_DATA(processor, thread_timer));
        }
 

+

        processor->quantum_end = ctime + thread->quantum_remaining;
        processor->quantum_end = ctime + thread->quantum_remaining;

-       timer_call_enter1(&processor->quantum_timer, thread,
-           processor->quantum_end, TIMER_CALL_SYS_CRITICAL | TIMER_CALL_LOCAL);

 
        /*
 
        /*

-        *      Context switch check.
+        * Context switch check
+        *
+        * non-urgent flags don't affect kernel threads, so upgrade to urgent
+        * to ensure that rebalancing and non-recommendation kick in quickly.

         */
         */

-       if ((preempt = csw_check(processor, AST_QUANTUM)) != AST_NONE)
+
+       ast_t check_reason = AST_QUANTUM;
+       if (thread->task == kernel_task)
+               check_reason |= AST_URGENT;
+
+       if ((preempt = csw_check(processor, check_reason)) != AST_NONE)

                ast_on(preempt);
 
                ast_on(preempt);
 

+       /*
+        * AST_KEVENT does not send an IPI when setting the AST,
+        * to avoid waiting for the next context switch to propagate the AST,
+        * the AST is propagated here at quantum expiration.
+        */
+       ast_propagate(thread);
+

        thread_unlock(thread);
 
        thread_unlock(thread);
 

+       timer_call_quantum_timer_enter(&processor->quantum_timer, thread,
+               processor->quantum_end, ctime);
+
+       /* Tell platform layer that we are still running this thread */
+       urgency = thread_get_urgency(thread, &ignore1, &ignore2);
+       machine_thread_going_on_core(thread, urgency, 0, 0, ctime);
+       machine_switch_perfcontrol_state_update(QUANTUM_EXPIRY, ctime,
+               0, thread);
+

 #if defined(CONFIG_SCHED_TIMESHARE_CORE)
 #if defined(CONFIG_SCHED_TIMESHARE_CORE)

-       sched_traditional_consider_maintenance(ctime);
-#endif /* CONFIG_SCHED_TIMESHARE_CORE */       
+       sched_timeshare_consider_maintenance(ctime);
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
+
+#if __arm__ || __arm64__
+       if (thread->sched_mode == TH_MODE_REALTIME)
+               sched_consider_recommended_cores(ctime, thread);
+#endif /* __arm__ || __arm64__ */
+
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_QUANTUM_EXPIRED) | DBG_FUNC_END, preempt, 0, 0, 0, 0);

 }
 
 /*
 }
 
 /*


@@ -212,35 +260,114 @@ thread_quantum_expire(
  *     Set the base priority of the thread
  *     and reset its scheduled priority.
  *
  *     Set the base priority of the thread
  *     and reset its scheduled priority.
  *

+ *     This is the only path to change base_pri.
+ *

  *     Called with the thread locked.
  */
 void
 sched_set_thread_base_priority(thread_t thread, int priority)
 {
  *     Called with the thread locked.
  */
 void
 sched_set_thread_base_priority(thread_t thread, int priority)
 {

-       thread->priority = priority;
-       SCHED(compute_priority)(thread, FALSE);
+       assert(priority >= MINPRI);
+       uint64_t ctime = 0;
+
+       if (thread->sched_mode == TH_MODE_REALTIME)
+               assert(priority <= BASEPRI_RTQUEUES);
+       else
+               assert(priority < BASEPRI_RTQUEUES);
+
+       int old_base_pri = thread->base_pri;
+       thread->base_pri = priority;
+
+       if ((thread->state & TH_RUN) == TH_RUN) {
+               assert(thread->last_made_runnable_time != THREAD_NOT_RUNNABLE);
+               ctime = mach_approximate_time();
+               thread->last_basepri_change_time = ctime;
+       } else {
+               assert(thread->last_basepri_change_time == THREAD_NOT_RUNNABLE);
+               assert(thread->last_made_runnable_time == THREAD_NOT_RUNNABLE);
+       }
+
+       /* 
+        * Currently the perfcontrol_attr depends on the base pri of the 
+        * thread. Therefore, we use this function as the hook for the 
+        * perfcontrol callout. 
+        */
+       if (thread == current_thread() && old_base_pri != priority) {
+               if (!ctime) {
+                   ctime = mach_approximate_time();
+               }
+               machine_switch_perfcontrol_state_update(PERFCONTROL_ATTR_UPDATE,
+                       ctime, PERFCONTROL_CALLOUT_WAKE_UNSAFE, thread);
+       }
+       sched_update_thread_bucket(thread);
+
+       thread_recompute_sched_pri(thread, FALSE);

 }
 
 }
 

+/*
+ *     thread_recompute_sched_pri:
+ *
+ *     Reset the scheduled priority of the thread
+ *     according to its base priority if the
+ *     thread has not been promoted or depressed.
+ *
+ *     This is the standard way to push base_pri changes into sched_pri,
+ *     or to recalculate the appropriate sched_pri after clearing
+ *     a promotion or depression.
+ *
+ *     Called at splsched with the thread locked.
+ */
+void
+thread_recompute_sched_pri(
+                           thread_t thread,
+                           boolean_t override_depress)
+{
+       int priority;
+
+       if (thread->sched_mode == TH_MODE_TIMESHARE)
+               priority = SCHED(compute_timeshare_priority)(thread);
+       else
+               priority = thread->base_pri;

 
 

-#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+       if ((!(thread->sched_flags & TH_SFLAG_PROMOTED_MASK)  || (priority > thread->sched_pri)) &&
+           (!(thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) || override_depress)) {
+               set_sched_pri(thread, priority);
+       }
+}

 
 void
 
 void

-sched_traditional_quantum_expire(thread_t      thread __unused)
+sched_default_quantum_expire(thread_t thread __unused)

 {
 {

-       /*
-        * No special behavior when a timeshare, fixed, or realtime thread
-        * uses up its entire quantum
-        */
+      /*
+       * No special behavior when a timeshare, fixed, or realtime thread
+       * uses up its entire quantum
+       */

 }
 
 }
 

+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+
+/*
+ *     lightweight_update_priority:
+ *
+ *     Update the scheduled priority for
+ *     a timesharing thread.
+ *
+ *     Only for use on the current thread.
+ *
+ *     Called with the thread locked.
+ */

 void
 lightweight_update_priority(thread_t thread)
 {
 void
 lightweight_update_priority(thread_t thread)
 {

+       assert(thread->runq == PROCESSOR_NULL);
+       assert(thread == current_thread());
+

        if (thread->sched_mode == TH_MODE_TIMESHARE) {
        if (thread->sched_mode == TH_MODE_TIMESHARE) {

-               register uint32_t       delta;
-               
+               int priority;
+               uint32_t delta;
+

                thread_timer_delta(thread, delta);
                thread_timer_delta(thread, delta);

-               
+

                /*
                 *      Accumulate timesharing usage only
                 *      during contention for processor
                /*
                 *      Accumulate timesharing usage only
                 *      during contention for processor


@@ -248,18 +375,29 @@ lightweight_update_priority(thread_t thread)
                 */
                if (thread->pri_shift < INT8_MAX)
                        thread->sched_usage += delta;
                 */
                if (thread->pri_shift < INT8_MAX)
                        thread->sched_usage += delta;

-               
+

                thread->cpu_delta += delta;
                thread->cpu_delta += delta;

-               
+
+               priority = sched_compute_timeshare_priority(thread);
+

                /*
                /*

-                * Adjust the scheduled priority if
-                * the thread has not been promoted
-                * and is not depressed.
+                * Adjust the scheduled priority like thread_recompute_sched_pri,
+                * except with the benefit of knowing the thread is on this core.

                 */
                 */

-               if (    !(thread->sched_flags & TH_SFLAG_PROMOTED_MASK) &&
-                       !(thread->sched_flags & TH_SFLAG_DEPRESSED_MASK)                )
-                       compute_my_priority(thread);
-       }       
+               if ((!(thread->sched_flags & TH_SFLAG_PROMOTED_MASK)  || (priority > thread->sched_pri)) &&
+                   (!(thread->sched_flags & TH_SFLAG_DEPRESSED_MASK)) &&
+                   priority != thread->sched_pri) {
+
+                       thread->sched_pri = priority;
+
+                       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_CHANGE_PRIORITY),
+                                             (uintptr_t)thread_tid(thread),
+                                             thread->base_pri,
+                                             thread->sched_pri,
+                                             0, /* eventually, 'reason' */
+                                             0);
+               }
+       }

 }
 
 /*
 }
 
 /*


@@ -283,84 +421,62 @@ static struct shift_data  sched_decay_shifts[SCHED_DECAY_TICKS] = {
 };
 
 /*
 };
 
 /*

- *     do_priority_computation:
+ *     sched_compute_timeshare_priority:

  *
  *     Calculate the timesharing priority based upon usage and load.
  */
 extern int sched_pri_decay_band_limit;
 
  *
  *     Calculate the timesharing priority based upon usage and load.
  */
 extern int sched_pri_decay_band_limit;
 

+#ifdef CONFIG_EMBEDDED

 
 

-static int do_priority_computation(thread_t th) {                                                                                                                      
-       register int priority = th->priority            /* start with base priority */          
-           - (th->sched_usage >> th->pri_shift);                               
-       if (priority < MINPRI_USER)                                                                                     
-               priority = MINPRI_USER;                                                                                 
-       else                                                                                                                            
-       if (priority > MAXPRI_KERNEL)                                                                                   
-               priority = MAXPRI_KERNEL;       
-
-       return priority;                                                                                
-}
+int
+sched_compute_timeshare_priority(thread_t thread)
+{
+       int decay_amount = (thread->sched_usage >> thread->pri_shift);
+       int decay_limit = sched_pri_decay_band_limit;

 
 

+       if (thread->base_pri > BASEPRI_FOREGROUND) {
+               decay_limit += (thread->base_pri - BASEPRI_FOREGROUND);
+       }

 
 

-/*
- *     compute_priority:
- *
- *     Reset the scheduled priority of the thread
- *     according to its base priority if the
- *     thread has not been promoted or depressed.
- *
- *     Called with the thread locked.
- */
-void
-compute_priority(
-       register thread_t       thread,
-       boolean_t                       override_depress)
-{
-       register int            priority;
+       if (decay_amount > decay_limit) {
+               decay_amount = decay_limit;
+       }

 
 

-       if (thread->sched_mode == TH_MODE_TIMESHARE)
-               priority = do_priority_computation(thread);
-       else
-               priority = thread->priority;
+       /* start with base priority */
+       int priority = thread->base_pri - decay_amount;

 
 

-       if ((!(thread->sched_flags & TH_SFLAG_PROMOTED_MASK) || (priority > thread->sched_pri)) &&
-               (!(thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) || override_depress)) {
-               set_sched_pri(thread, priority);
+       if (priority < MAXPRI_THROTTLE) {
+               if (thread->task->max_priority > MAXPRI_THROTTLE) {
+                       priority = MAXPRI_THROTTLE;
+               } else if (priority < MINPRI_USER) {
+                       priority = MINPRI_USER;
+               }
+       } else if (priority > MAXPRI_KERNEL) {
+               priority = MAXPRI_KERNEL;

        }
        }

+
+       return priority;

 }
 
 }
 

-/*
- *     compute_my_priority:
- *
- *     Reset the scheduled priority for
- *     a timesharing thread.
- *
- *     Only for use on the current thread
- *     if timesharing and not depressed.
- *
- *     Called with the thread locked.
- */
-void
-compute_my_priority(
-       register thread_t       thread)
+#else /* CONFIG_EMBEDDED */
+
+int
+sched_compute_timeshare_priority(thread_t thread)

 {
 {

-       register int            priority;
+       /* start with base priority */
+       int priority = thread->base_pri - (thread->sched_usage >> thread->pri_shift);

 
 

-       priority = do_priority_computation(thread);
-       assert(thread->runq == PROCESSOR_NULL);
+       if (priority < MINPRI_USER)
+               priority = MINPRI_USER;
+       else if (priority > MAXPRI_KERNEL)
+               priority = MAXPRI_KERNEL;

 
 

-       if (priority != thread->sched_pri) {
-               KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_DECAY_PRIORITY)|DBG_FUNC_NONE,
-                            (uintptr_t)thread_tid(thread),
-                            thread->priority,
-                            thread->sched_pri,
-                            priority,
-                            0);
-       }
-       thread->sched_pri = priority;
+       return priority;

 }
 
 }
 

+#endif /* CONFIG_EMBEDDED */
+

 /*
  *     can_update_priority
  *
 /*
  *     can_update_priority
  *


@@ -387,20 +503,16 @@ can_update_priority(
  */
 void
 update_priority(
  */
 void
 update_priority(

-       register thread_t       thread)
+       thread_t        thread)

 {
 {

-       register unsigned       ticks;
-       register uint32_t       delta;
+       uint32_t ticks, delta;

 
        ticks = sched_tick - thread->sched_stamp;
        assert(ticks != 0);
 
        ticks = sched_tick - thread->sched_stamp;
        assert(ticks != 0);

+

        thread->sched_stamp += ticks;
        thread->sched_stamp += ticks;

-       if (sched_use_combined_fgbg_decay)
-               thread->pri_shift = sched_combined_fgbg_pri_shift;
-       else if (thread->sched_flags & TH_SFLAG_THROTTLED)
-               thread->pri_shift = sched_background_pri_shift;
-       else
-               thread->pri_shift = sched_pri_shift;
+
+       thread->pri_shift = sched_pri_shifts[thread->th_sched_bucket];

 
        /* If requested, accelerate aging of sched_usage */
        if (sched_decay_usage_age_factor > 1)
 
        /* If requested, accelerate aging of sched_usage */
        if (sched_decay_usage_age_factor > 1)


@@ -411,8 +523,6 @@ update_priority(
         */
        thread_timer_delta(thread, delta);
        if (ticks < SCHED_DECAY_TICKS) {
         */
        thread_timer_delta(thread, delta);
        if (ticks < SCHED_DECAY_TICKS) {

-               register struct shift_data      *shiftp;
-

                /*
                 *      Accumulate timesharing usage only
                 *      during contention for processor
                /*
                 *      Accumulate timesharing usage only
                 *      during contention for processor


@@ -424,25 +534,20 @@ update_priority(
                thread->cpu_usage += delta + thread->cpu_delta;
                thread->cpu_delta = 0;
 
                thread->cpu_usage += delta + thread->cpu_delta;
                thread->cpu_delta = 0;
 

-               shiftp = &sched_decay_shifts[ticks];
+               struct shift_data *shiftp = &sched_decay_shifts[ticks];
+

                if (shiftp->shift2 > 0) {
                if (shiftp->shift2 > 0) {

-                   thread->cpu_usage =
-                                               (thread->cpu_usage >> shiftp->shift1) +
-                                               (thread->cpu_usage >> shiftp->shift2);
-                   thread->sched_usage =
-                                               (thread->sched_usage >> shiftp->shift1) +
-                                               (thread->sched_usage >> shiftp->shift2);
-               }
-               else {
-                   thread->cpu_usage =
-                                               (thread->cpu_usage >> shiftp->shift1) -
-                                               (thread->cpu_usage >> -(shiftp->shift2));
-                   thread->sched_usage =
-                                               (thread->sched_usage >> shiftp->shift1) -
-                                               (thread->sched_usage >> -(shiftp->shift2));
+                       thread->cpu_usage =   (thread->cpu_usage >> shiftp->shift1) +
+                                             (thread->cpu_usage >> shiftp->shift2);
+                       thread->sched_usage = (thread->sched_usage >> shiftp->shift1) +
+                                             (thread->sched_usage >> shiftp->shift2);
+               } else {
+                       thread->cpu_usage =   (thread->cpu_usage >>   shiftp->shift1) -
+                                             (thread->cpu_usage >> -(shiftp->shift2));
+                       thread->sched_usage = (thread->sched_usage >>   shiftp->shift1) -
+                                             (thread->sched_usage >> -(shiftp->shift2));

                }
                }

-       }
-       else {
+       } else {

                thread->cpu_usage = thread->cpu_delta = 0;
                thread->sched_usage = 0;
        }
                thread->cpu_usage = thread->cpu_delta = 0;
                thread->sched_usage = 0;
        }


@@ -455,127 +560,131 @@ update_priority(
                sched_thread_mode_undemote(thread, TH_SFLAG_FAILSAFE);
        }
 
                sched_thread_mode_undemote(thread, TH_SFLAG_FAILSAFE);
        }
 

-

        /*
         *      Recompute scheduled priority if appropriate.
         */
        /*
         *      Recompute scheduled priority if appropriate.
         */

-       if (    (thread->sched_mode == TH_MODE_TIMESHARE)       &&
-                       !(thread->sched_flags & TH_SFLAG_PROMOTED_MASK) &&
-                       !(thread->sched_flags & TH_SFLAG_DEPRESSED_MASK)                ) {
-               register int            new_pri;
-
-               new_pri = do_priority_computation(thread);
-               if (new_pri != thread->sched_pri) {
-                       boolean_t               removed = thread_run_queue_remove(thread);
-
-#if 0
-                       if (sched_use_combined_fgbg_decay && ((thread)->task->max_priority > MAXPRI_THROTTLE) && (new_pri == MAXPRI_THROTTLE)) {
-                               /* with the alternate (new) algorithm, would we have decayed this far? */
-                               int alt_pri = thread->priority - (thread->sched_usage >> sched_pri_shift);
-                               if ((alt_pri > new_pri) && (sched_background_count > 0)) {
-                                       printf("thread %p would have decayed to only %d instead of %d\n", thread, alt_pri, new_pri);
-                               }
-                       }
-#endif
+       if (thread->sched_mode == TH_MODE_TIMESHARE) {
+               int priority = sched_compute_timeshare_priority(thread);
+
+               /*
+                * Adjust the scheduled priority like thread_recompute_sched_pri,
+                * except without setting an AST.
+                */
+               if ((!(thread->sched_flags & TH_SFLAG_PROMOTED_MASK)  || (priority > thread->sched_pri)) &&
+                   (!(thread->sched_flags & TH_SFLAG_DEPRESSED_MASK)) &&
+                   priority != thread->sched_pri) {
+
+                       boolean_t removed = thread_run_queue_remove(thread);

 
 

-                       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_DECAY_PRIORITY)|DBG_FUNC_NONE,
-                                                         (uintptr_t)thread_tid(thread),
-                                                         thread->priority,
-                                                         thread->sched_pri,
-                                                         new_pri,
-                                                         0);
-                       thread->sched_pri = new_pri;
+                       thread->sched_pri = priority;
+
+                       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_CHANGE_PRIORITY),
+                                             (uintptr_t)thread_tid(thread),
+                                             thread->base_pri,
+                                             thread->sched_pri,
+                                             0, /* eventually, 'reason' */
+                                             0);

 
                        if (removed)
 
                        if (removed)

-                               thread_setrun(thread, SCHED_TAILQ);
+                               thread_run_queue_reinsert(thread, SCHED_TAILQ);

                }
        }
                }
        }

-       
+

        return;
 }
 
 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
 
        return;
 }
 
 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
 

-#if MACH_ASSERT
-/* sched_mode == TH_MODE_TIMESHARE controls whether a thread has a timeshare count when it has a run count */

 
 

-void sched_share_incr(thread_t thread) {
-       assert((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN);
-       assert(thread->sched_mode == TH_MODE_TIMESHARE);
-       assert(thread->SHARE_COUNT == 0);
-       thread->SHARE_COUNT++;
-       (void)hw_atomic_add(&sched_share_count, 1);
+/*
+ * TH_BUCKET_RUN is a count of *all* runnable non-idle threads.
+ * Each other bucket is a count of the runnable non-idle threads
+ * with that property.
+ */
+volatile uint32_t       sched_run_buckets[TH_BUCKET_MAX];
+
+static void
+sched_incr_bucket(sched_bucket_t bucket)
+{
+       assert(bucket >= TH_BUCKET_FIXPRI &&
+              bucket <= TH_BUCKET_SHARE_BG);
+
+       hw_atomic_add(&sched_run_buckets[bucket], 1);

 }
 
 }
 

-void sched_share_decr(thread_t thread) {
-       assert((thread->state & (TH_RUN|TH_IDLE)) != TH_RUN || thread->sched_mode != TH_MODE_TIMESHARE);
-       assert(thread->SHARE_COUNT == 1);
-       (void)hw_atomic_sub(&sched_share_count, 1);
-       thread->SHARE_COUNT--;
+static void
+sched_decr_bucket(sched_bucket_t bucket)
+{
+       assert(bucket >= TH_BUCKET_FIXPRI &&
+              bucket <= TH_BUCKET_SHARE_BG);
+
+       assert(sched_run_buckets[bucket] > 0);
+
+       hw_atomic_sub(&sched_run_buckets[bucket], 1);

 }
 
 }
 

-/* TH_SFLAG_THROTTLED controls whether a thread has a background count when it has a run count and a share count */
+/* TH_RUN & !TH_IDLE controls whether a thread has a run count */

 
 

-void sched_background_incr(thread_t thread) {
+uint32_t
+sched_run_incr(thread_t thread)
+{

        assert((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN);
        assert((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN);

-       assert(thread->sched_mode == TH_MODE_TIMESHARE);
-       assert((thread->sched_flags & TH_SFLAG_THROTTLED) == TH_SFLAG_THROTTLED);

 
 

-       assert(thread->BG_COUNT == 0);
-       thread->BG_COUNT++;
-       int val = hw_atomic_add(&sched_background_count, 1);
-       assert(val >= 0);
+       uint32_t new_count = hw_atomic_add(&sched_run_buckets[TH_BUCKET_RUN], 1);

 
 

-       /* Always do the background change while holding a share count */
-       assert(thread->SHARE_COUNT == 1);
-}
+       sched_incr_bucket(thread->th_sched_bucket);

 
 

-void sched_background_decr(thread_t thread) {
-       if ((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN && thread->sched_mode == TH_MODE_TIMESHARE)
-               assert((thread->sched_flags & TH_SFLAG_THROTTLED) != TH_SFLAG_THROTTLED);
-       assert(thread->BG_COUNT == 1);
-       int val = hw_atomic_sub(&sched_background_count, 1);
-       thread->BG_COUNT--;
-       assert(val >= 0);
-       assert(thread->BG_COUNT == 0);
-
-       /* Always do the background change while holding a share count */
-       assert(thread->SHARE_COUNT == 1);
+       return new_count;

 }
 
 }
 

+uint32_t
+sched_run_decr(thread_t thread)
+{
+       assert((thread->state & (TH_RUN|TH_IDLE)) != TH_RUN);

 
 

-void
-assert_thread_sched_count(thread_t thread) {
-       /* Only 0 or 1 are acceptable values */
-       assert(thread->BG_COUNT    == 0 || thread->BG_COUNT    == 1);
-       assert(thread->SHARE_COUNT == 0 || thread->SHARE_COUNT == 1);
-
-       /* BG is only allowed when you already have a share count */
-       if (thread->BG_COUNT == 1)
-               assert(thread->SHARE_COUNT == 1);
-       if (thread->SHARE_COUNT == 0)
-               assert(thread->BG_COUNT == 0);
-
-       if ((thread->state & (TH_RUN|TH_IDLE)) != TH_RUN ||
-           (thread->sched_mode != TH_MODE_TIMESHARE))
-               assert(thread->SHARE_COUNT == 0);
-
-       if ((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN &&
-           (thread->sched_mode == TH_MODE_TIMESHARE))
-               assert(thread->SHARE_COUNT == 1);
-
-       if ((thread->state & (TH_RUN|TH_IDLE)) != TH_RUN ||
-           (thread->sched_mode != TH_MODE_TIMESHARE)    ||
-           !(thread->sched_flags & TH_SFLAG_THROTTLED))
-               assert(thread->BG_COUNT == 0);
-
-       if ((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN &&
-           (thread->sched_mode == TH_MODE_TIMESHARE)    &&
-           (thread->sched_flags & TH_SFLAG_THROTTLED))
-               assert(thread->BG_COUNT == 1);
+       sched_decr_bucket(thread->th_sched_bucket);
+
+       uint32_t new_count = hw_atomic_sub(&sched_run_buckets[TH_BUCKET_RUN], 1);
+
+       return new_count;

 }
 
 }
 

-#endif /* MACH_ASSERT */
+static void
+sched_update_thread_bucket(thread_t thread)
+{
+       sched_bucket_t old_bucket = thread->th_sched_bucket;
+       sched_bucket_t new_bucket = TH_BUCKET_RUN;
+
+       switch (thread->sched_mode) {
+       case TH_MODE_FIXED:
+       case TH_MODE_REALTIME:
+               new_bucket = TH_BUCKET_FIXPRI;
+               break;
+
+       case TH_MODE_TIMESHARE:
+               if (thread->base_pri > BASEPRI_UTILITY)
+                       new_bucket = TH_BUCKET_SHARE_FG;
+               else if (thread->base_pri > MAXPRI_THROTTLE)
+                       new_bucket = TH_BUCKET_SHARE_UT;
+               else
+                       new_bucket = TH_BUCKET_SHARE_BG;
+               break;
+
+       default:
+               panic("unexpected mode: %d", thread->sched_mode);
+               break;
+       }
+
+       if (old_bucket != new_bucket) {
+               thread->th_sched_bucket = new_bucket;
+               thread->pri_shift = sched_pri_shifts[new_bucket];
+
+               if ((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN) {
+                       sched_decr_bucket(old_bucket);
+                       sched_incr_bucket(new_bucket);
+               }
+       }
+}

 
 /*
  * Set the thread's true scheduling mode
 
 /*
  * Set the thread's true scheduling mode


@@ -587,42 +696,22 @@ assert_thread_sched_count(thread_t thread) {
 void
 sched_set_thread_mode(thread_t thread, sched_mode_t new_mode)
 {
 void
 sched_set_thread_mode(thread_t thread, sched_mode_t new_mode)
 {

-       assert_thread_sched_count(thread);
-
-       sched_mode_t old_mode = thread->sched_mode;
-
-       thread->sched_mode = new_mode;
+       assert(thread->runq == PROCESSOR_NULL);

 
        switch (new_mode) {
 
        switch (new_mode) {

-               case TH_MODE_FIXED:
-               case TH_MODE_REALTIME:
-                       if (old_mode == TH_MODE_TIMESHARE) {
-                               if ((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN) {
-                                       if (thread->sched_flags & TH_SFLAG_THROTTLED)
-                                               sched_background_decr(thread);
-
-                                       sched_share_decr(thread);
-                               }
-                       }
-                       break;
-
-               case TH_MODE_TIMESHARE:
-                       if (old_mode != TH_MODE_TIMESHARE) {
-                               if ((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN) {
-                                       sched_share_incr(thread);
-
-                                       if (thread->sched_flags & TH_SFLAG_THROTTLED)
-                                               sched_background_incr(thread);
-                               }
-                       }
-                       break;
-
-               default:
-                       panic("unexpected mode: %d", new_mode);
-                       break;
+       case TH_MODE_FIXED:
+       case TH_MODE_REALTIME:
+       case TH_MODE_TIMESHARE:
+               break;
+
+       default:
+               panic("unexpected mode: %d", new_mode);
+               break;

        }
 
        }
 

-       assert_thread_sched_count(thread);
+       thread->sched_mode = new_mode;
+
+       sched_update_thread_bucket(thread);

 }
 
 /*
 }
 
 /*


@@ -633,7 +722,6 @@ sched_thread_mode_demote(thread_t thread, uint32_t reason)
 {
        assert(reason & TH_SFLAG_DEMOTED_MASK);
        assert((thread->sched_flags & reason) != reason);
 {
        assert(reason & TH_SFLAG_DEMOTED_MASK);
        assert((thread->sched_flags & reason) != reason);

-       assert_thread_sched_count(thread);

 
        if (thread->policy_reset)
                return;
 
        if (thread->policy_reset)
                return;


@@ -648,19 +736,16 @@ sched_thread_mode_demote(thread_t thread, uint32_t reason)
 
        boolean_t removed = thread_run_queue_remove(thread);
 
 
        boolean_t removed = thread_run_queue_remove(thread);
 

-       if (thread->sched_mode == TH_MODE_REALTIME)
-               thread->priority = DEPRESSPRI;
-

        thread->sched_flags |= reason;
 
        thread->saved_mode = thread->sched_mode;
 
        sched_set_thread_mode(thread, TH_MODE_TIMESHARE);
 
        thread->sched_flags |= reason;
 
        thread->saved_mode = thread->sched_mode;
 
        sched_set_thread_mode(thread, TH_MODE_TIMESHARE);
 

-       if (removed)
-               thread_setrun(thread, SCHED_TAILQ);
+       thread_recompute_priority(thread);

 
 

-       assert_thread_sched_count(thread);
+       if (removed)
+               thread_run_queue_reinsert(thread, SCHED_TAILQ);

 }
 
 /*
 }
 
 /*


@@ -675,8 +760,6 @@ sched_thread_mode_undemote(thread_t thread, uint32_t reason)
        assert(thread->sched_mode == TH_MODE_TIMESHARE);
        assert(thread->policy_reset == 0);
 
        assert(thread->sched_mode == TH_MODE_TIMESHARE);
        assert(thread->policy_reset == 0);
 

-       assert_thread_sched_count(thread);
-

        thread->sched_flags &= ~reason;
 
        if (thread->sched_flags & TH_SFLAG_DEMOTED_MASK) {
        thread->sched_flags &= ~reason;
 
        if (thread->sched_flags & TH_SFLAG_DEMOTED_MASK) {


@@ -690,58 +773,10 @@ sched_thread_mode_undemote(thread_t thread, uint32_t reason)
 
        thread->saved_mode = TH_MODE_NONE;
 
 
        thread->saved_mode = TH_MODE_NONE;
 

-       if (thread->sched_mode == TH_MODE_REALTIME) {
-               thread->priority = BASEPRI_RTQUEUES;
-       }
-
-       SCHED(compute_priority)(thread, FALSE);
+       thread_recompute_priority(thread);

 
        if (removed)
 
        if (removed)

-               thread_setrun(thread, SCHED_TAILQ);
+               thread_run_queue_reinsert(thread, SCHED_TAILQ);

 }
 
 }
 

-/*
- * Set the thread to be categorized as 'background'
- * Called with thread mutex and thread lock held
- *
- * TODO: Eventually, 'background' should be a true sched_mode.
- */
-void
-sched_set_thread_throttled(thread_t thread, boolean_t wants_throttle)
-{
-       if (thread->policy_reset)
-               return;
-
-       assert(((thread->sched_flags & TH_SFLAG_THROTTLED) ? TRUE : FALSE) != wants_throttle);
-
-       assert_thread_sched_count(thread);
-
-       /*
-        * When backgrounding a thread, iOS has the semantic that
-        * realtime and fixed priority threads should be demoted
-        * to timeshare background threads.
-        *
-        * On OSX, realtime and fixed priority threads don't lose their mode.
-        */
-
-       if (wants_throttle) {
-               thread->sched_flags |= TH_SFLAG_THROTTLED;
-               if ((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN && thread->sched_mode == TH_MODE_TIMESHARE) {
-                       sched_background_incr(thread);
-               }
-
-               assert_thread_sched_count(thread);
-
-       } else {
-               thread->sched_flags &= ~TH_SFLAG_THROTTLED;
-               if ((thread->state & (TH_RUN|TH_IDLE)) == TH_RUN && thread->sched_mode == TH_MODE_TIMESHARE) {
-                       sched_background_decr(thread);
-               }
-
-               assert_thread_sched_count(thread);
-
-       }
-
-       assert_thread_sched_count(thread);
-}