/*
 * The page_retire_thread loops forever, looking to see if there are
 * pages still waiting to be retired.
 */
static void
page_retire_thread(void)
{
	callb_cpr_t c;

	CALLB_CPR_INIT(&c, &pr_thread_mutex, callb_generic_cpr, "page_retire");

	mutex_enter(&pr_thread_mutex);
	for (;;) {
		if (pr_enable && PR_KSTAT_PENDING) {
			/*
			 * Sigh. It's SO broken how we have to try to shake
			 * loose the holder of the page. Since we have no
			 * idea who or what has it locked, we go bang on
			 * every door in the city to try to locate it.
			 */
			kmem_reap();
			seg_preap();
			page_retire_hunt(page_retire_thread_cb);
			CALLB_CPR_SAFE_BEGIN(&c);
			(void) cv_timedwait(&pr_cv, &pr_thread_mutex,
			    lbolt + pr_thread_shortwait);
			CALLB_CPR_SAFE_END(&c, &pr_thread_mutex);
		} else {
			CALLB_CPR_SAFE_BEGIN(&c);
			(void) cv_timedwait(&pr_cv, &pr_thread_mutex,
			    lbolt + pr_thread_longwait);
			CALLB_CPR_SAFE_END(&c, &pr_thread_mutex);
		}
	}
	/*NOTREACHED*/
}

/*
 * Delay this thread by 'ticks' if we are still in the open transaction
 * group and there is already a waiting txg quiesing or quiesced.  Abort
 * the delay if this txg stalls or enters the quiesing state.
 */
void
txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks)
{
	tx_state_t *tx = &dp->dp_tx;
	clock_t timeout = ddi_get_lbolt() + ticks;

	/* don't delay if this txg could transition to quiesing immediately */
	if (tx->tx_open_txg > txg ||
	    tx->tx_syncing_txg == txg-1 || tx->tx_synced_txg == txg-1)
		return;

	mutex_enter(&tx->tx_sync_lock);
	if (tx->tx_open_txg > txg || tx->tx_synced_txg == txg-1) {
		mutex_exit(&tx->tx_sync_lock);
		return;
	}

	while (ddi_get_lbolt() < timeout &&
	    tx->tx_syncing_txg < txg-1 && !txg_stalled(dp))
		(void) cv_timedwait(&tx->tx_quiesce_more_cv, &tx->tx_sync_lock,
		    timeout);

	DMU_TX_STAT_BUMP(dmu_tx_delay);

	mutex_exit(&tx->tx_sync_lock);
}

int
rumpuser_cv_timedwait(struct rumpuser_cv *cv, struct rumpuser_mtx *mtx,
	int64_t sec, int64_t nsec)
{

	return cv_timedwait(cv, mtx, sec, nsec);
}

int
testcall(struct lwp *l, void *uap, register_t *retval)
{
	int i;

	mutex_init(&test_mutex, MUTEX_DEFAULT, IPL_NONE);
	cv_init(&test_cv, "testcv");

	printf("test: creating threads\n");

	test_count = NTHREADS;
	test_exit = 0;

	for (i = 0; i < test_count; i++)
		kthread_create(0, KTHREAD_MPSAFE, NULL, thread1, &primes[i],
		    &test_threads[i], "thread%d", i);

	printf("test: sleeping\n");

	mutex_enter(&test_mutex);
	while (test_count != 0) {
		(void)cv_timedwait(&test_cv, &test_mutex, hz * SECONDS);
		test_exit = 1;
	}
	mutex_exit(&test_mutex);

	printf("test: finished\n");

	cv_destroy(&test_cv);
	mutex_destroy(&test_mutex);

	return 0;
}

int
_sema_timedwait(struct sema *sema, int timo, const char *file, int line)
{
	int ret, timed_out;

	mtx_lock(&sema->sema_mtx);

	/*
	 * A spurious wakeup will cause the timeout interval to start over.
	 * This isn't a big deal as long as spurious wakeups don't occur
	 * continuously, since the timeout period is merely a lower bound on how
	 * long to wait.
	 */
	for (timed_out = 0; sema->sema_value == 0 && timed_out == 0;) {
		sema->sema_waiters++;
		timed_out = cv_timedwait(&sema->sema_cv, &sema->sema_mtx, timo);
		sema->sema_waiters--;
	}
	if (sema->sema_value > 0) {
		/* Success. */
		sema->sema_value--;
		ret = 1;

		CTR6(KTR_LOCK, "%s(%p) \"%s\" v = %d at %s:%d", __func__, sema,
		    cv_wmesg(&sema->sema_cv), sema->sema_value, file, line);
	} else {
		ret = 0;
		
		CTR5(KTR_LOCK, "%s(%p) \"%s\" fail at %s:%d", __func__, sema,
		    cv_wmesg(&sema->sema_cv), file, line);
	}

	mtx_unlock(&sema->sema_mtx);
	return (ret);
}

/*
 * Thread to setup switching mode. This thread is created during vsw_attach()
 * initially. It invokes vsw_setup_switching() and keeps retrying while the
 * returned value is EAGAIN. The thread exits when the switching mode setup is
 * done successfully or when the error returned is not EAGAIN. This thread may
 * also get created from vsw_update_md_prop() if the switching mode needs to be
 * updated.
 */
void
vsw_setup_switching_thread(void *arg)
{
	callb_cpr_t	cprinfo;
	vsw_t		*vswp =  (vsw_t *)arg;
	clock_t		wait_time;
	clock_t		xwait;
	clock_t		wait_rv;
	int		rv;

	/* wait time used on successive retries */
	xwait = drv_usectohz(vsw_setup_switching_delay * MICROSEC);

	CALLB_CPR_INIT(&cprinfo, &vswp->sw_thr_lock, callb_generic_cpr,
	    "vsw_setup_sw_thread");

	mutex_enter(&vswp->sw_thr_lock);

	while ((vswp->sw_thr_flags & VSW_SWTHR_STOP) == 0) {

		CALLB_CPR_SAFE_BEGIN(&cprinfo);

		/* Wait for sometime before (re)trying setup_switching() */
		wait_time = ddi_get_lbolt() + xwait;
		while ((vswp->sw_thr_flags & VSW_SWTHR_STOP) == 0) {
			wait_rv = cv_timedwait(&vswp->sw_thr_cv,
			    &vswp->sw_thr_lock, wait_time);
			if (wait_rv == -1) {	/* timed out */
				break;
			}
		}

		CALLB_CPR_SAFE_END(&cprinfo, &vswp->sw_thr_lock)

		if ((vswp->sw_thr_flags & VSW_SWTHR_STOP) != 0) {
			/*
			 * If there is a stop request, process that first and
			 * exit the loop. Continue to hold the mutex which gets
			 * released in CALLB_CPR_EXIT().
			 */
			break;
		}

		mutex_exit(&vswp->sw_thr_lock);
		rv = vsw_setup_switching(vswp);
		if (rv == 0) {
			vsw_setup_switching_post_process(vswp);
		}
		mutex_enter(&vswp->sw_thr_lock);
		if (rv != EAGAIN) {
			break;
		}

	}

	vswp->sw_thr_flags &= ~VSW_SWTHR_STOP;
	vswp->sw_thread = NULL;
	CALLB_CPR_EXIT(&cprinfo);
	thread_exit();
}

/* afs_osi_TimedSleep
 * 
 * Arguments:
 * event - event to sleep on
 * ams --- max sleep time in milliseconds
 * aintok - 1 if should sleep interruptibly
 *
 * Returns 0 if timeout and EINTR if signalled.
 */
int
afs_osi_TimedSleep(void *event, afs_int32 ams, int aintok)
{
    int code = 0;
    struct afs_event *evp;
    clock_t ticks;

    ticks = (ams * afs_hz) / 1000;
#if defined(AFS_SUN510_ENV)
    ticks = ticks + ddi_get_lbolt();
#else
    ticks = ticks + lbolt;
#endif

    evp = afs_getevent(event);

    AFS_ASSERT_GLOCK();
    if (aintok) {
	if (cv_timedwait_sig(&evp->cond, &afs_global_lock, ticks) == 0)
	    code = EINTR;
    } else {
	cv_timedwait(&evp->cond, &afs_global_lock, ticks);
    }

    relevent(evp);
    return code;
}

static void
udf_discstrat_thread(void *arg)
{
	struct udf_mount *ump = (struct udf_mount *) arg;
	struct strat_private *priv = PRIV(ump);
	int empty;

	empty = 1;
	mutex_enter(&priv->discstrat_mutex);
	while (priv->run_thread || !empty) {
		/* process the current selected queue */
		udf_doshedule(ump);
		empty  = (bufq_peek(priv->queues[UDF_SHED_READING]) == NULL);
		empty &= (bufq_peek(priv->queues[UDF_SHED_WRITING]) == NULL);
		empty &= (bufq_peek(priv->queues[UDF_SHED_SEQWRITING]) == NULL);

		/* wait for more if needed */
		if (empty)
			cv_timedwait(&priv->discstrat_cv,
				&priv->discstrat_mutex, hz/8);
	}
	mutex_exit(&priv->discstrat_mutex);

	wakeup(&priv->run_thread);
	kthread_exit(0);
	/* not reached */
}

/*
 * ehci_wait_for_transfers_completion:
 *
 * Wait for processing all completed transfers and to send results
 * to upstream.
 */
static void
ehci_wait_for_isoc_completion(
	ehci_state_t		*ehcip,
	ehci_pipe_private_t	*pp)
{
	clock_t			xfer_cmpl_time_wait;

	ASSERT(mutex_owned(&ehcip->ehci_int_mutex));

	if (pp->pp_itw_head == NULL) {

		return;
	}

	/* Get the number of clock ticks to wait */
	xfer_cmpl_time_wait = drv_usectohz(EHCI_XFER_CMPL_TIMEWAIT * 1000000);

	(void) cv_timedwait(&pp->pp_xfer_cmpl_cv,
	    &ehcip->ehci_int_mutex,
	    ddi_get_lbolt() + xfer_cmpl_time_wait);

	if (pp->pp_itw_head) {
		USB_DPRINTF_L2(PRINT_MASK_LISTS, ehcip->ehci_log_hdl,
		    "ehci_wait_for_isoc_completion: "
		    "No transfers completion confirmation received");
	}
}

int
splat_condvar_test34_thread(void *arg)
{
	condvar_thr_t *ct = (condvar_thr_t *)arg;
	condvar_priv_t *cv = ct->ct_cvp;
	clock_t rc;

	ASSERT(cv->cv_magic == SPLAT_CONDVAR_TEST_MAGIC);

	mutex_enter(&cv->cv_mtx);
	splat_vprint(cv->cv_file, ct->ct_name,
	    "%s thread sleeping with %d waiters\n",
	    ct->ct_thread->comm, atomic_read(&cv->cv_condvar.cv_waiters));

	/* Sleep no longer than 3 seconds, for this test we should
	 * actually never sleep that long without being woken up. */
	rc = cv_timedwait(&cv->cv_condvar, &cv->cv_mtx, lbolt + HZ * 3);
	if (rc == -1) {
		ct->ct_rc = -ETIMEDOUT;
		splat_vprint(cv->cv_file, ct->ct_name, "%s thread timed out, "
		    "should have been woken\n", ct->ct_thread->comm);
	} else {
		splat_vprint(cv->cv_file, ct->ct_name,
		    "%s thread woken %d waiters remain\n",
		    ct->ct_thread->comm,
		    atomic_read(&cv->cv_condvar.cv_waiters));
	}

	mutex_exit(&cv->cv_mtx);

	/* wait for main thread reap us */
	while (!kthread_should_stop())
		schedule();
	return 0;
}

static void
trim_thread(void *arg)
{
	spa_t *spa = arg;
	zio_t *zio;

#ifdef _KERNEL
	(void) snprintf(curthread->td_name, sizeof(curthread->td_name),
	    "trim %s", spa_name(spa));
#endif

	for (;;) {
		mutex_enter(&spa->spa_trim_lock);
		if (spa->spa_trim_thread == NULL) {
			spa->spa_trim_thread = curthread;
			cv_signal(&spa->spa_trim_cv);
			mutex_exit(&spa->spa_trim_lock);
			thread_exit();
		}

		(void) cv_timedwait(&spa->spa_trim_cv, &spa->spa_trim_lock,
		    hz * trim_max_interval);
		mutex_exit(&spa->spa_trim_lock);

		zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);

		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
		trim_map_commit(spa, zio, spa->spa_root_vdev);
		(void) zio_wait(zio);
		trim_map_commit_done(spa, spa->spa_root_vdev);
		spa_config_exit(spa, SCL_STATE, FTAG);
	}
}

static void
npf_worker(void *arg)
{
	for (;;) {
		const bool finish = (worker_lwp == NULL);
		u_int i = NPF_MAX_WORKS;
		npf_workfunc_t work;

		/* Run the jobs. */
		while (i--) {
			if ((work = work_funcs[i]) != NULL) {
				work();
			}
		}

		/* Exit if requested and all jobs are done. */
		if (finish) {
			break;
		}

		/* Sleep and periodically wake up, unless we get notified. */
		mutex_enter(&worker_lock);
		worker_loop++;
		cv_broadcast(&worker_event_cv);
		cv_timedwait(&worker_cv, &worker_lock, WORKER_INTERVAL);
		mutex_exit(&worker_lock);
	}
	kthread_exit(0);
}

/* Performance monitor thread */
static void
spa_perfmon_thread(spa_t *spa)
{
	spa_perfmon_data_t *data = &spa->spa_perfmon;
	boolean_t done = B_FALSE;

	ASSERT(data);

	DTRACE_PROBE1(spa_pm_start, spa_t *, spa);

	/* take a reference against spa */
	mutex_enter(&spa_namespace_lock);
	spa_open_ref(spa, FTAG);
	mutex_exit(&spa_namespace_lock);

	/* CONSTCOND */
	while (1) {
		clock_t deadline = ddi_get_lbolt() +
		    spa_special_stat_update_ticks;

		/* wait for the next tick, check exit condition */
		mutex_enter(&data->perfmon_lock);
		(void) cv_timedwait(&data->perfmon_cv, &data->perfmon_lock,
		    deadline);
		if (spa->spa_state == POOL_STATE_UNINITIALIZED ||
		    data->perfmon_thr_exit)
			done = B_TRUE;
		mutex_exit(&data->perfmon_lock);

		if (done)
			goto out;

		/*
		 * do the monitoring work here: gather average
		 * latency and utilization statistics
		 */
		DTRACE_PROBE1(spa_pm_work, spa_t *, spa);
		spa_load_stats_update(spa);

		/* we can adjust load and dedup at the same time */
		if (spa_enable_data_placement_selection)
			spa_special_load_adjust(spa);
		if (spa->spa_dedup_best_effort)
			spa_special_dedup_adjust(spa);

		/* go to sleep until next tick */
		DTRACE_PROBE1(spa_pm_sleep, spa_t *, spa);
	}

out:
	/* release the reference against spa */
	mutex_enter(&spa_namespace_lock);
	spa_close(spa, FTAG);
	mutex_exit(&spa_namespace_lock);

	DTRACE_PROBE1(spa_pm_stop, spa_t *, spa);
	thread_exit();
}

void
usb_detach_wait(device_t dv, kcondvar_t *cv, kmutex_t *lock)
{
	DPRINTF(("usb_detach_wait: waiting for %s\n", device_xname(dv)));
	if (cv_timedwait(cv, lock, hz * 60))	// dv, PZERO, "usbdet", hz * 60
		printf("usb_detach_wait: %s didn't detach\n",
		        device_xname(dv));
	DPRINTF(("usb_detach_wait: %s done\n", device_xname(dv)));
}

static void cond(void *data) {
	int res;

	printf("%d: entered\n", (int) data);
	bsd_dde_prepare_thread("cond");

	mtx_lock(&mtx1);
	res = cv_timedwait(&cv1, &mtx1, hz/10);
	mtx_unlock(&mtx1);
	printf("%d: %s\n", (int) data, (res?"timeout":"success"));
}

void
usb_detach_wait(device_t dv, kcondvar_t *cv, kmutex_t *lock)
{
	USBHIST_FUNC(); USBHIST_CALLED(usbdebug);

	DPRINTFN(1, "waiting for dv %p", dv, 0, 0, 0);
	if (cv_timedwait(cv, lock, hz * 60))	// dv, PZERO, "usbdet", hz * 60
		printf("usb_detach_wait: %s didn't detach\n",
			device_xname(dv));
	DPRINTFN(1, "done", 0, 0, 0, 0);
}

static void
txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, uint64_t time)
{
	CALLB_CPR_SAFE_BEGIN(cpr);

	if (time)
		(void) cv_timedwait(cv, &tx->tx_sync_lock, time);
	else
		cv_wait(cv, &tx->tx_sync_lock);

	CALLB_CPR_SAFE_END(cpr, &tx->tx_sync_lock);
}

static int
splat_condvar_test5(struct file *file, void *arg)
{
        kcondvar_t condvar;
        kmutex_t mtx;
	clock_t time_left, time_before, time_after, time_delta;
	uint64_t whole_delta;
	uint32_t remain_delta;
	int rc = 0;

	mutex_init(&mtx, SPLAT_CONDVAR_TEST_NAME, MUTEX_DEFAULT, NULL);
	cv_init(&condvar, NULL, CV_DEFAULT, NULL);

        splat_vprint(file, SPLAT_CONDVAR_TEST5_NAME, "Thread going to sleep for "
	           "%d second and expecting to be woken by timeout\n", 1);

	/* Allow a 1 second timeout, plenty long to validate correctness. */
	time_before = lbolt;
	mutex_enter(&mtx);
	time_left = cv_timedwait(&condvar, &mtx, lbolt + HZ);
	mutex_exit(&mtx);
	time_after = lbolt;
	time_delta = time_after - time_before; /* XXX - Handle jiffie wrap */
	whole_delta  = time_delta;
	remain_delta = do_div(whole_delta, HZ);

	if (time_left == -1) {
		if (time_delta >= HZ) {
			splat_vprint(file, SPLAT_CONDVAR_TEST5_NAME,
			           "Thread correctly timed out and was asleep "
			           "for %d.%d seconds (%d second min)\n",
			           (int)whole_delta, (int)remain_delta, 1);
		} else {
			splat_vprint(file, SPLAT_CONDVAR_TEST5_NAME,
			           "Thread correctly timed out but was only "
			           "asleep for %d.%d seconds (%d second "
			           "min)\n", (int)whole_delta,
				   (int)remain_delta, 1);
			rc = -ETIMEDOUT;
		}
	} else {
		splat_vprint(file, SPLAT_CONDVAR_TEST5_NAME,
		           "Thread exited after only %d.%d seconds, it "
		           "did not hit the %d second timeout\n",
		           (int)whole_delta, (int)remain_delta, 1);
		rc = -ETIMEDOUT;
	}

	cv_destroy(&condvar);
	mutex_destroy(&mtx);

	return rc;
}

static int
awin_p2wi_wait(struct awin_p2wi_softc *sc, int flags)
{
	int error = 0, retry;

	/* Wait up to 5 seconds for a transfer to complete */
	sc->sc_stat = 0;
	for (retry = (flags & I2C_F_POLL) ? 100 : 5; retry > 0; retry--) {
		if (flags & I2C_F_POLL) {
			sc->sc_stat |= P2WI_READ(sc, AWIN_A31_P2WI_STAT_REG);
		} else {
			error = cv_timedwait(&sc->sc_cv, &sc->sc_lock, hz);
			if (error && error != EWOULDBLOCK) {
				break;
			}
		}
		if (sc->sc_stat & AWIN_A31_P2WI_STAT_MASK) {
			break;
		}
		if (flags & I2C_F_POLL) {
			delay(10000);
		}
	}
	if (retry == 0)
		error = EAGAIN;

	if (flags & I2C_F_POLL) {
		P2WI_WRITE(sc, AWIN_A31_P2WI_STAT_REG,
		    sc->sc_stat & AWIN_A31_P2WI_STAT_MASK);
	}

	if (error) {
		/* Abort transaction */
		device_printf(sc->sc_dev, "transfer timeout, error = %d\n",
		    error);
		P2WI_WRITE(sc, AWIN_A31_P2WI_CTRL_REG,
		    AWIN_A31_P2WI_CTRL_ABORT_TRANS);
		return error;
	}

	if (sc->sc_stat & AWIN_A31_P2WI_STAT_LOAD_BSY) {
		device_printf(sc->sc_dev, "transfer busy\n");
		return EBUSY;
	}
	if (sc->sc_stat & AWIN_A31_P2WI_STAT_TRANS_ERR) {
		device_printf(sc->sc_dev, "transfer error, id 0x%02llx\n",
		    __SHIFTOUT(sc->sc_stat, AWIN_A31_P2WI_STAT_TRANS_ERR_ID));
		return EIO;
	}

	return 0;
}

// Unlike cv_timedwait(), this timeout is relative to now.
clock_t cv_timedwait_hires(
        kcondvar_t * cv, kmutex_t * m, hrtime_t timeout, hrtime_t resolution, int flag)
{
    VERIFY0(flag); // We don't support flags.
    clock_t expiration = ddi_get_lbolt() + timeout;;

    if (resolution > 1) {
        expiration = (expiration / resolution) * resolution;
    }

    // clock_t is ticks (usec) and hrtime_t is nsec.
    return cv_timedwait(cv, m, USEC_TO_NSEC(expiration));
}