xref: /dflybsd-src/lib/libc/stdlib/nmalloc.c (revision 8b07b5e82a38bb7f04797b88aca14d3ccb08532c)
1 /*
2  * NMALLOC.C	- New Malloc (ported from kernel slab allocator)
3  *
4  * Copyright (c) 2003,2004,2009,2010 The DragonFly Project. All rights reserved.
5  *
6  * This code is derived from software contributed to The DragonFly Project
7  * by Matthew Dillon <dillon@backplane.com> and by
8  * Venkatesh Srinivas <me@endeavour.zapto.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  *
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in
18  *    the documentation and/or other materials provided with the
19  *    distribution.
20  * 3. Neither the name of The DragonFly Project nor the names of its
21  *    contributors may be used to endorse or promote products derived
22  *    from this software without specific, prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
27  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
28  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
29  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
30  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
31  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
32  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
33  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
34  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  * $Id: nmalloc.c,v 1.37 2010/07/23 08:20:35 vsrinivas Exp $
38  */
39 /*
40  * This module implements a slab allocator drop-in replacement for the
41  * libc malloc().
42  *
43  * A slab allocator reserves a ZONE for each chunk size, then lays the
44  * chunks out in an array within the zone.  Allocation and deallocation
45  * is nearly instantaneous, and overhead losses are limited to a fixed
46  * worst-case amount.
47  *
48  * The slab allocator does not have to pre-initialize the list of
49  * free chunks for each zone, and the underlying VM will not be
50  * touched at all beyond the zone header until an actual allocation
51  * needs it.
52  *
53  * Slab management and locking is done on a per-zone basis.
54  *
55  *	Alloc Size	Chunking        Number of zones
56  *	0-127		8		16
57  *	128-255		16		8
58  *	256-511		32		8
59  *	512-1023	64		8
60  *	1024-2047	128		8
61  *	2048-4095	256		8
62  *	4096-8191	512		8
63  *	8192-16383	1024		8
64  *	16384-32767	2048		8
65  *
66  *	Allocations >= ZoneLimit (16K) go directly to mmap and a hash table
67  *	is used to locate for free.  One and Two-page allocations use the
68  *	zone mechanic to avoid excessive mmap()/munmap() calls.
69  *
70  *			   API FEATURES AND SIDE EFFECTS
71  *
72  *    + power-of-2 sized allocations up to a page will be power-of-2 aligned.
73  *	Above that power-of-2 sized allocations are page-aligned.  Non
74  *	power-of-2 sized allocations are aligned the same as the chunk
75  *	size for their zone.
76  *    + malloc(0) returns a special non-NULL value
77  *    + ability to allocate arbitrarily large chunks of memory
78  *    + realloc will reuse the passed pointer if possible, within the
79  *	limitations of the zone chunking.
80  *
81  * Multithreaded enhancements for small allocations introduced August 2010.
82  * These are in the spirit of 'libumem'. See:
83  *	Bonwick, J.; Adams, J. (2001). "Magazines and Vmem: Extending the
84  *	slab allocator to many CPUs and arbitrary resources". In Proc. 2001
85  *	USENIX Technical Conference. USENIX Association.
86  *
87  * Oversized allocations employ the BIGCACHE mechanic whereby large
88  * allocations may be handed significantly larger buffers, allowing them
89  * to avoid mmap/munmap operations even through significant realloc()s.
90  * The excess space is only trimmed if too many large allocations have been
91  * given this treatment.
92  *
93  * TUNING
94  *
95  * The value of the environment variable MALLOC_OPTIONS is a character string
96  * containing various flags to tune nmalloc.
97  *
98  * 'U'   / ['u']	Generate / do not generate utrace entries for ktrace(1)
99  *			This will generate utrace events for all malloc,
100  *			realloc, and free calls. There are tools (mtrplay) to
101  *			replay and allocation pattern or to graph heap structure
102  *			(mtrgraph) which can interpret these logs.
103  * 'Z'   / ['z']	Zero out / do not zero all allocations.
104  *			Each new byte of memory allocated by malloc, realloc, or
105  *			reallocf will be initialized to 0. This is intended for
106  *			debugging and will affect performance negatively.
107  * 'H'	/  ['h']	Pass a hint to the kernel about pages unused by the
108  *			allocation functions.
109  */
110 
111 /* cc -shared -fPIC -g -O -I/usr/src/lib/libc/include -o nmalloc.so nmalloc.c */
112 
113 #include "namespace.h"
114 #include <sys/param.h>
115 #include <sys/types.h>
116 #include <sys/mman.h>
117 #include <sys/queue.h>
118 #include <sys/ktrace.h>
119 #include <stdio.h>
120 #include <stdint.h>
121 #include <stdlib.h>
122 #include <stdarg.h>
123 #include <stddef.h>
124 #include <unistd.h>
125 #include <string.h>
126 #include <fcntl.h>
127 #include <errno.h>
128 #include <pthread.h>
129 #include <machine/atomic.h>
130 #include "un-namespace.h"
131 
132 #include "libc_private.h"
133 #include "spinlock.h"
134 
135 void __free(void *);
136 void *__malloc(size_t);
137 void *__calloc(size_t, size_t);
138 void *__realloc(void *, size_t);
139 void *__aligned_alloc(size_t, size_t);
140 int __posix_memalign(void **, size_t, size_t);
141 
142 /*
143  * Linked list of large allocations
144  */
145 typedef struct bigalloc {
146 	struct bigalloc *next;	/* hash link */
147 	void	*base;		/* base pointer */
148 	u_long	active;		/* bytes active */
149 	u_long	bytes;		/* bytes allocated */
150 } *bigalloc_t;
151 
152 /*
153  * Note that any allocations which are exact multiples of PAGE_SIZE, or
154  * which are >= ZALLOC_ZONE_LIMIT, will fall through to the kmem subsystem.
155  */
156 #define ZALLOC_ZONE_LIMIT	(16 * 1024)	/* max slab-managed alloc */
157 #define ZALLOC_MIN_ZONE_SIZE	(32 * 1024)	/* minimum zone size */
158 #define ZALLOC_MAX_ZONE_SIZE	(128 * 1024)	/* maximum zone size */
159 #define ZALLOC_ZONE_SIZE	(64 * 1024)
160 #define ZALLOC_SLAB_MAGIC	0x736c6162	/* magic sanity */
161 #define ZALLOC_SLAB_SLIDE	20		/* L1-cache skip */
162 
163 #if ZALLOC_ZONE_LIMIT == 16384
164 #define NZONES			72
165 #elif ZALLOC_ZONE_LIMIT == 32768
166 #define NZONES			80
167 #else
168 #error "I couldn't figure out NZONES"
169 #endif
170 
171 /*
172  * Chunk structure for free elements
173  */
174 typedef struct slchunk {
175 	struct slchunk *c_Next;
176 } *slchunk_t;
177 
178 /*
179  * The IN-BAND zone header is placed at the beginning of each zone.
180  */
181 struct slglobaldata;
182 
183 typedef struct slzone {
184 	int32_t		z_Magic;	/* magic number for sanity check */
185 	int		z_NFree;	/* total free chunks / ualloc space */
186 	struct slzone *z_Next;		/* ZoneAry[] link if z_NFree non-zero */
187 	int		z_NMax;		/* maximum free chunks */
188 	char		*z_BasePtr;	/* pointer to start of chunk array */
189 	int		z_UIndex;	/* current initial allocation index */
190 	int		z_UEndIndex;	/* last (first) allocation index */
191 	int		z_ChunkSize;	/* chunk size for validation */
192 	int		z_FirstFreePg;	/* chunk list on a page-by-page basis */
193 	int		z_ZoneIndex;
194 	int		z_Flags;
195 	struct slchunk *z_PageAry[ZALLOC_ZONE_SIZE / PAGE_SIZE];
196 } *slzone_t;
197 
198 typedef struct slglobaldata {
199 	spinlock_t	Spinlock;
200 	slzone_t	ZoneAry[NZONES];/* linked list of zones NFree > 0 */
201 	int		JunkIndex;
202 } *slglobaldata_t;
203 
204 #define SLZF_UNOTZEROD		0x0001
205 
206 #define FASTSLABREALLOC		0x02
207 
208 /*
209  * Misc constants.  Note that allocations that are exact multiples of
210  * PAGE_SIZE, or exceed the zone limit, fall through to the kmem module.
211  * IN_SAME_PAGE_MASK is used to sanity-check the per-page free lists.
212  */
213 #define MIN_CHUNK_SIZE		8		/* in bytes */
214 #define MIN_CHUNK_MASK		(MIN_CHUNK_SIZE - 1)
215 #define IN_SAME_PAGE_MASK	(~(intptr_t)PAGE_MASK | MIN_CHUNK_MASK)
216 
217 /*
218  * WARNING: A limited number of spinlocks are available, BIGXSIZE should
219  *	    not be larger then 64.
220  */
221 #define BIGHSHIFT	10			/* bigalloc hash table */
222 #define BIGHSIZE	(1 << BIGHSHIFT)
223 #define BIGHMASK	(BIGHSIZE - 1)
224 #define BIGXSIZE	(BIGHSIZE / 16)		/* bigalloc lock table */
225 #define BIGXMASK	(BIGXSIZE - 1)
226 
227 /*
228  * BIGCACHE caches oversized allocations.  Note that a linear search is
229  * performed, so do not make the cache too large.
230  *
231  * BIGCACHE will garbage-collect excess space when the excess exceeds the
232  * specified value.  A relatively large number should be used here because
233  * garbage collection is expensive.
234  */
235 #define BIGCACHE	16
236 #define BIGCACHE_MASK	(BIGCACHE - 1)
237 #define BIGCACHE_LIMIT	(1024 * 1024)		/* size limit */
238 #define BIGCACHE_EXCESS	(16 * 1024 * 1024)	/* garbage collect */
239 
240 #define CACHE_CHUNKS	32
241 
242 #define SAFLAG_ZERO	0x0001
243 #define SAFLAG_PASSIVE	0x0002
244 #define SAFLAG_MAGS	0x0004
245 
246 /*
247  * Thread control
248  */
249 
250 #define arysize(ary)	(sizeof(ary)/sizeof((ary)[0]))
251 
252 /*
253  * The assertion macros try to pretty-print assertion failures
254  * which can be caused by corruption.  If a lock is held, we
255  * provide a macro that attempts to release it before asserting
256  * in order to prevent (e.g.) a reentrant SIGABRT calling malloc
257  * and deadlocking, resulting in the program freezing up.
258  */
259 #define MASSERT(exp)				\
260 	do { if (__predict_false(!(exp)))	\
261 	    _mpanic("assertion: %s in %s",	\
262 		    #exp, __func__);		\
263 	} while (0)
264 
265 #define MASSERT_WTHUNLK(exp, unlk)		\
266 	do { if (__predict_false(!(exp))) {	\
267 	    unlk;				\
268 	    _mpanic("assertion: %s in %s",	\
269 		    #exp, __func__);		\
270 	  }					\
271 	} while (0)
272 
273 /*
274  * Magazines
275  */
276 
277 #define M_MAX_ROUNDS	64
278 #define M_ZONE_ROUNDS	64
279 #define M_LOW_ROUNDS	32
280 #define M_INIT_ROUNDS	8
281 #define M_BURST_FACTOR  8
282 #define M_BURST_NSCALE	2
283 
284 #define M_BURST		0x0001
285 #define M_BURST_EARLY	0x0002
286 
287 struct magazine {
288 	SLIST_ENTRY(magazine) nextmagazine;
289 
290 	int		flags;
291 	int		capacity;	/* Max rounds in this magazine */
292 	int		rounds;		/* Current number of free rounds */
293 	int		burst_factor;	/* Number of blocks to prefill with */
294 	int		low_factor;	/* Free till low_factor from full mag */
295 	void		*objects[M_MAX_ROUNDS];
296 };
297 
298 SLIST_HEAD(magazinelist, magazine);
299 
300 static spinlock_t zone_mag_lock;
301 static spinlock_t depot_spinlock;
302 static struct magazine zone_magazine = {
303 	.flags = M_BURST | M_BURST_EARLY,
304 	.capacity = M_ZONE_ROUNDS,
305 	.rounds = 0,
306 	.burst_factor = M_BURST_FACTOR,
307 	.low_factor = M_LOW_ROUNDS
308 };
309 
310 #define MAGAZINE_FULL(mp)	(mp->rounds == mp->capacity)
311 #define MAGAZINE_NOTFULL(mp)	(mp->rounds < mp->capacity)
312 #define MAGAZINE_EMPTY(mp)	(mp->rounds == 0)
313 #define MAGAZINE_NOTEMPTY(mp)	(mp->rounds != 0)
314 
315 /*
316  * Each thread will have a pair of magazines per size-class (NZONES)
317  * The loaded magazine will support immediate allocations, the previous
318  * magazine will either be full or empty and can be swapped at need
319  */
320 typedef struct magazine_pair {
321 	struct magazine	*loaded;
322 	struct magazine	*prev;
323 } magazine_pair;
324 
325 /* A depot is a collection of magazines for a single zone. */
326 typedef struct magazine_depot {
327 	struct magazinelist full;
328 	struct magazinelist empty;
329 	spinlock_t	lock;
330 } magazine_depot;
331 
332 typedef struct thr_mags {
333 	magazine_pair	mags[NZONES];
334 	struct magazine	*newmag;
335 	int		init;
336 } thr_mags;
337 
338 static __thread thr_mags thread_mags TLS_ATTRIBUTE;
339 static pthread_key_t thread_mags_key;
340 static pthread_once_t thread_mags_once = PTHREAD_ONCE_INIT;
341 static magazine_depot depots[NZONES];
342 
343 /*
344  * Fixed globals (not per-cpu)
345  */
346 static const int ZoneSize = ZALLOC_ZONE_SIZE;
347 static const int ZoneLimit = ZALLOC_ZONE_LIMIT;
348 static const int ZonePageCount = ZALLOC_ZONE_SIZE / PAGE_SIZE;
349 static const int ZoneMask = ZALLOC_ZONE_SIZE - 1;
350 
351 static int opt_madvise = 0;
352 static int opt_utrace = 0;
353 static int g_malloc_flags = 0;
354 static struct slglobaldata SLGlobalData;
355 static bigalloc_t bigalloc_array[BIGHSIZE];
356 static spinlock_t bigspin_array[BIGXSIZE];
357 static volatile void *bigcache_array[BIGCACHE];		/* atomic swap */
358 static volatile size_t bigcache_size_array[BIGCACHE];	/* SMP races ok */
359 static volatile int bigcache_index;			/* SMP races ok */
360 static int malloc_panic;
361 static size_t excess_alloc;				/* excess big allocs */
362 
363 static void *_slaballoc(size_t size, int flags);
364 static void *_slabrealloc(void *ptr, size_t size);
365 static void _slabfree(void *ptr, int, bigalloc_t *);
366 static int _slabmemalign(void **memptr, size_t alignment, size_t size);
367 static void *_vmem_alloc(size_t bytes, size_t align, int flags);
368 static void _vmem_free(void *ptr, size_t bytes);
369 static void *magazine_alloc(struct magazine *, int *);
370 static int magazine_free(struct magazine *, void *);
371 static void *mtmagazine_alloc(int zi, int flags);
372 static int mtmagazine_free(int zi, void *);
373 static void mtmagazine_init(void);
374 static void mtmagazine_destructor(void *);
375 static slzone_t zone_alloc(int flags);
376 static void zone_free(void *z);
377 static void _mpanic(const char *ctl, ...) __printflike(1, 2);
378 static void malloc_init(void) __constructor(101);
379 
380 struct nmalloc_utrace {
381 	void *p;
382 	size_t s;
383 	void *r;
384 };
385 
386 #define UTRACE(a, b, c)						\
387 	if (opt_utrace) {					\
388 		struct nmalloc_utrace ut = {			\
389 			.p = (a),				\
390 			.s = (b),				\
391 			.r = (c)				\
392 		};						\
393 		utrace(&ut, sizeof(ut));			\
394 	}
395 
396 static void
397 malloc_init(void)
398 {
399 	const char *p = NULL;
400 
401 	if (issetugid() == 0)
402 		p = getenv("MALLOC_OPTIONS");
403 
404 	for (; p != NULL && *p != '\0'; p++) {
405 		switch(*p) {
406 		case 'u':	opt_utrace = 0; break;
407 		case 'U':	opt_utrace = 1; break;
408 		case 'h':	opt_madvise = 0; break;
409 		case 'H':	opt_madvise = 1; break;
410 		case 'z':	g_malloc_flags = 0; break;
411 		case 'Z':	g_malloc_flags = SAFLAG_ZERO; break;
412 		default:
413 			break;
414 		}
415 	}
416 
417 	UTRACE((void *) -1, 0, NULL);
418 }
419 
420 /*
421  * We have to install a handler for nmalloc thread teardowns when
422  * the thread is created.  We cannot delay this because destructors in
423  * sophisticated userland programs can call malloc() for the first time
424  * during their thread exit.
425  *
426  * This routine is called directly from pthreads.
427  */
428 void
429 _nmalloc_thr_init(void)
430 {
431 	thr_mags *tp;
432 
433 	/*
434 	 * Disallow mtmagazine operations until the mtmagazine is
435 	 * initialized.
436 	 */
437 	tp = &thread_mags;
438 	tp->init = -1;
439 
440 	_pthread_once(&thread_mags_once, mtmagazine_init);
441 	_pthread_setspecific(thread_mags_key, tp);
442 	tp->init = 1;
443 }
444 
445 void
446 _nmalloc_thr_prepfork(void)
447 {
448 	if (__isthreaded) {
449 		_SPINLOCK(&zone_mag_lock);
450 		_SPINLOCK(&depot_spinlock);
451 	}
452 }
453 
454 void
455 _nmalloc_thr_parentfork(void)
456 {
457 	if (__isthreaded) {
458 		_SPINUNLOCK(&depot_spinlock);
459 		_SPINUNLOCK(&zone_mag_lock);
460 	}
461 }
462 
463 void
464 _nmalloc_thr_childfork(void)
465 {
466 	if (__isthreaded) {
467 		_SPINUNLOCK(&depot_spinlock);
468 		_SPINUNLOCK(&zone_mag_lock);
469 	}
470 }
471 
472 /*
473  * Handle signal reentrancy safely whether we are threaded or not.
474  * This improves the stability for mono and will probably improve
475  * stability for other high-level languages which are becoming increasingly
476  * sophisticated.
477  *
478  * The sigblockall()/sigunblockall() implementation uses a counter on
479  * a per-thread shared user/kernel page, avoids system calls, and is thus
480  *  very fast.
481  */
482 static __inline void
483 nmalloc_sigblockall(void)
484 {
485 	sigblockall();
486 }
487 
488 static __inline void
489 nmalloc_sigunblockall(void)
490 {
491 	sigunblockall();
492 }
493 
494 /*
495  * Thread locks.
496  */
497 static __inline void
498 slgd_lock(slglobaldata_t slgd)
499 {
500 	if (__isthreaded)
501 		_SPINLOCK(&slgd->Spinlock);
502 	else
503 		sigblockall();
504 }
505 
506 static __inline void
507 slgd_unlock(slglobaldata_t slgd)
508 {
509 	if (__isthreaded)
510 		_SPINUNLOCK(&slgd->Spinlock);
511 	else
512 		sigunblockall();
513 }
514 
515 static __inline void
516 depot_lock(magazine_depot *dp __unused)
517 {
518 	if (__isthreaded)
519 		_SPINLOCK(&depot_spinlock);
520 	else
521 		sigblockall();
522 #if 0
523 	if (__isthreaded)
524 		_SPINLOCK(&dp->lock);
525 #endif
526 }
527 
528 static __inline void
529 depot_unlock(magazine_depot *dp __unused)
530 {
531 	if (__isthreaded)
532 		_SPINUNLOCK(&depot_spinlock);
533 	else
534 		sigunblockall();
535 #if 0
536 	if (__isthreaded)
537 		_SPINUNLOCK(&dp->lock);
538 #endif
539 }
540 
541 static __inline void
542 zone_magazine_lock(void)
543 {
544 	if (__isthreaded)
545 		_SPINLOCK(&zone_mag_lock);
546 	else
547 		sigblockall();
548 }
549 
550 static __inline void
551 zone_magazine_unlock(void)
552 {
553 	if (__isthreaded)
554 		_SPINUNLOCK(&zone_mag_lock);
555 	else
556 		sigunblockall();
557 }
558 
559 static __inline void
560 swap_mags(magazine_pair *mp)
561 {
562 	struct magazine *tmp;
563 	tmp = mp->loaded;
564 	mp->loaded = mp->prev;
565 	mp->prev = tmp;
566 }
567 
568 /*
569  * bigalloc hashing and locking support.
570  *
571  * Return an unmasked hash code for the passed pointer.
572  */
573 static __inline int
574 _bigalloc_hash(void *ptr)
575 {
576 	int hv;
577 
578 	hv = ((int)(intptr_t)ptr >> PAGE_SHIFT) ^
579 	      ((int)(intptr_t)ptr >> (PAGE_SHIFT + BIGHSHIFT));
580 
581 	return(hv);
582 }
583 
584 /*
585  * Lock the hash chain and return a pointer to its base for the specified
586  * address.
587  */
588 static __inline bigalloc_t *
589 bigalloc_lock(void *ptr)
590 {
591 	int hv = _bigalloc_hash(ptr);
592 	bigalloc_t *bigp;
593 
594 	bigp = &bigalloc_array[hv & BIGHMASK];
595 	if (__isthreaded)
596 		_SPINLOCK(&bigspin_array[hv & BIGXMASK]);
597 	return(bigp);
598 }
599 
600 /*
601  * Lock the hash chain and return a pointer to its base for the specified
602  * address.
603  *
604  * BUT, if the hash chain is empty, just return NULL and do not bother
605  * to lock anything.
606  */
607 static __inline bigalloc_t *
608 bigalloc_check_and_lock(void *ptr)
609 {
610 	int hv = _bigalloc_hash(ptr);
611 	bigalloc_t *bigp;
612 
613 	bigp = &bigalloc_array[hv & BIGHMASK];
614 	if (*bigp == NULL)
615 		return(NULL);
616 	if (__isthreaded) {
617 		_SPINLOCK(&bigspin_array[hv & BIGXMASK]);
618 	}
619 	return(bigp);
620 }
621 
622 static __inline void
623 bigalloc_unlock(void *ptr)
624 {
625 	int hv;
626 
627 	if (__isthreaded) {
628 		hv = _bigalloc_hash(ptr);
629 		_SPINUNLOCK(&bigspin_array[hv & BIGXMASK]);
630 	}
631 }
632 
633 /*
634  * Find a bigcache entry that might work for the allocation.  SMP races are
635  * ok here except for the swap (that is, it is ok if bigcache_size_array[i]
636  * is wrong or if a NULL or too-small big is returned).
637  *
638  * Generally speaking it is ok to find a large entry even if the bytes
639  * requested are relatively small (but still oversized), because we really
640  * don't know *what* the application is going to do with the buffer.
641  */
642 static __inline
643 bigalloc_t
644 bigcache_find_alloc(size_t bytes)
645 {
646 	bigalloc_t big = NULL;
647 	size_t test;
648 	int i;
649 
650 	for (i = 0; i < BIGCACHE; ++i) {
651 		test = bigcache_size_array[i];
652 		if (bytes <= test) {
653 			bigcache_size_array[i] = 0;
654 			big = atomic_swap_ptr(&bigcache_array[i], NULL);
655 			break;
656 		}
657 	}
658 	return big;
659 }
660 
661 /*
662  * Free a bigcache entry, possibly returning one that the caller really must
663  * free.  This is used to cache recent oversized memory blocks.  Only
664  * big blocks smaller than BIGCACHE_LIMIT will be cached this way, so try
665  * to collect the biggest ones we can that are under the limit.
666  */
667 static __inline
668 bigalloc_t
669 bigcache_find_free(bigalloc_t big)
670 {
671 	int i;
672 	int j;
673 	int b;
674 
675 	b = ++bigcache_index;
676 	for (i = 0; i < BIGCACHE; ++i) {
677 		j = (b + i) & BIGCACHE_MASK;
678 		if (bigcache_size_array[j] < big->bytes) {
679 			bigcache_size_array[j] = big->bytes;
680 			big = atomic_swap_ptr(&bigcache_array[j], big);
681 			break;
682 		}
683 	}
684 	return big;
685 }
686 
687 static __inline
688 void
689 handle_excess_big(void)
690 {
691 	int i;
692 	bigalloc_t big;
693 	bigalloc_t *bigp;
694 
695 	if (excess_alloc <= BIGCACHE_EXCESS)
696 		return;
697 
698 	for (i = 0; i < BIGHSIZE; ++i) {
699 		bigp = &bigalloc_array[i];
700 		if (*bigp == NULL)
701 			continue;
702 		if (__isthreaded)
703 			_SPINLOCK(&bigspin_array[i & BIGXMASK]);
704 		for (big = *bigp; big; big = big->next) {
705 			if (big->active < big->bytes) {
706 				MASSERT_WTHUNLK((big->active & PAGE_MASK) == 0,
707 				    _SPINUNLOCK(&bigspin_array[i & BIGXMASK]));
708 				MASSERT_WTHUNLK((big->bytes & PAGE_MASK) == 0,
709 				    _SPINUNLOCK(&bigspin_array[i & BIGXMASK]));
710 				munmap((char *)big->base + big->active,
711 				       big->bytes - big->active);
712 				atomic_add_long(&excess_alloc,
713 						big->active - big->bytes);
714 				big->bytes = big->active;
715 			}
716 		}
717 		if (__isthreaded)
718 			_SPINUNLOCK(&bigspin_array[i & BIGXMASK]);
719 	}
720 }
721 
722 /*
723  * Calculate the zone index for the allocation request size and set the
724  * allocation request size to that particular zone's chunk size.
725  */
726 static __inline int
727 zoneindex(size_t *bytes, size_t *chunking)
728 {
729 	size_t n = (unsigned int)*bytes;	/* unsigned for shift opt */
730 
731 	/*
732 	 * This used to be 8-byte chunks and 16 zones for n < 128.
733 	 * However some instructions may require 16-byte alignment
734 	 * (aka SIMD) and programs might not request an aligned size
735 	 * (aka GCC-7), so change this as follows:
736 	 *
737 	 * 0-15 bytes	8-byte alignment in two zones	(0-1)
738 	 * 16-127 bytes	16-byte alignment in four zones	(3-10)
739 	 * zone index 2 and 11-15 are currently unused.
740 	 */
741 	if (n < 16) {
742 		*bytes = n = (n + 7) & ~7;
743 		*chunking = 8;
744 		return(n / 8 - 1);		/* 8 byte chunks, 2 zones */
745 		/* zones 0,1, zone 2 is unused */
746 	}
747 	if (n < 128) {
748 		*bytes = n = (n + 15) & ~15;
749 		*chunking = 16;
750 		return(n / 16 + 2);		/* 16 byte chunks, 8 zones */
751 		/* zones 3-10, zones 11-15 unused */
752 	}
753 	if (n < 256) {
754 		*bytes = n = (n + 15) & ~15;
755 		*chunking = 16;
756 		return(n / 16 + 7);
757 	}
758 	if (n < 8192) {
759 		if (n < 512) {
760 			*bytes = n = (n + 31) & ~31;
761 			*chunking = 32;
762 			return(n / 32 + 15);
763 		}
764 		if (n < 1024) {
765 			*bytes = n = (n + 63) & ~63;
766 			*chunking = 64;
767 			return(n / 64 + 23);
768 		}
769 		if (n < 2048) {
770 			*bytes = n = (n + 127) & ~127;
771 			*chunking = 128;
772 			return(n / 128 + 31);
773 		}
774 		if (n < 4096) {
775 			*bytes = n = (n + 255) & ~255;
776 			*chunking = 256;
777 			return(n / 256 + 39);
778 		}
779 		*bytes = n = (n + 511) & ~511;
780 		*chunking = 512;
781 		return(n / 512 + 47);
782 	}
783 #if ZALLOC_ZONE_LIMIT > 8192
784 	if (n < 16384) {
785 		*bytes = n = (n + 1023) & ~1023;
786 		*chunking = 1024;
787 		return(n / 1024 + 55);
788 	}
789 #endif
790 #if ZALLOC_ZONE_LIMIT > 16384
791 	if (n < 32768) {
792 		*bytes = n = (n + 2047) & ~2047;
793 		*chunking = 2048;
794 		return(n / 2048 + 63);
795 	}
796 #endif
797 	_mpanic("Unexpected byte count %zu", n);
798 	return(0);
799 }
800 
801 /*
802  * malloc() - call internal slab allocator
803  */
804 void *
805 __malloc(size_t size)
806 {
807 	void *ptr;
808 
809 	nmalloc_sigblockall();
810 	ptr = _slaballoc(size, 0);
811 	if (ptr == NULL)
812 		errno = ENOMEM;
813 	else
814 		UTRACE(0, size, ptr);
815 	nmalloc_sigunblockall();
816 
817 	return(ptr);
818 }
819 
820 #define MUL_NO_OVERFLOW	(1UL << (sizeof(size_t) * 4))
821 
822 /*
823  * calloc() - call internal slab allocator
824  */
825 void *
826 __calloc(size_t number, size_t size)
827 {
828 	void *ptr;
829 
830 	if ((number >= MUL_NO_OVERFLOW || size >= MUL_NO_OVERFLOW) &&
831 	     number > 0 && SIZE_MAX / number < size) {
832 		errno = ENOMEM;
833 		return(NULL);
834 	}
835 
836 	nmalloc_sigblockall();
837 	ptr = _slaballoc(number * size, SAFLAG_ZERO);
838 	if (ptr == NULL)
839 		errno = ENOMEM;
840 	else
841 		UTRACE(0, number * size, ptr);
842 	nmalloc_sigunblockall();
843 
844 	return(ptr);
845 }
846 
847 /*
848  * realloc() (SLAB ALLOCATOR)
849  *
850  * We do not attempt to optimize this routine beyond reusing the same
851  * pointer if the new size fits within the chunking of the old pointer's
852  * zone.
853  */
854 void *
855 __realloc(void *ptr, size_t size)
856 {
857 	void *ret;
858 
859 	nmalloc_sigblockall();
860 	ret = _slabrealloc(ptr, size);
861 	if (ret == NULL)
862 		errno = ENOMEM;
863 	else
864 		UTRACE(ptr, size, ret);
865 	nmalloc_sigunblockall();
866 
867 	return(ret);
868 }
869 
870 /*
871  * aligned_alloc()
872  *
873  * Allocate (size) bytes with a alignment of (alignment).
874  */
875 void *
876 __aligned_alloc(size_t alignment, size_t size)
877 {
878 	void *ptr;
879 	int rc;
880 
881 	nmalloc_sigblockall();
882 	ptr = NULL;
883 	rc = _slabmemalign(&ptr, alignment, size);
884 	if (rc)
885 		errno = rc;
886 	nmalloc_sigunblockall();
887 
888 	return (ptr);
889 }
890 
891 /*
892  * posix_memalign()
893  *
894  * Allocate (size) bytes with a alignment of (alignment), where (alignment)
895  * is a power of 2 >= sizeof(void *).
896  */
897 int
898 __posix_memalign(void **memptr, size_t alignment, size_t size)
899 {
900 	int rc;
901 
902 	/*
903 	 * OpenGroup spec issue 6 check
904 	 */
905 	if (alignment < sizeof(void *)) {
906 		*memptr = NULL;
907 		return(EINVAL);
908 	}
909 
910 	nmalloc_sigblockall();
911 	rc = _slabmemalign(memptr, alignment, size);
912 	nmalloc_sigunblockall();
913 
914 	return (rc);
915 }
916 
917 /*
918  * The slab allocator will allocate on power-of-2 boundaries up to
919  * at least PAGE_SIZE.  We use the zoneindex mechanic to find a
920  * zone matching the requirements, and _vmem_alloc() otherwise.
921  */
922 static int
923 _slabmemalign(void **memptr, size_t alignment, size_t size)
924 {
925 	bigalloc_t *bigp;
926 	bigalloc_t big;
927 	size_t chunking;
928 	int zi __unused;
929 
930 	if (alignment < 1) {
931 		*memptr = NULL;
932 		return(EINVAL);
933 	}
934 
935 	/*
936 	 * OpenGroup spec issue 6 checks
937 	 */
938 	if ((alignment | (alignment - 1)) + 1 != (alignment << 1)) {
939 		*memptr = NULL;
940 		return(EINVAL);
941 	}
942 
943 	/*
944 	 * Our zone mechanism guarantees same-sized alignment for any
945 	 * power-of-2 allocation.  If size is a power-of-2 and reasonable
946 	 * we can just call _slaballoc() and be done.  We round size up
947 	 * to the nearest alignment boundary to improve our odds of
948 	 * it becoming a power-of-2 if it wasn't before.
949 	 */
950 	if (size <= alignment)
951 		size = alignment;
952 	else
953 		size = (size + alignment - 1) & ~(size_t)(alignment - 1);
954 
955 	/*
956 	 * If we have overflowed above when rounding to the nearest alignment
957 	 * boundary, just return ENOMEM, size should be == N * sizeof(void *).
958 	 *
959 	 * Power-of-2 allocations up to 8KB will be aligned to the allocation
960 	 * size and _slaballoc() can simply be used.  Please see line 1082
961 	 * for this special case: 'Align the storage in the zone based on
962 	 * the chunking' has a special case for powers of 2.
963 	 */
964 	if (size == 0)
965 		return(ENOMEM);
966 
967 	if (size <= PAGE_SIZE*2 && (size | (size - 1)) + 1 == (size << 1)) {
968 		*memptr = _slaballoc(size, 0);
969 		return(*memptr ? 0 : ENOMEM);
970 	}
971 
972 	/*
973 	 * Otherwise locate a zone with a chunking that matches
974 	 * the requested alignment, within reason.   Consider two cases:
975 	 *
976 	 * (1) A 1K allocation on a 32-byte alignment.  The first zoneindex
977 	 *     we find will be the best fit because the chunking will be
978 	 *     greater or equal to the alignment.
979 	 *
980 	 * (2) A 513 allocation on a 256-byte alignment.  In this case
981 	 *     the first zoneindex we find will be for 576 byte allocations
982 	 *     with a chunking of 64, which is not sufficient.  To fix this
983 	 *     we simply find the nearest power-of-2 >= size and use the
984 	 *     same side-effect of _slaballoc() which guarantees
985 	 *     same-alignment on a power-of-2 allocation.
986 	 */
987 	if (size < PAGE_SIZE) {
988 		zi = zoneindex(&size, &chunking);
989 		if (chunking >= alignment) {
990 			*memptr = _slaballoc(size, 0);
991 			return(*memptr ? 0 : ENOMEM);
992 		}
993 		if (size >= 1024)
994 			alignment = 1024;
995 		if (size >= 16384)
996 			alignment = 16384;
997 		while (alignment < size)
998 			alignment <<= 1;
999 		*memptr = _slaballoc(alignment, 0);
1000 		return(*memptr ? 0 : ENOMEM);
1001 	}
1002 
1003 	/*
1004 	 * If the slab allocator cannot handle it use vmem_alloc().
1005 	 *
1006 	 * Alignment must be adjusted up to at least PAGE_SIZE in this case.
1007 	 */
1008 	if (alignment < PAGE_SIZE)
1009 		alignment = PAGE_SIZE;
1010 	if (size < alignment)
1011 		size = alignment;
1012 	size = (size + PAGE_MASK) & ~(size_t)PAGE_MASK;
1013 	if (alignment == PAGE_SIZE && size <= BIGCACHE_LIMIT) {
1014 		big = bigcache_find_alloc(size);
1015 		if (big && big->bytes < size) {
1016 			_slabfree(big->base, FASTSLABREALLOC, &big);
1017 			big = NULL;
1018 		}
1019 		if (big) {
1020 			*memptr = big->base;
1021 			big->active = size;
1022 			if (big->active < big->bytes) {
1023 				atomic_add_long(&excess_alloc,
1024 						big->bytes - big->active);
1025 			}
1026 			bigp = bigalloc_lock(*memptr);
1027 			big->next = *bigp;
1028 			*bigp = big;
1029 			bigalloc_unlock(*memptr);
1030 			handle_excess_big();
1031 			return(0);
1032 		}
1033 	}
1034 	*memptr = _vmem_alloc(size, alignment, 0);
1035 	if (*memptr == NULL)
1036 		return(ENOMEM);
1037 
1038 	big = _slaballoc(sizeof(struct bigalloc), 0);
1039 	if (big == NULL) {
1040 		_vmem_free(*memptr, size);
1041 		*memptr = NULL;
1042 		return(ENOMEM);
1043 	}
1044 	bigp = bigalloc_lock(*memptr);
1045 	big->base = *memptr;
1046 	big->active = size;
1047 	big->bytes = size;		/* no excess */
1048 	big->next = *bigp;
1049 	*bigp = big;
1050 	bigalloc_unlock(*memptr);
1051 
1052 	return(0);
1053 }
1054 
1055 /*
1056  * free() (SLAB ALLOCATOR) - do the obvious
1057  */
1058 void
1059 __free(void *ptr)
1060 {
1061 	UTRACE(ptr, 0, 0);
1062 	nmalloc_sigblockall();
1063 	_slabfree(ptr, 0, NULL);
1064 	nmalloc_sigunblockall();
1065 }
1066 
1067 /*
1068  * _slaballoc()	(SLAB ALLOCATOR)
1069  *
1070  *	Allocate memory via the slab allocator.  If the request is too large,
1071  *	or if it page-aligned beyond a certain size, we fall back to the
1072  *	KMEM subsystem
1073  */
1074 static void *
1075 _slaballoc(size_t size, int flags)
1076 {
1077 	slzone_t z;
1078 	slchunk_t chunk;
1079 	slglobaldata_t slgd;
1080 	size_t chunking;
1081 	thr_mags *tp;
1082 	struct magazine *mp;
1083 	int count;
1084 	int zi;
1085 	int off;
1086 	void *obj;
1087 
1088 	/*
1089 	 * Handle the degenerate size == 0 case.  Yes, this does happen.
1090 	 * Return a special pointer.  This is to maintain compatibility with
1091 	 * the original malloc implementation.  Certain devices, such as the
1092 	 * adaptec driver, not only allocate 0 bytes, they check for NULL and
1093 	 * also realloc() later on.  Joy.
1094 	 */
1095 	if (size == 0)
1096 		size = 1;
1097 
1098 	/* Capture global flags */
1099 	flags |= g_malloc_flags;
1100 
1101 	/*
1102 	 * Handle large allocations directly.  There should not be very many
1103 	 * of these so performance is not a big issue.
1104 	 *
1105 	 * The backend allocator is pretty nasty on a SMP system.   Use the
1106 	 * slab allocator for one and two page-sized chunks even though we
1107 	 * lose some efficiency.
1108 	 *
1109 	 * NOTE: Please see posix_memalign around line 864, which assumes
1110 	 *	 that power-of-2 allocations of PAGE_SIZE and PAGE_SIZE*2
1111 	 *	 can use _slaballoc() and be aligned to the same.  The
1112 	 *	 zone cache can be used for this case, bigalloc does not
1113 	 *	 have to be used.
1114 	 */
1115 	if (size >= ZoneLimit ||
1116 	    ((size & PAGE_MASK) == 0 && size > PAGE_SIZE*2)) {
1117 		bigalloc_t big;
1118 		bigalloc_t *bigp;
1119 
1120 		/*
1121 		 * Page-align and cache-color in case of virtually indexed
1122 		 * physically tagged L1 caches (aka SandyBridge).  No sweat
1123 		 * otherwise, so just do it.
1124 		 *
1125 		 * (don't count as excess).
1126 		 */
1127 		size = (size + PAGE_MASK) & ~(size_t)PAGE_MASK;
1128 
1129 		/*
1130 		 * If we have overflowed above when rounding to the page
1131 		 * boundary, something has passed us (size_t)[-PAGE_MASK..-1]
1132 		 * so just return NULL, size at this point should be >= 0.
1133 		*/
1134 		if (size == 0)
1135 			return (NULL);
1136 
1137 		if ((size & (PAGE_SIZE * 2 - 1)) == 0)
1138 			size += PAGE_SIZE;
1139 
1140 		/*
1141 		 * Try to reuse a cached big block to avoid mmap'ing.  If it
1142 		 * turns out not to fit our requirements we throw it away
1143 		 * and allocate normally.
1144 		 */
1145 		big = NULL;
1146 		if (size <= BIGCACHE_LIMIT) {
1147 			big = bigcache_find_alloc(size);
1148 			if (big && big->bytes < size) {
1149 				_slabfree(big->base, FASTSLABREALLOC, &big);
1150 				big = NULL;
1151 			}
1152 		}
1153 		if (big) {
1154 			chunk = big->base;
1155 			if (flags & SAFLAG_ZERO)
1156 				bzero(chunk, size);
1157 		} else {
1158 			chunk = _vmem_alloc(size, PAGE_SIZE, flags);
1159 			if (chunk == NULL)
1160 				return(NULL);
1161 
1162 			big = _slaballoc(sizeof(struct bigalloc), 0);
1163 			if (big == NULL) {
1164 				_vmem_free(chunk, size);
1165 				return(NULL);
1166 			}
1167 			big->base = chunk;
1168 			big->bytes = size;
1169 		}
1170 		big->active = size;
1171 
1172 		bigp = bigalloc_lock(chunk);
1173 		if (big->active < big->bytes) {
1174 			atomic_add_long(&excess_alloc,
1175 					big->bytes - big->active);
1176 		}
1177 		big->next = *bigp;
1178 		*bigp = big;
1179 		bigalloc_unlock(chunk);
1180 		handle_excess_big();
1181 
1182 		return(chunk);
1183 	}
1184 
1185 	/* Compute allocation zone; zoneindex will panic on excessive sizes */
1186 	zi = zoneindex(&size, &chunking);
1187 	MASSERT(zi < NZONES);
1188 
1189 	obj = mtmagazine_alloc(zi, flags);
1190 	if (obj != NULL) {
1191 		if (flags & SAFLAG_ZERO)
1192 			bzero(obj, size);
1193 		return (obj);
1194 	}
1195 
1196 	/*
1197 	 * Attempt to allocate out of an existing global zone.  If all zones
1198 	 * are exhausted pull one off the free list or allocate a new one.
1199 	 */
1200 	slgd = &SLGlobalData;
1201 
1202 again:
1203 	if (slgd->ZoneAry[zi] == NULL) {
1204 		z = zone_alloc(flags);
1205 		if (z == NULL)
1206 			goto fail;
1207 
1208 		/*
1209 		 * How big is the base structure?
1210 		 */
1211 		off = sizeof(struct slzone);
1212 
1213 		/*
1214 		 * Align the storage in the zone based on the chunking.
1215 		 *
1216 		 * Guarantee power-of-2 alignment for power-of-2-sized
1217 		 * chunks.  Otherwise align based on the chunking size
1218 		 * (typically 8 or 16 bytes for small allocations).
1219 		 *
1220 		 * NOTE: Allocations >= ZoneLimit are governed by the
1221 		 * bigalloc code and typically only guarantee page-alignment.
1222 		 *
1223 		 * Set initial conditions for UIndex near the zone header
1224 		 * to reduce unecessary page faults, vs semi-randomization
1225 		 * to improve L1 cache saturation.
1226 		 *
1227 		 * NOTE: Please see posix_memalign around line 864-ish, which
1228 		 *	 assumes that power-of-2 allocations of PAGE_SIZE
1229 		 *	 and PAGE_SIZE*2 can use _slaballoc() and be aligned
1230 		 *	 to the same.  The zone cache can be used for this
1231 		 *	 case, bigalloc does not have to be used.
1232 		 *
1233 		 *	 ALL power-of-2 requests that fall through to this
1234 		 *	 code use this rule (conditionals above limit this
1235 		 *	 to <= PAGE_SIZE*2.
1236 		 */
1237 		if ((size | (size - 1)) + 1 == (size << 1))
1238 			off = roundup2(off, size);
1239 		else
1240 			off = roundup2(off, chunking);
1241 		z->z_Magic = ZALLOC_SLAB_MAGIC;
1242 		z->z_ZoneIndex = zi;
1243 		z->z_NMax = (ZoneSize - off) / size;
1244 		z->z_NFree = z->z_NMax;
1245 		z->z_BasePtr = (char *)z + off;
1246 		z->z_UIndex = z->z_UEndIndex = 0;
1247 		z->z_ChunkSize = size;
1248 		z->z_FirstFreePg = ZonePageCount;
1249 		if ((z->z_Flags & SLZF_UNOTZEROD) == 0) {
1250 			flags &= ~SAFLAG_ZERO;	/* already zero'd */
1251 			flags |= SAFLAG_PASSIVE;
1252 		}
1253 
1254 		/*
1255 		 * Slide the base index for initial allocations out of the
1256 		 * next zone we create so we do not over-weight the lower
1257 		 * part of the cpu memory caches.
1258 		 */
1259 		slgd_lock(slgd);
1260 		z->z_Next = slgd->ZoneAry[zi];
1261 		slgd->ZoneAry[zi] = z;
1262 		slgd->JunkIndex = (slgd->JunkIndex + ZALLOC_SLAB_SLIDE)
1263 					& (ZALLOC_MAX_ZONE_SIZE - 1);
1264 	} else {
1265 		slgd_lock(slgd);
1266 		z = slgd->ZoneAry[zi];
1267 		if (z == NULL) {
1268 			slgd_unlock(slgd);
1269 			goto again;
1270 		}
1271 	}
1272 
1273 	/*
1274 	 * Ok, we have a zone from which at least one chunk is available.
1275 	 */
1276 	MASSERT_WTHUNLK(z->z_NFree > 0, slgd_unlock(slgd));
1277 
1278 	/*
1279 	 * Try to cache <count> chunks, up to CACHE_CHUNKS (32 typ)
1280 	 * to avoid unnecessary global lock contention.
1281 	 */
1282 	tp = &thread_mags;
1283 	mp = tp->mags[zi].loaded;
1284 	count = 0;
1285 	if (mp && tp->init >= 0) {
1286 		count = mp->capacity - mp->rounds;
1287 		if (count >= z->z_NFree)
1288 			count = z->z_NFree - 1;
1289 		if (count > CACHE_CHUNKS)
1290 			count = CACHE_CHUNKS;
1291 	}
1292 
1293 	/*
1294 	 * Locate a chunk in a free page.  This attempts to localize
1295 	 * reallocations into earlier pages without us having to sort
1296 	 * the chunk list.  A chunk may still overlap a page boundary.
1297 	 */
1298 	while (z->z_FirstFreePg < ZonePageCount) {
1299 		if ((chunk = z->z_PageAry[z->z_FirstFreePg]) != NULL) {
1300 			if (((uintptr_t)chunk & ZoneMask) == 0) {
1301 				slgd_unlock(slgd);
1302 				_mpanic("assertion: corrupt malloc zone");
1303 			}
1304 			z->z_PageAry[z->z_FirstFreePg] = chunk->c_Next;
1305 			--z->z_NFree;
1306 
1307 			if (count == 0)
1308 				goto done;
1309 			mp->objects[mp->rounds++] = chunk;
1310 			--count;
1311 			continue;
1312 		}
1313 		++z->z_FirstFreePg;
1314 	}
1315 
1316 	/*
1317 	 * No chunks are available but NFree said we had some memory,
1318 	 * so it must be available in the never-before-used-memory
1319 	 * area governed by UIndex.  The consequences are very
1320 	 * serious if our zone got corrupted so we use an explicit
1321 	 * panic rather then a KASSERT.
1322 	 */
1323 	for (;;) {
1324 		chunk = (slchunk_t)(z->z_BasePtr + z->z_UIndex * size);
1325 		--z->z_NFree;
1326 		if (++z->z_UIndex == z->z_NMax)
1327 			z->z_UIndex = 0;
1328 		if (z->z_UIndex == z->z_UEndIndex) {
1329 			if (z->z_NFree != 0) {
1330 				slgd_unlock(slgd);
1331 				_mpanic("slaballoc: corrupted zone");
1332 			}
1333 		}
1334 		if (count == 0)
1335 			break;
1336 		mp->objects[mp->rounds++] = chunk;
1337 		--count;
1338 	}
1339 
1340 	if ((z->z_Flags & SLZF_UNOTZEROD) == 0) {
1341 		flags &= ~SAFLAG_ZERO;
1342 		flags |= SAFLAG_PASSIVE;
1343 	}
1344 
1345 done:
1346 	/*
1347 	 * Remove us from the ZoneAry[] when we become empty
1348 	 */
1349 	if (z->z_NFree == 0) {
1350 		slgd->ZoneAry[zi] = z->z_Next;
1351 		z->z_Next = NULL;
1352 	}
1353 	slgd_unlock(slgd);
1354 	if (flags & SAFLAG_ZERO)
1355 		bzero(chunk, size);
1356 
1357 	return(chunk);
1358 fail:
1359 	return(NULL);
1360 }
1361 
1362 /*
1363  * Reallocate memory within the chunk
1364  */
1365 static void *
1366 _slabrealloc(void *ptr, size_t size)
1367 {
1368 	bigalloc_t *bigp;
1369 	void *nptr;
1370 	slzone_t z;
1371 	size_t chunking;
1372 
1373 	if (ptr == NULL) {
1374 		return(_slaballoc(size, 0));
1375 	}
1376 
1377 	if (size == 0)
1378 		size = 1;
1379 
1380 	/*
1381 	 * Handle oversized allocations.
1382 	 */
1383 	if ((bigp = bigalloc_check_and_lock(ptr)) != NULL) {
1384 		bigalloc_t big;
1385 		size_t bigbytes;
1386 
1387 		while ((big = *bigp) != NULL) {
1388 			if (big->base == ptr) {
1389 				size = (size + PAGE_MASK) & ~(size_t)PAGE_MASK;
1390 				bigbytes = big->bytes;
1391 
1392 				/*
1393 				 * If it already fits determine if it makes
1394 				 * sense to shrink/reallocate.  Try to optimize
1395 				 * programs which stupidly make incremental
1396 				 * reallocations larger or smaller by scaling
1397 				 * the allocation.  Also deal with potential
1398 				 * coloring.
1399 				 */
1400 				if (size >= (bigbytes >> 1) &&
1401 				    size <= bigbytes) {
1402 					if (big->active != size) {
1403 						atomic_add_long(&excess_alloc,
1404 								big->active -
1405 								size);
1406 					}
1407 					big->active = size;
1408 					bigalloc_unlock(ptr);
1409 					return(ptr);
1410 				}
1411 
1412 				/*
1413 				 * For large reallocations, allocate more space
1414 				 * than we need to try to avoid excessive
1415 				 * reallocations later on.
1416 				 */
1417 				chunking = size + (size >> 3);
1418 				chunking = (chunking + PAGE_MASK) &
1419 					   ~(size_t)PAGE_MASK;
1420 
1421 				/*
1422 				 * Try to allocate adjacently in case the
1423 				 * program is idiotically realloc()ing a
1424 				 * huge memory block just slightly bigger.
1425 				 * (llvm's llc tends to do this a lot).
1426 				 *
1427 				 * (MAP_TRYFIXED forces mmap to fail if there
1428 				 *  is already something at the address).
1429 				 */
1430 				if (chunking > bigbytes) {
1431 					char *addr;
1432 					int errno_save = errno;
1433 
1434 					addr = mmap((char *)ptr + bigbytes,
1435 						    chunking - bigbytes,
1436 						    PROT_READ|PROT_WRITE,
1437 						    MAP_PRIVATE|MAP_ANON|
1438 						    MAP_TRYFIXED,
1439 						    -1, 0);
1440 					errno = errno_save;
1441 					if (addr == (char *)ptr + bigbytes) {
1442 						atomic_add_long(&excess_alloc,
1443 								big->active -
1444 								big->bytes +
1445 								chunking -
1446 								size);
1447 						big->bytes = chunking;
1448 						big->active = size;
1449 						bigalloc_unlock(ptr);
1450 
1451 						return(ptr);
1452 					}
1453 					MASSERT_WTHUNLK(
1454 						(void *)addr == MAP_FAILED,
1455 						bigalloc_unlock(ptr));
1456 				}
1457 
1458 				/*
1459 				 * Failed, unlink big and allocate fresh.
1460 				 * (note that we have to leave (big) intact
1461 				 * in case the slaballoc fails).
1462 				 */
1463 				*bigp = big->next;
1464 				bigalloc_unlock(ptr);
1465 				if ((nptr = _slaballoc(size, 0)) == NULL) {
1466 					/* Relink block */
1467 					bigp = bigalloc_lock(ptr);
1468 					big->next = *bigp;
1469 					*bigp = big;
1470 					bigalloc_unlock(ptr);
1471 					return(NULL);
1472 				}
1473 				if (size > bigbytes)
1474 					size = bigbytes;
1475 				bcopy(ptr, nptr, size);
1476 				atomic_add_long(&excess_alloc, big->active -
1477 							       big->bytes);
1478 				_slabfree(ptr, FASTSLABREALLOC, &big);
1479 
1480 				return(nptr);
1481 			}
1482 			bigp = &big->next;
1483 		}
1484 		bigalloc_unlock(ptr);
1485 		handle_excess_big();
1486 	}
1487 
1488 	/*
1489 	 * Get the original allocation's zone.  If the new request winds
1490 	 * up using the same chunk size we do not have to do anything.
1491 	 *
1492 	 * NOTE: We don't have to lock the globaldata here, the fields we
1493 	 * access here will not change at least as long as we have control
1494 	 * over the allocation.
1495 	 */
1496 	z = (slzone_t)((uintptr_t)ptr & ~(uintptr_t)ZoneMask);
1497 	MASSERT(z->z_Magic == ZALLOC_SLAB_MAGIC);
1498 
1499 	/*
1500 	 * Use zoneindex() to chunk-align the new size, as long as the
1501 	 * new size is not too large.
1502 	 */
1503 	if (size < ZoneLimit) {
1504 		zoneindex(&size, &chunking);
1505 		if (z->z_ChunkSize == size) {
1506 			return(ptr);
1507 		}
1508 	}
1509 
1510 	/*
1511 	 * Allocate memory for the new request size and copy as appropriate.
1512 	 */
1513 	if ((nptr = _slaballoc(size, 0)) != NULL) {
1514 		if (size > z->z_ChunkSize)
1515 			size = z->z_ChunkSize;
1516 		bcopy(ptr, nptr, size);
1517 		_slabfree(ptr, 0, NULL);
1518 	}
1519 
1520 	return(nptr);
1521 }
1522 
1523 /*
1524  * free (SLAB ALLOCATOR)
1525  *
1526  * Free a memory block previously allocated by malloc.  Note that we do not
1527  * attempt to uplodate ks_loosememuse as MP races could prevent us from
1528  * checking memory limits in malloc.
1529  *
1530  * flags:
1531  *	FASTSLABREALLOC		Fast call from realloc, *rbigp already
1532  *				unlinked.
1533  *
1534  * MPSAFE
1535  */
1536 static void
1537 _slabfree(void *ptr, int flags, bigalloc_t *rbigp)
1538 {
1539 	slzone_t z;
1540 	slchunk_t chunk;
1541 	bigalloc_t big;
1542 	bigalloc_t *bigp;
1543 	slglobaldata_t slgd;
1544 	size_t size;
1545 	int zi;
1546 	int pgno;
1547 
1548 	/* Fast realloc path for big allocations */
1549 	if (flags & FASTSLABREALLOC) {
1550 		big = *rbigp;
1551 		goto fastslabrealloc;
1552 	}
1553 
1554 	/*
1555 	 * Handle NULL frees and special 0-byte allocations
1556 	 */
1557 	if (ptr == NULL)
1558 		return;
1559 
1560 	/*
1561 	 * Handle oversized allocations.
1562 	 */
1563 	if ((bigp = bigalloc_check_and_lock(ptr)) != NULL) {
1564 		while ((big = *bigp) != NULL) {
1565 			if (big->base == ptr) {
1566 				*bigp = big->next;
1567 				atomic_add_long(&excess_alloc, big->active -
1568 							       big->bytes);
1569 				bigalloc_unlock(ptr);
1570 
1571 				/*
1572 				 * Try to stash the block we are freeing,
1573 				 * potentially receiving another block in
1574 				 * return which must be freed.
1575 				 */
1576 fastslabrealloc:
1577 				if (big->bytes <= BIGCACHE_LIMIT) {
1578 					big = bigcache_find_free(big);
1579 					if (big == NULL)
1580 						return;
1581 				}
1582 				ptr = big->base;	/* reload */
1583 				size = big->bytes;
1584 				_slabfree(big, 0, NULL);
1585 				_vmem_free(ptr, size);
1586 				return;
1587 			}
1588 			bigp = &big->next;
1589 		}
1590 		bigalloc_unlock(ptr);
1591 		handle_excess_big();
1592 	}
1593 
1594 	/*
1595 	 * Zone case.  Figure out the zone based on the fact that it is
1596 	 * ZoneSize aligned.
1597 	 */
1598 	z = (slzone_t)((uintptr_t)ptr & ~(uintptr_t)ZoneMask);
1599 	MASSERT(z->z_Magic == ZALLOC_SLAB_MAGIC);
1600 
1601 	size = z->z_ChunkSize;
1602 	zi = z->z_ZoneIndex;
1603 
1604 	if (g_malloc_flags & SAFLAG_ZERO)
1605 		bzero(ptr, size);
1606 
1607 	if (mtmagazine_free(zi, ptr) == 0)
1608 		return;
1609 
1610 	pgno = ((char *)ptr - (char *)z) >> PAGE_SHIFT;
1611 	chunk = ptr;
1612 
1613 	/*
1614 	 * Add this free non-zero'd chunk to a linked list for reuse, adjust
1615 	 * z_FirstFreePg.
1616 	 */
1617 	slgd = &SLGlobalData;
1618 	slgd_lock(slgd);
1619 
1620 	chunk->c_Next = z->z_PageAry[pgno];
1621 	z->z_PageAry[pgno] = chunk;
1622 	if (z->z_FirstFreePg > pgno)
1623 		z->z_FirstFreePg = pgno;
1624 
1625 	/*
1626 	 * Bump the number of free chunks.  If it becomes non-zero the zone
1627 	 * must be added back onto the appropriate list.
1628 	 */
1629 	if (z->z_NFree++ == 0) {
1630 		z->z_Next = slgd->ZoneAry[z->z_ZoneIndex];
1631 		slgd->ZoneAry[z->z_ZoneIndex] = z;
1632 	}
1633 
1634 	/*
1635 	 * If the zone becomes totally free we get rid of it.
1636 	 */
1637 	if (z->z_NFree == z->z_NMax) {
1638 		slzone_t *pz;
1639 
1640 		pz = &slgd->ZoneAry[z->z_ZoneIndex];
1641 		while (z != *pz)
1642 			pz = &(*pz)->z_Next;
1643 		*pz = z->z_Next;
1644 		z->z_Magic = -1;
1645 		z->z_Next = NULL;
1646 		slgd_unlock(slgd);
1647 		zone_free(z);
1648 	} else {
1649 		slgd_unlock(slgd);
1650 	}
1651 }
1652 
1653 /*
1654  * Allocate and return a magazine.  NULL is returned and *burst is adjusted
1655  * if the magazine is empty.
1656  */
1657 static __inline void *
1658 magazine_alloc(struct magazine *mp, int *burst)
1659 {
1660 	void *obj;
1661 
1662 	if (mp == NULL)
1663 		return(NULL);
1664 	if (MAGAZINE_NOTEMPTY(mp)) {
1665 		obj = mp->objects[--mp->rounds];
1666 		return(obj);
1667 	}
1668 
1669 	/*
1670 	 * Return burst factor to caller along with NULL
1671 	 */
1672 	if ((mp->flags & M_BURST) && (burst != NULL)) {
1673 		*burst = mp->burst_factor;
1674 	}
1675 	/* Reduce burst factor by NSCALE; if it hits 1, disable BURST */
1676 	if ((mp->flags & M_BURST) && (mp->flags & M_BURST_EARLY) &&
1677 	    (burst != NULL)) {
1678 		mp->burst_factor -= M_BURST_NSCALE;
1679 		if (mp->burst_factor <= 1) {
1680 			mp->burst_factor = 1;
1681 			mp->flags &= ~(M_BURST);
1682 			mp->flags &= ~(M_BURST_EARLY);
1683 		}
1684 	}
1685 	return (NULL);
1686 }
1687 
1688 static __inline int
1689 magazine_free(struct magazine *mp, void *p)
1690 {
1691 	if (mp != NULL && MAGAZINE_NOTFULL(mp)) {
1692 		mp->objects[mp->rounds++] = p;
1693 		return 0;
1694 	}
1695 
1696 	return -1;
1697 }
1698 
1699 static void *
1700 mtmagazine_alloc(int zi, int flags)
1701 {
1702 	thr_mags *tp;
1703 	struct magazine *mp, *emptymag;
1704 	magazine_depot *d;
1705 	void *obj;
1706 
1707 	/*
1708 	 * Do not try to access per-thread magazines while the mtmagazine
1709 	 * is being initialized or destroyed.
1710 	 */
1711 	tp = &thread_mags;
1712 	if (tp->init < 0)
1713 		return(NULL);
1714 
1715 	/*
1716 	 * Primary per-thread allocation loop
1717 	 */
1718 	nmalloc_sigblockall();
1719 	for (;;) {
1720 		/*
1721 		 * Make sure we have a magazine available for use.
1722 		 */
1723 		if (tp->newmag == NULL && (flags & SAFLAG_MAGS) == 0) {
1724 			mp = _slaballoc(sizeof(struct magazine),
1725 					SAFLAG_ZERO | SAFLAG_MAGS);
1726 			if (mp == NULL) {
1727 				obj = NULL;
1728 				break;
1729 			}
1730 			if (tp->newmag) {
1731 				_slabfree(mp, 0, NULL);
1732 			} else {
1733 				tp->newmag = mp;
1734 			}
1735 		}
1736 
1737 		/*
1738 		 * If the loaded magazine has rounds, allocate and return
1739 		 */
1740 		mp = tp->mags[zi].loaded;
1741 		obj = magazine_alloc(mp, NULL);
1742 		if (obj)
1743 			break;
1744 
1745 		/*
1746 		 * The prev magazine can only be completely empty or completely
1747 		 * full.  If it is full, swap it with the loaded magazine
1748 		 * and retry.
1749 		 */
1750 		mp = tp->mags[zi].prev;
1751 		if (mp && MAGAZINE_FULL(mp)) {
1752 			MASSERT(mp->rounds != 0);
1753 			swap_mags(&tp->mags[zi]);	/* prev now empty */
1754 			continue;
1755 		}
1756 
1757 		/*
1758 		 * If the depot has no loaded magazines ensure that tp->loaded
1759 		 * is not NULL and return NULL.  This will allow _slaballoc()
1760 		 * to cache referals to SLGlobalData in a magazine.
1761 		 */
1762 		d = &depots[zi];
1763 		if (SLIST_EMPTY(&d->full)) {	/* UNLOCKED TEST IS SAFE */
1764 			mp = tp->mags[zi].loaded;
1765 			if (mp == NULL && tp->newmag) {
1766 				mp = tp->newmag;
1767 				tp->newmag = NULL;
1768 				mp->capacity = M_MAX_ROUNDS;
1769 				mp->rounds = 0;
1770 				mp->flags = 0;
1771 				tp->mags[zi].loaded = mp;
1772 			}
1773 			break;
1774 		}
1775 
1776 		/*
1777 		 * Cycle: depot(loaded) -> loaded -> prev -> depot(empty)
1778 		 *
1779 		 * If we race and the depot has no full magazines, retry.
1780 		 */
1781 		depot_lock(d);
1782 		mp = SLIST_FIRST(&d->full);
1783 		if (mp) {
1784 			SLIST_REMOVE_HEAD(&d->full, nextmagazine);
1785 			emptymag = tp->mags[zi].prev;
1786 			if (emptymag) {
1787 				SLIST_INSERT_HEAD(&d->empty, emptymag,
1788 						  nextmagazine);
1789 			}
1790 			tp->mags[zi].prev = tp->mags[zi].loaded;
1791 			tp->mags[zi].loaded = mp;
1792 			MASSERT(MAGAZINE_NOTEMPTY(mp));
1793 		}
1794 		depot_unlock(d);
1795 		continue;
1796 	}
1797 	nmalloc_sigunblockall();
1798 
1799 	return (obj);
1800 }
1801 
1802 static int
1803 mtmagazine_free(int zi, void *ptr)
1804 {
1805 	thr_mags *tp;
1806 	struct magazine *mp, *loadedmag;
1807 	magazine_depot *d;
1808 	int rc = -1;
1809 
1810 	/*
1811 	 * Do not try to access per-thread magazines while the mtmagazine
1812 	 * is being initialized or destroyed.
1813 	 */
1814 	tp = &thread_mags;
1815 	if (tp->init < 0)
1816 		return(-1);
1817 
1818 	/*
1819 	 * Primary per-thread freeing loop
1820 	 */
1821 	nmalloc_sigblockall();
1822 	for (;;) {
1823 		/*
1824 		 * Make sure a new magazine is available in case we have
1825 		 * to use it.  Staging the newmag allows us to avoid
1826 		 * some locking/reentrancy complexity.
1827 		 *
1828 		 * Temporarily disable the per-thread caches for this
1829 		 * allocation to avoid reentrancy and/or to avoid a
1830 		 * stack overflow if the [zi] happens to be the same that
1831 		 * would be used to allocate the new magazine.
1832 		 */
1833 		if (tp->newmag == NULL) {
1834 			tp->newmag = _slaballoc(sizeof(struct magazine),
1835 						SAFLAG_ZERO);
1836 			if (tp->newmag == NULL) {
1837 				rc = -1;
1838 				break;
1839 			}
1840 		}
1841 
1842 		/*
1843 		 * If the loaded magazine has space, free directly to it
1844 		 */
1845 		rc = magazine_free(tp->mags[zi].loaded, ptr);
1846 		if (rc == 0)
1847 			break;
1848 
1849 		/*
1850 		 * The prev magazine can only be completely empty or completely
1851 		 * full.  If it is empty, swap it with the loaded magazine
1852 		 * and retry.
1853 		 */
1854 		mp = tp->mags[zi].prev;
1855 		if (mp && MAGAZINE_EMPTY(mp)) {
1856 			MASSERT(mp->rounds == 0);
1857 			swap_mags(&tp->mags[zi]);	/* prev now full */
1858 			continue;
1859 		}
1860 
1861 		/*
1862 		 * Try to get an empty magazine from the depot.  Cycle
1863 		 * through depot(empty)->loaded->prev->depot(full).
1864 		 * Retry if an empty magazine was available from the depot.
1865 		 */
1866 		d = &depots[zi];
1867 		depot_lock(d);
1868 
1869 		if ((loadedmag = tp->mags[zi].prev) != NULL)
1870 			SLIST_INSERT_HEAD(&d->full, loadedmag, nextmagazine);
1871 		tp->mags[zi].prev = tp->mags[zi].loaded;
1872 		mp = SLIST_FIRST(&d->empty);
1873 		if (mp) {
1874 			tp->mags[zi].loaded = mp;
1875 			SLIST_REMOVE_HEAD(&d->empty, nextmagazine);
1876 			depot_unlock(d);
1877 			MASSERT(MAGAZINE_NOTFULL(mp));
1878 		} else {
1879 			mp = tp->newmag;
1880 			tp->newmag = NULL;
1881 			mp->capacity = M_MAX_ROUNDS;
1882 			mp->rounds = 0;
1883 			mp->flags = 0;
1884 			tp->mags[zi].loaded = mp;
1885 			depot_unlock(d);
1886 		}
1887 	}
1888 	nmalloc_sigunblockall();
1889 
1890 	return rc;
1891 }
1892 
1893 static void
1894 mtmagazine_init(void)
1895 {
1896 	int error;
1897 
1898 	error = _pthread_key_create(&thread_mags_key, mtmagazine_destructor);
1899 	if (error)
1900 		abort();
1901 }
1902 
1903 /*
1904  * This function is only used by the thread exit destructor
1905  */
1906 static void
1907 mtmagazine_drain(struct magazine *mp)
1908 {
1909 	void *obj;
1910 
1911 	while (MAGAZINE_NOTEMPTY(mp)) {
1912 		obj = magazine_alloc(mp, NULL);
1913 		_slabfree(obj, 0, NULL);
1914 	}
1915 }
1916 
1917 /*
1918  * mtmagazine_destructor()
1919  *
1920  * When a thread exits, we reclaim all its resources; all its magazines are
1921  * drained and the structures are freed.
1922  *
1923  * WARNING!  The destructor can be called multiple times if the larger user
1924  *	     program has its own destructors which run after ours which
1925  *	     allocate or free memory.
1926  */
1927 static void
1928 mtmagazine_destructor(void *thrp)
1929 {
1930 	thr_mags *tp = thrp;
1931 	struct magazine *mp;
1932 	int i;
1933 
1934 	if (__isexiting)
1935 		return;
1936 
1937 	/*
1938 	 * Prevent further use of mtmagazines while we are destructing
1939 	 * them, as well as for any destructors which are run after us
1940 	 * prior to the thread actually being destroyed.
1941 	 */
1942 	tp->init = -1;
1943 
1944 	for (i = 0; i < NZONES; i++) {
1945 		mp = tp->mags[i].loaded;
1946 		tp->mags[i].loaded = NULL;
1947 		if (mp) {
1948 			if (MAGAZINE_NOTEMPTY(mp))
1949 				mtmagazine_drain(mp);
1950 			_slabfree(mp, 0, NULL);
1951 		}
1952 
1953 		mp = tp->mags[i].prev;
1954 		tp->mags[i].prev = NULL;
1955 		if (mp) {
1956 			if (MAGAZINE_NOTEMPTY(mp))
1957 				mtmagazine_drain(mp);
1958 			_slabfree(mp, 0, NULL);
1959 		}
1960 	}
1961 
1962 	if (tp->newmag) {
1963 		mp = tp->newmag;
1964 		tp->newmag = NULL;
1965 		_slabfree(mp, 0, NULL);
1966 	}
1967 }
1968 
1969 /*
1970  * zone_alloc()
1971  *
1972  * Attempt to allocate a zone from the zone magazine; the zone magazine has
1973  * M_BURST_EARLY enabled, so honor the burst request from the magazine.
1974  */
1975 static slzone_t
1976 zone_alloc(int flags)
1977 {
1978 	int burst = 1;
1979 	int i, j;
1980 	slzone_t z;
1981 
1982 	zone_magazine_lock();
1983 
1984 	z = magazine_alloc(&zone_magazine, &burst);
1985 	if (z == NULL && burst == 1) {
1986 		zone_magazine_unlock();
1987 		z = _vmem_alloc(ZoneSize * burst, ZoneSize, flags);
1988 	} else if (z == NULL) {
1989 		z = _vmem_alloc(ZoneSize * burst, ZoneSize, flags);
1990 		if (z) {
1991 			for (i = 1; i < burst; i++) {
1992 				j = magazine_free(&zone_magazine,
1993 						  (char *) z + (ZoneSize * i));
1994 				MASSERT_WTHUNLK(j == 0, zone_magazine_unlock());
1995 			}
1996 		}
1997 		zone_magazine_unlock();
1998 	} else {
1999 		z->z_Flags |= SLZF_UNOTZEROD;
2000 		zone_magazine_unlock();
2001 	}
2002 
2003 	return z;
2004 }
2005 
2006 /*
2007  * Free a zone.
2008  */
2009 static void
2010 zone_free(void *z)
2011 {
2012 	void *excess[M_ZONE_ROUNDS - M_LOW_ROUNDS] = {};
2013 	int i, j;
2014 
2015 	zone_magazine_lock();
2016 
2017 	bzero(z, sizeof(struct slzone));
2018 
2019 	if (opt_madvise)
2020 		madvise(z, ZoneSize, MADV_FREE);
2021 
2022 	i = magazine_free(&zone_magazine, z);
2023 
2024 	/*
2025 	 * If we failed to free, collect excess magazines; release the zone
2026 	 * magazine lock, and then free to the system via _vmem_free. Re-enable
2027 	 * BURST mode for the magazine.
2028 	 */
2029 	if (i == -1) {
2030 		j = zone_magazine.rounds - zone_magazine.low_factor;
2031 		for (i = 0; i < j; i++) {
2032 			excess[i] = magazine_alloc(&zone_magazine, NULL);
2033 			MASSERT_WTHUNLK(excess[i] !=  NULL,
2034 					zone_magazine_unlock());
2035 		}
2036 
2037 		zone_magazine_unlock();
2038 
2039 		for (i = 0; i < j; i++)
2040 			_vmem_free(excess[i], ZoneSize);
2041 
2042 		_vmem_free(z, ZoneSize);
2043 	} else {
2044 		zone_magazine_unlock();
2045 	}
2046 }
2047 
2048 /*
2049  * _vmem_alloc()
2050  *
2051  *	Directly map memory in PAGE_SIZE'd chunks with the specified
2052  *	alignment.
2053  *
2054  *	Alignment must be a multiple of PAGE_SIZE.
2055  *
2056  *	Size must be >= alignment.
2057  */
2058 static void *
2059 _vmem_alloc(size_t size, size_t align, int flags)
2060 {
2061 	static char *addr_hint;
2062 	static int reset_hint = 16;
2063 	char *addr;
2064 	char *save;
2065 
2066 	if (--reset_hint <= 0) {
2067 		addr_hint = NULL;
2068 		reset_hint = 16;
2069 	}
2070 
2071 	/*
2072 	 * Map anonymous private memory.
2073 	 */
2074 	save = mmap(addr_hint, size, PROT_READ|PROT_WRITE,
2075 		    MAP_PRIVATE|MAP_ANON, -1, 0);
2076 	if (save == MAP_FAILED)
2077 		goto worst_case;
2078 	if (((uintptr_t)save & (align - 1)) == 0)
2079 		return((void *)save);
2080 
2081 	addr_hint = (char *)(((size_t)save + (align - 1)) & ~(align - 1));
2082 	munmap(save, size);
2083 
2084 	save = mmap(addr_hint, size, PROT_READ|PROT_WRITE,
2085 		    MAP_PRIVATE|MAP_ANON, -1, 0);
2086 	if (save == MAP_FAILED)
2087 		goto worst_case;
2088 	if (((size_t)save & (align - 1)) == 0)
2089 		return((void *)save);
2090 	munmap(save, size);
2091 
2092 worst_case:
2093 	save = mmap(NULL, size + align, PROT_READ|PROT_WRITE,
2094 		    MAP_PRIVATE|MAP_ANON, -1, 0);
2095 	if (save == MAP_FAILED)
2096 		return NULL;
2097 
2098 	addr = (char *)(((size_t)save + (align - 1)) & ~(align - 1));
2099 	if (save != addr)
2100 		munmap(save, addr - save);
2101 	if (addr + size != save + size + align)
2102 		munmap(addr + size, save + align - addr);
2103 
2104 	addr_hint = addr + size;
2105 
2106 	return ((void *)addr);
2107 }
2108 
2109 /*
2110  * _vmem_free()
2111  *
2112  *	Free a chunk of memory allocated with _vmem_alloc()
2113  */
2114 static void
2115 _vmem_free(void *ptr, size_t size)
2116 {
2117 	munmap(ptr, size);
2118 }
2119 
2120 /*
2121  * Panic on fatal conditions
2122  */
2123 static void
2124 _mpanic(const char *ctl, ...)
2125 {
2126 	va_list va;
2127 
2128 	if (malloc_panic == 0) {
2129 		malloc_panic = 1;
2130 		va_start(va, ctl);
2131 		vfprintf(stderr, ctl, va);
2132 		fprintf(stderr, "\n");
2133 		fflush(stderr);
2134 		va_end(va);
2135 	}
2136 	abort();
2137 }
2138 
2139 __weak_reference(__aligned_alloc, aligned_alloc);
2140 __weak_reference(__malloc, malloc);
2141 __weak_reference(__calloc, calloc);
2142 __weak_reference(__posix_memalign, posix_memalign);
2143 __weak_reference(__realloc, realloc);
2144 __weak_reference(__free, free);
2145