]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/sched_prim.c
xnu-1228.12.14.tar.gz
[apple/xnu.git] / osfmk / kern / sched_prim.c
index c45c31048d49f32cba59e8cb75fceb1e491cd287..4e281607d965916f54c0699912848aeb07baf96e 100644 (file)
@@ -1,23 +1,29 @@
 /*
- * Copyright (c) 2000 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 <debug.h>
-#include <cpus.h>
 #include <mach_kdb.h>
-#include <simple_clock.h>
-#include <power_save.h>
-#include <task_swapper.h>
 
 #include <ddb/db_output.h>
+
+#include <mach/mach_types.h>
 #include <mach/machine.h>
+#include <mach/policy.h>
+#include <mach/sync_policy.h>
+
 #include <machine/machine_routines.h>
 #include <machine/sched_param.h>
-#include <kern/ast.h>
+#include <machine/machine_cpu.h>
+
+#include <kern/kern_types.h>
 #include <kern/clock.h>
 #include <kern/counters.h>
 #include <kern/cpu_number.h>
 #include <kern/cpu_data.h>
-#include <kern/etap_macros.h>
+#include <kern/debug.h>
 #include <kern/lock.h>
 #include <kern/macro_help.h>
 #include <kern/machine.h>
 #include <kern/syscall_subr.h>
 #include <kern/task.h>
 #include <kern/thread.h>
-#include <kern/thread_swap.h>
+#include <kern/wait_queue.h>
+
 #include <vm/pmap.h>
 #include <vm/vm_kern.h>
 #include <vm/vm_map.h>
-#include <mach/policy.h>
-#include <mach/sync_policy.h>
-#include <kern/mk_sp.h>        /*** ??? fix so this can be removed ***/
+
 #include <sys/kdebug.h>
 
-#if    TASK_SWAPPER
-#include <kern/task_swap.h>
-extern int     task_swap_on;
-#endif /* TASK_SWAPPER */
+#include <kern/pms.h>
 
-extern int     hz;
+struct run_queue       rt_runq;
+#define RT_RUNQ                ((processor_t)-1)
+decl_simple_lock_data(static,rt_lock);
 
 #define                DEFAULT_PREEMPTION_RATE         100             /* (1/s) */
 int                    default_preemption_rate = DEFAULT_PREEMPTION_RATE;
@@ -114,56 +121,69 @@ 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;
+
+uint32_t       std_quantum;
+uint32_t       min_std_quantum;
+
 uint32_t       std_quantum_us;
 
+uint32_t       max_rt_quantum;
+uint32_t       min_rt_quantum;
+
+uint32_t       sched_cswtime;
+
 unsigned       sched_tick;
+uint32_t       sched_tick_interval;
 
-#if    SIMPLE_CLOCK
-int                    sched_usec;
-#endif /* SIMPLE_CLOCK */
+uint32_t       sched_pri_shift = INT8_MAX;
+uint32_t       sched_fixed_shift;
+
+uint32_t       sched_run_count, sched_share_count;
+uint32_t       sched_load_average, sched_mach_factor;
 
 /* Forwards */
-void           wait_queues_init(void);
+void wait_queues_init(void) __attribute__((section("__TEXT, initcode")));
+
+static void load_shift_init(void) __attribute__((section("__TEXT, initcode")));
+static void preempt_pri_init(void) __attribute__((section("__TEXT, initcode")));
 
-thread_t       choose_pset_thread(
-                               processor_t                     myprocessor,
-                               processor_set_t         pset);
+static thread_t        run_queue_dequeue(
+                                       run_queue_t             runq,
+                                       integer_t               options);
 
-thread_t       choose_thread(
-                               processor_t             myprocessor);
+static thread_t        thread_select_idle(
+                                       thread_t                        thread,
+                                       processor_t                     processor);
 
-boolean_t      run_queue_enqueue(
-                               run_queue_t             runq,
-                               thread_t                thread,
-                               boolean_t               tail);
+static thread_t        processor_idle(
+                                       thread_t                        thread,
+                                       processor_t                     processor);
 
-void           do_thread_scan(void);
+static thread_t        steal_thread(
+                                       processor_set_t         pset);
+
+static thread_t        steal_processor_thread(
+                                       processor_t                     processor);
+
+static void            thread_update_scan(void);
 
 #if    DEBUG
-void           dump_run_queues(
-                               run_queue_t                     rq);
-void           dump_run_queue_struct(
-                               run_queue_t                     rq);
-void           dump_processor(
-                               processor_t             p);
-void           dump_processor_set(
-                               processor_set_t         ps);
-
-void           checkrq(
-                               run_queue_t             rq,
-                               char                    *msg);
-
-void           thread_check(
-                               thread_t                thread,
-                               run_queue_t             runq);
+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(
                                thread_t                thread);
 
 #endif /*DEBUG*/
 
-
 /*
  *     State machine
  *
@@ -223,6 +243,9 @@ struct wait_queue wait_queues[NUMQUEUES];
 #define wait_hash(event) \
        ((((int)(event) < 0)? ~(int)(event): (int)(event)) % NUMQUEUES)
 
+int8_t         sched_load_shifts[NRQS];
+int                    sched_preempt_pri[NRQBM];
+
 void
 sched_init(void)
 {
@@ -236,16 +259,65 @@ 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();
-       pset_sys_bootstrap();           /* initialize processor mgmt. */
-       processor_action();
+       load_shift_init();
+       preempt_pri_init();
+       simple_lock_init(&rt_lock, 0);
+       run_queue_init(&rt_runq);
        sched_tick = 0;
-#if    SIMPLE_CLOCK
-       sched_usec = 0;
-#endif /* SIMPLE_CLOCK */
        ast_init();
 }
 
+void
+sched_timebase_init(void)
+{
+       uint64_t        abstime;
+       uint32_t        shift;
+
+       /* standard timeslicing quantum */
+       clock_interval_to_absolutetime_interval(
+                                                       std_quantum_us, NSEC_PER_USEC, &abstime);
+       assert((abstime >> 32) == 0 && (uint32_t)abstime != 0);
+       std_quantum = 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;
+
+       /* 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;
+
+       /*
+        * Compute conversion factor from usage to
+        * timesharing priorities with 5/8 ** n aging.
+        */
+       abstime = (abstime * 5) / 3;
+       for (shift = 0; abstime > BASEPRI_DEFAULT; ++shift)
+               abstime >>= 1;
+       sched_fixed_shift = shift;
+
+       max_unsafe_computation = max_unsafe_quanta * std_quantum;
+       max_poll_computation = max_poll_quanta * std_quantum;
+}
+
 void
 wait_queues_init(void)
 {
@@ -256,32 +328,56 @@ wait_queues_init(void)
        }
 }
 
+/*
+ * Set up values for timeshare
+ * loading factors.
+ */
+static void
+load_shift_init(void)
+{
+       int8_t          k, *p = sched_load_shifts;
+       uint32_t        i, j;
+
+       *p++ = INT8_MIN; *p++ = 0;
+
+       for (i = j = 2, k = 1; i < NRQS; ++k) {
+               for (j <<= 1; i < j; ++i)
+                       *p++ = k;
+       }
+}
+
+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);
+}
+
 /*
  *     Thread wait timer expiration.
  */
 void
 thread_timer_expire(
-       timer_call_param_t              p0,
-       timer_call_param_t              p1)
+       void                    *p0,
+       __unused void   *p1)
 {
        thread_t                thread = p0;
        spl_t                   s;
 
        s = splsched();
-       wake_lock(thread);
-       if (--thread->wait_timer_active == 1) {
+       thread_lock(thread);
+       if (--thread->wait_timer_active == 0) {
                if (thread->wait_timer_is_set) {
                        thread->wait_timer_is_set = FALSE;
-                       thread_lock(thread);
-                       if (thread->active)
-                               clear_wait_internal(thread, THREAD_TIMED_OUT);
-                       thread_unlock(thread);
+                       clear_wait_internal(thread, THREAD_TIMED_OUT);
                }
        }
-       else
-       if (thread->wait_timer_active == 0)
-               thread_wakeup_one(&thread->wait_timer_active);
-       wake_unlock(thread);
+       thread_unlock(thread);
        splx(s);
 }
 
@@ -302,17 +398,14 @@ thread_set_timer(
        spl_t                   s;
 
        s = splsched();
-       wake_lock(thread);
        thread_lock(thread);
        if ((thread->state & TH_WAIT) != 0) {
                clock_interval_to_deadline(interval, scale_factor, &deadline);
-               timer_call_enter(&thread->wait_timer, deadline);
-               assert(!thread->wait_timer_is_set);
-               thread->wait_timer_active++;
+               if (!timer_call_enter(&thread->wait_timer, deadline))
+                       thread->wait_timer_active++;
                thread->wait_timer_is_set = TRUE;
        }
        thread_unlock(thread);
-       wake_unlock(thread);
        splx(s);
 }
 
@@ -324,16 +417,13 @@ thread_set_timer_deadline(
        spl_t                   s;
 
        s = splsched();
-       wake_lock(thread);
        thread_lock(thread);
        if ((thread->state & TH_WAIT) != 0) {
-               timer_call_enter(&thread->wait_timer, deadline);
-               assert(!thread->wait_timer_is_set);
-               thread->wait_timer_active++;
+               if (!timer_call_enter(&thread->wait_timer, deadline))
+                       thread->wait_timer_active++;
                thread->wait_timer_is_set = TRUE;
        }
        thread_unlock(thread);
-       wake_unlock(thread);
        splx(s);
 }
 
@@ -344,92 +434,104 @@ thread_cancel_timer(void)
        spl_t                   s;
 
        s = splsched();
-       wake_lock(thread);
+       thread_lock(thread);
        if (thread->wait_timer_is_set) {
                if (timer_call_cancel(&thread->wait_timer))
                        thread->wait_timer_active--;
                thread->wait_timer_is_set = FALSE;
        }
-       wake_unlock(thread);
+       thread_unlock(thread);
        splx(s);
 }
 
 /*
- * Set up thread timeout element when thread is created.
+ *     thread_unblock:
+ *
+ *     Unblock thread on wake up.
+ *
+ *     Returns TRUE if the thread is still running.
+ *
+ *     Thread must be locked.
  */
-void
-thread_timer_setup(
-        thread_t               thread)
+boolean_t
+thread_unblock(
+       thread_t                thread,
+       wait_result_t   wresult)
 {
-       extern void     thread_depress_expire(
-                                       timer_call_param_t      p0,
-                                       timer_call_param_t      p1);
-
-       timer_call_setup(&thread->wait_timer, thread_timer_expire, thread);
-       thread->wait_timer_is_set = FALSE;
-       thread->wait_timer_active = 1;
+       boolean_t               result = FALSE;
 
-       timer_call_setup(&thread->depress_timer, thread_depress_expire, thread);
-       thread->depress_timer_active = 1;
-
-       thread->ref_count++;
-}
-
-void
-thread_timer_terminate(void)
-{
-       thread_t                thread = current_thread();
-       wait_result_t   res;
-       spl_t                   s;
+       /*
+        *      Set wait_result.
+        */
+       thread->wait_result = wresult;
 
-       s = splsched();
-       wake_lock(thread);
+       /*
+        *      Cancel pending wait timer.
+        */
        if (thread->wait_timer_is_set) {
                if (timer_call_cancel(&thread->wait_timer))
                        thread->wait_timer_active--;
                thread->wait_timer_is_set = FALSE;
        }
 
-       thread->wait_timer_active--;
+       /*
+        *      Update scheduling state: not waiting,
+        *      set running.
+        */
+       thread->state &= ~(TH_WAIT|TH_UNINT);
 
-       while (thread->wait_timer_active > 0) {
-               res = assert_wait((event_t)&thread->wait_timer_active, THREAD_UNINT);
-               assert(res == THREAD_WAITING);
-               wake_unlock(thread);
-               splx(s);
+       if (!(thread->state & TH_RUN)) {
+               thread->state |= TH_RUN;
 
-               res = thread_block(THREAD_CONTINUE_NULL);
-               assert(res == THREAD_AWAKENED);
+               (*thread->sched_call)(SCHED_CALL_UNBLOCK, thread);
 
-               s = splsched();
-               wake_lock(thread);
+               /*
+                *      Update run counts.
+                */
+               sched_run_incr();
+               if (thread->sched_mode & TH_MODE_TIMESHARE)
+                       sched_share_incr();
        }
+       else {
+               /*
+                *      Signal if idling on another processor.
+                */
+               if (thread->state & TH_IDLE) {
+                       processor_t             processor = thread->last_processor;
 
-       thread->depress_timer_active--;
-
-       while (thread->depress_timer_active > 0) {
-               res = assert_wait((event_t)&thread->depress_timer_active, THREAD_UNINT);
-               assert(res == THREAD_WAITING);
-               wake_unlock(thread);
-               splx(s);
+                       if (processor != current_processor())
+                               machine_signal_idle(processor);
+               }
 
-               res = thread_block(THREAD_CONTINUE_NULL);
-               assert(res == THREAD_AWAKENED);
+               result = TRUE;
+       }
 
-               s = splsched();
-               wake_lock(thread);
+       /*
+        * Calculate deadline for real-time threads.
+        */
+       if (thread->sched_mode & TH_MODE_REALTIME) {
+               thread->realtime.deadline = mach_absolute_time();
+               thread->realtime.deadline += thread->realtime.constraint;
        }
 
-       wake_unlock(thread);
-       splx(s);
+       /*
+        * Clear old quantum, fail-safe computation, etc.
+        */
+       thread->current_quantum = 0;
+       thread->computation_metered = 0;
+       thread->reason = AST_NONE;
 
-       thread_deallocate(thread);
+       KERNEL_DEBUG_CONSTANT(
+               MACHDBG_CODE(DBG_MACH_SCHED,MACH_MAKE_RUNNABLE) | DBG_FUNC_NONE,
+                                       (int)thread, (int)thread->sched_pri, 0, 0, 0);
+
+       return (result);
 }
 
 /*
- *     Routine:        thread_go_locked
+ *     Routine:        thread_go
  *     Purpose:
- *             Start a thread running.
+ *             Unblock and dispatch thread.
  *     Conditions:
  *             thread lock held, IPC locks may be held.
  *             thread must have been pulled from wait queue under same lock hold.
@@ -438,32 +540,22 @@ thread_timer_terminate(void)
  *             KERN_NOT_WAITING - Thread was not waiting
  */
 kern_return_t
-thread_go_locked(
+thread_go(
        thread_t                thread,
-       wait_result_t   result)
+       wait_result_t   wresult)
 {
        assert(thread->at_safe_point == FALSE);
        assert(thread->wait_event == NO_EVENT64);
        assert(thread->wait_queue == WAIT_QUEUE_NULL);
 
        if ((thread->state & (TH_WAIT|TH_TERMINATE)) == TH_WAIT) {
-               thread->state &= ~(TH_WAIT|TH_UNINT);
-               if (!(thread->state & TH_RUN)) {
-                       thread->state |= TH_RUN;
-
-                       if (thread->active_callout)
-                               call_thread_unblock();
+               if (!thread_unblock(thread, wresult))
+                       thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
 
-                       if (!(thread->state & TH_IDLE)) {
-                               _mk_sp_thread_unblock(thread);
-                               hw_atomic_add(&thread->processor_set->run_count, 1);
-                       }
-               }
-
-               thread->wait_result = result;
-               return KERN_SUCCESS;
+               return (KERN_SUCCESS);
        }
-       return KERN_NOT_WAITING;
+
+       return (KERN_NOT_WAITING);
 }
 
 /*
@@ -481,10 +573,7 @@ thread_mark_wait_locked(
        thread_t                        thread,
        wait_interrupt_t        interruptible)
 {
-       wait_result_t   wait_result;
-       boolean_t                       at_safe_point;
-
-       assert(thread == current_thread());
+       boolean_t               at_safe_point;
 
        /*
         *      The thread may have certain types of interrupts/aborts masked
@@ -492,21 +581,23 @@ thread_mark_wait_locked(
         *      are OK, we have to honor mask settings (outer-scoped code may
         *      not be able to handle aborts at the moment).
         */
-       if (interruptible > thread->interrupt_level)
-               interruptible = thread->interrupt_level;
+       if (interruptible > (thread->options & TH_OPT_INTMASK))
+               interruptible = thread->options & TH_OPT_INTMASK;
 
        at_safe_point = (interruptible == THREAD_ABORTSAFE);
 
-       if ((interruptible == THREAD_UNINT) || 
-               !(thread->state & TH_ABORT) ||
-               (!at_safe_point && (thread->state & TH_ABORT_SAFELY))) {
+       if (    interruptible == THREAD_UNINT                   ||
+                       !(thread->sched_mode & TH_MODE_ABORT)   ||
+                       (!at_safe_point &&
+                               (thread->sched_mode & TH_MODE_ABORTSAFELY))) {
                thread->state |= (interruptible) ? TH_WAIT : (TH_WAIT | TH_UNINT);
                thread->at_safe_point = at_safe_point;
-               thread->sleep_stamp = sched_tick;
                return (thread->wait_result = THREAD_WAITING);
-       } else if (thread->state & TH_ABORT_SAFELY) {
-               thread->state &= ~(TH_ABORT|TH_ABORT_SAFELY);
        }
+       else
+       if (thread->sched_mode & TH_MODE_ABORTSAFELY)
+               thread->sched_mode &= ~TH_MODE_ISABORTED;
+
        return (thread->wait_result = THREAD_INTERRUPTED);
 }
 
@@ -529,31 +620,11 @@ thread_interrupt_level(
        wait_interrupt_t new_level)
 {
        thread_t thread = current_thread();
-       wait_interrupt_t result = thread->interrupt_level;
-
-       thread->interrupt_level = new_level;
-       return result;
-}
-
-/*
- *     Routine:        assert_wait_timeout
- *     Purpose:
- *             Assert that the thread intends to block,
- *             waiting for a timeout (no user known event).
- */
-unsigned int assert_wait_timeout_event;
+       wait_interrupt_t result = thread->options & TH_OPT_INTMASK;
 
-wait_result_t
-assert_wait_timeout(
-       mach_msg_timeout_t              msecs,
-       wait_interrupt_t                interruptible)
-{
-       wait_result_t res;
+       thread->options = (thread->options & ~TH_OPT_INTMASK) | (new_level & TH_OPT_INTMASK);
 
-       res = assert_wait((event_t)&assert_wait_timeout_event, interruptible);
-       if (res == THREAD_WAITING)
-               thread_set_timer(msecs, 1000*NSEC_PER_USEC);
-       return res;
+       return result;
 }
 
 /*
@@ -568,7 +639,6 @@ assert_wait_possible(void)
 {
 
        thread_t thread;
-       extern unsigned int debug_mode;
 
 #if    DEBUG
        if(debug_mode) return TRUE;             /* Always succeed in debug mode */
@@ -594,13 +664,70 @@ assert_wait(
        register int            index;
 
        assert(event != NO_EVENT);
-       assert(assert_wait_possible());
 
        index = wait_hash(event);
        wq = &wait_queues[index];
-       return wait_queue_assert_wait(wq, event, interruptible);
+       return wait_queue_assert_wait(wq, event, interruptible, 0);
+}
+
+wait_result_t
+assert_wait_timeout(
+       event_t                         event,
+       wait_interrupt_t        interruptible,
+       uint32_t                        interval,
+       uint32_t                        scale_factor)
+{
+       thread_t                        thread = current_thread();
+       wait_result_t           wresult;
+       wait_queue_t            wqueue;
+       uint64_t                        deadline;
+       spl_t                           s;
+
+       assert(event != NO_EVENT);
+       wqueue = &wait_queues[wait_hash(event)];
+
+       s = splsched();
+       wait_queue_lock(wqueue);
+       thread_lock(thread);
+
+       clock_interval_to_deadline(interval, scale_factor, &deadline);
+       wresult = wait_queue_assert_wait64_locked(wqueue, (uint32_t)event,
+                                                                                                       interruptible, deadline, thread);
+
+       thread_unlock(thread);
+       wait_queue_unlock(wqueue);
+       splx(s);
+
+       return (wresult);
 }
 
+wait_result_t
+assert_wait_deadline(
+       event_t                         event,
+       wait_interrupt_t        interruptible,
+       uint64_t                        deadline)
+{
+       thread_t                        thread = current_thread();
+       wait_result_t           wresult;
+       wait_queue_t            wqueue;
+       spl_t                           s;
+
+       assert(event != NO_EVENT);
+       wqueue = &wait_queues[wait_hash(event)];
+
+       s = splsched();
+       wait_queue_lock(wqueue);
+       thread_lock(thread);
+
+       wresult = wait_queue_assert_wait64_locked(wqueue, (uint32_t)event,
+                                                                                                       interruptible, deadline, thread);
+
+       thread_unlock(thread);
+       wait_queue_unlock(wqueue);
+       splx(s);
+
+       return (wresult);
+}
 
 /*
  *     thread_sleep_fast_usimple_lock:
@@ -689,8 +816,6 @@ thread_sleep_mutex(
  *     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.
- *
- *     JMM - Add hint to make sure mutex is available before rousting
  */
 wait_result_t
 thread_sleep_mutex_deadline(
@@ -701,13 +826,10 @@ thread_sleep_mutex_deadline(
 {
        wait_result_t   res;
 
-       res = assert_wait(event, interruptible);
+       res = assert_wait_deadline(event, interruptible, deadline);
        if (res == THREAD_WAITING) {
                mutex_unlock(mutex);
-               thread_set_timer_deadline(deadline);
                res = thread_block(THREAD_CONTINUE_NULL);
-               if (res != THREAD_TIMED_OUT)
-                       thread_cancel_timer();
                mutex_lock(mutex);
        }
        return res;
@@ -719,8 +841,6 @@ thread_sleep_mutex_deadline(
  *     Cause the current thread to wait until the specified event
  *     occurs.  The specified (write) lock is unlocked before releasing
  *     the cpu. The (write) lock will be re-acquired before returning.
- *
- *     JMM - Add hint to make sure mutex is available before rousting
  */
 wait_result_t
 thread_sleep_lock_write(
@@ -739,95 +859,66 @@ thread_sleep_lock_write(
        return res;
 }
 
-
-/*
- *     thread_sleep_funnel:
- *
- *     Cause the current thread to wait until the specified event
- *     occurs.  If the thread is funnelled, the funnel will be released
- *     before giving up the cpu. The funnel will be re-acquired before returning.
- *
- *     JMM - Right now the funnel is dropped and re-acquired inside
- *               thread_block().  At some point, this may give thread_block() a hint.
- */
-wait_result_t
-thread_sleep_funnel(
-       event_t                 event,
-       wait_interrupt_t interruptible)
-{
-       wait_result_t   res;
-
-       res = assert_wait(event, interruptible);
-       if (res == THREAD_WAITING) {
-               res = thread_block(THREAD_CONTINUE_NULL);
-       }
-       return res;
-}
-
 /*
- * thread_[un]stop(thread)
- *     Once a thread has blocked interruptibly (via assert_wait) prevent 
- *     it from running until thread_unstop.
+ * thread_stop:
  *
- *     If someone else has already stopped the thread, wait for the
- *     stop to be cleared, and then stop it again.
+ * Force a preemption point for a thread and wait
+ * for it to stop running.  Arbitrates access among
+ * multiple stop requests. (released by unstop)
  *
- *     Return FALSE if interrupted.
+ * The thread must enter a wait state and stop via a
+ * separate means.
  *
- * NOTE: thread_hold/thread_suspend should be called on the activation
- *     before calling thread_stop.  TH_SUSP is only recognized when
- *     a thread blocks and only prevents clear_wait/thread_wakeup
- *     from restarting an interruptible wait.  The wake_active flag is
- *     used to indicate that someone is waiting on the thread.
+ * Returns FALSE if interrupted.
  */
 boolean_t
 thread_stop(
-       thread_t        thread)
+       thread_t                thread)
 {
-       spl_t           s = splsched();
+       wait_result_t   wresult;
+       spl_t                   s = splsched();
 
        wake_lock(thread);
+       thread_lock(thread);
 
        while (thread->state & TH_SUSP) {
-               wait_result_t   result;
-
                thread->wake_active = TRUE;
-               result = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
+               thread_unlock(thread);
+
+               wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
                wake_unlock(thread);
                splx(s);
 
-               if (result == THREAD_WAITING)
-                       result = thread_block(THREAD_CONTINUE_NULL);
+               if (wresult == THREAD_WAITING)
+                       wresult = thread_block(THREAD_CONTINUE_NULL);
 
-               if (result != THREAD_AWAKENED)
+               if (wresult != THREAD_AWAKENED)
                        return (FALSE);
 
                s = splsched();
                wake_lock(thread);
+               thread_lock(thread);
        }
 
-       thread_lock(thread);
        thread->state |= TH_SUSP;
 
        while (thread->state & TH_RUN) {
-               wait_result_t   result;
                processor_t             processor = thread->last_processor;
 
-               if (    processor != PROCESSOR_NULL                                             &&
-                               processor->state == PROCESSOR_RUNNING                   &&
-                               processor->cpu_data->active_thread == thread    )
+               if (processor != PROCESSOR_NULL && processor->active_thread == thread)
                        cause_ast_check(processor);
-               thread_unlock(thread);
 
                thread->wake_active = TRUE;
-               result = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
+               thread_unlock(thread);
+
+               wresult = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
                wake_unlock(thread);
                splx(s);
 
-               if (result == THREAD_WAITING)
-                       result = thread_block(THREAD_CONTINUE_NULL);
+               if (wresult == THREAD_WAITING)
+                       wresult = thread_block(THREAD_CONTINUE_NULL);
 
-               if (result != THREAD_AWAKENED) {
+               if (wresult != THREAD_AWAKENED) {
                        thread_unstop(thread);
                        return (FALSE);
                }
@@ -845,8 +936,12 @@ thread_stop(
 }
 
 /*
- *     Clear TH_SUSP and if the thread has been stopped and is now runnable,
- *     put it back on the run queue.
+ * thread_unstop:
+ *
+ * Release a previous stop request and set
+ * the thread running if appropriate.
+ *
+ * Use only after a successful stop operation.
  */
 void
 thread_unstop(
@@ -859,11 +954,9 @@ thread_unstop(
 
        if ((thread->state & (TH_RUN|TH_WAIT|TH_SUSP)) == TH_SUSP) {
                thread->state &= ~TH_SUSP;
-               thread->state |= TH_RUN;
+               thread_unblock(thread, THREAD_AWAKENED);
 
-               assert(!(thread->state & TH_IDLE));
-               _mk_sp_thread_unblock(thread);
-               hw_atomic_add(&thread->processor_set->run_count, 1);
+               thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
        }
        else
        if (thread->state & TH_SUSP) {
@@ -872,10 +965,11 @@ 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;
                }
        }
@@ -886,37 +980,36 @@ thread_unstop(
 }
 
 /*
- * Wait for the thread's RUN bit to clear
+ * thread_wait:
+ *
+ * Wait for a thread to stop running. (non-interruptible)
+ *
  */
-boolean_t
+void
 thread_wait(
-       thread_t        thread)
+       thread_t                thread)
 {
-       spl_t           s = splsched();
+       wait_result_t   wresult;
+       spl_t                   s = splsched();
 
        wake_lock(thread);
        thread_lock(thread);
 
        while (thread->state & TH_RUN) {
-               wait_result_t   result;
                processor_t             processor = thread->last_processor;
 
-               if (    processor != PROCESSOR_NULL                                             &&
-                               processor->state == PROCESSOR_RUNNING                   &&
-                               processor->cpu_data->active_thread == thread    )
+               if (processor != PROCESSOR_NULL && processor->active_thread == thread)
                        cause_ast_check(processor);
-               thread_unlock(thread);
 
                thread->wake_active = TRUE;
-               result = assert_wait(&thread->wake_active, THREAD_ABORTSAFE);
+               thread_unlock(thread);
+
+               wresult = assert_wait(&thread->wake_active, THREAD_UNINT);
                wake_unlock(thread);
                splx(s);
 
-               if (result == THREAD_WAITING)
-                       result = thread_block(THREAD_CONTINUE_NULL);
-
-               if (result != THREAD_AWAKENED)
-                       return (FALSE);
+               if (wresult == THREAD_WAITING)
+                       thread_block(THREAD_CONTINUE_NULL);
 
                s = splsched();
                wake_lock(thread);
@@ -926,8 +1019,6 @@ thread_wait(
        thread_unlock(thread);
        wake_unlock(thread);
        splx(s);
-
-       return (TRUE);
 }
 
 /*
@@ -949,38 +1040,39 @@ thread_wait(
 __private_extern__ kern_return_t
 clear_wait_internal(
        thread_t                thread,
-       wait_result_t   result)
+       wait_result_t   wresult)
 {
        wait_queue_t    wq = thread->wait_queue;
-       kern_return_t   ret;
-       int                             loop_count;
+       int                             i = LockTimeOut;
 
-       loop_count = 0;
        do {
-               if ((result == THREAD_INTERRUPTED) && (thread->state & TH_UNINT))
-                       return KERN_FAILURE;
+               if (wresult == THREAD_INTERRUPTED && (thread->state & TH_UNINT))
+                       return (KERN_FAILURE);
 
                if (wq != WAIT_QUEUE_NULL) {
                        if (wait_queue_lock_try(wq)) {
                                wait_queue_pull_thread_locked(wq, thread, TRUE);
                                /* wait queue unlocked, thread still locked */
-                       } else {
+                       }
+                       else {
                                thread_unlock(thread);
                                delay(1);
+
                                thread_lock(thread);
+                               if (wq != thread->wait_queue)
+                                       return (KERN_NOT_WAITING);
 
-                               if (wq != thread->wait_queue) {
-                                       return KERN_NOT_WAITING; /* we know it moved */
-                               }
                                continue;
                        }
                }
-               ret = thread_go_locked(thread, result);
-               return ret; 
-       } while (++loop_count < LockTimeOut);
-       panic("clear_wait_internal: deadlock: thread=0x%x, wq=0x%x, cpu=%d\n",
+
+               return (thread_go(thread, wresult));
+       } while (--i > 0);
+
+       panic("clear_wait_internal: deadlock: thread=%p, wq=%p, cpu=%d\n",
                  thread, wq, cpu_number());
-       return KERN_FAILURE;
+
+       return (KERN_FAILURE);
 }
 
 
@@ -1038,397 +1130,510 @@ thread_wakeup_prim(
 /*
  *     thread_bind:
  *
- *     Force a thread to execute on the specified processor.
- *     If the thread is currently executing, it may wait until its
- *     time slice is up before switching onto the specified processor.
+ *     Force the current thread to execute on the specified processor.
  *
- *     A processor of PROCESSOR_NULL causes the thread to be unbound.
- *     xxx - DO NOT export this to users.
+ *     Returns the previous binding.  PROCESSOR_NULL means
+ *     not bound.
+ *
+ *     XXX - DO NOT export this to users - XXX
  */
-void
+processor_t
 thread_bind(
-       register thread_t       thread,
-       processor_t                     processor)
+       processor_t             processor)
 {
-       spl_t           s;
+       thread_t                self = current_thread();
+       processor_t             prev;
+       spl_t                   s;
 
        s = splsched();
-       thread_lock(thread);
-       thread_bind_locked(thread, processor);
-       thread_unlock(thread);
+       thread_lock(self);
+
+       prev = self->bound_processor;
+       self->bound_processor = processor;
+
+       thread_unlock(self);
        splx(s);
+
+       return (prev);
 }
 
 /*
- *     Select a thread for this processor (the current processor) to run.
- *     May select the current thread, which must already be locked.
+ *     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    myprocessor)
+       thread_t                        thread,
+       processor_t                     processor)
 {
-       register thread_t               thread;
-       processor_set_t                 pset;
-       register run_queue_t    runq = &myprocessor->runq;
-       boolean_t                               other_runnable;
+       processor_set_t         pset = processor->processor_set;
+       thread_t                        new_thread = THREAD_NULL;
+       boolean_t                       other_runnable, inactive_state;
 
-       /*
-        *      Check for other non-idle runnable threads.
-        */
-       pset = myprocessor->processor_set;
-       thread = myprocessor->cpu_data->active_thread;
+       do {
+               /*
+                *      Update the priority.
+                */
+               if (thread->sched_stamp != sched_tick)
+                       update_priority(thread);
 
-       /* Update the thread's priority */
-       if (thread->sched_stamp != sched_tick)
-               update_priority(thread);
+               processor->current_pri = thread->sched_pri;
+
+               pset_lock(pset);
+
+               inactive_state = processor->state != PROCESSOR_SHUTDOWN && machine_cpu_is_inactive(processor->cpu_num);
 
-       myprocessor->current_pri = thread->sched_pri;
+               simple_lock(&rt_lock);
 
-       simple_lock(&runq->lock);
-       simple_lock(&pset->runq.lock);
+               /*
+                *      Check for other runnable threads.
+                */
+               other_runnable = processor->runq.count > 0 || rt_runq.count > 0;
+
+               /*
+                *      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_RUN                                                 &&
+                               (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.highq >= BASEPRI_RTQUEUES) {
+                                       register run_queue_t    runq = &rt_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 = PROCESSOR_NULL;
+                                               runq->count--; 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);
+                                               }
+                                       }
+                               }
 
-       other_runnable = runq->count > 0 || pset->runq.count > 0;
+                               simple_unlock(&rt_lock);
 
-       if (    thread->state == TH_RUN                                                 &&
-                       (!other_runnable                                                        ||
-                        (runq->highq < thread->sched_pri               &&
-                         pset->runq.highq < thread->sched_pri))                &&
-                       thread->processor_set == pset                                   &&
-                       (thread->bound_processor == PROCESSOR_NULL      ||
-                        thread->bound_processor == myprocessor)                                ) {
+                               processor->deadline = thread->realtime.deadline;
 
-               /* I am the highest priority runnable (non-idle) thread */
-               simple_unlock(&pset->runq.lock);
-               simple_unlock(&runq->lock);
+                               pset_unlock(pset);
 
-               myprocessor->slice_quanta =
-                               (thread->sched_mode & TH_MODE_TIMESHARE)? pset->set_quanta: 1;
-       }
-       else
-       if (other_runnable)
-               thread = choose_thread(myprocessor);
-       else {
-               simple_unlock(&pset->runq.lock);
-               simple_unlock(&runq->lock);
+                               return (thread);
+                       }
+
+                       if (!inactive_state &&
+                                       (!other_runnable                                                        ||
+                                        (processor->runq.highq < thread->sched_pri             &&
+                                         rt_runq.highq < thread->sched_pri))                           ) {
+
+                               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, processor->runq.count);
+
+                               processor->deadline = UINT64_MAX;
+
+                               pset_unlock(pset);
+
+                               return (thread);
+                       }
+               }
+
+               if (other_runnable) {
+                       if (processor->runq.count > 0 && processor->runq.highq >= rt_runq.highq) {
+                               simple_unlock(&rt_lock);
+
+                               thread = run_queue_dequeue(&processor->runq, SCHED_HEADQ);
+
+                               if (!inactive_state) {
+                                       pset_pri_hint(pset, processor, thread->sched_pri);
+
+                                       pset_count_hint(pset, processor, processor->runq.count);
+                               }
+
+                               processor->deadline = UINT64_MAX;
+                               pset_unlock(pset);
+
+                               return (thread);
+                       }
+
+                       thread = run_queue_dequeue(&rt_runq, SCHED_HEADQ);
+                       simple_unlock(&rt_lock);
+
+                       processor->deadline = thread->realtime.deadline;
+                       pset_unlock(pset);
+
+                       return (thread);
+               }
+
+               simple_unlock(&rt_lock);
+
+               processor->deadline = UINT64_MAX;
+
+               if (inactive_state) {
+                       if (processor->state == PROCESSOR_RUNNING)
+                               remqueue(&pset->active_queue, (queue_entry_t)processor);
+                       else
+                       if (processor->state == PROCESSOR_IDLE)
+                               remqueue(&pset->idle_queue, (queue_entry_t)processor);
+
+                       processor->state = PROCESSOR_INACTIVE;
+
+                       pset_unlock(pset);
+
+                       return (processor->idle_thread);
+               }
 
                /*
-                *      Nothing is runnable, so set this processor idle if it
-                *      was running.  If it was in an assignment or shutdown,
-                *      leave it alone.  Return its idle thread.
+                *      No runnable threads, attempt to steal
+                *      from other processors.
                 */
-               simple_lock(&pset->sched_lock);
-               if (myprocessor->state == PROCESSOR_RUNNING) {
-                       remqueue(&pset->active_queue, (queue_entry_t)myprocessor);
-                       myprocessor->state = PROCESSOR_IDLE;
+               new_thread = steal_thread(pset);
+               if (new_thread != THREAD_NULL)
+                       return (new_thread);
 
-                       if (myprocessor == master_processor)
-                               enqueue_tail(&pset->idle_queue, (queue_entry_t)myprocessor);
-                       else
-                               enqueue_head(&pset->idle_queue, (queue_entry_t)myprocessor);
+               /*
+                *      If other threads have appeared, shortcut
+                *      around again.
+                */
+               if (processor->runq.count > 0 || rt_runq.count > 0)
+                       continue;
+
+               pset_lock(pset);
+
+               /*
+                *      Nothing is runnable, so set this processor idle if it
+                *      was running.
+                */
+               if (processor->state == PROCESSOR_RUNNING) {
+                       remqueue(&pset->active_queue, (queue_entry_t)processor);
+                       processor->state = PROCESSOR_IDLE;
 
-                       pset->idle_count++;
+                       enqueue_head(&pset->idle_queue, (queue_entry_t)processor);
+                       pset->low_pri = pset->low_count = processor;
                }
-               simple_unlock(&pset->sched_lock);
 
-               thread = myprocessor->idle_thread;
-       }
+               pset_unlock(pset);
 
-       return (thread);
-}
+               /*
+                *      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)
+                       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);
+
+       } while (new_thread == THREAD_NULL);
 
+       return (new_thread);
+}
 
 /*
- *     Stop running the current thread and start running the new thread.
- *     If continuation is non-zero, and the current thread is blocked,
- *     then it will resume by executing continuation on a new stack.
- *     Returns TRUE if the hand-off succeeds.
+ *     thread_select_idle:
+ *
+ *     Idle the processor using the current thread context.
  *
- *     Assumes splsched.
+ *     Called with thread locked, then dropped and relocked.
  */
-
 static thread_t
-__current_thread(void)
+thread_select_idle(
+       thread_t                thread,
+       processor_t             processor)
 {
-  return (current_thread());
+       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;
+
+       thread_unlock(thread);
+
+       /*
+        *      Switch execution timing to processor idle thread.
+        */
+       processor->last_dispatch = mach_absolute_time();
+       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);
+
+       /*
+        *      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 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);
+
+               processor->quantum_end = processor->last_dispatch + thread->current_quantum;
+               timer_call_enter1(&processor->quantum_timer, thread, processor->quantum_end);
+               processor->timeslice = 1;
+
+               thread->computation_epoch = processor->last_dispatch;
+       }
+
+       thread->state &= ~TH_IDLE;
+
+       sched_run_incr();
+       if (thread->sched_mode & TH_MODE_TIMESHARE)
+               sched_share_incr();
+
+       return (new_thread);
 }
 
-boolean_t
+/*
+ *     Perform a context switch and start executing the new thread.
+ *
+ *     Returns FALSE on failure, and the thread is re-dispatched.
+ *
+ *     Called at splsched.
+ */
+
+#define funnel_release_check(thread, debug)                            \
+MACRO_BEGIN                                                                                            \
+       if ((thread)->funnel_state & TH_FN_OWNED) {                     \
+               (thread)->funnel_state = TH_FN_REFUNNEL;                \
+               KERNEL_DEBUG(0x603242c | DBG_FUNC_NONE,                 \
+                       (thread)->funnel_lock, (debug), 0, 0, 0);       \
+               funnel_unlock((thread)->funnel_lock);                   \
+       }                                                                                                       \
+MACRO_END
+
+#define funnel_refunnel_check(thread, debug)                           \
+MACRO_BEGIN                                                                                                    \
+       if ((thread)->funnel_state & TH_FN_REFUNNEL) {                  \
+               kern_return_t   result = (thread)->wait_result;         \
+                                                                                                                       \
+               (thread)->funnel_state = 0;                                                     \
+               KERNEL_DEBUG(0x6032428 | DBG_FUNC_NONE,                         \
+                       (thread)->funnel_lock, (debug), 0, 0, 0);               \
+               funnel_lock((thread)->funnel_lock);                                     \
+               KERNEL_DEBUG(0x6032430 | DBG_FUNC_NONE,                         \
+                       (thread)->funnel_lock, (debug), 0, 0, 0);               \
+               (thread)->funnel_state = TH_FN_OWNED;                           \
+               (thread)->wait_result = result;                                         \
+       }                                                                                                               \
+MACRO_END
+
+static boolean_t
 thread_invoke(
-       register thread_t       old_thread,
-       register thread_t       new_thread,
-       int                                     reason,
-       thread_continue_t       old_cont)
+       register thread_t       self,
+       register thread_t       thread,
+       ast_t                           reason)
 {
-       thread_continue_t       new_cont;
+       thread_continue_t       continuation = self->continuation;
+       void                            *parameter = self->parameter;
        processor_t                     processor;
 
        if (get_preemption_level() != 0)
                panic("thread_invoke: preemption_level %d\n",
-                                                               get_preemption_level());
+                               get_preemption_level());
+
+       assert(self == current_thread());
 
        /*
         * Mark thread interruptible.
         */
-       thread_lock(new_thread);
-       new_thread->state &= ~TH_UNINT;
-
-       assert(thread_runnable(new_thread));
+       thread_lock(thread);
+       thread->state &= ~TH_UNINT;
 
-       assert(old_thread->continuation == NULL);       
+#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->stack_privilege                            ) {
-               old_thread->stack_privilege = old_thread->kernel_stack;
-       }
+       if ((self->sched_mode & TH_MODE_REALTIME) && !self->reserved_stack)
+               self->reserved_stack = self->kernel_stack;
 
-       if (old_cont != NULL) {
-               if (new_thread->state & TH_STACK_HANDOFF) {
+       if (continuation != NULL) {
+               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->stack_privilege &&
-                                                       !new_thread->stack_privilege)
+                       if (self->kernel_stack == self->reserved_stack && !thread->reserved_stack)
                                goto need_stack;
 
-                       new_thread->state &= ~TH_STACK_HANDOFF;
-                       new_cont = new_thread->continuation;
-                       new_thread->continuation = NULL;
-
                        /*
-                        * Set up ast context of new thread and switch
-                        * to its timer.
-                        */
-                       processor = current_processor();
-                       new_thread->last_processor = processor;
-                       processor->current_pri = new_thread->sched_pri;
-                       ast_context(new_thread->top_act, processor->slot_num);
-                       timer_switch(&new_thread->system_timer);
-                       thread_unlock(new_thread);
-               
-                       current_task()->csw++;
-
-                       old_thread->reason = reason;
-                       old_thread->continuation = old_cont;
-          
-                       _mk_sp_thread_done(old_thread, new_thread, processor);
-
-                       stack_handoff(old_thread, new_thread);
-
-                       _mk_sp_thread_begin(new_thread, processor);
-
-                       wake_lock(old_thread);
-                       thread_lock(old_thread);
-
-                       /* 
-                        * Inline thread_dispatch but
-                        * don't free stack.
+                        * Context switch by performing a stack handoff.
                         */
+                       continuation = thread->continuation;
+                       parameter = thread->parameter;
 
-                       switch (old_thread->state & (TH_RUN|TH_WAIT|TH_UNINT|TH_IDLE)) {
-                       case TH_RUN                             | TH_UNINT:
-                       case TH_RUN:
-                               /*
-                                * Still running, put back
-                                * onto a run queue.
-                                */
-                               old_thread->state |= TH_STACK_HANDOFF;
-                               _mk_sp_thread_dispatch(old_thread);
-
-                               thread_unlock(old_thread);
-                               wake_unlock(old_thread);
-                               break;
+                       processor = current_processor();
+                       processor->active_thread = thread;
+                       processor->current_pri = thread->sched_pri;
+                       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);
 
-                       case TH_RUN | TH_WAIT   | TH_UNINT:
-                       case TH_RUN | TH_WAIT:
-                       {
-                               boolean_t       reap, wake, callblock;
+                       self->reason = reason;
 
-                               /*
-                                * Waiting.
-                                */
-                               old_thread->sleep_stamp = sched_tick;
-                               old_thread->state |= TH_STACK_HANDOFF;
-                               old_thread->state &= ~TH_RUN;
-                               hw_atomic_sub(&old_thread->processor_set->run_count, 1);
-                               callblock = old_thread->active_callout;
-                               wake = old_thread->wake_active;
-                               old_thread->wake_active = FALSE;
-                               reap = (old_thread->state & TH_TERMINATE)? TRUE: FALSE;
-
-                               thread_unlock(old_thread);
-                               wake_unlock(old_thread);
-
-                               if (callblock)
-                                       call_thread_block();
-
-                               if (wake)
-                                       thread_wakeup((event_t)&old_thread->wake_active);
-
-                               if (reap)
-                                       thread_reaper_enqueue(old_thread);
-                               break;
-                       }
+                       processor->last_dispatch = mach_absolute_time();
+                       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, (int)thread, self->sched_pri, thread->sched_pri, 0);
 
-                       case TH_RUN                             | TH_IDLE:
-                               /*
-                                * The idle threads don't go
-                                * onto a run queue.
-                                */
-                               old_thread->state |= TH_STACK_HANDOFF;
-                               thread_unlock(old_thread);
-                               wake_unlock(old_thread);
-                               break;
+TLOG(1, "thread_invoke: calling machine_stack_handoff\n");
+                       machine_stack_handoff(self, thread);
 
-                       default:
-                               panic("thread_invoke: state 0x%x\n", old_thread->state);
-                       }
+                       thread_dispatch(self, thread);
 
-                       counter_always(c_thread_invoke_hits++);
+                       thread->continuation = thread->parameter = NULL;
 
-                       if (new_thread->funnel_state & TH_FN_REFUNNEL) {
-                               kern_return_t           wait_result = new_thread->wait_result;
+                       counter(c_thread_invoke_hits++);
 
-                               new_thread->funnel_state = 0;
-                               KERNEL_DEBUG(0x6032428 | DBG_FUNC_NONE,
-                                                                       new_thread->funnel_lock, 2, 0, 0, 0);
-                               funnel_lock(new_thread->funnel_lock);
-                               KERNEL_DEBUG(0x6032430 | DBG_FUNC_NONE,
-                                                                       new_thread->funnel_lock, 2, 0, 0, 0);
-                               new_thread->funnel_state = TH_FN_OWNED;
-                               new_thread->wait_result = wait_result;
-                       }
+                       funnel_refunnel_check(thread, 2);
                        (void) spllo();
 
-                       assert(new_cont);
-                       call_continuation(new_cont);
+                       assert(continuation);
+                       call_continuation(continuation, parameter, thread->wait_result);
                        /*NOTREACHED*/
-                       return (TRUE);
-               }
-               else
-               if (new_thread->state & TH_STACK_ALLOC) {
-                       /*
-                        * Waiting for a stack
-                        */
-                       counter_always(c_thread_invoke_misses++);
-                       thread_unlock(new_thread);
-                       return (FALSE);
                }
-               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);
 
-                       if (new_thread->funnel_state & TH_FN_REFUNNEL) {
-                               kern_return_t   wait_result = new_thread->wait_result;
-
-                               new_thread->funnel_state = 0;
-                               KERNEL_DEBUG(0x6032428 | DBG_FUNC_NONE,
-                                                                       new_thread->funnel_lock, 3, 0, 0, 0);
-                               funnel_lock(new_thread->funnel_lock);
-                               KERNEL_DEBUG(0x6032430 | DBG_FUNC_NONE,
-                                                                       new_thread->funnel_lock, 3, 0, 0, 0);
-                               new_thread->funnel_state = TH_FN_OWNED;
-                               new_thread->wait_result = wait_result;
-                       }
+                       self->continuation = self->parameter = NULL;
+
+                       funnel_refunnel_check(self, 3);
                        (void) spllo();
-                       call_continuation(old_cont);
+
+                       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->state & TH_STACK_HANDOFF) {
+               if (!thread->kernel_stack) {
 need_stack:
-                       if (!stack_alloc_try(new_thread, thread_continue)) {
-                               counter_always(c_thread_invoke_misses++);
-                               thread_swapin(new_thread);
+                       if (!stack_alloc_try(thread)) {
+                               counter(c_thread_invoke_misses++);
+                               thread_unlock(thread);
+                               thread_stack_enqueue(thread);
                                return (FALSE);
                        }
-        
-                       new_thread->state &= ~TH_STACK_HANDOFF;
-               }
-               else
-               if (new_thread->state & TH_STACK_ALLOC) {
-                       /*
-                        * Waiting for a stack
-                        */
-                       counter_always(c_thread_invoke_misses++);
-                       thread_unlock(new_thread);
-                       return (FALSE);
                }
-               else
-               if (old_thread == new_thread) {
+               else if (thread == self) {
+                       ast_context(self);
                        counter(++c_thread_invoke_same);
-                       thread_unlock(new_thread);
+                       thread_unlock(self);
                        return (TRUE);
                }
        }
 
        /*
-        * Set up ast context of new thread and switch to its timer.
+        * Context switch by full context save.
         */
        processor = current_processor();
-       new_thread->last_processor = processor;
-       processor->current_pri = new_thread->sched_pri;
-       ast_context(new_thread->top_act, processor->slot_num);
-       timer_switch(&new_thread->system_timer);
-       assert(thread_runnable(new_thread));
-       thread_unlock(new_thread);
+       processor->active_thread = thread;
+       processor->current_pri = thread->sched_pri;
+       if (thread->last_processor != processor && thread->last_processor != NULL) {
+               if (thread->last_processor->processor_set != processor->processor_set)
+                       thread->ps_switch++;
+               thread->p_switch++;
+       }
+       thread->last_processor = processor;
+       thread->c_switch++;
+       ast_context(thread);
+       thread_unlock(thread);
+
+       counter(c_thread_invoke_csw++);
 
-       counter_always(c_thread_invoke_csw++);
-       current_task()->csw++;
+       assert(self->runq == PROCESSOR_NULL);
+       self->reason = reason;
 
-       assert(old_thread->runq == RUN_QUEUE_NULL);
-       old_thread->reason = reason;
-       old_thread->continuation = old_cont;
+       processor->last_dispatch = mach_absolute_time();
+       thread_timer_event(processor->last_dispatch, &thread->system_timer);
+       PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
 
-       _mk_sp_thread_done(old_thread, new_thread, processor);
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED,MACH_SCHED) | DBG_FUNC_NONE,
+                                                       (int)self->reason, (int)thread, self->sched_pri, thread->sched_pri, 0);
 
        /*
-        *      switch_context is machine-dependent.  It does the
-        *      machine-dependent components of a context-switch, like
-        *      changing address spaces.  It updates active_threads.
+        * This is where we actually switch register context,
+        * and address space if required.  We will next run
+        * as a result of a subsequent context switch.
         */
-       old_thread = switch_context(old_thread, old_cont, new_thread);
-       
-       /* Now on new thread's stack.  Set a local variable to refer to it. */
-       new_thread = __current_thread();
-       assert(old_thread != new_thread);
-
-       assert(thread_runnable(new_thread));
-       _mk_sp_thread_begin(new_thread, new_thread->last_processor);
+       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're back.  Now old_thread is the thread that resumed
-        *      us, and we have to dispatch it.
+        * We have been resumed and are set to run.
         */
-       thread_dispatch(old_thread);
-
-       if (old_cont) {
-               if (new_thread->funnel_state & TH_FN_REFUNNEL) {
-                       kern_return_t           wait_result = new_thread->wait_result;
-
-                       new_thread->funnel_state = 0;
-                       KERNEL_DEBUG(0x6032428 | DBG_FUNC_NONE,
-                                                               new_thread->funnel_lock, 3, 0, 0, 0);
-                       funnel_lock(new_thread->funnel_lock);
-                       KERNEL_DEBUG(0x6032430 | DBG_FUNC_NONE,
-                                                               new_thread->funnel_lock, 3, 0, 0, 0);
-                       new_thread->funnel_state = TH_FN_OWNED;
-                       new_thread->wait_result = wait_result;
-               }
+       thread_dispatch(thread, self);
+
+       if (continuation) {
+               self->continuation = self->parameter = NULL;
+
+               funnel_refunnel_check(self, 3);
                (void) spllo();
-               call_continuation(old_cont);
+
+               call_continuation(continuation, parameter, self->wait_result);
                /*NOTREACHED*/
        }
 
@@ -1436,215 +1641,203 @@ need_stack:
 }
 
 /*
- *     thread_continue:
+ *     thread_dispatch:
+ *
+ *     Handle threads at context switch.  Re-dispatch other thread
+ *     if still running, otherwise update run state and perform
+ *     special actions.  Update quantum for other thread and begin
+ *     the quantum for ourselves.
  *
- *     Called when a thread gets a new stack, at splsched();
+ *     Called at splsched.
  */
 void
-thread_continue(
-       register thread_t       old_thread)
+thread_dispatch(
+       thread_t                thread,
+       thread_t                self)
 {
-       register thread_t                       self = current_thread();
-       register thread_continue_t      continuation;
-       
-       continuation = self->continuation;
-       self->continuation = NULL;
+       processor_t             processor = self->last_processor;
 
-       _mk_sp_thread_begin(self, self->last_processor);
-       
-       /*
-        *      We must dispatch the old thread and then
-        *      call the current thread's continuation.
-        *      There might not be an old thread, if we are
-        *      the first thread to run on this processor.
-        */
-       if (old_thread != THREAD_NULL)
-               thread_dispatch(old_thread);
-
-       if (self->funnel_state & TH_FN_REFUNNEL) {
-               kern_return_t           wait_result = self->wait_result;
-
-               self->funnel_state = 0;
-               KERNEL_DEBUG(0x6032428 | DBG_FUNC_NONE, self->funnel_lock, 4, 0, 0, 0);
-               funnel_lock(self->funnel_lock);
-               KERNEL_DEBUG(0x6032430 | DBG_FUNC_NONE, self->funnel_lock, 4, 0, 0, 0);
-               self->funnel_state = TH_FN_OWNED;
-               self->wait_result = wait_result;
-       }
-       (void)spllo();
-       assert(continuation);
-       call_continuation(continuation);
-       /*NOTREACHED*/
-}
+       if (thread != THREAD_NULL) {
+               /*
+                *      If blocked at a continuation, discard
+                *      the stack.
+                */
+               if (thread->continuation != NULL && thread->kernel_stack != 0)
+                       stack_free(thread);
 
-#if    MACH_LDEBUG || MACH_KDB
+               if (!(thread->state & TH_IDLE)) {
+                       wake_lock(thread);
+                       thread_lock(thread);
 
-#define THREAD_LOG_SIZE                300
+                       /*
+                        *      Compute remainder of current quantum.
+                        */
+                       if (    first_timeslice(processor)                                                      &&
+                                       processor->quantum_end > processor->last_dispatch               )
+                               thread->current_quantum = (processor->quantum_end - processor->last_dispatch);
+                       else
+                               thread->current_quantum = 0;
 
-struct t64 {
-       unsigned long h;
-       unsigned long l;
-};
+                       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 {
+                               /*
+                                *      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;
+                               }
+                       }
 
-struct {
-       struct t64      stamp;
-       thread_t        thread;
-       long            info1;
-       long            info2;
-       long            info3;
-       char            * action;
-} thread_log[THREAD_LOG_SIZE];
+                       /*
+                        *      If we are doing a direct handoff then
+                        *      take the remainder of the quantum.
+                        */
+                       if ((thread->reason & (AST_HANDOFF|AST_QUANTUM)) == AST_HANDOFF) {
+                               self->current_quantum = thread->current_quantum;
+                               thread->reason |= AST_QUANTUM;
+                               thread->current_quantum = 0;
+                       }
 
-int            thread_log_index;
+                       thread->last_switch = processor->last_dispatch;
 
-void           check_thread_time(long n);
+                       thread->computation_metered += (thread->last_switch - thread->computation_epoch);
 
+                       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);
 
-int    check_thread_time_crash;
+                               thread->reason = AST_NONE;
 
-#if 0
-void
-check_thread_time(long us)
-{
-       struct t64      temp;
+                               thread_unlock(thread);
+                               wake_unlock(thread);
+                       }
+                       else {
+                               /*
+                                *      Waiting.
+                                */
+                               thread->state &= ~TH_RUN;
 
-       if (!check_thread_time_crash)
-               return;
+                               if (thread->sched_mode & TH_MODE_TIMESHARE)
+                                       sched_share_decr();
+                               sched_run_decr();
 
-       temp = thread_log[0].stamp;
-       cyctm05_diff (&thread_log[1].stamp, &thread_log[0].stamp, &temp);
+                               if (thread->wake_active) {
+                                       thread->wake_active = FALSE;
+                                       thread_unlock(thread);
 
-       if (temp.l >= us && thread_log[1].info != 0x49) /* HACK!!! */
-               panic ("check_thread_time");
-}
-#endif
+                                       thread_wakeup(&thread->wake_active);
+                               }
+                               else
+                                       thread_unlock(thread);
 
-void
-log_thread_action(char * action, long info1, long info2, long info3)
-{
-       int     i;
-       spl_t   x;
-       static  unsigned int tstamp;
+                               wake_unlock(thread);
 
-       x = splhigh();
+                               (*thread->sched_call)(SCHED_CALL_BLOCK, thread);
 
-       for (i = THREAD_LOG_SIZE-1; i > 0; i--) {
-               thread_log[i] = thread_log[i-1];
+                               if (thread->state & TH_TERMINATE)
+                                       thread_terminate_enqueue(thread);
+                       }
+               }
        }
 
-       thread_log[0].stamp.h = 0;
-       thread_log[0].stamp.l = tstamp++;
-       thread_log[0].thread = current_thread();
-       thread_log[0].info1 = info1;
-       thread_log[0].info2 = info2;
-       thread_log[0].info3 = info3;
-       thread_log[0].action = action;
-/*     strcpy (&thread_log[0].action[0], action);*/
-
-       splx(x);
-}
-#endif /* MACH_LDEBUG || MACH_KDB */
+       if (!(self->state & TH_IDLE)) {
+               /*
+                *      Get a new quantum if none remaining.
+                */
+               if (self->current_quantum == 0)
+                       thread_quantum_init(self);
 
-#if    MACH_KDB
-#include <ddb/db_output.h>
-void           db_show_thread_log(void);
+               /*
+                *      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);
 
-void
-db_show_thread_log(void)
-{
-       int     i;
+               processor->timeslice = 1;
 
-       db_printf ("%s %s %s %s %s %s\n", " Thread ", "  Info1 ", "  Info2 ",
-                       "  Info3 ", "    Timestamp    ", "Action");
+               self->last_switch = processor->last_dispatch;
 
-       for (i = 0; i < THREAD_LOG_SIZE; i++) {
-               db_printf ("%08x %08x %08x %08x %08x/%08x %s\n",
-                       thread_log[i].thread,
-                       thread_log[i].info1,
-                       thread_log[i].info2,
-                       thread_log[i].info3,
-                       thread_log[i].stamp.h,
-                       thread_log[i].stamp.l,
-                       thread_log[i].action);
+               self->computation_epoch = self->last_switch;
+       }
+       else {
+               timer_call_cancel(&processor->quantum_timer);
+               processor->timeslice = 0;
        }
 }
-#endif /* MACH_KDB */
 
 /*
  *     thread_block_reason:
  *
- *     Block the current thread if a wait has been asserted,
- *     otherwise unconditionally yield the remainder of the
- *     current quantum unless reason contains AST_BLOCK.
- *     
- *     If a continuation is specified, then thread_block will
+ *     Forces a reschedule, blocking the caller if a wait
+ *     has been asserted.
+ *
+ *     If a continuation is specified, then thread_invoke will
  *     attempt to discard the thread's kernel stack.  When the
  *     thread resumes, it will execute the continuation function
  *     on a new kernel stack.
  */
 counter(mach_counter_t  c_thread_block_calls = 0;)
  
-int
+wait_result_t
 thread_block_reason(
        thread_continue_t       continuation,
+       void                            *parameter,
        ast_t                           reason)
 {
-       register thread_t               thread = current_thread();
-       register processor_t    myprocessor;
+       register thread_t               self = current_thread();
+       register processor_t    processor;
        register thread_t               new_thread;
        spl_t                                   s;
 
        counter(++c_thread_block_calls);
 
-       check_simple_locks();
-
-       machine_clock_assist();
-
        s = splsched();
 
-       if ((thread->funnel_state & TH_FN_OWNED) && !(reason & AST_PREEMPT)) {
-               thread->funnel_state = TH_FN_REFUNNEL;
-               KERNEL_DEBUG(
-                       0x603242c | DBG_FUNC_NONE, thread->funnel_lock, 2, 0, 0, 0);
-               funnel_unlock(thread->funnel_lock);
-       }
+       if (!(reason & AST_PREEMPT))
+               funnel_release_check(self, 2);
 
-       myprocessor = current_processor();
+       processor = current_processor();
 
        /* If we're explicitly yielding, force a subsequent quantum */
        if (reason & AST_YIELD)
-               myprocessor->slice_quanta = 0;
+               processor->timeslice = 0;
 
        /* We're handling all scheduling AST's */
        ast_off(AST_SCHEDULING);
 
-       thread_lock(thread);
-       new_thread = thread_select(myprocessor);
-       assert(new_thread && thread_runnable(new_thread));
-       thread_unlock(thread);
-       while (!thread_invoke(thread, new_thread, reason, continuation)) {
-               thread_lock(thread);
-               new_thread = thread_select(myprocessor);
-               assert(new_thread && thread_runnable(new_thread));
-               thread_unlock(thread);
-       }
+       self->continuation = continuation;
+       self->parameter = parameter;
 
-       if (thread->funnel_state & TH_FN_REFUNNEL) {
-               kern_return_t   wait_result = thread->wait_result;
-
-               thread->funnel_state = 0;
-               KERNEL_DEBUG(
-                       0x6032428 | DBG_FUNC_NONE, thread->funnel_lock, 5, 0, 0, 0);
-               funnel_lock(thread->funnel_lock);
-               KERNEL_DEBUG(
-                       0x6032430 | DBG_FUNC_NONE, thread->funnel_lock, 5, 0, 0, 0);
-               thread->funnel_state = TH_FN_OWNED;
-               thread->wait_result = wait_result;
-       }
+       do {
+               thread_lock(self);
+               new_thread = thread_select(self, processor);
+               thread_unlock(self);
+       } while (!thread_invoke(self, new_thread, reason));
 
+       funnel_refunnel_check(self, 5);
        splx(s);
 
-       return (thread->wait_result);
+       return (self->wait_result);
 }
 
 /*
@@ -1652,569 +1845,877 @@ thread_block_reason(
  *
  *     Block the current thread if a wait has been asserted.
  */
-int
+wait_result_t
 thread_block(
        thread_continue_t       continuation)
 {
-       return thread_block_reason(continuation, AST_NONE);
+       return thread_block_reason(continuation, NULL, AST_NONE);
+}
+
+wait_result_t
+thread_block_parameter(
+       thread_continue_t       continuation,
+       void                            *parameter)
+{
+       return thread_block_reason(continuation, parameter, AST_NONE);
 }
 
 /*
  *     thread_run:
  *
- *     Switch directly from the current (old) thread to the
- *     specified thread, handing off our quantum if possible.
+ *     Switch directly from the current thread to the
+ *     new thread, handing off our quantum if appropriate.
  *
  *     New thread must be runnable, and not on a run queue.
  *
- *  Assumption:
- *     at splsched.
+ *     Called at splsched.
  */
 int
 thread_run(
-       thread_t                        old_thread,
+       thread_t                        self,
        thread_continue_t       continuation,
+       void                            *parameter,
        thread_t                        new_thread)
 {
        ast_t           handoff = AST_HANDOFF;
 
-       assert(old_thread == current_thread());
-
-       machine_clock_assist();
+       funnel_release_check(self, 3);
 
-       if (old_thread->funnel_state & TH_FN_OWNED) {
-               old_thread->funnel_state = TH_FN_REFUNNEL;
-               KERNEL_DEBUG(
-                       0x603242c | DBG_FUNC_NONE, old_thread->funnel_lock, 3, 0, 0, 0);
-               funnel_unlock(old_thread->funnel_lock);
-       }
+       self->continuation = continuation;
+       self->parameter = parameter;
 
-       while (!thread_invoke(old_thread, new_thread, handoff, continuation)) {
-               register processor_t            myprocessor = current_processor();
+       while (!thread_invoke(self, new_thread, handoff)) {
+               processor_t             processor = current_processor();
 
-               thread_lock(old_thread);
-               new_thread = thread_select(myprocessor);
-               thread_unlock(old_thread);
+               thread_lock(self);
+               new_thread = thread_select(self, processor);
+               thread_unlock(self);
                handoff = AST_NONE;
        }
 
-       /* if we fell thru */
-       if (old_thread->funnel_state & TH_FN_REFUNNEL) {
-               kern_return_t   wait_result = old_thread->wait_result;
-
-               old_thread->funnel_state = 0;
-               KERNEL_DEBUG(
-                       0x6032428 | DBG_FUNC_NONE, old_thread->funnel_lock, 6, 0, 0, 0);
-               funnel_lock(old_thread->funnel_lock);
-               KERNEL_DEBUG(
-                       0x6032430 | DBG_FUNC_NONE, old_thread->funnel_lock, 6, 0, 0, 0);
-               old_thread->funnel_state = TH_FN_OWNED;
-               old_thread->wait_result = wait_result;
-       }
+       funnel_refunnel_check(self, 6);
 
-       return (old_thread->wait_result);
+       return (self->wait_result);
 }
 
 /*
- *     Dispatches a running thread that is not on a runq.
- *     Called at splsched.
+ *     thread_continue:
+ *
+ *     Called at splsched when a thread first receives
+ *     a new stack after a continuation.
  */
 void
-thread_dispatch(
+thread_continue(
        register thread_t       thread)
 {
-       wake_lock(thread);
-       thread_lock(thread);
+       register thread_t                       self = current_thread();
+       register thread_continue_t      continuation;
+       register void                           *parameter;
+       
+       continuation = self->continuation;
+       parameter = self->parameter;
 
-       /*
-        *      If we are discarding the thread's stack, we must do it
-        *      before the thread has a chance to run.
-        */
-#ifndef i386
-    if (thread->continuation != NULL) {
-               assert((thread->state & TH_STACK_STATE) == 0);
-               thread->state |= TH_STACK_HANDOFF;
-               stack_free(thread);
+       thread_dispatch(thread, self);
+
+       self->continuation = self->parameter = NULL;
+
+       funnel_refunnel_check(self, 4);
+
+       if (thread != THREAD_NULL)
+               (void)spllo();
+
+ TLOG(1, "thread_continue: calling call_continuation \n");
+       call_continuation(continuation, parameter, self->wait_result);
+       /*NOTREACHED*/
+}
+
+/*
+ *     run_queue_init:
+ *
+ *     Initialize a run queue before first use.
+ */
+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]);
+}
+
+/*
+ *     run_queue_dequeue:
+ *
+ *     Perform a dequeue operation on a run queue,
+ *     and return the resulting thread.
+ *
+ *     The run queue must be locked (see run_queue_remove()
+ *     for more info), and not empty.
+ */
+static 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)queue->next;
+               ((queue_entry_t)thread)->next->prev = queue;
+               queue->next = ((queue_entry_t)thread)->next;
+       }
+       else {
+               thread = (thread_t)queue->prev;
+               ((queue_entry_t)thread)->prev->next = queue;
+               queue->prev = ((queue_entry_t)thread)->prev;
        }
-#endif
 
-       switch (thread->state & (TH_RUN|TH_WAIT|TH_UNINT|TH_IDLE)) {
+       thread->runq = PROCESSOR_NULL;
+       rq->count--;
+       if (testbit(rq->highq, sched_preempt_pri)) {
+               rq->urgency--; assert(rq->urgency >= 0);
+       }
+       if (queue_empty(queue)) {
+               if (rq->highq != IDLEPRI)
+                       clrbit(MAXPRI - rq->highq, rq->bitmap);
+               rq->highq = MAXPRI - ffsbit(rq->bitmap);
+       }
 
-       case TH_RUN                              | TH_UNINT:
-       case TH_RUN:
-               /*
-                *      No reason to stop.  Put back on a run queue.
-                */
-               _mk_sp_thread_dispatch(thread);
-               break;
+       return (thread);
+}
 
-       case TH_RUN | TH_WAIT   | TH_UNINT:
-       case TH_RUN | TH_WAIT:
-       {
-               boolean_t       reap, wake, callblock;
-       
-               /*
-                *      Waiting
-                */
-               thread->sleep_stamp = sched_tick;
-               thread->state &= ~TH_RUN;
-               hw_atomic_sub(&thread->processor_set->run_count, 1);
-               callblock = thread->active_callout;
-               wake = thread->wake_active;
-               thread->wake_active = FALSE;
-               reap = (thread->state & TH_TERMINATE)? TRUE: FALSE;
+/*
+ *     realtime_queue_insert:
+ *
+ *     Enqueue a thread for realtime execution.
+ */
+static boolean_t
+realtime_queue_insert(
+       thread_t                        thread)
+{
+       run_queue_t                     rq = &rt_runq;
+       queue_t                         queue = rq->queues + thread->sched_pri;
+       uint64_t                        deadline = thread->realtime.deadline;
+       boolean_t                       preempt = FALSE;
 
-               thread_unlock(thread);
-               wake_unlock(thread);
+       simple_lock(&rt_lock);
+
+       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;
+               preempt = TRUE;
+       }
+       else {
+               register thread_t       entry = (thread_t)queue_first(queue);
 
-               if (callblock)
-                       call_thread_block();
+               while (TRUE) {
+                       if (    queue_end(queue, (queue_entry_t)entry)  ||
+                                               deadline < entry->realtime.deadline             ) {
+                               entry = (thread_t)queue_prev((queue_entry_t)entry);
+                               break;
+                       }
 
-               if (wake)
-                   thread_wakeup((event_t)&thread->wake_active);
+                       entry = (thread_t)queue_next((queue_entry_t)entry);
+               }
 
-               if (reap)
-                       thread_reaper_enqueue(thread);
+               if ((queue_entry_t)entry == queue)
+                       preempt = TRUE;
 
-               return;
+               insque((queue_entry_t)thread, (queue_entry_t)entry);
        }
 
-       case TH_RUN                                             | TH_IDLE:
-               /*
-                * The idle threads don't go
-                * onto a run queue.
-                */
-               break;
+       thread->runq = RT_RUNQ;
+       rq->count++; rq->urgency++;
+
+       simple_unlock(&rt_lock);
+
+       return (preempt);
+}
 
-       default:
-               panic("thread_dispatch: bad thread state 0x%x\n", thread->state);
+/*
+ *     realtime_setrun:
+ *
+ *     Dispatch a thread for realtime execution.
+ *
+ *     Thread must be locked.  Associated pset must
+ *     be locked, and is returned unlocked.
+ */
+static void
+realtime_setrun(
+       processor_t                     processor,
+       thread_t                        thread)
+{
+       processor_set_t         pset = processor->processor_set;
+
+       /*
+        *      Dispatch directly onto idle processor.
+        */
+       if (processor->state == PROCESSOR_IDLE) {
+               remqueue(&pset->idle_queue, (queue_entry_t)processor);
+               enqueue_tail(&pset->active_queue, (queue_entry_t)processor);
+
+               processor->next_thread = thread;
+               processor->deadline = thread->realtime.deadline;
+               processor->state = PROCESSOR_DISPATCHING;
+               pset_unlock(pset);
+
+               if (processor != current_processor())
+                       machine_signal_idle(processor);
+               return;
        }
 
-       thread_unlock(thread);
-       wake_unlock(thread);
+       if (realtime_queue_insert(thread)) {
+               if (processor == current_processor())
+                       ast_on(AST_PREEMPT | AST_URGENT);
+               else
+                       cause_ast_check(processor);
+       }
+
+       pset_unlock(pset);
 }
 
 /*
- * Enqueue thread on run queue.  Thread must be locked,
- * and not already be on a run queue.  Returns TRUE iff
- * the particular queue level was empty beforehand.
+ *     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 run_queue_remove()
+ *     for more info).
  */
-boolean_t
-run_queue_enqueue(
-       register run_queue_t    rq,
-       register thread_t               thread,
-       boolean_t                               tail)
-{
-       register int                    whichq = thread->sched_pri;
-       register queue_t                queue = &rq->queues[whichq];
-       boolean_t                               result = FALSE;
+static boolean_t
+processor_enqueue(
+       processor_t             processor,
+       thread_t                thread,
+       integer_t               options)
+{
+       run_queue_t             rq = &processor->runq;
+       queue_t                 queue = rq->queues + thread->sched_pri;
+       boolean_t               result = FALSE;
        
-       assert(whichq >= MINPRI && whichq <= MAXPRI);
-
-       simple_lock(&rq->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;
-               result = TRUE;
+               setbit(MAXPRI - thread->sched_pri, rq->bitmap);
+               if (thread->sched_pri > rq->highq) {
+                       rq->highq = thread->sched_pri;
+                       result = TRUE;
+               }
        }
        else
-       if (tail)
+       if (options & SCHED_TAILQ)
                enqueue_tail(queue, (queue_entry_t)thread);
        else
                enqueue_head(queue, (queue_entry_t)thread);
 
-       thread->runq = rq;
-       if (thread->sched_mode & TH_MODE_PREEMPT)
+       thread->runq = processor;
+       if (testbit(thread->sched_pri, sched_preempt_pri))
                rq->urgency++;
        rq->count++;
-#if    DEBUG
-       thread_check(thread, rq);
-#endif /* DEBUG */
-       simple_unlock(&rq->lock);
 
        return (result);
 }
 
-struct {
-       uint32_t        pset_idle_last,
-                               pset_idle_any,
-                               pset_self,
-                               pset_last,
-                               pset_other,
-                               bound_idle,
-                               bound_self,
-                               bound_other;
-} dispatch_counts;
+/*
+ *     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;
+
+       /*
+        *      Dispatch directly onto idle processor.
+        */
+       if (processor->state == PROCESSOR_IDLE) {
+               remqueue(&pset->idle_queue, (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 (testbit(thread->sched_pri, sched_preempt_pri))
+               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 (!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_RUNNING          ||
+                                processor->state == PROCESSOR_SHUTDOWN)                &&
+                               thread->sched_pri >= processor->current_pri             ) {
+                       cause_ast_check(processor);
+               }
+       }
+       else
+       if (    processor->state == PROCESSOR_SHUTDOWN          &&
+                       thread->sched_pri >= processor->current_pri     ) {
+               cause_ast_check(processor);
+       }
+
+       pset_unlock(pset);
+}
+
+#define next_pset(p)   (((p)->pset_list != PROCESSOR_SET_NULL)? (p)->pset_list: (p)->node->psets)
 
 /*
- *     thread_setrun:
+ *     choose_next_pset:
  *
- *     Dispatch thread for execution, directly onto an idle
- *     processor if possible.  Else put on appropriate run
- *     queue. (local if bound, else processor set)
+ *     Return the next sibling pset containing
+ *     available processors.
  *
- *     Thread must be locked.
+ *     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->processor_count < 1 && nset != pset);
+
+       return (nset);
+}
+
+/*
+ *     choose_processor:
+ *
+ *     Choose a processor for the thread, beginning at
+ *     the pset.
  *
- *     The tail parameter indicates the proper placement of
- *     the thread on a run queue.
+ *     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.
  */
-void
-thread_setrun(
-       register thread_t                       new_thread,
-       boolean_t                                       tail)
+static processor_t
+choose_processor(
+       processor_set_t         pset,
+       thread_t                        thread)
 {
-       register processor_t            processor;
-       register processor_set_t        pset;
-       register thread_t                       thread;
-       boolean_t                                       try_preempt = FALSE;
-       ast_t                                           preempt = AST_BLOCK;
+       processor_set_t         nset, cset = pset;
+       processor_t                     processor = thread->last_processor;
 
-       assert(thread_runnable(new_thread));
-       
        /*
-        *      Update priority if needed.
+        *      Prefer the last processor, when appropriate.
         */
-       if (new_thread->sched_stamp != sched_tick)
-               update_priority(new_thread);
+       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_DEFAULT && processor->current_pri < thread->sched_pri))
+                       return (processor);
+       }
 
        /*
-        *      Check for urgent preemption.
+        *      Iterate through the processor sets to locate
+        *      an appropriate processor.
         */
-       if (new_thread->sched_mode & TH_MODE_PREEMPT)
-               preempt |= AST_URGENT;
+       do {
+               /*
+                *      Choose an idle processor.
+                */
+               if (!queue_empty(&cset->idle_queue))
+                       return ((processor_t)queue_first(&cset->idle_queue));
 
-       assert(new_thread->runq == RUN_QUEUE_NULL);
+               if (thread->sched_pri >= BASEPRI_RTQUEUES) {
+                       /*
+                        *      For an RT thread, iterate through active processors, first fit.
+                        */
+                       processor = (processor_t)queue_first(&cset->active_queue);
+                       while (!queue_end(&cset->active_queue, (queue_entry_t)processor)) {
+                               if (thread->sched_pri > processor->current_pri ||
+                                               thread->realtime.deadline < processor->deadline)
+                                       return (processor);
 
-       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              ) {
-                       simple_lock(&processor->lock);
-                       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;
-                               processor->state = PROCESSOR_DISPATCHING;
-                               simple_unlock(&pset->sched_lock);
-                               simple_unlock(&processor->lock);
-                               if (processor != current_processor())
-                                       machine_signal_idle(processor);
-                               dispatch_counts.pset_idle_last++;
-                               return;
+                               processor = (processor_t)queue_next((queue_entry_t)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
+                       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 && cset->low_count->runq.count < processor->runq.count))) {
+                               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);
                        }
-                       simple_unlock(&processor->lock);
                }
-               else
-               simple_lock(&pset->sched_lock);
 
                /*
-                *      Next pick any idle processor
-                *      in the processor set.
+                *      Move onto the next processor set.
                 */
-               if (pset->idle_count > 0) {
-                       processor = (processor_t)dequeue_head(&pset->idle_queue);
-                       pset->idle_count--;
-                       processor->next_thread = new_thread;
-                       processor->state = PROCESSOR_DISPATCHING;
-                       simple_unlock(&pset->sched_lock);
-                       if (processor != current_processor())   
-                               machine_signal_idle(processor);
-                       dispatch_counts.pset_idle_any++;
-                       return;
+               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 {
                /*
-                * Place thread on run queue.
+                *      If we haven't been able to choose a processor,
+                *      pick the boot processor and return it.
                 */
-               if (run_queue_enqueue(&pset->runq, new_thread, tail))
-                       try_preempt = TRUE;
+               if (processor == PROCESSOR_NULL) {
+                       processor = master_processor;
+
+                       /*
+                        *      Check that the correct processor set is
+                        *      returned locked.
+                        */
+                       if (cset != processor->processor_set) {
+                               pset_unlock(cset);
+
+                               cset = processor->processor_set;
+                               pset_lock(cset);
+                       }
+
+                       return (processor);
+               }
 
                /*
-                *      Update the timesharing quanta.
+                *      Check that the processor set for the chosen
+                *      processor is locked.
                 */
-               pset_quanta_update(pset);
+               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
        
-           /*
-            *  Preempt check.
-            */
-           processor = current_processor();
-               thread = processor->cpu_data->active_thread;
-           if (try_preempt) {
+       /*
+        *      Update priority if needed.
+        */
+       if (thread->sched_stamp != sched_tick)
+               update_priority(thread);
+
+       assert(thread->runq == PROCESSOR_NULL);
+
+       if (thread->bound_processor == PROCESSOR_NULL) {
+               /*
+                *      Unbound case.
+                */
+               if (thread->affinity_set != AFFINITY_SET_NULL) {
                        /*
-                        * First try the current processor
-                        * if it is a member of the correct
-                        * processor set.
+                        * Use affinity set policy hint.
                         */
-                       if (    pset == processor->processor_set        &&
-                                       csw_needed(thread, processor)           ) {
-                               simple_unlock(&pset->sched_lock);
-
-                               ast_on(preempt);
-                               dispatch_counts.pset_self++;
-                               return;
-                       }
+                       pset = thread->affinity_set->aset_pset;
+                       pset_lock(pset);
 
+                       processor = choose_processor(pset, thread);
+               }
+               else
+               if (thread->last_processor != PROCESSOR_NULL) {
                        /*
-                        * If that failed and we have other
-                        * processors available keep trying.
+                        *      Simple (last processor) affinity case.
                         */
-                       if (    pset->processor_count > 1                       ||
-                                       pset != processor->processor_set        ) {
-                               queue_t                 active = &pset->active_queue;
-                               processor_t             myprocessor, lastprocessor;
-                               queue_entry_t   next;
+                       processor = thread->last_processor;
+                       pset = processor->processor_set;
+                       pset_lock(pset);
 
+                       /*
+                        *      Choose a different processor in certain cases.
+                        */
+                       if (thread->sched_pri >= BASEPRI_RTQUEUES) {
                                /*
-                                * Next try the last processor
-                                * dispatched on.
+                                *      If the processor is executing an RT thread with
+                                *      an earlier deadline, choose another.
                                 */
-                               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  ) {
-                                       cause_ast_check(processor);
-                                       simple_unlock(&pset->sched_lock);
-                                       dispatch_counts.pset_last++;
-                                       return;
-                               }
+                               if (thread->sched_pri <= processor->current_pri ||
+                                               thread->realtime.deadline >= processor->deadline)
+                                       processor = choose_processor(pset, thread);
+                       }
+                       else
+                               processor = choose_processor(pset, thread);
+               }
+               else {
+                       /*
+                        *      No Affinity case:
+                        *
+                        *      Utilitize a per task hint to spread threads
+                        *      among the available processor sets.
+                        */
+                       task_t          task = thread->task;
 
-                               /*
-                                * Lastly, pick any other
-                                * available processor.
-                                */
-                               lastprocessor = processor;
-                               processor = (processor_t)queue_first(active);
-                               while (!queue_end(active, (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(active, next)) {
-                                                       remqueue(active, (queue_entry_t)processor);
-                                                       enqueue_tail(active, (queue_entry_t)processor);
-                                               }
-                                               cause_ast_check(processor);
-                                               simple_unlock(&pset->sched_lock);
-                                               dispatch_counts.pset_other++;
-                                               return;
-                                       }
+                       pset = task->pset_hint;
+                       if (pset == PROCESSOR_SET_NULL)
+                               pset = current_processor()->processor_set;
 
-                                       processor = (processor_t)next;
-                               }
-                       }
-           }
+                       pset = choose_next_pset(pset);
+                       pset_lock(pset);
 
-               simple_unlock(&pset->sched_lock);
+                       processor = choose_processor(pset, 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.
-            */
-               if (processor->state == PROCESSOR_IDLE) {
-                       simple_lock(&processor->lock);
-                       pset = processor->processor_set;
-                       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->state = PROCESSOR_DISPATCHING;
-                               simple_unlock(&pset->sched_lock);
-                               simple_unlock(&processor->lock);
-                               if (processor != current_processor())   
-                                       machine_signal_idle(processor);
-                               dispatch_counts.bound_idle++;
-                               return;
-                       }
-                       simple_unlock(&pset->sched_lock);
-                       simple_unlock(&processor->lock);
-               }
-         
-               if (run_queue_enqueue(&processor->runq, new_thread, tail))
-                       try_preempt = TRUE;
+               /*
+                *      Bound case:
+                *
+                *      Unconditionally dispatch on the processor.
+                */
+               processor = thread->bound_processor;
+               pset = processor->processor_set;
+               pset_lock(pset);
+       }
 
-               if (processor == current_processor()) {
-                       if (try_preempt) {
-                               thread = processor->cpu_data->active_thread;
-                               if (csw_needed(thread, processor)) {
-                                       ast_on(preempt);
-                                       dispatch_counts.bound_self++;
+       /*
+        *      Dispatch the thread on the choosen processor.
+        */
+       if (thread->sched_pri >= BASEPRI_RTQUEUES)
+               realtime_setrun(processor, thread);
+       else
+               processor_setrun(processor, thread, options);
+}
+
+/*
+ *     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 = &processor->runq;
+       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) {
+                               remqueue(queue, (queue_entry_t)thread);
+
+                               thread->runq = PROCESSOR_NULL;
+                               rq->count--;
+                               if (testbit(pri, sched_preempt_pri)) {
+                                       rq->urgency--; assert(rq->urgency >= 0);
                                }
-                       }
-               }
-               else {
-                       if (try_preempt) {
-                               if (    processor->state == PROCESSOR_RUNNING                   &&
-                                               new_thread->sched_pri > processor->current_pri  ) {
-                                       cause_ast_check(processor);
-                                       dispatch_counts.bound_other++;
-                                       return;
+                               if (queue_empty(queue)) {
+                                       if (pri != IDLEPRI)
+                                               clrbit(MAXPRI - pri, rq->bitmap);
+                                       rq->highq = MAXPRI - ffsbit(rq->bitmap);
                                }
-                       }
 
-                       if (processor->state == PROCESSOR_IDLE) {
-                               machine_signal_idle(processor);
-                               dispatch_counts.bound_idle++;
+                               enqueue_tail(&tqueue, (queue_entry_t)thread);
                        }
+                       count--;
+
+                       thread = next;
                }
+
+               queue--; pri--;
+       }
+
+       pset_unlock(pset);
+
+       while ((thread = (thread_t)dequeue_head(&tqueue)) != THREAD_NULL) {
+               thread_lock(thread);
+
+               thread_setrun(thread, SCHED_TAILQ);
+
+               thread_unlock(thread);
        }
 }
 
 /*
- * Called at splsched by a thread on itself.
+ *     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_quantum(processor)) {
-               runq = &processor->processor_set->runq;
-               if (runq->highq > current_pri) {
+       if (first_timeslice(processor)) {
+               runq = &rt_runq;
+               if (runq->highq >= BASEPRI_RTQUEUES)
+                       return (AST_PREEMPT | AST_URGENT);
+
+               if (runq->highq > processor->current_pri) {
                        if (runq->urgency > 0)
-                               return (AST_BLOCK | AST_URGENT);
+                               return (AST_PREEMPT | AST_URGENT);
 
-                       result |= AST_BLOCK;
+                       result |= AST_PREEMPT;
                }
 
                runq = &processor->runq;
-               if (runq->highq > current_pri) {
+               if (runq->highq > processor->current_pri) {
                        if (runq->urgency > 0)
-                               return (AST_BLOCK | AST_URGENT);
+                               return (AST_PREEMPT | AST_URGENT);
 
-                       result |= AST_BLOCK;
+                       result |= AST_PREEMPT;
                }
        }
        else {
-               runq = &processor->processor_set->runq;
-               if (runq->highq >= current_pri) {
+               runq = &rt_runq;
+               if (runq->highq >= processor->current_pri) {
                        if (runq->urgency > 0)
-                               return (AST_BLOCK | AST_URGENT);
+                               return (AST_PREEMPT | AST_URGENT);
 
-                       result |= AST_BLOCK;
+                       result |= AST_PREEMPT;
                }
 
                runq = &processor->runq;
-               if (runq->highq >= current_pri) {
+               if (runq->highq >= processor->current_pri) {
                        if (runq->urgency > 0)
-                               return (AST_BLOCK | AST_URGENT);
+                               return (AST_PREEMPT | AST_URGENT);
 
-                       result |= AST_BLOCK;
+                       result |= AST_PREEMPT;
                }
        }
 
        if (result != AST_NONE)
                return (result);
 
-       if (thread->state & TH_SUSP)
-               result |= AST_BLOCK;
+       if (machine_cpu_is_inactive(processor->cpu_num))
+               return (AST_PREEMPT);
 
-       return (result);
+       if (processor->active_thread->state & TH_SUSP)
+               return (AST_PREEMPT);
+
+       return (AST_NONE);
 }
 
 /*
  *     set_sched_pri:
  *
- *     Set the current scheduled priority of the specified thread.
+ *     Set the scheduled priority of the specified thread.
+ *
  *     This may cause the thread to change queues.
  *
- *     The thread *must* be locked by the caller.
+ *     Thread must be locked.
  */
 void
 set_sched_pri(
-       thread_t                        thread,
-       int                                     priority)
+       thread_t                thread,
+       int                             priority)
 {
-       register struct run_queue       *rq = rem_runq(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_FORCEDPREEMPT)           )       )
-               thread->sched_mode |= TH_MODE_PREEMPT;
-       else
-               thread->sched_mode &= ~TH_MODE_PREEMPT;
+       boolean_t               removed = run_queue_remove(thread);
 
        thread->sched_pri = priority;
-       if (rq != RUN_QUEUE_NULL)
-               thread_setrun(thread, TAIL_Q);
+       if (removed)
+               thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ);
        else
-       if ((thread->state & (TH_RUN|TH_WAIT)) == TH_RUN) {
+       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;
+                       if ((preempt = csw_check(processor)) != AST_NONE)
+                               ast_on(preempt);
                }
                else
                if (    processor != PROCESSOR_NULL                                             &&
-                               processor->cpu_data->active_thread == thread    )
+                               processor->active_thread == thread      )
                        cause_ast_check(processor);
        }
 }
 
+#if            0
+
+static void
+run_queue_check(
+       run_queue_t             rq,
+       thread_t                thread)
+{
+       queue_t                 q;
+       queue_entry_t   qe;
+
+       if (rq != thread->runq)
+               panic("run_queue_check: thread runq");
+
+       if (thread->sched_pri > MAXPRI || thread->sched_pri < MINPRI)
+               panic("run_queue_check: thread sched_pri");
+
+       q = &rq->queues[thread->sched_pri];
+       qe = queue_first(q);
+       while (!queue_end(q, qe)) {
+               if (qe == (queue_entry_t)thread)
+                       return;
+
+               qe = queue_next(qe);
+       }
+
+       panic("run_queue_check: end");
+}
+
+#endif /* DEBUG */
+
 /*
- *     rem_runq:
+ *     run_queue_remove:
  *
- *     Remove a thread from its run queue.
- *     The run queue that the process was on is returned
- *     (or RUN_QUEUE_NULL if not on a run queue).  Thread *must* be locked
- *     before calling this routine.  Unusual locking protocol on runq
- *     field in thread structure makes this code interesting; see thread.h.
+ *     Remove a thread from a current run queue and
+ *     return TRUE if successful.
+ *
+ *     Thread must be locked.
  */
-run_queue_t
-rem_runq(
-       thread_t                        thread)
+boolean_t
+run_queue_remove(
+       thread_t                thread)
 {
-       register struct run_queue       *rq;
+       processor_t             processor = thread->runq;
 
-       rq = thread->runq;
        /*
-        *      If rq is RUN_QUEUE_NULL, the thread will stay out of the
-        *      run_queues because the caller locked the thread.  Otherwise
-        *      the thread is on a runq, but could leave.
+        *      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) {
-               simple_lock(&rq->lock);
-               if (rq == thread->runq) {
+       if (processor != PROCESSOR_NULL) {
+               void *                  rqlock;
+               run_queue_t             rq;
+
+               /*
+                *      The processor run queues are locked by the
+                *      processor set.  Real-time priorities use a
+                *      global queue with a dedicated lock.
+                */
+               if (thread->sched_pri < BASEPRI_RTQUEUES) {
+                       rqlock = &processor->processor_set->sched_lock;
+                       rq = &processor->runq;
+               }
+               else {
+                       rqlock = &rt_lock; rq = &rt_runq;
+               }
+
+               simple_lock(rqlock);
+
+               if (processor == thread->runq) {
                        /*
-                        *      Thread is in a runq and we have a lock on
-                        *      that runq.
+                        *      Thread is on a run queue and we have a lock on
+                        *      that run queue.
                         */
-#if    DEBUG
-                       thread_check(thread, rq);
-#endif /* DEBUG */
                        remqueue(&rq->queues[0], (queue_entry_t)thread);
                        rq->count--;
-                       if (thread->sched_mode & TH_MODE_PREEMPT)
-                               rq->urgency--;
-                       assert(rq->urgency >= 0);
+                       if (testbit(thread->sched_pri, sched_preempt_pri)) {
+                               rq->urgency--; assert(rq->urgency >= 0);
+                       }
 
                        if (queue_empty(rq->queues + thread->sched_pri)) {
                                /* update run queue status */
@@ -2222,470 +2723,453 @@ rem_runq(
                                        clrbit(MAXPRI - thread->sched_pri, rq->bitmap);
                                rq->highq = MAXPRI - ffsbit(rq->bitmap);
                        }
-                       thread->runq = RUN_QUEUE_NULL;
-                       simple_unlock(&rq->lock);
+
+                       thread->runq = PROCESSOR_NULL;
                }
                else {
                        /*
-                        *      The thread left the runq before we could
-                        *      lock the runq.  It is not on a runq now, and
-                        *      can't move again because this routine's
-                        *      caller locked the thread.
+                        *      The thread left the run queue before we could
+                        *      lock the run queue.
                         */
-                       assert(thread->runq == RUN_QUEUE_NULL);
-                       simple_unlock(&rq->lock);
-                       rq = RUN_QUEUE_NULL;
+                       assert(thread->runq == PROCESSOR_NULL);
+                       processor = PROCESSOR_NULL;
                }
+
+               simple_unlock(rqlock);
        }
 
-       return (rq);
+       return (processor != PROCESSOR_NULL);
 }
 
 /*
- *     choose_thread:
- *
- *     Choose a thread to execute.  The thread chosen is removed
- *     from its run queue.  Note that this requires only that the runq
- *     lock be held.
+ *     steal_processor_thread:
  *
- *     Strategy:
- *             Check processor runq first; if anything found, run it.
- *             Else check pset runq; if nothing found, return idle thread.
+ *     Locate a thread to steal from the processor and
+ *     return it.
  *
- *     Second line of strategy is implemented by choose_pset_thread.
- *
- *     Called with both the local & pset run queues locked, returned
- *     unlocked.
+ *     Associated pset must be locked.  Returns THREAD_NULL
+ *     on failure.
  */
-thread_t
-choose_thread(
-       processor_t             myprocessor)
+static thread_t
+steal_processor_thread(
+       processor_t             processor)
 {
-       thread_t                                thread;
-       register queue_t                q;
-       register run_queue_t    runq;
-       processor_set_t                 pset;
-
-       runq = &myprocessor->runq;
-       pset = myprocessor->processor_set;
-
-       if (runq->count > 0 && runq->highq >= pset->runq.highq) {
-               simple_unlock(&pset->runq.lock);
-               q = runq->queues + runq->highq;
-#if    MACH_ASSERT
-               if (!queue_empty(q)) {
-#endif /*MACH_ASSERT*/
-                       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);
+       run_queue_t             rq = &processor->runq;
+       queue_t                 queue = rq->queues + rq->highq;
+       int                             pri = rq->highq, count = rq->count;
+       thread_t                thread;
+
+       while (count > 0) {
+               thread = (thread_t)queue_first(queue);
+               while (!queue_end(queue, (queue_entry_t)thread)) {
+                       if (thread->bound_processor != processor) {
+                               remqueue(queue, (queue_entry_t)thread);
+
+                               thread->runq = PROCESSOR_NULL;
+                               rq->count--;
+                               if (testbit(pri, sched_preempt_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);
                        }
-                       simple_unlock(&runq->lock);
-                       return (thread);
-#if    MACH_ASSERT
+                       count--;
+
+                       thread = (thread_t)queue_next((queue_entry_t)thread);
                }
-               panic("choose_thread");
-#endif /*MACH_ASSERT*/
-               /*NOTREACHED*/
+
+               queue--; pri--;
        }
-       simple_unlock(&myprocessor->runq.lock);
 
-       return (choose_pset_thread(myprocessor, pset));
+       return (THREAD_NULL);
 }
 
 /*
- *     choose_pset_thread:  choose a thread from processor_set runq or
- *             set processor idle and choose its idle thread.
+ *     Locate and steal a thread, beginning
+ *     at the pset.
  *
- *     This routine chooses and removes a thread from the runq if there
- *     is one (and returns it), else it sets the processor idle and
- *     returns its idle thread.
- *
- *     Called with both local & pset run queues locked, returned
+ *     The pset must be locked, and is returned
  *     unlocked.
+ *
+ *     Returns the stolen thread, or THREAD_NULL on
+ *     failure.
  */
-thread_t
-choose_pset_thread(
-       register processor_t    myprocessor,
-       processor_set_t                 pset)
+static thread_t
+steal_thread(
+       processor_set_t         pset)
 {
-       register run_queue_t    runq;
-       register thread_t               thread;
-       register queue_t                q;
+       processor_set_t         nset, cset = pset;
+       processor_t                     processor;
+       thread_t                        thread;
+
+       do {
+               processor = (processor_t)queue_first(&cset->active_queue);
+               while (!queue_end(&cset->active_queue, (queue_entry_t)processor)) {
+                       if (processor->runq.count > 0) {
+                               thread = steal_processor_thread(processor);
+                               if (thread != THREAD_NULL) {
+                                       remqueue(&cset->active_queue, (queue_entry_t)processor);
+                                       enqueue_tail(&cset->active_queue, (queue_entry_t)processor);
 
-       runq = &pset->runq;
-       if (runq->count > 0) {
-               q = runq->queues + runq->highq;
-#if    MACH_ASSERT
-               if (!queue_empty(q)) {
-#endif /*MACH_ASSERT*/
-                       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);
+                                       pset_unlock(cset);
+
+                                       return (thread);
+                               }
                        }
-                       pset_quanta_update(pset);
-                       simple_unlock(&runq->lock);
-                       return (thread);
-#if    MACH_ASSERT
+
+                       processor = (processor_t)queue_next((queue_entry_t)processor);
                }
-               panic("choose_pset_thread");
-#endif /*MACH_ASSERT*/
-               /*NOTREACHED*/
-       }
-       simple_unlock(&runq->lock);
 
-       /*
-        *      Nothing is runnable, so set this processor idle if it
-        *      was running.  If it was in an assignment or shutdown,
-        *      leave it alone.  Return its idle thread.
-        */
-       simple_lock(&pset->sched_lock);
-       if (myprocessor->state == PROCESSOR_RUNNING) {
-               remqueue(&pset->active_queue, (queue_entry_t)myprocessor);
-           myprocessor->state = PROCESSOR_IDLE;
+               nset = next_pset(cset);
 
-           if (myprocessor == master_processor)
-                       enqueue_tail(&pset->idle_queue, (queue_entry_t)myprocessor);
-           else
-                       enqueue_head(&pset->idle_queue, (queue_entry_t)myprocessor);
+               if (nset != pset) {
+                       pset_unlock(cset);
 
-           pset->idle_count++;
-       }
-       simple_unlock(&pset->sched_lock);
+                       cset = nset;
+                       pset_lock(cset);
+               }
+       } while (nset != pset);
 
-       return (myprocessor->idle_thread);
-}
+       pset_unlock(cset);
 
-/*
- *     no_dispatch_count counts number of times processors go non-idle
- *     without being dispatched.  This should be very rare.
- */
-int    no_dispatch_count = 0;
+       return (THREAD_NULL);
+}
 
 /*
- *     This is the idle 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_continue(void)
+static thread_t
+processor_idle(
+       thread_t                        thread,
+       processor_t                     processor)
 {
-       register processor_t            myprocessor;
-       register volatile thread_t      *threadp;
-       register volatile int           *gcount;
-       register volatile int           *lcount;
-       register thread_t                       new_thread;
-       register int                            state;
-       register processor_set_t        pset;
-       int                                                     mycpu;
+       processor_set_t         pset = processor->processor_set;
+       thread_t                        new_thread;
+       int                                     state;
+
+       (void)splsched();
 
-       mycpu = cpu_number();
-       myprocessor = cpu_to_processor(mycpu);
-       threadp = (volatile thread_t *) &myprocessor->next_thread;
-       lcount = (volatile int *) &myprocessor->runq.count;
+#ifdef __ppc__
+       pmsDown();                                      /* Step power down */
+#endif
+
+       KERNEL_DEBUG_CONSTANT(
+               MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_START, (int)thread, 0, 0, 0, 0);
 
-       for (;;) {
-               gcount = (volatile int *)&myprocessor->processor_set->runq.count;
+       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 (processor->next_thread == THREAD_NULL && processor->runq.count == 0 && rt_runq.count == 0 &&
+                               (thread == THREAD_NULL || ((thread->state & (TH_WAIT|TH_SUSP)) == TH_WAIT && !thread->wake_active))) {
+               machine_idle();
 
                (void)splsched();
-               while ( (*threadp == (volatile thread_t)THREAD_NULL)    &&
-                                       (*gcount == 0) && (*lcount == 0)                                ) {
-
-                       /* check for ASTs while we wait */
-                       if (need_ast[mycpu] &~ (        AST_SCHEDULING | AST_BSD        )) {
-                               /* don't allow scheduling ASTs */
-                               need_ast[mycpu] &= ~(   AST_SCHEDULING | AST_BSD        );
-                               ast_taken(AST_ALL, TRUE);       /* back at spllo */
-                       }
-                       else
-#ifdef __ppc__
-                               machine_idle();
-#else
-                               (void)spllo();
+
+               if (processor->state == PROCESSOR_INACTIVE && !machine_cpu_is_inactive(processor->cpu_num))
+                       break;
+       }
+
+       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);
+
+       pset_lock(pset);
+
+#ifdef __ppc__
+       pmsStep(0);                                     /* Step up out of idle power */
 #endif
-               machine_clock_assist();
 
-                       (void)splsched();
+       state = processor->state;
+       if (state == PROCESSOR_DISPATCHING) {
+               /*
+                *      Commmon case -- cpu dispatched.
+                */
+               new_thread = processor->next_thread;
+               processor->next_thread = THREAD_NULL;
+               processor->state = PROCESSOR_RUNNING;
+
+               if (    processor->runq.highq > new_thread->sched_pri                                   ||
+                               (rt_runq.highq > 0 && rt_runq.highq >= new_thread->sched_pri)   ) {
+                       processor->deadline = UINT64_MAX;
+
+                       pset_unlock(pset);
+
+                       thread_lock(new_thread);
+                       thread_setrun(new_thread, SCHED_HEADQ);
+                       thread_unlock(new_thread);
+
+                       KERNEL_DEBUG_CONSTANT(
+                               MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (int)thread, (int)state, 0, 0, 0);
+       
+                       return (THREAD_NULL);
                }
 
+               pset_unlock(pset);
+
+               KERNEL_DEBUG_CONSTANT(
+                       MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (int)thread, (int)state, (int)new_thread, 0, 0);
+
+               return (new_thread);
+       }
+       else
+       if (state == PROCESSOR_IDLE) {
+               remqueue(&pset->idle_queue, (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);
+       }
+       else
+       if (state == PROCESSOR_SHUTDOWN) {
                /*
-                *      This is not a switch statement to avoid the
-                *      bounds checking code in the common case.
+                *      Going off-line.  Force a
+                *      reschedule.
                 */
-               pset = myprocessor->processor_set;
-               simple_lock(&pset->sched_lock);
-retry:
-               state = myprocessor->state;
-               if (state == PROCESSOR_DISPATCHING) {
-                       /*
-                        *      Commmon case -- cpu dispatched.
-                        */
-                       new_thread = *threadp;
-                       *threadp = (volatile thread_t) THREAD_NULL;
-                       myprocessor->state = PROCESSOR_RUNNING;
-                       enqueue_tail(&pset->active_queue, (queue_entry_t)myprocessor);
-                       simple_unlock(&pset->sched_lock);
-
-                       if (    myprocessor->runq.highq > new_thread->sched_pri         ||
-                                       pset->runq.highq > new_thread->sched_pri                                ) {
-                               thread_lock(new_thread);
-                               thread_setrun(new_thread, HEAD_Q);
-                               thread_unlock(new_thread);
-
-                               counter(c_idle_thread_block++);
-                               thread_block(idle_thread_continue);
-                               /* NOTREACHED */
-                       }
-                       else {
-                               counter(c_idle_thread_handoff++);
-                               thread_run(myprocessor->idle_thread,
-                                                                       idle_thread_continue, new_thread);
-                               /* NOTREACHED */
-                       }
-               }
-               else
-               if (state == PROCESSOR_IDLE) {
-                       if (myprocessor->state != PROCESSOR_IDLE) {
-                               /*
-                                *      Something happened, try again.
-                                */
-                               goto retry;
-                       }
-                       /*
-                        *      Processor was not dispatched (Rare).
-                        *      Set it running again.
-                        */
-                       no_dispatch_count++;
-                       pset->idle_count--;
-                       remqueue(&pset->idle_queue, (queue_entry_t)myprocessor);
-                       myprocessor->state = PROCESSOR_RUNNING;
-                       enqueue_tail(&pset->active_queue, (queue_entry_t)myprocessor);
-                       simple_unlock(&pset->sched_lock);
-
-                       counter(c_idle_thread_block++);
-                       thread_block(idle_thread_continue);
-                       /* NOTREACHED */
-               }
-               else
-               if (    state == PROCESSOR_ASSIGN               ||
-                               state == PROCESSOR_SHUTDOWN                     ) {
-                       /*
-                        *      Changing processor sets, or going off-line.
-                        *      Release next_thread if there is one.  Actual
-                        *      thread to run is on a runq.
-                        */
-                       if ((new_thread = (thread_t)*threadp) != THREAD_NULL) {
-                               *threadp = (volatile thread_t) THREAD_NULL;
-                               simple_unlock(&pset->sched_lock);
+               if ((new_thread = processor->next_thread) != THREAD_NULL) {
+                       processor->next_thread = THREAD_NULL;
+                       processor->deadline = UINT64_MAX;
 
-                               thread_lock(new_thread);
-                               thread_setrun(new_thread, TAIL_Q);
-                               thread_unlock(new_thread);
-                       }
-                       else
-                               simple_unlock(&pset->sched_lock);
+                       pset_unlock(pset);
 
-                       counter(c_idle_thread_block++);
-                       thread_block(idle_thread_continue);
-                       /* NOTREACHED */
-               }
-               else {
-                       simple_unlock(&pset->sched_lock);
+                       thread_lock(new_thread);
+                       thread_setrun(new_thread, SCHED_HEADQ);
+                       thread_unlock(new_thread);
 
-                       panic("idle_thread: bad processor state %d\n", cpu_state(mycpu));
-               }
+                       KERNEL_DEBUG_CONSTANT(
+                               MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (int)thread, (int)state, 0, 0, 0);
 
-               (void)spllo();
+                       return (THREAD_NULL);
+               }
        }
+
+       pset_unlock(pset);
+
+       KERNEL_DEBUG_CONSTANT(
+               MACHDBG_CODE(DBG_MACH_SCHED,MACH_IDLE) | DBG_FUNC_END, (int)thread, (int)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)
 {
-       thread_t                self = current_thread();
-       spl_t                   s;
-
-       stack_privilege(self);
+       processor_t             processor = current_processor();
+       thread_t                new_thread;
 
-       s = splsched();
-       thread_lock(self);
-       self->priority = IDLEPRI;
-       set_sched_pri(self, self->priority);
-       thread_unlock(self);
-       splx(s);
+       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*/
+       }
 
-       counter(c_idle_thread_block++);
-       thread_block(idle_thread_continue);
+       thread_block((thread_continue_t)idle_thread);
        /*NOTREACHED*/
 }
 
-static uint64_t                                sched_tick_interval, sched_tick_deadline;
+kern_return_t
+idle_thread_create(
+       processor_t             processor)
+{
+       kern_return_t   result;
+       thread_t                thread;
+       spl_t                   s;
 
-void   sched_tick_thread(void);
+       result = kernel_thread_create((thread_continue_t)idle_thread, NULL, MAXPRI_KERNEL, &thread);
+       if (result != KERN_SUCCESS)
+               return (result);
 
-void
-sched_tick_init(void)
-{
-       kernel_thread_with_priority(
-                                               kernel_task, MAXPRI_STANDARD,
-                                                               sched_tick_thread, TRUE, TRUE);
+       s = splsched();
+       thread_lock(thread);
+       thread->bound_processor = processor;
+       processor->idle_thread = thread;
+       thread->sched_pri = thread->priority = IDLEPRI;
+       thread->state = (TH_RUN | TH_IDLE);
+       thread_unlock(thread);
+       splx(s);
+
+       thread_deallocate(thread);
+
+       return (KERN_SUCCESS);
 }
 
+static uint64_t                sched_tick_deadline;
+
 /*
- *     sched_tick_thread
+ * sched_startup:
+ *
+ * Kicks off scheduler services.
  *
- *     Update the priorities of all threads periodically.
+ * Called at splsched.
  */
 void
-sched_tick_thread_continue(void)
+sched_startup(void)
 {
-       uint64_t                        abstime;
-#if    SIMPLE_CLOCK
-       int                                     new_usec;
-#endif /* SIMPLE_CLOCK */
+       kern_return_t   result;
+       thread_t                thread;
 
-       clock_get_uptime(&abstime);
+       result = kernel_thread_start_priority((thread_continue_t)sched_tick_thread, NULL, MAXPRI_KERNEL, &thread);
+       if (result != KERN_SUCCESS)
+               panic("sched_startup");
+
+       thread_deallocate(thread);
 
-       sched_tick++;           /* age usage one more time */
-#if    SIMPLE_CLOCK
        /*
-        *      Compensate for clock drift.  sched_usec is an
-        *      exponential average of the number of microseconds in
-        *      a second.  It decays in the same fashion as cpu_usage.
+        * Yield to the sched_tick_thread while it times
+        * a series of context switches back.  It stores
+        * the baseline value in sched_cswtime.
+        *
+        * The current thread is the only other thread
+        * active at this point.
         */
-       new_usec = sched_usec_elapsed();
-       sched_usec = (5*sched_usec + 3*new_usec)/8;
-#endif /* SIMPLE_CLOCK */
+       while (sched_cswtime == 0)
+               thread_block(THREAD_CONTINUE_NULL);
+
+       thread_daemon_init();
+
+       thread_call_initialize();
+}
+
+/*
+ *     sched_tick_thread:
+ *
+ *     Perform periodic bookkeeping functions about ten
+ *     times per second.
+ */
+static void
+sched_tick_continue(void)
+{
+       uint64_t                        abstime = mach_absolute_time();
+
+       sched_tick++;
 
        /*
-        *  Compute the scheduler load factors.
+        *  Compute various averages.
         */
-       compute_mach_factor();
+       compute_averages();
 
        /*
-        *  Scan the run queues for runnable threads that need to
-        *  have their priorities recalculated.
+        *  Scan the run queues for threads which
+        *  may need to be updated.
         */
-       do_thread_scan();
+       thread_update_scan();
 
        clock_deadline_for_periodic_event(sched_tick_interval, abstime,
                                                                                                                &sched_tick_deadline);
 
-       assert_wait((event_t)sched_tick_thread_continue, THREAD_INTERRUPTIBLE);
-       thread_set_timer_deadline(sched_tick_deadline);
-       thread_block(sched_tick_thread_continue);
+       assert_wait_deadline((event_t)sched_tick_thread, THREAD_UNINT, sched_tick_deadline);
+       thread_block((thread_continue_t)sched_tick_continue);
        /*NOTREACHED*/
 }
 
+/*
+ * Time a series of context switches to determine
+ * a baseline.  Toss the high and low and return
+ * the one-way value.
+ */
+static uint32_t
+time_cswitch(void)
+{
+       uint32_t        new, hi, low, accum;
+       uint64_t        abstime;
+       int                     i, tries = 7;
+
+       accum = hi = low = 0;
+       for (i = 0; i < tries; ++i) {
+               abstime = mach_absolute_time();
+               thread_block(THREAD_CONTINUE_NULL);
+
+               new = mach_absolute_time() - abstime;
+
+               if (i == 0)
+                       accum = hi = low = new;
+               else {
+                       if (new < low)
+                               low = new;
+                       else
+                       if (new > hi)
+                               hi = new;
+                       accum += new;
+               }
+       }
+
+       return ((accum - hi - low) / (2 * (tries - 2)));
+}
+
 void
 sched_tick_thread(void)
 {
-       thread_t                self = current_thread();
-       natural_t               rate;
-       spl_t                   s;
+       sched_cswtime = time_cswitch();
 
-       stack_privilege(self);
+       sched_tick_deadline = mach_absolute_time();
 
-       rate = (1000 >> SCHED_TICK_SHIFT);
-       clock_interval_to_absolutetime_interval(rate, USEC_PER_SEC,
-                                                                                               &sched_tick_interval);
-       clock_get_uptime(&sched_tick_deadline);
-
-       thread_block(sched_tick_thread_continue);
+       sched_tick_continue();
        /*NOTREACHED*/
 }
 
-#define        MAX_STUCK_THREADS       128
-
 /*
- *     do_thread_scan: scan for stuck threads.  A thread is stuck if
- *     it is runnable but its priority is so low that it has not
- *     run for several seconds.  Its priority should be higher, but
- *     won't be until it runs and calls update_priority.  The scanner
- *     finds these threads and does the updates.
+ *     thread_update_scan / runq_scan:
+ *
+ *     Scan the run queues to account for timesharing threads 
+ *     which need to be updated.
  *
  *     Scanner runs in two passes.  Pass one squirrels likely
- *     thread ids away in an array  (takes out references for them).
- *     Pass two does the priority updates.  This is necessary because
- *     the run queue lock is required for the candidate scan, but
- *     cannot be held during updates.
+ *     threads away in an array, pass two does the update.
  *
- *     Array length should be enough so that restart isn't necessary,
- *     but restart logic is included.
+ *     This is necessary because the run queue is locked for
+ *     the candidate scan, but the thread is locked for the update.
  *
+ *     Array should be sized to make forward progress, without
+ *     disabling preemption for long periods.
  */
-thread_t               stuck_threads[MAX_STUCK_THREADS];
-int                            stuck_count = 0;
+
+#define        THREAD_UPDATE_SIZE              128
+
+static thread_t                thread_update_array[THREAD_UPDATE_SIZE];
+static int                     thread_update_count = 0;
 
 /*
- *     do_runq_scan is the guts of pass 1.  It scans a runq for
- *     stuck threads.  A boolean is returned indicating whether
- *     a retry is needed.
+ *     Scan a runq for candidate threads.
+ *
+ *     Returns TRUE if retry is needed.
  */
-boolean_t
-do_runq_scan(
+static boolean_t
+runq_scan(
        run_queue_t                             runq)
 {
+       register int                    count;
        register queue_t                q;
        register thread_t               thread;
-       register int                    count;
-       spl_t                                   s;
-       boolean_t                               result = FALSE;
 
-       s = splsched();
-       simple_lock(&runq->lock);
        if ((count = runq->count) > 0) {
            q = runq->queues + runq->highq;
                while (count > 0) {
                        queue_iterate(q, thread, thread_t, links) {
-                               if (    !(thread->state & (TH_WAIT|TH_SUSP))            &&
+                               if (            thread->sched_stamp != sched_tick               &&
                                                (thread->sched_mode & TH_MODE_TIMESHARE)        ) {
-                                       if (thread->sched_stamp != sched_tick) {
-                                               /*
-                                                *      Stuck, save its id for later.
-                                                */
-                                               if (stuck_count == MAX_STUCK_THREADS) {
-                                                       /*
-                                                        *      !@#$% No more room.
-                                                        */
-                                                       simple_unlock(&runq->lock);
-                                                       splx(s);
-
-                                                       return (TRUE);
-                                               }
+                                       if (thread_update_count == THREAD_UPDATE_SIZE)
+                                               return (TRUE);
 
-                                               /*
-                                                * Inline version of thread_reference
-                                                * XXX - lock ordering problem here:
-                                                * thread locks should be taken before runq
-                                                * locks: just try and get the thread's locks
-                                                * and ignore this thread if we fail, we might
-                                                * have better luck next time.
-                                                */
-                                               if (thread_lock_try(thread)) {
-                                                       thread->ref_count++;
-                                                       thread_unlock(thread);
-                                                       stuck_threads[stuck_count++] = thread;
-                                               }
-                                               else
-                                                       result = TRUE;
-                                       }
+                                       thread_update_array[thread_update_count++] = thread;
+                                       thread_reference_internal(thread);
                                }
 
                                count--;
@@ -2694,73 +3178,63 @@ do_runq_scan(
                        q--;
                }
        }
-       simple_unlock(&runq->lock);
-       splx(s);
 
-       return (result);
+       return (FALSE);
 }
 
-boolean_t      thread_scan_enabled = TRUE;
-
-void
-do_thread_scan(void)
+static void
+thread_update_scan(void)
 {
-       register boolean_t                      restart_needed = FALSE;
-       register thread_t                       thread;
-       register processor_set_t        pset = &default_pset;
-       register processor_t            processor;
-       spl_t                                           s;
-
-       if (!thread_scan_enabled)
-               return;
+       boolean_t                       restart_needed = FALSE;
+       processor_t                     processor = processor_list;
+       processor_set_t         pset;
+       thread_t                        thread;
+       spl_t                           s;
 
        do {
-           restart_needed = do_runq_scan(&pset->runq);
-               if (!restart_needed) {
-                       simple_lock(&pset->processors_lock);
-                       processor = (processor_t)queue_first(&pset->processors);
-                       while (!queue_end(&pset->processors, (queue_entry_t)processor)) {
-                               if (restart_needed = do_runq_scan(&processor->runq))
-                                       break;
+               do {
+                       pset = processor->processor_set;
 
-                               thread = processor->idle_thread;
-                               if (thread->sched_stamp != sched_tick) {
-                                       if (stuck_count == MAX_STUCK_THREADS) {
-                                               restart_needed = TRUE;
-                                               break;
-                                       }
+                       s = splsched();
+                       pset_lock(pset);
+
+                       restart_needed = runq_scan(&processor->runq);
+
+                       pset_unlock(pset);
+                       splx(s);
 
-                                       stuck_threads[stuck_count++] = thread;
+                       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->processors_lock);
-               }
+               } while ((processor = processor->processor_list) != NULL);
 
            /*
             *  Ok, we now have a collection of candidates -- fix them.
             */
-           while (stuck_count > 0) {
-                       thread = stuck_threads[--stuck_count];
-                       stuck_threads[stuck_count] = THREAD_NULL;
+           while (thread_update_count > 0) {
+                       thread = thread_update_array[--thread_update_count];
+                       thread_update_array[thread_update_count] = THREAD_NULL;
+
                        s = splsched();
                        thread_lock(thread);
-                       if (    (thread->sched_mode & TH_MODE_TIMESHARE)        ||
-                                                       (thread->state & TH_IDLE)                               ) {
-                               if (    !(thread->state & (TH_WAIT|TH_SUSP))    &&
-                                               thread->sched_stamp != sched_tick               )
-                                       update_priority(thread);
-                       }
+                       if (    !(thread->state & (TH_WAIT|TH_SUSP))    &&
+                                               thread->sched_stamp != sched_tick       )
+                               update_priority(thread);
                        thread_unlock(thread);
                        splx(s);
-                       if (!(thread->state & TH_IDLE))
-                               thread_deallocate(thread);
-           }
 
-               if (restart_needed)
-                       delay(1);                       /* XXX */
-               
+                       thread_deallocate(thread);
+           }
        } while (restart_needed);
 }
                
@@ -2779,190 +3253,24 @@ thread_wakeup(
        thread_wakeup_with_result(x, THREAD_AWAKENED);
 }
 
+boolean_t
+preemption_enabled(void)
+{
+       return (get_preemption_level() == 0 && ml_get_interrupts_enabled());
+}
 
 #if    DEBUG
-
 static boolean_t
 thread_runnable(
        thread_t        thread)
 {
        return ((thread->state & (TH_RUN|TH_WAIT)) == TH_RUN);
 }
-
-void
-dump_processor_set(
-       processor_set_t ps)
-{
-    printf("processor_set: %08x\n",ps);
-    printf("idle_queue: %08x %08x, idle_count:      0x%x\n",
-       ps->idle_queue.next,ps->idle_queue.prev,ps->idle_count);
-    printf("processors: %08x %08x, processor_count: 0x%x\n",
-       ps->processors.next,ps->processors.prev,ps->processor_count);
-    printf("tasks:      %08x %08x, task_count:      0x%x\n",
-       ps->tasks.next,ps->tasks.prev,ps->task_count);
-    printf("threads:    %08x %08x, thread_count:    0x%x\n",
-       ps->threads.next,ps->threads.prev,ps->thread_count);
-    printf("ref_count: 0x%x, active: %x\n",
-       ps->ref_count,ps->active);
-    printf("pset_self: %08x, pset_name_self: %08x\n",ps->pset_self, ps->pset_name_self);
-    printf("set_quanta: 0x%x\n", ps->set_quanta);
-}
-
-#define processor_state(s) (((s)>PROCESSOR_SHUTDOWN)?"*unknown*":states[s])
-
-void
-dump_processor(
-       processor_t     p)
-{
-    char *states[]={"OFF_LINE","RUNNING","IDLE","DISPATCHING",
-                  "ASSIGN","SHUTDOWN"};
-
-    printf("processor: %08x\n",p);
-    printf("processor_queue: %08x %08x\n",
-       p->processor_queue.next,p->processor_queue.prev);
-    printf("state: %8s, next_thread: %08x, idle_thread: %08x\n",
-       processor_state(p->state), p->next_thread, p->idle_thread);
-    printf("slice_quanta: %x\n", p->slice_quanta);
-    printf("processor_set: %08x, processor_set_next: %08x\n",
-       p->processor_set, p->processor_set_next);
-    printf("processors: %08x %08x\n", p->processors.next,p->processors.prev);
-    printf("processor_self: %08x, slot_num: 0x%x\n", p->processor_self, p->slot_num);
-}
-
-void
-dump_run_queue_struct(
-       run_queue_t     rq)
-{
-    char dump_buf[80];
-    int i;
-
-    for( i=0; i < NRQS; ) {
-        int j;
-
-       printf("%6s",(i==0)?"runq:":"");
-       for( j=0; (j<8) && (i < NRQS); j++,i++ ) {
-           if( rq->queues[i].next == &rq->queues[i] )
-               printf( " --------");
-           else
-               printf(" %08x",rq->queues[i].next);
-       }
-       printf("\n");
-    }
-    for( i=0; i < NRQBM; ) {
-        register unsigned int mask;
-       char *d=dump_buf;
-
-       mask = ~0;
-       mask ^= (mask>>1);
-
-       do {
-           *d++ = ((rq->bitmap[i]&mask)?'r':'e');
-           mask >>=1;
-       } while( mask );
-       *d = '\0';
-       printf("%8s%s\n",((i==0)?"bitmap:":""),dump_buf);
-       i++;
-    }  
-    printf("highq: 0x%x, count: %u\n", rq->highq, rq->count);
-}
-void
-dump_run_queues(
-       run_queue_t     runq)
-{
-       register queue_t        q1;
-       register int            i;
-       register queue_entry_t  e;
-
-       q1 = runq->queues;
-       for (i = 0; i < NRQS; i++) {
-           if (q1->next != q1) {
-               int t_cnt;
-
-               printf("[%u]",i);
-               for (t_cnt=0, e = q1->next; e != q1; e = e->next) {
-                   printf("\t0x%08x",e);
-                   if( (t_cnt = ++t_cnt%4) == 0 )
-                       printf("\n");
-               }
-               if( t_cnt )
-                       printf("\n");
-           }
-           /* else
-               printf("[%u]\t<empty>\n",i);
-            */
-           q1++;
-       }
-}
-
-void
-checkrq(
-       run_queue_t     rq,
-       char            *msg)
-{
-       register queue_t        q1;
-       register int            i, j;
-       register queue_entry_t  e;
-       register int            highq;
-
-       highq = NRQS;
-       j = 0;
-       q1 = rq->queues;
-       for (i = MAXPRI; i >= 0; i--) {
-           if (q1->next == q1) {
-               if (q1->prev != q1) {
-                   panic("checkrq: empty at %s", msg);
-               }
-           }
-           else {
-               if (highq == -1)
-                   highq = i;
-               
-               for (e = q1->next; e != q1; e = e->next) {
-                   j++;
-                   if (e->next->prev != e)
-                       panic("checkrq-2 at %s", msg);
-                   if (e->prev->next != e)
-                       panic("checkrq-3 at %s", msg);
-               }
-           }
-           q1++;
-       }
-       if (j != rq->count)
-           panic("checkrq: count wrong at %s", msg);
-       if (rq->count != 0 && highq > rq->highq)
-           panic("checkrq: highq wrong at %s", msg);
-}
-
-void
-thread_check(
-       register thread_t               thread,
-       register run_queue_t    rq)
-{
-       register int                    whichq = thread->sched_pri;
-       register queue_entry_t  queue, entry;
-
-       if (whichq < MINPRI || whichq > MAXPRI)
-               panic("thread_check: bad pri");
-
-       queue = &rq->queues[whichq];
-       entry = queue_first(queue);
-       while (!queue_end(queue, entry)) {
-               if (entry == (queue_entry_t)thread)
-                       return;
-
-               entry = queue_next(entry);
-       }
-
-       panic("thread_check: not found");
-}
-
 #endif /* DEBUG */
 
 #if    MACH_KDB
 #include <ddb/db_output.h>
 #define        printf          kdbprintf
-extern int             db_indent;
 void                   db_sched(void);
 
 void
@@ -2975,11 +3283,19 @@ 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;
 }
+
+#include <ddb/db_output.h>
+void           db_show_thread_log(void);
+
+void
+db_show_thread_log(void)
+{
+}
 #endif /* MACH_KDB */