HelenOS sources

root/uspace/lib/c/generic/thread/fibril.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. _ready_debug_check
  2. _ready_up
  3. _ready_down
  4. _fibril_main
  5. fibril_alloc
  6. fibril_setup
  7. fibril_teardown
  8. _fibril_trigger_internal
  9. _ipc_wait
  10. _ready_list_pop
  11. _ready_list_pop_nonblocking
  12. _ready_list_push
  13. _wait_ipc
  14. _handle_expired_timeouts
  15. _fibril_cleanup_dead
  16. _fibril_switch_to
  17. _helper_fibril_fn
  18. fibril_create_generic
  19. fibril_create
  20. fibril_destroy
  21. _insert_timeout
  22. fibril_wait_timeout
  23. fibril_wait_for
  24. fibril_notify
  25. fibril_start
  26. fibril_add_ready
  27. fibril_self
  28. fibril_get_id
  29. fibril_yield
  30. _runner_fn
  31. fibril_test_spawn_runners
  32. fibril_enable_multithreaded
  33. fibril_detach
  34. fibril_exit
  35. __fibrils_init
  36. __fibrils_fini
  37. fibril_usleep
  38. fibril_sleep
  39. fibril_ipc_poke
  40. fibril_ipc_wait

/*
 * Copyright (c) 2006 Ondrej Palkovsky
 * Copyright (c) 2007 Jakub Jermar
 * Copyright (c) 2018 CZ.NIC, z.s.p.o.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * - Redistributions of source code must retain the above copyright
 *   notice, this list of conditions and the following disclaimer.
 * - Redistributions in binary form must reproduce the above copyright
 *   notice, this list of conditions and the following disclaimer in the
 *   documentation and/or other materials provided with the distribution.
 * - The name of the author may not be used to endorse or promote products
 *   derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

/** @addtogroup libc
 * @{
 */
/** @file
 */

#include <adt/list.h>
#include <fibril.h>
#include <stack.h>
#include <tls.h>
#include <stdlib.h>
#include <as.h>
#include <context.h>
#include <assert.h>

#include <mem.h>
#include <str.h>
#include <ipc/ipc.h>
#include <libarch/faddr.h>

#include "../private/thread.h"
#include "../private/futex.h"
#include "../private/fibril.h"
#include "../private/libc.h"

#define DPRINTF(...) ((void)0)
#undef READY_DEBUG

/** Member of timeout_list. */
typedef struct {
        link_t link;
        struct timespec expires;
        fibril_event_t *event;
} _timeout_t;

typedef struct {
        errno_t rc;
        link_t link;
        ipc_call_t *call;
        fibril_event_t event;
} _ipc_waiter_t;

typedef struct {
        errno_t rc;
        link_t link;
        ipc_call_t call;
} _ipc_buffer_t;

typedef enum {
        SWITCH_FROM_DEAD,
        SWITCH_FROM_HELPER,
        SWITCH_FROM_YIELD,
        SWITCH_FROM_BLOCKED,
} _switch_type_t;

static bool multithreaded = false;

/* This futex serializes access to global data. */
static futex_t fibril_futex;
static futex_t ready_semaphore;
static long ready_st_count;

static LIST_INITIALIZE(ready_list);
static LIST_INITIALIZE(fibril_list);
static LIST_INITIALIZE(timeout_list);

static futex_t ipc_lists_futex;
static LIST_INITIALIZE(ipc_waiter_list);
static LIST_INITIALIZE(ipc_buffer_list);
static LIST_INITIALIZE(ipc_buffer_free_list);

/* Only used as unique markers for triggered events. */
static fibril_t _fibril_event_triggered;
static fibril_t _fibril_event_timed_out;
#define _EVENT_INITIAL   (NULL)
#define _EVENT_TRIGGERED (&_fibril_event_triggered)
#define _EVENT_TIMED_OUT (&_fibril_event_timed_out)

static inline void _ready_debug_check(void)
{
#ifdef READY_DEBUG
        assert(!multithreaded);
        long count = (long) list_count(&ready_list) +
            (long) list_count(&ipc_buffer_free_list);
        assert(ready_st_count == count);
#endif
}

static inline void _ready_up(void)
{
        if (multithreaded) {
                futex_up(&ready_semaphore);
        } else {
                ready_st_count++;
                _ready_debug_check();
        }
}

static inline errno_t _ready_down(const struct timespec *expires)
{
        if (multithreaded)
                return futex_down_timeout(&ready_semaphore, expires);

        _ready_debug_check();
        ready_st_count--;
        return EOK;
}

static atomic_int threads_in_ipc_wait;

/** Function that spans the whole life-cycle of a fibril.
 *
 * Each fibril begins execution in this function. Then the function implementing
 * the fibril logic is called.  After its return, the return value is saved.
 * The fibril then switches to another fibril, which cleans up after it.
 *
 */
static void _fibril_main(void)
{
        /* fibril_futex is locked when a fibril is started. */
        futex_unlock(&fibril_futex);

        fibril_t *fibril = fibril_self();

        /* Call the implementing function. */
        fibril_exit(fibril->func(fibril->arg));

        /* Not reached */
}

/** Allocate a fibril structure and TCB, but don't do anything else with it. */
fibril_t *fibril_alloc(void)
{
        tcb_t *tcb = tls_make(__progsymbols.elfstart);
        if (!tcb)
                return NULL;

        fibril_t *fibril = calloc(1, sizeof(fibril_t));
        if (!fibril) {
                tls_free(tcb);
                return NULL;
        }

        tcb->fibril_data = fibril;
        fibril->tcb = tcb;
        fibril->is_freeable = true;

        fibril_setup(fibril);
        return fibril;
}

/**
 * Put the fibril into fibril_list.
 */
void fibril_setup(fibril_t *f)
{
        futex_lock(&fibril_futex);
        list_append(&f->all_link, &fibril_list);
        futex_unlock(&fibril_futex);
}

void fibril_teardown(fibril_t *fibril)
{
        futex_lock(&fibril_futex);
        list_remove(&fibril->all_link);
        futex_unlock(&fibril_futex);

        if (fibril->is_freeable) {
                tls_free(fibril->tcb);
                free(fibril);
        }
}

/**
 * Event notification with a given reason.
 *
 * @param reason  Reason of the notification.
 *                Can be either _EVENT_TRIGGERED or _EVENT_TIMED_OUT.
 */
static fibril_t *_fibril_trigger_internal(fibril_event_t *event, fibril_t *reason)
{
        assert(reason != _EVENT_INITIAL);
        assert(reason == _EVENT_TIMED_OUT || reason == _EVENT_TRIGGERED);

        futex_assert_is_locked(&fibril_futex);

        if (event->fibril == _EVENT_INITIAL) {
                event->fibril = reason;
                return NULL;
        }

        if (event->fibril == _EVENT_TIMED_OUT) {
                assert(reason == _EVENT_TRIGGERED);
                event->fibril = reason;
                return NULL;
        }

        if (event->fibril == _EVENT_TRIGGERED) {
                /* Already triggered. Nothing to do. */
                return NULL;
        }

        fibril_t *f = event->fibril;
        event->fibril = reason;

        assert(f->sleep_event == event);
        return f;
}

static errno_t _ipc_wait(ipc_call_t *call, const struct timespec *expires)
{
        if (!expires)
                return ipc_wait(call, SYNCH_NO_TIMEOUT, SYNCH_FLAGS_NONE);

        if (expires->tv_sec == 0)
                return ipc_wait(call, SYNCH_NO_TIMEOUT, SYNCH_FLAGS_NON_BLOCKING);

        struct timespec now;
        getuptime(&now);

        if (ts_gteq(&now, expires))
                return ipc_wait(call, SYNCH_NO_TIMEOUT, SYNCH_FLAGS_NON_BLOCKING);

        return ipc_wait(call, NSEC2USEC(ts_sub_diff(expires, &now)),
            SYNCH_FLAGS_NONE);
}

/*
 * Waits until a ready fibril is added to the list, or an IPC message arrives.
 * Returns NULL on timeout and may also return NULL if returning from IPC
 * wait after new ready fibrils are added.
 */
static fibril_t *_ready_list_pop(const struct timespec *expires, bool locked)
{
        if (locked) {
                futex_assert_is_locked(&fibril_futex);
                assert(expires);
                /* Must be nonblocking. */
                assert(expires->tv_sec == 0);
        } else {
                futex_assert_is_not_locked(&fibril_futex);
        }

        errno_t rc = _ready_down(expires);
        if (rc != EOK)
                return NULL;

        /*
         * Once we acquire a token from ready_semaphore, there are two options.
         * Either there is a ready fibril in the list, or it's our turn to
         * call `ipc_wait_cycle()`. There is one extra token on the semaphore
         * for each entry of the call buffer.
         */

        if (!locked)
                futex_lock(&fibril_futex);
        fibril_t *f = list_pop(&ready_list, fibril_t, link);
        if (!f)
                atomic_fetch_add_explicit(&threads_in_ipc_wait, 1,
                    memory_order_relaxed);
        if (!locked)
                futex_unlock(&fibril_futex);

        if (f)
                return f;

        if (!multithreaded)
                assert(list_empty(&ipc_buffer_list));

        /* No fibril is ready, IPC wait it is. */
        ipc_call_t call = { 0 };
        rc = _ipc_wait(&call, expires);

        atomic_fetch_sub_explicit(&threads_in_ipc_wait, 1,
            memory_order_relaxed);

        if (rc != EOK && rc != ENOENT) {
                /* Return token. */
                _ready_up();
                return NULL;
        }

        /*
         * We might get ENOENT due to a poke.
         * In that case, we propagate the null call out of fibril_ipc_wait(),
         * because poke must result in that call returning.
         */

        /*
         * If a fibril is already waiting for IPC, we wake up the fibril,
         * and return the token to ready_semaphore.
         * If there is no fibril waiting, we pop a buffer bucket and
         * put our call there. The token then returns when the bucket is
         * returned.
         */

        if (!locked)
                futex_lock(&fibril_futex);

        futex_lock(&ipc_lists_futex);

        _ipc_waiter_t *w = list_pop(&ipc_waiter_list, _ipc_waiter_t, link);
        if (w) {
                *w->call = call;
                w->rc = rc;
                /* We switch to the woken up fibril immediately if possible. */
                f = _fibril_trigger_internal(&w->event, _EVENT_TRIGGERED);

                /* Return token. */
                _ready_up();
        } else {
                _ipc_buffer_t *buf = list_pop(&ipc_buffer_free_list, _ipc_buffer_t, link);
                assert(buf);
                *buf = (_ipc_buffer_t) { .call = call, .rc = rc };
                list_append(&buf->link, &ipc_buffer_list);
        }

        futex_unlock(&ipc_lists_futex);

        if (!locked)
                futex_unlock(&fibril_futex);

        return f;
}

static fibril_t *_ready_list_pop_nonblocking(bool locked)
{
        struct timespec tv = { .tv_sec = 0, .tv_nsec = 0 };
        return _ready_list_pop(&tv, locked);
}

static void _ready_list_push(fibril_t *f)
{
        if (!f)
                return;

        futex_assert_is_locked(&fibril_futex);

        /* Enqueue in ready_list. */
        list_append(&f->link, &ready_list);
        _ready_up();

        if (atomic_load_explicit(&threads_in_ipc_wait, memory_order_relaxed)) {
                DPRINTF("Poking.\n");
                /* Wakeup one thread sleeping in SYS_IPC_WAIT. */
                ipc_poke();
        }
}

/* Blocks the current fibril until an IPC call arrives. */
static errno_t _wait_ipc(ipc_call_t *call, const struct timespec *expires)
{
        futex_assert_is_not_locked(&fibril_futex);

        futex_lock(&ipc_lists_futex);
        _ipc_buffer_t *buf = list_pop(&ipc_buffer_list, _ipc_buffer_t, link);
        if (buf) {
                *call = buf->call;
                errno_t rc = buf->rc;

                /* Return to freelist. */
                list_append(&buf->link, &ipc_buffer_free_list);
                /* Return IPC wait token. */
                _ready_up();

                futex_unlock(&ipc_lists_futex);
                return rc;
        }

        _ipc_waiter_t w = { .call = call };
        list_append(&w.link, &ipc_waiter_list);
        futex_unlock(&ipc_lists_futex);

        errno_t rc = fibril_wait_timeout(&w.event, expires);
        if (rc == EOK)
                return w.rc;

        futex_lock(&ipc_lists_futex);
        if (link_in_use(&w.link))
                list_remove(&w.link);
        else
                rc = w.rc;
        futex_unlock(&ipc_lists_futex);
        return rc;
}

/** Fire all timeouts that expired. */
static struct timespec *_handle_expired_timeouts(struct timespec *next_timeout)
{
        struct timespec ts;
        getuptime(&ts);

        futex_lock(&fibril_futex);

        while (!list_empty(&timeout_list)) {
                link_t *cur = list_first(&timeout_list);
                _timeout_t *to = list_get_instance(cur, _timeout_t, link);

                if (ts_gt(&to->expires, &ts)) {
                        *next_timeout = to->expires;
                        futex_unlock(&fibril_futex);
                        return next_timeout;
                }

                list_remove(&to->link);

                _ready_list_push(_fibril_trigger_internal(
                    to->event, _EVENT_TIMED_OUT));
        }

        futex_unlock(&fibril_futex);
        return NULL;
}

/**
 * Clean up after a dead fibril from which we restored context, if any.
 * Called after a switch is made and fibril_futex is unlocked.
 */
static void _fibril_cleanup_dead(void)
{
        fibril_t *srcf = fibril_self();
        if (!srcf->clean_after_me)
                return;

        void *stack = srcf->clean_after_me->stack;
        assert(stack);
        as_area_destroy(stack);
        fibril_teardown(srcf->clean_after_me);
        srcf->clean_after_me = NULL;
}

/** Switch to a fibril. */
static void _fibril_switch_to(_switch_type_t type, fibril_t *dstf, bool locked)
{
        assert(fibril_self()->rmutex_locks == 0);

        if (!locked)
                futex_lock(&fibril_futex);
        else
                futex_assert_is_locked(&fibril_futex);

        fibril_t *srcf = fibril_self();
        assert(srcf);
        assert(dstf);

        switch (type) {
        case SWITCH_FROM_YIELD:
                _ready_list_push(srcf);
                break;
        case SWITCH_FROM_DEAD:
                dstf->clean_after_me = srcf;
                break;
        case SWITCH_FROM_HELPER:
        case SWITCH_FROM_BLOCKED:
                break;
        }

        dstf->thread_ctx = srcf->thread_ctx;
        srcf->thread_ctx = NULL;

        /* Just some bookkeeping to allow better debugging of futex locks. */
        futex_give_to(&fibril_futex, dstf);

        /* Swap to the next fibril. */
        context_swap(&srcf->ctx, &dstf->ctx);

        assert(srcf == fibril_self());
        assert(srcf->thread_ctx);

        if (!locked) {
                /* Must be after context_swap()! */
                futex_unlock(&fibril_futex);
                _fibril_cleanup_dead();
        }
}

/**
 * Main function for a helper fibril.
 * The helper fibril executes on threads in the lightweight fibril pool when
 * there is no fibril ready to run. Its only purpose is to block until
 * another fibril is ready, or a timeout expires, or an IPC message arrives.
 *
 * There is at most one helper fibril per thread.
 *
 */
static errno_t _helper_fibril_fn(void *arg)
{
        /* Set itself as the thread's own context. */
        fibril_self()->thread_ctx = fibril_self();

        (void) arg;

        struct timespec next_timeout;
        while (true) {
                struct timespec *to = _handle_expired_timeouts(&next_timeout);
                fibril_t *f = _ready_list_pop(to, false);
                if (f) {
                        _fibril_switch_to(SWITCH_FROM_HELPER, f, false);
                }
        }

        return EOK;
}

/** Create a new fibril.
 *
 * @param func Implementing function of the new fibril.
 * @param arg Argument to pass to func.
 * @param stksz Stack size in bytes.
 *
 * @return 0 on failure or TLS of the new fibril.
 *
 */
fid_t fibril_create_generic(errno_t (*func)(void *), void *arg, size_t stksz)
{
        fibril_t *fibril;

        fibril = fibril_alloc();
        if (fibril == NULL)
                return 0;

        fibril->stack_size = stksz;
        fibril->stack = as_area_create(AS_AREA_ANY, fibril->stack_size,
            AS_AREA_READ | AS_AREA_WRITE | AS_AREA_CACHEABLE | AS_AREA_GUARD |
            AS_AREA_LATE_RESERVE, AS_AREA_UNPAGED);
        if (fibril->stack == AS_MAP_FAILED) {
                fibril_teardown(fibril);
                return 0;
        }

        fibril->func = func;
        fibril->arg = arg;

        context_create_t sctx = {
                .fn = _fibril_main,
                .stack_base = fibril->stack,
                .stack_size = fibril->stack_size,
                .tls = fibril->tcb,
        };

        context_create(&fibril->ctx, &sctx);
        return (fid_t) fibril;
}

fid_t fibril_create(errno_t (*func)(void *), void *arg)
{
        return fibril_create_generic(func, arg, stack_size_get());
}

/** Delete a fibril that has never run.
 *
 * Free resources of a fibril that has been created with fibril_create()
 * but never started using fibril_start().
 *
 * @param fid Pointer to the fibril structure of the fibril to be
 *            added.
 */
void fibril_destroy(fid_t fid)
{
        fibril_t *fibril = (fibril_t *) fid;

        assert(!fibril->is_running);
        assert(fibril->stack);
        as_area_destroy(fibril->stack);
        fibril_teardown(fibril);
}

static void _insert_timeout(_timeout_t *timeout)
{
        futex_assert_is_locked(&fibril_futex);
        assert(timeout);

        link_t *tmp = timeout_list.head.next;
        while (tmp != &timeout_list.head) {
                _timeout_t *cur = list_get_instance(tmp, _timeout_t, link);

                if (ts_gteq(&cur->expires, &timeout->expires))
                        break;

                tmp = tmp->next;
        }

        list_insert_before(&timeout->link, tmp);
}

/**
 * Same as `fibril_wait_for()`, except with a timeout.
 *
 * It is guaranteed that timing out cannot cause another thread's
 * `fibril_notify()` to be lost. I.e. the function returns success if and
 * only if `fibril_notify()` was called after the last call to
 * wait/wait_timeout returned, and before the call timed out.
 *
 * @return ETIMEOUT if timed out. EOK otherwise.
 */
errno_t fibril_wait_timeout(fibril_event_t *event,
    const struct timespec *expires)
{
        assert(fibril_self()->rmutex_locks == 0);

        DPRINTF("### Fibril %p sleeping on event %p.\n", fibril_self(), event);

        if (!fibril_self()->thread_ctx) {
                fibril_self()->thread_ctx =
                    fibril_create_generic(_helper_fibril_fn, NULL, PAGE_SIZE);
                if (!fibril_self()->thread_ctx)
                        return ENOMEM;
        }

        futex_lock(&fibril_futex);

        if (event->fibril == _EVENT_TRIGGERED) {
                DPRINTF("### Already triggered. Returning. \n");
                event->fibril = _EVENT_INITIAL;
                futex_unlock(&fibril_futex);
                return EOK;
        }

        assert(event->fibril == _EVENT_INITIAL);

        fibril_t *srcf = fibril_self();
        fibril_t *dstf = NULL;

        /*
         * We cannot block here waiting for another fibril becoming
         * ready, since that would require unlocking the fibril_futex,
         * and that in turn would allow another thread to restore
         * the source fibril before this thread finished switching.
         *
         * Instead, we switch to an internal "helper" fibril whose only
         * job is to wait for an event, freeing the source fibril for
         * wakeups. There is always one for each running thread.
         */

        dstf = _ready_list_pop_nonblocking(true);
        if (!dstf) {
                // XXX: It is possible for the _ready_list_pop_nonblocking() to
                //      check for IPC, find a pending message, and trigger the
                //      event on which we are currently trying to sleep.
                if (event->fibril == _EVENT_TRIGGERED) {
                        event->fibril = _EVENT_INITIAL;
                        futex_unlock(&fibril_futex);
                        return EOK;
                }

                dstf = srcf->thread_ctx;
                assert(dstf);
        }

        _timeout_t timeout = { 0 };
        if (expires) {
                timeout.expires = *expires;
                timeout.event = event;
                _insert_timeout(&timeout);
        }

        assert(srcf);

        event->fibril = srcf;
        srcf->sleep_event = event;

        assert(event->fibril != _EVENT_INITIAL);

        _fibril_switch_to(SWITCH_FROM_BLOCKED, dstf, true);

        assert(event->fibril != srcf);
        assert(event->fibril != _EVENT_INITIAL);
        assert(event->fibril == _EVENT_TIMED_OUT || event->fibril == _EVENT_TRIGGERED);

        list_remove(&timeout.link);
        errno_t rc = (event->fibril == _EVENT_TIMED_OUT) ? ETIMEOUT : EOK;
        event->fibril = _EVENT_INITIAL;

        futex_unlock(&fibril_futex);
        _fibril_cleanup_dead();
        return rc;
}

void fibril_wait_for(fibril_event_t *event)
{
        assert(fibril_self()->rmutex_locks == 0);

        (void) fibril_wait_timeout(event, NULL);
}

/**
 * Wake up the fibril waiting for the given event.
 * Up to one wakeup is remembered if the fibril is not currently waiting.
 *
 * This function is safe for use under restricted mutex lock.
 */
void fibril_notify(fibril_event_t *event)
{
        futex_lock(&fibril_futex);
        _ready_list_push(_fibril_trigger_internal(event, _EVENT_TRIGGERED));
        futex_unlock(&fibril_futex);
}

/** Start a fibril that has not been running yet. */
void fibril_start(fibril_t *fibril)
{
        futex_lock(&fibril_futex);
        assert(!fibril->is_running);
        fibril->is_running = true;

        if (!link_in_use(&fibril->all_link))
                list_append(&fibril->all_link, &fibril_list);

        _ready_list_push(fibril);

        futex_unlock(&fibril_futex);
}

/** Start a fibril that has not been running yet. (obsolete) */
void fibril_add_ready(fibril_t *fibril)
{
        fibril_start(fibril);
}

/** @return the currently running fibril. */
fibril_t *fibril_self(void)
{
        assert(__tcb_is_set());
        tcb_t *tcb = __tcb_get();
        assert(tcb->fibril_data);
        return tcb->fibril_data;
}

/**
 * Obsolete, use fibril_self().
 *
 * @return ID of the currently running fibril.
 */
fid_t fibril_get_id(void)
{
        return (fid_t) fibril_self();
}

/**
 * Switch to another fibril, if one is ready to run.
 * Has no effect on a heavy fibril.
 */
void fibril_yield(void)
{
        if (fibril_self()->rmutex_locks > 0)
                return;

        fibril_t *f = _ready_list_pop_nonblocking(false);
        if (f)
                _fibril_switch_to(SWITCH_FROM_YIELD, f, false);
}

static errno_t _runner_fn(void *arg)
{
        _helper_fibril_fn(arg);
        return EOK;
}

/**
 * Spawn a given number of runners (i.e. OS threads) immediately, and
 * unconditionally. This is meant to be used for tests and debugging.
 * Regular programs should just use `fibril_enable_multithreaded()`.
 *
 * @param n  Number of runners to spawn.
 * @return   Number of runners successfully spawned.
 */
int fibril_test_spawn_runners(int n)
{
        assert(fibril_self()->rmutex_locks == 0);

        if (!multithreaded) {
                _ready_debug_check();
                if (futex_initialize(&ready_semaphore, ready_st_count) != EOK)
                        abort();
                multithreaded = true;
        }

        errno_t rc;

        for (int i = 0; i < n; i++) {
                rc = thread_create(_runner_fn, NULL, "fibril runner");
                if (rc != EOK)
                        return i;
        }

        return n;
}

/**
 * Opt-in to have more than one runner thread.
 *
 * Currently, a task only ever runs in one thread because multithreading
 * might break some existing code.
 *
 * Eventually, the number of runner threads for a given task should become
 * configurable in the environment and this function becomes no-op.
 */
void fibril_enable_multithreaded(void)
{
        // TODO: Implement better.
        //       For now, 4 total runners is a sensible default.
        if (!multithreaded) {
                fibril_test_spawn_runners(3);
        }
}

/**
 * Detach a fibril.
 */
void fibril_detach(fid_t f)
{
        // TODO: Currently all fibrils are detached by default, but they
        //       won't always be. Code that explicitly spawns fibrils with
        //       limited lifetime should call this function.
}

/**
 * Exit a fibril. Never returns.
 *
 * @param retval  Value to return from fibril_join() called on this fibril.
 */
_Noreturn void fibril_exit(long retval)
{
        // TODO: implement fibril_join() and remember retval
        (void) retval;

        fibril_t *f = _ready_list_pop_nonblocking(false);
        if (!f)
                f = fibril_self()->thread_ctx;

        _fibril_switch_to(SWITCH_FROM_DEAD, f, false);
        __builtin_unreachable();
}

void __fibrils_init(void)
{
        if (futex_initialize(&fibril_futex, 1) != EOK)
                abort();
        if (futex_initialize(&ipc_lists_futex, 1) != EOK)
                abort();

        /*
         * We allow a fixed, small amount of parallelism for IPC reads, but
         * since IPC is currently serialized in kernel, there's not much
         * we can get from more threads reading messages.
         */

#define IPC_BUFFER_COUNT 1024
        static _ipc_buffer_t buffers[IPC_BUFFER_COUNT];

        for (int i = 0; i < IPC_BUFFER_COUNT; i++) {
                list_append(&buffers[i].link, &ipc_buffer_free_list);
                _ready_up();
        }
}

void __fibrils_fini(void)
{
        futex_destroy(&fibril_futex);
        futex_destroy(&ipc_lists_futex);
}

void fibril_usleep(usec_t timeout)
{
        struct timespec expires;
        getuptime(&expires);
        ts_add_diff(&expires, USEC2NSEC(timeout));

        fibril_event_t event = FIBRIL_EVENT_INIT;
        fibril_wait_timeout(&event, &expires);
}

void fibril_sleep(sec_t sec)
{
        struct timespec expires;
        getuptime(&expires);
        expires.tv_sec += sec;

        fibril_event_t event = FIBRIL_EVENT_INIT;
        fibril_wait_timeout(&event, &expires);
}

void fibril_ipc_poke(void)
{
        DPRINTF("Poking.\n");
        /* Wakeup one thread sleeping in SYS_IPC_WAIT. */
        ipc_poke();
}

errno_t fibril_ipc_wait(ipc_call_t *call, const struct timespec *expires)
{
        return _wait_ipc(call, expires);
}

/** @}
 */

/* [<][>][^][v][top][bottom][index][help] */
HelenOS homepage, sources at GitHub