Merge llvm-project release/18.x llvmorg-18.1.3-0-gc13b7485b879

[FreeBSD/FreeBSD.git] / sys / kern / subr_sleepqueue.c

/*-
 * SPDX-License-Identifier: BSD-2-Clause
 *
 * Copyright (c) 2004 John Baldwin <jhb@FreeBSD.org>
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. 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.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``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 OR CONTRIBUTORS 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.
 */

/*
 * Implementation of sleep queues used to hold queue of threads blocked on
 * a wait channel.  Sleep queues are different from turnstiles in that wait
 * channels are not owned by anyone, so there is no priority propagation.
 * Sleep queues can also provide a timeout and can also be interrupted by
 * signals.  That said, there are several similarities between the turnstile
 * and sleep queue implementations.  (Note: turnstiles were implemented
 * first.)  For example, both use a hash table of the same size where each
 * bucket is referred to as a "chain" that contains both a spin lock and
 * a linked list of queues.  An individual queue is located by using a hash
 * to pick a chain, locking the chain, and then walking the chain searching
 * for the queue.  This means that a wait channel object does not need to
 * embed its queue head just as locks do not embed their turnstile queue
 * head.  Threads also carry around a sleep queue that they lend to the
 * wait channel when blocking.  Just as in turnstiles, the queue includes
 * a free list of the sleep queues of other threads blocked on the same
 * wait channel in the case of multiple waiters.
 *
 * Some additional functionality provided by sleep queues include the
 * ability to set a timeout.  The timeout is managed using a per-thread
 * callout that resumes a thread if it is asleep.  A thread may also
 * catch signals while it is asleep (aka an interruptible sleep).  The
 * signal code uses sleepq_abort() to interrupt a sleeping thread.  Finally,
 * sleep queues also provide some extra assertions.  One is not allowed to
 * mix the sleep/wakeup and cv APIs for a given wait channel.  Also, one
 * must consistently use the same lock to synchronize with a wait channel,
 * though this check is currently only a warning for sleep/wakeup due to
 * pre-existing abuse of that API.  The same lock must also be held when
 * awakening threads, though that is currently only enforced for condition
 * variables.
 */

#include <sys/cdefs.h>
#include "opt_sleepqueue_profiling.h"
#include "opt_ddb.h"
#include "opt_sched.h"
#include "opt_stack.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/lock.h>
#include <sys/kernel.h>
#include <sys/ktr.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/sbuf.h>
#include <sys/sched.h>
#include <sys/sdt.h>
#include <sys/signalvar.h>
#include <sys/sleepqueue.h>
#include <sys/stack.h>
#include <sys/sysctl.h>
#include <sys/time.h>
#ifdef EPOCH_TRACE
#include <sys/epoch.h>
#endif

#include <machine/atomic.h>

#include <vm/uma.h>

#ifdef DDB
#include <ddb/ddb.h>
#endif

/*
 * Constants for the hash table of sleep queue chains.
 * SC_TABLESIZE must be a power of two for SC_MASK to work properly.
 */
#ifndef SC_TABLESIZE
#define SC_TABLESIZE    256
#endif
CTASSERT(powerof2(SC_TABLESIZE));
#define SC_MASK         (SC_TABLESIZE - 1)
#define SC_SHIFT        8
#define SC_HASH(wc)     ((((uintptr_t)(wc) >> SC_SHIFT) ^ (uintptr_t)(wc)) & \
                            SC_MASK)
#define SC_LOOKUP(wc)   &sleepq_chains[SC_HASH(wc)]
#define NR_SLEEPQS      2
/*
 * There are two different lists of sleep queues.  Both lists are connected
 * via the sq_hash entries.  The first list is the sleep queue chain list
 * that a sleep queue is on when it is attached to a wait channel.  The
 * second list is the free list hung off of a sleep queue that is attached
 * to a wait channel.
 *
 * Each sleep queue also contains the wait channel it is attached to, the
 * list of threads blocked on that wait channel, flags specific to the
 * wait channel, and the lock used to synchronize with a wait channel.
 * The flags are used to catch mismatches between the various consumers
 * of the sleep queue API (e.g. sleep/wakeup and condition variables).
 * The lock pointer is only used when invariants are enabled for various
 * debugging checks.
 *
 * Locking key:
 *  c - sleep queue chain lock
 */
struct sleepqueue {
        struct threadqueue sq_blocked[NR_SLEEPQS]; /* (c) Blocked threads. */
        u_int sq_blockedcnt[NR_SLEEPQS];        /* (c) N. of blocked threads. */
        LIST_ENTRY(sleepqueue) sq_hash;         /* (c) Chain and free list. */
        LIST_HEAD(, sleepqueue) sq_free;        /* (c) Free queues. */
        const void      *sq_wchan;              /* (c) Wait channel. */
        int     sq_type;                        /* (c) Queue type. */
#ifdef INVARIANTS
        struct lock_object *sq_lock;            /* (c) Associated lock. */
#endif
};

struct sleepqueue_chain {
        LIST_HEAD(, sleepqueue) sc_queues;      /* List of sleep queues. */
        struct mtx sc_lock;                     /* Spin lock for this chain. */
#ifdef SLEEPQUEUE_PROFILING
        u_int   sc_depth;                       /* Length of sc_queues. */
        u_int   sc_max_depth;                   /* Max length of sc_queues. */
#endif
} __aligned(CACHE_LINE_SIZE);

#ifdef SLEEPQUEUE_PROFILING
static SYSCTL_NODE(_debug, OID_AUTO, sleepq, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
    "sleepq profiling");
static SYSCTL_NODE(_debug_sleepq, OID_AUTO, chains,
    CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
    "sleepq chain stats");
static u_int sleepq_max_depth;
SYSCTL_UINT(_debug_sleepq, OID_AUTO, max_depth, CTLFLAG_RD, &sleepq_max_depth,
    0, "maxmimum depth achieved of a single chain");

static void     sleepq_profile(const char *wmesg);
static int      prof_enabled;
#endif
static struct sleepqueue_chain sleepq_chains[SC_TABLESIZE];
static uma_zone_t sleepq_zone;

/*
 * Prototypes for non-exported routines.
 */
static int      sleepq_catch_signals(const void *wchan, int pri);
static inline int sleepq_check_signals(void);
static inline int sleepq_check_timeout(void);
#ifdef INVARIANTS
static void     sleepq_dtor(void *mem, int size, void *arg);
#endif
static int      sleepq_init(void *mem, int size, int flags);
static int      sleepq_resume_thread(struct sleepqueue *sq, struct thread *td,
                    int pri, int srqflags);
static void     sleepq_remove_thread(struct sleepqueue *sq, struct thread *td);
static void     sleepq_switch(const void *wchan, int pri);
static void     sleepq_timeout(void *arg);

SDT_PROBE_DECLARE(sched, , , sleep);
SDT_PROBE_DECLARE(sched, , , wakeup);

/*
 * Initialize SLEEPQUEUE_PROFILING specific sysctl nodes.
 * Note that it must happen after sleepinit() has been fully executed, so
 * it must happen after SI_SUB_KMEM SYSINIT() subsystem setup.
 */
#ifdef SLEEPQUEUE_PROFILING
static void
init_sleepqueue_profiling(void)
{
        char chain_name[10];
        struct sysctl_oid *chain_oid;
        u_int i;

        for (i = 0; i < SC_TABLESIZE; i++) {
                snprintf(chain_name, sizeof(chain_name), "%u", i);
                chain_oid = SYSCTL_ADD_NODE(NULL,
                    SYSCTL_STATIC_CHILDREN(_debug_sleepq_chains), OID_AUTO,
                    chain_name, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
                    "sleepq chain stats");
                SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
                    "depth", CTLFLAG_RD, &sleepq_chains[i].sc_depth, 0, NULL);
                SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
                    "max_depth", CTLFLAG_RD, &sleepq_chains[i].sc_max_depth, 0,
                    NULL);
        }
}

SYSINIT(sleepqueue_profiling, SI_SUB_LOCK, SI_ORDER_ANY,
    init_sleepqueue_profiling, NULL);
#endif

/*
 * Early initialization of sleep queues that is called from the sleepinit()
 * SYSINIT.
 */
void
init_sleepqueues(void)
{
        int i;

        for (i = 0; i < SC_TABLESIZE; i++) {
                LIST_INIT(&sleepq_chains[i].sc_queues);
                mtx_init(&sleepq_chains[i].sc_lock, "sleepq chain", NULL,
                    MTX_SPIN);
        }
        sleepq_zone = uma_zcreate("SLEEPQUEUE", sizeof(struct sleepqueue),
#ifdef INVARIANTS
            NULL, sleepq_dtor, sleepq_init, NULL, UMA_ALIGN_CACHE, 0);
#else
            NULL, NULL, sleepq_init, NULL, UMA_ALIGN_CACHE, 0);
#endif

        thread0.td_sleepqueue = sleepq_alloc();
}

/*
 * Get a sleep queue for a new thread.
 */
struct sleepqueue *
sleepq_alloc(void)
{

        return (uma_zalloc(sleepq_zone, M_WAITOK));
}

/*
 * Free a sleep queue when a thread is destroyed.
 */
void
sleepq_free(struct sleepqueue *sq)
{

        uma_zfree(sleepq_zone, sq);
}

/*
 * Lock the sleep queue chain associated with the specified wait channel.
 */
void
sleepq_lock(const void *wchan)
{
        struct sleepqueue_chain *sc;

        sc = SC_LOOKUP(wchan);
        mtx_lock_spin(&sc->sc_lock);
}

/*
 * Look up the sleep queue associated with a given wait channel in the hash
 * table locking the associated sleep queue chain.  If no queue is found in
 * the table, NULL is returned.
 */
struct sleepqueue *
sleepq_lookup(const void *wchan)
{
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq;

        KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
        sc = SC_LOOKUP(wchan);
        mtx_assert(&sc->sc_lock, MA_OWNED);
        LIST_FOREACH(sq, &sc->sc_queues, sq_hash)
                if (sq->sq_wchan == wchan)
                        return (sq);
        return (NULL);
}

/*
 * Unlock the sleep queue chain associated with a given wait channel.
 */
void
sleepq_release(const void *wchan)
{
        struct sleepqueue_chain *sc;

        sc = SC_LOOKUP(wchan);
        mtx_unlock_spin(&sc->sc_lock);
}

/*
 * Places the current thread on the sleep queue for the specified wait
 * channel.  If INVARIANTS is enabled, then it associates the passed in
 * lock with the sleepq to make sure it is held when that sleep queue is
 * woken up.
 */
void
sleepq_add(const void *wchan, struct lock_object *lock, const char *wmesg,
    int flags, int queue)
{
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq;
        struct thread *td;

        td = curthread;
        sc = SC_LOOKUP(wchan);
        mtx_assert(&sc->sc_lock, MA_OWNED);
        MPASS(td->td_sleepqueue != NULL);
        MPASS(wchan != NULL);
        MPASS((queue >= 0) && (queue < NR_SLEEPQS));

        /* If this thread is not allowed to sleep, die a horrible death. */
        if (__predict_false(!THREAD_CAN_SLEEP())) {
#ifdef EPOCH_TRACE
                epoch_trace_list(curthread);
#endif
                KASSERT(0,
                    ("%s: td %p to sleep on wchan %p with sleeping prohibited",
                    __func__, td, wchan));
        }

        /* Look up the sleep queue associated with the wait channel 'wchan'. */
        sq = sleepq_lookup(wchan);

        /*
         * If the wait channel does not already have a sleep queue, use
         * this thread's sleep queue.  Otherwise, insert the current thread
         * into the sleep queue already in use by this wait channel.
         */
        if (sq == NULL) {
#ifdef INVARIANTS
                int i;

                sq = td->td_sleepqueue;
                for (i = 0; i < NR_SLEEPQS; i++) {
                        KASSERT(TAILQ_EMPTY(&sq->sq_blocked[i]),
                            ("thread's sleep queue %d is not empty", i));
                        KASSERT(sq->sq_blockedcnt[i] == 0,
                            ("thread's sleep queue %d count mismatches", i));
                }
                KASSERT(LIST_EMPTY(&sq->sq_free),
                    ("thread's sleep queue has a non-empty free list"));
                KASSERT(sq->sq_wchan == NULL, ("stale sq_wchan pointer"));
                sq->sq_lock = lock;
#endif
#ifdef SLEEPQUEUE_PROFILING
                sc->sc_depth++;
                if (sc->sc_depth > sc->sc_max_depth) {
                        sc->sc_max_depth = sc->sc_depth;
                        if (sc->sc_max_depth > sleepq_max_depth)
                                sleepq_max_depth = sc->sc_max_depth;
                }
#endif
                sq = td->td_sleepqueue;
                LIST_INSERT_HEAD(&sc->sc_queues, sq, sq_hash);
                sq->sq_wchan = wchan;
                sq->sq_type = flags & SLEEPQ_TYPE;
        } else {
                MPASS(wchan == sq->sq_wchan);
                MPASS(lock == sq->sq_lock);
                MPASS((flags & SLEEPQ_TYPE) == sq->sq_type);
                LIST_INSERT_HEAD(&sq->sq_free, td->td_sleepqueue, sq_hash);
        }
        thread_lock(td);
        TAILQ_INSERT_TAIL(&sq->sq_blocked[queue], td, td_slpq);
        sq->sq_blockedcnt[queue]++;
        td->td_sleepqueue = NULL;
        td->td_sqqueue = queue;
        td->td_wchan = wchan;
        td->td_wmesg = wmesg;
        if (flags & SLEEPQ_INTERRUPTIBLE) {
                td->td_intrval = 0;
                td->td_flags |= TDF_SINTR;
        }
        td->td_flags &= ~TDF_TIMEOUT;
        thread_unlock(td);
}

/*
 * Sets a timeout that will remove the current thread from the
 * specified sleep queue at the specified time if the thread has not
 * already been awakened.  Flags are from C_* (callout) namespace.
 */
void
sleepq_set_timeout_sbt(const void *wchan, sbintime_t sbt, sbintime_t pr,
    int flags)
{
        struct sleepqueue_chain *sc __unused;
        struct thread *td;
        sbintime_t pr1;

        td = curthread;
        sc = SC_LOOKUP(wchan);
        mtx_assert(&sc->sc_lock, MA_OWNED);
        MPASS(TD_ON_SLEEPQ(td));
        MPASS(td->td_sleepqueue == NULL);
        MPASS(wchan != NULL);
        if (cold && td == &thread0)
                panic("timed sleep before timers are working");
        KASSERT(td->td_sleeptimo == 0, ("td %d %p td_sleeptimo %jx",
            td->td_tid, td, (uintmax_t)td->td_sleeptimo));
        thread_lock(td);
        callout_when(sbt, pr, flags, &td->td_sleeptimo, &pr1);
        thread_unlock(td);
        callout_reset_sbt_on(&td->td_slpcallout, td->td_sleeptimo, pr1,
            sleepq_timeout, td, PCPU_GET(cpuid), flags | C_PRECALC |
            C_DIRECT_EXEC);
}

/*
 * Return the number of actual sleepers for the specified queue.
 */
u_int
sleepq_sleepcnt(const void *wchan, int queue)
{
        struct sleepqueue *sq;

        KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
        MPASS((queue >= 0) && (queue < NR_SLEEPQS));
        sq = sleepq_lookup(wchan);
        if (sq == NULL)
                return (0);
        return (sq->sq_blockedcnt[queue]);
}

static int
sleepq_check_ast_sc_locked(struct thread *td, struct sleepqueue_chain *sc)
{
        struct proc *p;
        int ret;

        mtx_assert(&sc->sc_lock, MA_OWNED);

        if ((td->td_pflags & TDP_WAKEUP) != 0) {
                td->td_pflags &= ~TDP_WAKEUP;
                thread_lock(td);
                return (EINTR);
        }

        /*
         * See if there are any pending signals or suspension requests for this
         * thread.  If not, we can switch immediately.
         */
        thread_lock(td);
        if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
                return (0);

        thread_unlock(td);
        mtx_unlock_spin(&sc->sc_lock);

        p = td->td_proc;
        CTR3(KTR_PROC, "sleepq catching signals: thread %p (pid %ld, %s)",
            (void *)td, (long)p->p_pid, td->td_name);
        PROC_LOCK(p);

        /*
         * Check for suspension first. Checking for signals and then
         * suspending could result in a missed signal, since a signal
         * can be delivered while this thread is suspended.
         */
        ret = sig_ast_checksusp(td);
        if (ret != 0) {
                PROC_UNLOCK(p);
                mtx_lock_spin(&sc->sc_lock);
                thread_lock(td);
                return (ret);
        }

        ret = sig_ast_needsigchk(td);

        /*
         * Lock the per-process spinlock prior to dropping the
         * PROC_LOCK to avoid a signal delivery race.
         * PROC_LOCK, PROC_SLOCK, and thread_lock() are
         * currently held in tdsendsignal() and thread_single().
         */
        PROC_SLOCK(p);
        mtx_lock_spin(&sc->sc_lock);
        PROC_UNLOCK(p);
        thread_lock(td);
        PROC_SUNLOCK(p);

        return (ret);
}

/*
 * Marks the pending sleep of the current thread as interruptible and
 * makes an initial check for pending signals before putting a thread
 * to sleep. Enters and exits with the thread lock held.  Thread lock
 * may have transitioned from the sleepq lock to a run lock.
 */
static int
sleepq_catch_signals(const void *wchan, int pri)
{
        struct thread *td;
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq;
        int ret;

        sc = SC_LOOKUP(wchan);
        mtx_assert(&sc->sc_lock, MA_OWNED);
        MPASS(wchan != NULL);
        td = curthread;

        ret = sleepq_check_ast_sc_locked(td, sc);
        THREAD_LOCK_ASSERT(td, MA_OWNED);
        mtx_assert(&sc->sc_lock, MA_OWNED);

        if (ret == 0) {
                /*
                 * No pending signals and no suspension requests found.
                 * Switch the thread off the cpu.
                 */
                sleepq_switch(wchan, pri);
        } else {
                /*
                 * There were pending signals and this thread is still
                 * on the sleep queue, remove it from the sleep queue.
                 */
                if (TD_ON_SLEEPQ(td)) {
                        sq = sleepq_lookup(wchan);
                        sleepq_remove_thread(sq, td);
                }
                MPASS(td->td_lock != &sc->sc_lock);
                mtx_unlock_spin(&sc->sc_lock);
                thread_unlock(td);
        }
        return (ret);
}

/*
 * Switches to another thread if we are still asleep on a sleep queue.
 * Returns with thread lock.
 */
static void
sleepq_switch(const void *wchan, int pri)
{
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq;
        struct thread *td;
        bool rtc_changed;

        td = curthread;
        sc = SC_LOOKUP(wchan);
        mtx_assert(&sc->sc_lock, MA_OWNED);
        THREAD_LOCK_ASSERT(td, MA_OWNED);

        /*
         * If we have a sleep queue, then we've already been woken up, so
         * just return.
         */
        if (td->td_sleepqueue != NULL) {
                mtx_unlock_spin(&sc->sc_lock);
                thread_unlock(td);
                return;
        }

        /*
         * If TDF_TIMEOUT is set, then our sleep has been timed out
         * already but we are still on the sleep queue, so dequeue the
         * thread and return.
         *
         * Do the same if the real-time clock has been adjusted since this
         * thread calculated its timeout based on that clock.  This handles
         * the following race:
         * - The Ts thread needs to sleep until an absolute real-clock time.
         *   It copies the global rtc_generation into curthread->td_rtcgen,
         *   reads the RTC, and calculates a sleep duration based on that time.
         *   See umtxq_sleep() for an example.
         * - The Tc thread adjusts the RTC, bumps rtc_generation, and wakes
         *   threads that are sleeping until an absolute real-clock time.
         *   See tc_setclock() and the POSIX specification of clock_settime().
         * - Ts reaches the code below.  It holds the sleepqueue chain lock,
         *   so Tc has finished waking, so this thread must test td_rtcgen.
         * (The declaration of td_rtcgen refers to this comment.)
         */
        rtc_changed = td->td_rtcgen != 0 && td->td_rtcgen != rtc_generation;
        if ((td->td_flags & TDF_TIMEOUT) || rtc_changed) {
                if (rtc_changed) {
                        td->td_rtcgen = 0;
                }
                MPASS(TD_ON_SLEEPQ(td));
                sq = sleepq_lookup(wchan);
                sleepq_remove_thread(sq, td);
                mtx_unlock_spin(&sc->sc_lock);
                thread_unlock(td);
                return;
        }
#ifdef SLEEPQUEUE_PROFILING
        if (prof_enabled)
                sleepq_profile(td->td_wmesg);
#endif
        MPASS(td->td_sleepqueue == NULL);
        sched_sleep(td, pri);
        thread_lock_set(td, &sc->sc_lock);
        SDT_PROBE0(sched, , , sleep);
        TD_SET_SLEEPING(td);
        mi_switch(SW_VOL | SWT_SLEEPQ);
        KASSERT(TD_IS_RUNNING(td), ("running but not TDS_RUNNING"));
        CTR3(KTR_PROC, "sleepq resume: thread %p (pid %ld, %s)",
            (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name);
}

/*
 * Check to see if we timed out.
 */
static inline int
sleepq_check_timeout(void)
{
        struct thread *td;
        int res;

        res = 0;
        td = curthread;
        if (td->td_sleeptimo != 0) {
                if (td->td_sleeptimo <= sbinuptime())
                        res = EWOULDBLOCK;
                td->td_sleeptimo = 0;
        }
        return (res);
}

/*
 * Check to see if we were awoken by a signal.
 */
static inline int
sleepq_check_signals(void)
{
        struct thread *td;

        td = curthread;
        KASSERT((td->td_flags & TDF_SINTR) == 0,
            ("thread %p still in interruptible sleep?", td));

        return (td->td_intrval);
}

/*
 * Block the current thread until it is awakened from its sleep queue.
 */
void
sleepq_wait(const void *wchan, int pri)
{
        struct thread *td;

        td = curthread;
        MPASS(!(td->td_flags & TDF_SINTR));
        thread_lock(td);
        sleepq_switch(wchan, pri);
}

/*
 * Block the current thread until it is awakened from its sleep queue
 * or it is interrupted by a signal.
 */
int
sleepq_wait_sig(const void *wchan, int pri)
{
        int rcatch;

        rcatch = sleepq_catch_signals(wchan, pri);
        if (rcatch)
                return (rcatch);
        return (sleepq_check_signals());
}

/*
 * Block the current thread until it is awakened from its sleep queue
 * or it times out while waiting.
 */
int
sleepq_timedwait(const void *wchan, int pri)
{
        struct thread *td;

        td = curthread;
        MPASS(!(td->td_flags & TDF_SINTR));

        thread_lock(td);
        sleepq_switch(wchan, pri);

        return (sleepq_check_timeout());
}

/*
 * Block the current thread until it is awakened from its sleep queue,
 * it is interrupted by a signal, or it times out waiting to be awakened.
 */
int
sleepq_timedwait_sig(const void *wchan, int pri)
{
        int rcatch, rvalt, rvals;

        rcatch = sleepq_catch_signals(wchan, pri);
        /* We must always call check_timeout() to clear sleeptimo. */
        rvalt = sleepq_check_timeout();
        rvals = sleepq_check_signals();
        if (rcatch)
                return (rcatch);
        if (rvals)
                return (rvals);
        return (rvalt);
}

/*
 * Returns the type of sleepqueue given a waitchannel.
 */
int
sleepq_type(const void *wchan)
{
        struct sleepqueue *sq;
        int type;

        MPASS(wchan != NULL);

        sq = sleepq_lookup(wchan);
        if (sq == NULL)
                return (-1);
        type = sq->sq_type;

        return (type);
}

/*
 * Removes a thread from a sleep queue and makes it
 * runnable.
 *
 * Requires the sc chain locked on entry.  If SRQ_HOLD is specified it will
 * be locked on return.  Returns without the thread lock held.
 */
static int
sleepq_resume_thread(struct sleepqueue *sq, struct thread *td, int pri,
    int srqflags)
{
        struct sleepqueue_chain *sc;
        bool drop;

        MPASS(td != NULL);
        MPASS(sq->sq_wchan != NULL);
        MPASS(td->td_wchan == sq->sq_wchan);

        sc = SC_LOOKUP(sq->sq_wchan);
        mtx_assert(&sc->sc_lock, MA_OWNED);

        /*
         * Avoid recursing on the chain lock.  If the locks don't match we
         * need to acquire the thread lock which setrunnable will drop for
         * us.  In this case we need to drop the chain lock afterwards.
         *
         * There is no race that will make td_lock equal to sc_lock because
         * we hold sc_lock.
         */
        drop = false;
        if (!TD_IS_SLEEPING(td)) {
                thread_lock(td);
                drop = true;
        } else
                thread_lock_block_wait(td);

        /* Remove thread from the sleepq. */
        sleepq_remove_thread(sq, td);

        /* If we're done with the sleepqueue release it. */
        if ((srqflags & SRQ_HOLD) == 0 && drop)
                mtx_unlock_spin(&sc->sc_lock);

        /* Adjust priority if requested. */
        MPASS(pri == 0 || (pri >= PRI_MIN && pri <= PRI_MAX));
        if (pri != 0 && td->td_priority > pri &&
            PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
                sched_prio(td, pri);

        /*
         * Note that thread td might not be sleeping if it is running
         * sleepq_catch_signals() on another CPU or is blocked on its
         * proc lock to check signals.  There's no need to mark the
         * thread runnable in that case.
         */
        if (TD_IS_SLEEPING(td)) {
                MPASS(!drop);
                TD_CLR_SLEEPING(td);
                return (setrunnable(td, srqflags));
        }
        MPASS(drop);
        thread_unlock(td);

        return (0);
}

static void
sleepq_remove_thread(struct sleepqueue *sq, struct thread *td)
{
        struct sleepqueue_chain *sc __unused;

        MPASS(td != NULL);
        MPASS(sq->sq_wchan != NULL);
        MPASS(td->td_wchan == sq->sq_wchan);
        MPASS(td->td_sqqueue < NR_SLEEPQS && td->td_sqqueue >= 0);
        THREAD_LOCK_ASSERT(td, MA_OWNED);
        sc = SC_LOOKUP(sq->sq_wchan);
        mtx_assert(&sc->sc_lock, MA_OWNED);

        SDT_PROBE2(sched, , , wakeup, td, td->td_proc);

        /* Remove the thread from the queue. */
        sq->sq_blockedcnt[td->td_sqqueue]--;
        TAILQ_REMOVE(&sq->sq_blocked[td->td_sqqueue], td, td_slpq);

        /*
         * Get a sleep queue for this thread.  If this is the last waiter,
         * use the queue itself and take it out of the chain, otherwise,
         * remove a queue from the free list.
         */
        if (LIST_EMPTY(&sq->sq_free)) {
                td->td_sleepqueue = sq;
#ifdef INVARIANTS
                sq->sq_wchan = NULL;
#endif
#ifdef SLEEPQUEUE_PROFILING
                sc->sc_depth--;
#endif
        } else
                td->td_sleepqueue = LIST_FIRST(&sq->sq_free);
        LIST_REMOVE(td->td_sleepqueue, sq_hash);

        if ((td->td_flags & TDF_TIMEOUT) == 0 && td->td_sleeptimo != 0 &&
            td->td_lock == &sc->sc_lock) {
                /*
                 * We ignore the situation where timeout subsystem was
                 * unable to stop our callout.  The struct thread is
                 * type-stable, the callout will use the correct
                 * memory when running.  The checks of the
                 * td_sleeptimo value in this function and in
                 * sleepq_timeout() ensure that the thread does not
                 * get spurious wakeups, even if the callout was reset
                 * or thread reused.
                 *
                 * We also cannot safely stop the callout if a scheduler
                 * lock is held since softclock_thread() forces a lock
                 * order of callout lock -> scheduler lock.  The thread
                 * lock will be a scheduler lock only if the thread is
                 * preparing to go to sleep, so this is hopefully a rare
                 * scenario.
                 */
                callout_stop(&td->td_slpcallout);
        }

        td->td_wmesg = NULL;
        td->td_wchan = NULL;
        td->td_flags &= ~(TDF_SINTR | TDF_TIMEOUT);

        CTR3(KTR_PROC, "sleepq_wakeup: thread %p (pid %ld, %s)",
            (void *)td, (long)td->td_proc->p_pid, td->td_name);
}

void
sleepq_remove_nested(struct thread *td)
{
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq;
        const void *wchan;

        MPASS(TD_ON_SLEEPQ(td));

        wchan = td->td_wchan;
        sc = SC_LOOKUP(wchan);
        mtx_lock_spin(&sc->sc_lock);
        sq = sleepq_lookup(wchan);
        MPASS(sq != NULL);
        thread_lock(td);
        sleepq_remove_thread(sq, td);
        mtx_unlock_spin(&sc->sc_lock);
        /* Returns with the thread lock owned. */
}

#ifdef INVARIANTS
/*
 * UMA zone item deallocator.
 */
static void
sleepq_dtor(void *mem, int size, void *arg)
{
        struct sleepqueue *sq;
        int i;

        sq = mem;
        for (i = 0; i < NR_SLEEPQS; i++) {
                MPASS(TAILQ_EMPTY(&sq->sq_blocked[i]));
                MPASS(sq->sq_blockedcnt[i] == 0);
        }
}
#endif

/*
 * UMA zone item initializer.
 */
static int
sleepq_init(void *mem, int size, int flags)
{
        struct sleepqueue *sq;
        int i;

        bzero(mem, size);
        sq = mem;
        for (i = 0; i < NR_SLEEPQS; i++) {
                TAILQ_INIT(&sq->sq_blocked[i]);
                sq->sq_blockedcnt[i] = 0;
        }
        LIST_INIT(&sq->sq_free);
        return (0);
}

/*
 * Find thread sleeping on a wait channel and resume it.
 */
int
sleepq_signal(const void *wchan, int flags, int pri, int queue)
{
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq;
        struct threadqueue *head;
        struct thread *td, *besttd;
        int wakeup_swapper;

        CTR2(KTR_PROC, "sleepq_signal(%p, %d)", wchan, flags);
        KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
        MPASS((queue >= 0) && (queue < NR_SLEEPQS));
        sq = sleepq_lookup(wchan);
        if (sq == NULL) {
                if (flags & SLEEPQ_DROP)
                        sleepq_release(wchan);
                return (0);
        }
        KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE),
            ("%s: mismatch between sleep/wakeup and cv_*", __func__));

        head = &sq->sq_blocked[queue];
        if (flags & SLEEPQ_UNFAIR) {
                /*
                 * Find the most recently sleeping thread, but try to
                 * skip threads still in process of context switch to
                 * avoid spinning on the thread lock.
                 */
                sc = SC_LOOKUP(wchan);
                besttd = TAILQ_LAST_FAST(head, thread, td_slpq);
                while (besttd->td_lock != &sc->sc_lock) {
                        td = TAILQ_PREV_FAST(besttd, head, thread, td_slpq);
                        if (td == NULL)
                                break;
                        besttd = td;
                }
        } else {
                /*
                 * Find the highest priority thread on the queue.  If there
                 * is a tie, use the thread that first appears in the queue
                 * as it has been sleeping the longest since threads are
                 * always added to the tail of sleep queues.
                 */
                besttd = td = TAILQ_FIRST(head);
                while ((td = TAILQ_NEXT(td, td_slpq)) != NULL) {
                        if (td->td_priority < besttd->td_priority)
                                besttd = td;
                }
        }
        MPASS(besttd != NULL);
        wakeup_swapper = sleepq_resume_thread(sq, besttd, pri,
            (flags & SLEEPQ_DROP) ? 0 : SRQ_HOLD);
        return (wakeup_swapper);
}

static bool
match_any(struct thread *td __unused)
{

        return (true);
}

/*
 * Resume all threads sleeping on a specified wait channel.
 */
int
sleepq_broadcast(const void *wchan, int flags, int pri, int queue)
{
        struct sleepqueue *sq;

        CTR2(KTR_PROC, "sleepq_broadcast(%p, %d)", wchan, flags);
        KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
        MPASS((queue >= 0) && (queue < NR_SLEEPQS));
        sq = sleepq_lookup(wchan);
        if (sq == NULL)
                return (0);
        KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE),
            ("%s: mismatch between sleep/wakeup and cv_*", __func__));

        return (sleepq_remove_matching(sq, queue, match_any, pri));
}

/*
 * Resume threads on the sleep queue that match the given predicate.
 */
int
sleepq_remove_matching(struct sleepqueue *sq, int queue,
    bool (*matches)(struct thread *), int pri)
{
        struct thread *td, *tdn;
        int wakeup_swapper;

        /*
         * The last thread will be given ownership of sq and may
         * re-enqueue itself before sleepq_resume_thread() returns,
         * so we must cache the "next" queue item at the beginning
         * of the final iteration.
         */
        wakeup_swapper = 0;
        TAILQ_FOREACH_SAFE(td, &sq->sq_blocked[queue], td_slpq, tdn) {
                if (matches(td))
                        wakeup_swapper |= sleepq_resume_thread(sq, td, pri,
                            SRQ_HOLD);
        }

        return (wakeup_swapper);
}

/*
 * Time sleeping threads out.  When the timeout expires, the thread is
 * removed from the sleep queue and made runnable if it is still asleep.
 */
static void
sleepq_timeout(void *arg)
{
        struct sleepqueue_chain *sc __unused;
        struct sleepqueue *sq;
        struct thread *td;
        const void *wchan;
        int wakeup_swapper;

        td = arg;
        CTR3(KTR_PROC, "sleepq_timeout: thread %p (pid %ld, %s)",
            (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name);

        thread_lock(td);
        if (td->td_sleeptimo == 0 ||
            td->td_sleeptimo > td->td_slpcallout.c_time) {
                /*
                 * The thread does not want a timeout (yet).
                 */
        } else if (TD_IS_SLEEPING(td) && TD_ON_SLEEPQ(td)) {
                /*
                 * See if the thread is asleep and get the wait
                 * channel if it is.
                 */
                wchan = td->td_wchan;
                sc = SC_LOOKUP(wchan);
                THREAD_LOCKPTR_ASSERT(td, &sc->sc_lock);
                sq = sleepq_lookup(wchan);
                MPASS(sq != NULL);
                td->td_flags |= TDF_TIMEOUT;
                wakeup_swapper = sleepq_resume_thread(sq, td, 0, 0);
                if (wakeup_swapper)
                        kick_proc0();
                return;
        } else if (TD_ON_SLEEPQ(td)) {
                /*
                 * If the thread is on the SLEEPQ but isn't sleeping
                 * yet, it can either be on another CPU in between
                 * sleepq_add() and one of the sleepq_*wait*()
                 * routines or it can be in sleepq_catch_signals().
                 */
                td->td_flags |= TDF_TIMEOUT;
        }
        thread_unlock(td);
}

/*
 * Resumes a specific thread from the sleep queue associated with a specific
 * wait channel if it is on that queue.
 */
void
sleepq_remove(struct thread *td, const void *wchan)
{
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq;
        int wakeup_swapper;

        /*
         * Look up the sleep queue for this wait channel, then re-check
         * that the thread is asleep on that channel, if it is not, then
         * bail.
         */
        MPASS(wchan != NULL);
        sc = SC_LOOKUP(wchan);
        mtx_lock_spin(&sc->sc_lock);
        /*
         * We can not lock the thread here as it may be sleeping on a
         * different sleepq.  However, holding the sleepq lock for this
         * wchan can guarantee that we do not miss a wakeup for this
         * channel.  The asserts below will catch any false positives.
         */
        if (!TD_ON_SLEEPQ(td) || td->td_wchan != wchan) {
                mtx_unlock_spin(&sc->sc_lock);
                return;
        }

        /* Thread is asleep on sleep queue sq, so wake it up. */
        sq = sleepq_lookup(wchan);
        MPASS(sq != NULL);
        MPASS(td->td_wchan == wchan);
        wakeup_swapper = sleepq_resume_thread(sq, td, 0, 0);
        if (wakeup_swapper)
                kick_proc0();
}

/*
 * Abort a thread as if an interrupt had occurred.  Only abort
 * interruptible waits (unfortunately it isn't safe to abort others).
 *
 * Requires thread lock on entry, releases on return.
 */
int
sleepq_abort(struct thread *td, int intrval)
{
        struct sleepqueue *sq;
        const void *wchan;

        THREAD_LOCK_ASSERT(td, MA_OWNED);
        MPASS(TD_ON_SLEEPQ(td));
        MPASS(td->td_flags & TDF_SINTR);
        MPASS((intrval == 0 && (td->td_flags & TDF_SIGWAIT) != 0) ||
            intrval == EINTR || intrval == ERESTART);

        /*
         * If the TDF_TIMEOUT flag is set, just leave. A
         * timeout is scheduled anyhow.
         */
        if (td->td_flags & TDF_TIMEOUT) {
                thread_unlock(td);
                return (0);
        }

        CTR3(KTR_PROC, "sleepq_abort: thread %p (pid %ld, %s)",
            (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name);
        td->td_intrval = intrval;

        /*
         * If the thread has not slept yet it will find the signal in
         * sleepq_catch_signals() and call sleepq_resume_thread.  Otherwise
         * we have to do it here.
         */
        if (!TD_IS_SLEEPING(td)) {
                thread_unlock(td);
                return (0);
        }
        wchan = td->td_wchan;
        MPASS(wchan != NULL);
        sq = sleepq_lookup(wchan);
        MPASS(sq != NULL);

        /* Thread is asleep on sleep queue sq, so wake it up. */
        return (sleepq_resume_thread(sq, td, 0, 0));
}

void
sleepq_chains_remove_matching(bool (*matches)(struct thread *))
{
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq, *sq1;
        int i, wakeup_swapper;

        wakeup_swapper = 0;
        for (sc = &sleepq_chains[0]; sc < sleepq_chains + SC_TABLESIZE; ++sc) {
                if (LIST_EMPTY(&sc->sc_queues)) {
                        continue;
                }
                mtx_lock_spin(&sc->sc_lock);
                LIST_FOREACH_SAFE(sq, &sc->sc_queues, sq_hash, sq1) {
                        for (i = 0; i < NR_SLEEPQS; ++i) {
                                wakeup_swapper |= sleepq_remove_matching(sq, i,
                                    matches, 0);
                        }
                }
                mtx_unlock_spin(&sc->sc_lock);
        }
        if (wakeup_swapper) {
                kick_proc0();
        }
}

/*
 * Prints the stacks of all threads presently sleeping on wchan/queue to
 * the sbuf sb.  Sets count_stacks_printed to the number of stacks actually
 * printed.  Typically, this will equal the number of threads sleeping on the
 * queue, but may be less if sb overflowed before all stacks were printed.
 */
#ifdef STACK
int
sleepq_sbuf_print_stacks(struct sbuf *sb, const void *wchan, int queue,
    int *count_stacks_printed)
{
        struct thread *td, *td_next;
        struct sleepqueue *sq;
        struct stack **st;
        struct sbuf **td_infos;
        int i, stack_idx, error, stacks_to_allocate;
        bool finished;

        error = 0;
        finished = false;

        KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
        MPASS((queue >= 0) && (queue < NR_SLEEPQS));

        stacks_to_allocate = 10;
        for (i = 0; i < 3 && !finished ; i++) {
                /* We cannot malloc while holding the queue's spinlock, so
                 * we do our mallocs now, and hope it is enough.  If it
                 * isn't, we will free these, drop the lock, malloc more,
                 * and try again, up to a point.  After that point we will
                 * give up and report ENOMEM. We also cannot write to sb
                 * during this time since the client may have set the
                 * SBUF_AUTOEXTEND flag on their sbuf, which could cause a
                 * malloc as we print to it.  So we defer actually printing
                 * to sb until after we drop the spinlock.
                 */

                /* Where we will store the stacks. */
                st = malloc(sizeof(struct stack *) * stacks_to_allocate,
                    M_TEMP, M_WAITOK);
                for (stack_idx = 0; stack_idx < stacks_to_allocate;
                    stack_idx++)
                        st[stack_idx] = stack_create(M_WAITOK);

                /* Where we will store the td name, tid, etc. */
                td_infos = malloc(sizeof(struct sbuf *) * stacks_to_allocate,
                    M_TEMP, M_WAITOK);
                for (stack_idx = 0; stack_idx < stacks_to_allocate;
                    stack_idx++)
                        td_infos[stack_idx] = sbuf_new(NULL, NULL,
                            MAXCOMLEN + sizeof(struct thread *) * 2 + 40,
                            SBUF_FIXEDLEN);

                sleepq_lock(wchan);
                sq = sleepq_lookup(wchan);
                if (sq == NULL) {
                        /* This sleepq does not exist; exit and return ENOENT. */
                        error = ENOENT;
                        finished = true;
                        sleepq_release(wchan);
                        goto loop_end;
                }

                stack_idx = 0;
                /* Save thread info */
                TAILQ_FOREACH_SAFE(td, &sq->sq_blocked[queue], td_slpq,
                    td_next) {
                        if (stack_idx >= stacks_to_allocate)
                                goto loop_end;

                        /* Note the td_lock is equal to the sleepq_lock here. */
                        (void)stack_save_td(st[stack_idx], td);

                        sbuf_printf(td_infos[stack_idx], "%d: %s %p",
                            td->td_tid, td->td_name, td);

                        ++stack_idx;
                }

                finished = true;
                sleepq_release(wchan);

                /* Print the stacks */
                for (i = 0; i < stack_idx; i++) {
                        sbuf_finish(td_infos[i]);
                        sbuf_printf(sb, "--- thread %s: ---\n", sbuf_data(td_infos[i]));
                        stack_sbuf_print(sb, st[i]);
                        sbuf_putc(sb, '\n');

                        error = sbuf_error(sb);
                        if (error == 0)
                                *count_stacks_printed = stack_idx;
                }

loop_end:
                if (!finished)
                        sleepq_release(wchan);
                for (stack_idx = 0; stack_idx < stacks_to_allocate;
                    stack_idx++)
                        stack_destroy(st[stack_idx]);
                for (stack_idx = 0; stack_idx < stacks_to_allocate;
                    stack_idx++)
                        sbuf_delete(td_infos[stack_idx]);
                free(st, M_TEMP);
                free(td_infos, M_TEMP);
                stacks_to_allocate *= 10;
        }

        if (!finished && error == 0)
                error = ENOMEM;

        return (error);
}
#endif

#ifdef SLEEPQUEUE_PROFILING
#define SLEEPQ_PROF_LOCATIONS   1024
#define SLEEPQ_SBUFSIZE         512
struct sleepq_prof {
        LIST_ENTRY(sleepq_prof) sp_link;
        const char      *sp_wmesg;
        long            sp_count;
};

LIST_HEAD(sqphead, sleepq_prof);

struct sqphead sleepq_prof_free;
struct sqphead sleepq_hash[SC_TABLESIZE];
static struct sleepq_prof sleepq_profent[SLEEPQ_PROF_LOCATIONS];
static struct mtx sleepq_prof_lock;
MTX_SYSINIT(sleepq_prof_lock, &sleepq_prof_lock, "sleepq_prof", MTX_SPIN);

static void
sleepq_profile(const char *wmesg)
{
        struct sleepq_prof *sp;

        mtx_lock_spin(&sleepq_prof_lock);
        if (prof_enabled == 0)
                goto unlock;
        LIST_FOREACH(sp, &sleepq_hash[SC_HASH(wmesg)], sp_link)
                if (sp->sp_wmesg == wmesg)
                        goto done;
        sp = LIST_FIRST(&sleepq_prof_free);
        if (sp == NULL)
                goto unlock;
        sp->sp_wmesg = wmesg;
        LIST_REMOVE(sp, sp_link);
        LIST_INSERT_HEAD(&sleepq_hash[SC_HASH(wmesg)], sp, sp_link);
done:
        sp->sp_count++;
unlock:
        mtx_unlock_spin(&sleepq_prof_lock);
        return;
}

static void
sleepq_prof_reset(void)
{
        struct sleepq_prof *sp;
        int enabled;
        int i;

        mtx_lock_spin(&sleepq_prof_lock);
        enabled = prof_enabled;
        prof_enabled = 0;
        for (i = 0; i < SC_TABLESIZE; i++)
                LIST_INIT(&sleepq_hash[i]);
        LIST_INIT(&sleepq_prof_free);
        for (i = 0; i < SLEEPQ_PROF_LOCATIONS; i++) {
                sp = &sleepq_profent[i];
                sp->sp_wmesg = NULL;
                sp->sp_count = 0;
                LIST_INSERT_HEAD(&sleepq_prof_free, sp, sp_link);
        }
        prof_enabled = enabled;
        mtx_unlock_spin(&sleepq_prof_lock);
}

static int
enable_sleepq_prof(SYSCTL_HANDLER_ARGS)
{
        int error, v;

        v = prof_enabled;
        error = sysctl_handle_int(oidp, &v, v, req);
        if (error)
                return (error);
        if (req->newptr == NULL)
                return (error);
        if (v == prof_enabled)
                return (0);
        if (v == 1)
                sleepq_prof_reset();
        mtx_lock_spin(&sleepq_prof_lock);
        prof_enabled = !!v;
        mtx_unlock_spin(&sleepq_prof_lock);

        return (0);
}

static int
reset_sleepq_prof_stats(SYSCTL_HANDLER_ARGS)
{
        int error, v;

        v = 0;
        error = sysctl_handle_int(oidp, &v, 0, req);
        if (error)
                return (error);
        if (req->newptr == NULL)
                return (error);
        if (v == 0)
                return (0);
        sleepq_prof_reset();

        return (0);
}

static int
dump_sleepq_prof_stats(SYSCTL_HANDLER_ARGS)
{
        struct sleepq_prof *sp;
        struct sbuf *sb;
        int enabled;
        int error;
        int i;

        error = sysctl_wire_old_buffer(req, 0);
        if (error != 0)
                return (error);
        sb = sbuf_new_for_sysctl(NULL, NULL, SLEEPQ_SBUFSIZE, req);
        sbuf_cat(sb, "\nwmesg\tcount\n");
        enabled = prof_enabled;
        mtx_lock_spin(&sleepq_prof_lock);
        prof_enabled = 0;
        mtx_unlock_spin(&sleepq_prof_lock);
        for (i = 0; i < SC_TABLESIZE; i++) {
                LIST_FOREACH(sp, &sleepq_hash[i], sp_link) {
                        sbuf_printf(sb, "%s\t%ld\n",
                            sp->sp_wmesg, sp->sp_count);
                }
        }
        mtx_lock_spin(&sleepq_prof_lock);
        prof_enabled = enabled;
        mtx_unlock_spin(&sleepq_prof_lock);

        error = sbuf_finish(sb);
        sbuf_delete(sb);
        return (error);
}

SYSCTL_PROC(_debug_sleepq, OID_AUTO, stats,
    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0,
    dump_sleepq_prof_stats, "A",
    "Sleepqueue profiling statistics");
SYSCTL_PROC(_debug_sleepq, OID_AUTO, reset,
    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0,
    reset_sleepq_prof_stats, "I",
    "Reset sleepqueue profiling statistics");
SYSCTL_PROC(_debug_sleepq, OID_AUTO, enable,
    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0,
    enable_sleepq_prof, "I",
    "Enable sleepqueue profiling");
#endif

#ifdef DDB
DB_SHOW_COMMAND(sleepq, db_show_sleepqueue)
{
        struct sleepqueue_chain *sc;
        struct sleepqueue *sq;
#ifdef INVARIANTS
        struct lock_object *lock;
#endif
        struct thread *td;
        void *wchan;
        int i;

        if (!have_addr)
                return;

        /*
         * First, see if there is an active sleep queue for the wait channel
         * indicated by the address.
         */
        wchan = (void *)addr;
        sc = SC_LOOKUP(wchan);
        LIST_FOREACH(sq, &sc->sc_queues, sq_hash)
                if (sq->sq_wchan == wchan)
                        goto found;

        /*
         * Second, see if there is an active sleep queue at the address
         * indicated.
         */
        for (i = 0; i < SC_TABLESIZE; i++)
                LIST_FOREACH(sq, &sleepq_chains[i].sc_queues, sq_hash) {
                        if (sq == (struct sleepqueue *)addr)
                                goto found;
                }

        db_printf("Unable to locate a sleep queue via %p\n", (void *)addr);
        return;
found:
        db_printf("Wait channel: %p\n", sq->sq_wchan);
        db_printf("Queue type: %d\n", sq->sq_type);
#ifdef INVARIANTS
        if (sq->sq_lock) {
                lock = sq->sq_lock;
                db_printf("Associated Interlock: %p - (%s) %s\n", lock,
                    LOCK_CLASS(lock)->lc_name, lock->lo_name);
        }
#endif
        db_printf("Blocked threads:\n");
        for (i = 0; i < NR_SLEEPQS; i++) {
                db_printf("\nQueue[%d]:\n", i);
                if (TAILQ_EMPTY(&sq->sq_blocked[i]))
                        db_printf("\tempty\n");
                else
                        TAILQ_FOREACH(td, &sq->sq_blocked[i],
                                      td_slpq) {
                                db_printf("\t%p (tid %d, pid %d, \"%s\")\n", td,
                                          td->td_tid, td->td_proc->p_pid,
                                          td->td_name);
                        }
                db_printf("(expected: %u)\n", sq->sq_blockedcnt[i]);
        }
}

/* Alias 'show sleepqueue' to 'show sleepq'. */
DB_SHOW_ALIAS(sleepqueue, db_show_sleepqueue);
#endif