source: mainline/kernel/generic/src/proc/thread.c@ 111b9b9

/*
 * Copyright (c) 2010 Jakub Jermar
 * Copyright (c) 2018 Jiri Svoboda
 * 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 kernel_generic_proc
 * @{
 */

/**
 * @file
 * @brief Thread management functions.
 */

#include <assert.h>
#include <proc/scheduler.h>
#include <proc/thread.h>
#include <proc/task.h>
#include <mm/frame.h>
#include <mm/page.h>
#include <arch/asm.h>
#include <arch/cycle.h>
#include <arch.h>
#include <synch/spinlock.h>
#include <synch/waitq.h>
#include <synch/syswaitq.h>
#include <cpu.h>
#include <str.h>
#include <context.h>
#include <adt/list.h>
#include <adt/odict.h>
#include <time/clock.h>
#include <time/timeout.h>
#include <time/delay.h>
#include <config.h>
#include <arch/interrupt.h>
#include <smp/ipi.h>
#include <arch/faddr.h>
#include <atomic.h>
#include <mem.h>
#include <stdio.h>
#include <stdlib.h>
#include <main/uinit.h>
#include <syscall/copy.h>
#include <errno.h>
#include <debug.h>
#include <halt.h>

/** Thread states */
const char *thread_states[] = {
        "Invalid",
        "Running",
        "Sleeping",
        "Ready",
        "Entering",
        "Exiting",
        "Lingering"
};

enum sleep_state {
        SLEEP_INITIAL,
        SLEEP_ASLEEP,
        SLEEP_WOKE,
};

/** Lock protecting the @c threads ordered dictionary .
 *
 * For locking rules, see declaration thereof.
 */
IRQ_SPINLOCK_INITIALIZE(threads_lock);

/** Ordered dictionary of all threads by their address (i.e. pointer to
 * the thread_t structure).
 *
 * When a thread is found in the @c threads ordered dictionary, it is
 * guaranteed to exist as long as the @c threads_lock is held.
 *
 * Members are of type thread_t.
 *
 * This structure contains weak references. Any reference from it must not leave
 * threads_lock critical section unless strengthened via thread_try_ref().
 */
odict_t threads;

IRQ_SPINLOCK_STATIC_INITIALIZE(tidlock);
static thread_id_t last_tid = 0;

static slab_cache_t *thread_cache;

static void *threads_getkey(odlink_t *);
static int threads_cmp(void *, void *);

/** Thread wrapper.
 *
 * This wrapper is provided to ensure that every thread makes a call to
 * thread_exit() when its implementing function returns.
 *
 * interrupts_disable() is assumed.
 *
 */
static void cushion(void)
{
        void (*f)(void *) = THREAD->thread_code;
        void *arg = THREAD->thread_arg;
        THREAD->last_cycle = get_cycle();

        /* This is where each thread wakes up after its creation */
        irq_spinlock_unlock(&THREAD->lock, false);
        interrupts_enable();

        f(arg);

        /* Accumulate accounting to the task */
        irq_spinlock_lock(&THREAD->lock, true);
        if (!THREAD->uncounted) {
                thread_update_accounting(true);
                uint64_t ucycles = THREAD->ucycles;
                THREAD->ucycles = 0;
                uint64_t kcycles = THREAD->kcycles;
                THREAD->kcycles = 0;

                irq_spinlock_pass(&THREAD->lock, &TASK->lock);
                TASK->ucycles += ucycles;
                TASK->kcycles += kcycles;
                irq_spinlock_unlock(&TASK->lock, true);
        } else
                irq_spinlock_unlock(&THREAD->lock, true);

        thread_exit();

        /* Not reached */
}

/** Initialization and allocation for thread_t structure
 *
 */
static errno_t thr_constructor(void *obj, unsigned int kmflags)
{
        thread_t *thread = (thread_t *) obj;

        irq_spinlock_initialize(&thread->lock, "thread_t_lock");
        link_initialize(&thread->rq_link);
        link_initialize(&thread->wq_link);
        link_initialize(&thread->th_link);

        /* call the architecture-specific part of the constructor */
        thr_constructor_arch(thread);

        /*
         * Allocate the kernel stack from the low-memory to prevent an infinite
         * nesting of TLB-misses when accessing the stack from the part of the
         * TLB-miss handler written in C.
         *
         * Note that low-memory is safe to be used for the stack as it will be
         * covered by the kernel identity mapping, which guarantees not to
         * nest TLB-misses infinitely (either via some hardware mechanism or
         * by the construction of the assembly-language part of the TLB-miss
         * handler).
         *
         * This restriction can be lifted once each architecture provides
         * a similar guarantee, for example, by locking the kernel stack
         * in the TLB whenever it is allocated from the high-memory and the
         * thread is being scheduled to run.
         */
        kmflags |= FRAME_LOWMEM;
        kmflags &= ~FRAME_HIGHMEM;

        /*
         * NOTE: All kernel stacks must be aligned to STACK_SIZE,
         *       see CURRENT.
         */

        uintptr_t stack_phys =
            frame_alloc(STACK_FRAMES, kmflags, STACK_SIZE - 1);
        if (!stack_phys)
                return ENOMEM;

        thread->kstack = (uint8_t *) PA2KA(stack_phys);

#ifdef CONFIG_UDEBUG
        mutex_initialize(&thread->udebug.lock, MUTEX_PASSIVE);
#endif

        return EOK;
}

/** Destruction of thread_t object */
static size_t thr_destructor(void *obj)
{
        thread_t *thread = (thread_t *) obj;

        /* call the architecture-specific part of the destructor */
        thr_destructor_arch(thread);

        frame_free(KA2PA(thread->kstack), STACK_FRAMES);

        return STACK_FRAMES;  /* number of frames freed */
}

/** Initialize threads
 *
 * Initialize kernel threads support.
 *
 */
void thread_init(void)
{
        THREAD = NULL;

        atomic_store(&nrdy, 0);
        thread_cache = slab_cache_create("thread_t", sizeof(thread_t), _Alignof(thread_t),
            thr_constructor, thr_destructor, 0);

        odict_initialize(&threads, threads_getkey, threads_cmp);
}

/** Wire thread to the given CPU
 *
 * @param cpu CPU to wire the thread to.
 *
 */
void thread_wire(thread_t *thread, cpu_t *cpu)
{
        irq_spinlock_lock(&thread->lock, true);
        thread->cpu = cpu;
        thread->wired = true;
        irq_spinlock_unlock(&thread->lock, true);
}

/** Invoked right before thread_ready() readies the thread. thread is locked. */
static void before_thread_is_ready(thread_t *thread)
{
        assert(irq_spinlock_locked(&thread->lock));
}

/** Make thread ready
 *
 * Switch thread to the ready state. Consumes reference passed by the caller.
 *
 * @param thread Thread to make ready.
 *
 */
void thread_ready(thread_t *thread)
{
        irq_spinlock_lock(&thread->lock, true);

        assert(thread->state != Ready);

        before_thread_is_ready(thread);

        int i = (thread->priority < RQ_COUNT - 1) ?
            ++thread->priority : thread->priority;

        cpu_t *cpu;
        if (thread->wired || thread->nomigrate || thread->fpu_context_engaged) {
                /* Cannot ready to another CPU */
                assert(thread->cpu != NULL);
                cpu = thread->cpu;
        } else if (thread->stolen) {
                /* Ready to the stealing CPU */
                cpu = CPU;
        } else if (thread->cpu) {
                /* Prefer the CPU on which the thread ran last */
                assert(thread->cpu != NULL);
                cpu = thread->cpu;
        } else {
                cpu = CPU;
        }

        thread->state = Ready;

        irq_spinlock_pass(&thread->lock, &(cpu->rq[i].lock));

        /*
         * Append thread to respective ready queue
         * on respective processor.
         */

        list_append(&thread->rq_link, &cpu->rq[i].rq);
        cpu->rq[i].n++;
        irq_spinlock_unlock(&(cpu->rq[i].lock), true);

        atomic_inc(&nrdy);
        atomic_inc(&cpu->nrdy);
}

/** Create new thread
 *
 * Create a new thread.
 *
 * @param func      Thread's implementing function.
 * @param arg       Thread's implementing function argument.
 * @param task      Task to which the thread belongs. The caller must
 *                  guarantee that the task won't cease to exist during the
 *                  call. The task's lock may not be held.
 * @param flags     Thread flags.
 * @param name      Symbolic name (a copy is made).
 *
 * @return New thread's structure on success, NULL on failure.
 *
 */
thread_t *thread_create(void (*func)(void *), void *arg, task_t *task,
    thread_flags_t flags, const char *name)
{
        thread_t *thread = (thread_t *) slab_alloc(thread_cache, FRAME_ATOMIC);
        if (!thread)
                return NULL;

        refcount_init(&thread->refcount);

        if (thread_create_arch(thread, flags) != EOK) {
                slab_free(thread_cache, thread);
                return NULL;
        }

        /* Not needed, but good for debugging */
        memsetb(thread->kstack, STACK_SIZE, 0);

        irq_spinlock_lock(&tidlock, true);
        thread->tid = ++last_tid;
        irq_spinlock_unlock(&tidlock, true);

        memset(&thread->saved_context, 0, sizeof(thread->saved_context));
        context_set(&thread->saved_context, FADDR(cushion),
            (uintptr_t) thread->kstack, STACK_SIZE);

        current_initialize((current_t *) thread->kstack);

        ipl_t ipl = interrupts_disable();
        thread->saved_ipl = interrupts_read();
        interrupts_restore(ipl);

        str_cpy(thread->name, THREAD_NAME_BUFLEN, name);

        thread->thread_code = func;
        thread->thread_arg = arg;
        thread->ucycles = 0;
        thread->kcycles = 0;
        thread->uncounted =
            ((flags & THREAD_FLAG_UNCOUNTED) == THREAD_FLAG_UNCOUNTED);
        thread->priority = -1;          /* Start in rq[0] */
        thread->cpu = NULL;
        thread->wired = false;
        thread->stolen = false;
        thread->uspace =
            ((flags & THREAD_FLAG_USPACE) == THREAD_FLAG_USPACE);

        thread->nomigrate = 0;
        thread->state = Entering;

        atomic_init(&thread->sleep_queue, NULL);

        thread->in_copy_from_uspace = false;
        thread->in_copy_to_uspace = false;

        thread->interrupted = false;
        atomic_init(&thread->sleep_state, SLEEP_INITIAL);

        waitq_initialize(&thread->join_wq);

        thread->task = task;

        thread->fpu_context_exists = false;
        thread->fpu_context_engaged = false;

        odlink_initialize(&thread->lthreads);

#ifdef CONFIG_UDEBUG
        /* Initialize debugging stuff */
        thread->btrace = false;
        udebug_thread_initialize(&thread->udebug);
#endif

        if ((flags & THREAD_FLAG_NOATTACH) != THREAD_FLAG_NOATTACH)
                thread_attach(thread, task);

        return thread;
}

/** Destroy thread memory structure
 *
 * Detach thread from all queues, cpus etc. and destroy it.
 *
 * @param obj  Thread to be destroyed.
 *
 */
static void thread_destroy(void *obj)
{
        thread_t *thread = (thread_t *) obj;

        assert_link_not_used(&thread->rq_link);
        assert_link_not_used(&thread->wq_link);

        assert(thread->task);

        ipl_t ipl = interrupts_disable();

        /* Remove thread from task's list. */
        irq_spinlock_lock(&thread->task->lock, false);
        list_remove(&thread->th_link);
        irq_spinlock_unlock(&thread->task->lock, false);

        /* Remove thread from global list. */
        irq_spinlock_lock(&threads_lock, false);
        odict_remove(&thread->lthreads);
        irq_spinlock_unlock(&threads_lock, false);

        /* Clear cpu->fpu_owner if set to this thread. */
        irq_spinlock_lock(&thread->lock, false);

        assert((thread->state == Exiting) || (thread->state == Lingering));
        assert(thread->cpu);

        irq_spinlock_lock(&thread->cpu->lock, false);
        if (thread->cpu->fpu_owner == thread)
                thread->cpu->fpu_owner = NULL;
        irq_spinlock_unlock(&thread->cpu->lock, false);

        irq_spinlock_unlock(&thread->lock, false);

        interrupts_restore(ipl);

        /*
         * Drop the reference to the containing task.
         */
        task_release(thread->task);
        thread->task = NULL;

        slab_free(thread_cache, thread);
}

void thread_put(thread_t *thread)
{
        if (refcount_down(&thread->refcount)) {
                thread_destroy(thread);
        }
}

/** Make the thread visible to the system.
 *
 * Attach the thread structure to the current task and make it visible in the
 * threads_tree.
 *
 * @param t    Thread to be attached to the task.
 * @param task Task to which the thread is to be attached.
 *
 */
void thread_attach(thread_t *thread, task_t *task)
{
        ipl_t ipl = interrupts_disable();

        /*
         * Attach to the specified task.
         */
        irq_spinlock_lock(&task->lock, false);

        /* Hold a reference to the task. */
        task_hold(task);

        /* Must not count kbox thread into lifecount */
        if (thread->uspace)
                atomic_inc(&task->lifecount);

        list_append(&thread->th_link, &task->threads);

        irq_spinlock_unlock(&task->lock, false);

        /*
         * Register this thread in the system-wide dictionary.
         */
        irq_spinlock_lock(&threads_lock, false);
        odict_insert(&thread->lthreads, &threads, NULL);
        irq_spinlock_unlock(&threads_lock, false);

        interrupts_restore(ipl);
}

/** Terminate thread.
 *
 * End current thread execution and switch it to the exiting state.
 * All pending timeouts are executed.
 *
 */
void thread_exit(void)
{
        if (THREAD->uspace) {
#ifdef CONFIG_UDEBUG
                /* Generate udebug THREAD_E event */
                udebug_thread_e_event();

                /*
                 * This thread will not execute any code or system calls from
                 * now on.
                 */
                udebug_stoppable_begin();
#endif
                if (atomic_predec(&TASK->lifecount) == 0) {
                        /*
                         * We are the last userspace thread in the task that
                         * still has not exited. With the exception of the
                         * moment the task was created, new userspace threads
                         * can only be created by threads of the same task.
                         * We are safe to perform cleanup.
                         *
                         */
                        ipc_cleanup();
                        sys_waitq_task_cleanup();
                        LOG("Cleanup of task %" PRIu64 " completed.", TASK->taskid);
                }
        }

        irq_spinlock_lock(&THREAD->lock, true);
        THREAD->state = Exiting;
        irq_spinlock_unlock(&THREAD->lock, true);

        scheduler();

        panic("should never be reached");
}

/** Interrupts an existing thread so that it may exit as soon as possible.
 *
 * Threads that are blocked waiting for a synchronization primitive
 * are woken up with a return code of EINTR if the
 * blocking call was interruptable. See waitq_sleep_timeout().
 *
 * Interrupted threads automatically exit when returning back to user space.
 *
 * @param thread A valid thread object.
 */
void thread_interrupt(thread_t *thread)
{
        assert(thread != NULL);
        thread->interrupted = true;
        thread_wakeup(thread);
}

/** Prepare for putting the thread to sleep.
 *
 * @returns whether the thread is currently terminating. If THREAD_OK
 * is returned, the thread is guaranteed to be woken up instantly if the thread
 * is terminated at any time between this function's return and
 * thread_wait_finish(). If THREAD_TERMINATING is returned, the thread can still
 * go to sleep, but doing so will delay termination.
 */
thread_termination_state_t thread_wait_start(void)
{
        assert(THREAD != NULL);

        /*
         * This is an exchange rather than a store so that we can use the acquire
         * semantics, which is needed to ensure that code after this operation sees
         * memory ops made before thread_wakeup() in other thread, if that wakeup
         * was reset by this operation.
         *
         * In particular, we need this to ensure we can't miss the thread being
         * terminated concurrently with a synchronization primitive preparing to
         * sleep.
         */
        (void) atomic_exchange_explicit(&THREAD->sleep_state, SLEEP_INITIAL,
            memory_order_acquire);

        return THREAD->interrupted ? THREAD_TERMINATING : THREAD_OK;
}

static void thread_wait_internal(void)
{
        assert(THREAD != NULL);

        ipl_t ipl = interrupts_disable();

        if (atomic_load(&haltstate))
                halt();

        /*
         * Lock here to prevent a race between entering the scheduler and another
         * thread rescheduling this thread.
         */
        irq_spinlock_lock(&THREAD->lock, false);

        int expected = SLEEP_INITIAL;

        /* Only set SLEEP_ASLEEP in sleep pad if it's still in initial state */
        if (atomic_compare_exchange_strong_explicit(&THREAD->sleep_state, &expected,
            SLEEP_ASLEEP, memory_order_acq_rel, memory_order_acquire)) {
                THREAD->state = Sleeping;
                scheduler_locked(ipl);
        } else {
                assert(expected == SLEEP_WOKE);
                /* Return immediately. */
                irq_spinlock_unlock(&THREAD->lock, false);
                interrupts_restore(ipl);
        }
}

static void thread_wait_timeout_callback(void *arg)
{
        thread_wakeup(arg);
}

/**
 * Suspends this thread's execution until thread_wakeup() is called on it,
 * or deadline is reached.
 *
 * The way this would normally be used is that the current thread call
 * thread_wait_start(), and if interruption has not been signaled, stores
 * a reference to itself in a synchronized structure (such as waitq).
 * After that, it releases any spinlocks it might hold and calls this function.
 *
 * The thread doing the wakeup will acquire the thread's reference from said
 * synchronized structure and calls thread_wakeup() on it.
 *
 * Notably, there can be more than one thread performing wakeup.
 * The number of performed calls to thread_wakeup(), or their relative
 * ordering with thread_wait_finish(), does not matter. However, calls to
 * thread_wakeup() are expected to be synchronized with thread_wait_start()
 * with which they are associated, otherwise wakeups may be missed.
 * However, the operation of thread_wakeup() is defined at any time,
 * synchronization notwithstanding (in the sense of C un/defined behavior),
 * and is in fact used to interrupt waiting threads by external events.
 * The waiting thread must operate correctly in face of spurious wakeups,
 * and clean up its reference in the synchronization structure if necessary.
 *
 * Returns THREAD_WAIT_TIMEOUT if timeout fired, which is a necessary condition
 * for it to have been waken up by the timeout, but the caller must assume
 * that proper wakeups, timeouts and interrupts may occur concurrently, so
 * the fact timeout has been registered does not necessarily mean the thread
 * has not been woken up or interrupted.
 */
thread_wait_result_t thread_wait_finish(deadline_t deadline)
{
        assert(THREAD != NULL);

        timeout_t timeout;

        if (deadline != DEADLINE_NEVER) {
                /* Extra check to avoid setting up a deadline if we don't need to. */
                if (atomic_load_explicit(&THREAD->sleep_state, memory_order_acquire) !=
                    SLEEP_INITIAL)
                        return THREAD_WAIT_SUCCESS;

                timeout_initialize(&timeout);
                timeout_register_deadline(&timeout, deadline,
                    thread_wait_timeout_callback, THREAD);
        }

        thread_wait_internal();

        if (deadline != DEADLINE_NEVER && !timeout_unregister(&timeout)) {
                return THREAD_WAIT_TIMEOUT;
        } else {
                return THREAD_WAIT_SUCCESS;
        }
}

void thread_wakeup(thread_t *thread)
{
        assert(thread != NULL);

        int state = atomic_exchange_explicit(&thread->sleep_state, SLEEP_WOKE,
            memory_order_release);

        if (state == SLEEP_ASLEEP) {
                /*
                 * Only one thread gets to do this.
                 * The reference consumed here is the reference implicitly passed to
                 * the waking thread by the sleeper in thread_wait_finish().
                 */
                thread_ready(thread);
        }
}

/** Prevent the current thread from being migrated to another processor. */
void thread_migration_disable(void)
{
        assert(THREAD);

        THREAD->nomigrate++;
}

/** Allow the current thread to be migrated to another processor. */
void thread_migration_enable(void)
{
        assert(THREAD);
        assert(THREAD->nomigrate > 0);

        if (THREAD->nomigrate > 0)
                THREAD->nomigrate--;
}

/** Thread sleep
 *
 * Suspend execution of the current thread.
 *
 * @param sec Number of seconds to sleep.
 *
 */
void thread_sleep(uint32_t sec)
{
        /*
         * Sleep in 1000 second steps to support
         * full argument range
         */
        while (sec > 0) {
                uint32_t period = (sec > 1000) ? 1000 : sec;

                thread_usleep(period * 1000000);
                sec -= period;
        }
}

errno_t thread_join(thread_t *thread)
{
        return thread_join_timeout(thread, SYNCH_NO_TIMEOUT, SYNCH_FLAGS_NONE);
}

/** Wait for another thread to exit.
 * This function does not destroy the thread. Reference counting handles that.
 *
 * @param thread Thread to join on exit.
 * @param usec   Timeout in microseconds.
 * @param flags  Mode of operation.
 *
 * @return An error code from errno.h or an error code from synch.h.
 *
 */
errno_t thread_join_timeout(thread_t *thread, uint32_t usec, unsigned int flags)
{
        if (thread == THREAD)
                return EINVAL;

        irq_spinlock_lock(&thread->lock, true);
        state_t state = thread->state;
        irq_spinlock_unlock(&thread->lock, true);

        if (state == Exiting) {
                return EOK;
        } else {
                return _waitq_sleep_timeout(&thread->join_wq, usec, flags);
        }
}

/** Thread usleep
 *
 * Suspend execution of the current thread.
 *
 * @param usec Number of microseconds to sleep.
 *
 */
void thread_usleep(uint32_t usec)
{
        waitq_t wq;

        waitq_initialize(&wq);

        (void) waitq_sleep_timeout(&wq, usec);
}

static void thread_print(thread_t *thread, bool additional)
{
        uint64_t ucycles, kcycles;
        char usuffix, ksuffix;
        order_suffix(thread->ucycles, &ucycles, &usuffix);
        order_suffix(thread->kcycles, &kcycles, &ksuffix);

        char *name;
        if (str_cmp(thread->name, "uinit") == 0)
                name = thread->task->name;
        else
                name = thread->name;

        if (additional)
                printf("%-8" PRIu64 " %p %p %9" PRIu64 "%c %9" PRIu64 "%c ",
                    thread->tid, thread->thread_code, thread->kstack,
                    ucycles, usuffix, kcycles, ksuffix);
        else
                printf("%-8" PRIu64 " %-14s %p %-8s %p %-5" PRIu32 "\n",
                    thread->tid, name, thread, thread_states[thread->state],
                    thread->task, thread->task->container);

        if (additional) {
                if (thread->cpu)
                        printf("%-5u", thread->cpu->id);
                else
                        printf("none ");

                if (thread->state == Sleeping) {
                        printf(" %p", thread->sleep_queue);
                }

                printf("\n");
        }
}

/** Print list of threads debug info
 *
 * @param additional Print additional information.
 *
 */
void thread_print_list(bool additional)
{
        thread_t *thread;

        /* Accessing system-wide threads list through thread_first()/thread_next(). */
        irq_spinlock_lock(&threads_lock, true);

        if (sizeof(void *) <= 4) {
                if (additional)
                        printf("[id    ] [code    ] [stack   ] [ucycles ] [kcycles ]"
                            " [cpu] [waitqueue]\n");
                else
                        printf("[id    ] [name        ] [address ] [state ] [task    ]"
                            " [ctn]\n");
        } else {
                if (additional) {
                        printf("[id    ] [code            ] [stack           ] [ucycles ] [kcycles ]"
                            " [cpu] [waitqueue       ]\n");
                } else
                        printf("[id    ] [name        ] [address         ] [state ]"
                            " [task            ] [ctn]\n");
        }

        thread = thread_first();
        while (thread != NULL) {
                thread_print(thread, additional);
                thread = thread_next(thread);
        }

        irq_spinlock_unlock(&threads_lock, true);
}

static bool thread_exists(thread_t *thread)
{
        odlink_t *odlink = odict_find_eq(&threads, thread, NULL);
        return odlink != NULL;
}

/** Check whether the thread exists, and if so, return a reference to it.
 */
thread_t *thread_try_get(thread_t *thread)
{
        irq_spinlock_lock(&threads_lock, true);

        if (thread_exists(thread)) {
                /* Try to strengthen the reference. */
                thread = thread_try_ref(thread);
        } else {
                thread = NULL;
        }

        irq_spinlock_unlock(&threads_lock, true);

        return thread;
}

/** Update accounting of current thread.
 *
 * Note that thread_lock on THREAD must be already held and
 * interrupts must be already disabled.
 *
 * @param user True to update user accounting, false for kernel.
 *
 */
void thread_update_accounting(bool user)
{
        uint64_t time = get_cycle();

        assert(interrupts_disabled());
        assert(irq_spinlock_locked(&THREAD->lock));

        if (user)
                THREAD->ucycles += time - THREAD->last_cycle;
        else
                THREAD->kcycles += time - THREAD->last_cycle;

        THREAD->last_cycle = time;
}

/** Find thread structure corresponding to thread ID.
 *
 * The threads_lock must be already held by the caller of this function and
 * interrupts must be disabled.
 *
 * The returned reference is weak.
 * If the caller needs to keep it, thread_try_ref() must be used to upgrade
 * to a strong reference _before_ threads_lock is released.
 *
 * @param id Thread ID.
 *
 * @return Thread structure address or NULL if there is no such thread ID.
 *
 */
thread_t *thread_find_by_id(thread_id_t thread_id)
{
        thread_t *thread;

        assert(interrupts_disabled());
        assert(irq_spinlock_locked(&threads_lock));

        thread = thread_first();
        while (thread != NULL) {
                if (thread->tid == thread_id)
                        return thread;

                thread = thread_next(thread);
        }

        return NULL;
}

/** Get count of threads.
 *
 * @return Number of threads in the system
 */
size_t thread_count(void)
{
        assert(interrupts_disabled());
        assert(irq_spinlock_locked(&threads_lock));

        return odict_count(&threads);
}

/** Get first thread.
 *
 * @return Pointer to first thread or @c NULL if there are none.
 */
thread_t *thread_first(void)
{
        odlink_t *odlink;

        assert(interrupts_disabled());
        assert(irq_spinlock_locked(&threads_lock));

        odlink = odict_first(&threads);
        if (odlink == NULL)
                return NULL;

        return odict_get_instance(odlink, thread_t, lthreads);
}

/** Get next thread.
 *
 * @param cur Current thread
 * @return Pointer to next thread or @c NULL if there are no more threads.
 */
thread_t *thread_next(thread_t *cur)
{
        odlink_t *odlink;

        assert(interrupts_disabled());
        assert(irq_spinlock_locked(&threads_lock));

        odlink = odict_next(&cur->lthreads, &threads);
        if (odlink == NULL)
                return NULL;

        return odict_get_instance(odlink, thread_t, lthreads);
}

#ifdef CONFIG_UDEBUG

void thread_stack_trace(thread_id_t thread_id)
{
        irq_spinlock_lock(&threads_lock, true);
        thread_t *thread = thread_try_ref(thread_find_by_id(thread_id));
        irq_spinlock_unlock(&threads_lock, true);

        if (thread == NULL) {
                printf("No such thread.\n");
                return;
        }

        /*
         * Schedule a stack trace to be printed
         * just before the thread is scheduled next.
         *
         * If the thread is sleeping then try to interrupt
         * the sleep. Any request for printing an uspace stack
         * trace from within the kernel should be always
         * considered a last resort debugging means, therefore
         * forcing the thread's sleep to be interrupted
         * is probably justifiable.
         */

        irq_spinlock_lock(&thread->lock, true);

        bool sleeping = false;
        istate_t *istate = thread->udebug.uspace_state;
        if (istate != NULL) {
                printf("Scheduling thread stack trace.\n");
                thread->btrace = true;
                if (thread->state == Sleeping)
                        sleeping = true;
        } else
                printf("Thread interrupt state not available.\n");

        irq_spinlock_unlock(&thread->lock, true);

        if (sleeping)
                thread_wakeup(thread);

        thread_put(thread);
}

#endif /* CONFIG_UDEBUG */

/** Get key function for the @c threads ordered dictionary.
 *
 * @param odlink Link
 * @return Pointer to thread structure cast as 'void *'
 */
static void *threads_getkey(odlink_t *odlink)
{
        thread_t *thread = odict_get_instance(odlink, thread_t, lthreads);
        return (void *) thread;
}

/** Key comparison function for the @c threads ordered dictionary.
 *
 * @param a Pointer to thread A
 * @param b Pointer to thread B
 * @return -1, 0, 1 iff pointer to A is less than, equal to, greater than B
 */
static int threads_cmp(void *a, void *b)
{
        if (a > b)
                return -1;
        else if (a == b)
                return 0;
        else
                return +1;
}

/** Process syscall to create new thread.
 *
 */
sys_errno_t sys_thread_create(uspace_ptr_uspace_arg_t uspace_uarg, uspace_ptr_char uspace_name,
    size_t name_len, uspace_ptr_thread_id_t uspace_thread_id)
{
        if (name_len > THREAD_NAME_BUFLEN - 1)
                name_len = THREAD_NAME_BUFLEN - 1;

        char namebuf[THREAD_NAME_BUFLEN];
        errno_t rc = copy_from_uspace(namebuf, uspace_name, name_len);
        if (rc != EOK)
                return (sys_errno_t) rc;

        namebuf[name_len] = 0;

        /*
         * In case of failure, kernel_uarg will be deallocated in this function.
         * In case of success, kernel_uarg will be freed in uinit().
         */
        uspace_arg_t *kernel_uarg =
            (uspace_arg_t *) malloc(sizeof(uspace_arg_t));
        if (!kernel_uarg)
                return (sys_errno_t) ENOMEM;

        rc = copy_from_uspace(kernel_uarg, uspace_uarg, sizeof(uspace_arg_t));
        if (rc != EOK) {
                free(kernel_uarg);
                return (sys_errno_t) rc;
        }

        thread_t *thread = thread_create(uinit, kernel_uarg, TASK,
            THREAD_FLAG_USPACE | THREAD_FLAG_NOATTACH, namebuf);
        if (thread) {
                if (uspace_thread_id) {
                        rc = copy_to_uspace(uspace_thread_id, &thread->tid,
                            sizeof(thread->tid));
                        if (rc != EOK) {
                                /*
                                 * We have encountered a failure, but the thread
                                 * has already been created. We need to undo its
                                 * creation now.
                                 */

                                /*
                                 * The new thread structure is initialized, but
                                 * is still not visible to the system.
                                 * We can safely deallocate it.
                                 */
                                slab_free(thread_cache, thread);
                                free(kernel_uarg);

                                return (sys_errno_t) rc;
                        }
                }

#ifdef CONFIG_UDEBUG
                /*
                 * Generate udebug THREAD_B event and attach the thread.
                 * This must be done atomically (with the debug locks held),
                 * otherwise we would either miss some thread or receive
                 * THREAD_B events for threads that already existed
                 * and could be detected with THREAD_READ before.
                 */
                udebug_thread_b_event_attach(thread, TASK);
#else
                thread_attach(thread, TASK);
#endif
                thread_ready(thread);

                return 0;
        } else
                free(kernel_uarg);

        return (sys_errno_t) ENOMEM;
}

/** Process syscall to terminate thread.
 *
 */
sys_errno_t sys_thread_exit(int uspace_status)
{
        thread_exit();
}

/** Syscall for getting TID.
 *
 * @param uspace_thread_id Userspace address of 8-byte buffer where to store
 * current thread ID.
 *
 * @return 0 on success or an error code from @ref errno.h.
 *
 */
sys_errno_t sys_thread_get_id(uspace_ptr_thread_id_t uspace_thread_id)
{
        /*
         * No need to acquire lock on THREAD because tid
         * remains constant for the lifespan of the thread.
         *
         */
        return (sys_errno_t) copy_to_uspace(uspace_thread_id, &THREAD->tid,
            sizeof(THREAD->tid));
}

/** Syscall wrapper for sleeping. */
sys_errno_t sys_thread_usleep(uint32_t usec)
{
        thread_usleep(usec);
        return 0;
}

sys_errno_t sys_thread_udelay(uint32_t usec)
{
        delay(usec);
        return 0;
}

/** @}
 */