Home | History | Annotate | Line # | Download | only in kern
kern_lock.c revision 1.99.2.4
      1 /*	$NetBSD: kern_lock.c,v 1.99.2.4 2006/11/17 16:34:36 ad Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999, 2000, 2006 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 1.99.2.4 2006/11/17 16:34:36 ad Exp $");
     80 
     81 #include "opt_multiprocessor.h"
     82 #include "opt_ddb.h"
     83 
     84 #define	__MUTEX_PRIVATE
     85 
     86 #include <sys/param.h>
     87 #include <sys/proc.h>
     88 #include <sys/lock.h>
     89 #include <sys/systm.h>
     90 #include <sys/lockdebug.h>
     91 
     92 #include <machine/cpu.h>
     93 
     94 #include <dev/lockstat.h>
     95 
     96 #if defined(LOCKDEBUG)
     97 #include <sys/syslog.h>
     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 #include <machine/stdarg.h>
    104 
    105 void	lock_printf(const char *fmt, ...)
    106     __attribute__((__format__(__printf__,1,2)));
    107 
    108 static int acquire(volatile struct lock **, int *, int, int, int, uintptr_t);
    109 
    110 int	lock_debug_syslog = 0;	/* defaults to printf, but can be patched */
    111 
    112 #ifdef DDB
    113 #include <ddb/ddbvar.h>
    114 #include <machine/db_machdep.h>
    115 #include <ddb/db_command.h>
    116 #include <ddb/db_interface.h>
    117 #endif
    118 #endif /* defined(LOCKDEBUG) */
    119 
    120 #if defined(MULTIPROCESSOR)
    121 /*
    122  * IPL_BIGLOCK: block IPLs which need to grab kernel_mutex.
    123  * XXX IPL_VM or IPL_AUDIO should be enough.
    124  */
    125 #if !defined(__HAVE_SPLBIGLOCK)
    126 #define	splbiglock	splclock
    127 #endif
    128 __cpu_simple_lock_t kernel_lock;
    129 int kernel_lock_id;
    130 #endif
    131 
    132 /*
    133  * Locking primitives implementation.
    134  * Locks provide shared/exclusive synchronization.
    135  */
    136 
    137 #if defined(LOCKDEBUG) || defined(DIAGNOSTIC) /* { */
    138 #if defined(MULTIPROCESSOR) /* { */
    139 #define	COUNT_CPU(cpu_id, x)						\
    140 	curcpu()->ci_spin_locks += (x)
    141 #else
    142 u_long	spin_locks;
    143 #define	COUNT_CPU(cpu_id, x)	spin_locks += (x)
    144 #endif /* MULTIPROCESSOR */ /* } */
    145 
    146 #define	COUNT(lkp, l, cpu_id, x)					\
    147 do {									\
    148 	if ((lkp)->lk_flags & LK_SPIN)					\
    149 		COUNT_CPU((cpu_id), (x));				\
    150 	else								\
    151 		(l)->l_locks += (x);					\
    152 } while (/*CONSTCOND*/0)
    153 #else
    154 #define COUNT(lkp, p, cpu_id, x)
    155 #define COUNT_CPU(cpu_id, x)
    156 #endif /* LOCKDEBUG || DIAGNOSTIC */ /* } */
    157 
    158 #define	INTERLOCK_ACQUIRE(lkp, flags, s)				\
    159 do {									\
    160 	if ((flags) & LK_SPIN)						\
    161 		s = spllock();						\
    162 	simple_lock(&(lkp)->lk_interlock);				\
    163 } while (/*CONSTCOND*/ 0)
    164 
    165 #define	INTERLOCK_RELEASE(lkp, flags, s)				\
    166 do {									\
    167 	simple_unlock(&(lkp)->lk_interlock);				\
    168 	if ((flags) & LK_SPIN)						\
    169 		splx(s);						\
    170 } while (/*CONSTCOND*/ 0)
    171 
    172 #ifdef DDB /* { */
    173 #if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
    174 int simple_lock_debugger = 1;	/* more serious on MP */
    175 #else
    176 int simple_lock_debugger = 0;
    177 #endif
    178 #define	SLOCK_DEBUGGER()	if (simple_lock_debugger && db_onpanic) Debugger()
    179 #define	SLOCK_TRACE()							\
    180 	db_stack_trace_print((db_expr_t)__builtin_frame_address(0),	\
    181 	    TRUE, 65535, "", lock_printf);
    182 #else
    183 #define	SLOCK_DEBUGGER()	/* nothing */
    184 #define	SLOCK_TRACE()		/* nothing */
    185 #endif /* } */
    186 
    187 #if defined(LOCKDEBUG)
    188 #if defined(DDB)
    189 #define	SPINLOCK_SPINCHECK_DEBUGGER	if (db_onpanic) Debugger()
    190 #else
    191 #define	SPINLOCK_SPINCHECK_DEBUGGER	/* nothing */
    192 #endif
    193 
    194 #define	SPINLOCK_SPINCHECK_DECL						\
    195 	/* 32-bits of count -- wrap constitutes a "spinout" */		\
    196 	uint32_t __spinc = 0
    197 
    198 #define	SPINLOCK_SPINCHECK						\
    199 do {									\
    200 	if (++__spinc == 0) {						\
    201 		lock_printf("LK_SPIN spinout, excl %d, share %d\n",	\
    202 		    lkp->lk_exclusivecount, lkp->lk_sharecount);	\
    203 		if (lkp->lk_exclusivecount)				\
    204 			lock_printf("held by CPU %lu\n",		\
    205 			    (u_long) lkp->lk_cpu);			\
    206 		if (lkp->lk_lock_file)					\
    207 			lock_printf("last locked at %s:%d\n",		\
    208 			    lkp->lk_lock_file, lkp->lk_lock_line);	\
    209 		if (lkp->lk_unlock_file)				\
    210 			lock_printf("last unlocked at %s:%d\n",		\
    211 			    lkp->lk_unlock_file, lkp->lk_unlock_line);	\
    212 		SLOCK_TRACE();						\
    213 		SPINLOCK_SPINCHECK_DEBUGGER;				\
    214 	}								\
    215 } while (/*CONSTCOND*/ 0)
    216 #else
    217 #define	SPINLOCK_SPINCHECK_DECL			/* nothing */
    218 #define	SPINLOCK_SPINCHECK			/* nothing */
    219 #endif /* LOCKDEBUG && DDB */
    220 
    221 #define	RETURN_ADDRESS		((uintptr_t)__builtin_return_address(0))
    222 
    223 /*
    224  * Acquire a resource.
    225  */
    226 static int
    227 acquire(volatile struct lock **lkpp, int *s, int extflags,
    228     int drain, int wanted, uintptr_t ra)
    229 {
    230 	int error;
    231 	volatile struct lock *lkp = *lkpp;
    232 	LOCKSTAT_TIMER(slptime);
    233 
    234 	KASSERT(drain || (wanted & LK_WAIT_NONZERO) == 0);
    235 
    236 	if (extflags & LK_SPIN) {
    237 		int interlocked;
    238 
    239 		SPINLOCK_SPINCHECK_DECL;
    240 
    241 		if (!drain) {
    242 			lkp->lk_waitcount++;
    243 			lkp->lk_flags |= LK_WAIT_NONZERO;
    244 		}
    245 		for (interlocked = 1;;) {
    246 			SPINLOCK_SPINCHECK;
    247 			if ((lkp->lk_flags & wanted) != 0) {
    248 				if (interlocked) {
    249 					INTERLOCK_RELEASE(lkp, LK_SPIN, *s);
    250 					interlocked = 0;
    251 				}
    252 				SPINLOCK_SPIN_HOOK;
    253 			} else if (interlocked) {
    254 				break;
    255 			} else {
    256 				INTERLOCK_ACQUIRE(lkp, LK_SPIN, *s);
    257 				interlocked = 1;
    258 			}
    259 		}
    260 		if (!drain) {
    261 			lkp->lk_waitcount--;
    262 			if (lkp->lk_waitcount == 0)
    263 				lkp->lk_flags &= ~LK_WAIT_NONZERO;
    264 		}
    265 		KASSERT((lkp->lk_flags & wanted) == 0);
    266 		error = 0;	/* sanity */
    267 	} else {
    268 		for (error = 0; (lkp->lk_flags & wanted) != 0; ) {
    269 			if (drain)
    270 				lkp->lk_flags |= LK_WAITDRAIN;
    271 			else {
    272 				lkp->lk_waitcount++;
    273 				lkp->lk_flags |= LK_WAIT_NONZERO;
    274 			}
    275 			/* XXX Cast away volatile. */
    276 			LOCKSTAT_START_TIMER(slptime);
    277 			error = ltsleep(drain ?
    278 			    (volatile const void *)&lkp->lk_flags :
    279 			    (volatile const void *)lkp, lkp->lk_prio,
    280 			    lkp->lk_wmesg, lkp->lk_timo, &lkp->lk_interlock);
    281 			LOCKSTAT_STOP_TIMER(slptime);
    282 			LOCKSTAT_EVENT_RA((void *)(uintptr_t)lkp,
    283 			    LB_LOCKMGR | LB_SLEEP, 1, slptime, ra);
    284 			if (!drain) {
    285 				lkp->lk_waitcount--;
    286 				if (lkp->lk_waitcount == 0)
    287 					lkp->lk_flags &= ~LK_WAIT_NONZERO;
    288 			}
    289 			if (error)
    290 				break;
    291 			if (extflags & LK_SLEEPFAIL) {
    292 				error = ENOLCK;
    293 				break;
    294 			}
    295 			if (lkp->lk_newlock != NULL) {
    296 				simple_lock(&lkp->lk_newlock->lk_interlock);
    297 				simple_unlock(&lkp->lk_interlock);
    298 				if (lkp->lk_waitcount == 0)
    299 					wakeup(&lkp->lk_newlock);
    300 				*lkpp = lkp = lkp->lk_newlock;
    301 			}
    302 		}
    303 	}
    304 
    305 	return error;
    306 }
    307 
    308 #define	SETHOLDER(lkp, pid, lid, cpu_id)				\
    309 do {									\
    310 	if ((lkp)->lk_flags & LK_SPIN)					\
    311 		(lkp)->lk_cpu = cpu_id;					\
    312 	else {								\
    313 		(lkp)->lk_lockholder = pid;				\
    314 		(lkp)->lk_locklwp = lid;				\
    315 	}								\
    316 } while (/*CONSTCOND*/0)
    317 
    318 #define	WEHOLDIT(lkp, pid, lid, cpu_id)					\
    319 	(((lkp)->lk_flags & LK_SPIN) != 0 ?				\
    320 	 ((lkp)->lk_cpu == (cpu_id)) :					\
    321 	 ((lkp)->lk_lockholder == (pid) && (lkp)->lk_locklwp == (lid)))
    322 
    323 #define	WAKEUP_WAITER(lkp)						\
    324 do {									\
    325 	if (((lkp)->lk_flags & (LK_SPIN | LK_WAIT_NONZERO)) ==		\
    326 	    LK_WAIT_NONZERO) {						\
    327 		wakeup((lkp));						\
    328 	}								\
    329 } while (/*CONSTCOND*/0)
    330 
    331 #if defined(LOCKDEBUG) /* { */
    332 #if defined(MULTIPROCESSOR) /* { */
    333 struct simplelock spinlock_list_slock = SIMPLELOCK_INITIALIZER;
    334 
    335 #define	SPINLOCK_LIST_LOCK()						\
    336 	__cpu_simple_lock(&spinlock_list_slock.lock_data)
    337 
    338 #define	SPINLOCK_LIST_UNLOCK()						\
    339 	__cpu_simple_unlock(&spinlock_list_slock.lock_data)
    340 #else
    341 #define	SPINLOCK_LIST_LOCK()	/* nothing */
    342 
    343 #define	SPINLOCK_LIST_UNLOCK()	/* nothing */
    344 #endif /* MULTIPROCESSOR */ /* } */
    345 
    346 _TAILQ_HEAD(, struct lock, volatile) spinlock_list =
    347     TAILQ_HEAD_INITIALIZER(spinlock_list);
    348 
    349 #define	HAVEIT(lkp)							\
    350 do {									\
    351 	if ((lkp)->lk_flags & LK_SPIN) {				\
    352 		int sp = spllock();					\
    353 		SPINLOCK_LIST_LOCK();					\
    354 		TAILQ_INSERT_TAIL(&spinlock_list, (lkp), lk_list);	\
    355 		SPINLOCK_LIST_UNLOCK();					\
    356 		splx(sp);						\
    357 	}								\
    358 } while (/*CONSTCOND*/0)
    359 
    360 #define	DONTHAVEIT(lkp)							\
    361 do {									\
    362 	if ((lkp)->lk_flags & LK_SPIN) {				\
    363 		int sp = spllock();					\
    364 		SPINLOCK_LIST_LOCK();					\
    365 		TAILQ_REMOVE(&spinlock_list, (lkp), lk_list);		\
    366 		SPINLOCK_LIST_UNLOCK();					\
    367 		splx(sp);						\
    368 	}								\
    369 } while (/*CONSTCOND*/0)
    370 #else
    371 #define	HAVEIT(lkp)		/* nothing */
    372 
    373 #define	DONTHAVEIT(lkp)		/* nothing */
    374 #endif /* LOCKDEBUG */ /* } */
    375 
    376 #if defined(LOCKDEBUG)
    377 /*
    378  * Lock debug printing routine; can be configured to print to console
    379  * or log to syslog.
    380  */
    381 void
    382 lock_printf(const char *fmt, ...)
    383 {
    384 	char b[150];
    385 	va_list ap;
    386 
    387 	va_start(ap, fmt);
    388 	if (lock_debug_syslog)
    389 		vlog(LOG_DEBUG, fmt, ap);
    390 	else {
    391 		vsnprintf(b, sizeof(b), fmt, ap);
    392 		printf_nolog("%s", b);
    393 	}
    394 	va_end(ap);
    395 }
    396 #endif /* LOCKDEBUG */
    397 
    398 /*
    399  * Transfer any waiting processes from one lock to another.
    400  */
    401 void
    402 transferlockers(struct lock *from, struct lock *to)
    403 {
    404 
    405 	KASSERT(from != to);
    406 	KASSERT((from->lk_flags & LK_WAITDRAIN) == 0);
    407 	if (from->lk_waitcount == 0)
    408 		return;
    409 	from->lk_newlock = to;
    410 	wakeup((void *)from);
    411 	tsleep((void *)&from->lk_newlock, from->lk_prio, "lkxfer", 0);
    412 	from->lk_newlock = NULL;
    413 	from->lk_flags &= ~(LK_WANT_EXCL | LK_WANT_UPGRADE);
    414 	KASSERT(from->lk_waitcount == 0);
    415 }
    416 
    417 
    418 /*
    419  * Initialize a lock; required before use.
    420  */
    421 void
    422 lockinit(struct lock *lkp, int prio, const char *wmesg, int timo, int flags)
    423 {
    424 
    425 	memset(lkp, 0, sizeof(struct lock));
    426 	simple_lock_init(&lkp->lk_interlock);
    427 	lkp->lk_flags = flags & LK_EXTFLG_MASK;
    428 	if (flags & LK_SPIN)
    429 		lkp->lk_cpu = LK_NOCPU;
    430 	else {
    431 		lkp->lk_lockholder = LK_NOPROC;
    432 		lkp->lk_newlock = NULL;
    433 		lkp->lk_prio = prio;
    434 		lkp->lk_timo = timo;
    435 	}
    436 	lkp->lk_wmesg = wmesg;	/* just a name for spin locks */
    437 #if defined(LOCKDEBUG)
    438 	lkp->lk_lock_file = NULL;
    439 	lkp->lk_unlock_file = NULL;
    440 #endif
    441 }
    442 
    443 /*
    444  * Determine the status of a lock.
    445  */
    446 int
    447 lockstatus(struct lock *lkp)
    448 {
    449 	int s = 0; /* XXX: gcc */
    450 	int lock_type = 0;
    451 	struct lwp *l = curlwp; /* XXX */
    452 	pid_t pid;
    453 	lwpid_t lid;
    454 	cpuid_t cpu_num;
    455 
    456 	if ((lkp->lk_flags & LK_SPIN) || l == NULL) {
    457 		cpu_num = cpu_number();
    458 		pid = LK_KERNPROC;
    459 		lid = 0;
    460 	} else {
    461 		cpu_num = LK_NOCPU;
    462 		pid = l->l_proc->p_pid;
    463 		lid = l->l_lid;
    464 	}
    465 
    466 	INTERLOCK_ACQUIRE(lkp, lkp->lk_flags, s);
    467 	if (lkp->lk_exclusivecount != 0) {
    468 		if (WEHOLDIT(lkp, pid, lid, cpu_num))
    469 			lock_type = LK_EXCLUSIVE;
    470 		else
    471 			lock_type = LK_EXCLOTHER;
    472 	} else if (lkp->lk_sharecount != 0)
    473 		lock_type = LK_SHARED;
    474 	INTERLOCK_RELEASE(lkp, lkp->lk_flags, s);
    475 	return (lock_type);
    476 }
    477 
    478 #if defined(LOCKDEBUG)
    479 /*
    480  * Make sure no spin locks are held by a CPU that is about
    481  * to context switch.
    482  */
    483 void
    484 spinlock_switchcheck(void)
    485 {
    486 	u_long cnt;
    487 	int s;
    488 
    489 	s = spllock();
    490 #if defined(MULTIPROCESSOR)
    491 	cnt = curcpu()->ci_spin_locks;
    492 #else
    493 	cnt = spin_locks;
    494 #endif
    495 	splx(s);
    496 
    497 	if (cnt != 0)
    498 		panic("spinlock_switchcheck: CPU %lu has %lu spin locks",
    499 		    (u_long) cpu_number(), cnt);
    500 }
    501 #endif /* LOCKDEBUG */
    502 
    503 /*
    504  * Locks and IPLs (interrupt priority levels):
    505  *
    506  * Locks which may be taken from interrupt context must be handled
    507  * very carefully; you must spl to the highest IPL where the lock
    508  * is needed before acquiring the lock.
    509  *
    510  * It is also important to avoid deadlock, since certain (very high
    511  * priority) interrupts are often needed to keep the system as a whole
    512  * from deadlocking, and must not be blocked while you are spinning
    513  * waiting for a lower-priority lock.
    514  *
    515  * In addition, the lock-debugging hooks themselves need to use locks!
    516  *
    517  * A raw __cpu_simple_lock may be used from interrupts are long as it
    518  * is acquired and held at a single IPL.
    519  *
    520  * A simple_lock (which is a __cpu_simple_lock wrapped with some
    521  * debugging hooks) may be used at or below spllock(), which is
    522  * typically at or just below splhigh() (i.e. blocks everything
    523  * but certain machine-dependent extremely high priority interrupts).
    524  *
    525  * spinlockmgr spinlocks should be used at or below splsched().
    526  *
    527  * Some platforms may have interrupts of higher priority than splsched(),
    528  * including hard serial interrupts, inter-processor interrupts, and
    529  * kernel debugger traps.
    530  */
    531 
    532 /*
    533  * XXX XXX kludge around another kludge..
    534  *
    535  * vfs_shutdown() may be called from interrupt context, either as a result
    536  * of a panic, or from the debugger.   It proceeds to call
    537  * sys_sync(&proc0, ...), pretending its running on behalf of proc0
    538  *
    539  * We would like to make an attempt to sync the filesystems in this case, so
    540  * if this happens, we treat attempts to acquire locks specially.
    541  * All locks are acquired on behalf of proc0.
    542  *
    543  * If we've already paniced, we don't block waiting for locks, but
    544  * just barge right ahead since we're already going down in flames.
    545  */
    546 
    547 /*
    548  * Set, change, or release a lock.
    549  *
    550  * Shared requests increment the shared count. Exclusive requests set the
    551  * LK_WANT_EXCL flag (preventing further shared locks), and wait for already
    552  * accepted shared locks and shared-to-exclusive upgrades to go away.
    553  */
    554 int
    555 #if defined(LOCKDEBUG)
    556 _lockmgr(volatile struct lock *lkp, u_int flags,
    557     struct simplelock *interlkp, const char *file, int line)
    558 #else
    559 lockmgr(volatile struct lock *lkp, u_int flags,
    560     struct simplelock *interlkp)
    561 #endif
    562 {
    563 	int error;
    564 	pid_t pid;
    565 	lwpid_t lid;
    566 	int extflags;
    567 	cpuid_t cpu_num;
    568 	struct lwp *l = curlwp;
    569 	int lock_shutdown_noblock = 0;
    570 	int s = 0;
    571 
    572 	error = 0;
    573 
    574 	/* LK_RETRY is for vn_lock, not for lockmgr. */
    575 	KASSERT((flags & LK_RETRY) == 0);
    576 
    577 	INTERLOCK_ACQUIRE(lkp, lkp->lk_flags, s);
    578 	if (flags & LK_INTERLOCK)
    579 		simple_unlock(interlkp);
    580 	extflags = (flags | lkp->lk_flags) & LK_EXTFLG_MASK;
    581 
    582 #ifdef DIAGNOSTIC /* { */
    583 	/*
    584 	 * Don't allow spins on sleep locks and don't allow sleeps
    585 	 * on spin locks.
    586 	 */
    587 	if ((flags ^ lkp->lk_flags) & LK_SPIN)
    588 		panic("lockmgr: sleep/spin mismatch");
    589 #endif /* } */
    590 
    591 	if (extflags & LK_SPIN) {
    592 		pid = LK_KERNPROC;
    593 		lid = 0;
    594 	} else {
    595 		if (l == NULL) {
    596 			if (!doing_shutdown) {
    597 				panic("lockmgr: no context");
    598 			} else {
    599 				l = &lwp0;
    600 				if (panicstr && (!(flags & LK_NOWAIT))) {
    601 					flags |= LK_NOWAIT;
    602 					lock_shutdown_noblock = 1;
    603 				}
    604 			}
    605 		}
    606 		lid = l->l_lid;
    607 		pid = l->l_proc->p_pid;
    608 	}
    609 	cpu_num = cpu_number();
    610 
    611 	/*
    612 	 * Once a lock has drained, the LK_DRAINING flag is set and an
    613 	 * exclusive lock is returned. The only valid operation thereafter
    614 	 * is a single release of that exclusive lock. This final release
    615 	 * clears the LK_DRAINING flag and sets the LK_DRAINED flag. Any
    616 	 * further requests of any sort will result in a panic. The bits
    617 	 * selected for these two flags are chosen so that they will be set
    618 	 * in memory that is freed (freed memory is filled with 0xdeadbeef).
    619 	 * The final release is permitted to give a new lease on life to
    620 	 * the lock by specifying LK_REENABLE.
    621 	 */
    622 	if (lkp->lk_flags & (LK_DRAINING|LK_DRAINED)) {
    623 #ifdef DIAGNOSTIC /* { */
    624 		if (lkp->lk_flags & LK_DRAINED)
    625 			panic("lockmgr: using decommissioned lock");
    626 		if ((flags & LK_TYPE_MASK) != LK_RELEASE ||
    627 		    WEHOLDIT(lkp, pid, lid, cpu_num) == 0)
    628 			panic("lockmgr: non-release on draining lock: %d",
    629 			    flags & LK_TYPE_MASK);
    630 #endif /* DIAGNOSTIC */ /* } */
    631 		lkp->lk_flags &= ~LK_DRAINING;
    632 		if ((flags & LK_REENABLE) == 0)
    633 			lkp->lk_flags |= LK_DRAINED;
    634 	}
    635 
    636 	switch (flags & LK_TYPE_MASK) {
    637 
    638 	case LK_SHARED:
    639 		if (WEHOLDIT(lkp, pid, lid, cpu_num) == 0) {
    640 			/*
    641 			 * If just polling, check to see if we will block.
    642 			 */
    643 			if ((extflags & LK_NOWAIT) && (lkp->lk_flags &
    644 			    (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE))) {
    645 				error = EBUSY;
    646 				break;
    647 			}
    648 			/*
    649 			 * Wait for exclusive locks and upgrades to clear.
    650 			 */
    651 			error = acquire(&lkp, &s, extflags, 0,
    652 			    LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE,
    653 			    RETURN_ADDRESS);
    654 			if (error)
    655 				break;
    656 			lkp->lk_sharecount++;
    657 			lkp->lk_flags |= LK_SHARE_NONZERO;
    658 			COUNT(lkp, l, cpu_num, 1);
    659 			break;
    660 		}
    661 		/*
    662 		 * We hold an exclusive lock, so downgrade it to shared.
    663 		 * An alternative would be to fail with EDEADLK.
    664 		 */
    665 		lkp->lk_sharecount++;
    666 		lkp->lk_flags |= LK_SHARE_NONZERO;
    667 		COUNT(lkp, l, cpu_num, 1);
    668 		/* fall into downgrade */
    669 
    670 	case LK_DOWNGRADE:
    671 		if (WEHOLDIT(lkp, pid, lid, cpu_num) == 0 ||
    672 		    lkp->lk_exclusivecount == 0)
    673 			panic("lockmgr: not holding exclusive lock");
    674 		lkp->lk_sharecount += lkp->lk_exclusivecount;
    675 		lkp->lk_flags |= LK_SHARE_NONZERO;
    676 		lkp->lk_exclusivecount = 0;
    677 		lkp->lk_recurselevel = 0;
    678 		lkp->lk_flags &= ~LK_HAVE_EXCL;
    679 		SETHOLDER(lkp, LK_NOPROC, 0, LK_NOCPU);
    680 #if defined(LOCKDEBUG)
    681 		lkp->lk_unlock_file = file;
    682 		lkp->lk_unlock_line = line;
    683 #endif
    684 		DONTHAVEIT(lkp);
    685 		WAKEUP_WAITER(lkp);
    686 		break;
    687 
    688 	case LK_EXCLUPGRADE:
    689 		/*
    690 		 * If another process is ahead of us to get an upgrade,
    691 		 * then we want to fail rather than have an intervening
    692 		 * exclusive access.
    693 		 */
    694 		if (lkp->lk_flags & LK_WANT_UPGRADE) {
    695 			lkp->lk_sharecount--;
    696 			if (lkp->lk_sharecount == 0)
    697 				lkp->lk_flags &= ~LK_SHARE_NONZERO;
    698 			COUNT(lkp, l, cpu_num, -1);
    699 			error = EBUSY;
    700 			break;
    701 		}
    702 		/* fall into normal upgrade */
    703 
    704 	case LK_UPGRADE:
    705 		/*
    706 		 * Upgrade a shared lock to an exclusive one. If another
    707 		 * shared lock has already requested an upgrade to an
    708 		 * exclusive lock, our shared lock is released and an
    709 		 * exclusive lock is requested (which will be granted
    710 		 * after the upgrade). If we return an error, the file
    711 		 * will always be unlocked.
    712 		 */
    713 		if (WEHOLDIT(lkp, pid, lid, cpu_num) || lkp->lk_sharecount <= 0)
    714 			panic("lockmgr: upgrade exclusive lock");
    715 		lkp->lk_sharecount--;
    716 		if (lkp->lk_sharecount == 0)
    717 			lkp->lk_flags &= ~LK_SHARE_NONZERO;
    718 		COUNT(lkp, l, cpu_num, -1);
    719 		/*
    720 		 * If we are just polling, check to see if we will block.
    721 		 */
    722 		if ((extflags & LK_NOWAIT) &&
    723 		    ((lkp->lk_flags & LK_WANT_UPGRADE) ||
    724 		     lkp->lk_sharecount > 1)) {
    725 			error = EBUSY;
    726 			break;
    727 		}
    728 		if ((lkp->lk_flags & LK_WANT_UPGRADE) == 0) {
    729 			/*
    730 			 * We are first shared lock to request an upgrade, so
    731 			 * request upgrade and wait for the shared count to
    732 			 * drop to zero, then take exclusive lock.
    733 			 */
    734 			lkp->lk_flags |= LK_WANT_UPGRADE;
    735 			error = acquire(&lkp, &s, extflags, 0, LK_SHARE_NONZERO,
    736 			    RETURN_ADDRESS);
    737 			lkp->lk_flags &= ~LK_WANT_UPGRADE;
    738 			if (error) {
    739 				WAKEUP_WAITER(lkp);
    740 				break;
    741 			}
    742 			lkp->lk_flags |= LK_HAVE_EXCL;
    743 			SETHOLDER(lkp, pid, lid, cpu_num);
    744 #if defined(LOCKDEBUG)
    745 			lkp->lk_lock_file = file;
    746 			lkp->lk_lock_line = line;
    747 #endif
    748 			HAVEIT(lkp);
    749 			if (lkp->lk_exclusivecount != 0)
    750 				panic("lockmgr: non-zero exclusive count");
    751 			lkp->lk_exclusivecount = 1;
    752 			if (extflags & LK_SETRECURSE)
    753 				lkp->lk_recurselevel = 1;
    754 			COUNT(lkp, l, cpu_num, 1);
    755 			break;
    756 		}
    757 		/*
    758 		 * Someone else has requested upgrade. Release our shared
    759 		 * lock, awaken upgrade requestor if we are the last shared
    760 		 * lock, then request an exclusive lock.
    761 		 */
    762 		if (lkp->lk_sharecount == 0)
    763 			WAKEUP_WAITER(lkp);
    764 		/* fall into exclusive request */
    765 
    766 	case LK_EXCLUSIVE:
    767 		if (WEHOLDIT(lkp, pid, lid, cpu_num)) {
    768 			/*
    769 			 * Recursive lock.
    770 			 */
    771 			if ((extflags & LK_CANRECURSE) == 0 &&
    772 			     lkp->lk_recurselevel == 0) {
    773 				if (extflags & LK_RECURSEFAIL) {
    774 					error = EDEADLK;
    775 					break;
    776 				} else
    777 					panic("lockmgr: locking against myself");
    778 			}
    779 			lkp->lk_exclusivecount++;
    780 			if (extflags & LK_SETRECURSE &&
    781 			    lkp->lk_recurselevel == 0)
    782 				lkp->lk_recurselevel = lkp->lk_exclusivecount;
    783 			COUNT(lkp, l, cpu_num, 1);
    784 			break;
    785 		}
    786 		/*
    787 		 * If we are just polling, check to see if we will sleep.
    788 		 */
    789 		if ((extflags & LK_NOWAIT) && (lkp->lk_flags &
    790 		     (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
    791 		     LK_SHARE_NONZERO))) {
    792 			error = EBUSY;
    793 			break;
    794 		}
    795 		/*
    796 		 * Try to acquire the want_exclusive flag.
    797 		 */
    798 		error = acquire(&lkp, &s, extflags, 0,
    799 		    LK_HAVE_EXCL | LK_WANT_EXCL, RETURN_ADDRESS);
    800 		if (error)
    801 			break;
    802 		lkp->lk_flags |= LK_WANT_EXCL;
    803 		/*
    804 		 * Wait for shared locks and upgrades to finish.
    805 		 */
    806 		error = acquire(&lkp, &s, extflags, 0,
    807 		    LK_HAVE_EXCL | LK_WANT_UPGRADE | LK_SHARE_NONZERO,
    808 		    RETURN_ADDRESS);
    809 		lkp->lk_flags &= ~LK_WANT_EXCL;
    810 		if (error) {
    811 			WAKEUP_WAITER(lkp);
    812 			break;
    813 		}
    814 		lkp->lk_flags |= LK_HAVE_EXCL;
    815 		SETHOLDER(lkp, pid, lid, cpu_num);
    816 #if defined(LOCKDEBUG)
    817 		lkp->lk_lock_file = file;
    818 		lkp->lk_lock_line = line;
    819 #endif
    820 		HAVEIT(lkp);
    821 		if (lkp->lk_exclusivecount != 0)
    822 			panic("lockmgr: non-zero exclusive count");
    823 		lkp->lk_exclusivecount = 1;
    824 		if (extflags & LK_SETRECURSE)
    825 			lkp->lk_recurselevel = 1;
    826 		COUNT(lkp, l, cpu_num, 1);
    827 		break;
    828 
    829 	case LK_RELEASE:
    830 		if (lkp->lk_exclusivecount != 0) {
    831 			if (WEHOLDIT(lkp, pid, lid, cpu_num) == 0) {
    832 				if (lkp->lk_flags & LK_SPIN) {
    833 					panic("lockmgr: processor %lu, not "
    834 					    "exclusive lock holder %lu "
    835 					    "unlocking", cpu_num, lkp->lk_cpu);
    836 				} else {
    837 					panic("lockmgr: pid %d, not "
    838 					    "exclusive lock holder %d "
    839 					    "unlocking", pid,
    840 					    lkp->lk_lockholder);
    841 				}
    842 			}
    843 			if (lkp->lk_exclusivecount == lkp->lk_recurselevel)
    844 				lkp->lk_recurselevel = 0;
    845 			lkp->lk_exclusivecount--;
    846 			COUNT(lkp, l, cpu_num, -1);
    847 			if (lkp->lk_exclusivecount == 0) {
    848 				lkp->lk_flags &= ~LK_HAVE_EXCL;
    849 				SETHOLDER(lkp, LK_NOPROC, 0, LK_NOCPU);
    850 #if defined(LOCKDEBUG)
    851 				lkp->lk_unlock_file = file;
    852 				lkp->lk_unlock_line = line;
    853 #endif
    854 				DONTHAVEIT(lkp);
    855 			}
    856 		} else if (lkp->lk_sharecount != 0) {
    857 			lkp->lk_sharecount--;
    858 			if (lkp->lk_sharecount == 0)
    859 				lkp->lk_flags &= ~LK_SHARE_NONZERO;
    860 			COUNT(lkp, l, cpu_num, -1);
    861 		}
    862 #ifdef DIAGNOSTIC
    863 		else
    864 			panic("lockmgr: release of unlocked lock!");
    865 #endif
    866 		WAKEUP_WAITER(lkp);
    867 		break;
    868 
    869 	case LK_DRAIN:
    870 		/*
    871 		 * Check that we do not already hold the lock, as it can
    872 		 * never drain if we do. Unfortunately, we have no way to
    873 		 * check for holding a shared lock, but at least we can
    874 		 * check for an exclusive one.
    875 		 */
    876 		if (WEHOLDIT(lkp, pid, lid, cpu_num))
    877 			panic("lockmgr: draining against myself");
    878 		/*
    879 		 * If we are just polling, check to see if we will sleep.
    880 		 */
    881 		if ((extflags & LK_NOWAIT) && (lkp->lk_flags &
    882 		     (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
    883 		     LK_SHARE_NONZERO | LK_WAIT_NONZERO))) {
    884 			error = EBUSY;
    885 			break;
    886 		}
    887 		error = acquire(&lkp, &s, extflags, 1,
    888 		    LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
    889 		    LK_SHARE_NONZERO | LK_WAIT_NONZERO,
    890 		    RETURN_ADDRESS);
    891 		if (error)
    892 			break;
    893 		lkp->lk_flags |= LK_DRAINING | LK_HAVE_EXCL;
    894 		SETHOLDER(lkp, pid, lid, cpu_num);
    895 #if defined(LOCKDEBUG)
    896 		lkp->lk_lock_file = file;
    897 		lkp->lk_lock_line = line;
    898 #endif
    899 		HAVEIT(lkp);
    900 		lkp->lk_exclusivecount = 1;
    901 		/* XXX unlikely that we'd want this */
    902 		if (extflags & LK_SETRECURSE)
    903 			lkp->lk_recurselevel = 1;
    904 		COUNT(lkp, l, cpu_num, 1);
    905 		break;
    906 
    907 	default:
    908 		INTERLOCK_RELEASE(lkp, lkp->lk_flags, s);
    909 		panic("lockmgr: unknown locktype request %d",
    910 		    flags & LK_TYPE_MASK);
    911 		/* NOTREACHED */
    912 	}
    913 	if ((lkp->lk_flags & (LK_WAITDRAIN|LK_SPIN)) == LK_WAITDRAIN &&
    914 	    ((lkp->lk_flags &
    915 	      (LK_HAVE_EXCL | LK_WANT_EXCL | LK_WANT_UPGRADE |
    916 	      LK_SHARE_NONZERO | LK_WAIT_NONZERO)) == 0)) {
    917 		lkp->lk_flags &= ~LK_WAITDRAIN;
    918 		wakeup(&lkp->lk_flags);
    919 	}
    920 	/*
    921 	 * Note that this panic will be a recursive panic, since
    922 	 * we only set lock_shutdown_noblock above if panicstr != NULL.
    923 	 */
    924 	if (error && lock_shutdown_noblock)
    925 		panic("lockmgr: deadlock (see previous panic)");
    926 
    927 	INTERLOCK_RELEASE(lkp, lkp->lk_flags, s);
    928 	return (error);
    929 }
    930 
    931 /*
    932  * For a recursive spinlock held one or more times by the current CPU,
    933  * release all N locks, and return N.
    934  * Intended for use in mi_switch() shortly before context switching.
    935  */
    936 
    937 int
    938 #if defined(LOCKDEBUG)
    939 _spinlock_release_all(volatile struct lock *lkp, const char *file, int line)
    940 #else
    941 spinlock_release_all(volatile struct lock *lkp)
    942 #endif
    943 {
    944 	int s, count;
    945 	cpuid_t cpu_num;
    946 
    947 	KASSERT(lkp->lk_flags & LK_SPIN);
    948 
    949 	INTERLOCK_ACQUIRE(lkp, LK_SPIN, s);
    950 
    951 	cpu_num = cpu_number();
    952 	count = lkp->lk_exclusivecount;
    953 
    954 	if (count != 0) {
    955 #ifdef DIAGNOSTIC
    956 		if (WEHOLDIT(lkp, 0, 0, cpu_num) == 0) {
    957 			panic("spinlock_release_all: processor %lu, not "
    958 			    "exclusive lock holder %lu "
    959 			    "unlocking", (long)cpu_num, lkp->lk_cpu);
    960 		}
    961 #endif
    962 		lkp->lk_recurselevel = 0;
    963 		lkp->lk_exclusivecount = 0;
    964 		COUNT_CPU(cpu_num, -count);
    965 		lkp->lk_flags &= ~LK_HAVE_EXCL;
    966 		SETHOLDER(lkp, LK_NOPROC, 0, LK_NOCPU);
    967 #if defined(LOCKDEBUG)
    968 		lkp->lk_unlock_file = file;
    969 		lkp->lk_unlock_line = line;
    970 #endif
    971 		DONTHAVEIT(lkp);
    972 	}
    973 #ifdef DIAGNOSTIC
    974 	else if (lkp->lk_sharecount != 0)
    975 		panic("spinlock_release_all: release of shared lock!");
    976 	else
    977 		panic("spinlock_release_all: release of unlocked lock!");
    978 #endif
    979 	INTERLOCK_RELEASE(lkp, LK_SPIN, s);
    980 
    981 	return (count);
    982 }
    983 
    984 /*
    985  * For a recursive spinlock held one or more times by the current CPU,
    986  * release all N locks, and return N.
    987  * Intended for use in mi_switch() right after resuming execution.
    988  */
    989 
    990 void
    991 #if defined(LOCKDEBUG)
    992 _spinlock_acquire_count(volatile struct lock *lkp, int count,
    993     const char *file, int line)
    994 #else
    995 spinlock_acquire_count(volatile struct lock *lkp, int count)
    996 #endif
    997 {
    998 	int s, error;
    999 	cpuid_t cpu_num;
   1000 
   1001 	KASSERT(lkp->lk_flags & LK_SPIN);
   1002 
   1003 	INTERLOCK_ACQUIRE(lkp, LK_SPIN, s);
   1004 
   1005 	cpu_num = cpu_number();
   1006 
   1007 #ifdef DIAGNOSTIC
   1008 	if (WEHOLDIT(lkp, LK_NOPROC, 0, cpu_num))
   1009 		panic("spinlock_acquire_count: processor %lu already holds lock", (long)cpu_num);
   1010 #endif
   1011 	/*
   1012 	 * Try to acquire the want_exclusive flag.
   1013 	 */
   1014 	error = acquire(&lkp, &s, LK_SPIN, 0, LK_HAVE_EXCL | LK_WANT_EXCL,
   1015 	    RETURN_ADDRESS);
   1016 	lkp->lk_flags |= LK_WANT_EXCL;
   1017 	/*
   1018 	 * Wait for shared locks and upgrades to finish.
   1019 	 */
   1020 	error = acquire(&lkp, &s, LK_SPIN, 0,
   1021 	    LK_HAVE_EXCL | LK_SHARE_NONZERO | LK_WANT_UPGRADE,
   1022 	    RETURN_ADDRESS);
   1023 	lkp->lk_flags &= ~LK_WANT_EXCL;
   1024 	lkp->lk_flags |= LK_HAVE_EXCL;
   1025 	SETHOLDER(lkp, LK_NOPROC, 0, cpu_num);
   1026 #if defined(LOCKDEBUG)
   1027 	lkp->lk_lock_file = file;
   1028 	lkp->lk_lock_line = line;
   1029 #endif
   1030 	HAVEIT(lkp);
   1031 	if (lkp->lk_exclusivecount != 0)
   1032 		panic("lockmgr: non-zero exclusive count");
   1033 	lkp->lk_exclusivecount = count;
   1034 	lkp->lk_recurselevel = 1;
   1035 	COUNT_CPU(cpu_num, count);
   1036 
   1037 	INTERLOCK_RELEASE(lkp, lkp->lk_flags, s);
   1038 }
   1039 
   1040 
   1041 
   1042 /*
   1043  * Print out information about state of a lock. Used by VOP_PRINT
   1044  * routines to display ststus about contained locks.
   1045  */
   1046 void
   1047 lockmgr_printinfo(volatile struct lock *lkp)
   1048 {
   1049 
   1050 	if (lkp->lk_sharecount)
   1051 		printf(" lock type %s: SHARED (count %d)", lkp->lk_wmesg,
   1052 		    lkp->lk_sharecount);
   1053 	else if (lkp->lk_flags & LK_HAVE_EXCL) {
   1054 		printf(" lock type %s: EXCL (count %d) by ",
   1055 		    lkp->lk_wmesg, lkp->lk_exclusivecount);
   1056 		if (lkp->lk_flags & LK_SPIN)
   1057 			printf("processor %lu", lkp->lk_cpu);
   1058 		else
   1059 			printf("pid %d.%d", lkp->lk_lockholder,
   1060 			    lkp->lk_locklwp);
   1061 	} else
   1062 		printf(" not locked");
   1063 	if ((lkp->lk_flags & LK_SPIN) == 0 && lkp->lk_waitcount > 0)
   1064 		printf(" with %d pending", lkp->lk_waitcount);
   1065 }
   1066 
   1067 #if defined(LOCKDEBUG) /* { */
   1068 _TAILQ_HEAD(, struct simplelock, volatile) simplelock_list =
   1069     TAILQ_HEAD_INITIALIZER(simplelock_list);
   1070 
   1071 #if defined(MULTIPROCESSOR) /* { */
   1072 struct simplelock simplelock_list_slock = SIMPLELOCK_INITIALIZER;
   1073 
   1074 #define	SLOCK_LIST_LOCK()						\
   1075 	__cpu_simple_lock(&simplelock_list_slock.lock_data)
   1076 
   1077 #define	SLOCK_LIST_UNLOCK()						\
   1078 	__cpu_simple_unlock(&simplelock_list_slock.lock_data)
   1079 
   1080 #define	SLOCK_COUNT(x)							\
   1081 	curcpu()->ci_simple_locks += (x)
   1082 #else
   1083 u_long simple_locks;
   1084 
   1085 #define	SLOCK_LIST_LOCK()	/* nothing */
   1086 
   1087 #define	SLOCK_LIST_UNLOCK()	/* nothing */
   1088 
   1089 #define	SLOCK_COUNT(x)		simple_locks += (x)
   1090 #endif /* MULTIPROCESSOR */ /* } */
   1091 
   1092 #ifdef MULTIPROCESSOR
   1093 #define SLOCK_MP()		lock_printf("on CPU %ld\n", 		\
   1094 				    (u_long) cpu_number())
   1095 #else
   1096 #define SLOCK_MP()		/* nothing */
   1097 #endif
   1098 
   1099 #define	SLOCK_WHERE(str, alp, id, l)					\
   1100 do {									\
   1101 	lock_printf("\n");						\
   1102 	lock_printf(str);						\
   1103 	lock_printf("lock: %p, currently at: %s:%d\n", (alp), (id), (l)); \
   1104 	SLOCK_MP();							\
   1105 	if ((alp)->lock_file != NULL)					\
   1106 		lock_printf("last locked: %s:%d\n", (alp)->lock_file,	\
   1107 		    (alp)->lock_line);					\
   1108 	if ((alp)->unlock_file != NULL)					\
   1109 		lock_printf("last unlocked: %s:%d\n", (alp)->unlock_file, \
   1110 		    (alp)->unlock_line);				\
   1111 	SLOCK_TRACE()							\
   1112 	SLOCK_DEBUGGER();						\
   1113 } while (/*CONSTCOND*/0)
   1114 
   1115 /*
   1116  * Simple lock functions so that the debugger can see from whence
   1117  * they are being called.
   1118  */
   1119 void
   1120 simple_lock_init(volatile struct simplelock *alp)
   1121 {
   1122 
   1123 #if defined(MULTIPROCESSOR) /* { */
   1124 	__cpu_simple_lock_init(&alp->lock_data);
   1125 #else
   1126 	alp->lock_data = __SIMPLELOCK_UNLOCKED;
   1127 #endif /* } */
   1128 	alp->lock_file = NULL;
   1129 	alp->lock_line = 0;
   1130 	alp->unlock_file = NULL;
   1131 	alp->unlock_line = 0;
   1132 	alp->lock_holder = LK_NOCPU;
   1133 }
   1134 
   1135 void
   1136 _simple_lock(volatile struct simplelock *alp, const char *id, int l)
   1137 {
   1138 	cpuid_t cpu_num = cpu_number();
   1139 	int s;
   1140 
   1141 	s = spllock();
   1142 
   1143 	/*
   1144 	 * MULTIPROCESSOR case: This is `safe' since if it's not us, we
   1145 	 * don't take any action, and just fall into the normal spin case.
   1146 	 */
   1147 	if (alp->lock_data == __SIMPLELOCK_LOCKED) {
   1148 #if defined(MULTIPROCESSOR) /* { */
   1149 		if (alp->lock_holder == cpu_num) {
   1150 			SLOCK_WHERE("simple_lock: locking against myself\n",
   1151 			    alp, id, l);
   1152 			goto out;
   1153 		}
   1154 #else
   1155 		SLOCK_WHERE("simple_lock: lock held\n", alp, id, l);
   1156 		goto out;
   1157 #endif /* MULTIPROCESSOR */ /* } */
   1158 	}
   1159 
   1160 #if defined(MULTIPROCESSOR) /* { */
   1161 	/* Acquire the lock before modifying any fields. */
   1162 	splx(s);
   1163 	__cpu_simple_lock(&alp->lock_data);
   1164 	s = spllock();
   1165 #else
   1166 	alp->lock_data = __SIMPLELOCK_LOCKED;
   1167 #endif /* } */
   1168 
   1169 	if (alp->lock_holder != LK_NOCPU) {
   1170 		SLOCK_WHERE("simple_lock: uninitialized lock\n",
   1171 		    alp, id, l);
   1172 	}
   1173 	alp->lock_file = id;
   1174 	alp->lock_line = l;
   1175 	alp->lock_holder = cpu_num;
   1176 
   1177 	SLOCK_LIST_LOCK();
   1178 	TAILQ_INSERT_TAIL(&simplelock_list, alp, list);
   1179 	SLOCK_LIST_UNLOCK();
   1180 
   1181 	SLOCK_COUNT(1);
   1182 
   1183  out:
   1184 	splx(s);
   1185 }
   1186 
   1187 int
   1188 _simple_lock_held(volatile struct simplelock *alp)
   1189 {
   1190 #if defined(MULTIPROCESSOR) || defined(DIAGNOSTIC)
   1191 	cpuid_t cpu_num = cpu_number();
   1192 #endif
   1193 	int s, locked = 0;
   1194 
   1195 	s = spllock();
   1196 
   1197 #if defined(MULTIPROCESSOR)
   1198 	if (__cpu_simple_lock_try(&alp->lock_data) == 0)
   1199 		locked = (alp->lock_holder == cpu_num);
   1200 	else
   1201 		__cpu_simple_unlock(&alp->lock_data);
   1202 #else
   1203 	if (alp->lock_data == __SIMPLELOCK_LOCKED) {
   1204 		locked = 1;
   1205 		KASSERT(alp->lock_holder == cpu_num);
   1206 	}
   1207 #endif
   1208 
   1209 	splx(s);
   1210 
   1211 	return (locked);
   1212 }
   1213 
   1214 int
   1215 _simple_lock_try(volatile struct simplelock *alp, const char *id, int l)
   1216 {
   1217 	cpuid_t cpu_num = cpu_number();
   1218 	int s, rv = 0;
   1219 
   1220 	s = spllock();
   1221 
   1222 	/*
   1223 	 * MULTIPROCESSOR case: This is `safe' since if it's not us, we
   1224 	 * don't take any action.
   1225 	 */
   1226 #if defined(MULTIPROCESSOR) /* { */
   1227 	if ((rv = __cpu_simple_lock_try(&alp->lock_data)) == 0) {
   1228 		if (alp->lock_holder == cpu_num)
   1229 			SLOCK_WHERE("simple_lock_try: locking against myself\n",
   1230 			    alp, id, l);
   1231 		goto out;
   1232 	}
   1233 #else
   1234 	if (alp->lock_data == __SIMPLELOCK_LOCKED) {
   1235 		SLOCK_WHERE("simple_lock_try: lock held\n", alp, id, l);
   1236 		goto out;
   1237 	}
   1238 	alp->lock_data = __SIMPLELOCK_LOCKED;
   1239 #endif /* MULTIPROCESSOR */ /* } */
   1240 
   1241 	/*
   1242 	 * At this point, we have acquired the lock.
   1243 	 */
   1244 
   1245 	rv = 1;
   1246 
   1247 	alp->lock_file = id;
   1248 	alp->lock_line = l;
   1249 	alp->lock_holder = cpu_num;
   1250 
   1251 	SLOCK_LIST_LOCK();
   1252 	TAILQ_INSERT_TAIL(&simplelock_list, alp, list);
   1253 	SLOCK_LIST_UNLOCK();
   1254 
   1255 	SLOCK_COUNT(1);
   1256 
   1257  out:
   1258 	splx(s);
   1259 	return (rv);
   1260 }
   1261 
   1262 void
   1263 _simple_unlock(volatile struct simplelock *alp, const char *id, int l)
   1264 {
   1265 	int s;
   1266 
   1267 	s = spllock();
   1268 
   1269 	/*
   1270 	 * MULTIPROCESSOR case: This is `safe' because we think we hold
   1271 	 * the lock, and if we don't, we don't take any action.
   1272 	 */
   1273 	if (alp->lock_data == __SIMPLELOCK_UNLOCKED) {
   1274 		SLOCK_WHERE("simple_unlock: lock not held\n",
   1275 		    alp, id, l);
   1276 		goto out;
   1277 	}
   1278 
   1279 	SLOCK_LIST_LOCK();
   1280 	TAILQ_REMOVE(&simplelock_list, alp, list);
   1281 	SLOCK_LIST_UNLOCK();
   1282 
   1283 	SLOCK_COUNT(-1);
   1284 
   1285 	alp->list.tqe_next = NULL;	/* sanity */
   1286 	alp->list.tqe_prev = NULL;	/* sanity */
   1287 
   1288 	alp->unlock_file = id;
   1289 	alp->unlock_line = l;
   1290 
   1291 #if defined(MULTIPROCESSOR) /* { */
   1292 	alp->lock_holder = LK_NOCPU;
   1293 	/* Now that we've modified all fields, release the lock. */
   1294 	__cpu_simple_unlock(&alp->lock_data);
   1295 #else
   1296 	alp->lock_data = __SIMPLELOCK_UNLOCKED;
   1297 	KASSERT(alp->lock_holder == cpu_number());
   1298 	alp->lock_holder = LK_NOCPU;
   1299 #endif /* } */
   1300 
   1301  out:
   1302 	splx(s);
   1303 }
   1304 
   1305 void
   1306 simple_lock_dump(void)
   1307 {
   1308 	volatile struct simplelock *alp;
   1309 	int s;
   1310 
   1311 	s = spllock();
   1312 	SLOCK_LIST_LOCK();
   1313 	lock_printf("all simple locks:\n");
   1314 	TAILQ_FOREACH(alp, &simplelock_list, list) {
   1315 		lock_printf("%p CPU %lu %s:%d\n", alp, alp->lock_holder,
   1316 		    alp->lock_file, alp->lock_line);
   1317 	}
   1318 	SLOCK_LIST_UNLOCK();
   1319 	splx(s);
   1320 }
   1321 
   1322 void
   1323 simple_lock_freecheck(void *start, void *end)
   1324 {
   1325 	volatile struct simplelock *alp;
   1326 	int s;
   1327 
   1328 	s = spllock();
   1329 	SLOCK_LIST_LOCK();
   1330 	TAILQ_FOREACH(alp, &simplelock_list, list) {
   1331 		if ((volatile void *)alp >= start &&
   1332 		    (volatile void *)alp < end) {
   1333 			lock_printf("freeing simple_lock %p CPU %lu %s:%d\n",
   1334 			    alp, alp->lock_holder, alp->lock_file,
   1335 			    alp->lock_line);
   1336 			SLOCK_DEBUGGER();
   1337 		}
   1338 	}
   1339 	SLOCK_LIST_UNLOCK();
   1340 	splx(s);
   1341 }
   1342 
   1343 /*
   1344  * We must be holding exactly one lock: the sched_lock.
   1345  */
   1346 
   1347 void
   1348 simple_lock_switchcheck(void)
   1349 {
   1350 
   1351 	simple_lock_only_held(NULL, "switching");
   1352 }
   1353 
   1354 /*
   1355  * Drop into the debugger if lp isn't the only lock held.
   1356  * lp may be NULL.
   1357  */
   1358 void
   1359 simple_lock_only_held(volatile struct simplelock *lp, const char *where)
   1360 {
   1361 	volatile struct simplelock *alp;
   1362 	cpuid_t cpu_num = cpu_number();
   1363 	int s;
   1364 
   1365 	if (lp) {
   1366 		LOCK_ASSERT(simple_lock_held(lp));
   1367 	}
   1368 	s = spllock();
   1369 	SLOCK_LIST_LOCK();
   1370 	TAILQ_FOREACH(alp, &simplelock_list, list) {
   1371 		if (alp == lp)
   1372 			continue;
   1373 		if (alp->lock_holder == cpu_num)
   1374 			break;
   1375 	}
   1376 	SLOCK_LIST_UNLOCK();
   1377 	splx(s);
   1378 
   1379 	if (alp != NULL) {
   1380 		lock_printf("\n%s with held simple_lock %p "
   1381 		    "CPU %lu %s:%d\n",
   1382 		    where, alp, alp->lock_holder, alp->lock_file,
   1383 		    alp->lock_line);
   1384 		SLOCK_TRACE();
   1385 		SLOCK_DEBUGGER();
   1386 	}
   1387 }
   1388 
   1389 /*
   1390  * Set to 1 by simple_lock_assert_*().
   1391  * Can be cleared from ddb to avoid a panic.
   1392  */
   1393 int slock_assert_will_panic;
   1394 
   1395 /*
   1396  * If the lock isn't held, print a traceback, optionally drop into the
   1397  *  debugger, then panic.
   1398  * The panic can be avoided by clearing slock_assert_with_panic from the
   1399  *  debugger.
   1400  */
   1401 void
   1402 _simple_lock_assert_locked(volatile struct simplelock *alp,
   1403     const char *lockname, const char *id, int l)
   1404 {
   1405 	if (simple_lock_held(alp) == 0) {
   1406 		slock_assert_will_panic = 1;
   1407 		lock_printf("%s lock not held\n", lockname);
   1408 		SLOCK_WHERE("lock not held", alp, id, l);
   1409 		if (slock_assert_will_panic)
   1410 			panic("%s: not locked", lockname);
   1411 	}
   1412 }
   1413 
   1414 void
   1415 _simple_lock_assert_unlocked(volatile struct simplelock *alp,
   1416     const char *lockname, const char *id, int l)
   1417 {
   1418 	if (simple_lock_held(alp)) {
   1419 		slock_assert_will_panic = 1;
   1420 		lock_printf("%s lock held\n", lockname);
   1421 		SLOCK_WHERE("lock held", alp, id, l);
   1422 		if (slock_assert_will_panic)
   1423 			panic("%s: locked", lockname);
   1424 	}
   1425 }
   1426 
   1427 void
   1428 assert_sleepable(struct simplelock *interlock, const char *msg)
   1429 {
   1430 
   1431 	if (curlwp == NULL) {
   1432 		panic("assert_sleepable: NULL curlwp");
   1433 	}
   1434 	simple_lock_only_held(interlock, msg);
   1435 }
   1436 
   1437 #endif /* LOCKDEBUG */ /* } */
   1438 
   1439 #if defined(MULTIPROCESSOR)
   1440 
   1441 /*
   1442  * Functions for manipulating the kernel_lock.  We put them here
   1443  * so that they show up in profiles.
   1444  */
   1445 
   1446 #define	_KERNEL_LOCK_ABORT(msg)						\
   1447     LOCKDEBUG_ABORT(kernel_lock_id, &kernel_lock, &_kernel_lock_ops,	\
   1448         __FUNCTION__, msg)
   1449 
   1450 int	_kernel_lock_dump(volatile void *, char *, size_t);
   1451 
   1452 lockops_t _kernel_lock_ops = {
   1453 	"Kernel lock",
   1454 	0,
   1455 	_kernel_lock_dump
   1456 };
   1457 
   1458 /*
   1459  * Initialize the kernel lock.
   1460  */
   1461 void
   1462 _kernel_lock_init(void)
   1463 {
   1464 
   1465 	__cpu_simple_lock_init(&kernel_lock);
   1466 	kernel_lock_id = LOCKDEBUG_ALLOC(&kernel_lock, &_kernel_lock_ops);
   1467 }
   1468 
   1469 /*
   1470  * Print debugging information about the kernel lock.
   1471  */
   1472 int
   1473 _kernel_lock_dump(volatile void *junk, char *buf, size_t l)
   1474 {
   1475 	struct cpu_info *ci = curcpu();
   1476 
   1477 	(void)junk;
   1478 
   1479 	return snprintf(buf, l, "curcpu holds : %18d want cnt.: %18d\n",
   1480 	    ci->ci_biglock_count, ci->ci_biglock_wanted);
   1481 }
   1482 
   1483 /*
   1484  * Acquire 'nlocks' holds on the kernel lock.  If 'l' is non-null, the
   1485  * acquisition is from process context.
   1486  */
   1487 void
   1488 _kernel_lock(u_int nlocks, struct lwp *l)
   1489 {
   1490 	struct cpu_info *ci = curcpu();
   1491 	LOCKSTAT_TIMER(spintime);
   1492 	u_int count;
   1493 	int s;
   1494 
   1495 	(void)l;
   1496 
   1497 	KASSERT(nlocks > 0);
   1498 
   1499 	s = splbiglock();
   1500 
   1501 	if (ci->ci_biglock_count != 0) {
   1502 		KDASSERT(__cpu_simple_lock_held(&kernel_lock));
   1503 		ci->ci_biglock_count += nlocks;
   1504 		splx(s);
   1505 		return;
   1506 	}
   1507 
   1508 	if (__cpu_simple_lock_try(&kernel_lock)) {
   1509 		ci->ci_biglock_count = nlocks;
   1510 		LOCKDEBUG_LOCKED(kernel_lock_id,
   1511 		    (uintptr_t)__builtin_return_address(0), 0);
   1512 		splx(s);
   1513 		return;
   1514 	}
   1515 
   1516 	LOCKSTAT_START_TIMER(spintime);
   1517 	count = 0;
   1518 	ci->ci_biglock_wanted++;
   1519 
   1520 	while (!__cpu_simple_lock_try(&kernel_lock)) {
   1521 		splx(s);
   1522 #ifdef LOCKDEBUG
   1523 		SPINLOCK_BACKOFF(count);
   1524 #else
   1525 		if (++count > 0x0fffffff)
   1526 			_KERNEL_LOCK_ABORT("spinout");
   1527 #endif
   1528 		s = splbiglock();
   1529 	}
   1530 
   1531 	ci->ci_biglock_wanted--;
   1532 	ci->ci_biglock_count += nlocks;
   1533 	LOCKSTAT_STOP_TIMER(spintime);
   1534 	LOCKDEBUG_LOCKED(kernel_lock_id,
   1535 	    (uintptr_t)__builtin_return_address(0), 0);
   1536 	splx(s);
   1537 
   1538 	if (count == 0)
   1539 		return;
   1540 
   1541 	LOCKSTAT_EVENT(&kernel_lock, LB_KERNEL_LOCK | LB_SPIN, 1, spintime);
   1542 }
   1543 
   1544 /*
   1545  * Release 'nlocks' holds on the kernel lock.  If 'nlocks' is zero, release
   1546  * all holds.  If 'l' is non-null, the release is from process context.
   1547  */
   1548 u_int
   1549 _kernel_unlock(u_int nlocks, struct lwp *l)
   1550 {
   1551 	struct cpu_info *ci = curcpu();
   1552 	u_int olocks;
   1553 	int s;
   1554 
   1555 	(void)l;
   1556 
   1557 	KASSERT(nlocks < 2);
   1558 
   1559 	olocks = ci->ci_biglock_count;
   1560 
   1561 	if (olocks == 0) {
   1562 		KASSERT(nlocks == 0);
   1563 		return 0;
   1564 	}
   1565 
   1566 	KDASSERT(__cpu_simple_lock_held(&kernel_lock));
   1567 
   1568 	if (nlocks == 0)
   1569 		nlocks = olocks;
   1570 
   1571 	s = splbiglock();
   1572 	if ((ci->ci_biglock_count -= nlocks) == 0) {
   1573 		LOCKDEBUG_UNLOCKED(kernel_lock_id,
   1574 		    (uintptr_t)__builtin_return_address(0), 0);
   1575 		__cpu_simple_unlock(&kernel_lock);
   1576 	}
   1577 	splx(s);
   1578 
   1579 	return olocks;
   1580 }
   1581 
   1582 #if defined(DEBUG)
   1583 /*
   1584  * Assert that the kernel lock is held.
   1585  */
   1586 void
   1587 _kernel_lock_assert_locked(void)
   1588 {
   1589 
   1590 	if (!__cpu_simple_lock_held(&kernel_lock) ||
   1591 	    ci->ci_biglock_count == 0)
   1592 		_KERNEL_LOCK_ABORT("not locked");
   1593 }
   1594 #endif
   1595 
   1596 int
   1597 lock_owner_onproc(uintptr_t owner)
   1598 {
   1599 	CPU_INFO_ITERATOR cii;
   1600 	struct cpu_info *ci;
   1601 
   1602 	for (CPU_INFO_FOREACH(cii, ci))
   1603 		if (owner == (uintptr_t)ci->ci_curlwp)
   1604 			return ci->ci_biglock_wanted == 0;
   1605 
   1606 	return (0);
   1607 }
   1608 
   1609 #else	/* MULTIPROCESSOR */
   1610 
   1611 int
   1612 lock_owner_onproc(uintptr_t owner)
   1613 {
   1614 
   1615 	return 0;
   1616 }
   1617 
   1618 #endif /* MULTIPROCESSOR */
   1619