Home | History | Annotate | Line # | Download | only in kern
kern_synch.c revision 1.177.2.11
      1 /*	$NetBSD: kern_synch.c,v 1.177.2.11 2007/02/27 17:23:24 yamt Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999, 2000, 2004, 2006, 2007 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
      9  * NASA Ames Research Center, by Charles M. Hannum, Andrew Doran and
     10  * Daniel Sieger.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *	This product includes software developed by the NetBSD
     23  *	Foundation, Inc. and its contributors.
     24  * 4. Neither the name of The NetBSD Foundation nor the names of its
     25  *    contributors may be used to endorse or promote products derived
     26  *    from this software without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     29  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     30  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     31  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     32  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     33  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     34  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     35  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     36  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     37  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     38  * POSSIBILITY OF SUCH DAMAGE.
     39  */
     40 
     41 /*-
     42  * Copyright (c) 1982, 1986, 1990, 1991, 1993
     43  *	The Regents of the University of California.  All rights reserved.
     44  * (c) UNIX System Laboratories, Inc.
     45  * All or some portions of this file are derived from material licensed
     46  * to the University of California by American Telephone and Telegraph
     47  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     48  * the permission of UNIX System Laboratories, Inc.
     49  *
     50  * Redistribution and use in source and binary forms, with or without
     51  * modification, are permitted provided that the following conditions
     52  * are met:
     53  * 1. Redistributions of source code must retain the above copyright
     54  *    notice, this list of conditions and the following disclaimer.
     55  * 2. Redistributions in binary form must reproduce the above copyright
     56  *    notice, this list of conditions and the following disclaimer in the
     57  *    documentation and/or other materials provided with the distribution.
     58  * 3. Neither the name of the University nor the names of its contributors
     59  *    may be used to endorse or promote products derived from this software
     60  *    without specific prior written permission.
     61  *
     62  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     63  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     64  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     65  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     66  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     67  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     68  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     69  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     70  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     71  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     72  * SUCH DAMAGE.
     73  *
     74  *	@(#)kern_synch.c	8.9 (Berkeley) 5/19/95
     75  */
     76 
     77 #include <sys/cdefs.h>
     78 __KERNEL_RCSID(0, "$NetBSD: kern_synch.c,v 1.177.2.11 2007/02/27 17:23:24 yamt Exp $");
     79 
     80 #include "opt_kstack.h"
     81 #include "opt_lockdebug.h"
     82 #include "opt_multiprocessor.h"
     83 #include "opt_perfctrs.h"
     84 
     85 #define	__MUTEX_PRIVATE
     86 
     87 #include <sys/param.h>
     88 #include <sys/systm.h>
     89 #include <sys/proc.h>
     90 #include <sys/kernel.h>
     91 #if defined(PERFCTRS)
     92 #include <sys/pmc.h>
     93 #endif
     94 #include <sys/cpu.h>
     95 #include <sys/resourcevar.h>
     96 #include <sys/sched.h>
     97 #include <sys/syscall_stats.h>
     98 #include <sys/sleepq.h>
     99 #include <sys/lockdebug.h>
    100 
    101 #include <uvm/uvm_extern.h>
    102 
    103 int	lbolt;			/* once a second sleep address */
    104 
    105 /*
    106  * The global scheduler state.
    107  */
    108 kmutex_t	sched_mutex;		/* global sched state mutex */
    109 
    110 static void	sched_unsleep(struct lwp *);
    111 static void	sched_changepri(struct lwp *, pri_t);
    112 static void	sched_lendpri(struct lwp *, pri_t);
    113 
    114 syncobj_t sleep_syncobj = {
    115 	SOBJ_SLEEPQ_SORTED,
    116 	sleepq_unsleep,
    117 	sleepq_changepri,
    118 	sleepq_lendpri,
    119 	syncobj_noowner,
    120 };
    121 
    122 syncobj_t sched_syncobj = {
    123 	SOBJ_SLEEPQ_SORTED,
    124 	sched_unsleep,
    125 	sched_changepri,
    126 	sched_lendpri,
    127 	syncobj_noowner,
    128 };
    129 
    130 /*
    131  * During autoconfiguration or after a panic, a sleep will simply lower the
    132  * priority briefly to allow interrupts, then return.  The priority to be
    133  * used (safepri) is machine-dependent, thus this value is initialized and
    134  * maintained in the machine-dependent layers.  This priority will typically
    135  * be 0, or the lowest priority that is safe for use on the interrupt stack;
    136  * it can be made higher to block network software interrupts after panics.
    137  */
    138 int	safepri;
    139 
    140 /*
    141  * OBSOLETE INTERFACE
    142  *
    143  * General sleep call.  Suspends the current process until a wakeup is
    144  * performed on the specified identifier.  The process will then be made
    145  * runnable with the specified priority.  Sleeps at most timo/hz seconds (0
    146  * means no timeout).  If pri includes PCATCH flag, signals are checked
    147  * before and after sleeping, else signals are not checked.  Returns 0 if
    148  * awakened, EWOULDBLOCK if the timeout expires.  If PCATCH is set and a
    149  * signal needs to be delivered, ERESTART is returned if the current system
    150  * call should be restarted if possible, and EINTR is returned if the system
    151  * call should be interrupted by the signal (return EINTR).
    152  *
    153  * The interlock is held until we are on a sleep queue. The interlock will
    154  * be locked before returning back to the caller unless the PNORELOCK flag
    155  * is specified, in which case the interlock will always be unlocked upon
    156  * return.
    157  */
    158 int
    159 ltsleep(wchan_t ident, pri_t priority, const char *wmesg, int timo,
    160 	volatile struct simplelock *interlock)
    161 {
    162 	struct lwp *l = curlwp;
    163 	sleepq_t *sq;
    164 	int error, catch;
    165 
    166 	if (sleepq_dontsleep(l)) {
    167 		(void)sleepq_abort(NULL, 0);
    168 		if ((priority & PNORELOCK) != 0)
    169 			simple_unlock(interlock);
    170 		return 0;
    171 	}
    172 
    173 	sq = sleeptab_lookup(&sleeptab, ident);
    174 	sleepq_enter(sq, l);
    175 
    176 	if (interlock != NULL) {
    177 		LOCK_ASSERT(simple_lock_held(interlock));
    178 		simple_unlock(interlock);
    179 	}
    180 
    181 	catch = priority & PCATCH;
    182 	sleepq_block(sq, priority & PRIMASK, ident, wmesg, timo, catch,
    183 	    &sleep_syncobj);
    184 	error = sleepq_unblock(timo, catch);
    185 
    186 	if (interlock != NULL && (priority & PNORELOCK) == 0)
    187 		simple_lock(interlock);
    188 
    189 	return error;
    190 }
    191 
    192 /*
    193  * General sleep call for situations where a wake-up is not expected.
    194  */
    195 int
    196 kpause(const char *wmesg, bool intr, int timo, kmutex_t *mtx)
    197 {
    198 	struct lwp *l = curlwp;
    199 	sleepq_t *sq;
    200 	int error;
    201 
    202 	if (sleepq_dontsleep(l))
    203 		return sleepq_abort(NULL, 0);
    204 
    205 	if (mtx != NULL)
    206 		mutex_exit(mtx);
    207 	sq = sleeptab_lookup(&sleeptab, l);
    208 	sleepq_enter(sq, l);
    209 	sleepq_block(sq, sched_kpri(l), l, wmesg, timo, intr, &sleep_syncobj);
    210 	error = sleepq_unblock(timo, intr);
    211 	if (mtx != NULL)
    212 		mutex_enter(mtx);
    213 
    214 	return error;
    215 }
    216 
    217 /*
    218  * OBSOLETE INTERFACE
    219  *
    220  * Make all processes sleeping on the specified identifier runnable.
    221  */
    222 void
    223 wakeup(wchan_t ident)
    224 {
    225 	sleepq_t *sq;
    226 
    227 	if (cold)
    228 		return;
    229 
    230 	sq = sleeptab_lookup(&sleeptab, ident);
    231 	sleepq_wake(sq, ident, (u_int)-1);
    232 }
    233 
    234 /*
    235  * OBSOLETE INTERFACE
    236  *
    237  * Make the highest priority process first in line on the specified
    238  * identifier runnable.
    239  */
    240 void
    241 wakeup_one(wchan_t ident)
    242 {
    243 	sleepq_t *sq;
    244 
    245 	if (cold)
    246 		return;
    247 
    248 	sq = sleeptab_lookup(&sleeptab, ident);
    249 	sleepq_wake(sq, ident, 1);
    250 }
    251 
    252 
    253 /*
    254  * General yield call.  Puts the current process back on its run queue and
    255  * performs a voluntary context switch.  Should only be called when the
    256  * current process explicitly requests it (eg sched_yield(2) in compat code).
    257  */
    258 void
    259 yield(void)
    260 {
    261 	struct lwp *l = curlwp;
    262 
    263 	KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
    264 	lwp_lock(l);
    265 	if (l->l_stat == LSONPROC) {
    266 		KASSERT(lwp_locked(l, &sched_mutex));
    267 		l->l_priority = l->l_usrpri;
    268 	}
    269 	l->l_nvcsw++;
    270 	mi_switch(l, NULL);
    271 	KERNEL_LOCK(l->l_biglocks, l);
    272 }
    273 
    274 /*
    275  * General preemption call.  Puts the current process back on its run queue
    276  * and performs an involuntary context switch.
    277  */
    278 void
    279 preempt(void)
    280 {
    281 	struct lwp *l = curlwp;
    282 
    283 	KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
    284 	lwp_lock(l);
    285 	if (l->l_stat == LSONPROC) {
    286 		KASSERT(lwp_locked(l, &sched_mutex));
    287 		l->l_priority = l->l_usrpri;
    288 	}
    289 	l->l_nivcsw++;
    290 	(void)mi_switch(l, NULL);
    291 	KERNEL_LOCK(l->l_biglocks, l);
    292 }
    293 
    294 /*
    295  * sched_switch_unlock: update 'curlwp' and release old lwp.
    296  */
    297 
    298 void
    299 sched_switch_unlock(struct lwp *old, struct lwp *new)
    300 {
    301 
    302 	KASSERT(old == NULL || old == curlwp);
    303 
    304 	if (old != NULL) {
    305 		LOCKDEBUG_BARRIER(old->l_mutex, 1);
    306 	} else {
    307 		LOCKDEBUG_BARRIER(NULL, 1);
    308 	}
    309 
    310 	curlwp = new;
    311 	if (old != NULL) {
    312 		lwp_unlock(old);
    313 	}
    314 	spl0();
    315 }
    316 
    317 /*
    318  * Compute the amount of time during which the current lwp was running.
    319  *
    320  * - update l_rtime unless it's an idle lwp.
    321  * - update spc_runtime for the next lwp.
    322  */
    323 
    324 static inline void
    325 updatertime(struct lwp *l, struct schedstate_percpu *spc)
    326 {
    327 	struct timeval tv;
    328 	long s, u;
    329 
    330 	if ((l->l_flag & LW_IDLE) != 0) {
    331 		microtime(&spc->spc_runtime);
    332 		return;
    333 	}
    334 
    335 	microtime(&tv);
    336 	u = l->l_rtime.tv_usec + (tv.tv_usec - spc->spc_runtime.tv_usec);
    337 	s = l->l_rtime.tv_sec + (tv.tv_sec - spc->spc_runtime.tv_sec);
    338 	if (u < 0) {
    339 		u += 1000000;
    340 		s--;
    341 	} else if (u >= 1000000) {
    342 		u -= 1000000;
    343 		s++;
    344 	}
    345 	l->l_rtime.tv_usec = u;
    346 	l->l_rtime.tv_sec = s;
    347 
    348 	spc->spc_runtime = tv;
    349 }
    350 
    351 /*
    352  * The machine independent parts of context switch.  Switch to "new"
    353  * if non-NULL, otherwise let cpu_switch choose the next lwp.
    354  *
    355  * Returns 1 if another process was actually run.
    356  */
    357 int
    358 mi_switch(struct lwp *l, struct lwp *newl)
    359 {
    360 	struct schedstate_percpu *spc;
    361 	int retval, oldspl;
    362 
    363 	LOCK_ASSERT(lwp_locked(l, NULL));
    364 
    365 #ifdef LOCKDEBUG
    366 	spinlock_switchcheck();
    367 	simple_lock_switchcheck();
    368 #endif
    369 #ifdef KSTACK_CHECK_MAGIC
    370 	kstack_check_magic(l);
    371 #endif
    372 
    373 	/*
    374 	 * It's safe to read the per CPU schedstate unlocked here, as all we
    375 	 * are after is the run time and that's guarenteed to have been last
    376 	 * updated by this CPU.
    377 	 */
    378 	KDASSERT(l->l_cpu == curcpu());
    379 	spc = &l->l_cpu->ci_schedstate;
    380 
    381 	/* Count time spent in current system call */
    382 	SYSCALL_TIME_SLEEP(l);
    383 
    384 	/*
    385 	 * XXXSMP If we are using h/w performance counters, save context.
    386 	 */
    387 #if PERFCTRS
    388 	if (PMC_ENABLED(l->l_proc)) {
    389 		pmc_save_context(l->l_proc);
    390 	}
    391 #endif
    392 
    393 	/*
    394 	 * If on the CPU and we have gotten this far, then we must yield.
    395 	 */
    396 	KASSERT(l->l_stat != LSRUN);
    397 	if (l->l_stat == LSONPROC) {
    398 		KASSERT(lwp_locked(l, &sched_mutex));
    399 		l->l_stat = LSRUN;
    400 		if ((l->l_flag & LW_IDLE) == 0) {
    401 			sched_enqueue(l);
    402 		}
    403 	}
    404 	uvmexp.swtch++;
    405 
    406 	/*
    407 	 * Process is about to yield the CPU; clear the appropriate
    408 	 * scheduling flags.
    409 	 */
    410 	spc->spc_flags &= ~SPCF_SWITCHCLEAR;
    411 
    412 	LOCKDEBUG_BARRIER(l->l_mutex, 1);
    413 
    414 	/*
    415 	 * Switch to the new LWP if necessary.
    416 	 * When we run again, we'll return back here.
    417 	 */
    418 	oldspl = MUTEX_SPIN_OLDSPL(l->l_cpu);
    419 
    420 	/*
    421 	 * Acquire the sched_mutex if necessary.
    422 	 */
    423 #if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
    424 	if (l->l_mutex != &sched_mutex) {
    425 		mutex_enter(&sched_mutex);
    426 	}
    427 #endif
    428 
    429 	if (newl == NULL) {
    430 		newl = sched_nextlwp();
    431 	}
    432 	if (newl != NULL) {
    433 		KASSERT(lwp_locked(newl, &sched_mutex));
    434 		sched_dequeue(newl);
    435 	} else {
    436 		newl = l->l_cpu->ci_data.cpu_idlelwp;
    437 		KASSERT(newl != NULL);
    438 	}
    439 	KASSERT(lwp_locked(newl, &sched_mutex));
    440 	newl->l_stat = LSONPROC;
    441 	newl->l_cpu = l->l_cpu;
    442 
    443 #if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
    444 	if (l->l_mutex != &sched_mutex) {
    445 		mutex_exit(&sched_mutex);
    446 	}
    447 #endif
    448 
    449 	updatertime(l, spc);
    450 	if (l != newl) {
    451 		struct lwp *prevlwp;
    452 
    453 		uvmexp.swtch++;
    454 		pmap_deactivate(l);
    455 		prevlwp = cpu_switchto(l, newl);
    456 		sched_switch_unlock(prevlwp, l);
    457 		pmap_activate(l);
    458 		retval = 1;
    459 	} else {
    460 		sched_switch_unlock(l, l);
    461 		retval = 0;
    462 	}
    463 
    464 	KASSERT(l == curlwp);
    465 	KASSERT(l->l_stat == LSONPROC);
    466 
    467 	/*
    468 	 * XXXSMP If we are using h/w performance counters, restore context.
    469 	 */
    470 #if PERFCTRS
    471 	if (PMC_ENABLED(l->l_proc)) {
    472 		pmc_restore_context(l->l_proc);
    473 	}
    474 #endif
    475 
    476 	/*
    477 	 * We're running again; record our new start time.  We might
    478 	 * be running on a new CPU now, so don't use the cached
    479 	 * schedstate_percpu pointer.
    480 	 */
    481 	SYSCALL_TIME_WAKEUP(l);
    482 	KDASSERT(l->l_cpu == curcpu());
    483 
    484 	(void)splsched();
    485 	splx(oldspl);
    486 	return retval;
    487 }
    488 
    489 /*
    490  * Change process state to be runnable, placing it on the run queue if it is
    491  * in memory, and awakening the swapper if it isn't in memory.
    492  *
    493  * Call with the process and LWP locked.  Will return with the LWP unlocked.
    494  */
    495 void
    496 setrunnable(struct lwp *l)
    497 {
    498 	struct proc *p = l->l_proc;
    499 	sigset_t *ss;
    500 
    501 	KASSERT((l->l_flag & LW_IDLE) == 0);
    502 	KASSERT(mutex_owned(&p->p_smutex));
    503 	KASSERT(lwp_locked(l, NULL));
    504 
    505 	switch (l->l_stat) {
    506 	case LSSTOP:
    507 		/*
    508 		 * If we're being traced (possibly because someone attached us
    509 		 * while we were stopped), check for a signal from the debugger.
    510 		 */
    511 		if ((p->p_slflag & PSL_TRACED) != 0 && p->p_xstat != 0) {
    512 			if ((sigprop[p->p_xstat] & SA_TOLWP) != 0)
    513 				ss = &l->l_sigpend.sp_set;
    514 			else
    515 				ss = &p->p_sigpend.sp_set;
    516 			sigaddset(ss, p->p_xstat);
    517 			signotify(l);
    518 		}
    519 		p->p_nrlwps++;
    520 		break;
    521 	case LSSUSPENDED:
    522 		l->l_flag &= ~LW_WSUSPEND;
    523 		p->p_nrlwps++;
    524 		break;
    525 	case LSSLEEP:
    526 		KASSERT(l->l_wchan != NULL);
    527 		break;
    528 	default:
    529 		panic("setrunnable: lwp %p state was %d", l, l->l_stat);
    530 	}
    531 
    532 	/*
    533 	 * If the LWP was sleeping interruptably, then it's OK to start it
    534 	 * again.  If not, mark it as still sleeping.
    535 	 */
    536 	if (l->l_wchan != NULL) {
    537 		l->l_stat = LSSLEEP;
    538 		/* lwp_unsleep() will release the lock. */
    539 		lwp_unsleep(l);
    540 		return;
    541 	}
    542 
    543 	LOCK_ASSERT(lwp_locked(l, &sched_mutex));
    544 
    545 	/*
    546 	 * If the LWP is still on the CPU, mark it as LSONPROC.  It may be
    547 	 * about to call mi_switch(), in which case it will yield.
    548 	 *
    549 	 * XXXSMP Will need to change for preemption.
    550 	 */
    551 #ifdef MULTIPROCESSOR
    552 	if (l->l_cpu->ci_curlwp == l) {
    553 #else
    554 	if (l == curlwp) {
    555 #endif
    556 		l->l_stat = LSONPROC;
    557 		l->l_slptime = 0;
    558 		lwp_unlock(l);
    559 		return;
    560 	}
    561 
    562 	/*
    563 	 * Set the LWP runnable.  If it's swapped out, we need to wake the swapper
    564 	 * to bring it back in.  Otherwise, enter it into a run queue.
    565 	 */
    566 	sched_setrunnable(l);
    567 	l->l_stat = LSRUN;
    568 	l->l_slptime = 0;
    569 
    570 	if (l->l_flag & LW_INMEM) {
    571 		sched_enqueue(l);
    572 		resched_cpu(l);
    573 		lwp_unlock(l);
    574 	} else {
    575 		lwp_unlock(l);
    576 		uvm_kick_scheduler();
    577 	}
    578 }
    579 
    580 /*
    581  * suspendsched:
    582  *
    583  *	Convert all non-L_SYSTEM LSSLEEP or LSRUN LWPs to LSSUSPENDED.
    584  */
    585 void
    586 suspendsched(void)
    587 {
    588 #ifdef MULTIPROCESSOR
    589 	CPU_INFO_ITERATOR cii;
    590 	struct cpu_info *ci;
    591 #endif
    592 	struct lwp *l;
    593 	struct proc *p;
    594 
    595 	/*
    596 	 * We do this by process in order not to violate the locking rules.
    597 	 */
    598 	mutex_enter(&proclist_mutex);
    599 	PROCLIST_FOREACH(p, &allproc) {
    600 		mutex_enter(&p->p_smutex);
    601 
    602 		if ((p->p_flag & PK_SYSTEM) != 0) {
    603 			mutex_exit(&p->p_smutex);
    604 			continue;
    605 		}
    606 
    607 		p->p_stat = SSTOP;
    608 
    609 		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
    610 			if (l == curlwp)
    611 				continue;
    612 
    613 			lwp_lock(l);
    614 
    615 			/*
    616 			 * Set L_WREBOOT so that the LWP will suspend itself
    617 			 * when it tries to return to user mode.  We want to
    618 			 * try and get to get as many LWPs as possible to
    619 			 * the user / kernel boundary, so that they will
    620 			 * release any locks that they hold.
    621 			 */
    622 			l->l_flag |= (LW_WREBOOT | LW_WSUSPEND);
    623 
    624 			if (l->l_stat == LSSLEEP &&
    625 			    (l->l_flag & LW_SINTR) != 0) {
    626 				/* setrunnable() will release the lock. */
    627 				setrunnable(l);
    628 				continue;
    629 			}
    630 
    631 			lwp_unlock(l);
    632 		}
    633 
    634 		mutex_exit(&p->p_smutex);
    635 	}
    636 	mutex_exit(&proclist_mutex);
    637 
    638 	/*
    639 	 * Kick all CPUs to make them preempt any LWPs running in user mode.
    640 	 * They'll trap into the kernel and suspend themselves in userret().
    641 	 */
    642 	sched_lock(0);
    643 #ifdef MULTIPROCESSOR
    644 	for (CPU_INFO_FOREACH(cii, ci))
    645 		cpu_need_resched(ci, 0);
    646 #else
    647 	cpu_need_resched(curcpu(), 0);
    648 #endif
    649 	sched_unlock(0);
    650 }
    651 
    652 /*
    653  * sched_kpri:
    654  *
    655  *	Scale a priority level to a kernel priority level, usually
    656  *	for an LWP that is about to sleep.
    657  */
    658 pri_t
    659 sched_kpri(struct lwp *l)
    660 {
    661 	/*
    662 	 * Scale user priorities (127 -> 50) up to kernel priorities
    663 	 * in the range (49 -> 8).  Reserve the top 8 kernel priorities
    664 	 * for high priority kthreads.  Kernel priorities passed in
    665 	 * are left "as is".  XXX This is somewhat arbitrary.
    666 	 */
    667 	static const uint8_t kpri_tab[] = {
    668 		 0,   1,   2,   3,   4,   5,   6,   7,
    669 		 8,   9,  10,  11,  12,  13,  14,  15,
    670 		16,  17,  18,  19,  20,  21,  22,  23,
    671 		24,  25,  26,  27,  28,  29,  30,  31,
    672 		32,  33,  34,  35,  36,  37,  38,  39,
    673 		40,  41,  42,  43,  44,  45,  46,  47,
    674 		48,  49,   8,   8,   9,   9,  10,  10,
    675 		11,  11,  12,  12,  13,  14,  14,  15,
    676 		15,  16,  16,  17,  17,  18,  18,  19,
    677 		20,  20,  21,  21,  22,  22,  23,  23,
    678 		24,  24,  25,  26,  26,  27,  27,  28,
    679 		28,  29,  29,  30,  30,  31,  32,  32,
    680 		33,  33,  34,  34,  35,  35,  36,  36,
    681 		37,  38,  38,  39,  39,  40,  40,  41,
    682 		41,  42,  42,  43,  44,  44,  45,  45,
    683 		46,  46,  47,  47,  48,  48,  49,  49,
    684 	};
    685 
    686 	return (pri_t)kpri_tab[l->l_usrpri];
    687 }
    688 
    689 /*
    690  * sched_unsleep:
    691  *
    692  *	The is called when the LWP has not been awoken normally but instead
    693  *	interrupted: for example, if the sleep timed out.  Because of this,
    694  *	it's not a valid action for running or idle LWPs.
    695  */
    696 static void
    697 sched_unsleep(struct lwp *l)
    698 {
    699 
    700 	lwp_unlock(l);
    701 	panic("sched_unsleep");
    702 }
    703 
    704 inline void
    705 resched_cpu(struct lwp *l)
    706 {
    707 	struct cpu_info *ci;
    708 	const pri_t pri = lwp_eprio(l);
    709 
    710 	/*
    711 	 * XXXSMP
    712 	 * Since l->l_cpu persists across a context switch,
    713 	 * this gives us *very weak* processor affinity, in
    714 	 * that we notify the CPU on which the process last
    715 	 * ran that it should try to switch.
    716 	 *
    717 	 * This does not guarantee that the process will run on
    718 	 * that processor next, because another processor might
    719 	 * grab it the next time it performs a context switch.
    720 	 *
    721 	 * This also does not handle the case where its last
    722 	 * CPU is running a higher-priority process, but every
    723 	 * other CPU is running a lower-priority process.  There
    724 	 * are ways to handle this situation, but they're not
    725 	 * currently very pretty, and we also need to weigh the
    726 	 * cost of moving a process from one CPU to another.
    727 	 *
    728 	 * XXXSMP
    729 	 * There is also the issue of locking the other CPU's
    730 	 * sched state, which we currently do not do.
    731 	 */
    732 	ci = (l->l_cpu != NULL) ? l->l_cpu : curcpu();
    733 	if (pri < ci->ci_schedstate.spc_curpriority)
    734 		cpu_need_resched(ci, 0);
    735 }
    736 
    737 static void
    738 sched_changepri(struct lwp *l, pri_t pri)
    739 {
    740 
    741 	LOCK_ASSERT(lwp_locked(l, &sched_mutex));
    742 
    743 	l->l_usrpri = pri;
    744 	if (l->l_priority < PUSER)
    745 		return;
    746 
    747 	if (l->l_stat != LSRUN || (l->l_flag & LW_INMEM) == 0) {
    748 		l->l_priority = pri;
    749 		return;
    750 	}
    751 
    752 	sched_dequeue(l);
    753 	l->l_priority = pri;
    754 	sched_enqueue(l);
    755 	resched_cpu(l);
    756 }
    757 
    758 static void
    759 sched_lendpri(struct lwp *l, pri_t pri)
    760 {
    761 
    762 	LOCK_ASSERT(lwp_locked(l, &sched_mutex));
    763 
    764 	if (l->l_stat != LSRUN || (l->l_flag & LW_INMEM) == 0) {
    765 		l->l_inheritedprio = pri;
    766 		return;
    767 	}
    768 
    769 	sched_dequeue(l);
    770 	l->l_inheritedprio = pri;
    771 	sched_enqueue(l);
    772 	resched_cpu(l);
    773 }
    774 
    775 struct lwp *
    776 syncobj_noowner(wchan_t wchan)
    777 {
    778 
    779 	return NULL;
    780 }
    781