Home | History | Annotate | Line # | Download | only in kern
kern_synch.c revision 1.177.2.15
      1 /*	$NetBSD: kern_synch.c,v 1.177.2.15 2007/03/12 05:58:38 rmind 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 1.177.2.15 2007/03/12 05:58:38 rmind 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 int
    193 mtsleep(wchan_t ident, pri_t priority, const char *wmesg, int timo,
    194 	kmutex_t *mtx)
    195 {
    196 	struct lwp *l = curlwp;
    197 	sleepq_t *sq;
    198 	int error, catch;
    199 
    200 	if (sleepq_dontsleep(l)) {
    201 		(void)sleepq_abort(mtx, (priority & PNORELOCK) != 0);
    202 		return 0;
    203 	}
    204 
    205 	sq = sleeptab_lookup(&sleeptab, ident);
    206 	sleepq_enter(sq, l);
    207 	mutex_exit(mtx);
    208 
    209 	catch = priority & PCATCH;
    210 	sleepq_block(sq, priority & PRIMASK, ident, wmesg, timo, catch,
    211 	    &sleep_syncobj);
    212 	error = sleepq_unblock(timo, catch);
    213 
    214 	if ((priority & PNORELOCK) == 0)
    215 		mutex_enter(mtx);
    216 
    217 	return error;
    218 }
    219 
    220 /*
    221  * General sleep call for situations where a wake-up is not expected.
    222  */
    223 int
    224 kpause(const char *wmesg, bool intr, int timo, kmutex_t *mtx)
    225 {
    226 	struct lwp *l = curlwp;
    227 	sleepq_t *sq;
    228 	int error;
    229 
    230 	if (sleepq_dontsleep(l))
    231 		return sleepq_abort(NULL, 0);
    232 
    233 	if (mtx != NULL)
    234 		mutex_exit(mtx);
    235 	sq = sleeptab_lookup(&sleeptab, l);
    236 	sleepq_enter(sq, l);
    237 	sleepq_block(sq, sched_kpri(l), l, wmesg, timo, intr, &sleep_syncobj);
    238 	error = sleepq_unblock(timo, intr);
    239 	if (mtx != NULL)
    240 		mutex_enter(mtx);
    241 
    242 	return error;
    243 }
    244 
    245 /*
    246  * OBSOLETE INTERFACE
    247  *
    248  * Make all processes sleeping on the specified identifier runnable.
    249  */
    250 void
    251 wakeup(wchan_t ident)
    252 {
    253 	sleepq_t *sq;
    254 
    255 	if (cold)
    256 		return;
    257 
    258 	sq = sleeptab_lookup(&sleeptab, ident);
    259 	sleepq_wake(sq, ident, (u_int)-1);
    260 }
    261 
    262 /*
    263  * OBSOLETE INTERFACE
    264  *
    265  * Make the highest priority process first in line on the specified
    266  * identifier runnable.
    267  */
    268 void
    269 wakeup_one(wchan_t ident)
    270 {
    271 	sleepq_t *sq;
    272 
    273 	if (cold)
    274 		return;
    275 
    276 	sq = sleeptab_lookup(&sleeptab, ident);
    277 	sleepq_wake(sq, ident, 1);
    278 }
    279 
    280 
    281 /*
    282  * General yield call.  Puts the current process back on its run queue and
    283  * performs a voluntary context switch.  Should only be called when the
    284  * current process explicitly requests it (eg sched_yield(2) in compat code).
    285  */
    286 void
    287 yield(void)
    288 {
    289 	struct lwp *l = curlwp;
    290 
    291 	KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
    292 	lwp_lock(l);
    293 	if (l->l_stat == LSONPROC) {
    294 		KASSERT(lwp_locked(l, &sched_mutex));
    295 		l->l_priority = l->l_usrpri;
    296 	}
    297 	l->l_nvcsw++;
    298 	mi_switch(l);
    299 	KERNEL_LOCK(l->l_biglocks, l);
    300 }
    301 
    302 /*
    303  * General preemption call.  Puts the current process back on its run queue
    304  * and performs an involuntary context switch.
    305  */
    306 void
    307 preempt(void)
    308 {
    309 	struct lwp *l = curlwp;
    310 
    311 	KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
    312 	lwp_lock(l);
    313 	if (l->l_stat == LSONPROC) {
    314 		KASSERT(lwp_locked(l, &sched_mutex));
    315 		l->l_priority = l->l_usrpri;
    316 	}
    317 	l->l_nivcsw++;
    318 	(void)mi_switch(l);
    319 	KERNEL_LOCK(l->l_biglocks, l);
    320 }
    321 
    322 /*
    323  * sched_switch_unlock: update 'curlwp' and release old lwp.
    324  */
    325 
    326 void
    327 sched_switch_unlock(struct lwp *old, struct lwp *new)
    328 {
    329 
    330 	KASSERT(old == NULL || old == curlwp);
    331 	KASSERT(new != NULL);
    332 
    333 	if (old != NULL) {
    334 		LOCKDEBUG_BARRIER(old->l_mutex, 1);
    335 	} else {
    336 		LOCKDEBUG_BARRIER(NULL, 1);
    337 	}
    338 
    339 	curlwp = new;
    340 	if (old != NULL) {
    341 		lwp_unlock(old);
    342 	}
    343 	spl0();
    344 }
    345 
    346 /*
    347  * Compute the amount of time during which the current lwp was running.
    348  *
    349  * - update l_rtime unless it's an idle lwp.
    350  * - update spc_runtime for the next lwp.
    351  */
    352 
    353 static inline void
    354 updatertime(struct lwp *l, struct schedstate_percpu *spc)
    355 {
    356 	struct timeval tv;
    357 	long s, u;
    358 
    359 	if ((l->l_flag & LW_IDLE) != 0) {
    360 		microtime(&spc->spc_runtime);
    361 		return;
    362 	}
    363 
    364 	microtime(&tv);
    365 	u = l->l_rtime.tv_usec + (tv.tv_usec - spc->spc_runtime.tv_usec);
    366 	s = l->l_rtime.tv_sec + (tv.tv_sec - spc->spc_runtime.tv_sec);
    367 	if (u < 0) {
    368 		u += 1000000;
    369 		s--;
    370 	} else if (u >= 1000000) {
    371 		u -= 1000000;
    372 		s++;
    373 	}
    374 	l->l_rtime.tv_usec = u;
    375 	l->l_rtime.tv_sec = s;
    376 
    377 	spc->spc_runtime = tv;
    378 }
    379 
    380 /*
    381  * The machine independent parts of context switch.
    382  *
    383  * Returns 1 if another process was actually run.
    384  */
    385 int
    386 mi_switch(struct lwp *l)
    387 {
    388 	struct schedstate_percpu *spc;
    389 	struct lwp *newl;
    390 	int retval, oldspl;
    391 
    392 	LOCK_ASSERT(lwp_locked(l, NULL));
    393 
    394 #ifdef LOCKDEBUG
    395 	spinlock_switchcheck();
    396 	simple_lock_switchcheck();
    397 #endif
    398 #ifdef KSTACK_CHECK_MAGIC
    399 	kstack_check_magic(l);
    400 #endif
    401 
    402 	/*
    403 	 * It's safe to read the per CPU schedstate unlocked here, as all we
    404 	 * are after is the run time and that's guarenteed to have been last
    405 	 * updated by this CPU.
    406 	 */
    407 	KDASSERT(l->l_cpu == curcpu());
    408 	spc = &l->l_cpu->ci_schedstate;
    409 
    410 	/* Count time spent in current system call */
    411 	SYSCALL_TIME_SLEEP(l);
    412 
    413 	/*
    414 	 * XXXSMP If we are using h/w performance counters, save context.
    415 	 */
    416 #if PERFCTRS
    417 	if (PMC_ENABLED(l->l_proc)) {
    418 		pmc_save_context(l->l_proc);
    419 	}
    420 #endif
    421 
    422 	/*
    423 	 * Process is about to yield the CPU; clear the appropriate
    424 	 * scheduling flags.
    425 	 */
    426 	spc->spc_flags &= ~SPCF_SWITCHCLEAR;
    427 
    428 	LOCKDEBUG_BARRIER(l->l_mutex, 1);
    429 
    430 	/*
    431 	 * Switch to the new LWP if necessary.
    432 	 * When we run again, we'll return back here.
    433 	 */
    434 	oldspl = MUTEX_SPIN_OLDSPL(l->l_cpu);
    435 
    436 	/*
    437 	 * Acquire the sched_mutex if necessary.
    438 	 */
    439 #if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
    440 	if (l->l_mutex != &sched_mutex) {
    441 		mutex_enter(&sched_mutex);
    442 	}
    443 #endif
    444 	/*
    445 	 * Let sched_switch() select the LWP to run the CPU next.
    446 	 * If no LWP is runnable, switch to the idle LWP.
    447 	 * Please note that sched_switch() will enqueue the LWP.
    448 	 */
    449 	newl = sched_switch(l);
    450 	if (newl == NULL) {
    451 		newl = l->l_cpu->ci_data.cpu_idlelwp;
    452 		KASSERT(newl != NULL);
    453 	} else {
    454 		sched_dequeue(newl);
    455 	}
    456 	KASSERT(lwp_locked(newl, &sched_mutex));
    457 	newl->l_stat = LSONPROC;
    458 	newl->l_cpu = l->l_cpu;
    459 
    460 #if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
    461 	if (l->l_mutex != &sched_mutex) {
    462 		mutex_exit(&sched_mutex);
    463 	}
    464 #endif
    465 
    466 	updatertime(l, spc);
    467 	if (l != newl) {
    468 		struct lwp *prevlwp;
    469 
    470 		uvmexp.swtch++;
    471 		pmap_deactivate(l);
    472 		prevlwp = cpu_switchto(l, newl);
    473 		sched_switch_unlock(prevlwp, l);
    474 		pmap_activate(l);
    475 		retval = 1;
    476 	} else {
    477 		sched_switch_unlock(l, l);
    478 		retval = 0;
    479 	}
    480 
    481 	KASSERT(l == curlwp);
    482 	KASSERT(l->l_stat == LSONPROC);
    483 
    484 	/*
    485 	 * XXXSMP If we are using h/w performance counters, restore context.
    486 	 */
    487 #if PERFCTRS
    488 	if (PMC_ENABLED(l->l_proc)) {
    489 		pmc_restore_context(l->l_proc);
    490 	}
    491 #endif
    492 
    493 	/*
    494 	 * We're running again; record our new start time.  We might
    495 	 * be running on a new CPU now, so don't use the cached
    496 	 * schedstate_percpu pointer.
    497 	 */
    498 	SYSCALL_TIME_WAKEUP(l);
    499 	KDASSERT(l->l_cpu == curcpu());
    500 
    501 	(void)splsched();
    502 	splx(oldspl);
    503 	return retval;
    504 }
    505 
    506 /*
    507  * Change process state to be runnable, placing it on the run queue if it is
    508  * in memory, and awakening the swapper if it isn't in memory.
    509  *
    510  * Call with the process and LWP locked.  Will return with the LWP unlocked.
    511  */
    512 void
    513 setrunnable(struct lwp *l)
    514 {
    515 	struct proc *p = l->l_proc;
    516 	sigset_t *ss;
    517 
    518 	KASSERT((l->l_flag & LW_IDLE) == 0);
    519 	KASSERT(mutex_owned(&p->p_smutex));
    520 	KASSERT(lwp_locked(l, NULL));
    521 
    522 	switch (l->l_stat) {
    523 	case LSSTOP:
    524 		/*
    525 		 * If we're being traced (possibly because someone attached us
    526 		 * while we were stopped), check for a signal from the debugger.
    527 		 */
    528 		if ((p->p_slflag & PSL_TRACED) != 0 && p->p_xstat != 0) {
    529 			if ((sigprop[p->p_xstat] & SA_TOLWP) != 0)
    530 				ss = &l->l_sigpend.sp_set;
    531 			else
    532 				ss = &p->p_sigpend.sp_set;
    533 			sigaddset(ss, p->p_xstat);
    534 			signotify(l);
    535 		}
    536 		p->p_nrlwps++;
    537 		break;
    538 	case LSSUSPENDED:
    539 		l->l_flag &= ~LW_WSUSPEND;
    540 		p->p_nrlwps++;
    541 		break;
    542 	case LSSLEEP:
    543 		KASSERT(l->l_wchan != NULL);
    544 		break;
    545 	default:
    546 		panic("setrunnable: lwp %p state was %d", l, l->l_stat);
    547 	}
    548 
    549 	/*
    550 	 * If the LWP was sleeping interruptably, then it's OK to start it
    551 	 * again.  If not, mark it as still sleeping.
    552 	 */
    553 	if (l->l_wchan != NULL) {
    554 		l->l_stat = LSSLEEP;
    555 		/* lwp_unsleep() will release the lock. */
    556 		lwp_unsleep(l);
    557 		return;
    558 	}
    559 
    560 	LOCK_ASSERT(lwp_locked(l, &sched_mutex));
    561 
    562 	/*
    563 	 * If the LWP is still on the CPU, mark it as LSONPROC.  It may be
    564 	 * about to call mi_switch(), in which case it will yield.
    565 	 *
    566 	 * XXXSMP Will need to change for preemption.
    567 	 */
    568 #ifdef MULTIPROCESSOR
    569 	if (l->l_cpu->ci_curlwp == l) {
    570 #else
    571 	if (l == curlwp) {
    572 #endif
    573 		l->l_stat = LSONPROC;
    574 		l->l_slptime = 0;
    575 		lwp_unlock(l);
    576 		return;
    577 	}
    578 
    579 	/*
    580 	 * Set the LWP runnable.  If it's swapped out, we need to wake the swapper
    581 	 * to bring it back in.  Otherwise, enter it into a run queue.
    582 	 */
    583 	sched_setrunnable(l);
    584 	l->l_stat = LSRUN;
    585 	l->l_slptime = 0;
    586 
    587 	if (l->l_flag & LW_INMEM) {
    588 		sched_enqueue(l);
    589 		resched_cpu(l);
    590 		lwp_unlock(l);
    591 	} else {
    592 		lwp_unlock(l);
    593 		uvm_kick_scheduler();
    594 	}
    595 }
    596 
    597 /*
    598  * suspendsched:
    599  *
    600  *	Convert all non-L_SYSTEM LSSLEEP or LSRUN LWPs to LSSUSPENDED.
    601  */
    602 void
    603 suspendsched(void)
    604 {
    605 #ifdef MULTIPROCESSOR
    606 	CPU_INFO_ITERATOR cii;
    607 	struct cpu_info *ci;
    608 #endif
    609 	struct lwp *l;
    610 	struct proc *p;
    611 
    612 	/*
    613 	 * We do this by process in order not to violate the locking rules.
    614 	 */
    615 	mutex_enter(&proclist_mutex);
    616 	PROCLIST_FOREACH(p, &allproc) {
    617 		mutex_enter(&p->p_smutex);
    618 
    619 		if ((p->p_flag & PK_SYSTEM) != 0) {
    620 			mutex_exit(&p->p_smutex);
    621 			continue;
    622 		}
    623 
    624 		p->p_stat = SSTOP;
    625 
    626 		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
    627 			if (l == curlwp)
    628 				continue;
    629 
    630 			lwp_lock(l);
    631 
    632 			/*
    633 			 * Set L_WREBOOT so that the LWP will suspend itself
    634 			 * when it tries to return to user mode.  We want to
    635 			 * try and get to get as many LWPs as possible to
    636 			 * the user / kernel boundary, so that they will
    637 			 * release any locks that they hold.
    638 			 */
    639 			l->l_flag |= (LW_WREBOOT | LW_WSUSPEND);
    640 
    641 			if (l->l_stat == LSSLEEP &&
    642 			    (l->l_flag & LW_SINTR) != 0) {
    643 				/* setrunnable() will release the lock. */
    644 				setrunnable(l);
    645 				continue;
    646 			}
    647 
    648 			lwp_unlock(l);
    649 		}
    650 
    651 		mutex_exit(&p->p_smutex);
    652 	}
    653 	mutex_exit(&proclist_mutex);
    654 
    655 	/*
    656 	 * Kick all CPUs to make them preempt any LWPs running in user mode.
    657 	 * They'll trap into the kernel and suspend themselves in userret().
    658 	 */
    659 	sched_lock(0);
    660 #ifdef MULTIPROCESSOR
    661 	for (CPU_INFO_FOREACH(cii, ci))
    662 		cpu_need_resched(ci, 0);
    663 #else
    664 	cpu_need_resched(curcpu(), 0);
    665 #endif
    666 	sched_unlock(0);
    667 }
    668 
    669 /*
    670  * sched_kpri:
    671  *
    672  *	Scale a priority level to a kernel priority level, usually
    673  *	for an LWP that is about to sleep.
    674  */
    675 pri_t
    676 sched_kpri(struct lwp *l)
    677 {
    678 	/*
    679 	 * Scale user priorities (127 -> 50) up to kernel priorities
    680 	 * in the range (49 -> 8).  Reserve the top 8 kernel priorities
    681 	 * for high priority kthreads.  Kernel priorities passed in
    682 	 * are left "as is".  XXX This is somewhat arbitrary.
    683 	 */
    684 	static const uint8_t kpri_tab[] = {
    685 		 0,   1,   2,   3,   4,   5,   6,   7,
    686 		 8,   9,  10,  11,  12,  13,  14,  15,
    687 		16,  17,  18,  19,  20,  21,  22,  23,
    688 		24,  25,  26,  27,  28,  29,  30,  31,
    689 		32,  33,  34,  35,  36,  37,  38,  39,
    690 		40,  41,  42,  43,  44,  45,  46,  47,
    691 		48,  49,   8,   8,   9,   9,  10,  10,
    692 		11,  11,  12,  12,  13,  14,  14,  15,
    693 		15,  16,  16,  17,  17,  18,  18,  19,
    694 		20,  20,  21,  21,  22,  22,  23,  23,
    695 		24,  24,  25,  26,  26,  27,  27,  28,
    696 		28,  29,  29,  30,  30,  31,  32,  32,
    697 		33,  33,  34,  34,  35,  35,  36,  36,
    698 		37,  38,  38,  39,  39,  40,  40,  41,
    699 		41,  42,  42,  43,  44,  44,  45,  45,
    700 		46,  46,  47,  47,  48,  48,  49,  49,
    701 	};
    702 
    703 	return (pri_t)kpri_tab[l->l_usrpri];
    704 }
    705 
    706 /*
    707  * sched_unsleep:
    708  *
    709  *	The is called when the LWP has not been awoken normally but instead
    710  *	interrupted: for example, if the sleep timed out.  Because of this,
    711  *	it's not a valid action for running or idle LWPs.
    712  */
    713 static void
    714 sched_unsleep(struct lwp *l)
    715 {
    716 
    717 	lwp_unlock(l);
    718 	panic("sched_unsleep");
    719 }
    720 
    721 inline void
    722 resched_cpu(struct lwp *l)
    723 {
    724 	struct cpu_info *ci;
    725 	const pri_t pri = lwp_eprio(l);
    726 
    727 	/*
    728 	 * XXXSMP
    729 	 * Since l->l_cpu persists across a context switch,
    730 	 * this gives us *very weak* processor affinity, in
    731 	 * that we notify the CPU on which the process last
    732 	 * ran that it should try to switch.
    733 	 *
    734 	 * This does not guarantee that the process will run on
    735 	 * that processor next, because another processor might
    736 	 * grab it the next time it performs a context switch.
    737 	 *
    738 	 * This also does not handle the case where its last
    739 	 * CPU is running a higher-priority process, but every
    740 	 * other CPU is running a lower-priority process.  There
    741 	 * are ways to handle this situation, but they're not
    742 	 * currently very pretty, and we also need to weigh the
    743 	 * cost of moving a process from one CPU to another.
    744 	 *
    745 	 * XXXSMP
    746 	 * There is also the issue of locking the other CPU's
    747 	 * sched state, which we currently do not do.
    748 	 */
    749 	ci = (l->l_cpu != NULL) ? l->l_cpu : curcpu();
    750 	if (pri < ci->ci_schedstate.spc_curpriority)
    751 		cpu_need_resched(ci, 0);
    752 }
    753 
    754 static void
    755 sched_changepri(struct lwp *l, pri_t pri)
    756 {
    757 
    758 	LOCK_ASSERT(lwp_locked(l, &sched_mutex));
    759 
    760 	l->l_usrpri = pri;
    761 	if (l->l_priority < PUSER)
    762 		return;
    763 
    764 	if (l->l_stat != LSRUN || (l->l_flag & LW_INMEM) == 0) {
    765 		l->l_priority = pri;
    766 		return;
    767 	}
    768 
    769 	sched_dequeue(l);
    770 	l->l_priority = pri;
    771 	sched_enqueue(l);
    772 	resched_cpu(l);
    773 }
    774 
    775 static void
    776 sched_lendpri(struct lwp *l, pri_t pri)
    777 {
    778 
    779 	LOCK_ASSERT(lwp_locked(l, &sched_mutex));
    780 
    781 	if (l->l_stat != LSRUN || (l->l_flag & LW_INMEM) == 0) {
    782 		l->l_inheritedprio = pri;
    783 		return;
    784 	}
    785 
    786 	sched_dequeue(l);
    787 	l->l_inheritedprio = pri;
    788 	sched_enqueue(l);
    789 	resched_cpu(l);
    790 }
    791 
    792 struct lwp *
    793 syncobj_noowner(wchan_t wchan)
    794 {
    795 
    796 	return NULL;
    797 }
    798