]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/sched_prim.c
xnu-6153.81.5.tar.gz
[apple/xnu.git] / osfmk / kern / sched_prim.c
index acc37468cb27d4a918533a9413f1429f9b8f5ab0..42e73b4f00bc30258bef6a244fb2a3557a362abd 100644 (file)
@@ -1,8 +1,8 @@
 /*
 /*
- * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2016 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- * 
+ *
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.
- * 
+ *
  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.
- * 
+ *
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.
- * 
+ *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_FREE_COPYRIGHT@
  */
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_FREE_COPYRIGHT@
  */
-/* 
+/*
  * Mach Operating System
  * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
  * All Rights Reserved.
  * Mach Operating System
  * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
  * All Rights Reserved.
- * 
+ *
  * Permission to use, copy, modify and distribute this software and its
  * documentation is hereby granted, provided that both the copyright
  * notice and this permission notice appear in all copies of the
  * software, derivative works or modified versions, and any portions
  * thereof, and that both notices appear in supporting documentation.
  * Permission to use, copy, modify and distribute this software and its
  * documentation is hereby granted, provided that both the copyright
  * notice and this permission notice appear in all copies of the
  * software, derivative works or modified versions, and any portions
  * thereof, and that both notices appear in supporting documentation.
- * 
+ *
  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
- * 
+ *
  * Carnegie Mellon requests users of this software to return to
  * Carnegie Mellon requests users of this software to return to
- * 
+ *
  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  *  School of Computer Science
  *  Carnegie Mellon University
  *  Pittsburgh PA 15213-3890
  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  *  School of Computer Science
  *  Carnegie Mellon University
  *  Pittsburgh PA 15213-3890
- * 
+ *
  * any improvements or extensions that they make and grant Carnegie Mellon
  * the rights to redistribute these changes.
  */
  * any improvements or extensions that they make and grant Carnegie Mellon
  * the rights to redistribute these changes.
  */
  */
 
 #include <debug.h>
  */
 
 #include <debug.h>
-#include <mach_kdb.h>
-
-#include <ddb/db_output.h>
 
 #include <mach/mach_types.h>
 #include <mach/machine.h>
 #include <mach/policy.h>
 #include <mach/sync_policy.h>
 
 #include <mach/mach_types.h>
 #include <mach/machine.h>
 #include <mach/policy.h>
 #include <mach/sync_policy.h>
+#include <mach/thread_act.h>
 
 #include <machine/machine_routines.h>
 #include <machine/sched_param.h>
 #include <machine/machine_cpu.h>
 
 #include <machine/machine_routines.h>
 #include <machine/sched_param.h>
 #include <machine/machine_cpu.h>
+#include <machine/limits.h>
+#include <machine/atomic.h>
+
+#include <machine/commpage.h>
 
 #include <kern/kern_types.h>
 
 #include <kern/kern_types.h>
+#include <kern/backtrace.h>
 #include <kern/clock.h>
 #include <kern/counters.h>
 #include <kern/cpu_number.h>
 #include <kern/cpu_data.h>
 #include <kern/clock.h>
 #include <kern/counters.h>
 #include <kern/cpu_number.h>
 #include <kern/cpu_data.h>
+#include <kern/smp.h>
 #include <kern/debug.h>
 #include <kern/debug.h>
-#include <kern/lock.h>
 #include <kern/macro_help.h>
 #include <kern/machine.h>
 #include <kern/misc_protos.h>
 #include <kern/macro_help.h>
 #include <kern/machine.h>
 #include <kern/misc_protos.h>
+#if MONOTONIC
+#include <kern/monotonic.h>
+#endif /* MONOTONIC */
 #include <kern/processor.h>
 #include <kern/queue.h>
 #include <kern/sched.h>
 #include <kern/sched_prim.h>
 #include <kern/processor.h>
 #include <kern/queue.h>
 #include <kern/sched.h>
 #include <kern/sched_prim.h>
+#include <kern/sfi.h>
 #include <kern/syscall_subr.h>
 #include <kern/task.h>
 #include <kern/thread.h>
 #include <kern/syscall_subr.h>
 #include <kern/task.h>
 #include <kern/thread.h>
-#include <kern/wait_queue.h>
+#include <kern/ledger.h>
+#include <kern/timer_queue.h>
+#include <kern/waitq.h>
+#include <kern/policy_internal.h>
+#include <kern/cpu_quiesce.h>
 
 #include <vm/pmap.h>
 #include <vm/vm_kern.h>
 #include <vm/vm_map.h>
 
 #include <vm/pmap.h>
 #include <vm/vm_kern.h>
 #include <vm/vm_map.h>
+#include <vm/vm_pageout.h>
 
 
-#include <sys/kdebug.h>
+#include <mach/sdt.h>
+#include <mach/mach_host.h>
+#include <mach/host_info.h>
 
 
+#include <sys/kdebug.h>
+#include <kperf/kperf.h>
+#include <kern/kpc.h>
+#include <san/kasan.h>
 #include <kern/pms.h>
 #include <kern/pms.h>
+#include <kern/host.h>
+#include <stdatomic.h>
+
+int
+rt_runq_count(processor_set_t pset)
+{
+       return atomic_load_explicit(&SCHED(rt_runq)(pset)->count, memory_order_relaxed);
+}
+
+void
+rt_runq_count_incr(processor_set_t pset)
+{
+       atomic_fetch_add_explicit(&SCHED(rt_runq)(pset)->count, 1, memory_order_relaxed);
+}
+
+void
+rt_runq_count_decr(processor_set_t pset)
+{
+       atomic_fetch_sub_explicit(&SCHED(rt_runq)(pset)->count, 1, memory_order_relaxed);
+}
+
+#define         DEFAULT_PREEMPTION_RATE         100             /* (1/s) */
+int                     default_preemption_rate = DEFAULT_PREEMPTION_RATE;
+
+#define         DEFAULT_BG_PREEMPTION_RATE      400             /* (1/s) */
+int                     default_bg_preemption_rate = DEFAULT_BG_PREEMPTION_RATE;
+
+#define         MAX_UNSAFE_QUANTA                       800
+int                     max_unsafe_quanta = MAX_UNSAFE_QUANTA;
+
+#define         MAX_POLL_QUANTA                         2
+int                     max_poll_quanta = MAX_POLL_QUANTA;
+
+#define         SCHED_POLL_YIELD_SHIFT          4               /* 1/16 */
+int                     sched_poll_yield_shift = SCHED_POLL_YIELD_SHIFT;
+
+uint64_t        max_poll_computation;
 
 
-#define                DEFAULT_PREEMPTION_RATE         100             /* (1/s) */
-int                    default_preemption_rate = DEFAULT_PREEMPTION_RATE;
+uint64_t        max_unsafe_computation;
+uint64_t        sched_safe_duration;
 
 
-#define                MAX_UNSAFE_QUANTA                       800
-int                    max_unsafe_quanta = MAX_UNSAFE_QUANTA;
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
 
 
-#define                MAX_POLL_QUANTA                         2
-int                    max_poll_quanta = MAX_POLL_QUANTA;
+uint32_t        std_quantum;
+uint32_t        min_std_quantum;
+uint32_t        bg_quantum;
 
 
-#define                SCHED_POLL_YIELD_SHIFT          4               /* 1/16 */
-int                    sched_poll_yield_shift = SCHED_POLL_YIELD_SHIFT;
+uint32_t        std_quantum_us;
+uint32_t        bg_quantum_us;
 
 
-uint64_t       max_unsafe_computation;
-uint32_t       sched_safe_duration;
-uint64_t       max_poll_computation;
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
 
 
-uint32_t       std_quantum;
-uint32_t       min_std_quantum;
+uint32_t        thread_depress_time;
+uint32_t        default_timeshare_computation;
+uint32_t        default_timeshare_constraint;
 
 
-uint32_t       std_quantum_us;
+uint32_t        max_rt_quantum;
+uint32_t        min_rt_quantum;
 
 
-uint32_t       max_rt_quantum;
-uint32_t       min_rt_quantum;
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
 
 
-uint32_t       sched_cswtime;
+unsigned                sched_tick;
+uint32_t                sched_tick_interval;
 
 
-static uint32_t                delay_idle_limit, delay_idle_spin;
-static processor_t     delay_idle(
-                                               processor_t             processor,
-                                               thread_t                self);
+/* Timeshare load calculation interval (15ms) */
+uint32_t                sched_load_compute_interval_us = 15000;
+uint64_t                sched_load_compute_interval_abs;
+static _Atomic uint64_t sched_load_compute_deadline;
 
 
-unsigned       sched_tick;
-uint32_t       sched_tick_interval;
+uint32_t        sched_pri_shifts[TH_BUCKET_MAX];
+uint32_t        sched_fixed_shift;
 
 
-uint32_t       sched_pri_shift;
+uint32_t        sched_decay_usage_age_factor = 1; /* accelerate 5/8^n usage aging */
+
+/* Allow foreground to decay past default to resolve inversions */
+#define DEFAULT_DECAY_BAND_LIMIT ((BASEPRI_FOREGROUND - BASEPRI_DEFAULT) + 2)
+int             sched_pri_decay_band_limit = DEFAULT_DECAY_BAND_LIMIT;
+
+/* Defaults for timer deadline profiling */
+#define TIMER_DEADLINE_TRACKING_BIN_1_DEFAULT 2000000 /* Timers with deadlines <=
+                                                      * 2ms */
+#define TIMER_DEADLINE_TRACKING_BIN_2_DEFAULT 5000000 /* Timers with deadlines
+                                                      *   <= 5ms */
+
+uint64_t timer_deadline_tracking_bin_1;
+uint64_t timer_deadline_tracking_bin_2;
+
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
+
+thread_t sched_maintenance_thread;
+
+/* interrupts disabled lock to guard recommended cores state */
+decl_simple_lock_data(static, sched_recommended_cores_lock);
+static uint64_t    usercontrol_requested_recommended_cores = ALL_CORES_RECOMMENDED;
+static void sched_update_recommended_cores(uint64_t recommended_cores);
+
+#if __arm__ || __arm64__
+static void sched_recommended_cores_maintenance(void);
+uint64_t    perfcontrol_failsafe_starvation_threshold;
+extern char *proc_name_address(struct proc *p);
+#endif /* __arm__ || __arm64__ */
+
+uint64_t        sched_one_second_interval;
 
 /* Forwards */
 
 /* Forwards */
-void           wait_queues_init(void);
 
 
-static void            load_shift_init(void);
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
 
 
-static thread_t        choose_thread(
-                                       processor_set_t         pset,
-                                       processor_t                     processor);
+static void load_shift_init(void);
+static void preempt_pri_init(void);
 
 
-static void            thread_update_scan(void);
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
 
 
-#if    DEBUG
-static
-boolean_t      thread_runnable(
-                               thread_t                thread);
+thread_t        processor_idle(
+       thread_t                        thread,
+       processor_t                     processor);
 
 
-#endif /*DEBUG*/
+static ast_t
+csw_check_locked(
+       thread_t        thread,
+       processor_t     processor,
+       processor_set_t pset,
+       ast_t           check_reason);
 
 
+static void processor_setrun(
+       processor_t                    processor,
+       thread_t                       thread,
+       integer_t                      options);
 
 
-/*
- *     State machine
- *
- * states are combinations of:
- *  R  running
- *  W  waiting (or on wait queue)
- *  N  non-interruptible
- *  O  swapped out
- *  I  being swapped in
- *
- * init        action 
- *     assert_wait thread_block    clear_wait          swapout swapin
- *
- * R   RW, RWN     R;   setrun     -                   -
- * RN  RWN         RN;  setrun     -                   -
- *
- * RW              W               R                   -
- * RWN             WN              RN                  -
- *
- * W                               R;   setrun         WO
- * WN                              RN;  setrun         -
- *
- * RO                              -                   -       R
- *
- */
+static void
+sched_realtime_timebase_init(void);
+
+static void
+sched_timer_deadline_tracking_init(void);
+
+#if     DEBUG
+extern int debug_task;
+#define TLOG(a, fmt, args...) if(debug_task & a) kprintf(fmt, ## args)
+#else
+#define TLOG(a, fmt, args...) do {} while (0)
+#endif
+
+static processor_t
+thread_bind_internal(
+       thread_t                thread,
+       processor_t             processor);
+
+static void
+sched_vm_group_maintenance(void);
+
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+int8_t          sched_load_shifts[NRQS];
+bitmap_t        sched_preempt_pri[BITMAP_LEN(NRQS_MAX)];
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
 
 /*
 
 /*
- *     Waiting protocols and implementation:
- *
- *     Each thread may be waiting for exactly one event; this event
- *     is set using assert_wait().  That thread may be awakened either
- *     by performing a thread_wakeup_prim() on its event,
- *     or by directly waking that thread up with clear_wait().
+ * Statically allocate a buffer to hold the longest possible
+ * scheduler description string, as currently implemented.
+ * bsd/kern/kern_sysctl.c has a corresponding definition in bsd/
+ * to export to userspace via sysctl(3). If either version
+ * changes, update the other.
  *
  *
- *     The implementation of wait events uses a hash table.  Each
- *     bucket is queue of threads having the same hash function
- *     value; the chain for the queue (linked list) is the run queue
- *     field.  [It is not possible to be waiting and runnable at the
- *     same time.]
- *
- *     Locks on both the thread and on the hash buckets govern the
- *     wait event field and the queue chain field.  Because wakeup
- *     operations only have the event as an argument, the event hash
- *     bucket must be locked before any thread.
- *
- *     Scheduling operations may also occur at interrupt level; therefore,
- *     interrupts below splsched() must be prevented when holding
- *     thread or hash bucket locks.
- *
- *     The wait event hash table declarations are as follows:
+ * Note that in addition to being an upper bound on the strings
+ * in the kernel, it's also an exact parameter to PE_get_default(),
+ * which interrogates the device tree on some platforms. That
+ * API requires the caller know the exact size of the device tree
+ * property, so we need both a legacy size (32) and the current size
+ * (48) to deal with old and new device trees. The device tree property
+ * is similarly padded to a fixed size so that the same kernel image
+ * can run on multiple devices with different schedulers configured
+ * in the device tree.
  */
  */
+char sched_string[SCHED_STRING_MAX_LENGTH];
+
+uint32_t sched_debug_flags = SCHED_DEBUG_FLAG_CHOOSE_PROCESSOR_TRACEPOINTS;
 
 
-#define NUMQUEUES      59
+/* Global flag which indicates whether Background Stepper Context is enabled */
+static int cpu_throttle_enabled = 1;
 
 
-struct wait_queue wait_queues[NUMQUEUES];
+void
+sched_init(void)
+{
+       kprintf("Scheduler: Default of %s\n", SCHED(sched_name));
+
+       if (!PE_parse_boot_argn("sched_pri_decay_limit", &sched_pri_decay_band_limit, sizeof(sched_pri_decay_band_limit))) {
+               /* No boot-args, check in device tree */
+               if (!PE_get_default("kern.sched_pri_decay_limit",
+                   &sched_pri_decay_band_limit,
+                   sizeof(sched_pri_decay_band_limit))) {
+                       /* Allow decay all the way to normal limits */
+                       sched_pri_decay_band_limit = DEFAULT_DECAY_BAND_LIMIT;
+               }
+       }
+
+       kprintf("Setting scheduler priority decay band limit %d\n", sched_pri_decay_band_limit);
+
+       if (PE_parse_boot_argn("sched_debug", &sched_debug_flags, sizeof(sched_debug_flags))) {
+               kprintf("Scheduler: Debug flags 0x%08x\n", sched_debug_flags);
+       }
+       strlcpy(sched_string, SCHED(sched_name), sizeof(sched_string));
 
 
-#define wait_hash(event) \
-       ((((int)(event) < 0)? ~(int)(event): (int)(event)) % NUMQUEUES)
+       cpu_quiescent_counter_init();
 
 
-int8_t         sched_load_shifts[NRQS];
+       SCHED(init)();
+       SCHED(rt_init)(&pset0);
+       sched_timer_deadline_tracking_init();
+
+       SCHED(pset_init)(&pset0);
+       SCHED(processor_init)(master_processor);
+}
 
 void
 
 void
-sched_init(void)
+sched_timebase_init(void)
+{
+       uint64_t        abstime;
+
+       clock_interval_to_absolutetime_interval(1, NSEC_PER_SEC, &abstime);
+       sched_one_second_interval = abstime;
+
+       SCHED(timebase_init)();
+       sched_realtime_timebase_init();
+}
+
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+
+void
+sched_timeshare_init(void)
 {
        /*
         * Calculate the timeslicing quantum
         * in us.
         */
 {
        /*
         * Calculate the timeslicing quantum
         * in us.
         */
-       if (default_preemption_rate < 1)
+       if (default_preemption_rate < 1) {
                default_preemption_rate = DEFAULT_PREEMPTION_RATE;
                default_preemption_rate = DEFAULT_PREEMPTION_RATE;
+       }
        std_quantum_us = (1000 * 1000) / default_preemption_rate;
 
        printf("standard timeslicing quantum is %d us\n", std_quantum_us);
 
        std_quantum_us = (1000 * 1000) / default_preemption_rate;
 
        printf("standard timeslicing quantum is %d us\n", std_quantum_us);
 
-       sched_safe_duration = (2 * max_unsafe_quanta / default_preemption_rate) *
-                                                                                       (1 << SCHED_TICK_SHIFT);
+       if (default_bg_preemption_rate < 1) {
+               default_bg_preemption_rate = DEFAULT_BG_PREEMPTION_RATE;
+       }
+       bg_quantum_us = (1000 * 1000) / default_bg_preemption_rate;
+
+       printf("standard background quantum is %d us\n", bg_quantum_us);
 
 
-       wait_queues_init();
        load_shift_init();
        load_shift_init();
-       pset_init(&default_pset);
+       preempt_pri_init();
        sched_tick = 0;
        sched_tick = 0;
-       ast_init();
 }
 
 void
 }
 
 void
-sched_timebase_init(void)
+sched_timeshare_timebase_init(void)
 {
 {
-       uint64_t        abstime;
-       uint32_t        shift;
+       uint64_t        abstime;
+       uint32_t        shift;
 
        /* standard timeslicing quantum */
        clock_interval_to_absolutetime_interval(
 
        /* standard timeslicing quantum */
        clock_interval_to_absolutetime_interval(
-                                                       std_quantum_us, NSEC_PER_USEC, &abstime);
+               std_quantum_us, NSEC_PER_USEC, &abstime);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       std_quantum = abstime;
+       std_quantum = (uint32_t)abstime;
 
        /* smallest remaining quantum (250 us) */
        clock_interval_to_absolutetime_interval(250, NSEC_PER_USEC, &abstime);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
 
        /* smallest remaining quantum (250 us) */
        clock_interval_to_absolutetime_interval(250, NSEC_PER_USEC, &abstime);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       min_std_quantum = abstime;
-
-       /* smallest rt computaton (50 us) */
-       clock_interval_to_absolutetime_interval(50, NSEC_PER_USEC, &abstime);
-       assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       min_rt_quantum = abstime;
+       min_std_quantum = (uint32_t)abstime;
 
 
-       /* maximum rt computation (50 ms) */
+       /* quantum for background tasks */
        clock_interval_to_absolutetime_interval(
        clock_interval_to_absolutetime_interval(
-                                                       50, 1000*NSEC_PER_USEC, &abstime);
+               bg_quantum_us, NSEC_PER_USEC, &abstime);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       max_rt_quantum = abstime;
+       bg_quantum = (uint32_t)abstime;
 
        /* scheduler tick interval */
        clock_interval_to_absolutetime_interval(USEC_PER_SEC >> SCHED_TICK_SHIFT,
 
        /* scheduler tick interval */
        clock_interval_to_absolutetime_interval(USEC_PER_SEC >> SCHED_TICK_SHIFT,
-                                                                                                       NSEC_PER_USEC, &abstime);
+           NSEC_PER_USEC, &abstime);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       sched_tick_interval = abstime;
+       sched_tick_interval = (uint32_t)abstime;
+
+       /* timeshare load calculation interval & deadline initialization */
+       clock_interval_to_absolutetime_interval(sched_load_compute_interval_us, NSEC_PER_USEC, &sched_load_compute_interval_abs);
+       sched_load_compute_deadline = sched_load_compute_interval_abs;
 
        /*
         * Compute conversion factor from usage to
         * timesharing priorities with 5/8 ** n aging.
         */
        abstime = (abstime * 5) / 3;
 
        /*
         * Compute conversion factor from usage to
         * timesharing priorities with 5/8 ** n aging.
         */
        abstime = (abstime * 5) / 3;
-       for (shift = 0; abstime > BASEPRI_DEFAULT; ++shift)
+       for (shift = 0; abstime > BASEPRI_DEFAULT; ++shift) {
                abstime >>= 1;
                abstime >>= 1;
-       sched_pri_shift = shift;
+       }
+       sched_fixed_shift = shift;
+
+       for (uint32_t i = 0; i < TH_BUCKET_MAX; i++) {
+               sched_pri_shifts[i] = INT8_MAX;
+       }
 
 
-       max_unsafe_computation = max_unsafe_quanta * std_quantum;
-       max_poll_computation = max_poll_quanta * std_quantum;
+       max_unsafe_computation = ((uint64_t)max_unsafe_quanta) * std_quantum;
+       sched_safe_duration = 2 * ((uint64_t)max_unsafe_quanta) * std_quantum;
 
 
-       /* delay idle constant(s) (60, 1 us) */
-       clock_interval_to_absolutetime_interval(60, NSEC_PER_USEC, &abstime);
+       max_poll_computation = ((uint64_t)max_poll_quanta) * std_quantum;
+       thread_depress_time = 1 * std_quantum;
+       default_timeshare_computation = std_quantum / 2;
+       default_timeshare_constraint = std_quantum;
+
+#if __arm__ || __arm64__
+       perfcontrol_failsafe_starvation_threshold = (2 * sched_tick_interval);
+#endif /* __arm__ || __arm64__ */
+}
+
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
+
+void
+pset_rt_init(processor_set_t pset)
+{
+       rt_lock_init(pset);
+
+       os_atomic_init(&pset->rt_runq.count, 0);
+       queue_init(&pset->rt_runq.queue);
+       memset(&pset->rt_runq.runq_stats, 0, sizeof pset->rt_runq.runq_stats);
+}
+
+rt_queue_t
+sched_rtglobal_runq(processor_set_t pset)
+{
+       (void)pset;
+
+       return &pset0.rt_runq;
+}
+
+void
+sched_rtglobal_init(processor_set_t pset)
+{
+       if (pset == &pset0) {
+               return pset_rt_init(pset);
+       }
+
+       /* Only pset0 rt_runq is used, so make it easy to detect
+        * buggy accesses to others.
+        */
+       memset(&pset->rt_runq, 0xfd, sizeof pset->rt_runq);
+}
+
+void
+sched_rtglobal_queue_shutdown(processor_t processor)
+{
+       (void)processor;
+}
+
+static void
+sched_realtime_timebase_init(void)
+{
+       uint64_t abstime;
+
+       /* smallest rt computaton (50 us) */
+       clock_interval_to_absolutetime_interval(50, NSEC_PER_USEC, &abstime);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       delay_idle_limit = abstime;
+       min_rt_quantum = (uint32_t)abstime;
 
 
-       clock_interval_to_absolutetime_interval(1, NSEC_PER_USEC, &abstime);
+       /* maximum rt computation (50 ms) */
+       clock_interval_to_absolutetime_interval(
+               50, 1000 * NSEC_PER_USEC, &abstime);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       delay_idle_spin = abstime;
+       max_rt_quantum = (uint32_t)abstime;
 }
 
 void
 }
 
 void
-wait_queues_init(void)
+sched_check_spill(processor_set_t pset, thread_t thread)
 {
 {
-       register int    i;
+       (void)pset;
+       (void)thread;
 
 
-       for (i = 0; i < NUMQUEUES; i++) {
-               wait_queue_init(&wait_queues[i], SYNC_POLICY_FIFO);
-       }
+       return;
 }
 
 }
 
+bool
+sched_thread_should_yield(processor_t processor, thread_t thread)
+{
+       (void)thread;
+
+       return !SCHED(processor_queue_empty)(processor) || rt_runq_count(processor->processor_set) > 0;
+}
+
+/* Default implementations of .steal_thread_enabled */
+bool
+sched_steal_thread_DISABLED(processor_set_t pset)
+{
+       (void)pset;
+       return false;
+}
+
+bool
+sched_steal_thread_enabled(processor_set_t pset)
+{
+       return pset->node->pset_count > 1;
+}
+
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+
 /*
  * Set up values for timeshare
  * loading factors.
 /*
  * Set up values for timeshare
  * loading factors.
@@ -317,98 +529,85 @@ wait_queues_init(void)
 static void
 load_shift_init(void)
 {
 static void
 load_shift_init(void)
 {
-       int8_t          k, *p = sched_load_shifts;
-       uint32_t        i, j;
+       int8_t          k, *p = sched_load_shifts;
+       uint32_t        i, j;
 
 
-       *p++ = INT8_MIN; *p++ = 0;
+       uint32_t        sched_decay_penalty = 1;
 
 
-       for (i = j = 2, k = 1; i < NRQS; ++k) {
-               for (j <<= 1; i < j; ++i)
-                       *p++ = k;
+       if (PE_parse_boot_argn("sched_decay_penalty", &sched_decay_penalty, sizeof(sched_decay_penalty))) {
+               kprintf("Overriding scheduler decay penalty %u\n", sched_decay_penalty);
        }
        }
-}
 
 
-/*
- *     Thread wait timer expiration.
- */
-void
-thread_timer_expire(
-       void                    *p0,
-       __unused void   *p1)
-{
-       thread_t                thread = p0;
-       spl_t                   s;
+       if (PE_parse_boot_argn("sched_decay_usage_age_factor", &sched_decay_usage_age_factor, sizeof(sched_decay_usage_age_factor))) {
+               kprintf("Overriding scheduler decay usage age factor %u\n", sched_decay_usage_age_factor);
+       }
 
 
-       s = splsched();
-       thread_lock(thread);
-       if (--thread->wait_timer_active == 0) {
-               if (thread->wait_timer_is_set) {
-                       thread->wait_timer_is_set = FALSE;
-                       clear_wait_internal(thread, THREAD_TIMED_OUT);
+       if (sched_decay_penalty == 0) {
+               /*
+                * There is no penalty for timeshare threads for using too much
+                * CPU, so set all load shifts to INT8_MIN. Even under high load,
+                * sched_pri_shift will be >INT8_MAX, and there will be no
+                * penalty applied to threads (nor will sched_usage be updated per
+                * thread).
+                */
+               for (i = 0; i < NRQS; i++) {
+                       sched_load_shifts[i] = INT8_MIN;
                }
                }
+
+               return;
        }
        }
-       thread_unlock(thread);
-       splx(s);
-}
 
 
-/*
- *     thread_set_timer:
- *
- *     Set a timer for the current thread, if the thread
- *     is ready to wait.  Must be called between assert_wait()
- *     and thread_block().
- */
-void
-thread_set_timer(
-       uint32_t                interval,
-       uint32_t                scale_factor)
-{
-       thread_t                thread = current_thread();
-       uint64_t                deadline;
-       spl_t                   s;
+       *p++ = INT8_MIN; *p++ = 0;
 
 
-       s = splsched();
-       thread_lock(thread);
-       if ((thread->state & TH_WAIT) != 0) {
-               clock_interval_to_deadline(interval, scale_factor, &deadline);
-               if (!timer_call_enter(&thread->wait_timer, deadline))
-                       thread->wait_timer_active++;
-               thread->wait_timer_is_set = TRUE;
+       /*
+        * For a given system load "i", the per-thread priority
+        * penalty per quantum of CPU usage is ~2^k priority
+        * levels. "sched_decay_penalty" can cause more
+        * array entries to be filled with smaller "k" values
+        */
+       for (i = 2, j = 1 << sched_decay_penalty, k = 1; i < NRQS; ++k) {
+               for (j <<= 1; (i < j) && (i < NRQS); ++i) {
+                       *p++ = k;
+               }
        }
        }
-       thread_unlock(thread);
-       splx(s);
 }
 
 }
 
-void
-thread_set_timer_deadline(
-       uint64_t                deadline)
+static void
+preempt_pri_init(void)
 {
 {
-       thread_t                thread = current_thread();
-       spl_t                   s;
+       bitmap_t *p = sched_preempt_pri;
 
 
-       s = splsched();
-       thread_lock(thread);
-       if ((thread->state & TH_WAIT) != 0) {
-               if (!timer_call_enter(&thread->wait_timer, deadline))
-                       thread->wait_timer_active++;
-               thread->wait_timer_is_set = TRUE;
+       for (int i = BASEPRI_FOREGROUND; i < MINPRI_KERNEL; ++i) {
+               bitmap_set(p, i);
+       }
+
+       for (int i = BASEPRI_PREEMPT; i <= MAXPRI; ++i) {
+               bitmap_set(p, i);
        }
        }
-       thread_unlock(thread);
-       splx(s);
 }
 
 }
 
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
+
+/*
+ *     Thread wait timer expiration.
+ */
 void
 void
-thread_cancel_timer(void)
+thread_timer_expire(
+       void                    *p0,
+       __unused void   *p1)
 {
 {
-       thread_t                thread = current_thread();
-       spl_t                   s;
+       thread_t                thread = p0;
+       spl_t                   s;
+
+       assert_thread_magic(thread);
 
        s = splsched();
        thread_lock(thread);
 
        s = splsched();
        thread_lock(thread);
-       if (thread->wait_timer_is_set) {
-               if (timer_call_cancel(&thread->wait_timer))
-                       thread->wait_timer_active--;
-               thread->wait_timer_is_set = FALSE;
+       if (--thread->wait_timer_active == 0) {
+               if (thread->wait_timer_is_set) {
+                       thread->wait_timer_is_set = FALSE;
+                       clear_wait_internal(thread, THREAD_TIMED_OUT);
+               }
        }
        thread_unlock(thread);
        splx(s);
        }
        thread_unlock(thread);
        splx(s);
@@ -419,75 +618,155 @@ thread_cancel_timer(void)
  *
  *     Unblock thread on wake up.
  *
  *
  *     Unblock thread on wake up.
  *
- *     Returns TRUE if the thread is still running.
+ *     Returns TRUE if the thread should now be placed on the runqueue.
  *
  *     Thread must be locked.
  *
  *     Thread must be locked.
+ *
+ *     Called at splsched().
  */
 boolean_t
 thread_unblock(
  */
 boolean_t
 thread_unblock(
-       thread_t                thread,
-       wait_result_t   wresult)
+       thread_t                thread,
+       wait_result_t   wresult)
 {
 {
-       boolean_t               result = FALSE;
+       boolean_t               ready_for_runq = FALSE;
+       thread_t                cthread = current_thread();
+       uint32_t                new_run_count;
+       int                             old_thread_state;
 
        /*
 
        /*
-        * Set wait_result.
+        *      Set wait_result.
         */
        thread->wait_result = wresult;
 
        /*
         */
        thread->wait_result = wresult;
 
        /*
-        * Cancel pending wait timer.
+        *      Cancel pending wait timer.
         */
        if (thread->wait_timer_is_set) {
         */
        if (thread->wait_timer_is_set) {
-               if (timer_call_cancel(&thread->wait_timer))
+               if (timer_call_cancel(&thread->wait_timer)) {
                        thread->wait_timer_active--;
                        thread->wait_timer_active--;
+               }
                thread->wait_timer_is_set = FALSE;
        }
 
        /*
                thread->wait_timer_is_set = FALSE;
        }
 
        /*
-        * Update scheduling state.
+        *      Update scheduling state: not waiting,
+        *      set running.
         */
         */
-       thread->state &= ~(TH_WAIT|TH_UNINT);
+       old_thread_state = thread->state;
+       thread->state = (old_thread_state | TH_RUN) &
+           ~(TH_WAIT | TH_UNINT | TH_WAIT_REPORT);
 
 
-       if (!(thread->state & TH_RUN)) {
-               thread->state |= TH_RUN;
+       if ((old_thread_state & TH_RUN) == 0) {
+               uint64_t ctime = mach_approximate_time();
+               thread->last_made_runnable_time = thread->last_basepri_change_time = ctime;
+               timer_start(&thread->runnable_timer, ctime);
 
 
+               ready_for_runq = TRUE;
+
+               if (old_thread_state & TH_WAIT_REPORT) {
+                       (*thread->sched_call)(SCHED_CALL_UNBLOCK, thread);
+               }
+
+               /* Update the runnable thread count */
+               new_run_count = SCHED(run_count_incr)(thread);
+       } else {
                /*
                /*
-                * Mark unblocked if call out.
+                * Either the thread is idling in place on another processor,
+                * or it hasn't finished context switching yet.
                 */
                 */
-               if (thread->options & TH_OPT_CALLOUT)
-                       call_thread_unblock();
-
+               assert((thread->state & TH_IDLE) == 0);
                /*
                /*
-                * Update pset run counts.
+                * The run count is only dropped after the context switch completes
+                * and the thread is still waiting, so we should not run_incr here
                 */
                 */
-               pset_run_incr(thread->processor_set);
-               if (thread->sched_mode & TH_MODE_TIMESHARE)
-                       pset_share_incr(thread->processor_set);
+               new_run_count = os_atomic_load(&sched_run_buckets[TH_BUCKET_RUN], relaxed);
        }
        }
-       else
-               result = TRUE;
+
 
        /*
         * Calculate deadline for real-time threads.
         */
 
        /*
         * Calculate deadline for real-time threads.
         */
-       if (thread->sched_mode & TH_MODE_REALTIME) {
-               thread->realtime.deadline = mach_absolute_time();
-               thread->realtime.deadline += thread->realtime.constraint;
+       if (thread->sched_mode == TH_MODE_REALTIME) {
+               uint64_t ctime;
+
+               ctime = mach_absolute_time();
+               thread->realtime.deadline = thread->realtime.constraint + ctime;
        }
 
        /*
         * Clear old quantum, fail-safe computation, etc.
         */
        }
 
        /*
         * Clear old quantum, fail-safe computation, etc.
         */
-       thread->current_quantum = 0;
+       thread->quantum_remaining = 0;
        thread->computation_metered = 0;
        thread->reason = AST_NONE;
        thread->computation_metered = 0;
        thread->reason = AST_NONE;
+       thread->block_hint = kThreadWaitNone;
+
+       /* Obtain power-relevant interrupt and "platform-idle exit" statistics.
+        * We also account for "double hop" thread signaling via
+        * the thread callout infrastructure.
+        * DRK: consider removing the callout wakeup counters in the future
+        * they're present for verification at the moment.
+        */
+       boolean_t aticontext, pidle;
+       ml_get_power_state(&aticontext, &pidle);
+
+       if (__improbable(aticontext && !(thread_get_tag_internal(thread) & THREAD_TAG_CALLOUT))) {
+               DTRACE_SCHED2(iwakeup, struct thread *, thread, struct proc *, thread->task->bsd_info);
+
+               uint64_t ttd = PROCESSOR_DATA(current_processor(), timer_call_ttd);
+
+               if (ttd) {
+                       if (ttd <= timer_deadline_tracking_bin_1) {
+                               thread->thread_timer_wakeups_bin_1++;
+                       } else if (ttd <= timer_deadline_tracking_bin_2) {
+                               thread->thread_timer_wakeups_bin_2++;
+                       }
+               }
+
+               ledger_credit_thread(thread, thread->t_ledger,
+                   task_ledgers.interrupt_wakeups, 1);
+               if (pidle) {
+                       ledger_credit_thread(thread, thread->t_ledger,
+                           task_ledgers.platform_idle_wakeups, 1);
+               }
+       } else if (thread_get_tag_internal(cthread) & THREAD_TAG_CALLOUT) {
+               /* TODO: what about an interrupt that does a wake taken on a callout thread? */
+               if (cthread->callout_woken_from_icontext) {
+                       ledger_credit_thread(thread, thread->t_ledger,
+                           task_ledgers.interrupt_wakeups, 1);
+                       thread->thread_callout_interrupt_wakeups++;
+
+                       if (cthread->callout_woken_from_platform_idle) {
+                               ledger_credit_thread(thread, thread->t_ledger,
+                                   task_ledgers.platform_idle_wakeups, 1);
+                               thread->thread_callout_platform_idle_wakeups++;
+                       }
+
+                       cthread->callout_woke_thread = TRUE;
+               }
+       }
+
+       if (thread_get_tag_internal(thread) & THREAD_TAG_CALLOUT) {
+               thread->callout_woken_from_icontext = aticontext;
+               thread->callout_woken_from_platform_idle = pidle;
+               thread->callout_woke_thread = FALSE;
+       }
+
+#if KPERF
+       if (ready_for_runq) {
+               kperf_make_runnable(thread, aticontext);
+       }
+#endif /* KPERF */
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_MAKE_RUNNABLE) | DBG_FUNC_NONE,
+           (uintptr_t)thread_tid(thread), thread->sched_pri, thread->wait_result,
+           sched_run_buckets[TH_BUCKET_RUN], 0);
 
 
-       KERNEL_DEBUG_CONSTANT(
-               MACHDBG_CODE(DBG_MACH_SCHED,MACH_MAKE_RUNNABLE) | DBG_FUNC_NONE,
-                                       (int)thread, (int)thread->sched_pri, 0, 0, 0);
+       DTRACE_SCHED2(wakeup, struct thread *, thread, struct proc *, thread->task->bsd_info);
 
 
-       return (result);
+       return ready_for_runq;
 }
 
 /*
 }
 
 /*
@@ -497,27 +776,36 @@ thread_unblock(
  *     Conditions:
  *             thread lock held, IPC locks may be held.
  *             thread must have been pulled from wait queue under same lock hold.
  *     Conditions:
  *             thread lock held, IPC locks may be held.
  *             thread must have been pulled from wait queue under same lock hold.
- *  Returns:
+ *             thread must have been waiting
+ *     Returns:
  *             KERN_SUCCESS - Thread was set running
  *             KERN_SUCCESS - Thread was set running
- *             KERN_NOT_WAITING - Thread was not waiting
+ *
+ * TODO: This should return void
  */
 kern_return_t
 thread_go(
  */
 kern_return_t
 thread_go(
-       thread_t                thread,
-       wait_result_t   wresult)
+       thread_t        thread,
+       wait_result_t   wresult)
 {
 {
+       assert_thread_magic(thread);
+
        assert(thread->at_safe_point == FALSE);
        assert(thread->wait_event == NO_EVENT64);
        assert(thread->at_safe_point == FALSE);
        assert(thread->wait_event == NO_EVENT64);
-       assert(thread->wait_queue == WAIT_QUEUE_NULL);
+       assert(thread->waitq == NULL);
+
+       assert(!(thread->state & (TH_TERMINATE | TH_TERMINATE2)));
+       assert(thread->state & TH_WAIT);
 
 
-       if ((thread->state & (TH_WAIT|TH_TERMINATE)) == TH_WAIT) {
-               if (!thread_unblock(thread, wresult))
-                       thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
 
 
-               return (KERN_SUCCESS);
+       if (thread_unblock(thread, wresult)) {
+#if     SCHED_TRACE_THREAD_WAKEUPS
+               backtrace(&thread->thread_wakeup_bt[0],
+                   (sizeof(thread->thread_wakeup_bt) / sizeof(uintptr_t)), NULL);
+#endif
+               thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
        }
 
        }
 
-       return (KERN_NOT_WAITING);
+       return KERN_SUCCESS;
 }
 
 /*
 }
 
 /*
@@ -532,10 +820,17 @@ thread_go(
 __private_extern__
 wait_result_t
 thread_mark_wait_locked(
 __private_extern__
 wait_result_t
 thread_mark_wait_locked(
-       thread_t                        thread,
-       wait_interrupt_t        interruptible)
+       thread_t                        thread,
+       wait_interrupt_t        interruptible_orig)
 {
 {
-       boolean_t               at_safe_point;
+       boolean_t                       at_safe_point;
+       wait_interrupt_t        interruptible = interruptible_orig;
+
+       if (thread->state & TH_IDLE) {
+               panic("Invalid attempt to wait while running the idle thread");
+       }
+
+       assert(!(thread->state & (TH_WAIT | TH_IDLE | TH_UNINT | TH_TERMINATE2 | TH_WAIT_REPORT)));
 
        /*
         *      The thread may have certain types of interrupts/aborts masked
 
        /*
         *      The thread may have certain types of interrupts/aborts masked
@@ -543,24 +838,52 @@ thread_mark_wait_locked(
         *      are OK, we have to honor mask settings (outer-scoped code may
         *      not be able to handle aborts at the moment).
         */
         *      are OK, we have to honor mask settings (outer-scoped code may
         *      not be able to handle aborts at the moment).
         */
-       if (interruptible > (thread->options & TH_OPT_INTMASK))
+       interruptible &= TH_OPT_INTMASK;
+       if (interruptible > (thread->options & TH_OPT_INTMASK)) {
                interruptible = thread->options & TH_OPT_INTMASK;
                interruptible = thread->options & TH_OPT_INTMASK;
+       }
 
        at_safe_point = (interruptible == THREAD_ABORTSAFE);
 
 
        at_safe_point = (interruptible == THREAD_ABORTSAFE);
 
-       if (    interruptible == THREAD_UNINT                   ||
-                       !(thread->state & TH_ABORT)                             ||
-                       (!at_safe_point &&
-                        (thread->state & TH_ABORT_SAFELY))             ) {
-               thread->state |= (interruptible) ? TH_WAIT : (TH_WAIT | TH_UNINT);
+       if (interruptible == THREAD_UNINT ||
+           !(thread->sched_flags & TH_SFLAG_ABORT) ||
+           (!at_safe_point &&
+           (thread->sched_flags & TH_SFLAG_ABORTSAFELY))) {
+               if (!(thread->state & TH_TERMINATE)) {
+                       DTRACE_SCHED(sleep);
+               }
+
+               int state_bits = TH_WAIT;
+               if (!interruptible) {
+                       state_bits |= TH_UNINT;
+               }
+               if (thread->sched_call) {
+                       wait_interrupt_t mask = THREAD_WAIT_NOREPORT_USER;
+                       if (is_kerneltask(thread->task)) {
+                               mask = THREAD_WAIT_NOREPORT_KERNEL;
+                       }
+                       if ((interruptible_orig & mask) == 0) {
+                               state_bits |= TH_WAIT_REPORT;
+                       }
+               }
+               thread->state |= state_bits;
                thread->at_safe_point = at_safe_point;
                thread->at_safe_point = at_safe_point;
-               return (thread->wait_result = THREAD_WAITING);
+
+               /* TODO: pass this through assert_wait instead, have
+                * assert_wait just take a struct as an argument */
+               assert(!thread->block_hint);
+               thread->block_hint = thread->pending_block_hint;
+               thread->pending_block_hint = kThreadWaitNone;
+
+               return thread->wait_result = THREAD_WAITING;
+       } else {
+               if (thread->sched_flags & TH_SFLAG_ABORTSAFELY) {
+                       thread->sched_flags &= ~TH_SFLAG_ABORTED_MASK;
+               }
        }
        }
-       else
-       if (thread->state & TH_ABORT_SAFELY)
-               thread->state &= ~(TH_ABORT|TH_ABORT_SAFELY);
+       thread->pending_block_hint = kThreadWaitNone;
 
 
-       return (thread->wait_result = THREAD_INTERRUPTED);
+       return thread->wait_result = THREAD_INTERRUPTED;
 }
 
 /*
 }
 
 /*
@@ -576,7 +899,7 @@ thread_mark_wait_locked(
  *     Returns:
  *             The old interrupt level for the thread.
  */
  *     Returns:
  *             The old interrupt level for the thread.
  */
-__private_extern__ 
+__private_extern__
 wait_interrupt_t
 thread_interrupt_level(
        wait_interrupt_t new_level)
 wait_interrupt_t
 thread_interrupt_level(
        wait_interrupt_t new_level)
@@ -589,28 +912,6 @@ thread_interrupt_level(
        return result;
 }
 
        return result;
 }
 
-/*
- * Check to see if an assert wait is possible, without actually doing one.
- * This is used by debug code in locks and elsewhere to verify that it is
- * always OK to block when trying to take a blocking lock (since waiting
- * for the actual assert_wait to catch the case may make it hard to detect
- * this case.
- */
-boolean_t
-assert_wait_possible(void)
-{
-
-       thread_t thread;
-
-#if    DEBUG
-       if(debug_mode) return TRUE;             /* Always succeed in debug mode */
-#endif
-       
-       thread = current_thread();
-
-       return (thread == NULL || wait_queue_assert_possible(thread));
-}
-
 /*
  *     assert_wait:
  *
 /*
  *     assert_wait:
  *
@@ -619,213 +920,239 @@ assert_wait_possible(void)
  */
 wait_result_t
 assert_wait(
  */
 wait_result_t
 assert_wait(
-       event_t                         event,
-       wait_interrupt_t        interruptible)
+       event_t                         event,
+       wait_interrupt_t        interruptible)
 {
 {
-       register wait_queue_t   wq;
-       register int            index;
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
+       }
 
 
-       assert(event != NO_EVENT);
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_WAIT) | DBG_FUNC_NONE,
+           VM_KERNEL_UNSLIDE_OR_PERM(event), 0, 0, 0, 0);
 
 
-       index = wait_hash(event);
-       wq = &wait_queues[index];
-       return wait_queue_assert_wait(wq, event, interruptible, 0);
+       struct waitq *waitq;
+       waitq = global_eventq(event);
+       return waitq_assert_wait64(waitq, CAST_EVENT64_T(event), interruptible, TIMEOUT_WAIT_FOREVER);
 }
 
 }
 
-wait_result_t
+/*
+ *     assert_wait_queue:
+ *
+ *     Return the global waitq for the specified event
+ */
+struct waitq *
+assert_wait_queue(
+       event_t                         event)
+{
+       return global_eventq(event);
+}
+
+wait_result_t
 assert_wait_timeout(
 assert_wait_timeout(
-       event_t                         event,
-       wait_interrupt_t        interruptible,
-       uint32_t                        interval,
-       uint32_t                        scale_factor)
+       event_t                         event,
+       wait_interrupt_t        interruptible,
+       uint32_t                        interval,
+       uint32_t                        scale_factor)
 {
 {
-       thread_t                        thread = current_thread();
-       wait_result_t           wresult;
-       wait_queue_t            wqueue;
-       uint64_t                        deadline;
-       spl_t                           s;
+       thread_t                        thread = current_thread();
+       wait_result_t           wresult;
+       uint64_t                        deadline;
+       spl_t                           s;
+
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
+       }
 
 
-       assert(event != NO_EVENT);
-       wqueue = &wait_queues[wait_hash(event)];
+       struct waitq *waitq;
+       waitq = global_eventq(event);
 
        s = splsched();
 
        s = splsched();
-       wait_queue_lock(wqueue);
-       thread_lock(thread);
+       waitq_lock(waitq);
 
        clock_interval_to_deadline(interval, scale_factor, &deadline);
 
        clock_interval_to_deadline(interval, scale_factor, &deadline);
-       wresult = wait_queue_assert_wait64_locked(wqueue, (uint32_t)event,
-                                                                                                       interruptible, deadline, thread);
 
 
-       thread_unlock(thread);
-       wait_queue_unlock(wqueue);
-       splx(s);
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_WAIT) | DBG_FUNC_NONE,
+           VM_KERNEL_UNSLIDE_OR_PERM(event), interruptible, deadline, 0, 0);
 
 
-       return (wresult);
+       wresult = waitq_assert_wait64_locked(waitq, CAST_EVENT64_T(event),
+           interruptible,
+           TIMEOUT_URGENCY_SYS_NORMAL,
+           deadline, TIMEOUT_NO_LEEWAY,
+           thread);
+
+       waitq_unlock(waitq);
+       splx(s);
+       return wresult;
 }
 
 wait_result_t
 }
 
 wait_result_t
-assert_wait_deadline(
-       event_t                         event,
-       wait_interrupt_t        interruptible,
-       uint64_t                        deadline)
+assert_wait_timeout_with_leeway(
+       event_t                         event,
+       wait_interrupt_t        interruptible,
+       wait_timeout_urgency_t  urgency,
+       uint32_t                        interval,
+       uint32_t                        leeway,
+       uint32_t                        scale_factor)
 {
 {
-       thread_t                        thread = current_thread();
-       wait_result_t           wresult;
-       wait_queue_t            wqueue;
-       spl_t                           s;
+       thread_t                        thread = current_thread();
+       wait_result_t           wresult;
+       uint64_t                        deadline;
+       uint64_t                        abstime;
+       uint64_t                        slop;
+       uint64_t                        now;
+       spl_t                           s;
+
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
+       }
 
 
-       assert(event != NO_EVENT);
-       wqueue = &wait_queues[wait_hash(event)];
+       now = mach_absolute_time();
+       clock_interval_to_absolutetime_interval(interval, scale_factor, &abstime);
+       deadline = now + abstime;
+
+       clock_interval_to_absolutetime_interval(leeway, scale_factor, &slop);
+
+       struct waitq *waitq;
+       waitq = global_eventq(event);
 
        s = splsched();
 
        s = splsched();
-       wait_queue_lock(wqueue);
-       thread_lock(thread);
+       waitq_lock(waitq);
 
 
-       wresult = wait_queue_assert_wait64_locked(wqueue, (uint32_t)event,
-                                                                                                       interruptible, deadline, thread);
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_WAIT) | DBG_FUNC_NONE,
+           VM_KERNEL_UNSLIDE_OR_PERM(event), interruptible, deadline, 0, 0);
 
 
-       thread_unlock(thread);
-       wait_queue_unlock(wqueue);
-       splx(s);
+       wresult = waitq_assert_wait64_locked(waitq, CAST_EVENT64_T(event),
+           interruptible,
+           urgency, deadline, slop,
+           thread);
 
 
-       return (wresult);
+       waitq_unlock(waitq);
+       splx(s);
+       return wresult;
 }
 
 }
 
-/*
- *     thread_sleep_fast_usimple_lock:
- *
- *     Cause the current thread to wait until the specified event
- *     occurs.  The specified simple_lock is unlocked before releasing
- *     the cpu and re-acquired as part of waking up.
- *
- *     This is the simple lock sleep interface for components that use a
- *     faster version of simple_lock() than is provided by usimple_lock().
- */
-__private_extern__ wait_result_t
-thread_sleep_fast_usimple_lock(
-       event_t                 event,
-       simple_lock_t           lock,
-       wait_interrupt_t        interruptible)
+wait_result_t
+assert_wait_deadline(
+       event_t                         event,
+       wait_interrupt_t        interruptible,
+       uint64_t                        deadline)
 {
 {
-       wait_result_t res;
+       thread_t                        thread = current_thread();
+       wait_result_t           wresult;
+       spl_t                           s;
 
 
-       res = assert_wait(event, interruptible);
-       if (res == THREAD_WAITING) {
-               simple_unlock(lock);
-               res = thread_block(THREAD_CONTINUE_NULL);
-               simple_lock(lock);
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
        }
        }
-       return res;
-}
 
 
+       struct waitq *waitq;
+       waitq = global_eventq(event);
 
 
-/*
- *     thread_sleep_usimple_lock:
- *
- *     Cause the current thread to wait until the specified event
- *     occurs.  The specified usimple_lock is unlocked before releasing
- *     the cpu and re-acquired as part of waking up.
- *
- *     This is the simple lock sleep interface for components where
- *     simple_lock() is defined in terms of usimple_lock().
- */
-wait_result_t
-thread_sleep_usimple_lock(
-       event_t                 event,
-       usimple_lock_t          lock,
-       wait_interrupt_t        interruptible)
-{
-       wait_result_t res;
+       s = splsched();
+       waitq_lock(waitq);
 
 
-       res = assert_wait(event, interruptible);
-       if (res == THREAD_WAITING) {
-               usimple_unlock(lock);
-               res = thread_block(THREAD_CONTINUE_NULL);
-               usimple_lock(lock);
-       }
-       return res;
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_WAIT) | DBG_FUNC_NONE,
+           VM_KERNEL_UNSLIDE_OR_PERM(event), interruptible, deadline, 0, 0);
+
+       wresult = waitq_assert_wait64_locked(waitq, CAST_EVENT64_T(event),
+           interruptible,
+           TIMEOUT_URGENCY_SYS_NORMAL, deadline,
+           TIMEOUT_NO_LEEWAY, thread);
+       waitq_unlock(waitq);
+       splx(s);
+       return wresult;
 }
 
 }
 
-/*
- *     thread_sleep_mutex:
- *
- *     Cause the current thread to wait until the specified event
- *     occurs.  The specified mutex is unlocked before releasing
- *     the cpu. The mutex will be re-acquired before returning.
- *
- *     JMM - Add hint to make sure mutex is available before rousting
- */
 wait_result_t
 wait_result_t
-thread_sleep_mutex(
-       event_t                 event,
-       mutex_t                 *mutex,
-       wait_interrupt_t interruptible)
+assert_wait_deadline_with_leeway(
+       event_t                         event,
+       wait_interrupt_t        interruptible,
+       wait_timeout_urgency_t  urgency,
+       uint64_t                        deadline,
+       uint64_t                        leeway)
 {
 {
-       wait_result_t   res;
+       thread_t                        thread = current_thread();
+       wait_result_t           wresult;
+       spl_t                           s;
 
 
-       res = assert_wait(event, interruptible);
-       if (res == THREAD_WAITING) {
-               mutex_unlock(mutex);
-               res = thread_block(THREAD_CONTINUE_NULL);
-               mutex_lock(mutex);
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
        }
        }
-       return res;
-}
-  
-/*
- *     thread_sleep_mutex_deadline:
- *
- *     Cause the current thread to wait until the specified event
- *     (or deadline) occurs.  The specified mutex is unlocked before
- *     releasing the cpu. The mutex will be re-acquired before returning.
- */
-wait_result_t
-thread_sleep_mutex_deadline(
-       event_t                 event,
-       mutex_t                 *mutex,
-       uint64_t                deadline,
-       wait_interrupt_t interruptible)
-{
-       wait_result_t   res;
 
 
-       res = assert_wait_deadline(event, interruptible, deadline);
-       if (res == THREAD_WAITING) {
-               mutex_unlock(mutex);
-               res = thread_block(THREAD_CONTINUE_NULL);
-               mutex_lock(mutex);
-       }
-       return res;
+       struct waitq *waitq;
+       waitq = global_eventq(event);
+
+       s = splsched();
+       waitq_lock(waitq);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_WAIT) | DBG_FUNC_NONE,
+           VM_KERNEL_UNSLIDE_OR_PERM(event), interruptible, deadline, 0, 0);
+
+       wresult = waitq_assert_wait64_locked(waitq, CAST_EVENT64_T(event),
+           interruptible,
+           urgency, deadline, leeway,
+           thread);
+       waitq_unlock(waitq);
+       splx(s);
+       return wresult;
 }
 
 /*
 }
 
 /*
- *     thread_sleep_lock_write:
+ * thread_isoncpu:
+ *
+ * Return TRUE if a thread is running on a processor such that an AST
+ * is needed to pull it out of userspace execution, or if executing in
+ * the kernel, bring to a context switch boundary that would cause
+ * thread state to be serialized in the thread PCB.
  *
  *
- *     Cause the current thread to wait until the specified event
- *     occurs.  The specified (write) lock is unlocked before releasing
- *     the cpu. The (write) lock will be re-acquired before returning.
+ * Thread locked, returns the same way. While locked, fields
+ * like "state" cannot change. "runq" can change only from set to unset.
  */
  */
-wait_result_t
-thread_sleep_lock_write(
-       event_t                 event,
-       lock_t                  *lock,
-       wait_interrupt_t interruptible)
+static inline boolean_t
+thread_isoncpu(thread_t thread)
 {
 {
-       wait_result_t   res;
+       /* Not running or runnable */
+       if (!(thread->state & TH_RUN)) {
+               return FALSE;
+       }
+
+       /* Waiting on a runqueue, not currently running */
+       /* TODO: This is invalid - it can get dequeued without thread lock, but not context switched. */
+       if (thread->runq != PROCESSOR_NULL) {
+               return FALSE;
+       }
 
 
-       res = assert_wait(event, interruptible);
-       if (res == THREAD_WAITING) {
-               lock_write_done(lock);
-               res = thread_block(THREAD_CONTINUE_NULL);
-               lock_write(lock);
+       /*
+        * Thread does not have a stack yet
+        * It could be on the stack alloc queue or preparing to be invoked
+        */
+       if (!thread->kernel_stack) {
+               return FALSE;
        }
        }
-       return res;
+
+       /*
+        * Thread must be running on a processor, or
+        * about to run, or just did run. In all these
+        * cases, an AST to the processor is needed
+        * to guarantee that the thread is kicked out
+        * of userspace and the processor has
+        * context switched (and saved register state).
+        */
+       return TRUE;
 }
 
 /*
  * thread_stop:
  *
  * Force a preemption point for a thread and wait
 }
 
 /*
  * thread_stop:
  *
  * Force a preemption point for a thread and wait
- * for it to stop running.  Arbitrates access among
+ * for it to stop running on a CPU. If a stronger
+ * guarantee is requested, wait until no longer
+ * runnable. Arbitrates access among
  * multiple stop requests. (released by unstop)
  *
  * The thread must enter a wait state and stop via a
  * multiple stop requests. (released by unstop)
  *
  * The thread must enter a wait state and stop via a
@@ -835,53 +1162,63 @@ thread_sleep_lock_write(
  */
 boolean_t
 thread_stop(
  */
 boolean_t
 thread_stop(
-       thread_t                thread)
+       thread_t                thread,
+       boolean_t       until_not_runnable)
 {
 {
-       wait_result_t   wresult;
-       spl_t                   s;
+       wait_result_t   wresult;
+       spl_t                   s = splsched();
+       boolean_t               oncpu;
 
 
-       s = splsched();
        wake_lock(thread);
        wake_lock(thread);
+       thread_lock(thread);
 
        while (thread->state & TH_SUSP) {
                thread->wake_active = TRUE;
 
        while (thread->state & TH_SUSP) {
                thread->wake_active = TRUE;
+               thread_unlock(thread);
+
                wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
                wake_unlock(thread);
                splx(s);
 
                wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
                wake_unlock(thread);
                splx(s);
 
-               if (wresult == THREAD_WAITING)
+               if (wresult == THREAD_WAITING) {
                        wresult = thread_block(THREAD_CONTINUE_NULL);
                        wresult = thread_block(THREAD_CONTINUE_NULL);
+               }
 
 
-               if (wresult != THREAD_AWAKENED)
-                       return (FALSE);
+               if (wresult != THREAD_AWAKENED) {
+                       return FALSE;
+               }
 
                s = splsched();
                wake_lock(thread);
 
                s = splsched();
                wake_lock(thread);
+               thread_lock(thread);
        }
 
        }
 
-       thread_lock(thread);
        thread->state |= TH_SUSP;
 
        thread->state |= TH_SUSP;
 
-       while (thread->state & TH_RUN) {
-               processor_t             processor = thread->last_processor;
+       while ((oncpu = thread_isoncpu(thread)) ||
+           (until_not_runnable && (thread->state & TH_RUN))) {
+               processor_t             processor;
 
 
-               if (    processor != PROCESSOR_NULL                                     &&
-                               processor->state == PROCESSOR_RUNNING           &&
-                               processor->active_thread == thread                      )
+               if (oncpu) {
+                       assert(thread->state & TH_RUN);
+                       processor = thread->chosen_processor;
                        cause_ast_check(processor);
                        cause_ast_check(processor);
-               thread_unlock(thread);
+               }
 
                thread->wake_active = TRUE;
 
                thread->wake_active = TRUE;
+               thread_unlock(thread);
+
                wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
                wake_unlock(thread);
                splx(s);
 
                wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
                wake_unlock(thread);
                splx(s);
 
-               if (wresult == THREAD_WAITING)
+               if (wresult == THREAD_WAITING) {
                        wresult = thread_block(THREAD_CONTINUE_NULL);
                        wresult = thread_block(THREAD_CONTINUE_NULL);
+               }
 
                if (wresult != THREAD_AWAKENED) {
                        thread_unstop(thread);
 
                if (wresult != THREAD_AWAKENED) {
                        thread_unstop(thread);
-                       return (FALSE);
+                       return FALSE;
                }
 
                s = splsched();
                }
 
                s = splsched();
@@ -893,7 +1230,14 @@ thread_stop(
        wake_unlock(thread);
        splx(s);
 
        wake_unlock(thread);
        splx(s);
 
-       return (TRUE);
+       /*
+        * We return with the thread unlocked. To prevent it from
+        * transitioning to a runnable state (or from TH_RUN to
+        * being on the CPU), the caller must ensure the thread
+        * is stopped via an external means (such as an AST)
+        */
+
+       return TRUE;
 }
 
 /*
 }
 
 /*
@@ -906,30 +1250,26 @@ thread_stop(
  */
 void
 thread_unstop(
  */
 void
 thread_unstop(
-       thread_t        thread)
+       thread_t        thread)
 {
 {
-       spl_t           s = splsched();
+       spl_t           s = splsched();
 
        wake_lock(thread);
        thread_lock(thread);
 
 
        wake_lock(thread);
        thread_lock(thread);
 
-       if ((thread->state & (TH_RUN|TH_WAIT|TH_SUSP)) == TH_SUSP) {
-               thread->state &= ~TH_SUSP;
-               thread_unblock(thread, THREAD_AWAKENED);
+       assert((thread->state & (TH_RUN | TH_WAIT | TH_SUSP)) != TH_SUSP);
 
 
-               thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
-       }
-       else
        if (thread->state & TH_SUSP) {
                thread->state &= ~TH_SUSP;
 
                if (thread->wake_active) {
                        thread->wake_active = FALSE;
                        thread_unlock(thread);
        if (thread->state & TH_SUSP) {
                thread->state &= ~TH_SUSP;
 
                if (thread->wake_active) {
                        thread->wake_active = FALSE;
                        thread_unlock(thread);
+
+                       thread_wakeup(&thread->wake_active);
                        wake_unlock(thread);
                        splx(s);
 
                        wake_unlock(thread);
                        splx(s);
 
-                       thread_wakeup(&thread->wake_active);
                        return;
                }
        }
                        return;
                }
        }
@@ -947,30 +1287,42 @@ thread_unstop(
  */
 void
 thread_wait(
  */
 void
 thread_wait(
-       thread_t                thread)
+       thread_t        thread,
+       boolean_t       until_not_runnable)
 {
 {
-       wait_result_t   wresult;
-       spl_t                   s = splsched();
+       wait_result_t   wresult;
+       boolean_t       oncpu;
+       processor_t     processor;
+       spl_t           s = splsched();
 
        wake_lock(thread);
        thread_lock(thread);
 
 
        wake_lock(thread);
        thread_lock(thread);
 
-       while (thread->state & TH_RUN) {
-               processor_t             processor = thread->last_processor;
-
-               if (    processor != PROCESSOR_NULL                                     &&
-                               processor->state == PROCESSOR_RUNNING           &&
-                               processor->active_thread == thread                      )
+       /*
+        * Wait until not running on a CPU.  If stronger requirement
+        * desired, wait until not runnable.  Assumption: if thread is
+        * on CPU, then TH_RUN is set, so we're not waiting in any case
+        * where the original, pure "TH_RUN" check would have let us
+        * finish.
+        */
+       while ((oncpu = thread_isoncpu(thread)) ||
+           (until_not_runnable && (thread->state & TH_RUN))) {
+               if (oncpu) {
+                       assert(thread->state & TH_RUN);
+                       processor = thread->chosen_processor;
                        cause_ast_check(processor);
                        cause_ast_check(processor);
-               thread_unlock(thread);
+               }
 
                thread->wake_active = TRUE;
 
                thread->wake_active = TRUE;
+               thread_unlock(thread);
+
                wresult = assert_wait(&thread->wake_active, THREAD_UNINT);
                wake_unlock(thread);
                splx(s);
 
                wresult = assert_wait(&thread->wake_active, THREAD_UNINT);
                wake_unlock(thread);
                splx(s);
 
-               if (wresult == THREAD_WAITING)
+               if (wresult == THREAD_WAITING) {
                        thread_block(THREAD_CONTINUE_NULL);
                        thread_block(THREAD_CONTINUE_NULL);
+               }
 
                s = splsched();
                wake_lock(thread);
 
                s = splsched();
                wake_lock(thread);
@@ -1000,40 +1352,44 @@ thread_wait(
  */
 __private_extern__ kern_return_t
 clear_wait_internal(
  */
 __private_extern__ kern_return_t
 clear_wait_internal(
-       thread_t                thread,
-       wait_result_t   wresult)
+       thread_t                thread,
+       wait_result_t   wresult)
 {
 {
-       wait_queue_t    wq = thread->wait_queue;
-       int                             i = LockTimeOut;
+       uint32_t        i = LockTimeOutUsec;
+       struct waitq *waitq = thread->waitq;
 
        do {
 
        do {
-               if (wresult == THREAD_INTERRUPTED && (thread->state & TH_UNINT))
-                       return (KERN_FAILURE);
+               if (wresult == THREAD_INTERRUPTED && (thread->state & TH_UNINT)) {
+                       return KERN_FAILURE;
+               }
 
 
-               if (wq != WAIT_QUEUE_NULL) {
-                       if (wait_queue_lock_try(wq)) {
-                               wait_queue_pull_thread_locked(wq, thread, TRUE);
-                               /* wait queue unlocked, thread still locked */
-                       }
-                       else {
+               if (waitq != NULL) {
+                       if (!waitq_pull_thread_locked(waitq, thread)) {
                                thread_unlock(thread);
                                delay(1);
                                thread_unlock(thread);
                                delay(1);
-
+                               if (i > 0 && !machine_timeout_suspended()) {
+                                       i--;
+                               }
                                thread_lock(thread);
                                thread_lock(thread);
-                               if (wq != thread->wait_queue)
-                                       return (KERN_NOT_WAITING);
-
+                               if (waitq != thread->waitq) {
+                                       return KERN_NOT_WAITING;
+                               }
                                continue;
                        }
                }
 
                                continue;
                        }
                }
 
-               return (thread_go(thread, wresult));
-       } while (--i > 0);
+               /* TODO: Can we instead assert TH_TERMINATE is not set?  */
+               if ((thread->state & (TH_WAIT | TH_TERMINATE)) == TH_WAIT) {
+                       return thread_go(thread, wresult);
+               } else {
+                       return KERN_NOT_WAITING;
+               }
+       } while (i > 0);
 
 
-       panic("clear_wait_internal: deadlock: thread=0x%x, wq=0x%x, cpu=%d\n",
-                 thread, wq, cpu_number());
+       panic("clear_wait_internal: deadlock: thread=%p, wq=%p, cpu=%d\n",
+           thread, waitq, cpu_number());
 
 
-       return (KERN_FAILURE);
+       return KERN_FAILURE;
 }
 
 
 }
 
 
@@ -1049,11 +1405,11 @@ clear_wait_internal(
  */
 kern_return_t
 clear_wait(
  */
 kern_return_t
 clear_wait(
-       thread_t                thread,
-       wait_result_t   result)
+       thread_t                thread,
+       wait_result_t   result)
 {
        kern_return_t ret;
 {
        kern_return_t ret;
-       spl_t           s;
+       spl_t           s;
 
        s = splsched();
        thread_lock(thread);
 
        s = splsched();
        thread_lock(thread);
@@ -1073,25 +1429,88 @@ clear_wait(
  */
 kern_return_t
 thread_wakeup_prim(
  */
 kern_return_t
 thread_wakeup_prim(
-       event_t                 event,
-       boolean_t               one_thread,
-       wait_result_t   result)
+       event_t          event,
+       boolean_t        one_thread,
+       wait_result_t    result)
+{
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
+       }
+
+       struct waitq *wq = global_eventq(event);
+
+       if (one_thread) {
+               return waitq_wakeup64_one(wq, CAST_EVENT64_T(event), result, WAITQ_ALL_PRIORITIES);
+       } else {
+               return waitq_wakeup64_all(wq, CAST_EVENT64_T(event), result, WAITQ_ALL_PRIORITIES);
+       }
+}
+
+/*
+ * Wakeup a specified thread if and only if it's waiting for this event
+ */
+kern_return_t
+thread_wakeup_thread(
+       event_t         event,
+       thread_t        thread)
+{
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
+       }
+
+       if (__improbable(thread == THREAD_NULL)) {
+               panic("%s() called with THREAD_NULL", __func__);
+       }
+
+       struct waitq *wq = global_eventq(event);
+
+       return waitq_wakeup64_thread(wq, CAST_EVENT64_T(event), thread, THREAD_AWAKENED);
+}
+
+/*
+ * Wakeup a thread waiting on an event and promote it to a priority.
+ *
+ * Requires woken thread to un-promote itself when done.
+ */
+kern_return_t
+thread_wakeup_one_with_pri(
+       event_t      event,
+       int          priority)
+{
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
+       }
+
+       struct waitq *wq = global_eventq(event);
+
+       return waitq_wakeup64_one(wq, CAST_EVENT64_T(event), THREAD_AWAKENED, priority);
+}
+
+/*
+ * Wakeup a thread waiting on an event,
+ * promote it to a priority,
+ * and return a reference to the woken thread.
+ *
+ * Requires woken thread to un-promote itself when done.
+ */
+thread_t
+thread_wakeup_identify(event_t  event,
+    int      priority)
 {
 {
-       register wait_queue_t   wq;
-       register int                    index;
+       if (__improbable(event == NO_EVENT)) {
+               panic("%s() called with NO_EVENT", __func__);
+       }
+
+       struct waitq *wq = global_eventq(event);
 
 
-       index = wait_hash(event);
-       wq = &wait_queues[index];
-       if (one_thread)
-           return (wait_queue_wakeup_one(wq, event, result));
-       else
-           return (wait_queue_wakeup_all(wq, event, result));
+       return waitq_wakeup64_identify(wq, CAST_EVENT64_T(event), THREAD_AWAKENED, priority);
 }
 
 /*
  *     thread_bind:
  *
 }
 
 /*
  *     thread_bind:
  *
- *     Force a thread to execute on the specified processor.
+ *     Force the current thread to execute on the specified processor.
+ *     Takes effect after the next thread_block().
  *
  *     Returns the previous binding.  PROCESSOR_NULL means
  *     not bound.
  *
  *     Returns the previous binding.  PROCESSOR_NULL means
  *     not bound.
@@ -1100,303 +1519,910 @@ thread_wakeup_prim(
  */
 processor_t
 thread_bind(
  */
 processor_t
 thread_bind(
-       register thread_t       thread,
-       processor_t                     processor)
+       processor_t             processor)
 {
 {
-       processor_t             prev;
-       run_queue_t             runq = RUN_QUEUE_NULL;
-       spl_t                   s;
+       thread_t                self = current_thread();
+       processor_t             prev;
+       spl_t                   s;
 
        s = splsched();
 
        s = splsched();
-       thread_lock(thread);
-       prev = thread->bound_processor;
-       if (prev != PROCESSOR_NULL)
-               runq = run_queue_remove(thread);
+       thread_lock(self);
 
 
-       thread->bound_processor = processor;
+       prev = thread_bind_internal(self, processor);
 
 
-       if (runq != RUN_QUEUE_NULL)
-               thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
-       thread_unlock(thread);
+       thread_unlock(self);
        splx(s);
 
        splx(s);
 
-       return (prev);
+       return prev;
 }
 
 }
 
-struct {
-       uint32_t        idle_pset_last,
-                               idle_pset_any,
-                               idle_bound;
+/*
+ * thread_bind_internal:
+ *
+ * If the specified thread is not the current thread, and it is currently
+ * running on another CPU, a remote AST must be sent to that CPU to cause
+ * the thread to migrate to its bound processor. Otherwise, the migration
+ * will occur at the next quantum expiration or blocking point.
+ *
+ * When the thread is the current thread, and explicit thread_block() should
+ * be used to force the current processor to context switch away and
+ * let the thread migrate to the bound processor.
+ *
+ * Thread must be locked, and at splsched.
+ */
+
+static processor_t
+thread_bind_internal(
+       thread_t                thread,
+       processor_t             processor)
+{
+       processor_t             prev;
+
+       /* <rdar://problem/15102234> */
+       assert(thread->sched_pri < BASEPRI_RTQUEUES);
+       /* A thread can't be bound if it's sitting on a (potentially incorrect) runqueue */
+       assert(thread->runq == PROCESSOR_NULL);
 
 
-       uint32_t        pset_self,
-                               pset_last,
-                               pset_other,
-                               bound_self,
-                               bound_other;
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_THREAD_BIND), thread_tid(thread), processor ? (uintptr_t)processor->cpu_id : (uintptr_t)-1, 0, 0, 0);
 
 
-       uint32_t        realtime_self,
-                               realtime_last,
-                               realtime_other;
+       prev = thread->bound_processor;
+       thread->bound_processor = processor;
 
 
-       uint32_t        missed_realtime,
-                               missed_other;
-} dispatch_counts;
+       return prev;
+}
 
 /*
 
 /*
- *     Select a thread for the current processor to run.
+ * thread_vm_bind_group_add:
  *
  *
- *     May select the current thread, which must be locked.
+ * The "VM bind group" is a special mechanism to mark a collection
+ * of threads from the VM subsystem that, in general, should be scheduled
+ * with only one CPU of parallelism. To accomplish this, we initially
+ * bind all the threads to the master processor, which has the effect
+ * that only one of the threads in the group can execute at once, including
+ * preempting threads in the group that are a lower priority. Future
+ * mechanisms may use more dynamic mechanisms to prevent the collection
+ * of VM threads from using more CPU time than desired.
+ *
+ * The current implementation can result in priority inversions where
+ * compute-bound priority 95 or realtime threads that happen to have
+ * landed on the master processor prevent the VM threads from running.
+ * When this situation is detected, we unbind the threads for one
+ * scheduler tick to allow the scheduler to run the threads an
+ * additional CPUs, before restoring the binding (assuming high latency
+ * is no longer a problem).
  */
  */
-thread_t
-thread_select(
-       register processor_t    processor)
-{
-       register thread_t               thread;
-       processor_set_t                 pset;
-       boolean_t                               other_runnable;
 
 
-       /*
-        *      Check for other non-idle runnable threads.
-        */
-       pset = processor->processor_set;
-       thread = processor->active_thread;
-
-       /* Update the thread's priority */
-       if (thread->sched_stamp != sched_tick)
-               update_priority(thread);
-
-       processor->current_pri = thread->sched_pri;
-
-       simple_lock(&pset->sched_lock);
-
-       other_runnable = processor->runq.count > 0 || pset->runq.count > 0;
-
-       if (    thread->state == TH_RUN                                                 &&
-                       thread->processor_set == pset                                   &&
-                       (thread->bound_processor == PROCESSOR_NULL      ||
-                        thread->bound_processor == processor)                          ) {
-               if (    thread->sched_pri >= BASEPRI_RTQUEUES   &&
-                                               first_timeslice(processor)                      ) {
-                       if (pset->runq.highq >= BASEPRI_RTQUEUES) {
-                               register run_queue_t    runq = &pset->runq;
-                               register queue_t                q;
-
-                               q = runq->queues + runq->highq;
-                               if (((thread_t)q->next)->realtime.deadline <
-                                                                                               processor->deadline) {
-                                       thread = (thread_t)q->next;
-                                       ((queue_entry_t)thread)->next->prev = q;
-                                       q->next = ((queue_entry_t)thread)->next;
-                                       thread->runq = RUN_QUEUE_NULL;
-                                       assert(thread->sched_mode & TH_MODE_PREEMPT);
-                                       runq->count--; runq->urgency--;
-                                       if (queue_empty(q)) {
-                                               if (runq->highq != IDLEPRI)
-                                                       clrbit(MAXPRI - runq->highq, runq->bitmap);
-                                               runq->highq = MAXPRI - ffsbit(runq->bitmap);
-                                       }
-                               }
-                       }
+/*
+ * The current max is provisioned for:
+ * vm_compressor_swap_trigger_thread (92)
+ * 2 x vm_pageout_iothread_internal (92) when vm_restricted_to_single_processor==TRUE
+ * vm_pageout_continue (92)
+ * memorystatus_thread (95)
+ */
+#define MAX_VM_BIND_GROUP_COUNT (5)
+decl_simple_lock_data(static, sched_vm_group_list_lock);
+static thread_t sched_vm_group_thread_list[MAX_VM_BIND_GROUP_COUNT];
+static int sched_vm_group_thread_count;
+static boolean_t sched_vm_group_temporarily_unbound = FALSE;
 
 
-                       processor->deadline = thread->realtime.deadline;
+void
+thread_vm_bind_group_add(void)
+{
+       thread_t self = current_thread();
 
 
-                       simple_unlock(&pset->sched_lock);
+       thread_reference_internal(self);
+       self->options |= TH_OPT_SCHED_VM_GROUP;
 
 
-                       return (thread);
-               }
+       simple_lock(&sched_vm_group_list_lock, LCK_GRP_NULL);
+       assert(sched_vm_group_thread_count < MAX_VM_BIND_GROUP_COUNT);
+       sched_vm_group_thread_list[sched_vm_group_thread_count++] = self;
+       simple_unlock(&sched_vm_group_list_lock);
 
 
-               if (    (!other_runnable                                                        ||
-                                (processor->runq.highq < thread->sched_pri             &&
-                                 pset->runq.highq < thread->sched_pri))                        ) {
+       thread_bind(master_processor);
 
 
-                       /* I am the highest priority runnable (non-idle) thread */
+       /* Switch to bound processor if not already there */
+       thread_block(THREAD_CONTINUE_NULL);
+}
 
 
-                       processor->deadline = UINT64_MAX;
+static void
+sched_vm_group_maintenance(void)
+{
+       uint64_t ctime = mach_absolute_time();
+       uint64_t longtime = ctime - sched_tick_interval;
+       int i;
+       spl_t s;
+       boolean_t high_latency_observed = FALSE;
+       boolean_t runnable_and_not_on_runq_observed = FALSE;
+       boolean_t bind_target_changed = FALSE;
+       processor_t bind_target = PROCESSOR_NULL;
+
+       /* Make sure nobody attempts to add new threads while we are enumerating them */
+       simple_lock(&sched_vm_group_list_lock, LCK_GRP_NULL);
+
+       s = splsched();
 
 
-                       simple_unlock(&pset->sched_lock);
+       for (i = 0; i < sched_vm_group_thread_count; i++) {
+               thread_t thread = sched_vm_group_thread_list[i];
+               assert(thread != THREAD_NULL);
+               thread_lock(thread);
+               if ((thread->state & (TH_RUN | TH_WAIT)) == TH_RUN) {
+                       if (thread->runq != PROCESSOR_NULL && thread->last_made_runnable_time < longtime) {
+                               high_latency_observed = TRUE;
+                       } else if (thread->runq == PROCESSOR_NULL) {
+                               /* There are some cases where a thread be transitiong that also fall into this case */
+                               runnable_and_not_on_runq_observed = TRUE;
+                       }
+               }
+               thread_unlock(thread);
 
 
-                       return (thread);
+               if (high_latency_observed && runnable_and_not_on_runq_observed) {
+                       /* All the things we are looking for are true, stop looking */
+                       break;
                }
        }
 
                }
        }
 
-       if (other_runnable)
-               thread = choose_thread(pset, processor);
-       else {
+       splx(s);
+
+       if (sched_vm_group_temporarily_unbound) {
+               /* If we turned off binding, make sure everything is OK before rebinding */
+               if (!high_latency_observed) {
+                       /* rebind */
+                       bind_target_changed = TRUE;
+                       bind_target = master_processor;
+                       sched_vm_group_temporarily_unbound = FALSE; /* might be reset to TRUE if change cannot be completed */
+               }
+       } else {
                /*
                /*
-                *      Nothing is runnable, so set this processor idle if it
-                *      was running.  Return its idle thread.
+                * Check if we're in a bad state, which is defined by high
+                * latency with no core currently executing a thread. If a
+                * single thread is making progress on a CPU, that means the
+                * binding concept to reduce parallelism is working as
+                * designed.
                 */
                 */
-               if (processor->state == PROCESSOR_RUNNING) {
-                       remqueue(&pset->active_queue, (queue_entry_t)processor);
-                       processor->state = PROCESSOR_IDLE;
-
-                       enqueue_tail(&pset->idle_queue, (queue_entry_t)processor);
-                       pset->idle_count++;
+               if (high_latency_observed && !runnable_and_not_on_runq_observed) {
+                       /* unbind */
+                       bind_target_changed = TRUE;
+                       bind_target = PROCESSOR_NULL;
+                       sched_vm_group_temporarily_unbound = TRUE;
                }
                }
+       }
 
 
-               processor->deadline = UINT64_MAX;
+       if (bind_target_changed) {
+               s = splsched();
+               for (i = 0; i < sched_vm_group_thread_count; i++) {
+                       thread_t thread = sched_vm_group_thread_list[i];
+                       boolean_t removed;
+                       assert(thread != THREAD_NULL);
 
 
-               thread = processor->idle_thread;
-       }
+                       thread_lock(thread);
+                       removed = thread_run_queue_remove(thread);
+                       if (removed || ((thread->state & (TH_RUN | TH_WAIT)) == TH_WAIT)) {
+                               thread_bind_internal(thread, bind_target);
+                       } else {
+                               /*
+                                * Thread was in the middle of being context-switched-to,
+                                * or was in the process of blocking. To avoid switching the bind
+                                * state out mid-flight, defer the change if possible.
+                                */
+                               if (bind_target == PROCESSOR_NULL) {
+                                       thread_bind_internal(thread, bind_target);
+                               } else {
+                                       sched_vm_group_temporarily_unbound = TRUE; /* next pass will try again */
+                               }
+                       }
 
 
-       simple_unlock(&pset->sched_lock);
+                       if (removed) {
+                               thread_run_queue_reinsert(thread, SCHED_PREEMPT | SCHED_TAILQ);
+                       }
+                       thread_unlock(thread);
+               }
+               splx(s);
+       }
 
 
-       return (thread);
+       simple_unlock(&sched_vm_group_list_lock);
 }
 
 }
 
-/*
- *     Perform a context switch and start executing the new thread.
- *
- *     Returns FALSE on failure, and the thread is re-dispatched.
- *
- *     Called at splsched.
+/* Invoked prior to idle entry to determine if, on SMT capable processors, an SMT
+ * rebalancing opportunity exists when a core is (instantaneously) idle, but
+ * other SMT-capable cores may be over-committed. TODO: some possible negatives:
+ * IPI thrash if this core does not remain idle following the load balancing ASTs
+ * Idle "thrash", when IPI issue is followed by idle entry/core power down
+ * followed by a wakeup shortly thereafter.
  */
 
  */
 
-#define funnel_release_check(thread, debug)                            \
-MACRO_BEGIN                                                                                            \
-       if ((thread)->funnel_state & TH_FN_OWNED) {                     \
-               (thread)->funnel_state = TH_FN_REFUNNEL;                \
-               KERNEL_DEBUG(0x603242c | DBG_FUNC_NONE,                 \
-                       (thread)->funnel_lock, (debug), 0, 0, 0);       \
-               funnel_unlock((thread)->funnel_lock);                   \
-       }                                                                                                       \
-MACRO_END
-
-#define funnel_refunnel_check(thread, debug)                           \
-MACRO_BEGIN                                                                                                    \
-       if ((thread)->funnel_state & TH_FN_REFUNNEL) {                  \
-               kern_return_t   result = (thread)->wait_result;         \
-                                                                                                                       \
-               (thread)->funnel_state = 0;                                                     \
-               KERNEL_DEBUG(0x6032428 | DBG_FUNC_NONE,                         \
-                       (thread)->funnel_lock, (debug), 0, 0, 0);               \
-               funnel_lock((thread)->funnel_lock);                                     \
-               KERNEL_DEBUG(0x6032430 | DBG_FUNC_NONE,                         \
-                       (thread)->funnel_lock, (debug), 0, 0, 0);               \
-               (thread)->funnel_state = TH_FN_OWNED;                           \
-               (thread)->wait_result = result;                                         \
-       }                                                                                                               \
-MACRO_END
+#if (DEVELOPMENT || DEBUG)
+int sched_smt_balance = 1;
+#endif
 
 
-boolean_t
-thread_invoke(
-       register thread_t       old_thread,
-       register thread_t       new_thread,
-       ast_t                           reason)
+#if __SMP__
+/* Invoked with pset locked, returns with pset unlocked */
+void
+sched_SMT_balance(processor_t cprocessor, processor_set_t cpset)
 {
 {
-       thread_continue_t       new_cont, continuation = old_thread->continuation;
-       void                            *new_param, *parameter = old_thread->parameter;
-       processor_t                     processor;
-       thread_t                        prev_thread;
+       processor_t ast_processor = NULL;
 
 
-       if (get_preemption_level() != 0)
-               panic("thread_invoke: preemption_level %d\n",
-                                                               get_preemption_level());
-
-       assert(old_thread == current_thread());
+#if (DEVELOPMENT || DEBUG)
+       if (__improbable(sched_smt_balance == 0)) {
+               goto smt_balance_exit;
+       }
+#endif
 
 
-       /*
-        * Mark thread interruptible.
-        */
-       thread_lock(new_thread);
-       new_thread->state &= ~TH_UNINT;
+       assert(cprocessor == current_processor());
+       if (cprocessor->is_SMT == FALSE) {
+               goto smt_balance_exit;
+       }
 
 
-       assert(thread_runnable(new_thread));
+       processor_t sib_processor = cprocessor->processor_secondary ? cprocessor->processor_secondary : cprocessor->processor_primary;
 
 
-       /*
-        * Allow time constraint threads to hang onto
-        * a stack.
+       /* Determine if both this processor and its sibling are idle,
+        * indicating an SMT rebalancing opportunity.
         */
         */
-       if (    (old_thread->sched_mode & TH_MODE_REALTIME)             &&
-                                       !old_thread->reserved_stack                             ) {
-               old_thread->reserved_stack = old_thread->kernel_stack;
+       if (sib_processor->state != PROCESSOR_IDLE) {
+               goto smt_balance_exit;
        }
 
        }
 
-       if (continuation != NULL) {
-               if (!new_thread->kernel_stack) {
-                       /*
-                        * If the old thread is using a privileged stack,
-                        * check to see whether we can exchange it with
-                        * that of the new thread.
-                        */
-                       if (    old_thread->kernel_stack == old_thread->reserved_stack  &&
-                                                       !new_thread->reserved_stack)
-                               goto need_stack;
+       processor_t sprocessor;
+
+       sched_ipi_type_t ipi_type = SCHED_IPI_NONE;
+       uint64_t running_secondary_map = (cpset->cpu_state_map[PROCESSOR_RUNNING] &
+           ~cpset->primary_map);
+       for (int cpuid = lsb_first(running_secondary_map); cpuid >= 0; cpuid = lsb_next(running_secondary_map, cpuid)) {
+               sprocessor = processor_array[cpuid];
+               if ((sprocessor->processor_primary->state == PROCESSOR_RUNNING) &&
+                   (sprocessor->current_pri < BASEPRI_RTQUEUES)) {
+                       ipi_type = sched_ipi_action(sprocessor, NULL, false, SCHED_IPI_EVENT_SMT_REBAL);
+                       if (ipi_type != SCHED_IPI_NONE) {
+                               assert(sprocessor != cprocessor);
+                               ast_processor = sprocessor;
+                               break;
+                       }
+               }
+       }
 
 
-                       /*
-                        * Context switch by performing a stack handoff.
-                        */
-                       new_cont = new_thread->continuation;
-                       new_thread->continuation = NULL;
-                       new_param = new_thread->parameter;
-                       new_thread->parameter = NULL;
+smt_balance_exit:
+       pset_unlock(cpset);
 
 
-                       processor = current_processor();
-                       processor->active_thread = new_thread;
-                       processor->current_pri = new_thread->sched_pri;
-                       new_thread->last_processor = processor;
-                       ast_context(new_thread);
-                       thread_unlock(new_thread);
-               
-                       current_task()->csw++;
+       if (ast_processor) {
+               KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_SMT_BALANCE), ast_processor->cpu_id, ast_processor->state, ast_processor->processor_primary->state, 0, 0);
+               sched_ipi_perform(ast_processor, ipi_type);
+       }
+}
+#else
+/* Invoked with pset locked, returns with pset unlocked */
+void
+sched_SMT_balance(__unused processor_t cprocessor, processor_set_t cpset)
+{
+       pset_unlock(cpset);
+}
+#endif /* __SMP__ */
 
 
-                       old_thread->reason = reason;
+/*
+ * Called with pset locked, on a processor that is committing to run a new thread
+ * Will transition an idle or dispatching processor to running as it picks up
+ * the first new thread from the idle thread.
+ */
+static void
+pset_commit_processor_to_new_thread(processor_set_t pset, processor_t processor, thread_t new_thread)
+{
+       if (processor->state == PROCESSOR_DISPATCHING || processor->state == PROCESSOR_IDLE) {
+               assert(current_thread() == processor->idle_thread);
 
 
-                       processor->last_dispatch = mach_absolute_time();
-                       timer_event((uint32_t)processor->last_dispatch,
-                                                                               &new_thread->system_timer);
-          
-                       thread_done(old_thread, new_thread, processor);
+               /*
+                * Dispatching processor is now committed to running new_thread,
+                * so change its state to PROCESSOR_RUNNING.
+                */
+               pset_update_processor_state(pset, processor, PROCESSOR_RUNNING);
+       } else {
+               assert((processor->state == PROCESSOR_RUNNING) || (processor->state == PROCESSOR_SHUTDOWN));
+       }
 
 
-                       machine_stack_handoff(old_thread, new_thread);
+       processor_state_update_from_thread(processor, new_thread);
+}
 
 
-                       thread_begin(new_thread, processor);
+static processor_t choose_processor_for_realtime_thread(processor_set_t pset);
+static bool all_available_primaries_are_running_realtime_threads(processor_set_t pset);
+static bool these_processors_are_running_realtime_threads(processor_set_t pset, uint64_t these_map);
+static bool sched_ok_to_run_realtime_thread(processor_set_t pset, processor_t processor);
+int sched_allow_rt_smt = 1;
+int sched_avoid_cpu0 = 1;
 
 
-                       /*
-                        * Now dispatch the old thread.
-                        */
-                       thread_dispatch(old_thread);
+/*
+ *     thread_select:
+ *
+ *     Select a new thread for the current processor to execute.
+ *
+ *     May select the current thread, which must be locked.
+ */
+static thread_t
+thread_select(thread_t          thread,
+    processor_t       processor,
+    ast_t            *reason)
+{
+       processor_set_t         pset = processor->processor_set;
+       thread_t                        new_thread = THREAD_NULL;
 
 
-                       counter_always(c_thread_invoke_hits++);
+       assert(processor == current_processor());
+       assert((thread->state & (TH_RUN | TH_TERMINATE2)) == TH_RUN);
 
 
-                       funnel_refunnel_check(new_thread, 2);
-                       (void) spllo();
+       do {
+               /*
+                *      Update the priority.
+                */
+               if (SCHED(can_update_priority)(thread)) {
+                       SCHED(update_priority)(thread);
+               }
 
 
-                       assert(new_cont);
-                       call_continuation(new_cont, new_param, new_thread->wait_result);
-                       /*NOTREACHED*/
+               pset_lock(pset);
+
+               processor_state_update_from_thread(processor, thread);
+
+restart:
+               /* Acknowledge any pending IPIs here with pset lock held */
+               bit_clear(pset->pending_AST_URGENT_cpu_mask, processor->cpu_id);
+               bit_clear(pset->pending_AST_PREEMPT_cpu_mask, processor->cpu_id);
+
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+               bit_clear(pset->pending_deferred_AST_cpu_mask, processor->cpu_id);
+#endif
+
+               bool secondary_can_only_run_realtime_thread = false;
+
+               assert(processor->state != PROCESSOR_OFF_LINE);
+
+               if (!processor->is_recommended) {
+                       /*
+                        * The performance controller has provided a hint to not dispatch more threads,
+                        * unless they are bound to us (and thus we are the only option
+                        */
+                       if (!SCHED(processor_bound_count)(processor)) {
+                               goto idle;
+                       }
+               } else if (processor->processor_primary != processor) {
+                       /*
+                        * Should this secondary SMT processor attempt to find work? For pset runqueue systems,
+                        * we should look for work only under the same conditions that choose_processor()
+                        * would have assigned work, which is when all primary processors have been assigned work.
+                        *
+                        * An exception is that bound threads are dispatched to a processor without going through
+                        * choose_processor(), so in those cases we should continue trying to dequeue work.
+                        */
+                       if (!SCHED(processor_bound_count)(processor)) {
+                               if ((pset->recommended_bitmask & pset->primary_map & pset->cpu_state_map[PROCESSOR_IDLE]) != 0) {
+                                       goto idle;
+                               }
+
+                               /*
+                                * TODO: What if a secondary core beat an idle primary to waking up from an IPI?
+                                * Should it dequeue immediately, or spin waiting for the primary to wake up?
+                                */
+
+                               /* There are no idle primaries */
+
+                               if (processor->processor_primary->current_pri >= BASEPRI_RTQUEUES) {
+                                       bool secondary_can_run_realtime_thread = sched_allow_rt_smt && rt_runq_count(pset) && all_available_primaries_are_running_realtime_threads(pset);
+                                       if (!secondary_can_run_realtime_thread) {
+                                               goto idle;
+                                       }
+                                       secondary_can_only_run_realtime_thread = true;
+                               }
+                       }
+               }
+
+               /*
+                *      Test to see if the current thread should continue
+                *      to run on this processor.  Must not be attempting to wait, and not
+                *      bound to a different processor, nor be in the wrong
+                *      processor set, nor be forced to context switch by TH_SUSP.
+                *
+                *      Note that there are never any RT threads in the regular runqueue.
+                *
+                *      This code is very insanely tricky.
+                */
+
+               /* i.e. not waiting, not TH_SUSP'ed */
+               bool still_running = ((thread->state & (TH_TERMINATE | TH_IDLE | TH_WAIT | TH_RUN | TH_SUSP)) == TH_RUN);
+
+               /*
+                * Threads running on SMT processors are forced to context switch. Don't rebalance realtime threads.
+                * TODO: This should check if it's worth it to rebalance, i.e. 'are there any idle primary processors'
+                *       <rdar://problem/47907700>
+                *
+                * A yielding thread shouldn't be forced to context switch.
+                */
+
+               bool is_yielding         = (*reason & AST_YIELD) == AST_YIELD;
+
+               bool needs_smt_rebalance = !is_yielding && thread->sched_pri < BASEPRI_RTQUEUES && processor->processor_primary != processor;
+
+               bool affinity_mismatch   = thread->affinity_set != AFFINITY_SET_NULL && thread->affinity_set->aset_pset != pset;
+
+               bool bound_elsewhere     = thread->bound_processor != PROCESSOR_NULL && thread->bound_processor != processor;
+
+               bool avoid_processor     = !is_yielding && SCHED(avoid_processor_enabled) && SCHED(thread_avoid_processor)(processor, thread);
+
+               if (still_running && !needs_smt_rebalance && !affinity_mismatch && !bound_elsewhere && !avoid_processor) {
+                       /*
+                        * This thread is eligible to keep running on this processor.
+                        *
+                        * RT threads with un-expired quantum stay on processor,
+                        * unless there's a valid RT thread with an earlier deadline.
+                        */
+                       if (thread->sched_pri >= BASEPRI_RTQUEUES && processor->first_timeslice) {
+                               if (rt_runq_count(pset) > 0) {
+                                       rt_lock_lock(pset);
+
+                                       if (rt_runq_count(pset) > 0) {
+                                               thread_t next_rt = qe_queue_first(&SCHED(rt_runq)(pset)->queue, struct thread, runq_links);
+
+                                               if (next_rt->realtime.deadline < processor->deadline &&
+                                                   (next_rt->bound_processor == PROCESSOR_NULL ||
+                                                   next_rt->bound_processor == processor)) {
+                                                       /* The next RT thread is better, so pick it off the runqueue. */
+                                                       goto pick_new_rt_thread;
+                                               }
+                                       }
+
+                                       rt_lock_unlock(pset);
+                               }
+
+                               /* This is still the best RT thread to run. */
+                               processor->deadline = thread->realtime.deadline;
+
+                               sched_update_pset_load_average(pset);
+
+                               processor_t next_rt_processor = PROCESSOR_NULL;
+                               sched_ipi_type_t next_rt_ipi_type = SCHED_IPI_NONE;
+
+                               if (rt_runq_count(pset) > 0) {
+                                       next_rt_processor = choose_processor_for_realtime_thread(pset);
+                                       if (next_rt_processor) {
+                                               if (next_rt_processor->state == PROCESSOR_IDLE) {
+                                                       pset_update_processor_state(pset, next_rt_processor, PROCESSOR_DISPATCHING);
+                                               }
+                                               next_rt_ipi_type = sched_ipi_action(next_rt_processor, NULL, false, SCHED_IPI_EVENT_PREEMPT);
+                                       }
+                               }
+                               pset_unlock(pset);
+
+                               if (next_rt_processor) {
+                                       sched_ipi_perform(next_rt_processor, next_rt_ipi_type);
+                               }
+
+                               return thread;
+                       }
+
+                       if ((rt_runq_count(pset) == 0) &&
+                           SCHED(processor_queue_has_priority)(processor, thread->sched_pri, TRUE) == FALSE) {
+                               /* This thread is still the highest priority runnable (non-idle) thread */
+                               processor->deadline = UINT64_MAX;
+
+                               sched_update_pset_load_average(pset);
+                               pset_unlock(pset);
+
+                               return thread;
+                       }
+               } else {
+                       /*
+                        * This processor must context switch.
+                        * If it's due to a rebalance, we should aggressively find this thread a new home.
+                        */
+                       if (needs_smt_rebalance || affinity_mismatch || bound_elsewhere || avoid_processor) {
+                               *reason |= AST_REBALANCE;
+                       }
+               }
+
+               /* OK, so we're not going to run the current thread. Look at the RT queue. */
+               bool ok_to_run_realtime_thread = sched_ok_to_run_realtime_thread(pset, processor);
+               if ((rt_runq_count(pset) > 0) && ok_to_run_realtime_thread) {
+                       rt_lock_lock(pset);
+
+                       if ((rt_runq_count(pset) > 0) && ok_to_run_realtime_thread) {
+                               thread_t next_rt = qe_queue_first(&SCHED(rt_runq)(pset)->queue, struct thread, runq_links);
+
+                               if (__probable((next_rt->bound_processor == PROCESSOR_NULL ||
+                                   (next_rt->bound_processor == processor)))) {
+pick_new_rt_thread:
+                                       new_thread = qe_dequeue_head(&SCHED(rt_runq)(pset)->queue, struct thread, runq_links);
+
+                                       new_thread->runq = PROCESSOR_NULL;
+                                       SCHED_STATS_RUNQ_CHANGE(&SCHED(rt_runq)(pset)->runq_stats, rt_runq_count(pset));
+                                       rt_runq_count_decr(pset);
+
+                                       processor->deadline = new_thread->realtime.deadline;
+
+                                       pset_commit_processor_to_new_thread(pset, processor, new_thread);
+
+                                       rt_lock_unlock(pset);
+                                       sched_update_pset_load_average(pset);
+
+                                       processor_t ast_processor = PROCESSOR_NULL;
+                                       processor_t next_rt_processor = PROCESSOR_NULL;
+                                       sched_ipi_type_t ipi_type = SCHED_IPI_NONE;
+                                       sched_ipi_type_t next_rt_ipi_type = SCHED_IPI_NONE;
+
+                                       if (processor->processor_secondary != NULL) {
+                                               processor_t sprocessor = processor->processor_secondary;
+                                               if ((sprocessor->state == PROCESSOR_RUNNING) || (sprocessor->state == PROCESSOR_DISPATCHING)) {
+                                                       ipi_type = sched_ipi_action(sprocessor, NULL, false, SCHED_IPI_EVENT_SMT_REBAL);
+                                                       ast_processor = sprocessor;
+                                               }
+                                       }
+                                       if (rt_runq_count(pset) > 0) {
+                                               next_rt_processor = choose_processor_for_realtime_thread(pset);
+                                               if (next_rt_processor) {
+                                                       if (next_rt_processor->state == PROCESSOR_IDLE) {
+                                                               pset_update_processor_state(pset, next_rt_processor, PROCESSOR_DISPATCHING);
+                                                       }
+                                                       next_rt_ipi_type = sched_ipi_action(next_rt_processor, NULL, false, SCHED_IPI_EVENT_PREEMPT);
+                                               }
+                                       }
+                                       pset_unlock(pset);
+
+                                       if (ast_processor) {
+                                               sched_ipi_perform(ast_processor, ipi_type);
+                                       }
+
+                                       if (next_rt_processor) {
+                                               sched_ipi_perform(next_rt_processor, next_rt_ipi_type);
+                                       }
+
+                                       return new_thread;
+                               }
+                       }
+
+                       rt_lock_unlock(pset);
+               }
+               if (secondary_can_only_run_realtime_thread) {
+                       goto idle;
+               }
+
+               processor->deadline = UINT64_MAX;
+
+               /* No RT threads, so let's look at the regular threads. */
+               if ((new_thread = SCHED(choose_thread)(processor, MINPRI, *reason)) != THREAD_NULL) {
+                       sched_update_pset_load_average(pset);
+
+                       pset_commit_processor_to_new_thread(pset, processor, new_thread);
+
+                       processor_t ast_processor = PROCESSOR_NULL;
+                       sched_ipi_type_t ipi_type = SCHED_IPI_NONE;
+
+                       processor_t sprocessor = processor->processor_secondary;
+                       if ((sprocessor != NULL) && (sprocessor->state == PROCESSOR_RUNNING)) {
+                               if (thread_no_smt(new_thread)) {
+                                       ipi_type = sched_ipi_action(sprocessor, NULL, false, SCHED_IPI_EVENT_SMT_REBAL);
+                                       ast_processor = sprocessor;
+                               }
+                       }
+                       pset_unlock(pset);
+
+                       if (ast_processor) {
+                               sched_ipi_perform(ast_processor, ipi_type);
+                       }
+                       return new_thread;
+               }
+
+               if (processor->must_idle) {
+                       processor->must_idle = false;
+                       goto idle;
+               }
+
+#if __SMP__
+               if (SCHED(steal_thread_enabled)(pset)) {
+                       /*
+                        * No runnable threads, attempt to steal
+                        * from other processors. Returns with pset lock dropped.
+                        */
+
+                       if ((new_thread = SCHED(steal_thread)(pset)) != THREAD_NULL) {
+                               /*
+                                * Avoid taking the pset_lock unless it is necessary to change state.
+                                * It's safe to read processor->state here, as only the current processor can change state
+                                * from this point (interrupts are disabled and this processor is committed to run new_thread).
+                                */
+                               if (processor->state == PROCESSOR_DISPATCHING || processor->state == PROCESSOR_IDLE) {
+                                       pset_lock(pset);
+                                       pset_commit_processor_to_new_thread(pset, processor, new_thread);
+                                       pset_unlock(pset);
+                               } else {
+                                       assert((processor->state == PROCESSOR_RUNNING) || (processor->state == PROCESSOR_SHUTDOWN));
+                                       processor_state_update_from_thread(processor, new_thread);
+                               }
+
+                               return new_thread;
+                       }
+
+                       /*
+                        * If other threads have appeared, shortcut
+                        * around again.
+                        */
+                       if (!SCHED(processor_queue_empty)(processor) || (ok_to_run_realtime_thread && (rt_runq_count(pset) > 0))) {
+                               continue;
+                       }
+
+                       pset_lock(pset);
+
+                       /* Someone selected this processor while we had dropped the lock */
+                       if (bit_test(pset->pending_AST_URGENT_cpu_mask, processor->cpu_id)) {
+                               goto restart;
+                       }
+               }
+#endif
+
+idle:
+               /*
+                *      Nothing is runnable, so set this processor idle if it
+                *      was running.
+                */
+               if ((processor->state == PROCESSOR_RUNNING) || (processor->state == PROCESSOR_DISPATCHING)) {
+                       pset_update_processor_state(pset, processor, PROCESSOR_IDLE);
+                       processor_state_update_idle(processor);
                }
                }
-               else
-               if (new_thread == old_thread) {
+
+#if __SMP__
+               /* Invoked with pset locked, returns with pset unlocked */
+               SCHED(processor_balance)(processor, pset);
+#else
+               pset_unlock(pset);
+#endif
+
+               new_thread = processor->idle_thread;
+       } while (new_thread == THREAD_NULL);
+
+       return new_thread;
+}
+
+/*
+ * thread_invoke
+ *
+ * Called at splsched with neither thread locked.
+ *
+ * Perform a context switch and start executing the new thread.
+ *
+ * Returns FALSE when the context switch didn't happen.
+ * The reference to the new thread is still consumed.
+ *
+ * "self" is what is currently running on the processor,
+ * "thread" is the new thread to context switch to
+ * (which may be the same thread in some cases)
+ */
+static boolean_t
+thread_invoke(
+       thread_t                        self,
+       thread_t                        thread,
+       ast_t                           reason)
+{
+       if (__improbable(get_preemption_level() != 0)) {
+               int pl = get_preemption_level();
+               panic("thread_invoke: preemption_level %d, possible cause: %s",
+                   pl, (pl < 0 ? "unlocking an unlocked mutex or spinlock" :
+                   "blocking while holding a spinlock, or within interrupt context"));
+       }
+
+       thread_continue_t       continuation = self->continuation;
+       void                    *parameter   = self->parameter;
+       processor_t             processor;
+
+       uint64_t                ctime = mach_absolute_time();
+
+#ifdef CONFIG_MACH_APPROXIMATE_TIME
+       commpage_update_mach_approximate_time(ctime);
+#endif
+
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+       if ((thread->state & TH_IDLE) == 0) {
+               sched_timeshare_consider_maintenance(ctime);
+       }
+#endif
+
+#if MONOTONIC
+       mt_sched_update(self);
+#endif /* MONOTONIC */
+
+       assert_thread_magic(self);
+       assert(self == current_thread());
+       assert(self->runq == PROCESSOR_NULL);
+       assert((self->state & (TH_RUN | TH_TERMINATE2)) == TH_RUN);
+
+       thread_lock(thread);
+
+       assert_thread_magic(thread);
+       assert((thread->state & (TH_RUN | TH_WAIT | TH_UNINT | TH_TERMINATE | TH_TERMINATE2)) == TH_RUN);
+       assert(thread->bound_processor == PROCESSOR_NULL || thread->bound_processor == current_processor());
+       assert(thread->runq == PROCESSOR_NULL);
+
+       /* Reload precise timing global policy to thread-local policy */
+       thread->precise_user_kernel_time = use_precise_user_kernel_time(thread);
+
+       /* Update SFI class based on other factors */
+       thread->sfi_class = sfi_thread_classify(thread);
+
+       /* Update the same_pri_latency for the thread (used by perfcontrol callouts) */
+       thread->same_pri_latency = ctime - thread->last_basepri_change_time;
+       /*
+        * In case a base_pri update happened between the timestamp and
+        * taking the thread lock
+        */
+       if (ctime <= thread->last_basepri_change_time) {
+               thread->same_pri_latency = ctime - thread->last_made_runnable_time;
+       }
+
+       /* Allow realtime threads to hang onto a stack. */
+       if ((self->sched_mode == TH_MODE_REALTIME) && !self->reserved_stack) {
+               self->reserved_stack = self->kernel_stack;
+       }
+
+       /* Prepare for spin debugging */
+#if INTERRUPT_MASKED_DEBUG
+       ml_spin_debug_clear(thread);
+#endif
+
+       if (continuation != NULL) {
+               if (!thread->kernel_stack) {
+                       /*
+                        * If we are using a privileged stack,
+                        * check to see whether we can exchange it with
+                        * that of the other thread.
+                        */
+                       if (self->kernel_stack == self->reserved_stack && !thread->reserved_stack) {
+                               goto need_stack;
+                       }
+
+                       /*
+                        * Context switch by performing a stack handoff.
+                        * Requires both threads to be parked in a continuation.
+                        */
+                       continuation = thread->continuation;
+                       parameter = thread->parameter;
+
+                       processor = current_processor();
+                       processor->active_thread = thread;
+                       processor_state_update_from_thread(processor, thread);
+
+                       if (thread->last_processor != processor && thread->last_processor != NULL) {
+                               if (thread->last_processor->processor_set != processor->processor_set) {
+                                       thread->ps_switch++;
+                               }
+                               thread->p_switch++;
+                       }
+                       thread->last_processor = processor;
+                       thread->c_switch++;
+                       ast_context(thread);
+
+                       thread_unlock(thread);
+
+                       self->reason = reason;
+
+                       processor->last_dispatch = ctime;
+                       self->last_run_time = ctime;
+                       processor_timer_switch_thread(ctime, &thread->system_timer);
+                       timer_update(&thread->runnable_timer, ctime);
+                       PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
+
+                       /*
+                        * Since non-precise user/kernel time doesn't update the state timer
+                        * during privilege transitions, synthesize an event now.
+                        */
+                       if (!thread->precise_user_kernel_time) {
+                               timer_update(PROCESSOR_DATA(processor, current_state), ctime);
+                       }
+
+                       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                           MACHDBG_CODE(DBG_MACH_SCHED, MACH_STACK_HANDOFF) | DBG_FUNC_NONE,
+                           self->reason, (uintptr_t)thread_tid(thread), self->sched_pri, thread->sched_pri, 0);
+
+                       if ((thread->chosen_processor != processor) && (thread->chosen_processor != PROCESSOR_NULL)) {
+                               SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_MOVED) | DBG_FUNC_NONE,
+                                   (uintptr_t)thread_tid(thread), (uintptr_t)thread->chosen_processor->cpu_id, 0, 0, 0);
+                       }
+
+                       DTRACE_SCHED2(off__cpu, struct thread *, thread, struct proc *, thread->task->bsd_info);
+
+                       SCHED_STATS_CSW(processor, self->reason, self->sched_pri, thread->sched_pri);
+
+#if KPERF
+                       kperf_off_cpu(self);
+#endif /* KPERF */
+
+                       /*
+                        * This is where we actually switch thread identity,
+                        * and address space if required.  However, register
+                        * state is not switched - this routine leaves the
+                        * stack and register state active on the current CPU.
+                        */
+                       TLOG(1, "thread_invoke: calling stack_handoff\n");
+                       stack_handoff(self, thread);
+
+                       /* 'self' is now off core */
+                       assert(thread == current_thread_volatile());
+
+                       DTRACE_SCHED(on__cpu);
+
+#if KPERF
+                       kperf_on_cpu(thread, continuation, NULL);
+#endif /* KPERF */
+
+                       thread_dispatch(self, thread);
+
+#if KASAN
+                       /* Old thread's stack has been moved to the new thread, so explicitly
+                        * unpoison it. */
+                       kasan_unpoison_stack(thread->kernel_stack, kernel_stack_size);
+#endif
+
+                       thread->continuation = thread->parameter = NULL;
+
+                       counter(c_thread_invoke_hits++);
+
+                       boolean_t enable_interrupts = TRUE;
+
+                       /* idle thread needs to stay interrupts-disabled */
+                       if ((thread->state & TH_IDLE)) {
+                               enable_interrupts = FALSE;
+                       }
+
+                       assert(continuation);
+                       call_continuation(continuation, parameter,
+                           thread->wait_result, enable_interrupts);
+                       /*NOTREACHED*/
+               } else if (thread == self) {
                        /* same thread but with continuation */
                        /* same thread but with continuation */
+                       ast_context(self);
                        counter(++c_thread_invoke_same);
                        counter(++c_thread_invoke_same);
-                       thread_unlock(new_thread);
 
 
-                       funnel_refunnel_check(new_thread, 3);
-                       (void) spllo();
+                       thread_unlock(self);
+
+#if KPERF
+                       kperf_on_cpu(thread, continuation, NULL);
+#endif /* KPERF */
 
 
-                       call_continuation(continuation, parameter, new_thread->wait_result);
+                       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                           MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED) | DBG_FUNC_NONE,
+                           self->reason, (uintptr_t)thread_tid(thread), self->sched_pri, thread->sched_pri, 0);
+
+#if KASAN
+                       /* stack handoff to self - no thread_dispatch(), so clear the stack
+                        * and free the fakestack directly */
+                       kasan_fakestack_drop(self);
+                       kasan_fakestack_gc(self);
+                       kasan_unpoison_stack(self->kernel_stack, kernel_stack_size);
+#endif
+
+                       self->continuation = self->parameter = NULL;
+
+                       boolean_t enable_interrupts = TRUE;
+
+                       /* idle thread needs to stay interrupts-disabled */
+                       if ((self->state & TH_IDLE)) {
+                               enable_interrupts = FALSE;
+                       }
+
+                       call_continuation(continuation, parameter,
+                           self->wait_result, enable_interrupts);
                        /*NOTREACHED*/
                }
                        /*NOTREACHED*/
                }
-       }
-       else {
+       } else {
                /*
                /*
-                * Check that the new thread has a stack
+                * Check that the other thread has a stack
                 */
                 */
-               if (!new_thread->kernel_stack) {
+               if (!thread->kernel_stack) {
 need_stack:
 need_stack:
-                       if (!stack_alloc_try(new_thread)) {
-                               counter_always(c_thread_invoke_misses++);
-                               thread_unlock(new_thread);
-                               thread_stack_enqueue(new_thread);
-                               return (FALSE);
+                       if (!stack_alloc_try(thread)) {
+                               counter(c_thread_invoke_misses++);
+                               thread_unlock(thread);
+                               thread_stack_enqueue(thread);
+                               return FALSE;
                        }
                        }
-               }
-               else
-               if (new_thread == old_thread) {
+               } else if (thread == self) {
+                       ast_context(self);
                        counter(++c_thread_invoke_same);
                        counter(++c_thread_invoke_same);
-                       thread_unlock(new_thread);
-                       return (TRUE);
+                       thread_unlock(self);
+
+                       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                           MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED) | DBG_FUNC_NONE,
+                           self->reason, (uintptr_t)thread_tid(thread), self->sched_pri, thread->sched_pri, 0);
+
+                       return TRUE;
                }
        }
 
                }
        }
 
@@ -1404,1608 +2430,3880 @@ need_stack:
         * Context switch by full context save.
         */
        processor = current_processor();
         * Context switch by full context save.
         */
        processor = current_processor();
-       processor->active_thread = new_thread;
-       processor->current_pri = new_thread->sched_pri;
-       new_thread->last_processor = processor;
-       ast_context(new_thread);
-       assert(thread_runnable(new_thread));
-       thread_unlock(new_thread);
+       processor->active_thread = thread;
+       processor_state_update_from_thread(processor, thread);
+
+       if (thread->last_processor != processor && thread->last_processor != NULL) {
+               if (thread->last_processor->processor_set != processor->processor_set) {
+                       thread->ps_switch++;
+               }
+               thread->p_switch++;
+       }
+       thread->last_processor = processor;
+       thread->c_switch++;
+       ast_context(thread);
+
+       thread_unlock(thread);
+
+       counter(c_thread_invoke_csw++);
+
+       self->reason = reason;
+
+       processor->last_dispatch = ctime;
+       self->last_run_time = ctime;
+       processor_timer_switch_thread(ctime, &thread->system_timer);
+       timer_update(&thread->runnable_timer, ctime);
+       PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
+
+       /*
+        * Since non-precise user/kernel time doesn't update the state timer
+        * during privilege transitions, synthesize an event now.
+        */
+       if (!thread->precise_user_kernel_time) {
+               timer_update(PROCESSOR_DATA(processor, current_state), ctime);
+       }
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED) | DBG_FUNC_NONE,
+           self->reason, (uintptr_t)thread_tid(thread), self->sched_pri, thread->sched_pri, 0);
+
+       if ((thread->chosen_processor != processor) && (thread->chosen_processor != NULL)) {
+               SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_MOVED) | DBG_FUNC_NONE,
+                   (uintptr_t)thread_tid(thread), (uintptr_t)thread->chosen_processor->cpu_id, 0, 0, 0);
+       }
+
+       DTRACE_SCHED2(off__cpu, struct thread *, thread, struct proc *, thread->task->bsd_info);
+
+       SCHED_STATS_CSW(processor, self->reason, self->sched_pri, thread->sched_pri);
+
+#if KPERF
+       kperf_off_cpu(self);
+#endif /* KPERF */
+
+       /*
+        * This is where we actually switch register context,
+        * and address space if required.  We will next run
+        * as a result of a subsequent context switch.
+        *
+        * Once registers are switched and the processor is running "thread",
+        * the stack variables and non-volatile registers will contain whatever
+        * was there the last time that thread blocked. No local variables should
+        * be used after this point, except for the special case of "thread", which
+        * the platform layer returns as the previous thread running on the processor
+        * via the function call ABI as a return register, and "self", which may have
+        * been stored on the stack or a non-volatile register, but a stale idea of
+        * what was on the CPU is newly-accurate because that thread is again
+        * running on the CPU.
+        *
+        * If one of the threads is using a continuation, thread_continue
+        * is used to stitch up its context.
+        *
+        * If we are invoking a thread which is resuming from a continuation,
+        * the CPU will invoke thread_continue next.
+        *
+        * If the current thread is parking in a continuation, then its state
+        * won't be saved and the stack will be discarded. When the stack is
+        * re-allocated, it will be configured to resume from thread_continue.
+        */
+       assert(continuation == self->continuation);
+       thread = machine_switch_context(self, continuation, thread);
+       assert(self == current_thread_volatile());
+       TLOG(1, "thread_invoke: returning machine_switch_context: self %p continuation %p thread %p\n", self, continuation, thread);
+
+       assert(continuation == NULL && self->continuation == NULL);
+
+       DTRACE_SCHED(on__cpu);
+
+#if KPERF
+       kperf_on_cpu(self, NULL, __builtin_frame_address(0));
+#endif /* KPERF */
+
+       /* We have been resumed and are set to run. */
+       thread_dispatch(thread, self);
+
+       return TRUE;
+}
+
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+/*
+ *     pset_cancel_deferred_dispatch:
+ *
+ *     Cancels all ASTs that we can cancel for the given processor set
+ *     if the current processor is running the last runnable thread in the
+ *     system.
+ *
+ *     This function assumes the current thread is runnable.  This must
+ *     be called with the pset unlocked.
+ */
+static void
+pset_cancel_deferred_dispatch(
+       processor_set_t         pset,
+       processor_t             processor)
+{
+       processor_t             active_processor = NULL;
+       uint32_t                sampled_sched_run_count;
+
+       pset_lock(pset);
+       sampled_sched_run_count = os_atomic_load(&sched_run_buckets[TH_BUCKET_RUN], relaxed);
+
+       /*
+        * If we have emptied the run queue, and our current thread is runnable, we
+        * should tell any processors that are still DISPATCHING that they will
+        * probably not have any work to do.  In the event that there are no
+        * pending signals that we can cancel, this is also uninteresting.
+        *
+        * In the unlikely event that another thread becomes runnable while we are
+        * doing this (sched_run_count is atomically updated, not guarded), the
+        * codepath making it runnable SHOULD (a dangerous word) need the pset lock
+        * in order to dispatch it to a processor in our pset.  So, the other
+        * codepath will wait while we squash all cancelable ASTs, get the pset
+        * lock, and then dispatch the freshly runnable thread.  So this should be
+        * correct (we won't accidentally have a runnable thread that hasn't been
+        * dispatched to an idle processor), if not ideal (we may be restarting the
+        * dispatch process, which could have some overhead).
+        */
+
+       if ((sampled_sched_run_count == 1) && (pset->pending_deferred_AST_cpu_mask)) {
+               uint64_t dispatching_map = (pset->cpu_state_map[PROCESSOR_DISPATCHING] &
+                   pset->pending_deferred_AST_cpu_mask &
+                   ~pset->pending_AST_URGENT_cpu_mask);
+               for (int cpuid = lsb_first(dispatching_map); cpuid >= 0; cpuid = lsb_next(dispatching_map, cpuid)) {
+                       active_processor = processor_array[cpuid];
+                       /*
+                        * If a processor is DISPATCHING, it could be because of
+                        * a cancelable signal.
+                        *
+                        * IF the processor is not our
+                        * current processor (the current processor should not
+                        * be DISPATCHING, so this is a bit paranoid), AND there
+                        * is a cancelable signal pending on the processor, AND
+                        * there is no non-cancelable signal pending (as there is
+                        * no point trying to backtrack on bringing the processor
+                        * up if a signal we cannot cancel is outstanding), THEN
+                        * it should make sense to roll back the processor state
+                        * to the IDLE state.
+                        *
+                        * If the racey nature of this approach (as the signal
+                        * will be arbitrated by hardware, and can fire as we
+                        * roll back state) results in the core responding
+                        * despite being pushed back to the IDLE state, it
+                        * should be no different than if the core took some
+                        * interrupt while IDLE.
+                        */
+                       if (active_processor != processor) {
+                               /*
+                                * Squash all of the processor state back to some
+                                * reasonable facsimile of PROCESSOR_IDLE.
+                                */
+
+                               processor_state_update_idle(active_processor);
+                               active_processor->deadline = UINT64_MAX;
+                               pset_update_processor_state(pset, active_processor, PROCESSOR_IDLE);
+                               bit_clear(pset->pending_deferred_AST_cpu_mask, active_processor->cpu_id);
+                               machine_signal_idle_cancel(active_processor);
+                       }
+               }
+       }
+
+       pset_unlock(pset);
+}
+#else
+/* We don't support deferred ASTs; everything is candycanes and sunshine. */
+#endif
+
+static void
+thread_csw_callout(
+       thread_t            old,
+       thread_t            new,
+       uint64_t            timestamp)
+{
+       perfcontrol_event event = (new->state & TH_IDLE) ? IDLE : CONTEXT_SWITCH;
+       uint64_t same_pri_latency = (new->state & TH_IDLE) ? 0 : new->same_pri_latency;
+       machine_switch_perfcontrol_context(event, timestamp, 0,
+           same_pri_latency, old, new);
+}
+
+
+/*
+ *     thread_dispatch:
+ *
+ *     Handle threads at context switch.  Re-dispatch other thread
+ *     if still running, otherwise update run state and perform
+ *     special actions.  Update quantum for other thread and begin
+ *     the quantum for ourselves.
+ *
+ *      "thread" is the old thread that we have switched away from.
+ *      "self" is the new current thread that we have context switched to
+ *
+ *     Called at splsched.
+ *
+ */
+void
+thread_dispatch(
+       thread_t                thread,
+       thread_t                self)
+{
+       processor_t             processor = self->last_processor;
+
+       assert(processor == current_processor());
+       assert(self == current_thread_volatile());
+       assert(thread != self);
+
+       if (thread != THREAD_NULL) {
+               /*
+                * Do the perfcontrol callout for context switch.
+                * The reason we do this here is:
+                * - thread_dispatch() is called from various places that are not
+                *   the direct context switch path for eg. processor shutdown etc.
+                *   So adding the callout here covers all those cases.
+                * - We want this callout as early as possible to be close
+                *   to the timestamp taken in thread_invoke()
+                * - We want to avoid holding the thread lock while doing the
+                *   callout
+                * - We do not want to callout if "thread" is NULL.
+                */
+               thread_csw_callout(thread, self, processor->last_dispatch);
+
+#if KASAN
+               if (thread->continuation != NULL) {
+                       /*
+                        * Thread has a continuation and the normal stack is going away.
+                        * Unpoison the stack and mark all fakestack objects as unused.
+                        */
+                       kasan_fakestack_drop(thread);
+                       if (thread->kernel_stack) {
+                               kasan_unpoison_stack(thread->kernel_stack, kernel_stack_size);
+                       }
+               }
+
+               /*
+                * Free all unused fakestack objects.
+                */
+               kasan_fakestack_gc(thread);
+#endif
+
+               /*
+                *      If blocked at a continuation, discard
+                *      the stack.
+                */
+               if (thread->continuation != NULL && thread->kernel_stack != 0) {
+                       stack_free(thread);
+               }
+
+               if (thread->state & TH_IDLE) {
+                       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                           MACHDBG_CODE(DBG_MACH_SCHED, MACH_DISPATCH) | DBG_FUNC_NONE,
+                           (uintptr_t)thread_tid(thread), 0, thread->state,
+                           sched_run_buckets[TH_BUCKET_RUN], 0);
+               } else {
+                       int64_t consumed;
+                       int64_t remainder = 0;
+
+                       if (processor->quantum_end > processor->last_dispatch) {
+                               remainder = processor->quantum_end -
+                                   processor->last_dispatch;
+                       }
+
+                       consumed = thread->quantum_remaining - remainder;
+
+                       if ((thread->reason & AST_LEDGER) == 0) {
+                               /*
+                                * Bill CPU time to both the task and
+                                * the individual thread.
+                                */
+                               ledger_credit_thread(thread, thread->t_ledger,
+                                   task_ledgers.cpu_time, consumed);
+                               ledger_credit_thread(thread, thread->t_threadledger,
+                                   thread_ledgers.cpu_time, consumed);
+                               if (thread->t_bankledger) {
+                                       ledger_credit_thread(thread, thread->t_bankledger,
+                                           bank_ledgers.cpu_time,
+                                           (consumed - thread->t_deduct_bank_ledger_time));
+                               }
+                               thread->t_deduct_bank_ledger_time = 0;
+                       }
+
+                       wake_lock(thread);
+                       thread_lock(thread);
+
+                       /*
+                        * Apply a priority floor if the thread holds a kernel resource
+                        * Do this before checking starting_pri to avoid overpenalizing
+                        * repeated rwlock blockers.
+                        */
+                       if (__improbable(thread->rwlock_count != 0)) {
+                               lck_rw_set_promotion_locked(thread);
+                       }
+
+                       boolean_t keep_quantum = processor->first_timeslice;
+
+                       /*
+                        * Treat a thread which has dropped priority since it got on core
+                        * as having expired its quantum.
+                        */
+                       if (processor->starting_pri > thread->sched_pri) {
+                               keep_quantum = FALSE;
+                       }
+
+                       /* Compute remainder of current quantum. */
+                       if (keep_quantum &&
+                           processor->quantum_end > processor->last_dispatch) {
+                               thread->quantum_remaining = (uint32_t)remainder;
+                       } else {
+                               thread->quantum_remaining = 0;
+                       }
+
+                       if (thread->sched_mode == TH_MODE_REALTIME) {
+                               /*
+                                *      Cancel the deadline if the thread has
+                                *      consumed the entire quantum.
+                                */
+                               if (thread->quantum_remaining == 0) {
+                                       thread->realtime.deadline = UINT64_MAX;
+                               }
+                       } else {
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+                               /*
+                                *      For non-realtime threads treat a tiny
+                                *      remaining quantum as an expired quantum
+                                *      but include what's left next time.
+                                */
+                               if (thread->quantum_remaining < min_std_quantum) {
+                                       thread->reason |= AST_QUANTUM;
+                                       thread->quantum_remaining += SCHED(initial_quantum_size)(thread);
+                               }
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
+                       }
+
+                       /*
+                        *      If we are doing a direct handoff then
+                        *      take the remainder of the quantum.
+                        */
+                       if ((thread->reason & (AST_HANDOFF | AST_QUANTUM)) == AST_HANDOFF) {
+                               self->quantum_remaining = thread->quantum_remaining;
+                               thread->reason |= AST_QUANTUM;
+                               thread->quantum_remaining = 0;
+                       } else {
+#if defined(CONFIG_SCHED_MULTIQ)
+                               if (SCHED(sched_groups_enabled) &&
+                                   thread->sched_group == self->sched_group) {
+                                       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                                           MACHDBG_CODE(DBG_MACH_SCHED, MACH_QUANTUM_HANDOFF),
+                                           self->reason, (uintptr_t)thread_tid(thread),
+                                           self->quantum_remaining, thread->quantum_remaining, 0);
+
+                                       self->quantum_remaining = thread->quantum_remaining;
+                                       thread->quantum_remaining = 0;
+                                       /* Don't set AST_QUANTUM here - old thread might still want to preempt someone else */
+                               }
+#endif /* defined(CONFIG_SCHED_MULTIQ) */
+                       }
+
+                       thread->computation_metered += (processor->last_dispatch - thread->computation_epoch);
+
+                       if (!(thread->state & TH_WAIT)) {
+                               /*
+                                *      Still runnable.
+                                */
+                               thread->last_made_runnable_time = thread->last_basepri_change_time = processor->last_dispatch;
+
+                               machine_thread_going_off_core(thread, FALSE, processor->last_dispatch, TRUE);
+
+                               ast_t reason = thread->reason;
+                               sched_options_t options = SCHED_NONE;
+
+                               if (reason & AST_REBALANCE) {
+                                       options |= SCHED_REBALANCE;
+                                       if (reason & AST_QUANTUM) {
+                                               /*
+                                                * Having gone to the trouble of forcing this thread off a less preferred core,
+                                                * we should force the preferable core to reschedule immediately to give this
+                                                * thread a chance to run instead of just sitting on the run queue where
+                                                * it may just be stolen back by the idle core we just forced it off.
+                                                * But only do this at the end of a quantum to prevent cascading effects.
+                                                */
+                                               options |= SCHED_PREEMPT;
+                                       }
+                               }
+
+                               if (reason & AST_QUANTUM) {
+                                       options |= SCHED_TAILQ;
+                               } else if (reason & AST_PREEMPT) {
+                                       options |= SCHED_HEADQ;
+                               } else {
+                                       options |= (SCHED_PREEMPT | SCHED_TAILQ);
+                               }
+
+                               thread_setrun(thread, options);
+
+                               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                                   MACHDBG_CODE(DBG_MACH_SCHED, MACH_DISPATCH) | DBG_FUNC_NONE,
+                                   (uintptr_t)thread_tid(thread), thread->reason, thread->state,
+                                   sched_run_buckets[TH_BUCKET_RUN], 0);
+
+                               if (thread->wake_active) {
+                                       thread->wake_active = FALSE;
+                                       thread_unlock(thread);
+
+                                       thread_wakeup(&thread->wake_active);
+                               } else {
+                                       thread_unlock(thread);
+                               }
+
+                               wake_unlock(thread);
+                       } else {
+                               /*
+                                *      Waiting.
+                                */
+                               boolean_t should_terminate = FALSE;
+                               uint32_t new_run_count;
+                               int thread_state = thread->state;
+
+                               /* Only the first call to thread_dispatch
+                                * after explicit termination should add
+                                * the thread to the termination queue
+                                */
+                               if ((thread_state & (TH_TERMINATE | TH_TERMINATE2)) == TH_TERMINATE) {
+                                       should_terminate = TRUE;
+                                       thread_state |= TH_TERMINATE2;
+                               }
+
+                               timer_stop(&thread->runnable_timer, processor->last_dispatch);
+
+                               thread_state &= ~TH_RUN;
+                               thread->state = thread_state;
+
+                               thread->last_made_runnable_time = thread->last_basepri_change_time = THREAD_NOT_RUNNABLE;
+                               thread->chosen_processor = PROCESSOR_NULL;
+
+                               new_run_count = SCHED(run_count_decr)(thread);
+
+#if CONFIG_SCHED_SFI
+                               if (thread->reason & AST_SFI) {
+                                       thread->wait_sfi_begin_time = processor->last_dispatch;
+                               }
+#endif
+                               machine_thread_going_off_core(thread, should_terminate, processor->last_dispatch, FALSE);
+
+                               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                                   MACHDBG_CODE(DBG_MACH_SCHED, MACH_DISPATCH) | DBG_FUNC_NONE,
+                                   (uintptr_t)thread_tid(thread), thread->reason, thread_state,
+                                   new_run_count, 0);
+
+                               if (thread_state & TH_WAIT_REPORT) {
+                                       (*thread->sched_call)(SCHED_CALL_BLOCK, thread);
+                               }
+
+                               if (thread->wake_active) {
+                                       thread->wake_active = FALSE;
+                                       thread_unlock(thread);
+
+                                       thread_wakeup(&thread->wake_active);
+                               } else {
+                                       thread_unlock(thread);
+                               }
+
+                               wake_unlock(thread);
+
+                               if (should_terminate) {
+                                       thread_terminate_enqueue(thread);
+                               }
+                       }
+               }
+       }
+
+       int urgency = THREAD_URGENCY_NONE;
+       uint64_t latency = 0;
+
+       /* Update (new) current thread and reprogram quantum timer */
+       thread_lock(self);
+
+       if (!(self->state & TH_IDLE)) {
+               uint64_t        arg1, arg2;
+
+#if CONFIG_SCHED_SFI
+               ast_t                   new_ast;
+
+               new_ast = sfi_thread_needs_ast(self, NULL);
+
+               if (new_ast != AST_NONE) {
+                       ast_on(new_ast);
+               }
+#endif
+
+               assertf(processor->last_dispatch >= self->last_made_runnable_time,
+                   "Non-monotonic time? dispatch at 0x%llx, runnable at 0x%llx",
+                   processor->last_dispatch, self->last_made_runnable_time);
+
+               assert(self->last_made_runnable_time <= self->last_basepri_change_time);
+
+               latency = processor->last_dispatch - self->last_made_runnable_time;
+               assert(latency >= self->same_pri_latency);
+
+               urgency = thread_get_urgency(self, &arg1, &arg2);
+
+               thread_tell_urgency(urgency, arg1, arg2, latency, self);
+
+               /*
+                *      Get a new quantum if none remaining.
+                */
+               if (self->quantum_remaining == 0) {
+                       thread_quantum_init(self);
+               }
+
+               /*
+                *      Set up quantum timer and timeslice.
+                */
+               processor->quantum_end = processor->last_dispatch + self->quantum_remaining;
+               timer_call_quantum_timer_enter(&processor->quantum_timer, self,
+                   processor->quantum_end, processor->last_dispatch);
+
+               processor->first_timeslice = TRUE;
+       } else {
+               timer_call_quantum_timer_cancel(&processor->quantum_timer);
+               processor->first_timeslice = FALSE;
+
+               thread_tell_urgency(THREAD_URGENCY_NONE, 0, 0, 0, self);
+       }
+
+       assert(self->block_hint == kThreadWaitNone);
+       self->computation_epoch = processor->last_dispatch;
+       self->reason = AST_NONE;
+       processor->starting_pri = self->sched_pri;
+
+       thread_unlock(self);
+
+       machine_thread_going_on_core(self, urgency, latency, self->same_pri_latency,
+           processor->last_dispatch);
+
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+       /*
+        * TODO: Can we state that redispatching our old thread is also
+        * uninteresting?
+        */
+       if ((os_atomic_load(&sched_run_buckets[TH_BUCKET_RUN], relaxed) == 1) && !(self->state & TH_IDLE)) {
+               pset_cancel_deferred_dispatch(processor->processor_set, processor);
+       }
+#endif
+}
+
+/*
+ *     thread_block_reason:
+ *
+ *     Forces a reschedule, blocking the caller if a wait
+ *     has been asserted.
+ *
+ *     If a continuation is specified, then thread_invoke will
+ *     attempt to discard the thread's kernel stack.  When the
+ *     thread resumes, it will execute the continuation function
+ *     on a new kernel stack.
+ */
+counter(mach_counter_t  c_thread_block_calls = 0; )
+
+wait_result_t
+thread_block_reason(
+       thread_continue_t       continuation,
+       void                            *parameter,
+       ast_t                           reason)
+{
+       thread_t        self = current_thread();
+       processor_t     processor;
+       thread_t        new_thread;
+       spl_t           s;
+
+       counter(++c_thread_block_calls);
+
+       s = splsched();
+
+       processor = current_processor();
+
+       /* If we're explicitly yielding, force a subsequent quantum */
+       if (reason & AST_YIELD) {
+               processor->first_timeslice = FALSE;
+       }
+
+       /* We're handling all scheduling AST's */
+       ast_off(AST_SCHEDULING);
+
+#if PROC_REF_DEBUG
+       if ((continuation != NULL) && (self->task != kernel_task)) {
+               if (uthread_get_proc_refcount(self->uthread) != 0) {
+                       panic("thread_block_reason with continuation uthread %p with uu_proc_refcount != 0", self->uthread);
+               }
+       }
+#endif
+
+       self->continuation = continuation;
+       self->parameter = parameter;
+
+       if (self->state & ~(TH_RUN | TH_IDLE)) {
+               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                   MACHDBG_CODE(DBG_MACH_SCHED, MACH_BLOCK),
+                   reason, VM_KERNEL_UNSLIDE(continuation), 0, 0, 0);
+       }
+
+       do {
+               thread_lock(self);
+               new_thread = thread_select(self, processor, &reason);
+               thread_unlock(self);
+       } while (!thread_invoke(self, new_thread, reason));
+
+       splx(s);
+
+       return self->wait_result;
+}
+
+/*
+ *     thread_block:
+ *
+ *     Block the current thread if a wait has been asserted.
+ */
+wait_result_t
+thread_block(
+       thread_continue_t       continuation)
+{
+       return thread_block_reason(continuation, NULL, AST_NONE);
+}
+
+wait_result_t
+thread_block_parameter(
+       thread_continue_t       continuation,
+       void                            *parameter)
+{
+       return thread_block_reason(continuation, parameter, AST_NONE);
+}
+
+/*
+ *     thread_run:
+ *
+ *     Switch directly from the current thread to the
+ *     new thread, handing off our quantum if appropriate.
+ *
+ *     New thread must be runnable, and not on a run queue.
+ *
+ *     Called at splsched.
+ */
+int
+thread_run(
+       thread_t                        self,
+       thread_continue_t       continuation,
+       void                            *parameter,
+       thread_t                        new_thread)
+{
+       ast_t reason = AST_NONE;
+
+       if ((self->state & TH_IDLE) == 0) {
+               reason = AST_HANDOFF;
+       }
+
+       self->continuation = continuation;
+       self->parameter = parameter;
+
+       while (!thread_invoke(self, new_thread, reason)) {
+               /* the handoff failed, so we have to fall back to the normal block path */
+               processor_t processor = current_processor();
+
+               reason = AST_NONE;
+
+               thread_lock(self);
+               new_thread = thread_select(self, processor, &reason);
+               thread_unlock(self);
+       }
+
+       return self->wait_result;
+}
+
+/*
+ *     thread_continue:
+ *
+ *     Called at splsched when a thread first receives
+ *     a new stack after a continuation.
+ *
+ *     Called with THREAD_NULL as the old thread when
+ *     invoked by machine_load_context.
+ */
+void
+thread_continue(
+       thread_t        thread)
+{
+       thread_t                self = current_thread();
+       thread_continue_t       continuation;
+       void                    *parameter;
+
+       DTRACE_SCHED(on__cpu);
+
+       continuation = self->continuation;
+       parameter = self->parameter;
+
+       assert(continuation != NULL);
+
+#if KPERF
+       kperf_on_cpu(self, continuation, NULL);
+#endif
+
+       thread_dispatch(thread, self);
+
+       self->continuation = self->parameter = NULL;
+
+#if INTERRUPT_MASKED_DEBUG
+       /* Reset interrupt-masked spin debugging timeout */
+       ml_spin_debug_clear(self);
+#endif
+
+       TLOG(1, "thread_continue: calling call_continuation\n");
+
+       boolean_t enable_interrupts = TRUE;
+
+       /* bootstrap thread, idle thread need to stay interrupts-disabled */
+       if (thread == THREAD_NULL || (self->state & TH_IDLE)) {
+               enable_interrupts = FALSE;
+       }
+
+       call_continuation(continuation, parameter, self->wait_result, enable_interrupts);
+       /*NOTREACHED*/
+}
+
+void
+thread_quantum_init(thread_t thread)
+{
+       if (thread->sched_mode == TH_MODE_REALTIME) {
+               thread->quantum_remaining = thread->realtime.computation;
+       } else {
+               thread->quantum_remaining = SCHED(initial_quantum_size)(thread);
+       }
+}
+
+uint32_t
+sched_timeshare_initial_quantum_size(thread_t thread)
+{
+       if ((thread != THREAD_NULL) && thread->th_sched_bucket == TH_BUCKET_SHARE_BG) {
+               return bg_quantum;
+       } else {
+               return std_quantum;
+       }
+}
+
+/*
+ *     run_queue_init:
+ *
+ *     Initialize a run queue before first use.
+ */
+void
+run_queue_init(
+       run_queue_t             rq)
+{
+       rq->highq = NOPRI;
+       for (u_int i = 0; i < BITMAP_LEN(NRQS); i++) {
+               rq->bitmap[i] = 0;
+       }
+       rq->urgency = rq->count = 0;
+       for (int i = 0; i < NRQS; i++) {
+               circle_queue_init(&rq->queues[i]);
+       }
+}
+
+/*
+ *     run_queue_dequeue:
+ *
+ *     Perform a dequeue operation on a run queue,
+ *     and return the resulting thread.
+ *
+ *     The run queue must be locked (see thread_run_queue_remove()
+ *     for more info), and not empty.
+ */
+thread_t
+run_queue_dequeue(
+       run_queue_t     rq,
+       sched_options_t options)
+{
+       thread_t        thread;
+       circle_queue_t  queue = &rq->queues[rq->highq];
+
+       if (options & SCHED_HEADQ) {
+               thread = cqe_dequeue_head(queue, struct thread, runq_links);
+       } else {
+               thread = cqe_dequeue_tail(queue, struct thread, runq_links);
+       }
+
+       assert(thread != THREAD_NULL);
+       assert_thread_magic(thread);
+
+       thread->runq = PROCESSOR_NULL;
+       SCHED_STATS_RUNQ_CHANGE(&rq->runq_stats, rq->count);
+       rq->count--;
+       if (SCHED(priority_is_urgent)(rq->highq)) {
+               rq->urgency--; assert(rq->urgency >= 0);
+       }
+       if (circle_queue_empty(queue)) {
+               bitmap_clear(rq->bitmap, rq->highq);
+               rq->highq = bitmap_first(rq->bitmap, NRQS);
+       }
+
+       return thread;
+}
+
+/*
+ *     run_queue_enqueue:
+ *
+ *     Perform a enqueue operation on a run queue.
+ *
+ *     The run queue must be locked (see thread_run_queue_remove()
+ *     for more info).
+ */
+boolean_t
+run_queue_enqueue(
+       run_queue_t      rq,
+       thread_t         thread,
+       sched_options_t  options)
+{
+       circle_queue_t  queue = &rq->queues[thread->sched_pri];
+       boolean_t       result = FALSE;
+
+       assert_thread_magic(thread);
+
+       if (circle_queue_empty(queue)) {
+               circle_enqueue_tail(queue, &thread->runq_links);
+
+               rq_bitmap_set(rq->bitmap, thread->sched_pri);
+               if (thread->sched_pri > rq->highq) {
+                       rq->highq = thread->sched_pri;
+                       result = TRUE;
+               }
+       } else {
+               if (options & SCHED_TAILQ) {
+                       circle_enqueue_tail(queue, &thread->runq_links);
+               } else {
+                       circle_enqueue_head(queue, &thread->runq_links);
+               }
+       }
+       if (SCHED(priority_is_urgent)(thread->sched_pri)) {
+               rq->urgency++;
+       }
+       SCHED_STATS_RUNQ_CHANGE(&rq->runq_stats, rq->count);
+       rq->count++;
+
+       return result;
+}
+
+/*
+ *     run_queue_remove:
+ *
+ *     Remove a specific thread from a runqueue.
+ *
+ *     The run queue must be locked.
+ */
+void
+run_queue_remove(
+       run_queue_t    rq,
+       thread_t       thread)
+{
+       circle_queue_t  queue = &rq->queues[thread->sched_pri];
+
+       assert(thread->runq != PROCESSOR_NULL);
+       assert_thread_magic(thread);
+
+       circle_dequeue(queue, &thread->runq_links);
+       SCHED_STATS_RUNQ_CHANGE(&rq->runq_stats, rq->count);
+       rq->count--;
+       if (SCHED(priority_is_urgent)(thread->sched_pri)) {
+               rq->urgency--; assert(rq->urgency >= 0);
+       }
+
+       if (circle_queue_empty(queue)) {
+               /* update run queue status */
+               bitmap_clear(rq->bitmap, thread->sched_pri);
+               rq->highq = bitmap_first(rq->bitmap, NRQS);
+       }
+
+       thread->runq = PROCESSOR_NULL;
+}
+
+/*
+ *      run_queue_peek
+ *
+ *      Peek at the runq and return the highest
+ *      priority thread from the runq.
+ *
+ *     The run queue must be locked.
+ */
+thread_t
+run_queue_peek(
+       run_queue_t    rq)
+{
+       if (rq->count > 0) {
+               circle_queue_t queue = &rq->queues[rq->highq];
+               thread_t thread = cqe_queue_first(queue, struct thread, runq_links);
+               assert_thread_magic(thread);
+               return thread;
+       } else {
+               return THREAD_NULL;
+       }
+}
+
+/* Assumes RT lock is not held, and acquires splsched/rt_lock itself */
+void
+sched_rtglobal_runq_scan(sched_update_scan_context_t scan_context)
+{
+       spl_t           s;
+       thread_t        thread;
+
+       processor_set_t pset = &pset0;
+
+       s = splsched();
+       rt_lock_lock(pset);
+
+       qe_foreach_element_safe(thread, &pset->rt_runq.queue, runq_links) {
+               if (thread->last_made_runnable_time < scan_context->earliest_rt_make_runnable_time) {
+                       scan_context->earliest_rt_make_runnable_time = thread->last_made_runnable_time;
+               }
+       }
+
+       rt_lock_unlock(pset);
+       splx(s);
+}
+
+int64_t
+sched_rtglobal_runq_count_sum(void)
+{
+       return pset0.rt_runq.runq_stats.count_sum;
+}
+
+/*
+ *     realtime_queue_insert:
+ *
+ *     Enqueue a thread for realtime execution.
+ */
+static boolean_t
+realtime_queue_insert(processor_t processor, processor_set_t pset, thread_t thread)
+{
+       queue_t     queue       = &SCHED(rt_runq)(pset)->queue;
+       uint64_t    deadline    = thread->realtime.deadline;
+       boolean_t   preempt     = FALSE;
+
+       rt_lock_lock(pset);
+
+       if (queue_empty(queue)) {
+               enqueue_tail(queue, &thread->runq_links);
+               preempt = TRUE;
+       } else {
+               /* Insert into rt_runq in thread deadline order */
+               queue_entry_t iter;
+               qe_foreach(iter, queue) {
+                       thread_t iter_thread = qe_element(iter, struct thread, runq_links);
+                       assert_thread_magic(iter_thread);
+
+                       if (deadline < iter_thread->realtime.deadline) {
+                               if (iter == queue_first(queue)) {
+                                       preempt = TRUE;
+                               }
+                               insque(&thread->runq_links, queue_prev(iter));
+                               break;
+                       } else if (iter == queue_last(queue)) {
+                               enqueue_tail(queue, &thread->runq_links);
+                               break;
+                       }
+               }
+       }
+
+       thread->runq = processor;
+       SCHED_STATS_RUNQ_CHANGE(&SCHED(rt_runq)(pset)->runq_stats, rt_runq_count(pset));
+       rt_runq_count_incr(pset);
+
+       rt_lock_unlock(pset);
+
+       return preempt;
+}
+
+/*
+ *     realtime_setrun:
+ *
+ *     Dispatch a thread for realtime execution.
+ *
+ *     Thread must be locked.  Associated pset must
+ *     be locked, and is returned unlocked.
+ */
+static void
+realtime_setrun(
+       processor_t                     processor,
+       thread_t                        thread)
+{
+       processor_set_t pset = processor->processor_set;
+       pset_assert_locked(pset);
+       ast_t preempt;
+
+       sched_ipi_type_t ipi_type = SCHED_IPI_NONE;
+
+       thread->chosen_processor = processor;
+
+       /* <rdar://problem/15102234> */
+       assert(thread->bound_processor == PROCESSOR_NULL);
+
+       if (processor->current_pri < BASEPRI_RTQUEUES) {
+               preempt = (AST_PREEMPT | AST_URGENT);
+       } else if (thread->realtime.deadline < processor->deadline) {
+               preempt = (AST_PREEMPT | AST_URGENT);
+       } else {
+               preempt = AST_NONE;
+       }
+
+       realtime_queue_insert(processor, pset, thread);
+
+       ipi_type = SCHED_IPI_NONE;
+       if (preempt != AST_NONE) {
+               if (processor->state == PROCESSOR_IDLE) {
+                       processor_state_update_from_thread(processor, thread);
+                       processor->deadline = thread->realtime.deadline;
+                       pset_update_processor_state(pset, processor, PROCESSOR_DISPATCHING);
+                       if (processor == current_processor()) {
+                               ast_on(preempt);
+                       } else {
+                               ipi_type = sched_ipi_action(processor, thread, true, SCHED_IPI_EVENT_PREEMPT);
+                       }
+               } else if (processor->state == PROCESSOR_DISPATCHING) {
+                       if ((processor->current_pri < thread->sched_pri) || (processor->deadline > thread->realtime.deadline)) {
+                               processor_state_update_from_thread(processor, thread);
+                               processor->deadline = thread->realtime.deadline;
+                       }
+               } else {
+                       if (processor == current_processor()) {
+                               ast_on(preempt);
+
+                               if ((preempt & AST_URGENT) == AST_URGENT) {
+                                       bit_set(pset->pending_AST_URGENT_cpu_mask, processor->cpu_id);
+                               }
+
+                               if ((preempt & AST_PREEMPT) == AST_PREEMPT) {
+                                       bit_set(pset->pending_AST_PREEMPT_cpu_mask, processor->cpu_id);
+                               }
+                       } else {
+                               ipi_type = sched_ipi_action(processor, thread, false, SCHED_IPI_EVENT_PREEMPT);
+                       }
+               }
+       } else {
+               /* Selected processor was too busy, just keep thread enqueued and let other processors drain it naturally. */
+       }
+
+       pset_unlock(pset);
+       sched_ipi_perform(processor, ipi_type);
+}
+
+
+sched_ipi_type_t
+sched_ipi_deferred_policy(processor_set_t pset, processor_t dst,
+    __unused sched_ipi_event_t event)
+{
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+       if (!bit_test(pset->pending_deferred_AST_cpu_mask, dst->cpu_id)) {
+               return SCHED_IPI_DEFERRED;
+       }
+#else /* CONFIG_SCHED_DEFERRED_AST */
+       panic("Request for deferred IPI on an unsupported platform; pset: %p CPU: %d", pset, dst->cpu_id);
+#endif /* CONFIG_SCHED_DEFERRED_AST */
+       return SCHED_IPI_NONE;
+}
+
+sched_ipi_type_t
+sched_ipi_action(processor_t dst, thread_t thread, boolean_t dst_idle, sched_ipi_event_t event)
+{
+       sched_ipi_type_t ipi_type = SCHED_IPI_NONE;
+       assert(dst != NULL);
+
+       processor_set_t pset = dst->processor_set;
+       if (current_processor() == dst) {
+               return SCHED_IPI_NONE;
+       }
+
+       if (bit_test(pset->pending_AST_URGENT_cpu_mask, dst->cpu_id)) {
+               return SCHED_IPI_NONE;
+       }
+
+       ipi_type = SCHED(ipi_policy)(dst, thread, dst_idle, event);
+       switch (ipi_type) {
+       case SCHED_IPI_NONE:
+               return SCHED_IPI_NONE;
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+       case SCHED_IPI_DEFERRED:
+               bit_set(pset->pending_deferred_AST_cpu_mask, dst->cpu_id);
+               break;
+#endif /* CONFIG_SCHED_DEFERRED_AST */
+       default:
+               bit_set(pset->pending_AST_URGENT_cpu_mask, dst->cpu_id);
+               bit_set(pset->pending_AST_PREEMPT_cpu_mask, dst->cpu_id);
+               break;
+       }
+       return ipi_type;
+}
+
+sched_ipi_type_t
+sched_ipi_policy(processor_t dst, thread_t thread, boolean_t dst_idle, sched_ipi_event_t event)
+{
+       sched_ipi_type_t ipi_type = SCHED_IPI_NONE;
+       boolean_t deferred_ipi_supported = false;
+       processor_set_t pset = dst->processor_set;
+
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+       deferred_ipi_supported = true;
+#endif /* CONFIG_SCHED_DEFERRED_AST */
+
+       switch (event) {
+       case SCHED_IPI_EVENT_SPILL:
+       case SCHED_IPI_EVENT_SMT_REBAL:
+       case SCHED_IPI_EVENT_REBALANCE:
+       case SCHED_IPI_EVENT_BOUND_THR:
+               /*
+                * The spill, SMT rebalance, rebalance and the bound thread
+                * scenarios use immediate IPIs always.
+                */
+               ipi_type = dst_idle ? SCHED_IPI_IDLE : SCHED_IPI_IMMEDIATE;
+               break;
+       case SCHED_IPI_EVENT_PREEMPT:
+               /* In the preemption case, use immediate IPIs for RT threads */
+               if (thread && (thread->sched_pri >= BASEPRI_RTQUEUES)) {
+                       ipi_type = dst_idle ? SCHED_IPI_IDLE : SCHED_IPI_IMMEDIATE;
+                       break;
+               }
+
+               /*
+                * For Non-RT threads preemption,
+                * If the core is active, use immediate IPIs.
+                * If the core is idle, use deferred IPIs if supported; otherwise immediate IPI.
+                */
+               if (deferred_ipi_supported && dst_idle) {
+                       return sched_ipi_deferred_policy(pset, dst, event);
+               }
+               ipi_type = dst_idle ? SCHED_IPI_IDLE : SCHED_IPI_IMMEDIATE;
+               break;
+       default:
+               panic("Unrecognized scheduler IPI event type %d", event);
+       }
+       assert(ipi_type != SCHED_IPI_NONE);
+       return ipi_type;
+}
+
+void
+sched_ipi_perform(processor_t dst, sched_ipi_type_t ipi)
+{
+       switch (ipi) {
+       case SCHED_IPI_NONE:
+               break;
+       case SCHED_IPI_IDLE:
+               machine_signal_idle(dst);
+               break;
+       case SCHED_IPI_IMMEDIATE:
+               cause_ast_check(dst);
+               break;
+       case SCHED_IPI_DEFERRED:
+               machine_signal_idle_deferred(dst);
+               break;
+       default:
+               panic("Unrecognized scheduler IPI type: %d", ipi);
+       }
+}
+
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+
+boolean_t
+priority_is_urgent(int priority)
+{
+       return bitmap_test(sched_preempt_pri, priority) ? TRUE : FALSE;
+}
+
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
+
+/*
+ *     processor_setrun:
+ *
+ *     Dispatch a thread for execution on a
+ *     processor.
+ *
+ *     Thread must be locked.  Associated pset must
+ *     be locked, and is returned unlocked.
+ */
+static void
+processor_setrun(
+       processor_t                     processor,
+       thread_t                        thread,
+       integer_t                       options)
+{
+       processor_set_t pset = processor->processor_set;
+       pset_assert_locked(pset);
+       ast_t preempt;
+       enum { eExitIdle, eInterruptRunning, eDoNothing } ipi_action = eDoNothing;
+
+       sched_ipi_type_t ipi_type = SCHED_IPI_NONE;
+
+       thread->chosen_processor = processor;
+
+       /*
+        *      Set preemption mode.
+        */
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+       /* TODO: Do we need to care about urgency (see rdar://problem/20136239)? */
+#endif
+       if (SCHED(priority_is_urgent)(thread->sched_pri) && thread->sched_pri > processor->current_pri) {
+               preempt = (AST_PREEMPT | AST_URGENT);
+       } else if (processor->active_thread && thread_eager_preemption(processor->active_thread)) {
+               preempt = (AST_PREEMPT | AST_URGENT);
+       } else if ((thread->sched_mode == TH_MODE_TIMESHARE) && (thread->sched_pri < thread->base_pri)) {
+               if (SCHED(priority_is_urgent)(thread->base_pri) && thread->sched_pri > processor->current_pri) {
+                       preempt = (options & SCHED_PREEMPT)? AST_PREEMPT: AST_NONE;
+               } else {
+                       preempt = AST_NONE;
+               }
+       } else {
+               preempt = (options & SCHED_PREEMPT)? AST_PREEMPT: AST_NONE;
+       }
+
+       if ((options & (SCHED_PREEMPT | SCHED_REBALANCE)) == (SCHED_PREEMPT | SCHED_REBALANCE)) {
+               /*
+                * Having gone to the trouble of forcing this thread off a less preferred core,
+                * we should force the preferable core to reschedule immediately to give this
+                * thread a chance to run instead of just sitting on the run queue where
+                * it may just be stolen back by the idle core we just forced it off.
+                */
+               preempt |= AST_PREEMPT;
+       }
+
+       SCHED(processor_enqueue)(processor, thread, options);
+       sched_update_pset_load_average(pset);
+
+       if (preempt != AST_NONE) {
+               if (processor->state == PROCESSOR_IDLE) {
+                       processor_state_update_from_thread(processor, thread);
+                       processor->deadline = UINT64_MAX;
+                       pset_update_processor_state(pset, processor, PROCESSOR_DISPATCHING);
+                       ipi_action = eExitIdle;
+               } else if (processor->state == PROCESSOR_DISPATCHING) {
+                       if (processor->current_pri < thread->sched_pri) {
+                               processor_state_update_from_thread(processor, thread);
+                               processor->deadline = UINT64_MAX;
+                       }
+               } else if ((processor->state == PROCESSOR_RUNNING ||
+                   processor->state == PROCESSOR_SHUTDOWN) &&
+                   (thread->sched_pri >= processor->current_pri)) {
+                       ipi_action = eInterruptRunning;
+               }
+       } else {
+               /*
+                * New thread is not important enough to preempt what is running, but
+                * special processor states may need special handling
+                */
+               if (processor->state == PROCESSOR_SHUTDOWN &&
+                   thread->sched_pri >= processor->current_pri) {
+                       ipi_action = eInterruptRunning;
+               } else if (processor->state == PROCESSOR_IDLE) {
+                       processor_state_update_from_thread(processor, thread);
+                       processor->deadline = UINT64_MAX;
+                       pset_update_processor_state(pset, processor, PROCESSOR_DISPATCHING);
+
+                       ipi_action = eExitIdle;
+               }
+       }
+
+       if (ipi_action != eDoNothing) {
+               if (processor == current_processor()) {
+                       if ((preempt = csw_check_locked(processor->active_thread, processor, pset, AST_NONE)) != AST_NONE) {
+                               ast_on(preempt);
+                       }
+
+                       if ((preempt & AST_URGENT) == AST_URGENT) {
+                               bit_set(pset->pending_AST_URGENT_cpu_mask, processor->cpu_id);
+                       } else {
+                               bit_clear(pset->pending_AST_URGENT_cpu_mask, processor->cpu_id);
+                       }
+
+                       if ((preempt & AST_PREEMPT) == AST_PREEMPT) {
+                               bit_set(pset->pending_AST_PREEMPT_cpu_mask, processor->cpu_id);
+                       } else {
+                               bit_clear(pset->pending_AST_PREEMPT_cpu_mask, processor->cpu_id);
+                       }
+               } else {
+                       sched_ipi_event_t event = (options & SCHED_REBALANCE) ? SCHED_IPI_EVENT_REBALANCE : SCHED_IPI_EVENT_PREEMPT;
+                       ipi_type = sched_ipi_action(processor, thread, (ipi_action == eExitIdle), event);
+               }
+       }
+       pset_unlock(pset);
+       sched_ipi_perform(processor, ipi_type);
+}
+
+/*
+ *     choose_next_pset:
+ *
+ *     Return the next sibling pset containing
+ *     available processors.
+ *
+ *     Returns the original pset if none other is
+ *     suitable.
+ */
+static processor_set_t
+choose_next_pset(
+       processor_set_t         pset)
+{
+       processor_set_t         nset = pset;
+
+       do {
+               nset = next_pset(nset);
+       } while (nset->online_processor_count < 1 && nset != pset);
+
+       return nset;
+}
+
+/*
+ *     choose_processor:
+ *
+ *     Choose a processor for the thread, beginning at
+ *     the pset.  Accepts an optional processor hint in
+ *     the pset.
+ *
+ *     Returns a processor, possibly from a different pset.
+ *
+ *     The thread must be locked.  The pset must be locked,
+ *     and the resulting pset is locked on return.
+ */
+processor_t
+choose_processor(
+       processor_set_t         starting_pset,
+       processor_t             processor,
+       thread_t                thread)
+{
+       processor_set_t pset = starting_pset;
+       processor_set_t nset;
+
+       assert(thread->sched_pri <= BASEPRI_RTQUEUES);
+
+       /*
+        * Prefer the hinted processor, when appropriate.
+        */
+
+       /* Fold last processor hint from secondary processor to its primary */
+       if (processor != PROCESSOR_NULL) {
+               processor = processor->processor_primary;
+       }
+
+       /*
+        * Only consult platform layer if pset is active, which
+        * it may not be in some cases when a multi-set system
+        * is going to sleep.
+        */
+       if (pset->online_processor_count) {
+               if ((processor == PROCESSOR_NULL) || (processor->processor_set == pset && processor->state == PROCESSOR_IDLE)) {
+                       processor_t mc_processor = machine_choose_processor(pset, processor);
+                       if (mc_processor != PROCESSOR_NULL) {
+                               processor = mc_processor->processor_primary;
+                       }
+               }
+       }
+
+       /*
+        * At this point, we may have a processor hint, and we may have
+        * an initial starting pset. If the hint is not in the pset, or
+        * if the hint is for a processor in an invalid state, discard
+        * the hint.
+        */
+       if (processor != PROCESSOR_NULL) {
+               if (processor->processor_set != pset) {
+                       processor = PROCESSOR_NULL;
+               } else if (!processor->is_recommended) {
+                       processor = PROCESSOR_NULL;
+               } else if ((thread->sched_pri >= BASEPRI_RTQUEUES) && !sched_ok_to_run_realtime_thread(pset, processor)) {
+                       processor = PROCESSOR_NULL;
+               } else {
+                       switch (processor->state) {
+                       case PROCESSOR_START:
+                       case PROCESSOR_SHUTDOWN:
+                       case PROCESSOR_OFF_LINE:
+                               /*
+                                * Hint is for a processor that cannot support running new threads.
+                                */
+                               processor = PROCESSOR_NULL;
+                               break;
+                       case PROCESSOR_IDLE:
+                               /*
+                                * Hint is for an idle processor. Assume it is no worse than any other
+                                * idle processor. The platform layer had an opportunity to provide
+                                * the "least cost idle" processor above.
+                                */
+                               return processor;
+                       case PROCESSOR_RUNNING:
+                       case PROCESSOR_DISPATCHING:
+                               /*
+                                * Hint is for an active CPU. This fast-path allows
+                                * realtime threads to preempt non-realtime threads
+                                * to regain their previous executing processor.
+                                */
+                               if ((thread->sched_pri >= BASEPRI_RTQUEUES) &&
+                                   (processor->current_pri < BASEPRI_RTQUEUES)) {
+                                       return processor;
+                               }
+
+                               /* Otherwise, use hint as part of search below */
+                               break;
+                       default:
+                               processor = PROCESSOR_NULL;
+                               break;
+                       }
+               }
+       }
+
+       /*
+        * Iterate through the processor sets to locate
+        * an appropriate processor. Seed results with
+        * a last-processor hint, if available, so that
+        * a search must find something strictly better
+        * to replace it.
+        *
+        * A primary/secondary pair of SMT processors are
+        * "unpaired" if the primary is busy but its
+        * corresponding secondary is idle (so the physical
+        * core has full use of its resources).
+        */
+
+       integer_t lowest_priority = MAXPRI + 1;
+       integer_t lowest_secondary_priority = MAXPRI + 1;
+       integer_t lowest_unpaired_primary_priority = MAXPRI + 1;
+       integer_t lowest_idle_secondary_priority = MAXPRI + 1;
+       integer_t lowest_count = INT_MAX;
+       uint64_t  furthest_deadline = 1;
+       processor_t lp_processor = PROCESSOR_NULL;
+       processor_t lp_unpaired_primary_processor = PROCESSOR_NULL;
+       processor_t lp_idle_secondary_processor = PROCESSOR_NULL;
+       processor_t lp_paired_secondary_processor = PROCESSOR_NULL;
+       processor_t lc_processor = PROCESSOR_NULL;
+       processor_t fd_processor = PROCESSOR_NULL;
+
+       if (processor != PROCESSOR_NULL) {
+               /* All other states should be enumerated above. */
+               assert(processor->state == PROCESSOR_RUNNING || processor->state == PROCESSOR_DISPATCHING);
+
+               lowest_priority = processor->current_pri;
+               lp_processor = processor;
+
+               if (processor->current_pri >= BASEPRI_RTQUEUES) {
+                       furthest_deadline = processor->deadline;
+                       fd_processor = processor;
+               }
+
+               lowest_count = SCHED(processor_runq_count)(processor);
+               lc_processor = processor;
+       }
+
+       do {
+               int cpuid;
+
+               if (thread->sched_pri >= BASEPRI_RTQUEUES) {
+                       processor = choose_processor_for_realtime_thread(pset);
+                       if (processor) {
+                               return processor;
+                       }
+               } else {
+                       /*
+                        * Choose an idle processor, in pset traversal order
+                        */
+
+                       uint64_t idle_primary_map = (pset->cpu_state_map[PROCESSOR_IDLE] &
+                           pset->primary_map &
+                           pset->recommended_bitmask);
+
+                       /* there shouldn't be a pending AST if the processor is idle */
+                       assert((idle_primary_map & pset->pending_AST_URGENT_cpu_mask) == 0);
+
+                       cpuid = lsb_first(idle_primary_map);
+                       if (cpuid >= 0) {
+                               processor = processor_array[cpuid];
+                               return processor;
+                       }
+               }
+
+               /*
+                * Otherwise, enumerate active and idle processors to find primary candidates
+                * with lower priority/etc.
+                */
+
+               uint64_t active_map = ((pset->cpu_state_map[PROCESSOR_RUNNING] | pset->cpu_state_map[PROCESSOR_DISPATCHING]) &
+                   pset->recommended_bitmask &
+                   ~pset->pending_AST_URGENT_cpu_mask);
+
+               if (SCHED(priority_is_urgent)(thread->sched_pri) == FALSE) {
+                       active_map &= ~pset->pending_AST_PREEMPT_cpu_mask;
+               }
+
+               active_map = bit_ror64(active_map, (pset->last_chosen + 1));
+               for (int rotid = lsb_first(active_map); rotid >= 0; rotid = lsb_next(active_map, rotid)) {
+                       cpuid = ((rotid + pset->last_chosen + 1) & 63);
+                       processor = processor_array[cpuid];
+
+                       integer_t cpri = processor->current_pri;
+                       processor_t primary = processor->processor_primary;
+                       if (primary != processor) {
+                               /* If primary is running a NO_SMT thread, don't choose its secondary */
+                               if (!((primary->state == PROCESSOR_RUNNING) && processor_active_thread_no_smt(primary))) {
+                                       if (cpri < lowest_secondary_priority) {
+                                               lowest_secondary_priority = cpri;
+                                               lp_paired_secondary_processor = processor;
+                                       }
+                               }
+                       } else {
+                               if (cpri < lowest_priority) {
+                                       lowest_priority = cpri;
+                                       lp_processor = processor;
+                               }
+                       }
+
+                       if ((cpri >= BASEPRI_RTQUEUES) && (processor->deadline > furthest_deadline)) {
+                               furthest_deadline = processor->deadline;
+                               fd_processor = processor;
+                       }
+
+                       integer_t ccount = SCHED(processor_runq_count)(processor);
+                       if (ccount < lowest_count) {
+                               lowest_count = ccount;
+                               lc_processor = processor;
+                       }
+               }
+
+               /*
+                * For SMT configs, these idle secondary processors must have active primary. Otherwise
+                * the idle primary would have short-circuited the loop above
+                */
+               uint64_t idle_secondary_map = (pset->cpu_state_map[PROCESSOR_IDLE] &
+                   ~pset->primary_map &
+                   pset->recommended_bitmask);
+
+               /* there shouldn't be a pending AST if the processor is idle */
+               assert((idle_secondary_map & pset->pending_AST_URGENT_cpu_mask) == 0);
+               assert((idle_secondary_map & pset->pending_AST_PREEMPT_cpu_mask) == 0);
+
+               for (cpuid = lsb_first(idle_secondary_map); cpuid >= 0; cpuid = lsb_next(idle_secondary_map, cpuid)) {
+                       processor = processor_array[cpuid];
+
+                       processor_t cprimary = processor->processor_primary;
+
+                       integer_t primary_pri = cprimary->current_pri;
+
+                       /*
+                        * TODO: This should also make the same decisions
+                        * as secondary_can_run_realtime_thread
+                        *
+                        * TODO: Keep track of the pending preemption priority
+                        * of the primary to make this more accurate.
+                        */
+
+                       /* If the primary is running a no-smt thread, then don't choose its secondary */
+                       if (cprimary->state == PROCESSOR_RUNNING &&
+                           processor_active_thread_no_smt(cprimary)) {
+                               continue;
+                       }
+
+                       /*
+                        * Find the idle secondary processor with the lowest priority primary
+                        *
+                        * We will choose this processor as a fallback if we find no better
+                        * primary to preempt.
+                        */
+                       if (primary_pri < lowest_idle_secondary_priority) {
+                               lp_idle_secondary_processor = processor;
+                               lowest_idle_secondary_priority = primary_pri;
+                       }
+
+                       /* Find the the lowest priority active primary with idle secondary */
+                       if (primary_pri < lowest_unpaired_primary_priority) {
+                               /* If the primary processor is offline or starting up, it's not a candidate for this path */
+                               if (cprimary->state != PROCESSOR_RUNNING &&
+                                   cprimary->state != PROCESSOR_DISPATCHING) {
+                                       continue;
+                               }
+
+                               if (!cprimary->is_recommended) {
+                                       continue;
+                               }
+
+                               /* if the primary is pending preemption, don't try to re-preempt it */
+                               if (bit_test(pset->pending_AST_URGENT_cpu_mask, cprimary->cpu_id)) {
+                                       continue;
+                               }
+
+                               if (SCHED(priority_is_urgent)(thread->sched_pri) == FALSE &&
+                                   bit_test(pset->pending_AST_PREEMPT_cpu_mask, cprimary->cpu_id)) {
+                                       continue;
+                               }
+
+                               lowest_unpaired_primary_priority = primary_pri;
+                               lp_unpaired_primary_processor = cprimary;
+                       }
+               }
+
+               /*
+                * We prefer preempting a primary processor over waking up its secondary.
+                * The secondary will then be woken up by the preempted thread.
+                */
+               if (thread->sched_pri > lowest_unpaired_primary_priority) {
+                       pset->last_chosen = lp_unpaired_primary_processor->cpu_id;
+                       return lp_unpaired_primary_processor;
+               }
+
+               /*
+                * We prefer preempting a lower priority active processor over directly
+                * waking up an idle secondary.
+                * The preempted thread will then find the idle secondary.
+                */
+               if (thread->sched_pri > lowest_priority) {
+                       pset->last_chosen = lp_processor->cpu_id;
+                       return lp_processor;
+               }
+
+               if (thread->sched_pri >= BASEPRI_RTQUEUES) {
+                       /*
+                        * For realtime threads, the most important aspect is
+                        * scheduling latency, so we will pick an active
+                        * secondary processor in this pset, or preempt
+                        * another RT thread with a further deadline before
+                        * going to the next pset.
+                        */
+
+                       if (sched_allow_rt_smt && (thread->sched_pri > lowest_secondary_priority)) {
+                               pset->last_chosen = lp_paired_secondary_processor->cpu_id;
+                               return lp_paired_secondary_processor;
+                       }
+
+                       if (thread->realtime.deadline < furthest_deadline) {
+                               return fd_processor;
+                       }
+               }
+
+               /*
+                * lc_processor is used to indicate the best processor set run queue
+                * on which to enqueue a thread when all available CPUs are busy with
+                * higher priority threads, so try to make sure it is initialized.
+                */
+               if (lc_processor == PROCESSOR_NULL) {
+                       cpumap_t available_map = ((pset->cpu_state_map[PROCESSOR_IDLE] |
+                           pset->cpu_state_map[PROCESSOR_RUNNING] |
+                           pset->cpu_state_map[PROCESSOR_DISPATCHING]) &
+                           pset->recommended_bitmask);
+                       cpuid = lsb_first(available_map);
+                       if (cpuid >= 0) {
+                               lc_processor = processor_array[cpuid];
+                               lowest_count = SCHED(processor_runq_count)(lc_processor);
+                       }
+               }
+
+               /*
+                * Move onto the next processor set.
+                *
+                * If all primary processors in this pset are running a higher
+                * priority thread, move on to next pset. Only when we have
+                * exhausted the search for primary processors do we
+                * fall back to secondaries.
+                */
+               nset = next_pset(pset);
+
+               if (nset != starting_pset) {
+                       pset_unlock(pset);
+
+                       pset = nset;
+                       pset_lock(pset);
+               }
+       } while (nset != starting_pset);
+
+       /*
+        * Make sure that we pick a running processor,
+        * and that the correct processor set is locked.
+        * Since we may have unlocked the candidate processor's
+        * pset, it may have changed state.
+        *
+        * All primary processors are running a higher priority
+        * thread, so the only options left are enqueuing on
+        * the secondary processor that would perturb the least priority
+        * primary, or the least busy primary.
+        */
+       boolean_t fallback_processor = false;
+       do {
+               /* lowest_priority is evaluated in the main loops above */
+               if (lp_idle_secondary_processor != PROCESSOR_NULL) {
+                       processor = lp_idle_secondary_processor;
+                       lp_idle_secondary_processor = PROCESSOR_NULL;
+               } else if (lp_paired_secondary_processor != PROCESSOR_NULL) {
+                       processor = lp_paired_secondary_processor;
+                       lp_paired_secondary_processor = PROCESSOR_NULL;
+               } else if (lc_processor != PROCESSOR_NULL) {
+                       processor = lc_processor;
+                       lc_processor = PROCESSOR_NULL;
+               } else {
+                       /*
+                        * All processors are executing higher priority threads, and
+                        * the lowest_count candidate was not usable.
+                        *
+                        * For AMP platforms running the clutch scheduler always
+                        * return a processor from the requested pset to allow the
+                        * thread to be enqueued in the correct runq. For non-AMP
+                        * platforms, simply return the master_processor.
+                        */
+                       fallback_processor = true;
+#if CONFIG_SCHED_CLUTCH && __AMP__
+                       processor = processor_array[lsb_first(starting_pset->primary_map)];
+#else /* CONFIG_SCHED_CLUTCH && __AMP__ */
+                       processor = master_processor;
+#endif /* CONFIG_SCHED_CLUTCH && __AMP__ */
+               }
+
+               /*
+                * Check that the correct processor set is
+                * returned locked.
+                */
+               if (pset != processor->processor_set) {
+                       pset_unlock(pset);
+                       pset = processor->processor_set;
+                       pset_lock(pset);
+               }
+
+               /*
+                * We must verify that the chosen processor is still available.
+                * The cases where we pick the master_processor or the fallback
+                * processor are execptions, since we may need enqueue a thread
+                * on its runqueue if this is the last remaining processor
+                * during pset shutdown.
+                *
+                * <rdar://problem/47559304> would really help here since it
+                * gets rid of the weird last processor SHUTDOWN case where
+                * the pset is still schedulable.
+                */
+               if (processor != master_processor && (fallback_processor == false) && (processor->state == PROCESSOR_SHUTDOWN || processor->state == PROCESSOR_OFF_LINE)) {
+                       processor = PROCESSOR_NULL;
+               }
+       } while (processor == PROCESSOR_NULL);
+
+       pset->last_chosen = processor->cpu_id;
+       return processor;
+}
+
+/*
+ *     thread_setrun:
+ *
+ *     Dispatch thread for execution, onto an idle
+ *     processor or run queue, and signal a preemption
+ *     as appropriate.
+ *
+ *     Thread must be locked.
+ */
+void
+thread_setrun(
+       thread_t                        thread,
+       sched_options_t                 options)
+{
+       processor_t                     processor;
+       processor_set_t         pset;
+
+       assert((thread->state & (TH_RUN | TH_WAIT | TH_UNINT | TH_TERMINATE | TH_TERMINATE2)) == TH_RUN);
+       assert(thread->runq == PROCESSOR_NULL);
+
+       /*
+        *      Update priority if needed.
+        */
+       if (SCHED(can_update_priority)(thread)) {
+               SCHED(update_priority)(thread);
+       }
+
+       thread->sfi_class = sfi_thread_classify(thread);
+
+       assert(thread->runq == PROCESSOR_NULL);
+
+#if __SMP__
+       if (thread->bound_processor == PROCESSOR_NULL) {
+               /*
+                *      Unbound case.
+                */
+               if (thread->affinity_set != AFFINITY_SET_NULL) {
+                       /*
+                        * Use affinity set policy hint.
+                        */
+                       pset = thread->affinity_set->aset_pset;
+                       pset_lock(pset);
+
+                       processor = SCHED(choose_processor)(pset, PROCESSOR_NULL, thread);
+                       pset = processor->processor_set;
+
+                       SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_CHOOSE_PROCESSOR) | DBG_FUNC_NONE,
+                           (uintptr_t)thread_tid(thread), (uintptr_t)-1, processor->cpu_id, processor->state, 0);
+               } else if (thread->last_processor != PROCESSOR_NULL) {
+                       /*
+                        *      Simple (last processor) affinity case.
+                        */
+                       processor = thread->last_processor;
+                       pset = processor->processor_set;
+                       pset_lock(pset);
+                       processor = SCHED(choose_processor)(pset, processor, thread);
+                       pset = processor->processor_set;
+
+                       SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_CHOOSE_PROCESSOR) | DBG_FUNC_NONE,
+                           (uintptr_t)thread_tid(thread), thread->last_processor->cpu_id, processor->cpu_id, processor->state, 0);
+               } else {
+                       /*
+                        *      No Affinity case:
+                        *
+                        *      Utilitize a per task hint to spread threads
+                        *      among the available processor sets.
+                        */
+                       task_t          task = thread->task;
+
+                       pset = task->pset_hint;
+                       if (pset == PROCESSOR_SET_NULL) {
+                               pset = current_processor()->processor_set;
+                       }
+
+                       pset = choose_next_pset(pset);
+                       pset_lock(pset);
+
+                       processor = SCHED(choose_processor)(pset, PROCESSOR_NULL, thread);
+                       pset = processor->processor_set;
+                       task->pset_hint = pset;
+
+                       SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_CHOOSE_PROCESSOR) | DBG_FUNC_NONE,
+                           (uintptr_t)thread_tid(thread), (uintptr_t)-1, processor->cpu_id, processor->state, 0);
+               }
+       } else {
+               /*
+                *      Bound case:
+                *
+                *      Unconditionally dispatch on the processor.
+                */
+               processor = thread->bound_processor;
+               pset = processor->processor_set;
+               pset_lock(pset);
+
+               SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_CHOOSE_PROCESSOR) | DBG_FUNC_NONE,
+                   (uintptr_t)thread_tid(thread), (uintptr_t)-2, processor->cpu_id, processor->state, 0);
+       }
+#else /* !__SMP__ */
+       /* Only one processor to choose */
+       assert(thread->bound_processor == PROCESSOR_NULL || thread->bound_processor == master_processor);
+       processor = master_processor;
+       pset = processor->processor_set;
+       pset_lock(pset);
+#endif /* !__SMP__ */
+
+       /*
+        *      Dispatch the thread on the chosen processor.
+        *      TODO: This should be based on sched_mode, not sched_pri
+        */
+       if (thread->sched_pri >= BASEPRI_RTQUEUES) {
+               realtime_setrun(processor, thread);
+       } else {
+               processor_setrun(processor, thread, options);
+       }
+       /* pset is now unlocked */
+       if (thread->bound_processor == PROCESSOR_NULL) {
+               SCHED(check_spill)(pset, thread);
+       }
+}
+
+processor_set_t
+task_choose_pset(
+       task_t          task)
+{
+       processor_set_t         pset = task->pset_hint;
+
+       if (pset != PROCESSOR_SET_NULL) {
+               pset = choose_next_pset(pset);
+       }
+
+       return pset;
+}
+
+/*
+ *     Check for a preemption point in
+ *     the current context.
+ *
+ *     Called at splsched with thread locked.
+ */
+ast_t
+csw_check(
+       thread_t                thread,
+       processor_t             processor,
+       ast_t                   check_reason)
+{
+       processor_set_t pset = processor->processor_set;
+
+       assert(thread == processor->active_thread);
+
+       pset_lock(pset);
+
+       processor_state_update_from_thread(processor, thread);
+
+       ast_t preempt = csw_check_locked(thread, processor, pset, check_reason);
+
+       /* Acknowledge the IPI if we decided not to preempt */
+
+       if ((preempt & AST_URGENT) == 0) {
+               bit_clear(pset->pending_AST_URGENT_cpu_mask, processor->cpu_id);
+       }
+
+       if ((preempt & AST_PREEMPT) == 0) {
+               bit_clear(pset->pending_AST_PREEMPT_cpu_mask, processor->cpu_id);
+       }
+
+       pset_unlock(pset);
+
+       return preempt;
+}
+
+/*
+ * Check for preemption at splsched with
+ * pset and thread locked
+ */
+ast_t
+csw_check_locked(
+       thread_t                thread,
+       processor_t             processor,
+       processor_set_t         pset,
+       ast_t                   check_reason)
+{
+       ast_t                   result;
+
+       if (processor->first_timeslice) {
+               if (rt_runq_count(pset) > 0) {
+                       return check_reason | AST_PREEMPT | AST_URGENT;
+               }
+       } else {
+               if (rt_runq_count(pset) > 0) {
+                       if (BASEPRI_RTQUEUES > processor->current_pri) {
+                               return check_reason | AST_PREEMPT | AST_URGENT;
+                       } else {
+                               return check_reason | AST_PREEMPT;
+                       }
+               }
+       }
+
+#if __SMP__
+       /*
+        * If the current thread is running on a processor that is no longer recommended,
+        * urgently preempt it, at which point thread_select() should
+        * try to idle the processor and re-dispatch the thread to a recommended processor.
+        */
+       if (!processor->is_recommended) {
+               return check_reason | AST_PREEMPT | AST_URGENT;
+       }
+#endif
+
+       result = SCHED(processor_csw_check)(processor);
+       if (result != AST_NONE) {
+               return check_reason | result | (thread_eager_preemption(thread) ? AST_URGENT : AST_NONE);
+       }
+
+#if __SMP__
+       /*
+        * Same for avoid-processor
+        *
+        * TODO: Should these set AST_REBALANCE?
+        */
+       if (SCHED(avoid_processor_enabled) && SCHED(thread_avoid_processor)(processor, thread)) {
+               return check_reason | AST_PREEMPT;
+       }
+
+       /*
+        * Even though we could continue executing on this processor, a
+        * secondary SMT core should try to shed load to another primary core.
+        *
+        * TODO: Should this do the same check that thread_select does? i.e.
+        * if no bound threads target this processor, and idle primaries exist, preempt
+        * The case of RT threads existing is already taken care of above
+        */
+
+       if (processor->current_pri < BASEPRI_RTQUEUES &&
+           processor->processor_primary != processor) {
+               return check_reason | AST_PREEMPT;
+       }
+#endif
+
+       if (thread->state & TH_SUSP) {
+               return check_reason | AST_PREEMPT;
+       }
+
+#if CONFIG_SCHED_SFI
+       /*
+        * Current thread may not need to be preempted, but maybe needs
+        * an SFI wait?
+        */
+       result = sfi_thread_needs_ast(thread, NULL);
+       if (result != AST_NONE) {
+               return check_reason | result;
+       }
+#endif
+
+       return AST_NONE;
+}
+
+/*
+ * Handle preemption IPI or IPI in response to setting an AST flag
+ * Triggered by cause_ast_check
+ * Called at splsched
+ */
+void
+ast_check(processor_t processor)
+{
+       if (processor->state != PROCESSOR_RUNNING &&
+           processor->state != PROCESSOR_SHUTDOWN) {
+               return;
+       }
+
+       thread_t thread = processor->active_thread;
+
+       assert(thread == current_thread());
+
+       thread_lock(thread);
+
+       /*
+        * Propagate thread ast to processor.
+        * (handles IPI in response to setting AST flag)
+        */
+       ast_propagate(thread);
+
+       /*
+        * Stash the old urgency and perfctl values to find out if
+        * csw_check updates them.
+        */
+       thread_urgency_t old_urgency = processor->current_urgency;
+       perfcontrol_class_t old_perfctl_class = processor->current_perfctl_class;
+
+       ast_t preempt;
+
+       if ((preempt = csw_check(thread, processor, AST_NONE)) != AST_NONE) {
+               ast_on(preempt);
+       }
+
+       if (old_urgency != processor->current_urgency) {
+               /*
+                * Urgency updates happen with the thread lock held (ugh).
+                * TODO: This doesn't notice QoS changes...
+                */
+               uint64_t urgency_param1, urgency_param2;
+
+               thread_urgency_t urgency = thread_get_urgency(thread, &urgency_param1, &urgency_param2);
+               thread_tell_urgency(urgency, urgency_param1, urgency_param2, 0, thread);
+       }
+
+       thread_unlock(thread);
+
+       if (old_perfctl_class != processor->current_perfctl_class) {
+               /*
+                * We updated the perfctl class of this thread from another core.
+                * Let CLPC know that the currently running thread has a new
+                * class.
+                */
+
+               machine_switch_perfcontrol_state_update(PERFCONTROL_ATTR_UPDATE,
+                   mach_approximate_time(), 0, thread);
+       }
+}
+
+
+/*
+ *     set_sched_pri:
+ *
+ *     Set the scheduled priority of the specified thread.
+ *
+ *     This may cause the thread to change queues.
+ *
+ *     Thread must be locked.
+ */
+void
+set_sched_pri(
+       thread_t        thread,
+       int             new_priority,
+       set_sched_pri_options_t options)
+{
+       bool is_current_thread = (thread == current_thread());
+       bool removed_from_runq = false;
+       bool lazy_update = ((options & SETPRI_LAZY) == SETPRI_LAZY);
+
+       int old_priority = thread->sched_pri;
+
+       /* If we're already at this priority, no need to mess with the runqueue */
+       if (new_priority == old_priority) {
+#if CONFIG_SCHED_CLUTCH
+               /* For the first thread in the system, the priority is correct but
+                * th_sched_bucket is still TH_BUCKET_RUN. Since the clutch
+                * scheduler relies on the bucket being set for all threads, update
+                * its bucket here.
+                */
+               if (thread->th_sched_bucket == TH_BUCKET_RUN) {
+                       assert(is_current_thread);
+                       SCHED(update_thread_bucket)(thread);
+               }
+#endif /* CONFIG_SCHED_CLUTCH */
+
+               return;
+       }
+
+       if (is_current_thread) {
+               assert(thread->state & TH_RUN);
+               assert(thread->runq == PROCESSOR_NULL);
+       } else {
+               removed_from_runq = thread_run_queue_remove(thread);
+       }
+
+       thread->sched_pri = new_priority;
+
+#if CONFIG_SCHED_CLUTCH
+       /*
+        * Since for the clutch scheduler, the thread's bucket determines its runq
+        * in the hierarchy it is important to update the bucket when the thread
+        * lock is held and the thread has been removed from the runq hierarchy.
+        */
+       SCHED(update_thread_bucket)(thread);
+
+#endif /* CONFIG_SCHED_CLUTCH */
+
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_CHANGE_PRIORITY),
+           (uintptr_t)thread_tid(thread),
+           thread->base_pri,
+           thread->sched_pri,
+           thread->sched_usage,
+           0);
+
+       if (removed_from_runq) {
+               thread_run_queue_reinsert(thread, SCHED_PREEMPT | SCHED_TAILQ);
+       } else if (is_current_thread) {
+               processor_t processor = thread->last_processor;
+               assert(processor == current_processor());
+
+               thread_urgency_t old_urgency = processor->current_urgency;
+
+               /*
+                * When dropping in priority, check if the thread no longer belongs on core.
+                * If a thread raises its own priority, don't aggressively rebalance it.
+                * <rdar://problem/31699165>
+                *
+                * csw_check does a processor_state_update_from_thread, but
+                * we should do our own if we're being lazy.
+                */
+               if (!lazy_update && new_priority < old_priority) {
+                       ast_t preempt;
+
+                       if ((preempt = csw_check(thread, processor, AST_NONE)) != AST_NONE) {
+                               ast_on(preempt);
+                       }
+               } else {
+                       processor_state_update_from_thread(processor, thread);
+               }
 
 
-       counter_always(c_thread_invoke_csw++);
-       current_task()->csw++;
+               /*
+                * set_sched_pri doesn't alter RT params. We expect direct base priority/QoS
+                * class alterations from user space to occur relatively infrequently, hence
+                * those are lazily handled. QoS classes have distinct priority bands, and QoS
+                * inheritance is expected to involve priority changes.
+                */
+               if (processor->current_urgency != old_urgency) {
+                       uint64_t urgency_param1, urgency_param2;
 
 
-       assert(old_thread->runq == RUN_QUEUE_NULL);
-       old_thread->reason = reason;
+                       thread_urgency_t new_urgency = thread_get_urgency(thread,
+                           &urgency_param1, &urgency_param2);
 
 
-       processor->last_dispatch = mach_absolute_time();
-       timer_event((uint32_t)processor->last_dispatch, &new_thread->system_timer);
+                       thread_tell_urgency(new_urgency, urgency_param1,
+                           urgency_param2, 0, thread);
+               }
 
 
-       thread_done(old_thread, new_thread, processor);
+               /* TODO: only call this if current_perfctl_class changed */
+               uint64_t ctime = mach_approximate_time();
+               machine_thread_going_on_core(thread, processor->current_urgency, 0, 0, ctime);
+       } else if (thread->state & TH_RUN) {
+               processor_t processor = thread->last_processor;
 
 
-       /*
-        * This is where we actually switch register context,
-        * and address space if required.  Control will not
-        * return here immediately.
-        */
-       prev_thread = machine_switch_context(old_thread, continuation, new_thread);
+               if (!lazy_update &&
+                   processor != PROCESSOR_NULL &&
+                   processor != current_processor() &&
+                   processor->active_thread == thread) {
+                       cause_ast_check(processor);
+               }
+       }
+}
 
 
-       /*
-        * We are still old_thread, possibly on a different processor,
-        * and new_thread is now stale.
-        */
-       thread_begin(old_thread, old_thread->last_processor);
+/*
+ * thread_run_queue_remove_for_handoff
+ *
+ * Pull a thread or its (recursive) push target out of the runqueue
+ * so that it is ready for thread_run()
+ *
+ * Called at splsched
+ *
+ * Returns the thread that was pulled or THREAD_NULL if no thread could be pulled.
+ * This may be different than the thread that was passed in.
+ */
+thread_t
+thread_run_queue_remove_for_handoff(thread_t thread)
+{
+       thread_t pulled_thread = THREAD_NULL;
+
+       thread_lock(thread);
 
        /*
 
        /*
-        * Now dispatch the thread which resumed us.
+        * Check that the thread is not bound
+        * to a different processor, and that realtime
+        * is not involved.
+        *
+        * Next, pull it off its run queue.  If it
+        * doesn't come, it's not eligible.
         */
         */
-       thread_dispatch(prev_thread);
-
-       if (continuation) {
-               funnel_refunnel_check(old_thread, 3);
-               (void) spllo();
 
 
-               call_continuation(continuation, parameter, old_thread->wait_result);
-               /*NOTREACHED*/
+       processor_t processor = current_processor();
+       if (processor->current_pri < BASEPRI_RTQUEUES && thread->sched_pri < BASEPRI_RTQUEUES &&
+           (thread->bound_processor == PROCESSOR_NULL || thread->bound_processor == processor)) {
+               if (thread_run_queue_remove(thread)) {
+                       pulled_thread = thread;
+               }
        }
 
        }
 
-       return (TRUE);
+       thread_unlock(thread);
+
+       return pulled_thread;
 }
 
 /*
 }
 
 /*
- *     thread_done:
+ *     thread_run_queue_remove:
  *
  *
- *     Perform calculations for thread
- *     finishing execution on the current processor.
+ *     Remove a thread from its current run queue and
+ *     return TRUE if successful.
  *
  *
- *     Called at splsched.
+ *     Thread must be locked.
+ *
+ *     If thread->runq is PROCESSOR_NULL, the thread will not re-enter the
+ *     run queues because the caller locked the thread.  Otherwise
+ *     the thread is on a run queue, but could be chosen for dispatch
+ *     and removed by another processor under a different lock, which
+ *     will set thread->runq to PROCESSOR_NULL.
+ *
+ *     Hence the thread select path must not rely on anything that could
+ *     be changed under the thread lock after calling this function,
+ *     most importantly thread->sched_pri.
  */
  */
-void
-thread_done(
-       thread_t                        old_thread,
-       thread_t                        new_thread,
-       processor_t                     processor)
+boolean_t
+thread_run_queue_remove(
+       thread_t        thread)
 {
 {
-       if (!(old_thread->state & TH_IDLE)) {
+       boolean_t removed = FALSE;
+       processor_t processor = thread->runq;
+
+       if ((thread->state & (TH_RUN | TH_WAIT)) == TH_WAIT) {
+               /* Thread isn't runnable */
+               assert(thread->runq == PROCESSOR_NULL);
+               return FALSE;
+       }
+
+       if (processor == PROCESSOR_NULL) {
                /*
                /*
-                * Compute remainder of current quantum.
+                * The thread is either not on the runq,
+                * or is in the midst of being removed from the runq.
+                *
+                * runq is set to NULL under the pset lock, not the thread
+                * lock, so the thread may still be in the process of being dequeued
+                * from the runq. It will wait in invoke for the thread lock to be
+                * dropped.
                 */
                 */
-               if (    first_timeslice(processor)                                                      &&
-                               processor->quantum_end > processor->last_dispatch               )
-                       old_thread->current_quantum =
-                                       (processor->quantum_end - processor->last_dispatch);
-               else
-                       old_thread->current_quantum = 0;
-
-               if (old_thread->sched_mode & TH_MODE_REALTIME) {
-                       /*
-                        * Cancel the deadline if the thread has
-                        * consumed the entire quantum.
-                        */
-                       if (old_thread->current_quantum == 0) {
-                               old_thread->realtime.deadline = UINT64_MAX;
-                               old_thread->reason |= AST_QUANTUM;
-                       }
-               }
-               else {
-                       /*
-                        * For non-realtime threads treat a tiny
-                        * remaining quantum as an expired quantum
-                        * but include what's left next time.
-                        */
-                       if (old_thread->current_quantum < min_std_quantum) {
-                               old_thread->reason |= AST_QUANTUM;
-                               old_thread->current_quantum += std_quantum;
-                       }
-               }
 
 
+               return FALSE;
+       }
+
+       if (thread->sched_pri < BASEPRI_RTQUEUES) {
+               return SCHED(processor_queue_remove)(processor, thread);
+       }
+
+       processor_set_t pset = processor->processor_set;
+
+       rt_lock_lock(pset);
+
+       if (thread->runq != PROCESSOR_NULL) {
                /*
                /*
-                * If we are doing a direct handoff then
-                * give the remainder of our quantum to
-                * the next thread.
+                *      Thread is on the RT run queue and we have a lock on
+                *      that run queue.
                 */
                 */
-               if ((old_thread->reason & (AST_HANDOFF|AST_QUANTUM)) == AST_HANDOFF) {
-                       new_thread->current_quantum = old_thread->current_quantum;
-                       old_thread->reason |= AST_QUANTUM;
-                       old_thread->current_quantum = 0;
-               }
 
 
-               old_thread->last_switch = processor->last_dispatch;
+               remqueue(&thread->runq_links);
+               SCHED_STATS_RUNQ_CHANGE(&SCHED(rt_runq)(pset)->runq_stats, rt_runq_count(pset));
+               rt_runq_count_decr(pset);
+
+               thread->runq = PROCESSOR_NULL;
 
 
-               old_thread->computation_metered +=
-                               (old_thread->last_switch - old_thread->computation_epoch);
+               removed = TRUE;
        }
        }
+
+       rt_lock_unlock(pset);
+
+       return removed;
 }
 
 /*
 }
 
 /*
- *     thread_begin:
+ * Put the thread back where it goes after a thread_run_queue_remove
  *
  *
- *     Set up for thread beginning execution on
- *     the current processor.
+ * Thread must have been removed under the same thread lock hold
  *
  *
- *     Called at splsched.
+ * thread locked, at splsched
  */
 void
  */
 void
-thread_begin(
-       thread_t                        thread,
-       processor_t                     processor)
+thread_run_queue_reinsert(thread_t thread, sched_options_t options)
+{
+       assert(thread->runq == PROCESSOR_NULL);
+       assert(thread->state & (TH_RUN));
+
+       thread_setrun(thread, options);
+}
+
+void
+sys_override_cpu_throttle(boolean_t enable_override)
+{
+       if (enable_override) {
+               cpu_throttle_enabled = 0;
+       } else {
+               cpu_throttle_enabled = 1;
+       }
+}
+
+thread_urgency_t
+thread_get_urgency(thread_t thread, uint64_t *arg1, uint64_t *arg2)
 {
 {
-       if (!(thread->state & TH_IDLE)) {
+       uint64_t urgency_param1 = 0, urgency_param2 = 0;
+
+       thread_urgency_t urgency;
+
+       if (thread == NULL || (thread->state & TH_IDLE)) {
+               urgency_param1 = 0;
+               urgency_param2 = 0;
+
+               urgency = THREAD_URGENCY_NONE;
+       } else if (thread->sched_mode == TH_MODE_REALTIME) {
+               urgency_param1 = thread->realtime.period;
+               urgency_param2 = thread->realtime.deadline;
+
+               urgency = THREAD_URGENCY_REAL_TIME;
+       } else if (cpu_throttle_enabled &&
+           (thread->sched_pri <= MAXPRI_THROTTLE) &&
+           (thread->base_pri <= MAXPRI_THROTTLE)) {
                /*
                /*
-                * Give the thread a new quantum
-                * if none remaining.
+                * Threads that are running at low priority but are not
+                * tagged with a specific QoS are separated out from
+                * the "background" urgency. Performance management
+                * subsystem can decide to either treat these threads
+                * as normal threads or look at other signals like thermal
+                * levels for optimal power/perf tradeoffs for a platform.
                 */
                 */
-               if (thread->current_quantum == 0)
-                       thread_quantum_init(thread);
+               boolean_t thread_lacks_qos = (proc_get_effective_thread_policy(thread, TASK_POLICY_QOS) == THREAD_QOS_UNSPECIFIED); //thread_has_qos_policy(thread);
+               boolean_t task_is_suppressed = (proc_get_effective_task_policy(thread->task, TASK_POLICY_SUP_ACTIVE) == 0x1);
 
                /*
 
                /*
-                * Set up quantum timer and timeslice.
+                * Background urgency applied when thread priority is
+                * MAXPRI_THROTTLE or lower and thread is not promoted
+                * and thread has a QoS specified
                 */
                 */
-               processor->quantum_end =
-                               (processor->last_dispatch + thread->current_quantum);
-               timer_call_enter1(&processor->quantum_timer,
-                                                               thread, processor->quantum_end);
+               urgency_param1 = thread->sched_pri;
+               urgency_param2 = thread->base_pri;
+
+               if (thread_lacks_qos && !task_is_suppressed) {
+                       urgency = THREAD_URGENCY_LOWPRI;
+               } else {
+                       urgency = THREAD_URGENCY_BACKGROUND;
+               }
+       } else {
+               /* For otherwise unclassified threads, report throughput QoS parameters */
+               urgency_param1 = proc_get_effective_thread_policy(thread, TASK_POLICY_THROUGH_QOS);
+               urgency_param2 = proc_get_effective_task_policy(thread->task, TASK_POLICY_THROUGH_QOS);
+               urgency = THREAD_URGENCY_NORMAL;
+       }
 
 
-               processor_timeslice_setup(processor, thread);
+       if (arg1 != NULL) {
+               *arg1 = urgency_param1;
+       }
+       if (arg2 != NULL) {
+               *arg2 = urgency_param2;
+       }
 
 
-               thread->last_switch = processor->last_dispatch;
+       return urgency;
+}
 
 
-               thread->computation_epoch = thread->last_switch;
+perfcontrol_class_t
+thread_get_perfcontrol_class(thread_t thread)
+{
+       /* Special case handling */
+       if (thread->state & TH_IDLE) {
+               return PERFCONTROL_CLASS_IDLE;
+       }
+       if (thread->task == kernel_task) {
+               return PERFCONTROL_CLASS_KERNEL;
        }
        }
-       else {
-               timer_call_cancel(&processor->quantum_timer);
-               processor->timeslice = 1;
+       if (thread->sched_mode == TH_MODE_REALTIME) {
+               return PERFCONTROL_CLASS_REALTIME;
+       }
+
+       /* perfcontrol_class based on base_pri */
+       if (thread->base_pri <= MAXPRI_THROTTLE) {
+               return PERFCONTROL_CLASS_BACKGROUND;
+       } else if (thread->base_pri <= BASEPRI_UTILITY) {
+               return PERFCONTROL_CLASS_UTILITY;
+       } else if (thread->base_pri <= BASEPRI_DEFAULT) {
+               return PERFCONTROL_CLASS_NONUI;
+       } else if (thread->base_pri <= BASEPRI_FOREGROUND) {
+               return PERFCONTROL_CLASS_UI;
+       } else {
+               return PERFCONTROL_CLASS_ABOVEUI;
        }
 }
 
 /*
        }
 }
 
 /*
- *     thread_dispatch:
- *
- *     Handle previous thread at context switch.  Re-dispatch
- *     if still running, otherwise update run state and perform
- *     special actions.
+ *     This is the processor idle loop, which just looks for other threads
+ *     to execute.  Processor idle threads invoke this without supplying a
+ *     current thread to idle without an asserted wait state.
  *
  *
- *     Called at splsched.
+ *     Returns a the next thread to execute if dispatched directly.
  */
  */
-void
-thread_dispatch(
-       register thread_t       thread)
+
+#if 0
+#define IDLE_KERNEL_DEBUG_CONSTANT(...) KERNEL_DEBUG_CONSTANT(__VA_ARGS__)
+#else
+#define IDLE_KERNEL_DEBUG_CONSTANT(...) do { } while(0)
+#endif
+
+thread_t
+processor_idle(
+       thread_t                        thread,
+       processor_t                     processor)
 {
 {
-       /*
-        *      If blocked at a continuation, discard
-        *      the stack.
-        */
-    if (thread->continuation != NULL && thread->kernel_stack)
-               stack_free(thread);
+       processor_set_t         pset = processor->processor_set;
 
 
-       if (!(thread->state & TH_IDLE)) {
-               wake_lock(thread);
-               thread_lock(thread);
+       (void)splsched();
 
 
-               if (!(thread->state & TH_WAIT)) {
-                       /*
-                        *      Still running.
-                        */
-                       if (thread->reason & AST_QUANTUM)
-                               thread_setrun(thread, SCHED_TAILQ);
-                       else
-                       if (thread->reason & AST_PREEMPT)
-                               thread_setrun(thread, SCHED_HEADQ);
-                       else
-                               thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_IDLE) | DBG_FUNC_START,
+           (uintptr_t)thread_tid(thread), 0, 0, 0, 0);
 
 
-                       thread->reason = AST_NONE;
+       SCHED_STATS_CPU_IDLE_START(processor);
 
 
-                       thread_unlock(thread);
-                       wake_unlock(thread);
+       uint64_t ctime = mach_absolute_time();
+
+       timer_switch(&PROCESSOR_DATA(processor, system_state), ctime, &PROCESSOR_DATA(processor, idle_state));
+       PROCESSOR_DATA(processor, current_state) = &PROCESSOR_DATA(processor, idle_state);
+
+       cpu_quiescent_counter_leave(ctime);
+
+       while (1) {
+               /*
+                * Ensure that updates to my processor and pset state,
+                * made by the IPI source processor before sending the IPI,
+                * are visible on this processor now (even though we don't
+                * take the pset lock yet).
+                */
+               atomic_thread_fence(memory_order_acquire);
+
+               if (processor->state != PROCESSOR_IDLE) {
+                       break;
+               }
+               if (bit_test(pset->pending_AST_URGENT_cpu_mask, processor->cpu_id)) {
+                       break;
+               }
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+               if (bit_test(pset->pending_deferred_AST_cpu_mask, processor->cpu_id)) {
+                       break;
+               }
+#endif
+               if (processor->is_recommended && (processor->processor_primary == processor)) {
+                       if (rt_runq_count(pset)) {
+                               break;
+                       }
+               } else {
+                       if (SCHED(processor_bound_count)(processor)) {
+                               break;
+                       }
                }
                }
-               else {
-                       boolean_t               wake;
 
 
-                       /*
-                        *      Waiting.
-                        */
-                       thread->state &= ~TH_RUN;
+               IDLE_KERNEL_DEBUG_CONSTANT(
+                       MACHDBG_CODE(DBG_MACH_SCHED, MACH_IDLE) | DBG_FUNC_NONE, (uintptr_t)thread_tid(thread), rt_runq_count(pset), SCHED(processor_runq_count)(processor), -1, 0);
 
 
-                       wake = thread->wake_active;
-                       thread->wake_active = FALSE;
+               machine_track_platform_idle(TRUE);
 
 
-                       if (thread->sched_mode & TH_MODE_TIMESHARE)
-                               pset_share_decr(thread->processor_set);
-                       pset_run_decr(thread->processor_set);
+               machine_idle();
+               /* returns with interrupts enabled */
 
 
-                       thread_unlock(thread);
-                       wake_unlock(thread);
+               machine_track_platform_idle(FALSE);
+
+               (void)splsched();
 
 
-                       if (thread->options & TH_OPT_CALLOUT)
-                               call_thread_block();
+               /*
+                * Check if we should call sched_timeshare_consider_maintenance() here.
+                * The CPU was woken out of idle due to an interrupt and we should do the
+                * call only if the processor is still idle. If the processor is non-idle,
+                * the threads running on the processor would do the call as part of
+                * context swithing.
+                */
+               if (processor->state == PROCESSOR_IDLE) {
+                       sched_timeshare_consider_maintenance(mach_absolute_time());
+               }
 
 
-                       if (wake)
-                               thread_wakeup((event_t)&thread->wake_active);
+               IDLE_KERNEL_DEBUG_CONSTANT(
+                       MACHDBG_CODE(DBG_MACH_SCHED, MACH_IDLE) | DBG_FUNC_NONE, (uintptr_t)thread_tid(thread), rt_runq_count(pset), SCHED(processor_runq_count)(processor), -2, 0);
 
 
-                       if (thread->state & TH_TERMINATE)
-                               thread_terminate_enqueue(thread);
+               if (!SCHED(processor_queue_empty)(processor)) {
+                       /* Secondary SMT processors respond to directed wakeups
+                        * exclusively. Some platforms induce 'spurious' SMT wakeups.
+                        */
+                       if (processor->processor_primary == processor) {
+                               break;
+                       }
                }
        }
                }
        }
+
+       ctime = mach_absolute_time();
+
+       timer_switch(&PROCESSOR_DATA(processor, idle_state), ctime, &PROCESSOR_DATA(processor, system_state));
+       PROCESSOR_DATA(processor, current_state) = &PROCESSOR_DATA(processor, system_state);
+
+       cpu_quiescent_counter_join(ctime);
+
+       ast_t reason = AST_NONE;
+
+       /* We're handling all scheduling AST's */
+       ast_off(AST_SCHEDULING);
+
+       /*
+        * thread_select will move the processor from dispatching to running,
+        * or put it in idle if there's nothing to do.
+        */
+       thread_t current_thread = current_thread();
+
+       thread_lock(current_thread);
+       thread_t new_thread = thread_select(current_thread, processor, &reason);
+       thread_unlock(current_thread);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_IDLE) | DBG_FUNC_END,
+           (uintptr_t)thread_tid(thread), processor->state, (uintptr_t)thread_tid(new_thread), reason, 0);
+
+       return new_thread;
 }
 
 /*
 }
 
 /*
- *     thread_block_reason:
+ *     Each processor has a dedicated thread which
+ *     executes the idle loop when there is no suitable
+ *     previous context.
  *
  *
- *     Forces a reschedule, blocking the caller if a wait
- *     has been asserted.
- *
- *     If a continuation is specified, then thread_invoke will
- *     attempt to discard the thread's kernel stack.  When the
- *     thread resumes, it will execute the continuation function
- *     on a new kernel stack.
+ *     This continuation is entered with interrupts disabled.
  */
  */
-counter(mach_counter_t  c_thread_block_calls = 0;)
-wait_result_t
-thread_block_reason(
-       thread_continue_t       continuation,
-       void                            *parameter,
-       ast_t                           reason)
+void
+idle_thread(__assert_only void* parameter,
+    __unused wait_result_t result)
 {
 {
-       register thread_t               self = current_thread();
-       register processor_t    processor;
-       register thread_t               new_thread;
-       spl_t                                   s;
+       assert(ml_get_interrupts_enabled() == FALSE);
+       assert(parameter == NULL);
 
 
-       counter(++c_thread_block_calls);
+       processor_t processor = current_processor();
 
 
-       s = splsched();
+       /*
+        * Ensure that anything running in idle context triggers
+        * preemption-disabled checks.
+        */
+       disable_preemption();
 
 
-#if 0
-#if    MACH_KDB
-       {
-               extern void db_chkpmgr(void);
-               db_chkpmgr();                                           /* (BRINGUP) See if pm config changed */
+       /*
+        * Enable interrupts temporarily to handle any pending interrupts
+        * or IPIs before deciding to sleep
+        */
+       spllo();
 
 
-       }
-#endif
-#endif
+       thread_t new_thread = processor_idle(THREAD_NULL, processor);
+       /* returns with interrupts disabled */
 
 
-       if (!(reason & AST_PREEMPT))
-               funnel_release_check(self, 2);
+       enable_preemption();
 
 
-       processor = current_processor();
+       if (new_thread != THREAD_NULL) {
+               thread_run(processor->idle_thread,
+                   idle_thread, NULL, new_thread);
+               /*NOTREACHED*/
+       }
 
 
-       /*
-        * Delay switching to the idle thread under certain conditions.
-        */
-       if (s != FALSE && (self->state & (TH_IDLE|TH_TERMINATE|TH_WAIT)) == TH_WAIT) {
-               if (    processor->processor_set->processor_count > 1   &&
-                               processor->processor_set->runq.count == 0               &&
-                               processor->runq.count == 0                                              )
-                       processor = delay_idle(processor, self);
+       thread_block(idle_thread);
+       /*NOTREACHED*/
+}
+
+kern_return_t
+idle_thread_create(
+       processor_t             processor)
+{
+       kern_return_t   result;
+       thread_t                thread;
+       spl_t                   s;
+       char                    name[MAXTHREADNAMESIZE];
+
+       result = kernel_thread_create(idle_thread, NULL, MAXPRI_KERNEL, &thread);
+       if (result != KERN_SUCCESS) {
+               return result;
        }
 
        }
 
-       /* If we're explicitly yielding, force a subsequent quantum */
-       if (reason & AST_YIELD)
-               processor->timeslice = 0;
+       snprintf(name, sizeof(name), "idle #%d", processor->cpu_id);
+       thread_set_thread_name(thread, name);
 
 
-       /* We're handling all scheduling AST's */
-       ast_off(AST_SCHEDULING);
+       s = splsched();
+       thread_lock(thread);
+       thread->bound_processor = processor;
+       processor->idle_thread = thread;
+       thread->sched_pri = thread->base_pri = IDLEPRI;
+       thread->state = (TH_RUN | TH_IDLE);
+       thread->options |= TH_OPT_IDLE_THREAD;
+       thread_unlock(thread);
+       splx(s);
 
 
-       self->continuation = continuation;
-       self->parameter = parameter;
+       thread_deallocate(thread);
 
 
-       thread_lock(self);
-       new_thread = thread_select(processor);
-       assert(new_thread && thread_runnable(new_thread));
-       thread_unlock(self);
-       while (!thread_invoke(self, new_thread, reason)) {
-               thread_lock(self);
-               new_thread = thread_select(processor);
-               assert(new_thread && thread_runnable(new_thread));
-               thread_unlock(self);
+       return KERN_SUCCESS;
+}
+
+/*
+ * sched_startup:
+ *
+ * Kicks off scheduler services.
+ *
+ * Called at splsched.
+ */
+void
+sched_startup(void)
+{
+       kern_return_t   result;
+       thread_t                thread;
+
+       simple_lock_init(&sched_vm_group_list_lock, 0);
+
+#if __arm__ || __arm64__
+       simple_lock_init(&sched_recommended_cores_lock, 0);
+#endif /* __arm__ || __arm64__ */
+
+       result = kernel_thread_start_priority((thread_continue_t)sched_init_thread,
+           (void *)SCHED(maintenance_continuation), MAXPRI_KERNEL, &thread);
+       if (result != KERN_SUCCESS) {
+               panic("sched_startup");
        }
 
        }
 
-       funnel_refunnel_check(self, 5);
-       splx(s);
+       thread_deallocate(thread);
+
+       assert_thread_magic(thread);
 
 
-       return (self->wait_result);
+       /*
+        * Yield to the sched_init_thread once, to
+        * initialize our own thread after being switched
+        * back to.
+        *
+        * The current thread is the only other thread
+        * active at this point.
+        */
+       thread_block(THREAD_CONTINUE_NULL);
 }
 
 }
 
+#if __arm64__
+static _Atomic uint64_t sched_perfcontrol_callback_deadline;
+#endif /* __arm64__ */
+
+
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+
+static volatile uint64_t                sched_maintenance_deadline;
+static uint64_t                         sched_tick_last_abstime;
+static uint64_t                         sched_tick_delta;
+uint64_t                                sched_tick_max_delta;
+
+
 /*
 /*
- *     thread_block:
+ *     sched_init_thread:
  *
  *
- *     Block the current thread if a wait has been asserted.
+ *     Perform periodic bookkeeping functions about ten
+ *     times per second.
  */
  */
-wait_result_t
-thread_block(
-       thread_continue_t       continuation)
+void
+sched_timeshare_maintenance_continue(void)
 {
 {
-       return thread_block_reason(continuation, NULL, AST_NONE);
-}
+       uint64_t        sched_tick_ctime, late_time;
+
+       struct sched_update_scan_context scan_context = {
+               .earliest_bg_make_runnable_time = UINT64_MAX,
+               .earliest_normal_make_runnable_time = UINT64_MAX,
+               .earliest_rt_make_runnable_time = UINT64_MAX
+       };
+
+       sched_tick_ctime = mach_absolute_time();
+
+       if (__improbable(sched_tick_last_abstime == 0)) {
+               sched_tick_last_abstime = sched_tick_ctime;
+               late_time = 0;
+               sched_tick_delta = 1;
+       } else {
+               late_time = sched_tick_ctime - sched_tick_last_abstime;
+               sched_tick_delta = late_time / sched_tick_interval;
+               /* Ensure a delta of 1, since the interval could be slightly
+                * smaller than the sched_tick_interval due to dispatch
+                * latencies.
+                */
+               sched_tick_delta = MAX(sched_tick_delta, 1);
+
+               /* In the event interrupt latencies or platform
+                * idle events that advanced the timebase resulted
+                * in periods where no threads were dispatched,
+                * cap the maximum "tick delta" at SCHED_TICK_MAX_DELTA
+                * iterations.
+                */
+               sched_tick_delta = MIN(sched_tick_delta, SCHED_TICK_MAX_DELTA);
+
+               sched_tick_last_abstime = sched_tick_ctime;
+               sched_tick_max_delta = MAX(sched_tick_delta, sched_tick_max_delta);
+       }
+
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_MAINTENANCE) | DBG_FUNC_START,
+           sched_tick_delta, late_time, 0, 0, 0);
+
+       /* Add a number of pseudo-ticks corresponding to the elapsed interval
+        * This could be greater than 1 if substantial intervals where
+        * all processors are idle occur, which rarely occurs in practice.
+        */
+
+       sched_tick += sched_tick_delta;
+
+       update_vm_info();
+
+       /*
+        *  Compute various averages.
+        */
+       compute_averages(sched_tick_delta);
+
+       /*
+        *  Scan the run queues for threads which
+        *  may need to be updated, and find the earliest runnable thread on the runqueue
+        *  to report its latency.
+        */
+       SCHED(thread_update_scan)(&scan_context);
+
+       SCHED(rt_runq_scan)(&scan_context);
+
+       uint64_t ctime = mach_absolute_time();
+
+       uint64_t bg_max_latency       = (ctime > scan_context.earliest_bg_make_runnable_time) ?
+           ctime - scan_context.earliest_bg_make_runnable_time : 0;
+
+       uint64_t default_max_latency  = (ctime > scan_context.earliest_normal_make_runnable_time) ?
+           ctime - scan_context.earliest_normal_make_runnable_time : 0;
+
+       uint64_t realtime_max_latency = (ctime > scan_context.earliest_rt_make_runnable_time) ?
+           ctime - scan_context.earliest_rt_make_runnable_time : 0;
+
+       machine_max_runnable_latency(bg_max_latency, default_max_latency, realtime_max_latency);
+
+       /*
+        * Check to see if the special sched VM group needs attention.
+        */
+       sched_vm_group_maintenance();
+
+#if __arm__ || __arm64__
+       /* Check to see if the recommended cores failsafe is active */
+       sched_recommended_cores_maintenance();
+#endif /* __arm__ || __arm64__ */
+
 
 
-wait_result_t
-thread_block_parameter(
-       thread_continue_t       continuation,
-       void                            *parameter)
-{
-       return thread_block_reason(continuation, parameter, AST_NONE);
+#if DEBUG || DEVELOPMENT
+#if __x86_64__
+#include <i386/misc_protos.h>
+       /* Check for long-duration interrupts */
+       mp_interrupt_watchdog();
+#endif /* __x86_64__ */
+#endif /* DEBUG || DEVELOPMENT */
+
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_MAINTENANCE) | DBG_FUNC_END,
+           sched_pri_shifts[TH_BUCKET_SHARE_FG], sched_pri_shifts[TH_BUCKET_SHARE_BG],
+           sched_pri_shifts[TH_BUCKET_SHARE_UT], sched_pri_shifts[TH_BUCKET_SHARE_DF], 0);
+
+       assert_wait((event_t)sched_timeshare_maintenance_continue, THREAD_UNINT);
+       thread_block((thread_continue_t)sched_timeshare_maintenance_continue);
+       /*NOTREACHED*/
 }
 
 }
 
+static uint64_t sched_maintenance_wakeups;
+
 /*
 /*
- *     thread_run:
- *
- *     Switch directly from the current thread to the
- *     new thread, handing off our quantum if appropriate.
- *
- *     New thread must be runnable, and not on a run queue.
- *
- *     Called at splsched.
+ * Determine if the set of routines formerly driven by a maintenance timer
+ * must be invoked, based on a deadline comparison. Signals the scheduler
+ * maintenance thread on deadline expiration. Must be invoked at an interval
+ * lower than the "sched_tick_interval", currently accomplished by
+ * invocation via the quantum expiration timer and at context switch time.
+ * Performance matters: this routine reuses a timestamp approximating the
+ * current absolute time received from the caller, and should perform
+ * no more than a comparison against the deadline in the common case.
  */
  */
-int
-thread_run(
-       thread_t                        self,
-       thread_continue_t       continuation,
-       void                            *parameter,
-       thread_t                        new_thread)
+void
+sched_timeshare_consider_maintenance(uint64_t ctime)
 {
 {
-       ast_t           handoff = AST_HANDOFF;
+       cpu_quiescent_counter_checkin(ctime);
 
 
-       funnel_release_check(self, 3);
+       uint64_t deadline = sched_maintenance_deadline;
 
 
-       self->continuation = continuation;
-       self->parameter = parameter;
+       if (__improbable(ctime >= deadline)) {
+               if (__improbable(current_thread() == sched_maintenance_thread)) {
+                       return;
+               }
+               OSMemoryBarrier();
 
 
-       while (!thread_invoke(self, new_thread, handoff)) {
-               register processor_t            processor = current_processor();
+               uint64_t ndeadline = ctime + sched_tick_interval;
 
 
-               thread_lock(self);
-               new_thread = thread_select(processor);
-               thread_unlock(self);
-               handoff = AST_NONE;
+               if (__probable(os_atomic_cmpxchg(&sched_maintenance_deadline, deadline, ndeadline, seq_cst))) {
+                       thread_wakeup((event_t)sched_timeshare_maintenance_continue);
+                       sched_maintenance_wakeups++;
+               }
+       }
+
+#if !CONFIG_SCHED_CLUTCH
+       /*
+        * Only non-clutch schedulers use the global load calculation EWMA algorithm. For clutch
+        * scheduler, the load is maintained at the thread group and bucket level.
+        */
+       uint64_t load_compute_deadline = os_atomic_load_wide(&sched_load_compute_deadline, relaxed);
+
+       if (__improbable(load_compute_deadline && ctime >= load_compute_deadline)) {
+               uint64_t new_deadline = 0;
+               if (os_atomic_cmpxchg(&sched_load_compute_deadline, load_compute_deadline, new_deadline, relaxed)) {
+                       compute_sched_load();
+                       new_deadline = ctime + sched_load_compute_interval_abs;
+                       os_atomic_store_wide(&sched_load_compute_deadline, new_deadline, relaxed);
+               }
        }
        }
+#endif /* CONFIG_SCHED_CLUTCH */
 
 
-       funnel_refunnel_check(self, 6);
+#if __arm64__
+       uint64_t perf_deadline = os_atomic_load(&sched_perfcontrol_callback_deadline, relaxed);
 
 
-       return (self->wait_result);
+       if (__improbable(perf_deadline && ctime >= perf_deadline)) {
+               /* CAS in 0, if success, make callback. Otherwise let the next context switch check again. */
+               if (os_atomic_cmpxchg(&sched_perfcontrol_callback_deadline, perf_deadline, 0, relaxed)) {
+                       machine_perfcontrol_deadline_passed(perf_deadline);
+               }
+       }
+#endif /* __arm64__ */
 }
 
 }
 
-/*
- *     thread_continue:
- *
- *     Called at splsched when a thread first receives
- *     a new stack after a continuation.
- */
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
+
 void
 void
-thread_continue(
-       register thread_t       old_thread)
+sched_init_thread(void (*continuation)(void))
 {
 {
-       register thread_t                       self = current_thread();
-       register thread_continue_t      continuation;
-       register void                           *parameter;
-       
-       continuation = self->continuation;
-       self->continuation = NULL;
-       parameter = self->parameter;
-       self->parameter = NULL;
+       thread_block(THREAD_CONTINUE_NULL);
 
 
-       thread_begin(self, self->last_processor);
+       thread_t thread = current_thread();
 
 
-       if (old_thread != THREAD_NULL)
-               thread_dispatch(old_thread);
+       thread_set_thread_name(thread, "sched_maintenance_thread");
 
 
-       funnel_refunnel_check(self, 4);
+       sched_maintenance_thread = thread;
 
 
-       if (old_thread != THREAD_NULL)
-               (void)spllo();
+       continuation();
 
 
-       call_continuation(continuation, parameter, self->wait_result);
        /*NOTREACHED*/
 }
 
        /*NOTREACHED*/
 }
 
+#if defined(CONFIG_SCHED_TIMESHARE_CORE)
+
 /*
 /*
- *     Enqueue thread on run queue.  Thread must be locked,
- *     and not already be on a run queue.  Returns TRUE
- *     if a preemption is indicated based on the state
- *     of the run queue.
+ *     thread_update_scan / runq_scan:
+ *
+ *     Scan the run queues to account for timesharing threads
+ *     which need to be updated.
  *
  *
- *     Run queue must be locked, see run_queue_remove()
- *     for more info.
+ *     Scanner runs in two passes.  Pass one squirrels likely
+ *     threads away in an array, pass two does the update.
+ *
+ *     This is necessary because the run queue is locked for
+ *     the candidate scan, but the thread is locked for the update.
+ *
+ *     Array should be sized to make forward progress, without
+ *     disabling preemption for long periods.
  */
  */
-static boolean_t
-run_queue_enqueue(
-       register run_queue_t    rq,
-       register thread_t               thread,
-       integer_t                               options)
+
+#define THREAD_UPDATE_SIZE              128
+
+static thread_t thread_update_array[THREAD_UPDATE_SIZE];
+static uint32_t thread_update_count = 0;
+
+/* Returns TRUE if thread was added, FALSE if thread_update_array is full */
+boolean_t
+thread_update_add_thread(thread_t thread)
 {
 {
-       register int                    whichq = thread->sched_pri;
-       register queue_t                queue = &rq->queues[whichq];
-       boolean_t                               result = FALSE;
-       
-       assert(whichq >= MINPRI && whichq <= MAXPRI);
+       if (thread_update_count == THREAD_UPDATE_SIZE) {
+               return FALSE;
+       }
 
 
-       assert(thread->runq == RUN_QUEUE_NULL);
-       if (queue_empty(queue)) {
-               enqueue_tail(queue, (queue_entry_t)thread);
+       thread_update_array[thread_update_count++] = thread;
+       thread_reference_internal(thread);
+       return TRUE;
+}
 
 
-               setbit(MAXPRI - whichq, rq->bitmap);
-               if (whichq > rq->highq) {
-                       rq->highq = whichq;
-                       result = TRUE;
+void
+thread_update_process_threads(void)
+{
+       assert(thread_update_count <= THREAD_UPDATE_SIZE);
+
+       for (uint32_t i = 0; i < thread_update_count; i++) {
+               thread_t thread = thread_update_array[i];
+               assert_thread_magic(thread);
+               thread_update_array[i] = THREAD_NULL;
+
+               spl_t s = splsched();
+               thread_lock(thread);
+               if (!(thread->state & (TH_WAIT)) && thread->sched_stamp != sched_tick) {
+                       SCHED(update_priority)(thread);
                }
                }
-       }
-       else
-       if (options & SCHED_HEADQ)
-               enqueue_head(queue, (queue_entry_t)thread);
-       else
-               enqueue_tail(queue, (queue_entry_t)thread);
+               thread_unlock(thread);
+               splx(s);
 
 
-       thread->runq = rq;
-       if (thread->sched_mode & TH_MODE_PREEMPT)
-               rq->urgency++;
-       rq->count++;
+               thread_deallocate(thread);
+       }
 
 
-       return (result);
+       thread_update_count = 0;
 }
 
 /*
 }
 
 /*
- *     Enqueue a thread for realtime execution, similar
- *     to above.  Handles preemption directly.
+ *     Scan a runq for candidate threads.
+ *
+ *     Returns TRUE if retry is needed.
  */
  */
-static void
-realtime_schedule_insert(
-       register processor_set_t        pset,
-       register thread_t                       thread)
+boolean_t
+runq_scan(
+       run_queue_t                   runq,
+       sched_update_scan_context_t   scan_context)
 {
 {
-       register run_queue_t    rq = &pset->runq;
-       register int                    whichq = thread->sched_pri;
-       register queue_t                queue = &rq->queues[whichq];
-       uint64_t                                deadline = thread->realtime.deadline;
-       boolean_t                               try_preempt = FALSE;
-
-       assert(whichq >= BASEPRI_REALTIME && whichq <= MAXPRI);
+       int count       = runq->count;
+       int queue_index;
 
 
-       assert(thread->runq == RUN_QUEUE_NULL);
-       if (queue_empty(queue)) {
-               enqueue_tail(queue, (queue_entry_t)thread);
+       assert(count >= 0);
 
 
-               setbit(MAXPRI - whichq, rq->bitmap);
-               if (whichq > rq->highq)
-                       rq->highq = whichq;
-               try_preempt = TRUE;
+       if (count == 0) {
+               return FALSE;
        }
        }
-       else {
-               register thread_t       entry = (thread_t)queue_first(queue);
 
 
-               while (TRUE) {
-                       if (    queue_end(queue, (queue_entry_t)entry)  ||
-                                               deadline < entry->realtime.deadline             ) {
-                               entry = (thread_t)queue_prev((queue_entry_t)entry);
-                               break;
-                       }
+       for (queue_index = bitmap_first(runq->bitmap, NRQS);
+           queue_index >= 0;
+           queue_index = bitmap_next(runq->bitmap, queue_index)) {
+               thread_t thread;
+               circle_queue_t queue = &runq->queues[queue_index];
 
 
-                       entry = (thread_t)queue_next((queue_entry_t)entry);
-               }
+               cqe_foreach_element(thread, queue, runq_links) {
+                       assert(count > 0);
+                       assert_thread_magic(thread);
 
 
-               if ((queue_entry_t)entry == queue)
-                       try_preempt = TRUE;
+                       if (thread->sched_stamp != sched_tick &&
+                           thread->sched_mode == TH_MODE_TIMESHARE) {
+                               if (thread_update_add_thread(thread) == FALSE) {
+                                       return TRUE;
+                               }
+                       }
 
 
-               insque((queue_entry_t)thread, (queue_entry_t)entry);
+                       if (cpu_throttle_enabled && ((thread->sched_pri <= MAXPRI_THROTTLE) && (thread->base_pri <= MAXPRI_THROTTLE))) {
+                               if (thread->last_made_runnable_time < scan_context->earliest_bg_make_runnable_time) {
+                                       scan_context->earliest_bg_make_runnable_time = thread->last_made_runnable_time;
+                               }
+                       } else {
+                               if (thread->last_made_runnable_time < scan_context->earliest_normal_make_runnable_time) {
+                                       scan_context->earliest_normal_make_runnable_time = thread->last_made_runnable_time;
+                               }
+                       }
+                       count--;
+               }
        }
 
        }
 
-       thread->runq = rq;
-       assert(thread->sched_mode & TH_MODE_PREEMPT);
-       rq->count++; rq->urgency++;
+       return FALSE;
+}
+
+#endif /* CONFIG_SCHED_TIMESHARE_CORE */
 
 
-       if (try_preempt) {
-               register processor_t    processor;
+boolean_t
+thread_eager_preemption(thread_t thread)
+{
+       return (thread->sched_flags & TH_SFLAG_EAGERPREEMPT) != 0;
+}
 
 
-               processor = current_processor();
-               if (    pset == processor->processor_set                                &&
-                               (thread->sched_pri > processor->current_pri     ||
-                                       deadline < processor->deadline                  )               ) {
-                       dispatch_counts.realtime_self++;
-                       simple_unlock(&pset->sched_lock);
+void
+thread_set_eager_preempt(thread_t thread)
+{
+       spl_t x;
+       processor_t p;
+       ast_t ast = AST_NONE;
 
 
-                       ast_on(AST_PREEMPT | AST_URGENT);
-                       return;
-               }
+       x = splsched();
+       p = current_processor();
 
 
-               if (    pset->processor_count > 1                       ||
-                               pset != processor->processor_set        ) {
-                       processor_t             myprocessor, lastprocessor;
-                       queue_entry_t   next;
+       thread_lock(thread);
+       thread->sched_flags |= TH_SFLAG_EAGERPREEMPT;
 
 
-                       myprocessor = processor;
-                       processor = thread->last_processor;
-                       if (    processor != myprocessor                                                &&
-                                       processor != PROCESSOR_NULL                                             &&
-                                       processor->processor_set == pset                                &&
-                                       processor->state == PROCESSOR_RUNNING                   &&
-                                       (thread->sched_pri > processor->current_pri     ||
-                                               deadline < processor->deadline                  )               ) {
-                               dispatch_counts.realtime_last++;
-                               cause_ast_check(processor);
-                               simple_unlock(&pset->sched_lock);
-                               return;
-                       }
-
-                       lastprocessor = processor;
-                       queue = &pset->active_queue;
-                       processor = (processor_t)queue_first(queue);
-                       while (!queue_end(queue, (queue_entry_t)processor)) {
-                               next = queue_next((queue_entry_t)processor);
-
-                               if (    processor != myprocessor                                                &&
-                                               processor != lastprocessor                                              &&
-                                               (thread->sched_pri > processor->current_pri     ||
-                                                       deadline < processor->deadline                  )               ) {
-                                       if (!queue_end(queue, next)) {
-                                               remqueue(queue, (queue_entry_t)processor);
-                                               enqueue_tail(queue, (queue_entry_t)processor);
-                                       }
-                                       dispatch_counts.realtime_other++;
-                                       cause_ast_check(processor);
-                                       simple_unlock(&pset->sched_lock);
-                                       return;
-                               }
+       if (thread == current_thread()) {
+               ast = csw_check(thread, p, AST_NONE);
+               thread_unlock(thread);
+               if (ast != AST_NONE) {
+                       (void) thread_block_reason(THREAD_CONTINUE_NULL, NULL, ast);
+               }
+       } else {
+               p = thread->last_processor;
 
 
-                               processor = (processor_t)next;
-                       }
+               if (p != PROCESSOR_NULL && p->state == PROCESSOR_RUNNING &&
+                   p->active_thread == thread) {
+                       cause_ast_check(p);
                }
                }
+
+               thread_unlock(thread);
        }
 
        }
 
-       simple_unlock(&pset->sched_lock);
+       splx(x);
+}
+
+void
+thread_clear_eager_preempt(thread_t thread)
+{
+       spl_t x;
+
+       x = splsched();
+       thread_lock(thread);
+
+       thread->sched_flags &= ~TH_SFLAG_EAGERPREEMPT;
+
+       thread_unlock(thread);
+       splx(x);
 }
 
 /*
 }
 
 /*
- *     thread_setrun:
- *
- *     Dispatch thread for execution, directly onto an idle
- *     processor if possible.  Else put on appropriate run
- *     queue. (local if bound, else processor set)
- *
- *     Thread must be locked.
+ * Scheduling statistics
  */
 void
  */
 void
-thread_setrun(
-       register thread_t                       new_thread,
-       integer_t                                       options)
+sched_stats_handle_csw(processor_t processor, int reasons, int selfpri, int otherpri)
 {
 {
-       register processor_t            processor;
-       register processor_set_t        pset;
-       register thread_t                       thread;
-       ast_t                                           preempt = (options & SCHED_PREEMPT)?
-                                                                                                       AST_PREEMPT: AST_NONE;
+       struct processor_sched_statistics *stats;
+       boolean_t to_realtime = FALSE;
 
 
-       assert(thread_runnable(new_thread));
-       
-       /*
-        *      Update priority if needed.
-        */
-       if (new_thread->sched_stamp != sched_tick)
-               update_priority(new_thread);
+       stats = &processor->processor_data.sched_stats;
+       stats->csw_count++;
 
 
-       /*
-        *      Check for urgent preemption.
-        */
-       if (new_thread->sched_mode & TH_MODE_PREEMPT)
-               preempt = (AST_PREEMPT | AST_URGENT);
+       if (otherpri >= BASEPRI_REALTIME) {
+               stats->rt_sched_count++;
+               to_realtime = TRUE;
+       }
 
 
-       assert(new_thread->runq == RUN_QUEUE_NULL);
-
-       if ((processor = new_thread->bound_processor) == PROCESSOR_NULL) {
-           /*
-            *  First try to dispatch on
-                *      the last processor.
-            */
-           pset = new_thread->processor_set;
-               processor = new_thread->last_processor;
-               if (    pset->processor_count > 1                               &&
-                               processor != PROCESSOR_NULL                             &&
-                               processor->state == PROCESSOR_IDLE              ) {
-                       processor_lock(processor);
-                       simple_lock(&pset->sched_lock);
-                       if (    processor->processor_set == pset                &&
-                                       processor->state == PROCESSOR_IDLE              ) {
-                               remqueue(&pset->idle_queue, (queue_entry_t)processor);
-                               pset->idle_count--;
-                               processor->next_thread = new_thread;
-                               if (new_thread->sched_pri >= BASEPRI_RTQUEUES)
-                                       processor->deadline = new_thread->realtime.deadline;
-                               else
-                                       processor->deadline = UINT64_MAX;
-                               processor->state = PROCESSOR_DISPATCHING;
-                               dispatch_counts.idle_pset_last++;
-                               simple_unlock(&pset->sched_lock);
-                               processor_unlock(processor);
-                               if (processor != current_processor())
-                                       machine_signal_idle(processor);
-                               return;
-                       }
-                       processor_unlock(processor);
+       if ((reasons & AST_PREEMPT) != 0) {
+               stats->preempt_count++;
+
+               if (selfpri >= BASEPRI_REALTIME) {
+                       stats->preempted_rt_count++;
                }
                }
-               else
-               simple_lock(&pset->sched_lock);
 
 
-               /*
-                *      Next pick any idle processor
-                *      in the processor set.
-                */
-               if (pset->idle_count > 0) {
-                       processor = (processor_t)dequeue_head(&pset->idle_queue);
-                       pset->idle_count--;
-                       processor->next_thread = new_thread;
-                       if (new_thread->sched_pri >= BASEPRI_RTQUEUES)
-                               processor->deadline = new_thread->realtime.deadline;
-                       else
-                               processor->deadline = UINT64_MAX;
-                       processor->state = PROCESSOR_DISPATCHING;
-                       dispatch_counts.idle_pset_any++;
-                       simple_unlock(&pset->sched_lock);
-                       if (processor != current_processor())   
-                               machine_signal_idle(processor);
-                       return;
+               if (to_realtime) {
+                       stats->preempted_by_rt_count++;
                }
                }
+       }
+}
 
 
-               if (new_thread->sched_pri >= BASEPRI_RTQUEUES)
-                       realtime_schedule_insert(pset, new_thread);
-               else {
-                       if (!run_queue_enqueue(&pset->runq, new_thread, options))
-                               preempt = AST_NONE;
+void
+sched_stats_handle_runq_change(struct runq_stats *stats, int old_count)
+{
+       uint64_t timestamp = mach_absolute_time();
 
 
-                       /*
-                        *      Update the timesharing quanta.
-                        */
-                       timeshare_quanta_update(pset);
-       
-                       /*
-                        *      Preempt check.
-                        */
-                       if (preempt != AST_NONE) {
-                               /*
-                                * First try the current processor
-                                * if it is a member of the correct
-                                * processor set.
-                                */
-                               processor = current_processor();
-                               thread = processor->active_thread;
-                               if (    pset == processor->processor_set        &&
-                                               csw_needed(thread, processor)           ) {
-                                       dispatch_counts.pset_self++;
-                                       simple_unlock(&pset->sched_lock);
-
-                                       ast_on(preempt);
-                                       return;
-                               }
+       stats->count_sum += (timestamp - stats->last_change_timestamp) * old_count;
+       stats->last_change_timestamp = timestamp;
+}
 
 
-                               /*
-                                * If that failed and we have other
-                                * processors available keep trying.
-                                */
-                               if (    pset->processor_count > 1                       ||
-                                               pset != processor->processor_set        ) {
-                                       queue_t                 queue = &pset->active_queue;
-                                       processor_t             myprocessor, lastprocessor;
-                                       queue_entry_t   next;
-
-                                       /*
-                                        * Next try the last processor
-                                        * dispatched on.
-                                        */
-                                       myprocessor = processor;
-                                       processor = new_thread->last_processor;
-                                       if (    processor != myprocessor                                                &&
-                                                       processor != PROCESSOR_NULL                                             &&
-                                                       processor->processor_set == pset                                &&
-                                                       processor->state == PROCESSOR_RUNNING                   &&
-                                                       new_thread->sched_pri > processor->current_pri  ) {
-                                               dispatch_counts.pset_last++;
-                                               cause_ast_check(processor);
-                                               simple_unlock(&pset->sched_lock);
-                                               return;
-                                       }
+/*
+ *     For calls from assembly code
+ */
+#undef thread_wakeup
+void
+thread_wakeup(
+       event_t         x);
 
 
-                                       /*
-                                        * Lastly, pick any other
-                                        * available processor.
-                                        */
-                                       lastprocessor = processor;
-                                       processor = (processor_t)queue_first(queue);
-                                       while (!queue_end(queue, (queue_entry_t)processor)) {
-                                               next = queue_next((queue_entry_t)processor);
-
-                                               if (    processor != myprocessor                        &&
-                                                               processor != lastprocessor                      &&
-                                                               new_thread->sched_pri >
-                                                                                       processor->current_pri          ) {
-                                                       if (!queue_end(queue, next)) {
-                                                               remqueue(queue, (queue_entry_t)processor);
-                                                               enqueue_tail(queue, (queue_entry_t)processor);
-                                                       }
-                                                       dispatch_counts.pset_other++;
-                                                       cause_ast_check(processor);
-                                                       simple_unlock(&pset->sched_lock);
-                                                       return;
-                                               }
+void
+thread_wakeup(
+       event_t         x)
+{
+       thread_wakeup_with_result(x, THREAD_AWAKENED);
+}
 
 
-                                               processor = (processor_t)next;
-                                       }
-                               }
-                       }
+boolean_t
+preemption_enabled(void)
+{
+       return get_preemption_level() == 0 && ml_get_interrupts_enabled();
+}
 
 
-                       simple_unlock(&pset->sched_lock);
-               }
-       }
-       else {
-           /*
-            *  Bound, can only run on bound processor.  Have to lock
-            *  processor here because it may not be the current one.
-            */
-               processor_lock(processor);
-               pset = processor->processor_set;
-               if (pset != PROCESSOR_SET_NULL) {
-                       simple_lock(&pset->sched_lock);
-                       if (processor->state == PROCESSOR_IDLE) {
-                               remqueue(&pset->idle_queue, (queue_entry_t)processor);
-                               pset->idle_count--;
-                               processor->next_thread = new_thread;
-                               processor->deadline = UINT64_MAX;
-                               processor->state = PROCESSOR_DISPATCHING;
-                               dispatch_counts.idle_bound++;
-                               simple_unlock(&pset->sched_lock);
-                               processor_unlock(processor);
-                               if (processor != current_processor())   
-                                       machine_signal_idle(processor);
-                               return;
-                       }
-               }
-         
-               if (!run_queue_enqueue(&processor->runq, new_thread, options))
-                       preempt = AST_NONE;
+static void
+sched_timer_deadline_tracking_init(void)
+{
+       nanoseconds_to_absolutetime(TIMER_DEADLINE_TRACKING_BIN_1_DEFAULT, &timer_deadline_tracking_bin_1);
+       nanoseconds_to_absolutetime(TIMER_DEADLINE_TRACKING_BIN_2_DEFAULT, &timer_deadline_tracking_bin_2);
+}
 
 
-               if (preempt != AST_NONE) {
-                       if (processor == current_processor()) {
-                               thread = processor->active_thread;
-                               if (csw_needed(thread, processor)) {
-                                       dispatch_counts.bound_self++;
-                                       ast_on(preempt);
-                               }
-                       }
-                       else
-                       if (    processor->state == PROCESSOR_RUNNING                   &&
-                                       new_thread->sched_pri > processor->current_pri  ) {
-                               dispatch_counts.bound_other++;
-                               cause_ast_check(processor);
-                       }
-               }
+#if __arm__ || __arm64__
 
 
-               if (pset != PROCESSOR_SET_NULL)
-                       simple_unlock(&pset->sched_lock);
+uint32_t    perfcontrol_requested_recommended_cores = ALL_CORES_RECOMMENDED;
+uint32_t    perfcontrol_requested_recommended_core_count = MAX_CPUS;
+bool        perfcontrol_failsafe_active = false;
+bool        perfcontrol_sleep_override = false;
 
 
-               processor_unlock(processor);
-       }
-}
+uint64_t    perfcontrol_failsafe_maintenance_runnable_time;
+uint64_t    perfcontrol_failsafe_activation_time;
+uint64_t    perfcontrol_failsafe_deactivation_time;
+
+/* data covering who likely caused it and how long they ran */
+#define FAILSAFE_NAME_LEN       33 /* (2*MAXCOMLEN)+1 from size of p_name */
+char        perfcontrol_failsafe_name[FAILSAFE_NAME_LEN];
+int         perfcontrol_failsafe_pid;
+uint64_t    perfcontrol_failsafe_tid;
+uint64_t    perfcontrol_failsafe_thread_timer_at_start;
+uint64_t    perfcontrol_failsafe_thread_timer_last_seen;
+uint32_t    perfcontrol_failsafe_recommended_at_trigger;
 
 /*
 
 /*
- *     Check for a possible preemption point in
- *     the (current) thread.
+ * Perf controller calls here to update the recommended core bitmask.
+ * If the failsafe is active, we don't immediately apply the new value.
+ * Instead, we store the new request and use it after the failsafe deactivates.
  *
  *
- *     Called at splsched.
+ * If the failsafe is not active, immediately apply the update.
+ *
+ * No scheduler locks are held, no other locks are held that scheduler might depend on,
+ * interrupts are enabled
+ *
+ * currently prototype is in osfmk/arm/machine_routines.h
  */
  */
-ast_t
-csw_check(
-       thread_t                thread,
-       processor_t             processor)
+void
+sched_perfcontrol_update_recommended_cores(uint32_t recommended_cores)
 {
 {
-       int                             current_pri = thread->sched_pri;
-       ast_t                   result = AST_NONE;
-       run_queue_t             runq;
-
-       if (first_timeslice(processor)) {
-               runq = &processor->processor_set->runq;
-               if (runq->highq >= BASEPRI_RTQUEUES)
-                       return (AST_PREEMPT | AST_URGENT);
-
-               if (runq->highq > current_pri) {
-                       if (runq->urgency > 0)
-                               return (AST_PREEMPT | AST_URGENT);
+       assert(preemption_enabled());
 
 
-                       result |= AST_PREEMPT;
-               }
+       spl_t s = splsched();
+       simple_lock(&sched_recommended_cores_lock, LCK_GRP_NULL);
 
 
-               runq = &processor->runq;
-               if (runq->highq > current_pri) {
-                       if (runq->urgency > 0)
-                               return (AST_PREEMPT | AST_URGENT);
+       perfcontrol_requested_recommended_cores = recommended_cores;
+       perfcontrol_requested_recommended_core_count = __builtin_popcountll(recommended_cores);
 
 
-                       result |= AST_PREEMPT;
-               }
+       if ((perfcontrol_failsafe_active == false) && (perfcontrol_sleep_override == false)) {
+               sched_update_recommended_cores(perfcontrol_requested_recommended_cores & usercontrol_requested_recommended_cores);
+       } else {
+               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                   MACHDBG_CODE(DBG_MACH_SCHED, MACH_REC_CORES_FAILSAFE) | DBG_FUNC_NONE,
+                   perfcontrol_requested_recommended_cores,
+                   sched_maintenance_thread->last_made_runnable_time, 0, 0, 0);
        }
        }
-       else {
-               runq = &processor->processor_set->runq;
-               if (runq->highq >= current_pri) {
-                       if (runq->urgency > 0)
-                               return (AST_PREEMPT | AST_URGENT);
 
 
-                       result |= AST_PREEMPT;
-               }
+       simple_unlock(&sched_recommended_cores_lock);
+       splx(s);
+}
 
 
-               runq = &processor->runq;
-               if (runq->highq >= current_pri) {
-                       if (runq->urgency > 0)
-                               return (AST_PREEMPT | AST_URGENT);
+void
+sched_override_recommended_cores_for_sleep(void)
+{
+       spl_t s = splsched();
+       simple_lock(&sched_recommended_cores_lock, LCK_GRP_NULL);
 
 
-                       result |= AST_PREEMPT;
-               }
+       if (perfcontrol_sleep_override == false) {
+               perfcontrol_sleep_override = true;
+               sched_update_recommended_cores(ALL_CORES_RECOMMENDED);
        }
 
        }
 
-       if (result != AST_NONE)
-               return (result);
+       simple_unlock(&sched_recommended_cores_lock);
+       splx(s);
+}
+
+void
+sched_restore_recommended_cores_after_sleep(void)
+{
+       spl_t s = splsched();
+       simple_lock(&sched_recommended_cores_lock, LCK_GRP_NULL);
 
 
-       if (thread->state & TH_SUSP)
-               result |= AST_PREEMPT;
+       if (perfcontrol_sleep_override == true) {
+               perfcontrol_sleep_override = false;
+               sched_update_recommended_cores(perfcontrol_requested_recommended_cores & usercontrol_requested_recommended_cores);
+       }
 
 
-       return (result);
+       simple_unlock(&sched_recommended_cores_lock);
+       splx(s);
 }
 
 /*
 }
 
 /*
- *     set_sched_pri:
+ * Consider whether we need to activate the recommended cores failsafe
  *
  *
- *     Set the scheduled priority of the specified thread.
- *
- *     This may cause the thread to change queues.
- *
- *     Thread must be locked.
+ * Called from quantum timer interrupt context of a realtime thread
+ * No scheduler locks are held, interrupts are disabled
  */
 void
  */
 void
-set_sched_pri(
-       thread_t                        thread,
-       int                                     priority)
-{
-       register struct run_queue       *rq = run_queue_remove(thread);
-
-       if (    !(thread->sched_mode & TH_MODE_TIMESHARE)                               &&
-                       (priority >= BASEPRI_PREEMPT                                            ||
-                        (thread->task_priority < MINPRI_KERNEL                 &&
-                         thread->task_priority >= BASEPRI_BACKGROUND   &&
-                         priority > thread->task_priority)                                     )       )
-               thread->sched_mode |= TH_MODE_PREEMPT;
-       else
-               thread->sched_mode &= ~TH_MODE_PREEMPT;
-
-       thread->sched_pri = priority;
-       if (rq != RUN_QUEUE_NULL)
-               thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
-       else
-       if (thread->state & TH_RUN) {
-               processor_t             processor = thread->last_processor;
+sched_consider_recommended_cores(uint64_t ctime, thread_t cur_thread)
+{
+       /*
+        * Check if a realtime thread is starving the system
+        * and bringing up non-recommended cores would help
+        *
+        * TODO: Is this the correct check for recommended == possible cores?
+        * TODO: Validate the checks without the relevant lock are OK.
+        */
 
 
-               if (thread == current_thread()) {
-                       ast_t           preempt = csw_check(thread, processor);
+       if (__improbable(perfcontrol_failsafe_active == TRUE)) {
+               /* keep track of how long the responsible thread runs */
 
 
-                       if (preempt != AST_NONE)
-                               ast_on(preempt);
-                       processor->current_pri = priority;
+               simple_lock(&sched_recommended_cores_lock, LCK_GRP_NULL);
+
+               if (perfcontrol_failsafe_active == TRUE &&
+                   cur_thread->thread_id == perfcontrol_failsafe_tid) {
+                       perfcontrol_failsafe_thread_timer_last_seen = timer_grab(&cur_thread->user_timer) +
+                           timer_grab(&cur_thread->system_timer);
                }
                }
-               else
-               if (    processor != PROCESSOR_NULL                                             &&
-                               processor->active_thread == thread      )
-                       cause_ast_check(processor);
+
+               simple_unlock(&sched_recommended_cores_lock);
+
+               /* we're already trying to solve the problem, so bail */
+               return;
        }
        }
-}
 
 
-#if            0
+       /* The failsafe won't help if there are no more processors to enable */
+       if (__probable(perfcontrol_requested_recommended_core_count >= processor_count)) {
+               return;
+       }
 
 
-static void
-run_queue_check(
-       run_queue_t             rq,
-       thread_t                thread)
-{
-       queue_t                 q;
-       queue_entry_t   qe;
+       uint64_t too_long_ago = ctime - perfcontrol_failsafe_starvation_threshold;
 
 
-       if (rq != thread->runq)
-               panic("run_queue_check: thread runq");
+       /* Use the maintenance thread as our canary in the coal mine */
+       thread_t m_thread = sched_maintenance_thread;
 
 
-       if (thread->sched_pri > MAXPRI || thread->sched_pri < MINPRI)
-               panic("run_queue_check: thread sched_pri");
+       /* If it doesn't look bad, nothing to see here */
+       if (__probable(m_thread->last_made_runnable_time >= too_long_ago)) {
+               return;
+       }
 
 
-       q = &rq->queues[thread->sched_pri];
-       qe = queue_first(q);
-       while (!queue_end(q, qe)) {
-               if (qe == (queue_entry_t)thread)
-                       return;
+       /* It looks bad, take the lock to be sure */
+       thread_lock(m_thread);
 
 
-               qe = queue_next(qe);
+       if (m_thread->runq == PROCESSOR_NULL ||
+           (m_thread->state & (TH_RUN | TH_WAIT)) != TH_RUN ||
+           m_thread->last_made_runnable_time >= too_long_ago) {
+               /*
+                * Maintenance thread is either on cpu or blocked, and
+                * therefore wouldn't benefit from more cores
+                */
+               thread_unlock(m_thread);
+               return;
        }
 
        }
 
-       panic("run_queue_check: end");
-}
+       uint64_t maintenance_runnable_time = m_thread->last_made_runnable_time;
 
 
-#endif /* DEBUG */
+       thread_unlock(m_thread);
 
 
-/*
- *     run_queue_remove:
- *
- *     Remove a thread from its current run queue and
- *     return the run queue if successful.
- *
- *     Thread must be locked.
- */
-run_queue_t
-run_queue_remove(
-       thread_t                        thread)
-{
-       register run_queue_t    rq = thread->runq;
+       /*
+        * There are cores disabled at perfcontrol's recommendation, but the
+        * system is so overloaded that the maintenance thread can't run.
+        * That likely means that perfcontrol can't run either, so it can't fix
+        * the recommendation.  We have to kick in a failsafe to keep from starving.
+        *
+        * When the maintenance thread has been starved for too long,
+        * ignore the recommendation from perfcontrol and light up all the cores.
+        *
+        * TODO: Consider weird states like boot, sleep, or debugger
+        */
+
+       simple_lock(&sched_recommended_cores_lock, LCK_GRP_NULL);
+
+       if (perfcontrol_failsafe_active == TRUE) {
+               simple_unlock(&sched_recommended_cores_lock);
+               return;
+       }
 
 
-       /*
-        *      If rq is RUN_QUEUE_NULL, the thread will stay out of the
-        *      run queues because the caller locked the thread.  Otherwise
-        *      the thread is on a run queue, but could be chosen for dispatch
-        *      and removed.
-        */
-       if (rq != RUN_QUEUE_NULL) {
-               processor_set_t         pset = thread->processor_set;
-               processor_t                     processor = thread->bound_processor;
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_REC_CORES_FAILSAFE) | DBG_FUNC_START,
+           perfcontrol_requested_recommended_cores, maintenance_runnable_time, 0, 0, 0);
 
 
-               /*
-                *      The run queues are locked by the pset scheduling
-                *      lock, except when a processor is off-line the
-                *      local run queue is locked by the processor lock.
-                */
-               if (processor != PROCESSOR_NULL) {
-                       processor_lock(processor);
-                       pset = processor->processor_set;
-               }
+       perfcontrol_failsafe_active = TRUE;
+       perfcontrol_failsafe_activation_time = mach_absolute_time();
+       perfcontrol_failsafe_maintenance_runnable_time = maintenance_runnable_time;
+       perfcontrol_failsafe_recommended_at_trigger = perfcontrol_requested_recommended_cores;
 
 
-               if (pset != PROCESSOR_SET_NULL)
-                       simple_lock(&pset->sched_lock);
+       /* Capture some data about who screwed up (assuming that the thread on core is at fault) */
+       task_t task = cur_thread->task;
+       perfcontrol_failsafe_pid = task_pid(task);
+       strlcpy(perfcontrol_failsafe_name, proc_name_address(task->bsd_info), sizeof(perfcontrol_failsafe_name));
 
 
-               if (rq == thread->runq) {
-                       /*
-                        *      Thread is on a run queue and we have a lock on
-                        *      that run queue.
-                        */
-                       remqueue(&rq->queues[0], (queue_entry_t)thread);
-                       rq->count--;
-                       if (thread->sched_mode & TH_MODE_PREEMPT)
-                               rq->urgency--;
-                       assert(rq->urgency >= 0);
+       perfcontrol_failsafe_tid = cur_thread->thread_id;
 
 
-                       if (queue_empty(rq->queues + thread->sched_pri)) {
-                               /* update run queue status */
-                               if (thread->sched_pri != IDLEPRI)
-                                       clrbit(MAXPRI - thread->sched_pri, rq->bitmap);
-                               rq->highq = MAXPRI - ffsbit(rq->bitmap);
-                       }
+       /* Blame the thread for time it has run recently */
+       uint64_t recent_computation = (ctime - cur_thread->computation_epoch) + cur_thread->computation_metered;
 
 
-                       thread->runq = RUN_QUEUE_NULL;
-               }
-               else {
-                       /*
-                        *      The thread left the run queue before we could
-                        *      lock the run queue.
-                        */
-                       assert(thread->runq == RUN_QUEUE_NULL);
-                       rq = RUN_QUEUE_NULL;
-               }
+       uint64_t last_seen = timer_grab(&cur_thread->user_timer) + timer_grab(&cur_thread->system_timer);
 
 
-               if (pset != PROCESSOR_SET_NULL)
-                       simple_unlock(&pset->sched_lock);
+       /* Compute the start time of the bad behavior in terms of the thread's on core time */
+       perfcontrol_failsafe_thread_timer_at_start  = last_seen - recent_computation;
+       perfcontrol_failsafe_thread_timer_last_seen = last_seen;
 
 
-               if (processor != PROCESSOR_NULL)
-                       processor_unlock(processor);
-       }
+       /* Ignore the previously recommended core configuration */
+       sched_update_recommended_cores(ALL_CORES_RECOMMENDED);
 
 
-       return (rq);
+       simple_unlock(&sched_recommended_cores_lock);
 }
 
 /*
 }
 
 /*
- *     choose_thread:
- *
- *     Remove a thread to execute from the run queues
- *     and return it.
+ * Now that our bacon has been saved by the failsafe, consider whether to turn it off
  *
  *
- *     Called with pset scheduling lock held.
+ * Runs in the context of the maintenance thread, no locks held
  */
  */
-static thread_t
-choose_thread(
-       processor_set_t         pset,
-       processor_t                     processor)
+static void
+sched_recommended_cores_maintenance(void)
 {
 {
-       register run_queue_t    runq;
-       register thread_t               thread;
-       register queue_t                q;
+       /* Common case - no failsafe, nothing to be done here */
+       if (__probable(perfcontrol_failsafe_active == FALSE)) {
+               return;
+       }
 
 
-       runq = &processor->runq;
+       uint64_t ctime = mach_absolute_time();
 
 
-       if (runq->count > 0 && runq->highq >= pset->runq.highq) {
-               q = runq->queues + runq->highq;
+       boolean_t print_diagnostic = FALSE;
+       char p_name[FAILSAFE_NAME_LEN] = "";
 
 
-               thread = (thread_t)q->next;
-               ((queue_entry_t)thread)->next->prev = q;
-               q->next = ((queue_entry_t)thread)->next;
-               thread->runq = RUN_QUEUE_NULL;
-               runq->count--;
-               if (thread->sched_mode & TH_MODE_PREEMPT)
-                       runq->urgency--;
-               assert(runq->urgency >= 0);
-               if (queue_empty(q)) {
-                       if (runq->highq != IDLEPRI)
-                               clrbit(MAXPRI - runq->highq, runq->bitmap);
-                       runq->highq = MAXPRI - ffsbit(runq->bitmap);
-               }
+       spl_t s = splsched();
+       simple_lock(&sched_recommended_cores_lock, LCK_GRP_NULL);
 
 
-               processor->deadline = UINT64_MAX;
+       /* Check again, under the lock, to avoid races */
+       if (perfcontrol_failsafe_active == FALSE) {
+               goto out;
+       }
 
 
-               return (thread);
+       /*
+        * Ensure that the other cores get another few ticks to run some threads
+        * If we don't have this hysteresis, the maintenance thread is the first
+        * to run, and then it immediately kills the other cores
+        */
+       if ((ctime - perfcontrol_failsafe_activation_time) < perfcontrol_failsafe_starvation_threshold) {
+               goto out;
        }
 
        }
 
-       runq = &pset->runq;
+       /* Capture some diagnostic state under the lock so we can print it out later */
 
 
-       assert(runq->count > 0);
-       q = runq->queues + runq->highq;
+       int      pid = perfcontrol_failsafe_pid;
+       uint64_t tid = perfcontrol_failsafe_tid;
 
 
-       thread = (thread_t)q->next;
-       ((queue_entry_t)thread)->next->prev = q;
-       q->next = ((queue_entry_t)thread)->next;
-       thread->runq = RUN_QUEUE_NULL;
-       runq->count--;
-       if (runq->highq >= BASEPRI_RTQUEUES)
-               processor->deadline = thread->realtime.deadline;
-       else
-               processor->deadline = UINT64_MAX;
-       if (thread->sched_mode & TH_MODE_PREEMPT)
-               runq->urgency--;
-       assert(runq->urgency >= 0);
-       if (queue_empty(q)) {
-               if (runq->highq != IDLEPRI)
-                       clrbit(MAXPRI - runq->highq, runq->bitmap);
-               runq->highq = MAXPRI - ffsbit(runq->bitmap);
-       }
+       uint64_t thread_usage       = perfcontrol_failsafe_thread_timer_last_seen -
+           perfcontrol_failsafe_thread_timer_at_start;
+       uint32_t rec_cores_before   = perfcontrol_failsafe_recommended_at_trigger;
+       uint32_t rec_cores_after    = perfcontrol_requested_recommended_cores;
+       uint64_t failsafe_duration  = ctime - perfcontrol_failsafe_activation_time;
+       strlcpy(p_name, perfcontrol_failsafe_name, sizeof(p_name));
 
 
-       timeshare_quanta_update(pset);
+       print_diagnostic = TRUE;
 
 
-       return (thread);
-}
+       /* Deactivate the failsafe and reinstate the requested recommendation settings */
 
 
-static processor_t
-delay_idle(
-       processor_t             processor,
-       thread_t                self)
-{
-       int                             *gcount, *lcount;
-       uint64_t                abstime, spin, limit;
+       perfcontrol_failsafe_deactivation_time = ctime;
+       perfcontrol_failsafe_active = FALSE;
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           MACHDBG_CODE(DBG_MACH_SCHED, MACH_REC_CORES_FAILSAFE) | DBG_FUNC_END,
+           perfcontrol_requested_recommended_cores, failsafe_duration, 0, 0, 0);
 
 
-       lcount = &processor->runq.count;
-       gcount = &processor->processor_set->runq.count;
+       sched_update_recommended_cores(perfcontrol_requested_recommended_cores & usercontrol_requested_recommended_cores);
 
 
-       abstime = mach_absolute_time();
-       limit = abstime + delay_idle_limit;
-       spin = abstime + delay_idle_spin;
+out:
+       simple_unlock(&sched_recommended_cores_lock);
+       splx(s);
 
 
-       timer_event((uint32_t)abstime, &processor->idle_thread->system_timer);
+       if (print_diagnostic) {
+               uint64_t failsafe_duration_ms = 0, thread_usage_ms = 0;
 
 
-       self->options |= TH_OPT_DELAYIDLE;
+               absolutetime_to_nanoseconds(failsafe_duration, &failsafe_duration_ms);
+               failsafe_duration_ms = failsafe_duration_ms / NSEC_PER_MSEC;
 
 
-       while (         *gcount == 0 && *lcount == 0    &&
-                               (self->state & TH_WAIT) != 0    &&
-                                       abstime < limit                         ) {
-               if (abstime >= spin) {
-                       (void)spllo();
+               absolutetime_to_nanoseconds(thread_usage, &thread_usage_ms);
+               thread_usage_ms = thread_usage_ms / NSEC_PER_MSEC;
 
 
-                       (void)splsched();
-                       processor = current_processor();
-                       lcount = &processor->runq.count;
-                       gcount = &processor->processor_set->runq.count;
+               printf("recommended core failsafe kicked in for %lld ms "
+                   "likely due to %s[%d] thread 0x%llx spending "
+                   "%lld ms on cpu at realtime priority - "
+                   "new recommendation: 0x%x -> 0x%x\n",
+                   failsafe_duration_ms, p_name, pid, tid, thread_usage_ms,
+                   rec_cores_before, rec_cores_after);
+       }
+}
 
 
-                       abstime = mach_absolute_time();
-                       spin = abstime + delay_idle_spin;
+#endif /* __arm__ || __arm64__ */
 
 
-                       timer_event((uint32_t)abstime, &processor->idle_thread->system_timer);
-               }
-               else {
-                       cpu_pause();
-                       abstime = mach_absolute_time();
-               }
+kern_return_t
+sched_processor_enable(processor_t processor, boolean_t enable)
+{
+       assert(preemption_enabled());
+
+       spl_t s = splsched();
+       simple_lock(&sched_recommended_cores_lock, LCK_GRP_NULL);
+
+       if (enable) {
+               bit_set(usercontrol_requested_recommended_cores, processor->cpu_id);
+       } else {
+               bit_clear(usercontrol_requested_recommended_cores, processor->cpu_id);
        }
 
        }
 
-       timer_event((uint32_t)abstime, &self->system_timer);
+#if __arm__ || __arm64__
+       if ((perfcontrol_failsafe_active == false) && (perfcontrol_sleep_override == false)) {
+               sched_update_recommended_cores(perfcontrol_requested_recommended_cores & usercontrol_requested_recommended_cores);
+       } else {
+               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                   MACHDBG_CODE(DBG_MACH_SCHED, MACH_REC_CORES_FAILSAFE) | DBG_FUNC_NONE,
+                   perfcontrol_requested_recommended_cores,
+                   sched_maintenance_thread->last_made_runnable_time, 0, 0, 0);
+       }
+#else /* __arm__ || __arm64__ */
+       sched_update_recommended_cores(usercontrol_requested_recommended_cores);
+#endif /* !__arm__ || __arm64__ */
 
 
-       self->options &= ~TH_OPT_DELAYIDLE;
+       simple_unlock(&sched_recommended_cores_lock);
+       splx(s);
 
 
-       return (processor);
+       return KERN_SUCCESS;
 }
 
 }
 
-/*
- *     no_dispatch_count counts number of times processors go non-idle
- *     without being dispatched.  This should be very rare.
- */
-int    no_dispatch_count = 0;
 
 /*
 
 /*
- *     This is the idle processor thread, which just looks for other threads
- *     to execute.
+ * Apply a new recommended cores mask to the processors it affects
+ * Runs after considering failsafes and such
+ *
+ * Iterate over processors and update their ->is_recommended field.
+ * If a processor is running, we let it drain out at its next
+ * quantum expiration or blocking point. If a processor is idle, there
+ * may be more work for it to do, so IPI it.
+ *
+ * interrupts disabled, sched_recommended_cores_lock is held
  */
  */
-void
-idle_thread(void)
+static void
+sched_update_recommended_cores(uint64_t recommended_cores)
 {
 {
-       register processor_t            processor;
-       register thread_t                       *threadp;
-       register int                            *gcount;
-       register int                            *lcount;
-       register thread_t                       new_thread;
-       register int                            state;
-       register processor_set_t        pset;
-       ast_t                                           *myast = ast_pending();
+       processor_set_t pset, nset;
+       processor_t     processor;
+       uint64_t        needs_exit_idle_mask = 0x0;
+       uint32_t        avail_count;
 
 
-       processor = current_processor();
+       processor = processor_list;
+       pset = processor->processor_set;
 
 
-       threadp = &processor->next_thread;
-       lcount = &processor->runq.count;
-       gcount = &processor->processor_set->runq.count;
+       KDBG(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_UPDATE_REC_CORES) | DBG_FUNC_START,
+           recommended_cores,
+#if __arm__ || __arm64__
+           perfcontrol_failsafe_active, 0, 0);
+#else /* __arm__ || __arm64__ */
+           0, 0, 0);
+#endif /* ! __arm__ || __arm64__ */
 
 
+       if (__builtin_popcountll(recommended_cores) == 0) {
+               bit_set(recommended_cores, master_processor->cpu_id); /* add boot processor or we hang */
+       }
 
 
-       (void)splsched();                       /* Turn interruptions off */
+       boolean_t pset_newly_recommended = false;
 
 
-       pmsDown();                                      /* Step power down.  Note: interruptions must be disabled for this call */
+       /* First set recommended cores */
+       pset_lock(pset);
+       avail_count = 0;
+       do {
+               nset = processor->processor_set;
+               if (nset != pset) {
+                       pset_unlock(pset);
+                       pset = nset;
+                       pset_newly_recommended = false;
+                       pset_lock(pset);
+               }
 
 
-       while ( (*threadp == THREAD_NULL)                               &&
-                               (*gcount == 0) && (*lcount == 0)        ) {
+               if (bit_test(recommended_cores, processor->cpu_id)) {
+                       processor->is_recommended = TRUE;
+                       if (bit_first(pset->recommended_bitmask) == -1) {
+                               pset_newly_recommended = true;
+                       }
+                       bit_set(pset->recommended_bitmask, processor->cpu_id);
 
 
-               /* check for ASTs while we wait */
-               if (*myast &~ (AST_SCHEDULING | AST_BSD)) {
-                       /* no ASTs for us */
-                       *myast &= AST_NONE;
-                       (void)spllo();
+                       if (processor->state == PROCESSOR_IDLE) {
+                               if (processor != current_processor()) {
+                                       bit_set(needs_exit_idle_mask, processor->cpu_id);
+                               }
+                       }
+                       if (processor->state != PROCESSOR_OFF_LINE) {
+                               avail_count++;
+                       }
+                       if (pset_newly_recommended) {
+                               SCHED(pset_made_schedulable)(processor, pset, false);
+                       }
                }
                }
-               else
-                       machine_idle();
-
-               (void)splsched();
-       }
+       } while ((processor = processor->processor_list) != NULL);
+       pset_unlock(pset);
 
 
-       /*
-        *      This is not a switch statement to avoid the
-        *      bounds checking code in the common case.
-        */
+       /* Now shutdown not recommended cores */
+       processor = processor_list;
        pset = processor->processor_set;
        pset = processor->processor_set;
-       simple_lock(&pset->sched_lock);
 
 
-       pmsStep(0);                                     /* Step up out of idle power, may start timer for next step */
+       pset_lock(pset);
+       do {
+               nset = processor->processor_set;
+               if (nset != pset) {
+                       pset_unlock(pset);
+                       pset = nset;
+                       pset_lock(pset);
+               }
 
 
-       state = processor->state;
-       if (state == PROCESSOR_DISPATCHING) {
-               /*
-                *      Commmon case -- cpu dispatched.
-                */
-               new_thread = *threadp;
-               *threadp = (volatile thread_t) THREAD_NULL;
-               processor->state = PROCESSOR_RUNNING;
-               enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
-
-               if (    pset->runq.highq >= BASEPRI_RTQUEUES                    &&
-                               new_thread->sched_pri >= BASEPRI_RTQUEUES               ) {
-                       register run_queue_t    runq = &pset->runq;
-                       register queue_t                q;
-
-                       q = runq->queues + runq->highq;
-                       if (((thread_t)q->next)->realtime.deadline <
-                                                                                       processor->deadline) {
-                               thread_t        thread = new_thread;
-
-                               new_thread = (thread_t)q->next;
-                               ((queue_entry_t)new_thread)->next->prev = q;
-                               q->next = ((queue_entry_t)new_thread)->next;
-                               new_thread->runq = RUN_QUEUE_NULL;
-                               processor->deadline = new_thread->realtime.deadline;
-                               assert(new_thread->sched_mode & TH_MODE_PREEMPT);
-                               runq->count--; runq->urgency--;
-                               if (queue_empty(q)) {
-                                       if (runq->highq != IDLEPRI)
-                                               clrbit(MAXPRI - runq->highq, runq->bitmap);
-                                       runq->highq = MAXPRI - ffsbit(runq->bitmap);
-                               }
-                               dispatch_counts.missed_realtime++;
-                               simple_unlock(&pset->sched_lock);
+               if (!bit_test(recommended_cores, processor->cpu_id)) {
+                       sched_ipi_type_t ipi_type = SCHED_IPI_NONE;
 
 
-                               thread_lock(thread);
-                               thread_setrun(thread, SCHED_HEADQ);
-                               thread_unlock(thread);
+                       processor->is_recommended = FALSE;
+                       bit_clear(pset->recommended_bitmask, processor->cpu_id);
 
 
-                               counter(c_idle_thread_handoff++);
-                               thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
-                               /*NOTREACHED*/
+                       if ((processor->state == PROCESSOR_RUNNING) || (processor->state == PROCESSOR_DISPATCHING)) {
+                               ipi_type = SCHED_IPI_IMMEDIATE;
                        }
                        }
-                       simple_unlock(&pset->sched_lock);
-
-                       counter(c_idle_thread_handoff++);
-                       thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
-                       /*NOTREACHED*/
-               }
+                       SCHED(processor_queue_shutdown)(processor);
+                       /* pset unlocked */
 
 
-               if (    processor->runq.highq > new_thread->sched_pri           ||
-                               pset->runq.highq > new_thread->sched_pri                                ) {
-                       thread_t        thread = new_thread;
+                       SCHED(rt_queue_shutdown)(processor);
 
 
-                       new_thread = choose_thread(pset, processor);
-                       dispatch_counts.missed_other++;
-                       simple_unlock(&pset->sched_lock);
-
-                       thread_lock(thread);
-                       thread_setrun(thread, SCHED_HEADQ);
-                       thread_unlock(thread);
+                       if (ipi_type != SCHED_IPI_NONE) {
+                               if (processor == current_processor()) {
+                                       ast_on(AST_PREEMPT);
+                               } else {
+                                       sched_ipi_perform(processor, ipi_type);
+                               }
+                       }
 
 
-                       counter(c_idle_thread_handoff++);
-                       thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
-                       /* NOTREACHED */
+                       pset_lock(pset);
                }
                }
-               else {
-                       simple_unlock(&pset->sched_lock);
+       } while ((processor = processor->processor_list) != NULL);
 
 
-                       counter(c_idle_thread_handoff++);
-                       thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
-                       /* NOTREACHED */
-               }
-       }
-       else
-       if (state == PROCESSOR_IDLE) {
-               /*
-                *      Processor was not dispatched (Rare).
-                *      Set it running again and force a
-                *      reschedule.
-                */
-               no_dispatch_count++;
-               pset->idle_count--;
-               remqueue(&pset->idle_queue, (queue_entry_t)processor);
-               processor->state = PROCESSOR_RUNNING;
-               enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
-               simple_unlock(&pset->sched_lock);
-
-               counter(c_idle_thread_block++);
-               thread_block((thread_continue_t)idle_thread);
-               /* NOTREACHED */
-       }
-       else
-       if (state == PROCESSOR_SHUTDOWN) {
-               /*
-                *      Going off-line.  Force a
-                *      reschedule.
-                */
-               if ((new_thread = (thread_t)*threadp) != THREAD_NULL) {
-                       *threadp = (volatile thread_t) THREAD_NULL;
-                       processor->deadline = UINT64_MAX;
-                       simple_unlock(&pset->sched_lock);
+       processor_avail_count_user = avail_count;
+#if defined(__x86_64__)
+       commpage_update_active_cpus();
+#endif
 
 
-                       thread_lock(new_thread);
-                       thread_setrun(new_thread, SCHED_HEADQ);
-                       thread_unlock(new_thread);
-               }
-               else
-                       simple_unlock(&pset->sched_lock);
+       pset_unlock(pset);
 
 
-               counter(c_idle_thread_block++);
-               thread_block((thread_continue_t)idle_thread);
-               /* NOTREACHED */
+       /* Issue all pending IPIs now that the pset lock has been dropped */
+       for (int cpuid = lsb_first(needs_exit_idle_mask); cpuid >= 0; cpuid = lsb_next(needs_exit_idle_mask, cpuid)) {
+               processor = processor_array[cpuid];
+               machine_signal_idle(processor);
        }
 
        }
 
-       simple_unlock(&pset->sched_lock);
-
-       panic("idle_thread: state %d\n", processor->state);
-       /*NOTREACHED*/
+       KDBG(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_UPDATE_REC_CORES) | DBG_FUNC_END,
+           needs_exit_idle_mask, 0, 0, 0);
 }
 
 }
 
-kern_return_t
-idle_thread_create(
-       processor_t             processor)
+void
+thread_set_options(uint32_t thopt)
 {
 {
-       kern_return_t   result;
-       thread_t                thread;
-       spl_t                   s;
-
-       result = kernel_thread_create((thread_continue_t)idle_thread, NULL, MAXPRI_KERNEL, &thread);
-       if (result != KERN_SUCCESS)
-               return (result);
+       spl_t x;
+       thread_t t = current_thread();
 
 
-       s = splsched();
-       thread_lock(thread);
-       thread->bound_processor = processor;
-       processor->idle_thread = thread;
-       thread->sched_pri = thread->priority = IDLEPRI;
-       thread->state = (TH_RUN | TH_IDLE);
-       thread_unlock(thread);
-       splx(s);
+       x = splsched();
+       thread_lock(t);
 
 
-       thread_deallocate(thread);
+       t->options |= thopt;
 
 
-       return (KERN_SUCCESS);
+       thread_unlock(t);
+       splx(x);
 }
 
 }
 
-static uint64_t                sched_tick_deadline;
-
-/*
- * sched_startup:
- *
- * Kicks off scheduler services.
- *
- * Called at splsched.
- */
 void
 void
-sched_startup(void)
+thread_set_pending_block_hint(thread_t thread, block_hint_t block_hint)
 {
 {
-       kern_return_t   result;
-       thread_t                thread;
-
-       result = kernel_thread_start_priority((thread_continue_t)sched_tick_thread, NULL, MAXPRI_KERNEL, &thread);
-       if (result != KERN_SUCCESS)
-               panic("sched_startup");
+       thread->pending_block_hint = block_hint;
+}
 
 
-       thread_deallocate(thread);
+uint32_t
+qos_max_parallelism(int qos, uint64_t options)
+{
+       return SCHED(qos_max_parallelism)(qos, options);
+}
 
 
-       /*
-        * Yield to the sched_tick_thread while it times
-        * a series of context switches back.  It stores
-        * the baseline value in sched_cswtime.
-        *
-        * The current thread is the only other thread
-        * active at this point.
-        */
-       while (sched_cswtime == 0)
-               thread_block(THREAD_CONTINUE_NULL);
+uint32_t
+sched_qos_max_parallelism(__unused int qos, uint64_t options)
+{
+       host_basic_info_data_t hinfo;
+       mach_msg_type_number_t count = HOST_BASIC_INFO_COUNT;
+       /* Query the machine layer for core information */
+       __assert_only kern_return_t kret = host_info(host_self(), HOST_BASIC_INFO,
+           (host_info_t)&hinfo, &count);
+       assert(kret == KERN_SUCCESS);
+
+       if (options & QOS_PARALLELISM_COUNT_LOGICAL) {
+               return hinfo.logical_cpu;
+       } else {
+               return hinfo.physical_cpu;
+       }
+}
 
 
-       thread_daemon_init();
+int sched_allow_NO_SMT_threads = 1;
+bool
+thread_no_smt(thread_t thread)
+{
+#if DEBUG || DEVELOPMENT
+       return sched_allow_NO_SMT_threads && (thread->bound_processor == PROCESSOR_NULL) && ((thread->sched_flags & TH_SFLAG_NO_SMT) || (thread->task->t_flags & TF_NO_SMT));
+#else
+       return sched_allow_NO_SMT_threads && (thread->bound_processor == PROCESSOR_NULL) && (thread->sched_flags & TH_SFLAG_NO_SMT);
+#endif
+}
 
 
-       thread_call_initialize();
+bool
+processor_active_thread_no_smt(processor_t processor)
+{
+       return sched_allow_NO_SMT_threads && !processor->current_is_bound && processor->current_is_NO_SMT;
 }
 
 }
 
+#if __arm64__
+
 /*
 /*
- *     sched_tick_thread:
+ * Set up or replace old timer with new timer
  *
  *
- *     Perform periodic bookkeeping functions about ten
- *     times per second.
+ * Returns true if canceled old timer, false if it did not
  */
  */
-static void
-sched_tick_continue(void)
+boolean_t
+sched_perfcontrol_update_callback_deadline(uint64_t new_deadline)
 {
 {
-       uint64_t                        abstime = mach_absolute_time();
-
-       sched_tick++;
-
        /*
        /*
-        *  Compute various averages.
+        * Exchange deadline for new deadline, if old deadline was nonzero,
+        * then I cancelled the callback, otherwise I didn't
         */
         */
-       compute_averages();
 
 
-       /*
-        *  Scan the run queues for threads which
-        *  may need to be updated.
-        */
-       thread_update_scan();
+       return os_atomic_xchg(&sched_perfcontrol_callback_deadline, new_deadline,
+                  relaxed) != 0;
+}
 
 
-       clock_deadline_for_periodic_event(sched_tick_interval, abstime,
-                                                                                                               &sched_tick_deadline);
+#endif /* __arm64__ */
 
 
-       assert_wait_deadline((event_t)sched_tick_thread, THREAD_UNINT, sched_tick_deadline);
-       thread_block((thread_continue_t)sched_tick_continue);
-       /*NOTREACHED*/
+void
+sched_update_pset_load_average(processor_set_t pset)
+{
+#if CONFIG_SCHED_CLUTCH
+       int non_rt_load = sched_clutch_root_count(&pset->pset_clutch_root);
+#else /* CONFIG_SCHED_CLUTCH */
+       int non_rt_load = pset->pset_runq.count;
+#endif /* CONFIG_SCHED_CLUTCH */
+
+       int load = ((bit_count(pset->cpu_state_map[PROCESSOR_RUNNING]) + non_rt_load + rt_runq_count(pset)) << PSET_LOAD_NUMERATOR_SHIFT);
+       int new_load_average = (pset->load_average + load) >> 1;
+
+       pset->load_average = new_load_average;
+
+#if (DEVELOPMENT || DEBUG)
+#if __AMP__
+       if (pset->pset_cluster_type == PSET_AMP_P) {
+               KDBG(MACHDBG_CODE(DBG_MACH_SCHED, MACH_PSET_LOAD_AVERAGE) | DBG_FUNC_NONE, sched_get_pset_load_average(pset), (bit_count(pset->cpu_state_map[PROCESSOR_RUNNING]) + pset->pset_runq.count + rt_runq_count(pset)));
+       }
+#endif
+#endif
 }
 
 }
 
-/*
- * Time a series of context switches to determine
- * a baseline.  Toss the high and low and return
- * the one-way value.
- */
-static uint32_t
-time_cswitch(void)
+/* pset is locked */
+static processor_t
+choose_processor_for_realtime_thread(processor_set_t pset)
 {
 {
-       uint32_t        new, hi, low, accum;
-       uint64_t        abstime;
-       int                     i, tries = 7;
+#if defined(__x86_64__)
+       bool avoid_cpu0 = sched_avoid_cpu0 && bit_test(pset->cpu_bitmask, 0);
+#else
+       const bool avoid_cpu0 = false;
+#endif
+
+       uint64_t cpu_map = (pset->cpu_bitmask & pset->recommended_bitmask & ~pset->pending_AST_URGENT_cpu_mask);
+       if (avoid_cpu0) {
+               cpu_map = bit_ror64(cpu_map, 1);
+       }
 
 
-       accum = hi = low = 0;
-       for (i = 0; i < tries; ++i) {
-               abstime = mach_absolute_time();
-               thread_block(THREAD_CONTINUE_NULL);
+       for (int rotid = lsb_first(cpu_map); rotid >= 0; rotid = lsb_next(cpu_map, rotid)) {
+               int cpuid = avoid_cpu0 ?  ((rotid + 1) & 63) : rotid;
 
 
-               new = mach_absolute_time() - abstime;
+               processor_t processor = processor_array[cpuid];
 
 
-               if (i == 0)
-                       accum = hi = low = new;
-               else {
-                       if (new < low)
-                               low = new;
-                       else
-                       if (new > hi)
-                               hi = new;
-                       accum += new;
+               if (processor->processor_primary != processor) {
+                       continue;
                }
                }
-       }
 
 
-       return ((accum - hi - low) / (2 * (tries - 2)));
-}
+               if (processor->state == PROCESSOR_IDLE) {
+                       return processor;
+               }
 
 
-void
-sched_tick_thread(void)
-{
-       sched_cswtime = time_cswitch();
+               if ((processor->state != PROCESSOR_RUNNING) && (processor->state != PROCESSOR_DISPATCHING)) {
+                       continue;
+               }
 
 
-       sched_tick_deadline = mach_absolute_time();
+               if (processor->current_pri >= BASEPRI_RTQUEUES) {
+                       continue;
+               }
 
 
-       sched_tick_continue();
-       /*NOTREACHED*/
-}
+               return processor;
+       }
 
 
-/*
- *     thread_update_scan / runq_scan:
- *
- *     Scan the run queues to account for timesharing threads 
- *     which need to be updated.
- *
- *     Scanner runs in two passes.  Pass one squirrels likely
- *     threads away in an array, pass two does the update.
- *
- *     This is necessary because the run queue is locked for
- *     the candidate scan, but the thread is locked for the update.
- *
- *     Array should be sized to make forward progress, without
- *     disabling preemption for long periods.
- */
+       if (!sched_allow_rt_smt) {
+               return PROCESSOR_NULL;
+       }
 
 
-#define        THREAD_UPDATE_SIZE              128
+       /* Consider secondary processors */
+       if (avoid_cpu0) {
+               /* Also avoid cpu1 */
+               cpu_map = bit_ror64(cpu_map, 1);
+       }
+       for (int rotid = lsb_first(cpu_map); rotid >= 0; rotid = lsb_next(cpu_map, rotid)) {
+               int cpuid = avoid_cpu0 ?  ((rotid + 2) & 63) : rotid;
 
 
-static thread_t                thread_update_array[THREAD_UPDATE_SIZE];
-static int                     thread_update_count = 0;
+               processor_t processor = processor_array[cpuid];
 
 
-/*
- *     Scan a runq for candidate threads.
- *
- *     Returns TRUE if retry is needed.
- */
-static boolean_t
-runq_scan(
-       run_queue_t                             runq)
-{
-       register int                    count;
-       register queue_t                q;
-       register thread_t               thread;
-
-       if ((count = runq->count) > 0) {
-           q = runq->queues + runq->highq;
-               while (count > 0) {
-                       queue_iterate(q, thread, thread_t, links) {
-                               if (            thread->sched_stamp != sched_tick               &&
-                                               (thread->sched_mode & TH_MODE_TIMESHARE)        ) {
-                                       if (thread_update_count == THREAD_UPDATE_SIZE)
-                                               return (TRUE);
-
-                                       thread_update_array[thread_update_count++] = thread;
-                                       thread_reference_internal(thread);
-                               }
+               if (processor->processor_primary == processor) {
+                       continue;
+               }
 
 
-                               count--;
-                       }
+               if (processor->state == PROCESSOR_IDLE) {
+                       return processor;
+               }
+
+               if ((processor->state != PROCESSOR_RUNNING) && (processor->state != PROCESSOR_DISPATCHING)) {
+                       continue;
+               }
 
 
-                       q--;
+               if (processor->current_pri >= BASEPRI_RTQUEUES) {
+                       continue;
                }
                }
+
+               return processor;
        }
 
        }
 
-       return (FALSE);
+       return PROCESSOR_NULL;
 }
 
 }
 
-static void
-thread_update_scan(void)
+/* pset is locked */
+static bool
+all_available_primaries_are_running_realtime_threads(processor_set_t pset)
 {
 {
-       register boolean_t                      restart_needed;
-       register processor_set_t        pset = &default_pset;
-       register processor_t            processor;
-       register thread_t                       thread;
-       spl_t                                           s;
+       return these_processors_are_running_realtime_threads(pset, pset->primary_map);
+}
 
 
-       do {
-               s = splsched();
-               simple_lock(&pset->sched_lock);
-           restart_needed = runq_scan(&pset->runq);
-               simple_unlock(&pset->sched_lock);
-
-               if (!restart_needed) {
-                       simple_lock(&pset->sched_lock);
-                       processor = (processor_t)queue_first(&pset->processors);
-                       while (!queue_end(&pset->processors, (queue_entry_t)processor)) {
-                               if ((restart_needed = runq_scan(&processor->runq)) != 0)
-                                       break;
-
-                               thread = processor->idle_thread;
-                               if (thread->sched_stamp != sched_tick) {
-                                       if (thread_update_count == THREAD_UPDATE_SIZE) {
-                                               restart_needed = TRUE;
-                                               break;
-                                       }
+/* pset is locked */
+static bool
+these_processors_are_running_realtime_threads(processor_set_t pset, uint64_t these_map)
+{
+       uint64_t cpu_map = (pset->cpu_bitmask & pset->recommended_bitmask) & these_map;
 
 
-                                       thread_update_array[thread_update_count++] = thread;
-                                       thread_reference_internal(thread);
-                               }
+       for (int cpuid = lsb_first(cpu_map); cpuid >= 0; cpuid = lsb_next(cpu_map, cpuid)) {
+               processor_t processor = processor_array[cpuid];
 
 
-                               processor = (processor_t)queue_next(&processor->processors);
-                       }
-                       simple_unlock(&pset->sched_lock);
+               if (processor->state == PROCESSOR_IDLE) {
+                       return false;
                }
                }
-               splx(s);
 
 
-           /*
-            *  Ok, we now have a collection of candidates -- fix them.
-            */
-           while (thread_update_count > 0) {
-                       thread = thread_update_array[--thread_update_count];
-                       thread_update_array[thread_update_count] = THREAD_NULL;
+               if (processor->state == PROCESSOR_DISPATCHING) {
+                       return false;
+               }
 
 
-                       s = splsched();
-                       thread_lock(thread);
-                       if (    !(thread->state & (TH_WAIT|TH_SUSP))    &&
-                                               thread->sched_stamp != sched_tick       )
-                               update_priority(thread);
-                       thread_unlock(thread);
-                       splx(s);
+               if (processor->state != PROCESSOR_RUNNING) {
+                       /*
+                        * All other processor states are considered unavailable to run
+                        * realtime threads.  In particular, we prefer an available secondary
+                        * processor over the risk of leaving a realtime thread on the run queue
+                        * while waiting for a processor in PROCESSOR_START state,
+                        * which should anyway be a rare case.
+                        */
+                       continue;
+               }
 
 
-                       thread_deallocate(thread);
-           }
-       } while (restart_needed);
+               if (processor->current_pri < BASEPRI_RTQUEUES) {
+                       return false;
+               }
+       }
+
+       return true;
 }
 }
-               
-/*
- *     Just in case someone doesn't use the macro
- */
-#undef thread_wakeup
+
+static bool
+sched_ok_to_run_realtime_thread(processor_set_t pset, processor_t processor)
+{
+       bool ok_to_run_realtime_thread = true;
+#if defined(__x86_64__)
+       if (sched_avoid_cpu0 && processor->cpu_id == 0) {
+               ok_to_run_realtime_thread = these_processors_are_running_realtime_threads(pset, pset->primary_map & ~0x1);
+       } else if (sched_avoid_cpu0 && (processor->cpu_id == 1) && processor->is_SMT) {
+               ok_to_run_realtime_thread = sched_allow_rt_smt && these_processors_are_running_realtime_threads(pset, ~0x2);
+       } else if (processor->processor_primary != processor) {
+               ok_to_run_realtime_thread = sched_allow_rt_smt && all_available_primaries_are_running_realtime_threads(pset);
+       }
+#else
+       (void)pset;
+       (void)processor;
+#endif
+       return ok_to_run_realtime_thread;
+}
+
 void
 void
-thread_wakeup(
-       event_t         x);
+sched_pset_made_schedulable(__unused processor_t processor, processor_set_t pset, boolean_t drop_lock)
+{
+       if (drop_lock) {
+               pset_unlock(pset);
+       }
+}
 
 void
 
 void
-thread_wakeup(
-       event_t         x)
+thread_set_no_smt(bool set)
 {
 {
-       thread_wakeup_with_result(x, THREAD_AWAKENED);
+       thread_t thread = current_thread();
+
+       spl_t s = splsched();
+       thread_lock(thread);
+       if (set) {
+               thread->sched_flags |= TH_SFLAG_NO_SMT;
+       } else {
+               thread->sched_flags &= ~TH_SFLAG_NO_SMT;
+       }
+       thread_unlock(thread);
+       splx(s);
 }
 
 }
 
-boolean_t
-preemption_enabled(void)
+bool
+thread_get_no_smt(void)
 {
 {
-       return (get_preemption_level() == 0 && ml_get_interrupts_enabled());
+       return current_thread()->sched_flags & TH_SFLAG_NO_SMT;
 }
 
 }
 
-#if    DEBUG
-static boolean_t
-thread_runnable(
-       thread_t        thread)
+#if DEBUG || DEVELOPMENT
+extern void sysctl_task_set_no_smt(char no_smt);
+void
+sysctl_task_set_no_smt(char no_smt)
 {
 {
-       return ((thread->state & (TH_RUN|TH_WAIT)) == TH_RUN);
+       thread_t thread = current_thread();
+       task_t task = thread->task;
+
+       if (no_smt == '1') {
+               task->t_flags |= TF_NO_SMT;
+       } else {
+               task->t_flags &= ~TF_NO_SMT;
+       }
 }
 }
-#endif /* DEBUG */
 
 
-#if    MACH_KDB
-#include <ddb/db_output.h>
-#define        printf          kdbprintf
-void                   db_sched(void);
+extern char sysctl_task_get_no_smt(void);
+char
+sysctl_task_get_no_smt(void)
+{
+       thread_t thread = current_thread();
+       task_t task = thread->task;
 
 
-void
-db_sched(void)
-{
-       iprintf("Scheduling Statistics:\n");
-       db_indent += 2;
-       iprintf("Thread invocations:  csw %d same %d\n",
-               c_thread_invoke_csw, c_thread_invoke_same);
-#if    MACH_COUNTERS
-       iprintf("Thread block:  calls %d\n",
-               c_thread_block_calls);
-       iprintf("Idle thread:\n\thandoff %d block %d no_dispatch %d\n",
-               c_idle_thread_handoff,
-               c_idle_thread_block, no_dispatch_count);
-       iprintf("Sched thread blocks:  %d\n", c_sched_thread_block);
-#endif /* MACH_COUNTERS */
-       db_indent -= 2;
-}
-
-#include <ddb/db_output.h>
-void           db_show_thread_log(void);
+       if (task->t_flags & TF_NO_SMT) {
+               return '1';
+       }
+       return '0';
+}
+#endif /* DEVELOPMENT || DEBUG */
 
 
-void
-db_show_thread_log(void)
+
+__private_extern__ void
+thread_bind_cluster_type(char cluster_type)
 {
 {
+#if __AMP__
+       thread_t thread = current_thread();
+
+       spl_t s = splsched();
+       thread_lock(thread);
+       thread->sched_flags &= ~(TH_SFLAG_ECORE_ONLY | TH_SFLAG_PCORE_ONLY);
+       switch (cluster_type) {
+       case 'e':
+       case 'E':
+               thread->sched_flags |= TH_SFLAG_ECORE_ONLY;
+               break;
+       case 'p':
+       case 'P':
+               thread->sched_flags |= TH_SFLAG_PCORE_ONLY;
+               break;
+       default:
+               break;
+       }
+       thread_unlock(thread);
+       splx(s);
+
+       thread_block(THREAD_CONTINUE_NULL);
+#else /* __AMP__ */
+       (void)cluster_type;
+#endif /* __AMP__ */
 }
 }
-#endif /* MACH_KDB */