]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/sched_prim.c
xnu-1699.26.8.tar.gz
[apple/xnu.git] / osfmk / kern / sched_prim.c
index aaece698a70baf3bdaa5028468f59effe968a206..d7b959249aabd874f45413858fbcd9df123ee79c 100644 (file)
@@ -1,23 +1,29 @@
 /*
- * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2009 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License").  You may not use this file except in compliance with the
- * License.  Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * 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
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * 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.
  * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * 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,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE 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_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_FREE_COPYRIGHT@
 #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/machlimits.h>
 
 #include <kern/kern_types.h>
 #include <kern/clock.h>
 #include <vm/vm_kern.h>
 #include <vm/vm_map.h>
 
+#include <mach/sdt.h>
+
 #include <sys/kdebug.h>
 
-#ifdef __ppc__
-#include <ppc/pms.h>
+#include <kern/pms.h>
+
+struct rt_queue        rt_runq;
+#define RT_RUNQ                ((processor_t)-1)
+decl_simple_lock_data(static,rt_lock);
+
+#if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_PROTO) || defined(CONFIG_SCHED_GRRR) || defined(CONFIG_SCHED_FIXEDPRIORITY)
+static struct fairshare_queue  fs_runq;
+#define FS_RUNQ                ((processor_t)-2)
+decl_simple_lock_data(static,fs_lock);
 #endif
 
 #define                DEFAULT_PREEMPTION_RATE         100             /* (1/s) */
@@ -112,41 +131,210 @@ 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_unsafe_computation;
-uint32_t       sched_safe_duration;
 uint64_t       max_poll_computation;
 
+uint64_t       max_unsafe_computation;
+uint64_t       sched_safe_duration;
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
 uint32_t       std_quantum;
 uint32_t       min_std_quantum;
 
 uint32_t       std_quantum_us;
 
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
+uint32_t       thread_depress_time;
+uint32_t       default_timeshare_computation;
+uint32_t       default_timeshare_constraint;
+
 uint32_t       max_rt_quantum;
 uint32_t       min_rt_quantum;
 
 uint32_t       sched_cswtime;
 
-static uint32_t                delay_idle_limit, delay_idle_spin;
-static processor_t     delay_idle(
-                                               processor_t             processor,
-                                               thread_t                self);
+#if defined(CONFIG_SCHED_TRADITIONAL)
 
 unsigned       sched_tick;
 uint32_t       sched_tick_interval;
 
-uint32_t       sched_pri_shift;
+uint32_t       sched_pri_shift = INT8_MAX;
+uint32_t       sched_fixed_shift;
+
+static boolean_t sched_traditional_use_pset_runqueue = FALSE;
+
+__attribute__((always_inline))
+static inline run_queue_t runq_for_processor(processor_t processor)
+{
+       if (sched_traditional_use_pset_runqueue)
+               return &processor->processor_set->pset_runq;
+       else
+               return &processor->runq;
+}
+
+__attribute__((always_inline))
+static inline void runq_consider_incr_bound_count(processor_t processor, thread_t thread)
+{
+       if (thread->bound_processor == PROCESSOR_NULL)
+               return;
+    
+       assert(thread->bound_processor == processor);
+    
+       if (sched_traditional_use_pset_runqueue)
+               processor->processor_set->pset_runq_bound_count++;
+    
+       processor->runq_bound_count++;
+}
+
+__attribute__((always_inline))
+static inline void runq_consider_decr_bound_count(processor_t processor, thread_t thread)
+{
+       if (thread->bound_processor == PROCESSOR_NULL)
+               return;
+    
+       assert(thread->bound_processor == processor);
+    
+       if (sched_traditional_use_pset_runqueue)
+               processor->processor_set->pset_runq_bound_count--;
+    
+       processor->runq_bound_count--;
+}
+
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
+uint64_t       sched_one_second_interval;
+
+uint32_t       sched_run_count, sched_share_count;
+uint32_t       sched_load_average, sched_mach_factor;
 
 /* Forwards */
-void           wait_queues_init(void);
 
-static void            load_shift_init(void);
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
+static void load_shift_init(void) __attribute__((section("__TEXT, initcode")));
+static void preempt_pri_init(void) __attribute__((section("__TEXT, initcode")));
+
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
+static thread_t        thread_select(
+                                       thread_t                        thread,
+                                       processor_t                     processor);
+
+#if CONFIG_SCHED_IDLE_IN_PLACE
+static thread_t        thread_select_idle(
+                                       thread_t                        thread,
+                                       processor_t                     processor);
+#endif
+
+thread_t       processor_idle(
+                                       thread_t                        thread,
+                                       processor_t                     processor);
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
+static thread_t        steal_thread(
+                                       processor_set_t         pset);
+
+static thread_t        steal_thread_disabled(
+                                       processor_set_t         pset) __attribute__((unused));
 
-static thread_t        choose_thread(
-                                       processor_set_t         pset,
+
+static thread_t        steal_processor_thread(
                                        processor_t                     processor);
 
 static void            thread_update_scan(void);
 
+static void processor_setrun(
+                                processor_t                    processor,
+                                thread_t                       thread,
+                                integer_t                      options);
+
+static boolean_t
+processor_enqueue(
+                                 processor_t           processor,
+                                 thread_t              thread,
+                                 integer_t             options);
+
+static boolean_t
+processor_queue_remove(
+                                          processor_t                  processor,
+                                          thread_t             thread);
+
+static boolean_t       processor_queue_empty(processor_t               processor);
+
+static boolean_t       priority_is_urgent(int priority);
+
+static ast_t           processor_csw_check(processor_t processor);
+
+static boolean_t       processor_queue_has_priority(processor_t                processor,
+                                                                                       int                             priority,
+                                                                                       boolean_t               gte);
+
+static boolean_t       should_current_thread_rechoose_processor(processor_t                    processor);
+
+static int     sched_traditional_processor_runq_count(processor_t   processor);
+
+static boolean_t       sched_traditional_with_pset_runqueue_processor_queue_empty(processor_t          processor);
+
+static uint64_t     sched_traditional_processor_runq_stats_count_sum(processor_t   processor);
+
+static uint64_t                sched_traditional_with_pset_runqueue_processor_runq_stats_count_sum(processor_t   processor);
+#endif
+       
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
+static void
+sched_traditional_init(void);
+
+static void
+sched_traditional_timebase_init(void);
+
+static void
+sched_traditional_processor_init(processor_t processor);
+
+static void
+sched_traditional_pset_init(processor_set_t pset);
+
+static void
+sched_traditional_with_pset_runqueue_init(void);
+
+#endif
+
+static void
+sched_realtime_init(void)  __attribute__((section("__TEXT, initcode")));
+
+static void
+sched_realtime_timebase_init(void);
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+static void
+sched_traditional_tick_continue(void);
+
+static uint32_t
+sched_traditional_initial_quantum_size(thread_t thread);
+
+static sched_mode_t
+sched_traditional_initial_thread_sched_mode(task_t parent_task);
+
+static boolean_t
+sched_traditional_supports_timeshare_mode(void);
+
+static thread_t
+sched_traditional_choose_thread(
+                                                               processor_t             processor,
+                                                               int                             priority);
+
+#endif
+
+#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
+
 #if    DEBUG
 static
 boolean_t      thread_runnable(
@@ -154,7 +342,6 @@ boolean_t   thread_runnable(
 
 #endif /*DEBUG*/
 
-
 /*
  *     State machine
  *
@@ -181,43 +368,221 @@ boolean_t        thread_runnable(
  *
  */
 
+#if defined(CONFIG_SCHED_TRADITIONAL)
+int8_t         sched_load_shifts[NRQS];
+int            sched_preempt_pri[NRQBM];
+#endif
+
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
+const struct sched_dispatch_table sched_traditional_dispatch = {
+       sched_traditional_init,
+       sched_traditional_timebase_init,
+       sched_traditional_processor_init,
+       sched_traditional_pset_init,
+       sched_traditional_tick_continue,
+       sched_traditional_choose_thread,
+       steal_thread,
+       compute_priority,
+       choose_processor,
+       processor_enqueue,
+       processor_queue_shutdown,
+       processor_queue_remove,
+       processor_queue_empty,
+       priority_is_urgent,
+       processor_csw_check,
+       processor_queue_has_priority,
+       sched_traditional_initial_quantum_size,
+       sched_traditional_initial_thread_sched_mode,
+       sched_traditional_supports_timeshare_mode,
+       can_update_priority,
+       update_priority,
+       lightweight_update_priority,
+       sched_traditional_quantum_expire,
+       should_current_thread_rechoose_processor,
+       sched_traditional_processor_runq_count,
+       sched_traditional_processor_runq_stats_count_sum,
+       sched_traditional_fairshare_init,
+       sched_traditional_fairshare_runq_count,
+       sched_traditional_fairshare_runq_stats_count_sum,
+       sched_traditional_fairshare_enqueue,
+       sched_traditional_fairshare_dequeue,
+       sched_traditional_fairshare_queue_remove,
+       TRUE /* direct_dispatch_to_idle_processors */
+};
+
+const struct sched_dispatch_table sched_traditional_with_pset_runqueue_dispatch = {
+       sched_traditional_with_pset_runqueue_init,
+       sched_traditional_timebase_init,
+       sched_traditional_processor_init,
+       sched_traditional_pset_init,
+       sched_traditional_tick_continue,
+       sched_traditional_choose_thread,
+       steal_thread,
+       compute_priority,
+       choose_processor,
+       processor_enqueue,
+       processor_queue_shutdown,
+       processor_queue_remove,
+       sched_traditional_with_pset_runqueue_processor_queue_empty,
+       priority_is_urgent,
+       processor_csw_check,
+       processor_queue_has_priority,
+       sched_traditional_initial_quantum_size,
+       sched_traditional_initial_thread_sched_mode,
+       sched_traditional_supports_timeshare_mode,
+       can_update_priority,
+       update_priority,
+       lightweight_update_priority,
+       sched_traditional_quantum_expire,
+       should_current_thread_rechoose_processor,
+       sched_traditional_processor_runq_count,
+       sched_traditional_with_pset_runqueue_processor_runq_stats_count_sum,
+       sched_traditional_fairshare_init,
+       sched_traditional_fairshare_runq_count,
+       sched_traditional_fairshare_runq_stats_count_sum,
+       sched_traditional_fairshare_enqueue,
+       sched_traditional_fairshare_dequeue,
+       sched_traditional_fairshare_queue_remove,
+       FALSE /* direct_dispatch_to_idle_processors */
+};
+
+#endif
+
+const struct sched_dispatch_table *sched_current_dispatch = NULL;
+
 /*
- *     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().
- *
- *     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.
+ * 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.
  *
- *     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.
  */
+#define SCHED_STRING_MAX_LENGTH (48)
 
-#define NUMQUEUES      59
-
-struct wait_queue wait_queues[NUMQUEUES];
+char sched_string[SCHED_STRING_MAX_LENGTH];
+static enum sched_enum _sched_enum = sched_enum_unknown;
 
-#define wait_hash(event) \
-       ((((int)(event) < 0)? ~(int)(event): (int)(event)) % NUMQUEUES)
+void
+sched_init(void)
+{
+       char sched_arg[SCHED_STRING_MAX_LENGTH] = { '\0' };
+
+       /* Check for runtime selection of the scheduler algorithm */
+       if (!PE_parse_boot_argn("sched", sched_arg, sizeof (sched_arg))) {
+               /* If no boot-args override, look in device tree */
+               if (!PE_get_default("kern.sched", sched_arg,
+                                                       SCHED_STRING_MAX_LENGTH)) {
+                       sched_arg[0] = '\0';
+               }
+       }
 
-int8_t         sched_load_shifts[NRQS];
+       if (strlen(sched_arg) > 0) {
+               if (0) {
+                       /* Allow pattern below */
+#if defined(CONFIG_SCHED_TRADITIONAL)
+               } else if (0 == strcmp(sched_arg, kSchedTraditionalString)) {
+                       sched_current_dispatch = &sched_traditional_dispatch;
+                       _sched_enum = sched_enum_traditional;
+                       strlcpy(sched_string, kSchedTraditionalString, sizeof(sched_string));
+                       kprintf("Scheduler: Runtime selection of %s\n", kSchedTraditionalString);
+               } else if (0 == strcmp(sched_arg, kSchedTraditionalWithPsetRunqueueString)) {
+                       sched_current_dispatch = &sched_traditional_with_pset_runqueue_dispatch;
+                       _sched_enum = sched_enum_traditional_with_pset_runqueue;
+                       strlcpy(sched_string, kSchedTraditionalWithPsetRunqueueString, sizeof(sched_string));
+                       kprintf("Scheduler: Runtime selection of %s\n", kSchedTraditionalWithPsetRunqueueString);
+#endif
+#if defined(CONFIG_SCHED_PROTO)
+               } else if (0 == strcmp(sched_arg, kSchedProtoString)) {
+                       sched_current_dispatch = &sched_proto_dispatch;
+                       _sched_enum = sched_enum_proto;
+                       strlcpy(sched_string, kSchedProtoString, sizeof(sched_string));
+                       kprintf("Scheduler: Runtime selection of %s\n", kSchedProtoString);
+#endif
+#if defined(CONFIG_SCHED_GRRR)
+               } else if (0 == strcmp(sched_arg, kSchedGRRRString)) {
+                       sched_current_dispatch = &sched_grrr_dispatch;
+                       _sched_enum = sched_enum_grrr;
+                       strlcpy(sched_string, kSchedGRRRString, sizeof(sched_string));
+                       kprintf("Scheduler: Runtime selection of %s\n", kSchedGRRRString);
+#endif
+#if defined(CONFIG_SCHED_FIXEDPRIORITY)
+               } else if (0 == strcmp(sched_arg, kSchedFixedPriorityString)) {
+                       sched_current_dispatch = &sched_fixedpriority_dispatch;
+                       _sched_enum = sched_enum_fixedpriority;
+                       strlcpy(sched_string, kSchedFixedPriorityString, sizeof(sched_string));
+                       kprintf("Scheduler: Runtime selection of %s\n", kSchedFixedPriorityString);
+               } else if (0 == strcmp(sched_arg, kSchedFixedPriorityWithPsetRunqueueString)) {
+                       sched_current_dispatch = &sched_fixedpriority_with_pset_runqueue_dispatch;
+                       _sched_enum = sched_enum_fixedpriority_with_pset_runqueue;
+                       strlcpy(sched_string, kSchedFixedPriorityWithPsetRunqueueString, sizeof(sched_string));
+                       kprintf("Scheduler: Runtime selection of %s\n", kSchedFixedPriorityWithPsetRunqueueString);
+#endif
+               } else {
+                       panic("Unrecognized scheduler algorithm: %s", sched_arg);
+               }
+       } else {
+#if   defined(CONFIG_SCHED_TRADITIONAL)
+               sched_current_dispatch = &sched_traditional_dispatch;
+               _sched_enum = sched_enum_traditional;
+               strlcpy(sched_string, kSchedTraditionalString, sizeof(sched_string));
+               kprintf("Scheduler: Default of %s\n", kSchedTraditionalString);
+#elif defined(CONFIG_SCHED_PROTO)
+               sched_current_dispatch = &sched_proto_dispatch;
+               _sched_enum = sched_enum_proto;
+               strlcpy(sched_string, kSchedProtoString, sizeof(sched_string));
+               kprintf("Scheduler: Default of %s\n", kSchedProtoString);
+#elif defined(CONFIG_SCHED_GRRR)
+               sched_current_dispatch = &sched_grrr_dispatch;
+               _sched_enum = sched_enum_grrr;
+               strlcpy(sched_string, kSchedGRRRString, sizeof(sched_string));
+               kprintf("Scheduler: Default of %s\n", kSchedGRRRString);
+#elif defined(CONFIG_SCHED_FIXEDPRIORITY)
+               sched_current_dispatch = &sched_fixedpriority_dispatch;
+               _sched_enum = sched_enum_fixedpriority;
+               strlcpy(sched_string, kSchedFixedPriorityString, sizeof(sched_string));
+               kprintf("Scheduler: Default of %s\n", kSchedFixedPriorityString);
+#else
+#error No default scheduler implementation
+#endif
+       }
+       
+       SCHED(init)();
+       SCHED(fairshare_init)();
+       sched_realtime_init();
+       ast_init();
+       
+       SCHED(pset_init)(&pset0);
+       SCHED(processor_init)(master_processor);
+}
 
 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_TRADITIONAL)
+
+static void
+sched_traditional_init(void)
 {
        /*
         * Calculate the timeslicing quantum
@@ -229,18 +594,13 @@ sched_init(void)
 
        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);
-
-       wait_queues_init();
        load_shift_init();
-       pset_init(&default_pset);
+       preempt_pri_init();
        sched_tick = 0;
-       ast_init();
 }
 
-void
-sched_timebase_init(void)
+static void
+sched_traditional_timebase_init(void)
 {
        uint64_t        abstime;
        uint32_t        shift;
@@ -249,29 +609,18 @@ sched_timebase_init(void)
        clock_interval_to_absolutetime_interval(
                                                        std_quantum_us, NSEC_PER_USEC, &abstime);
        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);
-       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;
-
-       /* maximum rt computation (50 ms) */
-       clock_interval_to_absolutetime_interval(
-                                                       50, 1000*NSEC_PER_USEC, &abstime);
-       assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       max_rt_quantum = abstime;
+       min_std_quantum = (uint32_t)abstime;
 
        /* scheduler tick interval */
        clock_interval_to_absolutetime_interval(USEC_PER_SEC >> SCHED_TICK_SHIFT,
                                                                                                        NSEC_PER_USEC, &abstime);
        assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       sched_tick_interval = abstime;
+       sched_tick_interval = (uint32_t)abstime;
 
        /*
         * Compute conversion factor from usage to
@@ -280,31 +629,85 @@ sched_timebase_init(void)
        abstime = (abstime * 5) / 3;
        for (shift = 0; abstime > BASEPRI_DEFAULT; ++shift)
                abstime >>= 1;
-       sched_pri_shift = shift;
+       sched_fixed_shift = shift;
 
        max_unsafe_computation = max_unsafe_quanta * std_quantum;
+       sched_safe_duration = 2 * max_unsafe_quanta * std_quantum;
+       
        max_poll_computation = max_poll_quanta * std_quantum;
+       thread_depress_time = 1 * std_quantum;
+       default_timeshare_computation = std_quantum / 2;
+       default_timeshare_constraint = std_quantum;
 
-       /* delay idle constant(s) (60, 1 us) */
-       clock_interval_to_absolutetime_interval(60, NSEC_PER_USEC, &abstime);
-       assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       delay_idle_limit = abstime;
+}
 
-       clock_interval_to_absolutetime_interval(1, NSEC_PER_USEC, &abstime);
-       assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
-       delay_idle_spin = abstime;
+static void
+sched_traditional_processor_init(processor_t processor)
+{
+       if (!sched_traditional_use_pset_runqueue) {
+               run_queue_init(&processor->runq);
+       }
+       processor->runq_bound_count = 0;
+}
+
+static void
+sched_traditional_pset_init(processor_set_t pset)
+{
+       if (sched_traditional_use_pset_runqueue) {
+               run_queue_init(&pset->pset_runq);
+       }
+       pset->pset_runq_bound_count = 0;
+}
+
+static void
+sched_traditional_with_pset_runqueue_init(void)
+{
+       sched_traditional_init();
+       sched_traditional_use_pset_runqueue = TRUE;
 }
 
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
+#if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_PROTO) || defined(CONFIG_SCHED_GRRR) || defined(CONFIG_SCHED_FIXEDPRIORITY)
 void
-wait_queues_init(void)
+sched_traditional_fairshare_init(void)
 {
-       register int    i;
+       simple_lock_init(&fs_lock, 0);
+       
+       fs_runq.count = 0;
+       queue_init(&fs_runq.queue);
+}
+#endif
+
+static void
+sched_realtime_init(void)
+{
+       simple_lock_init(&rt_lock, 0);
+
+       rt_runq.count = 0;
+       queue_init(&rt_runq.queue);
+}
+
+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);
+       min_rt_quantum = (uint32_t)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);
+       max_rt_quantum = (uint32_t)abstime;
 
-       for (i = 0; i < NUMQUEUES; i++) {
-               wait_queue_init(&wait_queues[i], SYNC_POLICY_FIFO);
-       }
 }
 
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
 /*
  * Set up values for timeshare
  * loading factors.
@@ -323,6 +726,20 @@ load_shift_init(void)
        }
 }
 
+static void
+preempt_pri_init(void)
+{
+       int             i, *p = sched_preempt_pri;
+
+       for (i = BASEPRI_FOREGROUND + 1; i < MINPRI_KERNEL; ++i)
+               setbit(i, p);
+
+       for (i = BASEPRI_PREEMPT; i <= MAXPRI; ++i)
+               setbit(i, p);
+}
+
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
 /*
  *     Thread wait timer expiration.
  */
@@ -346,6 +763,8 @@ thread_timer_expire(
        splx(s);
 }
 
+#ifndef        __LP64__
+
 /*
  *     thread_set_timer:
  *
@@ -366,7 +785,7 @@ thread_set_timer(
        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))
+               if (!timer_call_enter(&thread->wait_timer, deadline, thread->sched_pri >= BASEPRI_RTQUEUES ? TIMER_CALL_CRITICAL : 0))
                        thread->wait_timer_active++;
                thread->wait_timer_is_set = TRUE;
        }
@@ -384,7 +803,7 @@ thread_set_timer_deadline(
        s = splsched();
        thread_lock(thread);
        if ((thread->state & TH_WAIT) != 0) {
-               if (!timer_call_enter(&thread->wait_timer, deadline))
+               if (!timer_call_enter(&thread->wait_timer, deadline, thread->sched_pri >= BASEPRI_RTQUEUES ? TIMER_CALL_CRITICAL : 0))
                        thread->wait_timer_active++;
                thread->wait_timer_is_set = TRUE;
        }
@@ -409,6 +828,8 @@ thread_cancel_timer(void)
        splx(s);
 }
 
+#endif /* __LP64__ */
+
 /*
  *     thread_unblock:
  *
@@ -426,12 +847,12 @@ thread_unblock(
        boolean_t               result = FALSE;
 
        /*
-        * Set wait_result.
+        *      Set wait_result.
         */
        thread->wait_result = wresult;
 
        /*
-        * Cancel pending wait timer.
+        *      Cancel pending wait timer.
         */
        if (thread->wait_timer_is_set) {
                if (timer_call_cancel(&thread->wait_timer))
@@ -440,33 +861,45 @@ thread_unblock(
        }
 
        /*
-        * Update scheduling state.
+        *      Update scheduling state: not waiting,
+        *      set running.
         */
        thread->state &= ~(TH_WAIT|TH_UNINT);
 
        if (!(thread->state & TH_RUN)) {
                thread->state |= TH_RUN;
 
-               /*
-                * Mark unblocked if call out.
-                */
-               if (thread->options & TH_OPT_CALLOUT)
-                       call_thread_unblock();
+               (*thread->sched_call)(SCHED_CALL_UNBLOCK, thread);
 
                /*
-                * Update pset run counts.
+                *      Update run counts.
                 */
-               pset_run_incr(thread->processor_set);
-               if (thread->sched_mode & TH_MODE_TIMESHARE)
-                       pset_share_incr(thread->processor_set);
+               sched_run_incr();
+               if (thread->sched_mode == TH_MODE_TIMESHARE)
+                       sched_share_incr();
        }
-       else
+       else {
+               /*
+                *      Signal if idling on another processor.
+                */
+#if CONFIG_SCHED_IDLE_IN_PLACE
+               if (thread->state & TH_IDLE) {
+                       processor_t             processor = thread->last_processor;
+
+                       if (processor != current_processor())
+                               machine_signal_idle(processor);
+               }
+#else
+               assert((thread->state & TH_IDLE) == 0);
+#endif
+
                result = TRUE;
+       }
 
        /*
         * Calculate deadline for real-time threads.
         */
-       if (thread->sched_mode & TH_MODE_REALTIME) {
+       if (thread->sched_mode == TH_MODE_REALTIME) {
                thread->realtime.deadline = mach_absolute_time();
                thread->realtime.deadline += thread->realtime.constraint;
        }
@@ -480,7 +913,9 @@ thread_unblock(
 
        KERNEL_DEBUG_CONSTANT(
                MACHDBG_CODE(DBG_MACH_SCHED,MACH_MAKE_RUNNABLE) | DBG_FUNC_NONE,
-                                       (int)thread, (int)thread->sched_pri, 0, 0, 0);
+                                       (uintptr_t)thread_tid(thread), thread->sched_pri, 0, 0, 0);
+
+       DTRACE_SCHED2(wakeup, struct thread *, thread, struct proc *, thread->task->bsd_info);
 
        return (result);
 }
@@ -532,6 +967,8 @@ thread_mark_wait_locked(
 {
        boolean_t               at_safe_point;
 
+       assert(thread == current_thread());
+
        /*
         *      The thread may have certain types of interrupts/aborts masked
         *      off.  Even if the wait location says these types of interrupts
@@ -544,16 +981,19 @@ thread_mark_wait_locked(
        at_safe_point = (interruptible == THREAD_ABORTSAFE);
 
        if (    interruptible == THREAD_UNINT                   ||
-                       !(thread->state & TH_ABORT)                             ||
+                       !(thread->sched_flags & TH_SFLAG_ABORT) ||
                        (!at_safe_point &&
-                        (thread->state & TH_ABORT_SAFELY))             ) {
+                               (thread->sched_flags & TH_SFLAG_ABORTSAFELY))) {
+
+               DTRACE_SCHED(sleep);
+
                thread->state |= (interruptible) ? TH_WAIT : (TH_WAIT | TH_UNINT);
                thread->at_safe_point = at_safe_point;
                return (thread->wait_result = THREAD_WAITING);
        }
        else
-       if (thread->state & TH_ABORT_SAFELY)
-               thread->state &= ~(TH_ABORT|TH_ABORT_SAFELY);
+       if (thread->sched_flags & TH_SFLAG_ABORTSAFELY)
+               thread->sched_flags &= ~TH_SFLAG_ABORTED_MASK;
 
        return (thread->wait_result = THREAD_INTERRUPTED);
 }
@@ -648,7 +1088,7 @@ assert_wait_timeout(
        thread_lock(thread);
 
        clock_interval_to_deadline(interval, scale_factor, &deadline);
-       wresult = wait_queue_assert_wait64_locked(wqueue, (uint32_t)event,
+       wresult = wait_queue_assert_wait64_locked(wqueue, CAST_DOWN(event64_t, event),
                                                                                                        interruptible, deadline, thread);
 
        thread_unlock(thread);
@@ -676,7 +1116,7 @@ assert_wait_deadline(
        wait_queue_lock(wqueue);
        thread_lock(thread);
 
-       wresult = wait_queue_assert_wait64_locked(wqueue, (uint32_t)event,
+       wresult = wait_queue_assert_wait64_locked(wqueue, CAST_DOWN(event64_t,event),
                                                                                                        interruptible, deadline, thread);
 
        thread_unlock(thread);
@@ -741,57 +1181,6 @@ thread_sleep_usimple_lock(
        return res;
 }
 
-/*
- *     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
-thread_sleep_mutex(
-       event_t                 event,
-       mutex_t                 *mutex,
-       wait_interrupt_t interruptible)
-{
-       wait_result_t   res;
-
-       res = assert_wait(event, interruptible);
-       if (res == THREAD_WAITING) {
-               mutex_unlock(mutex);
-               res = thread_block(THREAD_CONTINUE_NULL);
-               mutex_lock(mutex);
-       }
-       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;
-}
-
 /*
  *     thread_sleep_lock_write:
  *
@@ -833,13 +1222,15 @@ thread_stop(
        thread_t                thread)
 {
        wait_result_t   wresult;
-       spl_t                   s;
+       spl_t                   s = splsched();
 
-       s = splsched();
        wake_lock(thread);
+       thread_lock(thread);
 
        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);
@@ -852,21 +1243,20 @@ thread_stop(
 
                s = splsched();
                wake_lock(thread);
+               thread_lock(thread);
        }
 
-       thread_lock(thread);
        thread->state |= TH_SUSP;
 
        while (thread->state & TH_RUN) {
                processor_t             processor = thread->last_processor;
 
-               if (    processor != PROCESSOR_NULL                                     &&
-                               processor->state == PROCESSOR_RUNNING           &&
-                               processor->active_thread == thread                      )
+               if (processor != PROCESSOR_NULL && processor->active_thread == thread)
                        cause_ast_check(processor);
-               thread_unlock(thread);
 
                thread->wake_active = TRUE;
+               thread_unlock(thread);
+
                wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
                wake_unlock(thread);
                splx(s);
@@ -921,10 +1311,11 @@ thread_unstop(
                if (thread->wake_active) {
                        thread->wake_active = FALSE;
                        thread_unlock(thread);
+
+                       thread_wakeup(&thread->wake_active);
                        wake_unlock(thread);
                        splx(s);
 
-                       thread_wakeup(&thread->wake_active);
                        return;
                }
        }
@@ -953,13 +1344,12 @@ thread_wait(
        while (thread->state & TH_RUN) {
                processor_t             processor = thread->last_processor;
 
-               if (    processor != PROCESSOR_NULL                                     &&
-                               processor->state == PROCESSOR_RUNNING           &&
-                               processor->active_thread == thread                      )
+               if (processor != PROCESSOR_NULL && processor->active_thread == thread)
                        cause_ast_check(processor);
-               thread_unlock(thread);
 
                thread->wake_active = TRUE;
+               thread_unlock(thread);
+
                wresult = assert_wait(&thread->wake_active, THREAD_UNINT);
                wake_unlock(thread);
                splx(s);
@@ -999,7 +1389,7 @@ clear_wait_internal(
        wait_result_t   wresult)
 {
        wait_queue_t    wq = thread->wait_queue;
-       int                             i = LockTimeOut;
+       uint32_t        i = LockTimeOut;
 
        do {
                if (wresult == THREAD_INTERRUPTED && (thread->state & TH_UNINT))
@@ -1023,9 +1413,9 @@ clear_wait_internal(
                }
 
                return (thread_go(thread, wresult));
-       } while (--i > 0);
+       } while ((--i > 0) || machine_timeout_suspended());
 
-       panic("clear_wait_internal: deadlock: thread=0x%x, wq=0x%x, cpu=%d\n",
+       panic("clear_wait_internal: deadlock: thread=%p, wq=%p, cpu=%d\n",
                  thread, wq, cpu_number());
 
        return (KERN_FAILURE);
@@ -1070,23 +1460,34 @@ kern_return_t
 thread_wakeup_prim(
        event_t                 event,
        boolean_t               one_thread,
-       wait_result_t   result)
+       wait_result_t           result)
 {
-       register wait_queue_t   wq;
-       register int                    index;
+       return (thread_wakeup_prim_internal(event, one_thread, result, -1));
+}
 
-       index = wait_hash(event);
+
+kern_return_t
+thread_wakeup_prim_internal(
+       event_t                 event,
+       boolean_t               one_thread,
+       wait_result_t           result,
+       int                     priority)
+{
+       register wait_queue_t   wq;
+       register int                    index;
+
+       index = wait_hash(event);
        wq = &wait_queues[index];
        if (one_thread)
-           return (wait_queue_wakeup_one(wq, event, result));
+               return (wait_queue_wakeup_one(wq, event, result, priority));
        else
-           return (wait_queue_wakeup_all(wq, event, result));
+               return (wait_queue_wakeup_all(wq, event, result));
 }
 
 /*
  *     thread_bind:
  *
- *     Force a thread to execute on the specified processor.
+ *     Force the current thread to execute on the specified processor.
  *
  *     Returns the previous binding.  PROCESSOR_NULL means
  *     not bound.
@@ -1095,150 +1496,429 @@ thread_wakeup_prim(
  */
 processor_t
 thread_bind(
-       register thread_t       thread,
-       processor_t                     processor)
+       processor_t             processor)
 {
+       thread_t                self = current_thread();
        processor_t             prev;
-       run_queue_t             runq = RUN_QUEUE_NULL;
        spl_t                   s;
 
        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 = self->bound_processor;
+       self->bound_processor = processor;
 
-       if (runq != RUN_QUEUE_NULL)
-               thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
-       thread_unlock(thread);
+       thread_unlock(self);
        splx(s);
 
        return (prev);
 }
 
-struct {
-       uint32_t        idle_pset_last,
-                               idle_pset_any,
-                               idle_bound;
-
-       uint32_t        pset_self,
-                               pset_last,
-                               pset_other,
-                               bound_self,
-                               bound_other;
-
-       uint32_t        realtime_self,
-                               realtime_last,
-                               realtime_other;
-
-       uint32_t        missed_realtime,
-                               missed_other;
-} dispatch_counts;
-
 /*
- *     Select a thread for the current processor to run.
+ *     thread_select:
+ *
+ *     Select a new thread for the current processor to execute.
  *
  *     May select the current thread, which must be locked.
  */
-thread_t
+static thread_t
 thread_select(
-       register processor_t    processor)
+       thread_t                        thread,
+       processor_t                     processor)
 {
-       register thread_t               thread;
-       processor_set_t                 pset;
-       boolean_t                               other_runnable;
+       processor_set_t         pset = processor->processor_set;
+       thread_t                        new_thread = THREAD_NULL;
+       boolean_t                       inactive_state;
 
-       /*
-        *      Check for other non-idle runnable threads.
-        */
-       pset = processor->processor_set;
-       thread = processor->active_thread;
+       assert(processor == current_processor());
+
+       do {
+               /*
+                *      Update the priority.
+                */
+               if (SCHED(can_update_priority)(thread))
+                       SCHED(update_priority)(thread);
+               
+               processor->current_pri = thread->sched_pri;
+               processor->current_thmode = thread->sched_mode;
 
-       /* Update the thread's priority */
-       if (thread->sched_stamp != sched_tick)
-               update_priority(thread);
+               pset_lock(pset);
 
-       processor->current_pri = thread->sched_pri;
+               assert(pset->low_count);
+               assert(pset->low_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);
+               inactive_state = processor->state != PROCESSOR_SHUTDOWN && machine_processor_is_inactive(processor);
+
+               simple_lock(&rt_lock);
+
+               /*
+                *      Test to see if the current thread should continue
+                *      to run on this processor.  Must be runnable, and not
+                *      bound to a different processor, nor be in the wrong
+                *      processor set.
+                */
+               if (    ((thread->state & ~TH_SUSP) == TH_RUN)                                  &&
+                               (thread->sched_pri >= BASEPRI_RTQUEUES          ||
+                                processor->processor_meta == PROCESSOR_META_NULL ||
+                                processor->processor_meta->primary == processor)               &&
+                               (thread->bound_processor == PROCESSOR_NULL      ||
+                                thread->bound_processor == processor)                  &&
+                               (thread->affinity_set == AFFINITY_SET_NULL      ||
+                                thread->affinity_set->aset_pset == pset)                       ) {
+                       if (    thread->sched_pri >= BASEPRI_RTQUEUES   &&
+                                               first_timeslice(processor)                              ) {
+                               if (rt_runq.count > 0) {
+                                       register queue_t                q;
+
+                                       q = &rt_runq.queue;
+                                       if (((thread_t)q->next)->realtime.deadline <
+                                                                                                       processor->deadline) {
+                                               thread = (thread_t)dequeue_head(q);
+                                               thread->runq = PROCESSOR_NULL;
+                                               SCHED_STATS_RUNQ_CHANGE(&rt_runq.runq_stats, rt_runq.count);
+                                               rt_runq.count--;
                                        }
                                }
+
+                               simple_unlock(&rt_lock);
+
+                               processor->deadline = thread->realtime.deadline;
+
+                               pset_unlock(pset);
+
+                               return (thread);
                        }
 
-                       processor->deadline = thread->realtime.deadline;
+                       if (!inactive_state && (thread->sched_mode != TH_MODE_FAIRSHARE || SCHED(fairshare_runq_count)() == 0) && (rt_runq.count == 0 || BASEPRI_RTQUEUES < thread->sched_pri) &&
+                                       (new_thread = SCHED(choose_thread)(processor, thread->sched_mode == TH_MODE_FAIRSHARE ? MINPRI : thread->sched_pri)) == THREAD_NULL) {
+
+                               simple_unlock(&rt_lock);
+
+                               /* I am the highest priority runnable (non-idle) thread */
+
+                               pset_pri_hint(pset, processor, processor->current_pri);
+
+                               pset_count_hint(pset, processor, SCHED(processor_runq_count)(processor));
+
+                               processor->deadline = UINT64_MAX;
+
+                               pset_unlock(pset);
+
+                               return (thread);
+                       }
+               }
+
+               if (new_thread != THREAD_NULL ||
+                               (SCHED(processor_queue_has_priority)(processor, rt_runq.count == 0 ? IDLEPRI : BASEPRI_RTQUEUES, TRUE) &&
+                                        (new_thread = SCHED(choose_thread)(processor, MINPRI)) != THREAD_NULL)) {
+                               simple_unlock(&rt_lock);
+
+                               if (!inactive_state) {
+                                       pset_pri_hint(pset, processor, new_thread->sched_pri);
+
+                                       pset_count_hint(pset, processor, SCHED(processor_runq_count)(processor));
+                               }
+
+                               processor->deadline = UINT64_MAX;
+                               pset_unlock(pset);
+
+                               return (new_thread);
+               }
+
+               if (rt_runq.count > 0) {
+                       thread = (thread_t)dequeue_head(&rt_runq.queue);
 
-                       simple_unlock(&pset->sched_lock);
+                       thread->runq = PROCESSOR_NULL;
+                       SCHED_STATS_RUNQ_CHANGE(&rt_runq.runq_stats, rt_runq.count);
+                       rt_runq.count--;
+
+                       simple_unlock(&rt_lock);
+
+                       processor->deadline = thread->realtime.deadline;
+                       pset_unlock(pset);
 
                        return (thread);
                }
 
-               if (    (!other_runnable                                                        ||
-                                (processor->runq.highq < thread->sched_pri             &&
-                                 pset->runq.highq < thread->sched_pri))                        ) {
+               simple_unlock(&rt_lock);
 
-                       /* I am the highest priority runnable (non-idle) thread */
+               /* No realtime threads and no normal threads on the per-processor
+                * runqueue. Finally check for global fairshare threads.
+                */
+               if ((new_thread = SCHED(fairshare_dequeue)()) != THREAD_NULL) {
 
                        processor->deadline = UINT64_MAX;
+                       pset_unlock(pset);
+                       
+                       return (new_thread);
+               }
+                       
+               processor->deadline = UINT64_MAX;
+
+               /*
+                *      Set processor inactive based on
+                *      indication from the platform code.
+                */
+               if (inactive_state) {
+                       if (processor->state == PROCESSOR_RUNNING)
+                               remqueue((queue_entry_t)processor);
+                       else
+                       if (processor->state == PROCESSOR_IDLE)
+                               remqueue((queue_entry_t)processor);
 
-                       simple_unlock(&pset->sched_lock);
+                       processor->state = PROCESSOR_INACTIVE;
 
-                       return (thread);
+                       pset_unlock(pset);
+
+                       return (processor->idle_thread);
                }
-       }
 
-       if (other_runnable)
-               thread = choose_thread(pset, processor);
-       else {
+               /*
+                *      No runnable threads, attempt to steal
+                *      from other processors.
+                */
+               new_thread = SCHED(steal_thread)(pset);
+               if (new_thread != THREAD_NULL) {
+                       return (new_thread);
+               }
+
+               /*
+                *      If other threads have appeared, shortcut
+                *      around again.
+                */
+               if (!SCHED(processor_queue_empty)(processor) || rt_runq.count > 0 || SCHED(fairshare_runq_count)() > 0)
+                       continue;
+
+               pset_lock(pset);
+
                /*
                 *      Nothing is runnable, so set this processor idle if it
-                *      was running.  Return its idle thread.
+                *      was running.
                 */
                if (processor->state == PROCESSOR_RUNNING) {
-                       remqueue(&pset->active_queue, (queue_entry_t)processor);
+                       remqueue((queue_entry_t)processor);
                        processor->state = PROCESSOR_IDLE;
 
-                       enqueue_tail(&pset->idle_queue, (queue_entry_t)processor);
-                       pset->idle_count++;
+                       if (processor->processor_meta == PROCESSOR_META_NULL || processor->processor_meta->primary == processor) {
+                               enqueue_head(&pset->idle_queue, (queue_entry_t)processor);
+                               pset_pri_init_hint(pset, processor);
+                               pset_count_init_hint(pset, processor);
+                       }
+                       else {
+                               enqueue_head(&processor->processor_meta->idle_queue, (queue_entry_t)processor);
+                                       pset_unlock(pset);
+                                       return (processor->idle_thread);
+                       }
                }
 
-               processor->deadline = UINT64_MAX;
+               pset_unlock(pset);
+
+#if CONFIG_SCHED_IDLE_IN_PLACE
+               /*
+                *      Choose idle thread if fast idle is not possible.
+                */
+               if ((thread->state & (TH_IDLE|TH_TERMINATE|TH_SUSP)) || !(thread->state & TH_WAIT) || thread->wake_active || thread->sched_pri >= BASEPRI_RTQUEUES)
+                       return (processor->idle_thread);
+
+               /*
+                *      Perform idling activities directly without a
+                *      context switch.  Return dispatched thread,
+                *      else check again for a runnable thread.
+                */
+               new_thread = thread_select_idle(thread, processor);
+
+#else /* !CONFIG_SCHED_IDLE_IN_PLACE */
+               
+               /*
+                * Do a full context switch to idle so that the current
+                * thread can start running on another processor without
+                * waiting for the fast-idled processor to wake up.
+                */
+               return (processor->idle_thread);
+
+#endif /* !CONFIG_SCHED_IDLE_IN_PLACE */
+
+       } while (new_thread == THREAD_NULL);
+
+       return (new_thread);
+}
+
+#if CONFIG_SCHED_IDLE_IN_PLACE
+/*
+ *     thread_select_idle:
+ *
+ *     Idle the processor using the current thread context.
+ *
+ *     Called with thread locked, then dropped and relocked.
+ */
+static thread_t
+thread_select_idle(
+       thread_t                thread,
+       processor_t             processor)
+{
+       thread_t                new_thread;
+
+       if (thread->sched_mode == TH_MODE_TIMESHARE)
+               sched_share_decr();
+       sched_run_decr();
+
+       thread->state |= TH_IDLE;
+       processor->current_pri = IDLEPRI;
+       processor->current_thmode = TH_MODE_NONE;
+
+       thread_unlock(thread);
+
+       /*
+        *      Switch execution timing to processor idle thread.
+        */
+       processor->last_dispatch = mach_absolute_time();
+       thread->last_run_time = processor->last_dispatch;
+       thread_timer_event(processor->last_dispatch, &processor->idle_thread->system_timer);
+       PROCESSOR_DATA(processor, kernel_timer) = &processor->idle_thread->system_timer;
+
+       /*
+        *      Cancel the quantum timer while idling.
+        */
+       timer_call_cancel(&processor->quantum_timer);
+       processor->timeslice = 0;
+
+       (*thread->sched_call)(SCHED_CALL_BLOCK, thread);
+
+       thread_tell_urgency(THREAD_URGENCY_NONE, 0, 0);
+
+       /*
+        *      Enable interrupts and perform idling activities.  No
+        *      preemption due to TH_IDLE being set.
+        */
+       spllo(); new_thread = processor_idle(thread, processor);
+
+       /*
+        *      Return at splsched.
+        */
+       (*thread->sched_call)(SCHED_CALL_UNBLOCK, thread);
+
+       thread_lock(thread);
+
+       /*
+        * If we idled in place, simulate a context switch back
+        * to the original priority of the thread so that the
+        * platform layer cannot distinguish this from a true
+        * switch to the idle thread.
+        */
+       if (thread->sched_mode == TH_MODE_REALTIME)
+               thread_tell_urgency(THREAD_URGENCY_REAL_TIME, thread->realtime.period, thread->realtime.deadline);
+       /* Identify non-promoted threads which have requested a
+        * "background" priority.
+        */
+       else if ((thread->sched_pri <= MAXPRI_THROTTLE) &&
+                        (thread->priority <= MAXPRI_THROTTLE))
+               thread_tell_urgency(THREAD_URGENCY_BACKGROUND, thread->sched_pri, thread->priority);
+       else
+               thread_tell_urgency(THREAD_URGENCY_NORMAL, thread->sched_pri, thread->priority);
+
+       /*
+        *      If awakened, switch to thread timer and start a new quantum.
+        *      Otherwise skip; we will context switch to another thread or return here.
+        */
+       if (!(thread->state & TH_WAIT)) {
+               processor->last_dispatch = mach_absolute_time();
+               thread_timer_event(processor->last_dispatch, &thread->system_timer);
+               PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
+
+               thread_quantum_init(thread);
+               thread->last_quantum_refill_time = processor->last_dispatch;
+
+               processor->quantum_end = processor->last_dispatch + thread->current_quantum;
+               timer_call_enter1(&processor->quantum_timer, thread, processor->quantum_end, TIMER_CALL_CRITICAL);
+               processor->timeslice = 1;
 
-               thread = processor->idle_thread;
+               thread->computation_epoch = processor->last_dispatch;
        }
 
-       simple_unlock(&pset->sched_lock);
+       thread->state &= ~TH_IDLE;
 
-       return (thread);
+       sched_run_incr();
+       if (thread->sched_mode == TH_MODE_TIMESHARE)
+               sched_share_incr();
+
+       return (new_thread);
+}
+#endif /* CONFIG_SCHED_IDLE_IN_PLACE */
+
+#if defined(CONFIG_SCHED_TRADITIONAL) 
+static thread_t
+sched_traditional_choose_thread(
+                                                               processor_t             processor,
+                                                               int                             priority)
+{
+       thread_t thread;
+       
+       thread = choose_thread(processor, runq_for_processor(processor), priority);
+       if (thread != THREAD_NULL) {
+               runq_consider_decr_bound_count(processor, thread);
+       }
+       
+       return thread;
+}
+
+#endif /* defined(CONFIG_SCHED_TRADITIONAL)  */
+
+#if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_FIXEDPRIORITY)
+
+/*
+ *     choose_thread:
+ *
+ *     Locate a thread to execute from the processor run queue
+ *     and return it.  Only choose a thread with greater or equal
+ *     priority.
+ *
+ *     Associated pset must be locked.  Returns THREAD_NULL
+ *     on failure.
+ */
+thread_t
+choose_thread(
+       processor_t             processor,
+       run_queue_t             rq,
+       int                             priority)
+{
+       queue_t                 queue = rq->queues + rq->highq;
+       int                             pri = rq->highq, count = rq->count;
+       thread_t                thread;
+
+       while (count > 0 && pri >= priority) {
+               thread = (thread_t)queue_first(queue);
+               while (!queue_end(queue, (queue_entry_t)thread)) {
+                       if (thread->bound_processor == PROCESSOR_NULL ||
+                                                       thread->bound_processor == processor) {
+                               remqueue((queue_entry_t)thread);
+
+                               thread->runq = PROCESSOR_NULL;
+                               SCHED_STATS_RUNQ_CHANGE(&rq->runq_stats, rq->count);
+                               rq->count--;
+                               if (SCHED(priority_is_urgent)(pri)) {
+                                       rq->urgency--; assert(rq->urgency >= 0);
+                               }
+                               if (queue_empty(queue)) {
+                                       if (pri != IDLEPRI)
+                                               clrbit(MAXPRI - pri, rq->bitmap);
+                                       rq->highq = MAXPRI - ffsbit(rq->bitmap);
+                               }
+
+                               return (thread);
+                       }
+                       count--;
+
+                       thread = (thread_t)queue_next((queue_entry_t)thread);
+               }
+
+               queue--; pri--;
+       }
+
+       return (THREAD_NULL);
 }
 
+#endif /* defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_FIXEDPRIORITY) */
+
 /*
  *     Perform a context switch and start executing the new thread.
  *
@@ -1273,124 +1953,148 @@ MACRO_BEGIN                                                                                                   \
        }                                                                                                               \
 MACRO_END
 
-boolean_t
+static boolean_t
 thread_invoke(
-       register thread_t       old_thread,
-       register thread_t       new_thread,
+       register thread_t       self,
+       register thread_t       thread,
        ast_t                           reason)
 {
-       thread_continue_t       new_cont, continuation = old_thread->continuation;
-       void                            *new_param, *parameter = old_thread->parameter;
+       thread_continue_t       continuation = self->continuation;
+       void                            *parameter = self->parameter;
        processor_t                     processor;
-       thread_t                        prev_thread;
 
-       if (get_preemption_level() != 0)
-               panic("thread_invoke: preemption_level %d\n",
-                                                               get_preemption_level());
+       if (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"));
+       }
 
-       assert(old_thread == current_thread());
+       assert(self == current_thread());
 
        /*
         * Mark thread interruptible.
         */
-       thread_lock(new_thread);
-       new_thread->state &= ~TH_UNINT;
+       thread_lock(thread);
+       thread->state &= ~TH_UNINT;
 
-       assert(thread_runnable(new_thread));
+#if DEBUG
+       assert(thread_runnable(thread));
+#endif
 
        /*
         * Allow time constraint threads to hang onto
         * a stack.
         */
-       if (    (old_thread->sched_mode & TH_MODE_REALTIME)             &&
-                                       !old_thread->reserved_stack                             ) {
-               old_thread->reserved_stack = old_thread->kernel_stack;
-       }
+       if ((self->sched_mode == TH_MODE_REALTIME) && !self->reserved_stack)
+               self->reserved_stack = self->kernel_stack;
 
        if (continuation != NULL) {
-               if (!new_thread->kernel_stack) {
+               if (!thread->kernel_stack) {
                        /*
-                        * If the old thread is using a privileged stack,
+                        * If we are using a privileged stack,
                         * check to see whether we can exchange it with
-                        * that of the new thread.
+                        * that of the other thread.
                         */
-                       if (    old_thread->kernel_stack == old_thread->reserved_stack  &&
-                                                       !new_thread->reserved_stack)
+                       if (self->kernel_stack == self->reserved_stack && !thread->reserved_stack)
                                goto need_stack;
 
                        /*
                         * 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;
+                       continuation = thread->continuation;
+                       parameter = thread->parameter;
 
                        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++;
+                       processor->active_thread = thread;
+                       processor->current_pri = thread->sched_pri;
+                       processor->current_thmode = thread->sched_mode;
+                       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);
 
-                       old_thread->reason = reason;
+                       self->reason = reason;
 
                        processor->last_dispatch = mach_absolute_time();
-                       timer_event((uint32_t)processor->last_dispatch,
-                                                                               &new_thread->system_timer);
-          
-                       thread_done(old_thread, new_thread, processor);
+                       self->last_run_time = processor->last_dispatch;
+                       thread_timer_event(processor->last_dispatch, &thread->system_timer);
+                       PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
+       
+                       KERNEL_DEBUG_CONSTANT(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);
 
-                       machine_stack_handoff(old_thread, new_thread);
+                       if ((thread->chosen_processor != processor) && (thread->chosen_processor != NULL)) {
+                               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);
+                       }
 
-                       thread_begin(new_thread, processor);
+                       DTRACE_SCHED2(off__cpu, struct thread *, thread, struct proc *, thread->task->bsd_info);
 
-                       /*
-                        * Now dispatch the old thread.
-                        */
-                       thread_dispatch(old_thread);
+                       SCHED_STATS_CSW(processor, self->reason, self->sched_pri, thread->sched_pri);
+
+                       TLOG(1, "thread_invoke: calling stack_handoff\n");
+                       stack_handoff(self, thread);
+
+                       DTRACE_SCHED(on__cpu);
+
+                       thread_dispatch(self, thread);
+
+                       thread->continuation = thread->parameter = NULL;
 
-                       counter_always(c_thread_invoke_hits++);
+                       counter(c_thread_invoke_hits++);
 
-                       funnel_refunnel_check(new_thread, 2);
+                       funnel_refunnel_check(thread, 2);
                        (void) spllo();
 
-                       assert(new_cont);
-                       call_continuation(new_cont, new_param, new_thread->wait_result);
+                       assert(continuation);
+                       call_continuation(continuation, parameter, thread->wait_result);
                        /*NOTREACHED*/
                }
-               else
-               if (new_thread == old_thread) {
+               else if (thread == self) {
                        /* same thread but with continuation */
+                       ast_context(self);
                        counter(++c_thread_invoke_same);
-                       thread_unlock(new_thread);
+                       thread_unlock(self);
+
+                       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED,MACH_SCHED) | DBG_FUNC_NONE,
+                                                               self->reason, (uintptr_t)thread_tid(thread), self->sched_pri, thread->sched_pri, 0);
+
+                       self->continuation = self->parameter = NULL;
 
-                       funnel_refunnel_check(new_thread, 3);
+                       funnel_refunnel_check(self, 3);
                        (void) spllo();
 
-                       call_continuation(continuation, parameter, new_thread->wait_result);
+                       call_continuation(continuation, parameter, self->wait_result);
                        /*NOTREACHED*/
                }
        }
        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:
-                       if (!stack_alloc_try(new_thread)) {
-                               counter_always(c_thread_invoke_misses++);
-                               thread_unlock(new_thread);
-                               thread_stack_enqueue(new_thread);
+                       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);
-                       thread_unlock(new_thread);
+                       thread_unlock(self);
+
+                       KERNEL_DEBUG_CONSTANT(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);
                }
        }
@@ -1399,47 +2103,63 @@ need_stack:
         * 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->current_pri = thread->sched_pri;
+       processor->current_thmode = thread->sched_mode;
+       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_always(c_thread_invoke_csw++);
-       current_task()->csw++;
+       counter(c_thread_invoke_csw++);
 
-       assert(old_thread->runq == RUN_QUEUE_NULL);
-       old_thread->reason = reason;
+       assert(self->runq == PROCESSOR_NULL);
+       self->reason = reason;
 
        processor->last_dispatch = mach_absolute_time();
-       timer_event((uint32_t)processor->last_dispatch, &new_thread->system_timer);
+       self->last_run_time = processor->last_dispatch;
+       thread_timer_event(processor->last_dispatch, &thread->system_timer);
+       PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
+
+       KERNEL_DEBUG_CONSTANT(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)) {
+               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);
 
-       thread_done(old_thread, new_thread, processor);
+       SCHED_STATS_CSW(processor, self->reason, self->sched_pri, thread->sched_pri);
 
        /*
         * This is where we actually switch register context,
-        * and address space if required.  Control will not
-        * return here immediately.
+        * and address space if required.  We will next run
+        * as a result of a subsequent context switch.
         */
-       prev_thread = machine_switch_context(old_thread, continuation, new_thread);
+       thread = machine_switch_context(self, continuation, thread);
+       TLOG(1,"thread_invoke: returning machine_switch_context: self %p continuation %p thread %p\n", self, continuation, thread);
 
-       /*
-        * We are still old_thread, possibly on a different processor,
-        * and new_thread is now stale.
-        */
-       thread_begin(old_thread, old_thread->last_processor);
+       DTRACE_SCHED(on__cpu);
 
        /*
-        * Now dispatch the thread which resumed us.
+        * We have been resumed and are set to run.
         */
-       thread_dispatch(prev_thread);
+       thread_dispatch(thread, self);
 
        if (continuation) {
-               funnel_refunnel_check(old_thread, 3);
+               self->continuation = self->parameter = NULL;
+
+               funnel_refunnel_check(self, 3);
                (void) spllo();
 
-               call_continuation(continuation, parameter, old_thread->wait_result);
+               call_continuation(continuation, parameter, self->wait_result);
                /*NOTREACHED*/
        }
 
@@ -1447,184 +2167,178 @@ need_stack:
 }
 
 /*
- *     thread_done:
+ *     thread_dispatch:
  *
- *     Perform calculations for thread
- *     finishing execution on the current processor.
+ *     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.
  *
  *     Called at splsched.
  */
 void
-thread_done(
-       thread_t                        old_thread,
-       thread_t                        new_thread,
-       processor_t                     processor)
+thread_dispatch(
+       thread_t                thread,
+       thread_t                self)
 {
-       if (!(old_thread->state & TH_IDLE)) {
+       processor_t             processor = self->last_processor;
+
+       if (thread != THREAD_NULL) {
                /*
-                * Compute remainder of current quantum.
+                *      If blocked at a continuation, discard
+                *      the stack.
                 */
-               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 (thread->continuation != NULL && thread->kernel_stack != 0)
+                       stack_free(thread);
+
+               if (!(thread->state & TH_IDLE)) {
+                       wake_lock(thread);
+                       thread_lock(thread);
 
-               if (old_thread->sched_mode & TH_MODE_REALTIME) {
                        /*
-                        * Cancel the deadline if the thread has
-                        * consumed the entire quantum.
+                        *      Compute remainder of current quantum.
                         */
-                       if (old_thread->current_quantum == 0) {
-                               old_thread->realtime.deadline = UINT64_MAX;
-                               old_thread->reason |= AST_QUANTUM;
+                       if (    first_timeslice(processor)                                                      &&
+                                       processor->quantum_end > processor->last_dispatch               )
+                               thread->current_quantum = (uint32_t)(processor->quantum_end - processor->last_dispatch);
+                       else
+                               thread->current_quantum = 0;
+
+                       if (thread->sched_mode == TH_MODE_REALTIME) {
+                               /*
+                                *      Cancel the deadline if the thread has
+                                *      consumed the entire quantum.
+                                */
+                               if (thread->current_quantum == 0) {
+                                       thread->realtime.deadline = UINT64_MAX;
+                                       thread->reason |= AST_QUANTUM;
+                               }
+                       } else {
+#if defined(CONFIG_SCHED_TRADITIONAL)
+                               /*
+                                *      For non-realtime threads treat a tiny
+                                *      remaining quantum as an expired quantum
+                                *      but include what's left next time.
+                                */
+                               if (thread->current_quantum < min_std_quantum) {
+                                       thread->reason |= AST_QUANTUM;
+                                       thread->current_quantum += std_quantum;
+                               }
+#endif
                        }
-               }
-               else {
+
                        /*
-                        * For non-realtime threads treat a tiny
-                        * remaining quantum as an expired quantum
-                        * but include what's left next time.
+                        *      If we are doing a direct handoff then
+                        *      take the remainder of the quantum.
                         */
-                       if (old_thread->current_quantum < min_std_quantum) {
-                               old_thread->reason |= AST_QUANTUM;
-                               old_thread->current_quantum += std_quantum;
+                       if ((thread->reason & (AST_HANDOFF|AST_QUANTUM)) == AST_HANDOFF) {
+                               self->current_quantum = thread->current_quantum;
+                               thread->reason |= AST_QUANTUM;
+                               thread->current_quantum = 0;
                        }
-               }
 
-               /*
-                * If we are doing a direct handoff then
-                * give the remainder of our quantum to
-                * the next thread.
-                */
-               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;
-               }
+                       thread->computation_metered += (processor->last_dispatch - thread->computation_epoch);
 
-               old_thread->last_switch = processor->last_dispatch;
+                       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);
 
-               old_thread->computation_metered +=
-                               (old_thread->last_switch - old_thread->computation_epoch);
-       }
-}
+                               thread->reason = AST_NONE;
 
-/*
- *     thread_begin:
- *
- *     Set up for thread beginning execution on
- *     the current processor.
- *
- *     Called at splsched.
- */
-void
-thread_begin(
-       thread_t                        thread,
-       processor_t                     processor)
-{
-       if (!(thread->state & TH_IDLE)) {
-               /*
-                * Give the thread a new quantum
-                * if none remaining.
-                */
-               if (thread->current_quantum == 0)
-                       thread_quantum_init(thread);
-
-               /*
-                * Set up quantum timer and timeslice.
-                */
-               processor->quantum_end =
-                               (processor->last_dispatch + thread->current_quantum);
-               timer_call_enter1(&processor->quantum_timer,
-                                                               thread, processor->quantum_end);
+                               thread_unlock(thread);
+                               wake_unlock(thread);
+                       }
+                       else {
+                               /*
+                                *      Waiting.
+                                */
+                               boolean_t should_terminate = FALSE;
 
-               processor_timeslice_setup(processor, thread);
+                               /* 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;
+                               }
 
-               thread->last_switch = processor->last_dispatch;
+                               thread->state &= ~TH_RUN;
 
-               thread->computation_epoch = thread->last_switch;
-       }
-       else {
-               timer_call_cancel(&processor->quantum_timer);
-               processor->timeslice = 1;
-       }
-}
+                               if (thread->sched_mode == TH_MODE_TIMESHARE)
+                                       sched_share_decr();
+                               sched_run_decr();
 
-/*
- *     thread_dispatch:
- *
- *     Handle previous thread at context switch.  Re-dispatch
- *     if still running, otherwise update run state and perform
- *     special actions.
- *
- *     Called at splsched.
- */
-void
-thread_dispatch(
-       register thread_t       thread)
-{
-       /*
-        *      If blocked at a continuation, discard
-        *      the stack.
-        */
-#ifndef i386
-    if (thread->continuation != NULL && thread->kernel_stack)
-               stack_free(thread);
-#endif
+                               (*thread->sched_call)(SCHED_CALL_BLOCK, thread);
 
-       if (!(thread->state & TH_IDLE)) {
-               wake_lock(thread);
-               thread_lock(thread);
+                               if (thread->wake_active) {
+                                       thread->wake_active = FALSE;
+                                       thread_unlock(thread);
 
-               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);
+                                       thread_wakeup(&thread->wake_active);
+                               }
+                               else
+                                       thread_unlock(thread);
 
-                       thread->reason = AST_NONE;
+                               wake_unlock(thread);
 
-                       thread_unlock(thread);
-                       wake_unlock(thread);
+                               if (should_terminate)
+                                       thread_terminate_enqueue(thread);
+                       }
                }
-               else {
-                       boolean_t               wake;
-
-                       /*
-                        *      Waiting.
-                        */
-                       thread->state &= ~TH_RUN;
+       }
 
-                       wake = thread->wake_active;
-                       thread->wake_active = FALSE;
+       if (!(self->state & TH_IDLE)) {
 
-                       if (thread->sched_mode & TH_MODE_TIMESHARE)
-                               pset_share_decr(thread->processor_set);
-                       pset_run_decr(thread->processor_set);
+               if (self->sched_mode == TH_MODE_REALTIME)
+                       thread_tell_urgency(THREAD_URGENCY_REAL_TIME, self->realtime.period, self->realtime.deadline);
+               /* Identify non-promoted threads which have requested a
+                * "background" priority.
+                */
+               else if ((self->sched_pri <= MAXPRI_THROTTLE) &&
+                       (self->priority <= MAXPRI_THROTTLE))
+                       thread_tell_urgency(THREAD_URGENCY_BACKGROUND, self->sched_pri, self->priority);
+               else
+                       thread_tell_urgency(THREAD_URGENCY_NORMAL, self->sched_pri, self->priority);
+               /*
+                *      Get a new quantum if none remaining.
+                */
+               if (self->current_quantum == 0) {
+                       thread_quantum_init(self);
+                       self->last_quantum_refill_time = processor->last_dispatch;
+               }
 
-                       thread_unlock(thread);
-                       wake_unlock(thread);
+               /*
+                *      Set up quantum timer and timeslice.
+                */
+               processor->quantum_end = (processor->last_dispatch + self->current_quantum);
+               timer_call_enter1(&processor->quantum_timer, self, processor->quantum_end, TIMER_CALL_CRITICAL);
 
-                       if (thread->options & TH_OPT_CALLOUT)
-                               call_thread_block();
+               processor->timeslice = 1;
 
-                       if (wake)
-                               thread_wakeup((event_t)&thread->wake_active);
+               self->computation_epoch = processor->last_dispatch;
+       }
+       else {
+               timer_call_cancel(&processor->quantum_timer);
+               processor->timeslice = 0;
 
-                       if (thread->state & TH_TERMINATE)
-                               thread_terminate_enqueue(thread);
-               }
+               thread_tell_urgency(THREAD_URGENCY_NONE, 0, 0);
        }
 }
 
+#include <libkern/OSDebug.h>
+
+uint32_t       kdebug_thread_block = 0;
+
+
 /*
  *     thread_block_reason:
  *
@@ -1658,16 +2372,6 @@ thread_block_reason(
 
        processor = current_processor();
 
-       /*
-        * 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);
-       }
-
        /* If we're explicitly yielding, force a subsequent quantum */
        if (reason & AST_YIELD)
                processor->timeslice = 0;
@@ -1678,16 +2382,20 @@ thread_block_reason(
        self->continuation = continuation;
        self->parameter = parameter;
 
-       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)) {
+       if (__improbable(kdebug_thread_block && kdebug_enable && self->state != TH_RUN)) {
+               uint32_t        bt[8];
+
+               OSBacktrace((void **)&bt[0], 8);
+
+               KERNEL_DEBUG_CONSTANT(0x140004c | DBG_FUNC_START, bt[0], bt[1], bt[2], bt[3], 0);
+               KERNEL_DEBUG_CONSTANT(0x140004c | DBG_FUNC_END, bt[4], bt[5], bt[6], bt[7], 0);
+       }
+
+       do {
                thread_lock(self);
-               new_thread = thread_select(processor);
-               assert(new_thread && thread_runnable(new_thread));
+               new_thread = thread_select(self, processor);
                thread_unlock(self);
-       }
+       } while (!thread_invoke(self, new_thread, reason));
 
        funnel_refunnel_check(self, 5);
        splx(s);
@@ -1740,10 +2448,10 @@ thread_run(
        self->parameter = parameter;
 
        while (!thread_invoke(self, new_thread, handoff)) {
-               register processor_t            processor = current_processor();
+               processor_t             processor = current_processor();
 
                thread_lock(self);
-               new_thread = thread_select(processor);
+               new_thread = thread_select(self, processor);
                thread_unlock(self);
                handoff = AST_NONE;
        }
@@ -1761,101 +2469,320 @@ thread_run(
  */
 void
 thread_continue(
-       register thread_t       old_thread)
+       register thread_t       thread)
 {
        register thread_t                       self = current_thread();
        register thread_continue_t      continuation;
        register void                           *parameter;
-       
+
+       DTRACE_SCHED(on__cpu);
+
        continuation = self->continuation;
-       self->continuation = NULL;
        parameter = self->parameter;
-       self->parameter = NULL;
 
-       thread_begin(self, self->last_processor);
+       thread_dispatch(thread, self);
 
-       if (old_thread != THREAD_NULL)
-               thread_dispatch(old_thread);
+       self->continuation = self->parameter = NULL;
 
        funnel_refunnel_check(self, 4);
 
-       if (old_thread != THREAD_NULL)
+       if (thread != THREAD_NULL)
                (void)spllo();
 
+ TLOG(1, "thread_continue: calling call_continuation \n");
        call_continuation(continuation, parameter, self->wait_result);
        /*NOTREACHED*/
 }
 
+void
+thread_quantum_init(thread_t thread)
+{
+       if (thread->sched_mode == TH_MODE_REALTIME) {
+               thread->current_quantum = thread->realtime.computation;
+       } else {
+               thread->current_quantum = SCHED(initial_quantum_size)(thread);
+       }
+}
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+static uint32_t
+sched_traditional_initial_quantum_size(thread_t thread __unused)
+{
+       return std_quantum;
+}
+
+static sched_mode_t
+sched_traditional_initial_thread_sched_mode(task_t parent_task)
+{
+       if (parent_task == kernel_task)
+               return TH_MODE_FIXED;
+       else
+               return TH_MODE_TIMESHARE;
+}
+
+static boolean_t
+sched_traditional_supports_timeshare_mode(void)
+{
+       return TRUE;
+}
+
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
 /*
- *     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.
+ *     run_queue_init:
  *
- *     Run queue must be locked, see run_queue_remove()
- *     for more info.
+ *     Initialize a run queue before first use.
  */
-static boolean_t
-run_queue_enqueue(
-       register run_queue_t    rq,
-       register thread_t               thread,
-       integer_t                               options)
+void
+run_queue_init(
+       run_queue_t             rq)
+{
+       int                             i;
+
+       rq->highq = IDLEPRI;
+       for (i = 0; i < NRQBM; i++)
+               rq->bitmap[i] = 0;
+       setbit(MAXPRI - IDLEPRI, rq->bitmap);
+       rq->urgency = rq->count = 0;
+       for (i = 0; i < NRQS; i++)
+               queue_init(&rq->queues[i]);
+}
+
+#if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_PROTO) || defined(CONFIG_SCHED_GRRR) || defined(CONFIG_SCHED_FIXEDPRIORITY)
+int
+sched_traditional_fairshare_runq_count(void)
+{
+       return fs_runq.count;
+}
+
+uint64_t
+sched_traditional_fairshare_runq_stats_count_sum(void)
+{
+       return fs_runq.runq_stats.count_sum;
+}
+
+void
+sched_traditional_fairshare_enqueue(thread_t thread)
+{
+       queue_t                         queue = &fs_runq.queue;
+       
+       simple_lock(&fs_lock);
+       
+       enqueue_tail(queue, (queue_entry_t)thread);
+       
+       thread->runq = FS_RUNQ;
+       SCHED_STATS_RUNQ_CHANGE(&fs_runq.runq_stats, fs_runq.count);
+       fs_runq.count++;
+       
+       simple_unlock(&fs_lock);        
+}
+
+thread_t
+sched_traditional_fairshare_dequeue(void)
 {
-       register int                    whichq = thread->sched_pri;
-       register queue_t                queue = &rq->queues[whichq];
-       boolean_t                               result = FALSE;
+       thread_t thread;
+       
+       simple_lock(&fs_lock);
+       if (fs_runq.count > 0) {
+               thread = (thread_t)dequeue_head(&fs_runq.queue);
+               
+               thread->runq = PROCESSOR_NULL;
+               SCHED_STATS_RUNQ_CHANGE(&fs_runq.runq_stats, fs_runq.count);
+               fs_runq.count--;
+               
+               simple_unlock(&fs_lock);
+               
+               return (thread);
+       }
+       simple_unlock(&fs_lock);                
+
+       return THREAD_NULL;
+}
+
+boolean_t
+sched_traditional_fairshare_queue_remove(thread_t thread)
+{
+       queue_t                 q;
+
+       simple_lock(&fs_lock);
+       q = &fs_runq.queue;
        
-       assert(whichq >= MINPRI && whichq <= MAXPRI);
+       if (FS_RUNQ == thread->runq) {
+               remqueue((queue_entry_t)thread);
+               SCHED_STATS_RUNQ_CHANGE(&fs_runq.runq_stats, fs_runq.count);
+               fs_runq.count--;
+               
+               thread->runq = PROCESSOR_NULL;
+               simple_unlock(&fs_lock);
+               return (TRUE);
+       }
+       else {
+               /*
+                *      The thread left the run queue before we could
+                *      lock the run queue.
+                */
+               assert(thread->runq == PROCESSOR_NULL);
+               simple_unlock(&fs_lock);
+               return (FALSE);
+       }       
+}
+
+#endif /* defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_PROTO) || defined(CONFIG_SCHED_GRRR) || defined(CONFIG_SCHED_FIXEDPRIORITY) */
+
+/*
+ *     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,
+       integer_t               options)
+{
+       thread_t                thread;
+       queue_t                 queue = rq->queues + rq->highq;
+
+       if (options & SCHED_HEADQ) {
+               thread = (thread_t)dequeue_head(queue);
+       }
+       else {
+               thread = (thread_t)dequeue_tail(queue);
+       }
 
-       assert(thread->runq == RUN_QUEUE_NULL);
+       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 (queue_empty(queue)) {
-               enqueue_tail(queue, (queue_entry_t)thread);
+               if (rq->highq != IDLEPRI)
+                       clrbit(MAXPRI - rq->highq, rq->bitmap);
+               rq->highq = MAXPRI - ffsbit(rq->bitmap);
+       }
+
+       return (thread);
+}
 
-               setbit(MAXPRI - whichq, rq->bitmap);
-               if (whichq > rq->highq) {
-                       rq->highq = whichq;
+/*
+ *     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,
+                                                         integer_t             options)
+{
+       queue_t                 queue = rq->queues + thread->sched_pri;
+       boolean_t               result = FALSE;
+       
+       if (queue_empty(queue)) {
+               enqueue_tail(queue, (queue_entry_t)thread);
+               
+               setbit(MAXPRI - thread->sched_pri, rq->bitmap);
+               if (thread->sched_pri > rq->highq) {
+                       rq->highq = thread->sched_pri;
                        result = TRUE;
                }
        }
        else
-       if (options & SCHED_HEADQ)
-               enqueue_head(queue, (queue_entry_t)thread);
-       else
-               enqueue_tail(queue, (queue_entry_t)thread);
-
-       thread->runq = rq;
-       if (thread->sched_mode & TH_MODE_PREEMPT)
+               if (options & SCHED_TAILQ)
+                       enqueue_tail(queue, (queue_entry_t)thread);
+               else
+                       enqueue_head(queue, (queue_entry_t)thread);
+       
+       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)
+{
+
+       remqueue((queue_entry_t)thread);
+       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 (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);
+       }
+       
+       thread->runq = PROCESSOR_NULL;
 }
 
 /*
- *     Enqueue a thread for realtime execution, similar
- *     to above.  Handles preemption directly.
+ *     fairshare_setrun:
+ *
+ *     Dispatch a thread for round-robin execution.
+ *
+ *     Thread must be locked.  Associated pset must
+ *     be locked, and is returned unlocked.
  */
 static void
-realtime_schedule_insert(
-       register processor_set_t        pset,
-       register thread_t                       thread)
+fairshare_setrun(
+                                 processor_t                   processor,
+                                 thread_t                      thread)
+{
+       processor_set_t         pset = processor->processor_set;
+               
+       thread->chosen_processor = processor;
+
+       SCHED(fairshare_enqueue)(thread);
+       
+       if (processor != current_processor())
+               machine_signal_idle(processor);
+
+       pset_unlock(pset);
+
+}
+
+/*
+ *     realtime_queue_insert:
+ *
+ *     Enqueue a thread for realtime execution.
+ */
+static boolean_t
+realtime_queue_insert(
+       thread_t                        thread)
 {
-       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;
+       queue_t                         queue = &rt_runq.queue;
+       uint64_t                        deadline = thread->realtime.deadline;
+       boolean_t                       preempt = FALSE;
 
-       assert(whichq >= BASEPRI_REALTIME && whichq <= MAXPRI);
+       simple_lock(&rt_lock);
 
-       assert(thread->runq == RUN_QUEUE_NULL);
        if (queue_empty(queue)) {
                enqueue_tail(queue, (queue_entry_t)thread);
-
-               setbit(MAXPRI - whichq, rq->bitmap);
-               if (whichq > rq->highq)
-                       rq->highq = whichq;
-               try_preempt = TRUE;
+               preempt = TRUE;
        }
        else {
                register thread_t       entry = (thread_t)queue_first(queue);
@@ -1871,365 +2798,768 @@ realtime_schedule_insert(
                }
 
                if ((queue_entry_t)entry == queue)
-                       try_preempt = TRUE;
+                       preempt = TRUE;
 
                insque((queue_entry_t)thread, (queue_entry_t)entry);
        }
 
-       thread->runq = rq;
-       assert(thread->sched_mode & TH_MODE_PREEMPT);
-       rq->count++; rq->urgency++;
-
-       if (try_preempt) {
-               register processor_t    processor;
-
-               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);
-
-                       ast_on(AST_PREEMPT | AST_URGENT);
-                       return;
-               }
-
-               if (    pset->processor_count > 1                       ||
-                               pset != processor->processor_set        ) {
-                       processor_t             myprocessor, lastprocessor;
-                       queue_entry_t   next;
-
-                       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;
-                               }
+       thread->runq = RT_RUNQ;
+       SCHED_STATS_RUNQ_CHANGE(&rt_runq.runq_stats, rt_runq.count);
+       rt_runq.count++;
 
-                               processor = (processor_t)next;
-                       }
-               }
-       }
+       simple_unlock(&rt_lock);
 
-       simple_unlock(&pset->sched_lock);
+       return (preempt);
 }
 
 /*
- *     thread_setrun:
+ *     realtime_setrun:
  *
- *     Dispatch thread for execution, directly onto an idle
- *     processor if possible.  Else put on appropriate run
- *     queue. (local if bound, else processor set)
+ *     Dispatch a thread for realtime execution.
  *
- *     Thread must be locked.
+ *     Thread must be locked.  Associated pset must
+ *     be locked, and is returned unlocked.
  */
-void
-thread_setrun(
-       register thread_t                       new_thread,
-       integer_t                                       options)
+static void
+realtime_setrun(
+       processor_t                     processor,
+       thread_t                        thread)
 {
-       register processor_t            processor;
-       register processor_set_t        pset;
-       register thread_t                       thread;
-       ast_t                                           preempt = (options & SCHED_PREEMPT)?
-                                                                                                       AST_PREEMPT: AST_NONE;
+       processor_set_t         pset = processor->processor_set;
 
-       assert(thread_runnable(new_thread));
-       
-       /*
-        *      Update priority if needed.
-        */
-       if (new_thread->sched_stamp != sched_tick)
-               update_priority(new_thread);
+       thread->chosen_processor = processor;
 
        /*
-        *      Check for urgent preemption.
+        *      Dispatch directly onto idle processor.
         */
-       if (new_thread->sched_mode & TH_MODE_PREEMPT)
-               preempt = (AST_PREEMPT | AST_URGENT);
+       if ( (thread->bound_processor == processor)
+               && processor->state == PROCESSOR_IDLE) {
+               remqueue((queue_entry_t)processor);
+               enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
 
-       assert(new_thread->runq == RUN_QUEUE_NULL);
+               processor->next_thread = thread;
+               processor->deadline = thread->realtime.deadline;
+               processor->state = PROCESSOR_DISPATCHING;
+               pset_unlock(pset);
 
-       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);
-               }
-               else
-               simple_lock(&pset->sched_lock);
+               if (processor != current_processor())
+                       machine_signal_idle(processor);
+               return;
+       }
 
-               /*
-                *      Next pick any idle processor
-                *      in the processor set.
+       if (realtime_queue_insert(thread)) {
+               int prstate = processor->state;
+               if (processor == current_processor())
+                       ast_on(AST_PREEMPT | AST_URGENT);
+               else if ((prstate == PROCESSOR_DISPATCHING)  || (prstate == PROCESSOR_IDLE))
+                       machine_signal_idle(processor);
+               else
+                       cause_ast_check(processor);
+       }
+
+       pset_unlock(pset);
+}
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
+static boolean_t
+priority_is_urgent(int priority)
+{
+       return testbit(priority, sched_preempt_pri) ? TRUE : FALSE;
+}
+
+/*
+ *     processor_enqueue:
+ *
+ *     Enqueue thread on a processor 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.
+ *
+ *     The run queue must be locked (see thread_run_queue_remove()
+ *     for more info).
+ */
+static boolean_t
+processor_enqueue(
+       processor_t             processor,
+       thread_t                thread,
+       integer_t               options)
+{
+       run_queue_t             rq = runq_for_processor(processor);
+       boolean_t               result;
+       
+       result = run_queue_enqueue(rq, thread, options);
+       thread->runq = processor;
+       runq_consider_incr_bound_count(processor, thread);
+
+       return (result);
+}
+
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
+/*
+ *     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;
+       ast_t                           preempt;
+
+       thread->chosen_processor = processor;
+
+       /*
+        *      Dispatch directly onto idle processor.
+        */
+       if ( (SCHED(direct_dispatch_to_idle_processors) ||
+                 thread->bound_processor == processor)
+               && processor->state == PROCESSOR_IDLE) {
+               remqueue((queue_entry_t)processor);
+               enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
+
+               processor->next_thread = thread;
+               processor->deadline = UINT64_MAX;
+               processor->state = PROCESSOR_DISPATCHING;
+               pset_unlock(pset);
+
+               if (processor != current_processor())
+                       machine_signal_idle(processor);
+               return;
+       }
+
+       /*
+        *      Set preemption mode.
+        */
+       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->priority)
+               preempt = AST_NONE;
+       else
+               preempt = (options & SCHED_PREEMPT)? AST_PREEMPT: AST_NONE;
+
+       if (!SCHED(processor_enqueue)(processor, thread, options))
+               preempt = AST_NONE;
+
+       if (preempt != AST_NONE) {
+               if (processor == current_processor()) {
+                       if (csw_check(processor) != AST_NONE)
+                               ast_on(preempt);
+               }
+               else
+               if (    processor->state == PROCESSOR_IDLE || processor->state == PROCESSOR_DISPATCHING) {
+                       machine_signal_idle(processor);
+               }
+               else
+               if (    (processor->state == PROCESSOR_RUNNING          ||
+                                processor->state == PROCESSOR_SHUTDOWN)                &&
+                               (thread->sched_pri >= processor->current_pri    ||
+                               processor->current_thmode == TH_MODE_FAIRSHARE)) {
+                       cause_ast_check(processor);
+               }
+       }
+       else
+       if (    processor->state == PROCESSOR_SHUTDOWN          &&
+                       thread->sched_pri >= processor->current_pri     ) {
+               cause_ast_check(processor);
+       }
+       else
+       if (    processor->state == PROCESSOR_IDLE      &&
+                       processor != current_processor()        ) {
+                               machine_signal_idle(processor);
+       }
+
+       pset_unlock(pset);
+}
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
+static boolean_t
+processor_queue_empty(processor_t              processor)
+{
+       return runq_for_processor(processor)->count == 0;
+       
+}
+
+static boolean_t
+sched_traditional_with_pset_runqueue_processor_queue_empty(processor_t         processor)
+{
+       processor_set_t pset = processor->processor_set;
+       int count = runq_for_processor(processor)->count;
+
+       /*
+        * The pset runq contains the count of all runnable threads
+        * for all processors in the pset. However, for threads that
+        * are bound to another processor, the current "processor"
+        * is not eligible to execute the thread. So we only
+        * include bound threads that our bound to the current
+        * "processor". This allows the processor to idle when the
+        * count of eligible threads drops to 0, even if there's
+        * a runnable thread bound to a different processor in the
+        * shared runq.
+        */
+
+       count -= pset->pset_runq_bound_count;
+       count += processor->runq_bound_count;
+
+       return count == 0;
+}
+
+static ast_t
+processor_csw_check(processor_t processor)
+{
+       run_queue_t             runq;
+
+       assert(processor->active_thread != NULL);
+       
+       runq = runq_for_processor(processor);
+       if (runq->highq > processor->current_pri) {
+               if (runq->urgency > 0)
+                       return (AST_PREEMPT | AST_URGENT);
+               
+               if (processor->active_thread && thread_eager_preemption(processor->active_thread))
+                       return (AST_PREEMPT | AST_URGENT);
+
+               return AST_PREEMPT;
+       }
+
+       return AST_NONE;
+}
+
+static boolean_t
+processor_queue_has_priority(processor_t               processor,
+                                                        int                            priority,
+                                                        boolean_t                      gte)
+{
+       if (gte)
+               return runq_for_processor(processor)->highq >= priority;
+       else
+               return runq_for_processor(processor)->highq > priority;
+}
+
+static boolean_t
+should_current_thread_rechoose_processor(processor_t                   processor)
+{
+       return (processor->current_pri < BASEPRI_RTQUEUES
+                       && processor->processor_meta != PROCESSOR_META_NULL
+                       && processor->processor_meta->primary != processor);
+}
+
+static int
+sched_traditional_processor_runq_count(processor_t   processor)
+{
+       return runq_for_processor(processor)->count;
+}
+
+
+static uint64_t
+sched_traditional_processor_runq_stats_count_sum(processor_t   processor)
+{
+       return runq_for_processor(processor)->runq_stats.count_sum;
+}
+
+static uint64_t
+sched_traditional_with_pset_runqueue_processor_runq_stats_count_sum(processor_t   processor)
+{
+       if (processor->cpu_id == processor->processor_set->cpu_set_low)
+               return runq_for_processor(processor)->runq_stats.count_sum;
+       else
+               return 0ULL;
+}
+
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
+#define next_pset(p)   (((p)->pset_list != PROCESSOR_SET_NULL)? (p)->pset_list: (p)->node->psets)
+
+/*
+ *     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         pset,
+       processor_t                     processor,
+       thread_t                        thread)
+{
+       processor_set_t         nset, cset = pset;
+       processor_meta_t        pmeta = PROCESSOR_META_NULL;
+       processor_t             mprocessor;
+       
+       /*
+        *      Prefer the hinted processor, when appropriate.
+        */
+
+       if (processor != PROCESSOR_NULL) {
+               if (processor->processor_meta != PROCESSOR_META_NULL)
+                       processor = processor->processor_meta->primary;
+       }
+
+       mprocessor = machine_choose_processor(pset, processor);
+       if (mprocessor != PROCESSOR_NULL)
+               processor = mprocessor;
+
+       if (processor != PROCESSOR_NULL) {
+               if (processor->processor_set != pset ||
+                   processor->state == PROCESSOR_INACTIVE ||
+                   processor->state == PROCESSOR_SHUTDOWN ||
+                   processor->state == PROCESSOR_OFF_LINE)
+                       processor = PROCESSOR_NULL;
+               else
+                       if (processor->state == PROCESSOR_IDLE ||
+                           ((thread->sched_pri >= BASEPRI_RTQUEUES) &&
+                           (processor->current_pri < BASEPRI_RTQUEUES)))
+                               return (processor);
+       }
+
+       /*
+        *      Iterate through the processor sets to locate
+        *      an appropriate processor.
+        */
+       do {
+               /*
+                *      Choose an idle processor.
                 */
-               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;
+               if (!queue_empty(&cset->idle_queue))
+                       return ((processor_t)queue_first(&cset->idle_queue));
+
+               if (thread->sched_pri >= BASEPRI_RTQUEUES) {
+                       integer_t lowest_priority = MAXPRI + 1;
+                       integer_t lowest_unpaired = MAXPRI + 1;
+                       uint64_t  furthest_deadline = 1;
+                       processor_t lp_processor = PROCESSOR_NULL;
+                       processor_t lp_unpaired = PROCESSOR_NULL;
+                       processor_t fd_processor = PROCESSOR_NULL;
+
+                       lp_processor = cset->low_pri;
+                       /* Consider hinted processor */
+                       if (lp_processor != PROCESSOR_NULL &&
+                           ((lp_processor->processor_meta == PROCESSOR_META_NULL) ||
+                           ((lp_processor == lp_processor->processor_meta->primary) &&
+                           !queue_empty(&lp_processor->processor_meta->idle_queue))) &&
+                           lp_processor->state != PROCESSOR_INACTIVE &&
+                           lp_processor->state != PROCESSOR_SHUTDOWN &&
+                           lp_processor->state != PROCESSOR_OFF_LINE &&
+                           (lp_processor->current_pri < thread->sched_pri))
+                               return lp_processor;
+
+                       processor = (processor_t)queue_first(&cset->active_queue);
+                       while (!queue_end(&cset->active_queue, (queue_entry_t)processor)) {
+                               /* Discover the processor executing the
+                                * thread with the lowest priority within
+                                * this pset, or the one with the furthest
+                                * deadline
+                                */
+                               integer_t cpri = processor->current_pri;
+                               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;
+                               }
+
+
+                               if (processor->processor_meta != PROCESSOR_META_NULL &&
+                                   !queue_empty(&processor->processor_meta->idle_queue)) {
+                                       if (cpri < lowest_unpaired) {
+                                               lowest_unpaired = cpri;
+                                               lp_unpaired = processor;
+                                               pmeta = processor->processor_meta;
+                                       }
+                                       else
+                                               if (pmeta == PROCESSOR_META_NULL)
+                                                       pmeta = processor->processor_meta;
+                               }
+                               processor = (processor_t)queue_next((queue_entry_t)processor);
+                       }
+
+                       if (thread->sched_pri > lowest_unpaired)
+                               return lp_unpaired;
+
+                       if (pmeta != PROCESSOR_META_NULL)
+                               return ((processor_t)queue_first(&pmeta->idle_queue));
+                       if (thread->sched_pri > lowest_priority)
+                               return lp_processor;
+                       if (thread->realtime.deadline < furthest_deadline)
+                               return fd_processor;
+
+                       processor = PROCESSOR_NULL;
+               }
+               else {
+                       /*
+                        *      Check any hinted processors in the processor set if available.
+                        */
+                       if (cset->low_pri != PROCESSOR_NULL && cset->low_pri->state != PROCESSOR_INACTIVE &&
+                                       cset->low_pri->state != PROCESSOR_SHUTDOWN && cset->low_pri->state != PROCESSOR_OFF_LINE &&
+                                               (processor == PROCESSOR_NULL ||
+                                                       (thread->sched_pri > BASEPRI_DEFAULT && cset->low_pri->current_pri < thread->sched_pri))) {
+                               processor = cset->low_pri;
+                       }
                        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 (cset->low_count != PROCESSOR_NULL && cset->low_count->state != PROCESSOR_INACTIVE &&
+                                       cset->low_count->state != PROCESSOR_SHUTDOWN && cset->low_count->state != PROCESSOR_OFF_LINE &&
+                                               (processor == PROCESSOR_NULL || (thread->sched_pri <= BASEPRI_DEFAULT &&
+                                                                                                                       SCHED(processor_runq_count)(cset->low_count) < SCHED(processor_runq_count)(processor)))) {
+                               processor = cset->low_count;
+                       }
+
+                       /*
+                        *      Otherwise, choose an available processor in the set.
+                        */
+                       if (processor == PROCESSOR_NULL) {
+                               processor = (processor_t)dequeue_head(&cset->active_queue);
+                               if (processor != PROCESSOR_NULL)
+                                       enqueue_tail(&cset->active_queue, (queue_entry_t)processor);
+                       }
+
+                       if (processor != PROCESSOR_NULL && pmeta == PROCESSOR_META_NULL) {
+                               if (processor->processor_meta != PROCESSOR_META_NULL &&
+                                                                                       !queue_empty(&processor->processor_meta->idle_queue))
+                                       pmeta = processor->processor_meta;
+                       }
                }
 
-               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;
+               /*
+                *      Move onto the next processor set.
+                */
+               nset = next_pset(cset);
+
+               if (nset != pset) {
+                       pset_unlock(cset);
+
+                       cset = nset;
+                       pset_lock(cset);
+               }
+       } while (nset != pset);
+
+       /*
+        *      Make sure that we pick a running processor,
+        *      and that the correct processor set is locked.
+        */
+       do {
+               if (pmeta != PROCESSOR_META_NULL) {
+                       if (cset != pmeta->primary->processor_set) {
+                               pset_unlock(cset);
+
+                               cset = pmeta->primary->processor_set;
+                               pset_lock(cset);
+                       }
+
+                       if (!queue_empty(&pmeta->idle_queue))
+                               return ((processor_t)queue_first(&pmeta->idle_queue));
+
+                       pmeta = PROCESSOR_META_NULL;
+               }
+
+               /*
+                *      If we haven't been able to choose a processor,
+                *      pick the boot processor and return it.
+                */
+               if (processor == PROCESSOR_NULL) {
+                       processor = master_processor;
 
                        /*
-                        *      Update the timesharing quanta.
+                        *      Check that the correct processor set is
+                        *      returned locked.
                         */
-                       timeshare_quanta_update(pset);
+                       if (cset != processor->processor_set) {
+                               pset_unlock(cset);
+
+                               cset = processor->processor_set;
+                               pset_lock(cset);
+                       }
+
+                       return (processor);
+               }
+
+               /*
+                *      Check that the processor set for the chosen
+                *      processor is locked.
+                */
+               if (cset != processor->processor_set) {
+                       pset_unlock(cset);
+
+                       cset = processor->processor_set;
+                       pset_lock(cset);
+               }
+
+               /*
+                *      We must verify that the chosen processor is still available.
+                */
+               if (processor->state == PROCESSOR_INACTIVE ||
+                                       processor->state == PROCESSOR_SHUTDOWN || processor->state == PROCESSOR_OFF_LINE)
+                       processor = PROCESSOR_NULL;
+       } while (processor == PROCESSOR_NULL);
+
+       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,
+       integer_t                       options)
+{
+       processor_t                     processor;
+       processor_set_t         pset;
+
+#if DEBUG
+       assert(thread_runnable(thread));
+#endif
        
+       /*
+        *      Update priority if needed.
+        */
+       if (SCHED(can_update_priority)(thread))
+               SCHED(update_priority)(thread);
+
+       assert(thread->runq == PROCESSOR_NULL);
+
+       if (thread->bound_processor == PROCESSOR_NULL) {
+               /*
+                *      Unbound case.
+                */
+               if (thread->affinity_set != AFFINITY_SET_NULL) {
                        /*
-                        *      Preempt check.
+                        * Use affinity set policy hint.
                         */
-                       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;
-                               }
+                       pset = thread->affinity_set->aset_pset;
+                       pset_lock(pset);
 
-                               /*
-                                * 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;
-                                       }
+                       processor = SCHED(choose_processor)(pset, PROCESSOR_NULL, thread);
+               }
+               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);
 
-                                       /*
-                                        * 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;
-                                               }
-
-                                               processor = (processor_t)next;
-                                       }
-                               }
+                       if ((thread->last_processor != processor) && (thread->last_processor != PROCESSOR_NULL)) {
+                               KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_LPA_BROKEN)|DBG_FUNC_NONE,
+                                                                         (uintptr_t)thread_tid(thread), (uintptr_t)thread->last_processor->cpu_id, (uintptr_t)processor->cpu_id, thread->last_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;
 
-                       simple_unlock(&pset->sched_lock);
+                       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);
+                       task->pset_hint = processor->processor_set;
                }
        }
        else {
-           /*
-            *  Bound, can only run on bound processor.  Have to lock
-            *  processor here because it may not be the current one.
-            */
-               processor_lock(processor);
+               /*
+                *      Bound case:
+                *
+                *      Unconditionally dispatch on the processor.
+                */
+               processor = thread->bound_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;
-
-               if (preempt != AST_NONE) {
-                       if (processor == current_processor()) {
-                               thread = processor->active_thread;
-                               if (csw_needed(thread, processor)) {
-                                       dispatch_counts.bound_self++;
-                                       ast_on(preempt);
+               pset_lock(pset);
+       }
+
+       /*
+        *      Dispatch the thread on the choosen processor.
+        */
+       if (thread->sched_pri >= BASEPRI_RTQUEUES)
+               realtime_setrun(processor, thread);
+       else if (thread->sched_mode == TH_MODE_FAIRSHARE)
+               fairshare_setrun(processor, thread);
+       else
+               processor_setrun(processor, thread, options);
+}
+
+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);
+}
+
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
+/*
+ *     processor_queue_shutdown:
+ *
+ *     Shutdown a processor run queue by
+ *     re-dispatching non-bound threads.
+ *
+ *     Associated pset must be locked, and is
+ *     returned unlocked.
+ */
+void
+processor_queue_shutdown(
+       processor_t                     processor)
+{
+       processor_set_t         pset = processor->processor_set;
+       run_queue_t                     rq = runq_for_processor(processor);
+       queue_t                         queue = rq->queues + rq->highq;
+       int                                     pri = rq->highq, count = rq->count;
+       thread_t                        next, thread;
+       queue_head_t            tqueue;
+
+       queue_init(&tqueue);
+       
+       while (count > 0) {
+               thread = (thread_t)queue_first(queue);
+               while (!queue_end(queue, (queue_entry_t)thread)) {
+                       next = (thread_t)queue_next((queue_entry_t)thread);
+
+                       if (thread->bound_processor == PROCESSOR_NULL) {
+                               remqueue((queue_entry_t)thread);
+
+                               thread->runq = PROCESSOR_NULL;
+                               SCHED_STATS_RUNQ_CHANGE(&rq->runq_stats, rq->count);
+                               runq_consider_decr_bound_count(processor, thread);
+                               rq->count--;
+                               if (SCHED(priority_is_urgent)(pri)) {
+                                       rq->urgency--; assert(rq->urgency >= 0);
                                }
+                               if (queue_empty(queue)) {
+                                       if (pri != IDLEPRI)
+                                               clrbit(MAXPRI - pri, rq->bitmap);
+                                       rq->highq = MAXPRI - ffsbit(rq->bitmap);
+                               }
+
+                               enqueue_tail(&tqueue, (queue_entry_t)thread);
                        }
-                       else
-                       if (    processor->state == PROCESSOR_RUNNING                   &&
-                                       new_thread->sched_pri > processor->current_pri  ) {
-                               dispatch_counts.bound_other++;
-                               cause_ast_check(processor);
-                       }
+                       count--;
+
+                       thread = next;
                }
 
-               if (pset != PROCESSOR_SET_NULL)
-                       simple_unlock(&pset->sched_lock);
+               queue--; pri--;
+       }
 
-               processor_unlock(processor);
+       pset_unlock(pset);
+
+       while ((thread = (thread_t)dequeue_head(&tqueue)) != THREAD_NULL) {
+               thread_lock(thread);
+
+               thread_setrun(thread, SCHED_TAILQ);
+
+               thread_unlock(thread);
        }
 }
 
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
 /*
- *     Check for a possible preemption point in
- *     the (current) thread.
+ *     Check for a preemption point in
+ *     the current context.
  *
  *     Called at splsched.
  */
 ast_t
 csw_check(
-       thread_t                thread,
        processor_t             processor)
 {
-       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);
-
-                       result |= AST_PREEMPT;
-               }
+       ast_t                   result = AST_NONE;
 
-               runq = &processor->runq;
-               if (runq->highq > current_pri) {
-                       if (runq->urgency > 0)
-                               return (AST_PREEMPT | AST_URGENT);
+       if (first_timeslice(processor)) {
+               if (rt_runq.count > 0)
+                       return (AST_PREEMPT | AST_URGENT);
 
-                       result |= AST_PREEMPT;
-               }
+               result |= SCHED(processor_csw_check)(processor);
+               if (result & AST_URGENT)
+                       return result;
        }
        else {
-               runq = &processor->processor_set->runq;
-               if (runq->highq >= current_pri) {
-                       if (runq->urgency > 0)
-                               return (AST_PREEMPT | AST_URGENT);
-
-                       result |= AST_PREEMPT;
-               }
-
-               runq = &processor->runq;
-               if (runq->highq >= current_pri) {
-                       if (runq->urgency > 0)
-                               return (AST_PREEMPT | AST_URGENT);
+               if (rt_runq.count > 0 && BASEPRI_RTQUEUES >= processor->current_pri)
+                       return (AST_PREEMPT | AST_URGENT);
 
-                       result |= AST_PREEMPT;
-               }
+               result |= SCHED(processor_csw_check)(processor);
+               if (result & AST_URGENT)
+                       return result;
        }
 
        if (result != AST_NONE)
                return (result);
 
-       if (thread->state & TH_SUSP)
-               result |= AST_PREEMPT;
+       if (SCHED(should_current_thread_rechoose_processor)(processor))
+               return (AST_PREEMPT);
+       
+       if (machine_processor_is_inactive(processor))
+               return (AST_PREEMPT);
+
+       if (processor->active_thread->state & TH_SUSP)
+               return (AST_PREEMPT);
 
-       return (result);
+       return (AST_NONE);
 }
 
 /*
@@ -2243,33 +3573,25 @@ csw_check(
  */
 void
 set_sched_pri(
-       thread_t                        thread,
-       int                                     priority)
+       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;
+       boolean_t               removed = thread_run_queue_remove(thread);
 
        thread->sched_pri = priority;
-       if (rq != RUN_QUEUE_NULL)
+       if (removed)
                thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
        else
        if (thread->state & TH_RUN) {
                processor_t             processor = thread->last_processor;
 
                if (thread == current_thread()) {
-                       ast_t           preempt = csw_check(thread, processor);
+                       ast_t                   preempt;
 
-                       if (preempt != AST_NONE)
-                               ast_on(preempt);
                        processor->current_pri = priority;
+                       processor->current_thmode = thread->sched_mode;
+                       if ((preempt = csw_check(processor)) != AST_NONE)
+                               ast_on(preempt);
                }
                else
                if (    processor != PROCESSOR_NULL                                             &&
@@ -2308,349 +3630,362 @@ run_queue_check(
 
 #endif /* DEBUG */
 
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
+/* locks the runqueue itself */
+
+static boolean_t
+processor_queue_remove(
+                                          processor_t                  processor,
+                                          thread_t             thread)
+{
+       void *                  rqlock;
+       run_queue_t             rq;
+       
+       rqlock = &processor->processor_set->sched_lock;
+       rq = runq_for_processor(processor);
+
+       simple_lock(rqlock);
+       if (processor == thread->runq) {
+               /*
+                *      Thread is on a run queue and we have a lock on
+                *      that run queue.
+                */
+               runq_consider_decr_bound_count(processor, thread);
+               run_queue_remove(rq, thread);
+       }
+       else {
+               /*
+                *      The thread left the run queue before we could
+                *      lock the run queue.
+                */
+               assert(thread->runq == PROCESSOR_NULL);
+               processor = PROCESSOR_NULL;
+       }
+       
+       simple_unlock(rqlock);
+
+       return (processor != PROCESSOR_NULL);
+}
+
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
 /*
- *     run_queue_remove:
+ *     thread_run_queue_remove:
  *
- *     Remove a thread from its current run queue and
- *     return the run queue if successful.
+ *     Remove a thread from a current run queue and
+ *     return TRUE if successful.
  *
  *     Thread must be locked.
  */
-run_queue_t
-run_queue_remove(
-       thread_t                        thread)
+boolean_t
+thread_run_queue_remove(
+       thread_t                thread)
 {
-       register run_queue_t    rq = thread->runq;
+       processor_t             processor = thread->runq;
 
        /*
-        *      If rq is RUN_QUEUE_NULL, the thread will stay out of the
+        *      If processor is PROCESSOR_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;
+       if (processor != PROCESSOR_NULL) {
+               queue_t                 q;
 
                /*
-                *      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.
+                *      The processor run queues are locked by the
+                *      processor set.  Real-time priorities use a
+                *      global queue with a dedicated lock.
                 */
-               if (processor != PROCESSOR_NULL) {
-                       processor_lock(processor);
-                       pset = processor->processor_set;
+               if (thread->sched_mode == TH_MODE_FAIRSHARE) {
+                       return SCHED(fairshare_queue_remove)(thread);
+               }
+               
+               if (thread->sched_pri < BASEPRI_RTQUEUES) {
+                       return SCHED(processor_queue_remove)(processor, thread);
                }
 
-               if (pset != PROCESSOR_SET_NULL)
-                       simple_lock(&pset->sched_lock);
+               simple_lock(&rt_lock);
+               q = &rt_runq.queue;
 
-               if (rq == thread->runq) {
+               if (processor == 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);
-
-                       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);
-                       }
+                       remqueue((queue_entry_t)thread);
+                       SCHED_STATS_RUNQ_CHANGE(&rt_runq.runq_stats, rt_runq.count);
+                       rt_runq.count--;
 
-                       thread->runq = RUN_QUEUE_NULL;
+                       thread->runq = PROCESSOR_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;
+                       assert(thread->runq == PROCESSOR_NULL);
+                       processor = PROCESSOR_NULL;
                }
 
-               if (pset != PROCESSOR_SET_NULL)
-                       simple_unlock(&pset->sched_lock);
-
-               if (processor != PROCESSOR_NULL)
-                       processor_unlock(processor);
+               simple_unlock(&rt_lock);
        }
 
-       return (rq);
+       return (processor != PROCESSOR_NULL);
 }
 
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
 /*
- *     choose_thread:
+ *     steal_processor_thread:
  *
- *     Remove a thread to execute from the run queues
- *     and return it.
+ *     Locate a thread to steal from the processor and
+ *     return it.
  *
- *     Called with pset scheduling lock held.
+ *     Associated pset must be locked.  Returns THREAD_NULL
+ *     on failure.
  */
 static thread_t
-choose_thread(
-       processor_set_t         pset,
-       processor_t                     processor)
+steal_processor_thread(
+       processor_t             processor)
 {
-       register run_queue_t    runq;
-       register thread_t               thread;
-       register queue_t                q;
+       run_queue_t             rq = runq_for_processor(processor);
+       queue_t                 queue = rq->queues + rq->highq;
+       int                             pri = rq->highq, count = rq->count;
+       thread_t                thread;
 
-       runq = &processor->runq;
-
-       if (runq->count > 0 && runq->highq >= pset->runq.highq) {
-               q = runq->queues + runq->highq;
-
-               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);
-               }
+       while (count > 0) {
+               thread = (thread_t)queue_first(queue);
+               while (!queue_end(queue, (queue_entry_t)thread)) {
+                       if (thread->bound_processor == PROCESSOR_NULL) {
+                               remqueue((queue_entry_t)thread);
+
+                               thread->runq = PROCESSOR_NULL;
+                               SCHED_STATS_RUNQ_CHANGE(&rq->runq_stats, rq->count);
+                               runq_consider_decr_bound_count(processor, thread);
+                               rq->count--;
+                               if (SCHED(priority_is_urgent)(pri)) {
+                                       rq->urgency--; assert(rq->urgency >= 0);
+                               }
+                               if (queue_empty(queue)) {
+                                       if (pri != IDLEPRI)
+                                               clrbit(MAXPRI - pri, rq->bitmap);
+                                       rq->highq = MAXPRI - ffsbit(rq->bitmap);
+                               }
 
-               processor->deadline = UINT64_MAX;
+                               return (thread);
+                       }
+                       count--;
 
-               return (thread);
+                       thread = (thread_t)queue_next((queue_entry_t)thread);
+               }
+
+               queue--; pri--;
        }
 
-       runq = &pset->runq;
+       return (THREAD_NULL);
+}
+
+/*
+ *     Locate and steal a thread, beginning
+ *     at the pset.
+ *
+ *     The pset must be locked, and is returned
+ *     unlocked.
+ *
+ *     Returns the stolen thread, or THREAD_NULL on
+ *     failure.
+ */
+static thread_t
+steal_thread(
+       processor_set_t         pset)
+{
+       processor_set_t         nset, cset = pset;
+       processor_t                     processor;
+       thread_t                        thread;
 
-       assert(runq->count > 0);
-       q = runq->queues + runq->highq;
+       do {
+               processor = (processor_t)queue_first(&cset->active_queue);
+               while (!queue_end(&cset->active_queue, (queue_entry_t)processor)) {
+                       if (runq_for_processor(processor)->count > 0) {
+                               thread = steal_processor_thread(processor);
+                               if (thread != THREAD_NULL) {
+                                       remqueue((queue_entry_t)processor);
+                                       enqueue_tail(&cset->active_queue, (queue_entry_t)processor);
 
-       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);
-       }
+                                       pset_unlock(cset);
 
-       timeshare_quanta_update(pset);
+                                       return (thread);
+                               }
+                       }
 
-       return (thread);
+                       processor = (processor_t)queue_next((queue_entry_t)processor);
+               }
+
+               nset = next_pset(cset);
+
+               if (nset != pset) {
+                       pset_unlock(cset);
+
+                       cset = nset;
+                       pset_lock(cset);
+               }
+       } while (nset != pset);
+
+       pset_unlock(cset);
+
+       return (THREAD_NULL);
 }
 
-static processor_t
-delay_idle(
-       processor_t             processor,
-       thread_t                self)
+static thread_t        steal_thread_disabled(
+                                       processor_set_t         pset)
 {
-       int                             *gcount, *lcount;
-       uint64_t                abstime, spin, limit;
-
-       lcount = &processor->runq.count;
-       gcount = &processor->processor_set->runq.count;
+       pset_unlock(pset);
 
-       abstime = mach_absolute_time();
-       limit = abstime + delay_idle_limit;
-       spin = abstime + delay_idle_spin;
+       return (THREAD_NULL);
+}
 
-       timer_event((uint32_t)abstime, &processor->idle_thread->system_timer);
+#endif /* CONFIG_SCHED_TRADITIONAL */
 
-       self->options |= TH_OPT_DELAYIDLE;
 
-       while (         *gcount == 0 && *lcount == 0    &&
-                               (self->state & TH_WAIT) != 0    &&
-                                       abstime < limit                         ) {
-               if (abstime >= spin) {
-                       (void)spllo();
+int
+thread_get_urgency(uint64_t *rt_period, uint64_t *rt_deadline)
+{
+       processor_t     processor;
+       thread_t        thread;
+       
+       processor = current_processor();
 
-                       (void)splsched();
-                       processor = current_processor();
-                       lcount = &processor->runq.count;
-                       gcount = &processor->processor_set->runq.count;
+       thread = processor->next_thread;
 
-                       abstime = mach_absolute_time();
-                       spin = abstime + delay_idle_spin;
+       if (thread != NULL) {
+               if (thread->sched_mode == TH_MODE_REALTIME) {
 
-                       timer_event((uint32_t)abstime, &processor->idle_thread->system_timer);
-               }
-               else
-                       abstime = mach_absolute_time();
-       }
+                       if (rt_period != NULL)
+                               *rt_period = thread->realtime.period;
+                       if (rt_deadline != NULL)
+                               *rt_deadline = thread->realtime.deadline;
 
-       timer_event((uint32_t)abstime, &self->system_timer);
+                       KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_GET_URGENCY), THREAD_URGENCY_REAL_TIME, thread->realtime.period,
+                           (thread->realtime.deadline >> 32), thread->realtime.deadline, 0);
 
-       self->options &= ~TH_OPT_DELAYIDLE;
+                       return (THREAD_URGENCY_REAL_TIME);
+               } else if ((thread->sched_pri <= MAXPRI_THROTTLE) &&
+                   (thread->priority <= MAXPRI_THROTTLE)) {
+                       KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_GET_URGENCY), THREAD_URGENCY_BACKGROUND, thread->sched_pri, thread->priority, 0, 0);
+                       return (THREAD_URGENCY_BACKGROUND);
+               }
+               else 
+                       KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_GET_URGENCY), THREAD_URGENCY_NORMAL, 0, 0, 0, 0);
 
-       return (processor);
+               return (THREAD_URGENCY_NORMAL);
+       }
+       else
+               KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_GET_URGENCY), THREAD_URGENCY_NONE, 0, 0, 0, 0);
+       return (THREAD_URGENCY_NONE);
 }
 
-/*
- *     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.
+ *     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.
+ *
+ *     Returns a the next thread to execute if dispatched directly.
  */
-void
-idle_thread(void)
-{
-       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 = current_processor();
+#if 0
+#define IDLE_KERNEL_DEBUG_CONSTANT(...) KERNEL_DEBUG_CONSTANT(__VA_ARGS__)
+#else
+#define IDLE_KERNEL_DEBUG_CONSTANT(...) do { } while(0)
+#endif
 
-       threadp = &processor->next_thread;
-       lcount = &processor->runq.count;
-       gcount = &processor->processor_set->runq.count;
+thread_t
+processor_idle(
+       thread_t                        thread,
+       processor_t                     processor)
+{
+       processor_set_t         pset = processor->processor_set;
+       thread_t                        new_thread;
+       int                                     state;
+       (void)splsched();
 
+       KERNEL_DEBUG_CONSTANT(
+               MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_START, (uintptr_t)thread_tid(thread), 0, 0, 0, 0);
 
-       (void)splsched();                       /* Turn interruptions off */
+       SCHED_STATS_CPU_IDLE_START(processor);
 
-#ifdef __ppc__
-       pmsDown();                                      /* Step power down.  Note: interruptions must be disabled for this call */
-#endif
+       timer_switch(&PROCESSOR_DATA(processor, system_state),
+                                                                       mach_absolute_time(), &PROCESSOR_DATA(processor, idle_state));
+       PROCESSOR_DATA(processor, current_state) = &PROCESSOR_DATA(processor, idle_state);
 
-       while ( (*threadp == THREAD_NULL)                               &&
-                               (*gcount == 0) && (*lcount == 0)        ) {
+       while (processor->next_thread == THREAD_NULL && SCHED(processor_queue_empty)(processor) && rt_runq.count == 0 && SCHED(fairshare_runq_count)() == 0 &&
+                               (thread == THREAD_NULL || ((thread->state & (TH_WAIT|TH_SUSP)) == TH_WAIT && !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, SCHED(processor_runq_count)(processor), -1, 0);
 
-               /* check for ASTs while we wait */
-               if (*myast &~ (AST_SCHEDULING | AST_BSD)) {
-                       /* no ASTs for us */
-                       *myast &= AST_NONE;
-                       (void)spllo();
-               }
-               else
-                       machine_idle();
+               machine_idle();
 
                (void)splsched();
+
+               IDLE_KERNEL_DEBUG_CONSTANT(
+                       MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_NONE, (uintptr_t)thread_tid(thread), rt_runq.count, SCHED(processor_runq_count)(processor), -2, 0);
+
+               if (processor->state == PROCESSOR_INACTIVE && !machine_processor_is_inactive(processor))
+                       break;
        }
 
-       /*
-        *      This is not a switch statement to avoid the
-        *      bounds checking code in the common case.
-        */
-       pset = processor->processor_set;
-       simple_lock(&pset->sched_lock);
+       timer_switch(&PROCESSOR_DATA(processor, idle_state),
+                                                                       mach_absolute_time(), &PROCESSOR_DATA(processor, system_state));
+       PROCESSOR_DATA(processor, current_state) = &PROCESSOR_DATA(processor, system_state);
 
-#ifdef __ppc__
-       pmsStep(0);                                     /* Step up out of idle power, may start timer for next step */
-#endif
+       pset_lock(pset);
 
        state = processor->state;
        if (state == PROCESSOR_DISPATCHING) {
                /*
                 *      Commmon case -- cpu dispatched.
                 */
-               new_thread = *threadp;
-               *threadp = (volatile thread_t) THREAD_NULL;
+               new_thread = processor->next_thread;
+               processor->next_thread = 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);
-
-                               thread_lock(thread);
-                               thread_setrun(thread, SCHED_HEADQ);
-                               thread_unlock(thread);
-
-                               counter(c_idle_thread_handoff++);
-                               thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
-                               /*NOTREACHED*/
-                       }
-                       simple_unlock(&pset->sched_lock);
 
-                       counter(c_idle_thread_handoff++);
-                       thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
-                       /*NOTREACHED*/
-               }
+               if (SCHED(processor_queue_has_priority)(processor, new_thread->sched_pri, FALSE)                                        ||
+                               (rt_runq.count > 0 && BASEPRI_RTQUEUES >= new_thread->sched_pri)        ) {
+                       processor->deadline = UINT64_MAX;
 
-               if (    processor->runq.highq > new_thread->sched_pri           ||
-                               pset->runq.highq > new_thread->sched_pri                                ) {
-                       thread_t        thread = new_thread;
+                       pset_unlock(pset);
 
-                       new_thread = choose_thread(pset, processor);
-                       dispatch_counts.missed_other++;
-                       simple_unlock(&pset->sched_lock);
+                       thread_lock(new_thread);
+                       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_REDISPATCH), (uintptr_t)thread_tid(new_thread), new_thread->sched_pri, rt_runq.count, 0, 0);
+                       thread_setrun(new_thread, SCHED_HEADQ);
+                       thread_unlock(new_thread);
 
-                       thread_lock(thread);
-                       thread_setrun(thread, SCHED_HEADQ);
-                       thread_unlock(thread);
+                       KERNEL_DEBUG_CONSTANT(
+                               MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (uintptr_t)thread_tid(thread), state, 0, 0, 0);
 
-                       counter(c_idle_thread_handoff++);
-                       thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
-                       /* NOTREACHED */
+                       return (THREAD_NULL);
                }
-               else {
-                       simple_unlock(&pset->sched_lock);
 
-                       counter(c_idle_thread_handoff++);
-                       thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
-                       /* NOTREACHED */
-               }
+               pset_unlock(pset);
+
+               KERNEL_DEBUG_CONSTANT(
+                                     MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (uintptr_t)thread_tid(thread), state, (uintptr_t)thread_tid(new_thread), 0, 0);
+                       
+               return (new_thread);
        }
        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);
+               remqueue((queue_entry_t)processor);
+
+               processor->state = PROCESSOR_RUNNING;
+               enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
+       }
+       else
+       if (state == PROCESSOR_INACTIVE) {
                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) {
@@ -2658,26 +3993,49 @@ idle_thread(void)
                 *      Going off-line.  Force a
                 *      reschedule.
                 */
-               if ((new_thread = (thread_t)*threadp) != THREAD_NULL) {
-                       *threadp = (volatile thread_t) THREAD_NULL;
+               if ((new_thread = processor->next_thread) != THREAD_NULL) {
+                       processor->next_thread = THREAD_NULL;
                        processor->deadline = UINT64_MAX;
-                       simple_unlock(&pset->sched_lock);
+
+                       pset_unlock(pset);
 
                        thread_lock(new_thread);
                        thread_setrun(new_thread, SCHED_HEADQ);
                        thread_unlock(new_thread);
-               }
-               else
-                       simple_unlock(&pset->sched_lock);
 
-               counter(c_idle_thread_block++);
-               thread_block((thread_continue_t)idle_thread);
-               /* NOTREACHED */
+                       KERNEL_DEBUG_CONSTANT(
+                               MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (uintptr_t)thread_tid(thread), state, 0, 0, 0);
+               
+                       return (THREAD_NULL);
+               }
        }
 
-       simple_unlock(&pset->sched_lock);
+       pset_unlock(pset);
+
+       KERNEL_DEBUG_CONSTANT(
+               MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (uintptr_t)thread_tid(thread), state, 0, 0, 0);
+               
+       return (THREAD_NULL);
+}
+
+/*
+ *     Each processor has a dedicated thread which
+ *     executes the idle loop when there is no suitable
+ *     previous context.
+ */
+void
+idle_thread(void)
+{
+       processor_t             processor = current_processor();
+       thread_t                new_thread;
+
+       new_thread = processor_idle(THREAD_NULL, processor);
+       if (new_thread != THREAD_NULL) {
+               thread_run(processor->idle_thread, (thread_continue_t)idle_thread, NULL, new_thread);
+               /*NOTREACHED*/
+       }
 
-       panic("idle_thread: state %d\n", processor->state);
+       thread_block((thread_continue_t)idle_thread);
        /*NOTREACHED*/
 }
 
@@ -2707,8 +4065,6 @@ idle_thread_create(
        return (KERN_SUCCESS);
 }
 
-static uint64_t                sched_tick_deadline;
-
 /*
  * sched_startup:
  *
@@ -2722,14 +4078,16 @@ sched_startup(void)
        kern_return_t   result;
        thread_t                thread;
 
-       result = kernel_thread_start_priority((thread_continue_t)sched_tick_thread, NULL, MAXPRI_KERNEL, &thread);
+       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");
 
        thread_deallocate(thread);
 
        /*
-        * Yield to the sched_tick_thread while it times
+        * Yield to the sched_init_thread while it times
         * a series of context switches back.  It stores
         * the baseline value in sched_cswtime.
         *
@@ -2738,20 +4096,20 @@ sched_startup(void)
         */
        while (sched_cswtime == 0)
                thread_block(THREAD_CONTINUE_NULL);
+}
 
-       thread_daemon_init();
+#if defined(CONFIG_SCHED_TRADITIONAL)
 
-       thread_call_initialize();
-}
+static uint64_t                        sched_tick_deadline = 0;
 
 /*
- *     sched_tick_thread:
+ *     sched_init_thread:
  *
  *     Perform periodic bookkeeping functions about ten
  *     times per second.
  */
 static void
-sched_tick_continue(void)
+sched_traditional_tick_continue(void)
 {
        uint64_t                        abstime = mach_absolute_time();
 
@@ -2768,14 +4126,44 @@ sched_tick_continue(void)
         */
        thread_update_scan();
 
+       if (sched_tick_deadline == 0)
+               sched_tick_deadline = abstime;
+       
        clock_deadline_for_periodic_event(sched_tick_interval, abstime,
                                                                                                                &sched_tick_deadline);
 
-       assert_wait_deadline((event_t)sched_tick_thread, THREAD_UNINT, sched_tick_deadline);
-       thread_block((thread_continue_t)sched_tick_continue);
+       assert_wait_deadline((event_t)sched_traditional_tick_continue, THREAD_UNINT, sched_tick_deadline);
+       thread_block((thread_continue_t)sched_traditional_tick_continue);
        /*NOTREACHED*/
 }
 
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
+static uint32_t
+time_individual_cswitch(void)
+{
+       uint32_t switches = 0;
+       uint64_t newtime, starttime;
+
+       /* Wait for absolute time to increase. */
+       starttime = mach_absolute_time();
+       do {
+               newtime = mach_absolute_time();
+       } while (newtime == starttime);
+
+       /* Measure one or more context switches until time increases again.
+        * This ensures we get non-zero timings even if absolute time
+        * increases very infrequently compared to CPU clock. */
+       starttime = newtime;
+       do {
+               thread_block(THREAD_CONTINUE_NULL);
+               newtime = mach_absolute_time();
+               ++switches;
+       } while (newtime == starttime);
+       /* Round up. */
+       return (uint32_t) ((newtime - starttime + switches - 1) / switches);
+}
+
 /*
  * Time a series of context switches to determine
  * a baseline.  Toss the high and low and return
@@ -2785,15 +4173,11 @@ static uint32_t
 time_cswitch(void)
 {
        uint32_t        new, hi, low, accum;
-       uint64_t        abstime;
-       int                     i, tries = 7;
+       int                     i, tries = 7, denom;
 
        accum = hi = low = 0;
        for (i = 0; i < tries; ++i) {
-               abstime = mach_absolute_time();
-               thread_block(THREAD_CONTINUE_NULL);
-
-               new = mach_absolute_time() - abstime;
+               new = time_individual_cswitch();
 
                if (i == 0)
                        accum = hi = low = new;
@@ -2806,21 +4190,24 @@ time_cswitch(void)
                        accum += new;
                }
        }
-
-       return ((accum - hi - low) / (2 * (tries - 2)));
+       /* Round up. */
+       denom = 2 * (tries - 2);
+       return (accum - hi - low + denom - 1) / denom;
 }
 
 void
-sched_tick_thread(void)
+sched_init_thread(void (*continuation)(void))
 {
        sched_cswtime = time_cswitch();
+       assert(sched_cswtime > 0);
 
-       sched_tick_deadline = mach_absolute_time();
+       continuation();
 
-       sched_tick_continue();
        /*NOTREACHED*/
 }
 
+#if defined(CONFIG_SCHED_TRADITIONAL)
+
 /*
  *     thread_update_scan / runq_scan:
  *
@@ -2860,7 +4247,7 @@ runq_scan(
                while (count > 0) {
                        queue_iterate(q, thread, thread_t, links) {
                                if (            thread->sched_stamp != sched_tick               &&
-                                               (thread->sched_mode & TH_MODE_TIMESHARE)        ) {
+                                               (thread->sched_mode == TH_MODE_TIMESHARE)       ) {
                                        if (thread_update_count == THREAD_UPDATE_SIZE)
                                                return (TRUE);
 
@@ -2881,41 +4268,38 @@ runq_scan(
 static void
 thread_update_scan(void)
 {
-       register boolean_t                      restart_needed;
-       register processor_set_t        pset = &default_pset;
-       register processor_t            processor;
-       register thread_t                       thread;
-       spl_t                                           s;
+       boolean_t                       restart_needed = FALSE;
+       processor_t                     processor = processor_list;
+       processor_set_t         pset;
+       thread_t                        thread;
+       spl_t                           s;
 
        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;
+               do {
+                       pset = processor->processor_set;
 
-                               thread = processor->idle_thread;
-                               if (thread->sched_stamp != sched_tick) {
-                                       if (thread_update_count == THREAD_UPDATE_SIZE) {
-                                               restart_needed = TRUE;
-                                               break;
-                                       }
+                       s = splsched();
+                       pset_lock(pset);
 
-                                       thread_update_array[thread_update_count++] = thread;
-                                       thread_reference_internal(thread);
+                       restart_needed = runq_scan(runq_for_processor(processor));
+
+                       pset_unlock(pset);
+                       splx(s);
+
+                       if (restart_needed)
+                               break;
+
+                       thread = processor->idle_thread;
+                       if (thread != THREAD_NULL && thread->sched_stamp != sched_tick) {
+                               if (thread_update_count == THREAD_UPDATE_SIZE) {
+                                       restart_needed = TRUE;
+                                       break;
                                }
 
-                               processor = (processor_t)queue_next(&processor->processors);
+                               thread_update_array[thread_update_count++] = thread;
+                               thread_reference_internal(thread);
                        }
-                       simple_unlock(&pset->sched_lock);
-               }
-               splx(s);
+               } while ((processor = processor->processor_list) != NULL);
 
            /*
             *  Ok, we now have a collection of candidates -- fix them.
@@ -2926,9 +4310,10 @@ thread_update_scan(void)
 
                        s = splsched();
                        thread_lock(thread);
-                       if (    !(thread->state & (TH_WAIT|TH_SUSP))    &&
-                                               thread->sched_stamp != sched_tick       )
-                               update_priority(thread);
+                       if (    !(thread->state & (TH_WAIT))    ) {
+                               if (SCHED(can_update_priority)(thread))
+                                       SCHED(update_priority)(thread);
+                       }
                        thread_unlock(thread);
                        splx(s);
 
@@ -2936,20 +4321,115 @@ thread_update_scan(void)
            }
        } while (restart_needed);
 }
+
+#endif /* CONFIG_SCHED_TRADITIONAL */
+
+boolean_t
+thread_eager_preemption(thread_t thread) 
+{
+       return ((thread->sched_flags & TH_SFLAG_EAGERPREEMPT) != 0);
+}
+
+void
+thread_set_eager_preempt(thread_t thread) 
+{
+       spl_t x;
+       processor_t p;
+       ast_t ast = AST_NONE;
+
+       x = splsched();
+       p = current_processor();
+
+       thread_lock(thread);
+       thread->sched_flags |= TH_SFLAG_EAGERPREEMPT;
+
+       if (thread == current_thread()) {
+               thread_unlock(thread);
+
+               ast = csw_check(p);
+               if (ast != AST_NONE) {
+                       (void) thread_block_reason(THREAD_CONTINUE_NULL, NULL, ast);
+               }
+       } else {
+               p = thread->last_processor;
+
+               if (p != PROCESSOR_NULL && p->state == PROCESSOR_RUNNING &&
+                       p->active_thread == thread) {
+                       cause_ast_check(p);
+               }
                
+               thread_unlock(thread);
+       }
+
+       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);
+}
+/*
+ * Scheduling statistics
+ */
+void
+sched_stats_handle_csw(processor_t processor, int reasons, int selfpri, int otherpri)
+{
+       struct processor_sched_statistics *stats;
+       boolean_t to_realtime = FALSE;
+       
+       stats = &processor->processor_data.sched_stats;
+       stats->csw_count++;
+
+       if (otherpri >= BASEPRI_REALTIME) {
+               stats->rt_sched_count++;
+               to_realtime = TRUE;
+       }
+
+       if ((reasons & AST_PREEMPT) != 0) {
+               stats->preempt_count++;
+
+               if (selfpri >= BASEPRI_REALTIME) {
+                       stats->preempted_rt_count++;
+               } 
+
+               if (to_realtime) {
+                       stats->preempted_by_rt_count++;
+               }
+
+       }
+}
+
+void
+sched_stats_handle_runq_change(struct runq_stats *stats, int old_count) 
+{
+       uint64_t timestamp = mach_absolute_time();
+
+       stats->count_sum += (timestamp - stats->last_change_timestamp) * old_count;
+       stats->last_change_timestamp = timestamp;
+}
+
 /*
- *     Just in case someone doesn't use the macro
+ *     For calls from assembly code
  */
-#undef thread_wakeup
+#undef thread_wakeup
 void
 thread_wakeup(
-       event_t         x);
+       event_t         x);
 
 void
 thread_wakeup(
-       event_t         x)
+       event_t         x)
 {
-       thread_wakeup_with_result(x, THREAD_AWAKENED);
+       thread_wakeup_with_result(x, THREAD_AWAKENED);
 }
 
 boolean_t
@@ -2982,9 +4462,9 @@ db_sched(void)
 #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",
+       iprintf("Idle thread:\n\thandoff %d block %d\n",
                c_idle_thread_handoff,
-               c_idle_thread_block, no_dispatch_count);
+               c_idle_thread_block);
        iprintf("Sched thread blocks:  %d\n", c_sched_thread_block);
 #endif /* MACH_COUNTERS */
        db_indent -= 2;