Home | History | Annotate | Line # | Download | only in kern
kern_lock.c revision 1.131
      1 /*	$NetBSD: kern_lock.c,v 1.131 2008/01/04 21:18:08 ad Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999, 2000, 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, and by Andrew Doran.
     10  *
     11  * This code is derived from software contributed to The NetBSD Foundation
     12  * by Ross Harvey.
     13  *
     14  * Redistribution and use in source and binary forms, with or without
     15  * modification, are permitted provided that the following conditions
     16  * are met:
     17  * 1. Redistributions of source code must retain the above copyright
     18  *    notice, this list of conditions and the following disclaimer.
     19  * 2. Redistributions in binary form must reproduce the above copyright
     20  *    notice, this list of conditions and the following disclaimer in the
     21  *    documentation and/or other materials provided with the distribution.
     22  * 3. All advertising materials mentioning features or use of this software
     23  *    must display the following acknowledgement:
     24  *	This product includes software developed by the NetBSD
     25  *	Foundation, Inc. and its contributors.
     26  * 4. Neither the name of The NetBSD Foundation nor the names of its
     27  *    contributors may be used to endorse or promote products derived
     28  *    from this software without specific prior written permission.
     29  *
     30  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     31  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     32  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     33  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     34  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     35  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     36  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     37  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     38  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     39  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     40  * POSSIBILITY OF SUCH DAMAGE.
     41  */
     42 
     43 /*
     44  * Copyright (c) 1995
     45  *	The Regents of the University of California.  All rights reserved.
     46  *
     47  * This code contains ideas from software contributed to Berkeley by
     48  * Avadis Tevanian, Jr., Michael Wayne Young, and the Mach Operating
     49  * System project at Carnegie-Mellon University.
     50  *
     51  * Redistribution and use in source and binary forms, with or without
     52  * modification, are permitted provided that the following conditions
     53  * are met:
     54  * 1. Redistributions of source code must retain the above copyright
     55  *    notice, this list of conditions and the following disclaimer.
     56  * 2. Redistributions in binary form must reproduce the above copyright
     57  *    notice, this list of conditions and the following disclaimer in the
     58  *    documentation and/or other materials provided with the distribution.
     59  * 3. Neither the name of the University nor the names of its contributors
     60  *    may be used to endorse or promote products derived from this software
     61  *    without specific prior written permission.
     62  *
     63  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     64  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     65  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     66  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     67  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     68  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     69  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     70  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     71  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     72  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     73  * SUCH DAMAGE.
     74  *
     75  *	@(#)kern_lock.c	8.18 (Berkeley) 5/21/95
     76  */
     77 
     78 #include <sys/cdefs.h>
     79 __KERNEL_RCSID(0, "$NetBSD: kern_lock.c,v 1.131 2008/01/04 21:18:08 ad Exp $");
     80 
     81 #include "opt_multiprocessor.h"
     82 
     83 #include <sys/param.h>
     84 #include <sys/proc.h>
     85 #include <sys/lock.h>
     86 #include <sys/systm.h>
     87 #include <sys/kernel.h>
     88 #include <sys/lockdebug.h>
     89 #include <sys/cpu.h>
     90 #include <sys/syslog.h>
     91 #include <sys/atomic.h>
     92 
     93 #include <machine/stdarg.h>
     94 #include <machine/lock.h>
     95 
     96 #include <dev/lockstat.h>
     97 
     98 /*
     99  * note that stdarg.h and the ansi style va_start macro is used for both
    100  * ansi and traditional c compiles.
    101  * XXX: this requires that stdarg.h define: va_alist and va_dcl
    102  */
    103 void	lock_printf(const char *fmt, ...)
    104     __attribute__((__format__(__printf__,1,2)));
    105 
    106 static int acquire(struct lock **, int *, int, int, int, uintptr_t);
    107 
    108 int	lock_debug_syslog = 0;	/* defaults to printf, but can be patched */
    109 bool	kernel_lock_dodebug;
    110 __cpu_simple_lock_t kernel_lock;
    111 
    112 #ifdef LOCKDEBUG
    113 static lockops_t lockmgr_lockops = {
    114 	"lockmgr",
    115 	1,
    116 	(void *)nullop
    117 };
    118 #endif
    119 
    120 #if defined(LOCKDEBUG) || defined(DIAGNOSTIC) /* { */
    121 #define	COUNT(lkp, l, cpu_id, x)	(l)->l_locks += (x)
    122 #else
    123 #define COUNT(lkp, p, cpu_id, x)
    124 #endif /* LOCKDEBUG || DIAGNOSTIC */ /* } */
    125 
    126 #define	RETURN_ADDRESS		((uintptr_t)__builtin_return_address(0))
    127 
    128 /*
    129  * Acquire a resource.
    130  */
    131 static int
    132 acquire(struct lock **lkpp, int *s, int extflags,
    133 	int drain, int wanted, uintptr_t ra)
    134 {
    135 	int error;
    136 	struct lock *lkp = *lkpp;
    137 	LOCKSTAT_TIMER(slptime);
    138 	LOCKSTAT_FLAG(lsflag);
    139 
    140 	KASSERT(drain || (wanted & LK_WAIT_NONZERO) == 0);
    141 
    142 	LOCKSTAT_ENTER(lsflag);
    143 
    144 	for (error = 0; (lkp->lk_flags & wanted) != 0; ) {
    145 		if (drain)
    146 			lkp->lk_flags |= LK_WAITDRAIN;
    147 		else {
    148 			lkp->lk_waitcount++;
    149 			lkp->lk_flags |= LK_WAIT_NONZERO;
    150 		}
    151 		LOCKSTAT_START_TIMER(lsflag, slptime);
    152 		error = mtsleep(drain ? (void *)&lkp->lk_flags : (void *)lkp,
    153 		    lkp->lk_prio, lkp->lk_wmesg, lkp->lk_timo,
    154 		    __UNVOLATILE(&lkp->lk_interlock));
    155 		LOCKSTAT_STOP_TIMER(lsflag, slptime);
    156 		LOCKSTAT_EVENT_RA(lsflag, (void *)(uintptr_t)lkp,
    157 		    LB_LOCKMGR | LB_SLEEP1, 1, slptime, ra);
    158 		if (!drain) {
    159 			lkp->lk_waitcount--;
    160 			if (lkp->lk_waitcount == 0)
    161 				lkp->lk_flags &= ~LK_WAIT_NONZERO;
    162 		}
    163 		if (error)
    164 			break;
    165 		if (extflags & LK_SLEEPFAIL) {
    166 			error = ENOLCK;
    167 			break;
    168 		}
    169 	}
    170 
    171 	LOCKSTAT_EXIT(lsflag);
    172 
    173 	return error;
    174 }
    175 
    176 #define	SETHOLDER(lkp, pid, lid, cpu_id)				\
    177 do {									\
    178 	(lkp)->lk_lockholder = pid;					\
    179 	(lkp)->lk_locklwp = lid;					\
    180 } while (/*CONSTCOND*/0)
    181 
    182 #define	WEHOLDIT(lkp, pid, lid, cpu_id)					\
    183 	 ((lkp)->lk_lockholder == (pid) && (lkp)->lk_locklwp == (lid))
    184 
    185 #define	WAKEUP_WAITER(lkp)						\
    186 do {									\
    187 	if (((lkp)->lk_flags & LK_WAIT_NONZERO) != 0) {			\
    188 		wakeup((lkp));						\
    189 	}								\
    190 } while (/*CONSTCOND*/0)
    191 
    192 #if defined(LOCKDEBUG)
    193 /*
    194  * Lock debug printing routine; can be configured to print to console
    195  * or log to syslog.
    196  */
    197 void
    198 lock_printf(const char *fmt, ...)
    199 {
    200 	char b[150];
    201 	va_list ap;
    202 
    203 	va_start(ap, fmt);
    204 	if (lock_debug_syslog)
    205 		vlog(LOG_DEBUG, fmt, ap);
    206 	else {
    207 		vsnprintf(b, sizeof(b), fmt, ap);
    208 		printf_nolog("%s", b);
    209 	}
    210 	va_end(ap);
    211 }
    212 #endif /* LOCKDEBUG */
    213 
    214 static void
    215 lockpanic(struct lock *lkp, const char *fmt, ...)
    216 {
    217 	char s[150], b[150];
    218 	static const char *locktype[] = {
    219 	    "*0*", "shared", "exclusive", "*3*", "*4*", "downgrade",
    220 	    "*release*", "drain", "exclother", "*9*", "*10*",
    221 	    "*11*", "*12*", "*13*", "*14*", "*15*"
    222 	};
    223 	va_list ap;
    224 	va_start(ap, fmt);
    225 	vsnprintf(s, sizeof(s), fmt, ap);
    226 	va_end(ap);
    227 	bitmask_snprintf(lkp->lk_flags, __LK_FLAG_BITS, b, sizeof(b));
    228 	panic("%s ("
    229 	    "type %s flags %s, sharecount %d, exclusivecount %d, "
    230 	    "recurselevel %d, waitcount %d, wmesg %s"
    231 	    ", lock_addr %p, unlock_addr %p"
    232 	    ")\n",
    233 	    s, locktype[lkp->lk_flags & LK_TYPE_MASK],
    234 	    b, lkp->lk_sharecount, lkp->lk_exclusivecount,
    235 	    lkp->lk_recurselevel, lkp->lk_waitcount, lkp->lk_wmesg,
    236 	    (void *)lkp->lk_lock_addr, (void *)lkp->lk_unlock_addr
    237 	);
    238 }
    239 
    240 /*
    241  * Initialize a lock; required before use.
    242  */
    243 void
    244 lockinit(struct lock *lkp, pri_t prio, const char *wmesg, int timo, int flags)
    245 {
    246 
    247 	memset(lkp, 0, sizeof(struct lock));
    248 	lkp->lk_flags = flags & LK_EXTFLG_MASK;
    249 	mutex_init(&lkp->lk_interlock, MUTEX_DEFAULT, IPL_NONE);
    250 	lkp->lk_lockholder = LK_NOPROC;
    251 	lkp->lk_prio = prio;
    252 	lkp->lk_timo = timo;
    253 	lkp->lk_wmesg = wmesg;
    254 	lkp->lk_lock_addr = 0;
    255 	lkp->lk_unlock_addr = 0;
    256 
    257 	if (LOCKDEBUG_ALLOC(lkp, &lockmgr_lockops,
    258 	    (uintptr_t)__builtin_return_address(0))) {
    259 		lkp->lk_flags |= LK_DODEBUG;
    260 	}
    261 }
    262 
    263 void
    264 lockdestroy(struct lock *lkp)
    265 {
    266 
    267 	LOCKDEBUG_FREE(((lkp->lk_flags & LK_DODEBUG) != 0), lkp);
    268 	mutex_destroy(&lkp->lk_interlock);
    269 }
    270 
    271 /*
    272  * Determine the status of a lock.
    273  */
    274 int
    275 lockstatus(struct lock *lkp)
    276 {
    277 	int lock_type = 0;
    278 	struct lwp *l = curlwp; /* XXX */
    279 	pid_t pid;
    280 	lwpid_t lid;
    281 	cpuid_t cpu_num;
    282 
    283 	if (l == NULL) {
    284 		cpu_num = cpu_number();
    285 		pid = LK_KERNPROC;
    286 		lid = 0;
    287 	} else {
    288 		cpu_num = LK_NOCPU;
    289 		pid = l->l_proc->p_pid;
    290 		lid = l->l_lid;
    291 	}
    292 
    293 	mutex_enter(&lkp->lk_interlock);
    294 	if (lkp->lk_exclusivecount != 0) {
    295 		if (WEHOLDIT(lkp, pid, lid, cpu_num))
    296 			lock_type = LK_EXCLUSIVE;
    297 		else
    298 			lock_type = LK_EXCLOTHER;
    299 	} else if (lkp->lk_sharecount != 0)
    300 		lock_type = LK_SHARED;
    301 	else if (lkp->lk_flags & LK_WANT_EXCL)
    302 		lock_type = LK_EXCLOTHER;
    303 	mutex_exit(&lkp->lk_interlock);
    304 	return (lock_type);
    305 }
    306 
    307 /*
    308  * XXX XXX kludge around another kludge..
    309  *
    310  * vfs_shutdown() may be called from interrupt context, either as a result
    311  * of a panic, or from the debugger.   It proceeds to call
    312  * sys_sync(&proc0, ...), pretending its running on behalf of proc0
    313  *
    314  * We would like to make an attempt to sync the filesystems in this case, so
    315  * if this happens, we treat attempts to acquire locks specially.
    316  * All locks are acquired on behalf of proc0.
    317  *
    318  * If we've already paniced, we don't block waiting for locks, but
    319  * just barge right ahead since we're already going down in flames.
    320  */
    321 
    322 /*
    323  * Set, change, or release a lock.
    324  *
    325  * Shared requests increment the shared count. Exclusive requests set the
    326  * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
    327  * accepted shared locks to go away.
    328  */
    329 int
    330 lockmgr(struct lock *lkp, u_int flags, kmutex_t *interlkp)
    331 {
    332 	int error;
    333 	pid_t pid;
    334 	lwpid_t lid;
    335 	int extflags;
    336 	cpuid_t cpu_num;
    337 	struct lwp *l = curlwp;
    338 	int lock_shutdown_noblock = 0;
    339 	int s = 0;
    340 
    341 	error = 0;
    342 
    343 	/* LK_RETRY is for vn_lock, not for lockmgr. */
    344 	KASSERT((flags & LK_RETRY) == 0);
    345 	KASSERT((l->l_pflag & LP_INTR) == 0 || panicstr != NULL);
    346 
    347 	mutex_enter(&lkp->lk_interlock);
    348 	if (flags & LK_INTERLOCK)
    349 		mutex_exit(interlkp);
    350 	extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
    351 
    352 	if (l == NULL) {
    353 		if (!doing_shutdown) {
    354 			panic("lockmgr: no context");
    355 		} else {
    356 			l = &lwp0;
    357 			if (panicstr && (!(flags & LK_NOWAIT))) {
    358 				flags |= LK_NOWAIT;
    359 				lock_shutdown_noblock = 1;
    360 			}
    361 		}
    362 	}
    363 	lid = l->l_lid;
    364 	pid = l->l_proc->p_pid;
    365 	cpu_num = cpu_number();
    366 
    367 	/*
    368 	 * Once a lock has drained, the LK_DRAINING flag is set and an
    369 	 * exclusive lock is returned. The only valid operation thereafter
    370 	 * is a single release of that exclusive lock. This final release
    371 	 * clears the LK_DRAINING flag and sets the LK_DRAINED flag. Any
    372 	 * further requests of any sort will result in a panic. The bits
    373 	 * selected for these two flags are chosen so that they will be set
    374 	 * in memory that is freed (freed memory is filled with 0xdeadbeef).
    375 	 * The final release is permitted to give a new lease on life to
    376 	 * the lock by specifying LK_REENABLE.
    377 	 */
    378 	if (lkp->lk_flags & (LK_DRAINING|LK_DRAINED)) {
    379 #ifdef DIAGNOSTIC /* { */
    380 		if (lkp->lk_flags & LK_DRAINED)
    381 			lockpanic(lkp, "lockmgr: using decommissioned lock");
    382 		if ((flags & LK_TYPE_MASK) != LK_RELEASE ||
    383 		    WEHOLDIT(lkp, pid, lid, cpu_num) == 0)
    384 			lockpanic(lkp, "lockmgr: non-release on draining lock: %d",
    385 			    flags & LK_TYPE_MASK);
    386 #endif /* DIAGNOSTIC */ /* } */
    387 		lkp->lk_flags &= ~LK_DRAINING;
    388 		if ((flags & LK_REENABLE) == 0)
    389 			lkp->lk_flags |= LK_DRAINED;
    390 	}
    391 
    392 	switch (flags & LK_TYPE_MASK) {
    393 
    394 	case LK_SHARED:
    395 		if (WEHOLDIT(lkp, pid, lid, cpu_num) == 0) {
    396 			/*
    397 			 * If just polling, check to see if we will block.
    398 			 */
    399 			if ((extflags & LK_NOWAIT) && (lkp->lk_flags &
    400 			    (LK_HAVE_EXCL | LK_WANT_EXCL))) {
    401 				error = EBUSY;
    402 				break;
    403 			}
    404 			/*
    405 			 * Wait for exclusive locks to clear.
    406 			 */
    407 			error = acquire(&lkp, &s, extflags, 0,
    408 			    LK_HAVE_EXCL | LK_WANT_EXCL,
    409 			    RETURN_ADDRESS);
    410 			if (error)
    411 				break;
    412 			lkp->lk_sharecount++;
    413 			lkp->lk_flags |= LK_SHARE_NONZERO;
    414 			COUNT(lkp, l, cpu_num, 1);
    415 			break;
    416 		}
    417 		/*
    418 		 * We hold an exclusive lock, so downgrade it to shared.
    419 		 * An alternative would be to fail with EDEADLK.
    420 		 */
    421 		lkp->lk_sharecount++;
    422 		lkp->lk_flags |= LK_SHARE_NONZERO;
    423 		COUNT(lkp, l, cpu_num, 1);
    424 		/* fall into downgrade */
    425 
    426 	case LK_DOWNGRADE:
    427 		if (WEHOLDIT(lkp, pid, lid, cpu_num) == 0 ||
    428 		    lkp->lk_exclusivecount == 0)
    429 			lockpanic(lkp, "lockmgr: not holding exclusive lock");
    430 		lkp->lk_sharecount += lkp->lk_exclusivecount;
    431 		lkp->lk_flags |= LK_SHARE_NONZERO;
    432 		lkp->lk_exclusivecount = 0;
    433 		lkp->lk_recurselevel = 0;
    434 		lkp->lk_flags &= ~LK_HAVE_EXCL;
    435 		SETHOLDER(lkp, LK_NOPROC, 0, LK_NOCPU);
    436 #if defined(LOCKDEBUG)
    437 		lkp->lk_unlock_addr = RETURN_ADDRESS;
    438 #endif
    439 		WAKEUP_WAITER(lkp);
    440 		break;
    441 
    442 	case LK_EXCLUSIVE:
    443 		if (WEHOLDIT(lkp, pid, lid, cpu_num)) {
    444 			/*
    445 			 * Recursive lock.
    446 			 */
    447 			if ((extflags & LK_CANRECURSE) == 0 &&
    448 			     lkp->lk_recurselevel == 0) {
    449 				if (extflags & LK_RECURSEFAIL) {
    450 					error = EDEADLK;
    451 					break;
    452 				} else
    453 					lockpanic(lkp, "lockmgr: locking against myself");
    454 			}
    455 			lkp->lk_exclusivecount++;
    456 			COUNT(lkp, l, cpu_num, 1);
    457 			break;
    458 		}
    459 		/*
    460 		 * If we are just polling, check to see if we will sleep.
    461 		 */
    462 		if ((extflags & LK_NOWAIT) && (lkp->lk_flags &
    463 		     (LK_HAVE_EXCL | LK_WANT_EXCL | LK_SHARE_NONZERO))) {
    464 			error = EBUSY;
    465 			break;
    466 		}
    467 		/*
    468 		 * Try to acquire the want_exclusive flag.
    469 		 */
    470 		error = acquire(&lkp, &s, extflags, 0,
    471 		    LK_HAVE_EXCL | LK_WANT_EXCL, RETURN_ADDRESS);
    472 		if (error)
    473 			break;
    474 		lkp->lk_flags |= LK_WANT_EXCL;
    475 		/*
    476 		 * Wait for shared locks to finish.
    477 		 */
    478 		error = acquire(&lkp, &s, extflags, 0,
    479 		    LK_HAVE_EXCL | LK_SHARE_NONZERO,
    480 		    RETURN_ADDRESS);
    481 		lkp->lk_flags &= ~LK_WANT_EXCL;
    482 		if (error) {
    483 			WAKEUP_WAITER(lkp);
    484 			break;
    485 		}
    486 		lkp->lk_flags |= LK_HAVE_EXCL;
    487 		SETHOLDER(lkp, pid, lid, cpu_num);
    488 #if defined(LOCKDEBUG)
    489 		lkp->lk_lock_addr = RETURN_ADDRESS;
    490 #endif
    491 		if (lkp->lk_exclusivecount != 0)
    492 			lockpanic(lkp, "lockmgr: non-zero exclusive count");
    493 		lkp->lk_exclusivecount = 1;
    494 		COUNT(lkp, l, cpu_num, 1);
    495 		break;
    496 
    497 	case LK_RELEASE:
    498 		if (lkp->lk_exclusivecount != 0) {
    499 			if (WEHOLDIT(lkp, pid, lid, cpu_num) == 0) {
    500 				lockpanic(lkp, "lockmgr: pid %d.%d, not "
    501 				    "exclusive lock holder %d.%d "
    502 				    "unlocking", pid, lid,
    503 				    lkp->lk_lockholder,
    504 				    lkp->lk_locklwp);
    505 			}
    506 			if (lkp->lk_exclusivecount == lkp->lk_recurselevel)
    507 				lkp->lk_recurselevel = 0;
    508 			lkp->lk_exclusivecount--;
    509 			COUNT(lkp, l, cpu_num, -1);
    510 			if (lkp->lk_exclusivecount == 0) {
    511 				lkp->lk_flags &= ~LK_HAVE_EXCL;
    512 				SETHOLDER(lkp, LK_NOPROC, 0, LK_NOCPU);
    513 #if defined(LOCKDEBUG)
    514 				lkp->lk_unlock_addr = RETURN_ADDRESS;
    515 #endif
    516 			}
    517 		} else if (lkp->lk_sharecount != 0) {
    518 			lkp->lk_sharecount--;
    519 			if (lkp->lk_sharecount == 0)
    520 				lkp->lk_flags &= ~LK_SHARE_NONZERO;
    521 			COUNT(lkp, l, cpu_num, -1);
    522 		}
    523 #ifdef DIAGNOSTIC
    524 		else
    525 			lockpanic(lkp, "lockmgr: release of unlocked lock!");
    526 #endif
    527 		WAKEUP_WAITER(lkp);
    528 		break;
    529 
    530 	case LK_DRAIN:
    531 		/*
    532 		 * Check that we do not already hold the lock, as it can
    533 		 * never drain if we do. Unfortunately, we have no way to
    534 		 * check for holding a shared lock, but at least we can
    535 		 * check for an exclusive one.
    536 		 */
    537 		if (WEHOLDIT(lkp, pid, lid, cpu_num))
    538 			lockpanic(lkp, "lockmgr: draining against myself");
    539 		/*
    540 		 * If we are just polling, check to see if we will sleep.
    541 		 */
    542 		if ((extflags & LK_NOWAIT) && (lkp->lk_flags &
    543 		     (LK_HAVE_EXCL | LK_WANT_EXCL |
    544 		     LK_SHARE_NONZERO | LK_WAIT_NONZERO))) {
    545 			error = EBUSY;
    546 			break;
    547 		}
    548 		error = acquire(&lkp, &s, extflags, 1,
    549 		    LK_HAVE_EXCL | LK_WANT_EXCL |
    550 		    LK_SHARE_NONZERO | LK_WAIT_NONZERO,
    551 		    RETURN_ADDRESS);
    552 		if (error)
    553 			break;
    554 		lkp->lk_flags |= LK_HAVE_EXCL;
    555 		if ((extflags & LK_RESURRECT) == 0)
    556 			lkp->lk_flags |= LK_DRAINING;
    557 		SETHOLDER(lkp, pid, lid, cpu_num);
    558 #if defined(LOCKDEBUG)
    559 		lkp->lk_lock_addr = RETURN_ADDRESS;
    560 #endif
    561 		lkp->lk_exclusivecount = 1;
    562 		COUNT(lkp, l, cpu_num, 1);
    563 		break;
    564 
    565 	default:
    566 		mutex_exit(&lkp->lk_interlock);
    567 		lockpanic(lkp, "lockmgr: unknown locktype request %d",
    568 		    flags & LK_TYPE_MASK);
    569 		/* NOTREACHED */
    570 	}
    571 	if ((lkp->lk_flags & LK_WAITDRAIN) != 0 &&
    572 	    ((lkp->lk_flags &
    573 	      (LK_HAVE_EXCL | LK_WANT_EXCL |
    574 	      LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0)) {
    575 		lkp->lk_flags &= ~LK_WAITDRAIN;
    576 		wakeup(&lkp->lk_flags);
    577 	}
    578 	/*
    579 	 * Note that this panic will be a recursive panic, since
    580 	 * we only set lock_shutdown_noblock above if panicstr != NULL.
    581 	 */
    582 	if (error && lock_shutdown_noblock)
    583 		lockpanic(lkp, "lockmgr: deadlock (see previous panic)");
    584 
    585 	mutex_exit(&lkp->lk_interlock);
    586 	return (error);
    587 }
    588 
    589 /*
    590  * Print out information about state of a lock. Used by VOP_PRINT
    591  * routines to display ststus about contained locks.
    592  */
    593 void
    594 lockmgr_printinfo(struct lock *lkp)
    595 {
    596 
    597 	if (lkp->lk_sharecount)
    598 		printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
    599 		    lkp->lk_sharecount);
    600 	else if (lkp->lk_flags & LK_HAVE_EXCL) {
    601 		printf(" lock type %s: EXCL (count %d) by ",
    602 		    lkp->lk_wmesg, lkp->lk_exclusivecount);
    603 		printf("pid %d.%d", lkp->lk_lockholder,
    604 		    lkp->lk_locklwp);
    605 	} else
    606 		printf(" not locked");
    607 	if (lkp->lk_waitcount > 0)
    608 		printf(" with %d pending", lkp->lk_waitcount);
    609 }
    610 
    611 #if defined(LOCKDEBUG)
    612 void
    613 assert_sleepable(struct simplelock *interlock, const char *msg)
    614 {
    615 
    616 	if (panicstr != NULL)
    617 		return;
    618 	LOCKDEBUG_BARRIER(&kernel_lock, 1);
    619 	if (CURCPU_IDLE_P() && !cold) {
    620 		panic("assert_sleepable: idle");
    621 	}
    622 }
    623 #endif
    624 
    625 /*
    626  * rump doesn't need the kernel lock so force it out.  We cannot
    627  * currently easily include it for compilation because of
    628  * a) SPINLOCK_* b) membar_producer().  They are defined in different
    629  * places / way for each arch, so just simply do not bother to
    630  * fight a lot for no gain (i.e. pain but still no gain).
    631  */
    632 #ifndef _RUMPKERNEL
    633 /*
    634  * Functions for manipulating the kernel_lock.  We put them here
    635  * so that they show up in profiles.
    636  */
    637 
    638 #define	_KERNEL_LOCK_ABORT(msg)						\
    639     LOCKDEBUG_ABORT(&kernel_lock, &_kernel_lock_ops, __func__, msg)
    640 
    641 #ifdef LOCKDEBUG
    642 #define	_KERNEL_LOCK_ASSERT(cond)					\
    643 do {									\
    644 	if (!(cond))							\
    645 		_KERNEL_LOCK_ABORT("assertion failed: " #cond);		\
    646 } while (/* CONSTCOND */ 0)
    647 #else
    648 #define	_KERNEL_LOCK_ASSERT(cond)	/* nothing */
    649 #endif
    650 
    651 void	_kernel_lock_dump(volatile void *);
    652 
    653 lockops_t _kernel_lock_ops = {
    654 	"Kernel lock",
    655 	0,
    656 	_kernel_lock_dump
    657 };
    658 
    659 /*
    660  * Initialize the kernel lock.
    661  */
    662 void
    663 kernel_lock_init(void)
    664 {
    665 
    666 	__cpu_simple_lock_init(&kernel_lock);
    667 	kernel_lock_dodebug = LOCKDEBUG_ALLOC(&kernel_lock, &_kernel_lock_ops,
    668 	    RETURN_ADDRESS);
    669 }
    670 
    671 /*
    672  * Print debugging information about the kernel lock.
    673  */
    674 void
    675 _kernel_lock_dump(volatile void *junk)
    676 {
    677 	struct cpu_info *ci = curcpu();
    678 
    679 	(void)junk;
    680 
    681 	printf_nolog("curcpu holds : %18d wanted by: %#018lx\n",
    682 	    ci->ci_biglock_count, (long)ci->ci_biglock_wanted);
    683 }
    684 
    685 /*
    686  * Acquire 'nlocks' holds on the kernel lock.  If 'l' is non-null, the
    687  * acquisition is from process context.
    688  */
    689 void
    690 _kernel_lock(int nlocks, struct lwp *l)
    691 {
    692 	struct cpu_info *ci = curcpu();
    693 	LOCKSTAT_TIMER(spintime);
    694 	LOCKSTAT_FLAG(lsflag);
    695 	struct lwp *owant;
    696 #ifdef LOCKDEBUG
    697 	u_int spins;
    698 #endif
    699 	int s;
    700 
    701 	if (nlocks == 0)
    702 		return;
    703 	_KERNEL_LOCK_ASSERT(nlocks > 0);
    704 
    705 	l = curlwp;
    706 
    707 	if (ci->ci_biglock_count != 0) {
    708 		_KERNEL_LOCK_ASSERT(__SIMPLELOCK_LOCKED_P(&kernel_lock));
    709 		ci->ci_biglock_count += nlocks;
    710 		l->l_blcnt += nlocks;
    711 		return;
    712 	}
    713 
    714 	_KERNEL_LOCK_ASSERT(l->l_blcnt == 0);
    715 	LOCKDEBUG_WANTLOCK(kernel_lock_dodebug, &kernel_lock, RETURN_ADDRESS,
    716 	    0);
    717 
    718 	s = splvm();
    719 	if (__cpu_simple_lock_try(&kernel_lock)) {
    720 		ci->ci_biglock_count = nlocks;
    721 		l->l_blcnt = nlocks;
    722 		LOCKDEBUG_LOCKED(kernel_lock_dodebug, &kernel_lock,
    723 		    RETURN_ADDRESS, 0);
    724 		splx(s);
    725 		return;
    726 	}
    727 
    728 	LOCKSTAT_ENTER(lsflag);
    729 	LOCKSTAT_START_TIMER(lsflag, spintime);
    730 
    731 	/*
    732 	 * Before setting ci_biglock_wanted we must post a store
    733 	 * fence (see kern_mutex.c).  This is accomplished by the
    734 	 * __cpu_simple_lock_try() above.
    735 	 */
    736 	owant = ci->ci_biglock_wanted;
    737 	ci->ci_biglock_wanted = curlwp;	/* XXXAD */
    738 
    739 #ifdef LOCKDEBUG
    740 	spins = 0;
    741 #endif
    742 
    743 	do {
    744 		splx(s);
    745 		while (__SIMPLELOCK_LOCKED_P(&kernel_lock)) {
    746 #ifdef LOCKDEBUG
    747 			if (SPINLOCK_SPINOUT(spins))
    748 				_KERNEL_LOCK_ABORT("spinout");
    749 #endif
    750 			SPINLOCK_BACKOFF_HOOK;
    751 			SPINLOCK_SPIN_HOOK;
    752 		}
    753 		(void)splvm();
    754 	} while (!__cpu_simple_lock_try(&kernel_lock));
    755 
    756 	ci->ci_biglock_wanted = owant;
    757 	ci->ci_biglock_count = nlocks;
    758 	l->l_blcnt = nlocks;
    759 	LOCKSTAT_STOP_TIMER(lsflag, spintime);
    760 	LOCKDEBUG_LOCKED(kernel_lock_dodebug, &kernel_lock, RETURN_ADDRESS, 0);
    761 	splx(s);
    762 
    763 	/*
    764 	 * Again, another store fence is required (see kern_mutex.c).
    765 	 */
    766 	membar_producer();
    767 	if (owant == NULL) {
    768 		LOCKSTAT_EVENT(lsflag, &kernel_lock, LB_KERNEL_LOCK | LB_SPIN,
    769 		    1, spintime);
    770 	}
    771 	LOCKSTAT_EXIT(lsflag);
    772 }
    773 
    774 /*
    775  * Release 'nlocks' holds on the kernel lock.  If 'nlocks' is zero, release
    776  * all holds.  If 'l' is non-null, the release is from process context.
    777  */
    778 void
    779 _kernel_unlock(int nlocks, struct lwp *l, int *countp)
    780 {
    781 	struct cpu_info *ci = curcpu();
    782 	u_int olocks;
    783 	int s;
    784 
    785 	l = curlwp;
    786 
    787 	_KERNEL_LOCK_ASSERT(nlocks < 2);
    788 
    789 	olocks = l->l_blcnt;
    790 
    791 	if (olocks == 0) {
    792 		_KERNEL_LOCK_ASSERT(nlocks <= 0);
    793 		if (countp != NULL)
    794 			*countp = 0;
    795 		return;
    796 	}
    797 
    798 	_KERNEL_LOCK_ASSERT(__SIMPLELOCK_LOCKED_P(&kernel_lock));
    799 
    800 	if (nlocks == 0)
    801 		nlocks = olocks;
    802 	else if (nlocks == -1) {
    803 		nlocks = 1;
    804 		_KERNEL_LOCK_ASSERT(olocks == 1);
    805 	}
    806 
    807 	_KERNEL_LOCK_ASSERT(ci->ci_biglock_count >= l->l_blcnt);
    808 
    809 	l->l_blcnt -= nlocks;
    810 	if (ci->ci_biglock_count == nlocks) {
    811 		s = splvm();
    812 		LOCKDEBUG_UNLOCKED(kernel_lock_dodebug, &kernel_lock,
    813 		    RETURN_ADDRESS, 0);
    814 		ci->ci_biglock_count = 0;
    815 		__cpu_simple_unlock(&kernel_lock);
    816 		splx(s);
    817 	} else
    818 		ci->ci_biglock_count -= nlocks;
    819 
    820 	if (countp != NULL)
    821 		*countp = olocks;
    822 }
    823 #endif /* !_RUMPKERNEL */
    824