Subversion Repositories HelenOS

/*

/*

 * Copyright (c) 2001-2007 Jakub Jermar

 * Copyright (c) 2001-2007 Jakub Jermar

 * All rights reserved.

 * All rights reserved.

 *

 *

 * Redistribution and use in source and binary forms, with or without

 * Redistribution and use in source and binary forms, with or without

 * modification, are permitted provided that the following conditions

 * modification, are permitted provided that the following conditions

 * are met:

 * are met:

 *

 *

 * - Redistributions of source code must retain the above copyright

 * - Redistributions of source code must retain the above copyright

 *   notice, this list of conditions and the following disclaimer.

 *   notice, this list of conditions and the following disclaimer.

 * - Redistributions in binary form must reproduce the above copyright

 * - Redistributions in binary form must reproduce the above copyright

 *   notice, this list of conditions and the following disclaimer in the

 *   notice, this list of conditions and the following disclaimer in the

 *   documentation and/or other materials provided with the distribution.

 *   documentation and/or other materials provided with the distribution.

 * - The name of the author may not be used to endorse or promote products

 * - The name of the author may not be used to endorse or promote products

 *   derived from this software without specific prior written permission.

 *   derived from this software without specific prior written permission.

 *

 *

 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR

 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR

 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES

 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES

 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.

 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.

 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,

 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,

 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT

 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT

 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,

 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,

 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY

 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY

 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT

 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT

 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF

 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF

 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 */

 */

/** @addtogroup genericproc

/** @addtogroup genericproc

 * @{

 * @{

 */

 */

/**

/**

 * @file

 * @file

 * @brief   Scheduler and load balancing.

 * @brief   Scheduler and load balancing.

 *

 *

 * This file contains the scheduler and kcpulb kernel thread which

 * This file contains the scheduler and kcpulb kernel thread which

 * performs load-balancing of per-CPU run queues.

 * performs load-balancing of per-CPU run queues.

 */

 */

#include <proc/scheduler.h>

#include <proc/scheduler.h>

#include <proc/thread.h>

#include <proc/thread.h>

#include <proc/task.h>

#include <proc/task.h>

#include <mm/frame.h>

#include <mm/frame.h>

#include <mm/page.h>

#include <mm/page.h>

#include <mm/as.h>

#include <mm/as.h>

#include <time/delay.h>

#include <time/delay.h>

#include <arch/asm.h>

#include <arch/asm.h>

#include <arch/faddr.h>

#include <arch/faddr.h>

#include <arch/cycle.h>

#include <arch/cycle.h>

#include <atomic.h>

#include <atomic.h>

#include <synch/spinlock.h>

#include <synch/spinlock.h>

#include <config.h>

#include <config.h>

#include <context.h>

#include <context.h>

#include <func.h>

#include <func.h>

#include <arch.h>

#include <arch.h>

#include <adt/list.h>

#include <adt/list.h>

#include <panic.h>

#include <panic.h>

#include <cpu.h>

#include <cpu.h>

#include <print.h>

#include <print.h>

#include <debug.h>

#include <debug.h>

static void before_task_runs(void);

static void before_task_runs(void);

static void before_thread_runs(void);

static void before_thread_runs(void);

static void after_thread_ran(void);

static void after_thread_ran(void);

static void scheduler_separated_stack(void);

static void scheduler_separated_stack(void);

atomic_t nrdy;  /**< Number of ready threads in the system. */

atomic_t nrdy;  /**< Number of ready threads in the system. */

/** Carry out actions before new task runs. */

/** Carry out actions before new task runs. */

void before_task_runs(void)

void before_task_runs(void)

{

{

    before_task_runs_arch();

    before_task_runs_arch();

}

}

/** Take actions before new thread runs.

/** Take actions before new thread runs.

 *

 *

 * Perform actions that need to be

 * Perform actions that need to be

 * taken before the newly selected

 * taken before the newly selected

 * tread is passed control.

 * tread is passed control.

 *

 *

 * THREAD->lock is locked on entry

 * THREAD->lock is locked on entry

 *

 *

 */

 */

void before_thread_runs(void)

void before_thread_runs(void)

{

{

    before_thread_runs_arch();

    before_thread_runs_arch();

#ifdef CONFIG_FPU_LAZY

#ifdef CONFIG_FPU_LAZY

    if(THREAD == CPU->fpu_owner)

    if(THREAD == CPU->fpu_owner)

        fpu_enable();

        fpu_enable();

    else

    else

        fpu_disable();

        fpu_disable();

#else

#else

    fpu_enable();

    fpu_enable();

    if (THREAD->fpu_context_exists)

    if (THREAD->fpu_context_exists)

        fpu_context_restore(THREAD->saved_fpu_context);

        fpu_context_restore(THREAD->saved_fpu_context);

    else {

    else {

        fpu_init();

        fpu_init();

        THREAD->fpu_context_exists = 1;

        THREAD->fpu_context_exists = 1;

    }

    }

#endif

#endif

}

}

/** Take actions after THREAD had run.

/** Take actions after THREAD had run.

 *

 *

 * Perform actions that need to be

 * Perform actions that need to be

 * taken after the running thread

 * taken after the running thread

 * had been preempted by the scheduler.

 * had been preempted by the scheduler.

 *

 *

 * THREAD->lock is locked on entry

 * THREAD->lock is locked on entry

 *

 *

 */

 */

void after_thread_ran(void)

void after_thread_ran(void)

{

{

    after_thread_ran_arch();

    after_thread_ran_arch();

}

}

#ifdef CONFIG_FPU_LAZY

#ifdef CONFIG_FPU_LAZY

void scheduler_fpu_lazy_request(void)

void scheduler_fpu_lazy_request(void)

{

{

restart:

restart:

    fpu_enable();

    fpu_enable();

    spinlock_lock(&CPU->lock);

    spinlock_lock(&CPU->lock);

    /* Save old context */

    /* Save old context */

    if (CPU->fpu_owner != NULL) {  

    if (CPU->fpu_owner != NULL) {  

        spinlock_lock(&CPU->fpu_owner->lock);

        spinlock_lock(&CPU->fpu_owner->lock);

        fpu_context_save(CPU->fpu_owner->saved_fpu_context);

        fpu_context_save(CPU->fpu_owner->saved_fpu_context);

        /* don't prevent migration */

        /* don't prevent migration */

        CPU->fpu_owner->fpu_context_engaged = 0;

        CPU->fpu_owner->fpu_context_engaged = 0;

        spinlock_unlock(&CPU->fpu_owner->lock);

        spinlock_unlock(&CPU->fpu_owner->lock);

        CPU->fpu_owner = NULL;

        CPU->fpu_owner = NULL;

    }

    }

    spinlock_lock(&THREAD->lock);

    spinlock_lock(&THREAD->lock);

    if (THREAD->fpu_context_exists) {

    if (THREAD->fpu_context_exists) {

        fpu_context_restore(THREAD->saved_fpu_context);

        fpu_context_restore(THREAD->saved_fpu_context);

    } else {

    } else {

        /* Allocate FPU context */

        /* Allocate FPU context */

        if (!THREAD->saved_fpu_context) {

        if (!THREAD->saved_fpu_context) {

            /* Might sleep */

            /* Might sleep */

            spinlock_unlock(&THREAD->lock);

            spinlock_unlock(&THREAD->lock);

            spinlock_unlock(&CPU->lock);

            spinlock_unlock(&CPU->lock);

            THREAD->saved_fpu_context =

            THREAD->saved_fpu_context =

                slab_alloc(fpu_context_slab, 0);

                slab_alloc(fpu_context_slab, 0);

            /* We may have switched CPUs during slab_alloc */

            /* We may have switched CPUs during slab_alloc */

            goto restart;

            goto restart;

        }

        }

        fpu_init();

        fpu_init();

        THREAD->fpu_context_exists = 1;

        THREAD->fpu_context_exists = 1;

    }

    }

    CPU->fpu_owner = THREAD;

    CPU->fpu_owner = THREAD;

    THREAD->fpu_context_engaged = 1;

    THREAD->fpu_context_engaged = 1;

    spinlock_unlock(&THREAD->lock);

    spinlock_unlock(&THREAD->lock);

    spinlock_unlock(&CPU->lock);

    spinlock_unlock(&CPU->lock);

}

}

#endif

#endif

/** Initialize scheduler

/** Initialize scheduler

 *

 *

 * Initialize kernel scheduler.

 * Initialize kernel scheduler.

 *

 *

 */

 */

void scheduler_init(void)

void scheduler_init(void)

{

{

}

}

/** Get thread to be scheduled

/** Get thread to be scheduled

 *

 *

 * Get the optimal thread to be scheduled

 * Get the optimal thread to be scheduled

 * according to thread accounting and scheduler

 * according to thread accounting and scheduler

 * policy.

 * policy.

 *

 *

 * @return Thread to be scheduled.

 * @return Thread to be scheduled.

 *

 *

 */

 */

static thread_t *find_best_thread(void)

static thread_t *find_best_thread(void)

{

{

    thread_t *t;

    thread_t *t;

    runq_t *r;

    runq_t *r;

    int i;

    int i;

    ASSERT(CPU != NULL);

    ASSERT(CPU != NULL);

loop:

loop:

    interrupts_enable();

    interrupts_enable();

    if (atomic_get(&CPU->nrdy) == 0) {

    if (atomic_get(&CPU->nrdy) == 0) {

        /*

        /*

         * For there was nothing to run, the CPU goes to sleep

         * For there was nothing to run, the CPU goes to sleep

         * until a hardware interrupt or an IPI comes.

         * until a hardware interrupt or an IPI comes.

         * This improves energy saving and hyperthreading.

         * This improves energy saving and hyperthreading.

         */

         */

        /*

        /*

         * An interrupt might occur right now and wake up a thread.

         * An interrupt might occur right now and wake up a thread.

         * In such case, the CPU will continue to go to sleep

         * In such case, the CPU will continue to go to sleep

         * even though there is a runnable thread.

         * even though there is a runnable thread.

         */

         */

         cpu_sleep();

         cpu_sleep();

         goto loop;

         goto loop;

    }

    }

    interrupts_disable();

    interrupts_disable();

    for (i = 0; i<RQ_COUNT; i++) {

    for (i = 0; i<RQ_COUNT; i++) {

        r = &CPU->rq[i];

        r = &CPU->rq[i];

        spinlock_lock(&r->lock);

        spinlock_lock(&r->lock);

        if (r->n == 0) {

        if (r->n == 0) {

            /*

            /*

             * If this queue is empty, try a lower-priority queue.

             * If this queue is empty, try a lower-priority queue.

             */

             */

            spinlock_unlock(&r->lock);

            spinlock_unlock(&r->lock);

            continue;

            continue;

        }

        }

        atomic_dec(&CPU->nrdy);

        atomic_dec(&CPU->nrdy);

        atomic_dec(&nrdy);

        atomic_dec(&nrdy);

        r->n--;

        r->n--;

        /*

        /*

         * Take the first thread from the queue.

         * Take the first thread from the queue.

         */

         */

        t = list_get_instance(r->rq_head.next, thread_t, rq_link);

        t = list_get_instance(r->rq_head.next, thread_t, rq_link);

        list_remove(&t->rq_link);

        list_remove(&t->rq_link);

        spinlock_unlock(&r->lock);

        spinlock_unlock(&r->lock);

        spinlock_lock(&t->lock);

        spinlock_lock(&t->lock);

        t->cpu = CPU;

        t->cpu = CPU;

        t->ticks = us2ticks((i + 1) * 10000);

        t->ticks = us2ticks((i + 1) * 10000);

        t->priority = i;    /* correct rq index */

        t->priority = i;    /* correct rq index */

        /*

        /*

         * Clear the THREAD_FLAG_STOLEN flag so that t can be migrated

         * Clear the THREAD_FLAG_STOLEN flag so that t can be migrated

         * when load balancing needs emerge.

         * when load balancing needs emerge.

         */

         */

        t->flags &= ~THREAD_FLAG_STOLEN;

        t->flags &= ~THREAD_FLAG_STOLEN;

        spinlock_unlock(&t->lock);

        spinlock_unlock(&t->lock);

        return t;

        return t;

    }

    }

    goto loop;

    goto loop;

}

}

/** Prevent rq starvation

/** Prevent rq starvation

 *

 *

 * Prevent low priority threads from starving in rq's.

 * Prevent low priority threads from starving in rq's.

 *

 *

 * When the function decides to relink rq's, it reconnects

 * When the function decides to relink rq's, it reconnects

 * respective pointers so that in result threads with 'pri'

 * respective pointers so that in result threads with 'pri'

 * greater or equal start are moved to a higher-priority queue.

 * greater or equal start are moved to a higher-priority queue.

 *

 *

 * @param start Threshold priority.

 * @param start Threshold priority.

 *

 *

 */

 */

static void relink_rq(int start)

static void relink_rq(int start)

{

{

    link_t head;

    link_t head;

    runq_t *r;

    runq_t *r;

    int i, n;

    int i, n;

    list_initialize(&head);

    list_initialize(&head);

    spinlock_lock(&CPU->lock);

    spinlock_lock(&CPU->lock);

    if (CPU->needs_relink > NEEDS_RELINK_MAX) {

    if (CPU->needs_relink > NEEDS_RELINK_MAX) {

        for (i = start; i < RQ_COUNT - 1; i++) {

        for (i = start; i < RQ_COUNT - 1; i++) {

            /* remember and empty rq[i + 1] */

            /* remember and empty rq[i + 1] */

            r = &CPU->rq[i + 1];

            r = &CPU->rq[i + 1];

            spinlock_lock(&r->lock);

            spinlock_lock(&r->lock);

            list_concat(&head, &r->rq_head);

            list_concat(&head, &r->rq_head);

            n = r->n;

            n = r->n;

            r->n = 0;

            r->n = 0;

            spinlock_unlock(&r->lock);

            spinlock_unlock(&r->lock);

            /* append rq[i + 1] to rq[i] */

            /* append rq[i + 1] to rq[i] */

            r = &CPU->rq[i];

            r = &CPU->rq[i];

            spinlock_lock(&r->lock);

            spinlock_lock(&r->lock);

            list_concat(&r->rq_head, &head);

            list_concat(&r->rq_head, &head);

            r->n += n;

            r->n += n;

            spinlock_unlock(&r->lock);

            spinlock_unlock(&r->lock);

        }

        }

        CPU->needs_relink = 0;

        CPU->needs_relink = 0;

    }

    }

    spinlock_unlock(&CPU->lock);

    spinlock_unlock(&CPU->lock);

}

}

/** The scheduler

/** The scheduler

 *

 *

 * The thread scheduling procedure.

 * The thread scheduling procedure.

 * Passes control directly to

 * Passes control directly to

 * scheduler_separated_stack().

 * scheduler_separated_stack().

 *

 *

 */

 */

void scheduler(void)

void scheduler(void)

{

{

    volatile ipl_t ipl;

    volatile ipl_t ipl;

    ASSERT(CPU != NULL);

    ASSERT(CPU != NULL);

    ipl = interrupts_disable();

    ipl = interrupts_disable();

    if (atomic_get(&haltstate))

    if (atomic_get(&haltstate))

        halt();

        halt();

    if (THREAD) {

    if (THREAD) {

        spinlock_lock(&THREAD->lock);

        spinlock_lock(&THREAD->lock);

        /* Update thread accounting */

        /* Update thread accounting */

        THREAD->cycles += get_cycle() - THREAD->last_cycle;

        THREAD->cycles += get_cycle() - THREAD->last_cycle;

#ifndef CONFIG_FPU_LAZY

#ifndef CONFIG_FPU_LAZY

        fpu_context_save(THREAD->saved_fpu_context);

        fpu_context_save(THREAD->saved_fpu_context);

#endif

#endif

        if (!context_save(&THREAD->saved_context)) {

        if (!context_save(&THREAD->saved_context)) {

            /*

            /*

             * This is the place where threads leave scheduler();

             * This is the place where threads leave scheduler();

             */

             */

            /* Save current CPU cycle */

            /* Save current CPU cycle */

            THREAD->last_cycle = get_cycle();

            THREAD->last_cycle = get_cycle();

            spinlock_unlock(&THREAD->lock);

            spinlock_unlock(&THREAD->lock);

            interrupts_restore(THREAD->saved_context.ipl);

            interrupts_restore(THREAD->saved_context.ipl);

            return;

            return;

        }

        }

        /*

        /*

         * Interrupt priority level of preempted thread is recorded

         * Interrupt priority level of preempted thread is recorded

         * here to facilitate scheduler() invocations from

         * here to facilitate scheduler() invocations from

         * interrupts_disable()'d code (e.g. waitq_sleep_timeout()).

         * interrupts_disable()'d code (e.g. waitq_sleep_timeout()).

         */

         */

        THREAD->saved_context.ipl = ipl;

        THREAD->saved_context.ipl = ipl;

    }

    }

    /*

    /*

     * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM

     * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM

     * and preemption counter. At this point THE could be coming either

     * and preemption counter. At this point THE could be coming either

     * from THREAD's or CPU's stack.

     * from THREAD's or CPU's stack.

     */

     */

    the_copy(THE, (the_t *) CPU->stack);

    the_copy(THE, (the_t *) CPU->stack);

    /*

    /*

     * We may not keep the old stack.

     * We may not keep the old stack.

     * Reason: If we kept the old stack and got blocked, for instance, in

     * Reason: If we kept the old stack and got blocked, for instance, in

     * find_best_thread(), the old thread could get rescheduled by another

     * find_best_thread(), the old thread could get rescheduled by another

     * CPU and overwrite the part of its own stack that was also used by

     * CPU and overwrite the part of its own stack that was also used by

     * the scheduler on this CPU.

     * the scheduler on this CPU.

     *

     *

     * Moreover, we have to bypass the compiler-generated POP sequence

     * Moreover, we have to bypass the compiler-generated POP sequence

     * which is fooled by SP being set to the very top of the stack.

     * which is fooled by SP being set to the very top of the stack.

     * Therefore the scheduler() function continues in

     * Therefore the scheduler() function continues in

     * scheduler_separated_stack().

     * scheduler_separated_stack().

     */

     */

    context_save(&CPU->saved_context);

    context_save(&CPU->saved_context);

    context_set(&CPU->saved_context, FADDR(scheduler_separated_stack),

    context_set(&CPU->saved_context, FADDR(scheduler_separated_stack),

        (uintptr_t) CPU->stack, CPU_STACK_SIZE);

        (uintptr_t) CPU->stack, CPU_STACK_SIZE);

    context_restore(&CPU->saved_context);

    context_restore(&CPU->saved_context);

    /* not reached */

    /* not reached */

}

}

/** Scheduler stack switch wrapper

/** Scheduler stack switch wrapper

 *

 *

 * Second part of the scheduler() function

 * Second part of the scheduler() function

 * using new stack. Handling the actual context

 * using new stack. Handling the actual context

 * switch to a new thread.

 * switch to a new thread.

 *

 *

 * Assume THREAD->lock is held.

 * Assume THREAD->lock is held.

 */

 */

void scheduler_separated_stack(void)

void scheduler_separated_stack(void)

{

{

    int priority;

    int priority;

    ASSERT(CPU != NULL);

    ASSERT(CPU != NULL);

    if (THREAD) {

    if (THREAD) {

        /* must be run after the switch to scheduler stack */

        /* must be run after the switch to scheduler stack */

        after_thread_ran();

        after_thread_ran();

        switch (THREAD->state) {

        switch (THREAD->state) {

        case Running:

        case Running:

            spinlock_unlock(&THREAD->lock);

            spinlock_unlock(&THREAD->lock);

            thread_ready(THREAD);

            thread_ready(THREAD);

            break;

            break;

        case Exiting:

        case Exiting:

repeat:

repeat:

            if (THREAD->detached) {

            if (THREAD->detached) {

                thread_destroy(THREAD);

                thread_destroy(THREAD);

            } else {

            } else {

                /*

                /*

                 * The thread structure is kept allocated until

                 * The thread structure is kept allocated until

                 * somebody calls thread_detach() on it.

                 * somebody calls thread_detach() on it.

                 */

                 */

                if (!spinlock_trylock(&THREAD->join_wq.lock)) {

                if (!spinlock_trylock(&THREAD->join_wq.lock)) {

                    /*

                    /*

                     * Avoid deadlock.

                     * Avoid deadlock.

                     */

                     */

                    spinlock_unlock(&THREAD->lock);

                    spinlock_unlock(&THREAD->lock);

                    delay(10);

                    delay(10);

                    spinlock_lock(&THREAD->lock);

                    spinlock_lock(&THREAD->lock);

                    goto repeat;

                    goto repeat;

                }

                }

                _waitq_wakeup_unsafe(&THREAD->join_wq, false);

                _waitq_wakeup_unsafe(&THREAD->join_wq, false);

                spinlock_unlock(&THREAD->join_wq.lock);

                spinlock_unlock(&THREAD->join_wq.lock);

                THREAD->state = Undead;

                THREAD->state = Undead;

                spinlock_unlock(&THREAD->lock);

                spinlock_unlock(&THREAD->lock);

            }

            }

            break;

            break;

        case Sleeping:

        case Sleeping:

            /*

            /*

             * Prefer the thread after it's woken up.

             * Prefer the thread after it's woken up.

             */

             */

            THREAD->priority = -1;

            THREAD->priority = -1;

            /*

            /*

             * We need to release wq->lock which we locked in

             * We need to release wq->lock which we locked in

             * waitq_sleep(). Address of wq->lock is kept in

             * waitq_sleep(). Address of wq->lock is kept in

             * THREAD->sleep_queue.

             * THREAD->sleep_queue.

             */

             */

            spinlock_unlock(&THREAD->sleep_queue->lock);

            spinlock_unlock(&THREAD->sleep_queue->lock);

            /*

            /*

             * Check for possible requests for out-of-context

             * Check for possible requests for out-of-context

             * invocation.

             * invocation.

             */

             */

            if (THREAD->call_me) {

            if (THREAD->call_me) {

                THREAD->call_me(THREAD->call_me_with);

                THREAD->call_me(THREAD->call_me_with);

                THREAD->call_me = NULL;

                THREAD->call_me = NULL;

                THREAD->call_me_with = NULL;

                THREAD->call_me_with = NULL;

            }

            }

            spinlock_unlock(&THREAD->lock);

            spinlock_unlock(&THREAD->lock);

            break;

            break;

        default:

        default:

            /*

            /*

             * Entering state is unexpected.

             * Entering state is unexpected.

             */

             */

            panic("tid%d: unexpected state %s\n", THREAD->tid,

            panic("tid%d: unexpected state %s\n", THREAD->tid,

                thread_states[THREAD->state]);

                thread_states[THREAD->state]);

            break;

            break;

        }

        }

        THREAD = NULL;

        THREAD = NULL;

    }

    }

    THREAD = find_best_thread();

    THREAD = find_best_thread();

    spinlock_lock(&THREAD->lock);

    spinlock_lock(&THREAD->lock);

    priority = THREAD->priority;

    priority = THREAD->priority;

    spinlock_unlock(&THREAD->lock);

    spinlock_unlock(&THREAD->lock);

    relink_rq(priority);       

    relink_rq(priority);       

    /*

    /*

     * If both the old and the new task are the same, lots of work is

     * If both the old and the new task are the same, lots of work is

     * avoided.

     * avoided.

     */

     */

    if (TASK != THREAD->task) {

    if (TASK != THREAD->task) {

        as_t *as1 = NULL;

        as_t *as1 = NULL;

        as_t *as2;

        as_t *as2;

        if (TASK) {

        if (TASK) {

            spinlock_lock(&TASK->lock);

            spinlock_lock(&TASK->lock);

            as1 = TASK->as;

            as1 = TASK->as;

            spinlock_unlock(&TASK->lock);

            spinlock_unlock(&TASK->lock);

        }

        }

        spinlock_lock(&THREAD->task->lock);

        spinlock_lock(&THREAD->task->lock);

        as2 = THREAD->task->as;

        as2 = THREAD->task->as;

        spinlock_unlock(&THREAD->task->lock);

        spinlock_unlock(&THREAD->task->lock);

        /*

        /*

         * Note that it is possible for two tasks to share one address

         * Note that it is possible for two tasks to share one address

         * space.

         * space.

         */

         */

        if (as1 != as2) {

        if (as1 != as2) {

            /*

            /*

             * Both tasks and address spaces are different.

             * Both tasks and address spaces are different.

             * Replace the old one with the new one.

             * Replace the old one with the new one.

             */

             */

            as_switch(as1, as2);

            as_switch(as1, as2);

        }

        }

        TASK = THREAD->task;

        TASK = THREAD->task;

        before_task_runs();

        before_task_runs();

    }

    }

    spinlock_lock(&THREAD->lock);  

    spinlock_lock(&THREAD->lock);  

    THREAD->state = Running;

    THREAD->state = Running;

#ifdef SCHEDULER_VERBOSE

#ifdef SCHEDULER_VERBOSE

    printf("cpu%d: tid %d (priority=%d, ticks=%lld, nrdy=%ld)\n",

    printf("cpu%d: tid %d (priority=%d, ticks=%lld, nrdy=%ld)\n",

        CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks,

        CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks,

        atomic_get(&CPU->nrdy));

        atomic_get(&CPU->nrdy));

#endif  

#endif  

    /*

    /*

     * Some architectures provide late kernel PA2KA(identity)

     * Some architectures provide late kernel PA2KA(identity)

     * mapping in a page fault handler. However, the page fault

     * mapping in a page fault handler. However, the page fault

     * handler uses the kernel stack of the running thread and

     * handler uses the kernel stack of the running thread and

     * therefore cannot be used to map it. The kernel stack, if

     * therefore cannot be used to map it. The kernel stack, if

     * necessary, is to be mapped in before_thread_runs(). This

     * necessary, is to be mapped in before_thread_runs(). This

     * function must be executed before the switch to the new stack.

     * function must be executed before the switch to the new stack.

     */

     */

    before_thread_runs();

    before_thread_runs();

    /*

    /*

     * Copy the knowledge of CPU, TASK, THREAD and preemption counter to

     * Copy the knowledge of CPU, TASK, THREAD and preemption counter to

     * thread's stack.

     * thread's stack.

     */

     */

    the_copy(THE, (the_t *) THREAD->kstack);

    the_copy(THE, (the_t *) THREAD->kstack);

    context_restore(&THREAD->saved_context);

    context_restore(&THREAD->saved_context);

    /* not reached */

    /* not reached */

}

}

#ifdef CONFIG_SMP

#ifdef CONFIG_SMP

/** Load balancing thread

/** Load balancing thread

 *

 *

 * SMP load balancing thread, supervising thread supplies

 * SMP load balancing thread, supervising thread supplies

 * for the CPU it's wired to.

 * for the CPU it's wired to.

 *

 *

 * @param arg Generic thread argument (unused).

 * @param arg Generic thread argument (unused).

 *

 *

 */

 */

void kcpulb(void *arg)

void kcpulb(void *arg)

{

{

    thread_t *t;

    thread_t *t;

    int count, average, i, j, k = 0;

    int count, average, i, j, k = 0;

    ipl_t ipl;

    ipl_t ipl;

    /*

    /*

     * Detach kcpulb as nobody will call thread_join_timeout() on it.

     * Detach kcpulb as nobody will call thread_join_timeout() on it.

     */

     */

    thread_detach(THREAD);

    thread_detach(THREAD);

loop:

loop:

    /*

    /*

     * Work in 1s intervals.

     * Work in 1s intervals.

     */

     */

    thread_sleep(1);

    thread_sleep(1);

not_satisfied:

not_satisfied:

    /*

    /*

     * Calculate the number of threads that will be migrated/stolen from

     * Calculate the number of threads that will be migrated/stolen from

     * other CPU's. Note that situation can have changed between two

     * other CPU's. Note that situation can have changed between two

     * passes. Each time get the most up to date counts.

     * passes. Each time get the most up to date counts.

     */

     */

    average = atomic_get(&nrdy) / config.cpu_active + 1;

    average = atomic_get(&nrdy) / config.cpu_active + 1;

    count = average - atomic_get(&CPU->nrdy);

    count = average - atomic_get(&CPU->nrdy);

    if (count <= 0)

    if (count <= 0)

        goto satisfied;

        goto satisfied;

    /*

    /*

     * Searching least priority queues on all CPU's first and most priority

     * Searching least priority queues on all CPU's first and most priority

     * queues on all CPU's last.

     * queues on all CPU's last.

     */

     */

    for (j= RQ_COUNT - 1; j >= 0; j--) {

    for (j= RQ_COUNT - 1; j >= 0; j--) {

        for (i = 0; i < config.cpu_active; i++) {

        for (i = 0; i < config.cpu_active; i++) {

            link_t *l;

            link_t *l;

            runq_t *r;

            runq_t *r;

            cpu_t *cpu;

            cpu_t *cpu;

            cpu = &cpus[(i + k) % config.cpu_active];

            cpu = &cpus[(i + k) % config.cpu_active];

            /*

            /*

             * Not interested in ourselves.

             * Not interested in ourselves.

             * Doesn't require interrupt disabling for kcpulb has

             * Doesn't require interrupt disabling for kcpulb has

             * THREAD_FLAG_WIRED.

             * THREAD_FLAG_WIRED.

             */

             */

            if (CPU == cpu)

            if (CPU == cpu)

                continue;

                continue;

            if (atomic_get(&cpu->nrdy) <= average)

            if (atomic_get(&cpu->nrdy) <= average)

                continue;

                continue;

            ipl = interrupts_disable();

            ipl = interrupts_disable();

            r = &cpu->rq[j];

            r = &cpu->rq[j];

            spinlock_lock(&r->lock);

            spinlock_lock(&r->lock);

            if (r->n == 0) {

            if (r->n == 0) {

                spinlock_unlock(&r->lock);

                spinlock_unlock(&r->lock);

                interrupts_restore(ipl);

                interrupts_restore(ipl);

                continue;

                continue;

            }

            }

            t = NULL;

            t = NULL;

            l = r->rq_head.prev;    /* search rq from the back */

            l = r->rq_head.prev;    /* search rq from the back */

            while (l != &r->rq_head) {

            while (l != &r->rq_head) {

                t = list_get_instance(l, thread_t, rq_link);

                t = list_get_instance(l, thread_t, rq_link);

                /*

                /*

                 * We don't want to steal CPU-wired threads

                 * We don't want to steal CPU-wired threads

                 * neither threads already stolen. The latter

                 * neither threads already stolen. The latter

                 * prevents threads from migrating between CPU's

                 * prevents threads from migrating between CPU's

                 * without ever being run. We don't want to

                 * without ever being run. We don't want to

                 * steal threads whose FPU context is still in

                 * steal threads whose FPU context is still in

                 * CPU.

                 * CPU.

                 */

                 */

                spinlock_lock(&t->lock);

                spinlock_lock(&t->lock);

                if ((!(t->flags & (THREAD_FLAG_WIRED |

                if ((!(t->flags & (THREAD_FLAG_WIRED |

                    THREAD_FLAG_STOLEN))) &&

                    THREAD_FLAG_STOLEN))) &&

                    (!(t->fpu_context_engaged)) ) {

                    (!(t->fpu_context_engaged)) ) {

                    /*

                    /*

                     * Remove t from r.

                     * Remove t from r.

                     */

                     */

                    spinlock_unlock(&t->lock);

                    spinlock_unlock(&t->lock);

                    atomic_dec(&cpu->nrdy);

                    atomic_dec(&cpu->nrdy);

                    atomic_dec(&nrdy);

                    atomic_dec(&nrdy);

                    r->n--;

                    r->n--;

                    list_remove(&t->rq_link);

                    list_remove(&t->rq_link);

                    break;

                    break;

                }

                }

                spinlock_unlock(&t->lock);

                spinlock_unlock(&t->lock);

                l = l->prev;

                l = l->prev;

                t = NULL;

                t = NULL;

            }

            }

            spinlock_unlock(&r->lock);

            spinlock_unlock(&r->lock);

            if (t) {

            if (t) {

                /*

                /*

                 * Ready t on local CPU

                 * Ready t on local CPU

                 */

                 */

                spinlock_lock(&t->lock);

                spinlock_lock(&t->lock);

#ifdef KCPULB_VERBOSE

#ifdef KCPULB_VERBOSE

                printf("kcpulb%d: TID %d -> cpu%d, nrdy=%ld, "

                printf("kcpulb%d: TID %d -> cpu%d, nrdy=%ld, "

                    "avg=%nd\n", CPU->id, t->tid, CPU->id,

                    "avg=%nd\n", CPU->id, t->tid, CPU->id,

                    atomic_get(&CPU->nrdy),

                    atomic_get(&CPU->nrdy),

                    atomic_get(&nrdy) / config.cpu_active);

                    atomic_get(&nrdy) / config.cpu_active);

#endif

#endif

                t->flags |= THREAD_FLAG_STOLEN;

                t->flags |= THREAD_FLAG_STOLEN;

                t->state = Entering;

                t->state = Entering;

                spinlock_unlock(&t->lock);

                spinlock_unlock(&t->lock);

                thread_ready(t);

                thread_ready(t);

                interrupts_restore(ipl);

                interrupts_restore(ipl);

                if (--count == 0)

                if (--count == 0)

                    goto satisfied;

                    goto satisfied;

                /*

                /*

                 * We are not satisfied yet, focus on another

                 * We are not satisfied yet, focus on another

                 * CPU next time.

                 * CPU next time.

                 */

                 */

                k++;

                k++;

                continue;

                continue;

            }

            }

            interrupts_restore(ipl);

            interrupts_restore(ipl);

        }

        }

    }

    }

    if (atomic_get(&CPU->nrdy)) {

    if (atomic_get(&CPU->nrdy)) {

        /*

        /*

         * Be a little bit light-weight and let migrated threads run.

         * Be a little bit light-weight and let migrated threads run.

         */

         */

        scheduler();

        scheduler();

    } else {

    } else {

        /*

        /*

         * We failed to migrate a single thread.

         * We failed to migrate a single thread.

         * Give up this turn.

         * Give up this turn.

         */

         */

        goto loop;

        goto loop;

    }

    }

    goto not_satisfied;

    goto not_satisfied;

satisfied:

satisfied:

    goto loop;

    goto loop;

}

}

#endif /* CONFIG_SMP */

#endif /* CONFIG_SMP */

/** Print information about threads & scheduler queues */

/** Print information about threads & scheduler queues */

void sched_print_list(void)

void sched_print_list(void)

{

{

    ipl_t ipl;

    ipl_t ipl;

    int cpu,i;

    int cpu,i;

    runq_t *r;

    runq_t *r;

    thread_t *t;

    thread_t *t;

    link_t *cur;

    link_t *cur;

    /* We are going to mess with scheduler structures,

    /* We are going to mess with scheduler structures,

     * let's not be interrupted */

     * let's not be interrupted */

    ipl = interrupts_disable();

    ipl = interrupts_disable();

    for (cpu=0;cpu < config.cpu_count; cpu++) {

    for (cpu=0;cpu < config.cpu_count; cpu++) {

        if (!cpus[cpu].active)

        if (!cpus[cpu].active)