Home | History | Annotate | Line # | Download | only in kern
kern_malloc.c revision 1.78
      1 /*	$NetBSD: kern_malloc.c,v 1.78 2003/02/14 21:51:36 pk Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1996 Christopher G. Demetriou.  All rights reserved.
      5  * Copyright (c) 1987, 1991, 1993
      6  *	The Regents of the University of California.  All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. All advertising materials mentioning features or use of this software
     17  *    must display the following acknowledgement:
     18  *	This product includes software developed by the University of
     19  *	California, Berkeley and its contributors.
     20  * 4. Neither the name of the University nor the names of its contributors
     21  *    may be used to endorse or promote products derived from this software
     22  *    without specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  * SUCH DAMAGE.
     35  *
     36  *	@(#)kern_malloc.c	8.4 (Berkeley) 5/20/95
     37  */
     38 
     39 #include <sys/cdefs.h>
     40 __KERNEL_RCSID(0, "$NetBSD: kern_malloc.c,v 1.78 2003/02/14 21:51:36 pk Exp $");
     41 
     42 #include "opt_lockdebug.h"
     43 
     44 #include <sys/param.h>
     45 #include <sys/proc.h>
     46 #include <sys/kernel.h>
     47 #include <sys/malloc.h>
     48 #include <sys/systm.h>
     49 
     50 #include <uvm/uvm_extern.h>
     51 
     52 static struct vm_map kmem_map_store;
     53 struct vm_map *kmem_map = NULL;
     54 
     55 #include "opt_kmempages.h"
     56 
     57 #ifdef NKMEMCLUSTERS
     58 #error NKMEMCLUSTERS is obsolete; remove it from your kernel config file and use NKMEMPAGES instead or let the kernel auto-size
     59 #endif
     60 
     61 /*
     62  * Default number of pages in kmem_map.  We attempt to calculate this
     63  * at run-time, but allow it to be either patched or set in the kernel
     64  * config file.
     65  */
     66 #ifndef NKMEMPAGES
     67 #define	NKMEMPAGES	0
     68 #endif
     69 int	nkmempages = NKMEMPAGES;
     70 
     71 /*
     72  * Defaults for lower- and upper-bounds for the kmem_map page count.
     73  * Can be overridden by kernel config options.
     74  */
     75 #ifndef	NKMEMPAGES_MIN
     76 #define	NKMEMPAGES_MIN	NKMEMPAGES_MIN_DEFAULT
     77 #endif
     78 
     79 #ifndef NKMEMPAGES_MAX
     80 #define	NKMEMPAGES_MAX	NKMEMPAGES_MAX_DEFAULT
     81 #endif
     82 
     83 #include "opt_kmemstats.h"
     84 #include "opt_malloclog.h"
     85 #include "opt_malloc_debug.h"
     86 
     87 struct kmembuckets bucket[MINBUCKET + 16];
     88 struct kmemusage *kmemusage;
     89 char *kmembase, *kmemlimit;
     90 
     91 struct malloc_type *kmemstatistics;
     92 
     93 #ifdef MALLOCLOG
     94 #ifndef MALLOCLOGSIZE
     95 #define	MALLOCLOGSIZE	100000
     96 #endif
     97 
     98 struct malloclog {
     99 	void *addr;
    100 	long size;
    101 	struct malloc_type *type;
    102 	int action;
    103 	const char *file;
    104 	long line;
    105 } malloclog[MALLOCLOGSIZE];
    106 
    107 long	malloclogptr;
    108 
    109 static void
    110 domlog(void *a, long size, struct malloc_type *type, int action,
    111     const char *file, long line)
    112 {
    113 
    114 	malloclog[malloclogptr].addr = a;
    115 	malloclog[malloclogptr].size = size;
    116 	malloclog[malloclogptr].type = type;
    117 	malloclog[malloclogptr].action = action;
    118 	malloclog[malloclogptr].file = file;
    119 	malloclog[malloclogptr].line = line;
    120 	malloclogptr++;
    121 	if (malloclogptr >= MALLOCLOGSIZE)
    122 		malloclogptr = 0;
    123 }
    124 
    125 static void
    126 hitmlog(void *a)
    127 {
    128 	struct malloclog *lp;
    129 	long l;
    130 
    131 #define	PRT do { \
    132 	if (malloclog[l].addr == a && malloclog[l].action) { \
    133 		lp = &malloclog[l]; \
    134 		printf("malloc log entry %ld:\n", l); \
    135 		printf("\taddr = %p\n", lp->addr); \
    136 		printf("\tsize = %ld\n", lp->size); \
    137 		printf("\ttype = %s\n", lp->type->ks_shortdesc); \
    138 		printf("\taction = %s\n", lp->action == 1 ? "alloc" : "free"); \
    139 		printf("\tfile = %s\n", lp->file); \
    140 		printf("\tline = %ld\n", lp->line); \
    141 	} \
    142 } while (/* CONSTCOND */0)
    143 
    144 	for (l = malloclogptr; l < MALLOCLOGSIZE; l++)
    145 		PRT;
    146 
    147 	for (l = 0; l < malloclogptr; l++)
    148 		PRT;
    149 }
    150 #endif /* MALLOCLOG */
    151 
    152 #ifdef DIAGNOSTIC
    153 /*
    154  * This structure provides a set of masks to catch unaligned frees.
    155  */
    156 const long addrmask[] = { 0,
    157 	0x00000001, 0x00000003, 0x00000007, 0x0000000f,
    158 	0x0000001f, 0x0000003f, 0x0000007f, 0x000000ff,
    159 	0x000001ff, 0x000003ff, 0x000007ff, 0x00000fff,
    160 	0x00001fff, 0x00003fff, 0x00007fff, 0x0000ffff,
    161 };
    162 
    163 /*
    164  * The WEIRD_ADDR is used as known text to copy into free objects so
    165  * that modifications after frees can be detected.
    166  */
    167 #define	WEIRD_ADDR	((uint32_t) 0xdeadbeef)
    168 #ifdef DEBUG
    169 #define	MAX_COPY	PAGE_SIZE
    170 #else
    171 #define	MAX_COPY	32
    172 #endif
    173 
    174 /*
    175  * Normally the freelist structure is used only to hold the list pointer
    176  * for free objects.  However, when running with diagnostics, the first
    177  * 8/16 bytes of the structure is unused except for diagnostic information,
    178  * and the free list pointer is at offset 8/16 in the structure.  Since the
    179  * first 8 bytes is the portion of the structure most often modified, this
    180  * helps to detect memory reuse problems and avoid free list corruption.
    181  */
    182 struct freelist {
    183 	uint32_t spare0;
    184 #ifdef _LP64
    185 	uint32_t spare1;		/* explicit padding */
    186 #endif
    187 	struct malloc_type *type;
    188 	caddr_t	next;
    189 };
    190 #else /* !DIAGNOSTIC */
    191 struct freelist {
    192 	caddr_t	next;
    193 };
    194 #endif /* DIAGNOSTIC */
    195 
    196 /*
    197  * The following are standard, build-in malloc types are are not
    198  * specific to any one subsystem.
    199  */
    200 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
    201 MALLOC_DEFINE(M_DMAMAP, "DMA map", "bus_dma(9) structures");
    202 MALLOC_DEFINE(M_FREE, "free", "should be on free list");
    203 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
    204 MALLOC_DEFINE(M_SOFTINTR, "softintr", "Softinterrupt structures");
    205 MALLOC_DEFINE(M_TEMP, "temp", "misc. temporary data buffers");
    206 
    207 /* XXX These should all be elsewhere. */
    208 MALLOC_DEFINE(M_RTABLE, "routetbl", "routing tables");
    209 MALLOC_DEFINE(M_FTABLE, "fragtbl", "fragment reassembly header");
    210 MALLOC_DEFINE(M_UFSMNT, "UFS mount", "UFS mount structure");
    211 MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
    212 MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
    213 MALLOC_DEFINE(M_IPMADDR, "in_multi", "internet multicast address");
    214 MALLOC_DEFINE(M_MRTABLE, "mrt", "multicast routing tables");
    215 MALLOC_DEFINE(M_1394DATA, "1394data", "IEEE 1394 data buffers");
    216 
    217 struct simplelock malloc_slock = SIMPLELOCK_INITIALIZER;
    218 
    219 /*
    220  * Allocate a block of memory
    221  */
    222 #ifdef MALLOCLOG
    223 void *
    224 _malloc(unsigned long size, struct malloc_type *ksp, int flags,
    225     const char *file, long line)
    226 #else
    227 void *
    228 malloc(unsigned long size, struct malloc_type *ksp, int flags)
    229 #endif /* MALLOCLOG */
    230 {
    231 	struct kmembuckets *kbp;
    232 	struct kmemusage *kup;
    233 	struct freelist *freep;
    234 	long indx, npg, allocsize;
    235 	int s;
    236 	caddr_t va, cp, savedlist;
    237 #ifdef DIAGNOSTIC
    238 	uint32_t *end, *lp;
    239 	int copysize;
    240 	const char *savedtype;
    241 #endif
    242 
    243 #ifdef LOCKDEBUG
    244 	if ((flags & M_NOWAIT) == 0)
    245 		simple_lock_only_held(NULL, "malloc");
    246 #endif
    247 #ifdef MALLOC_DEBUG
    248 	if (debug_malloc(size, ksp, flags, (void **) &va))
    249 		return ((void *) va);
    250 #endif
    251 	indx = BUCKETINDX(size);
    252 	kbp = &bucket[indx];
    253 	s = splvm();
    254 	simple_lock(&malloc_slock);
    255 #ifdef KMEMSTATS
    256 	while (ksp->ks_memuse >= ksp->ks_limit) {
    257 		if (flags & M_NOWAIT) {
    258 			simple_unlock(&malloc_slock);
    259 			splx(s);
    260 			return ((void *) NULL);
    261 		}
    262 		if (ksp->ks_limblocks < 65535)
    263 			ksp->ks_limblocks++;
    264 		ltsleep((caddr_t)ksp, PSWP+2, ksp->ks_shortdesc, 0,
    265 			&malloc_slock);
    266 	}
    267 	ksp->ks_size |= 1 << indx;
    268 #endif
    269 #ifdef DIAGNOSTIC
    270 	copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
    271 #endif
    272 	if (kbp->kb_next == NULL) {
    273 		kbp->kb_last = NULL;
    274 		if (size > MAXALLOCSAVE)
    275 			allocsize = round_page(size);
    276 		else
    277 			allocsize = 1 << indx;
    278 		npg = btoc(allocsize);
    279 		simple_unlock(&malloc_slock);
    280 		va = (caddr_t) uvm_km_kmemalloc(kmem_map, NULL,
    281 		    (vsize_t)ctob(npg),
    282 		    ((flags & M_NOWAIT) ? UVM_KMF_NOWAIT : 0) |
    283 		    ((flags & M_CANFAIL) ? UVM_KMF_CANFAIL : 0));
    284 		if (__predict_false(va == NULL)) {
    285 			/*
    286 			 * Kmem_malloc() can return NULL, even if it can
    287 			 * wait, if there is no map space avaiable, because
    288 			 * it can't fix that problem.  Neither can we,
    289 			 * right now.  (We should release pages which
    290 			 * are completely free and which are in buckets
    291 			 * with too many free elements.)
    292 			 */
    293 			if ((flags & (M_NOWAIT|M_CANFAIL)) == 0)
    294 				panic("malloc: out of space in kmem_map");
    295 			splx(s);
    296 			return (NULL);
    297 		}
    298 		simple_lock(&malloc_slock);
    299 #ifdef KMEMSTATS
    300 		kbp->kb_total += kbp->kb_elmpercl;
    301 #endif
    302 		kup = btokup(va);
    303 		kup->ku_indx = indx;
    304 		if (allocsize > MAXALLOCSAVE) {
    305 			if (npg > 65535)
    306 				panic("malloc: allocation too large");
    307 			kup->ku_pagecnt = npg;
    308 #ifdef KMEMSTATS
    309 			ksp->ks_memuse += allocsize;
    310 #endif
    311 			goto out;
    312 		}
    313 #ifdef KMEMSTATS
    314 		kup->ku_freecnt = kbp->kb_elmpercl;
    315 		kbp->kb_totalfree += kbp->kb_elmpercl;
    316 #endif
    317 		/*
    318 		 * Just in case we blocked while allocating memory,
    319 		 * and someone else also allocated memory for this
    320 		 * bucket, don't assume the list is still empty.
    321 		 */
    322 		savedlist = kbp->kb_next;
    323 		kbp->kb_next = cp = va + (npg << PAGE_SHIFT) - allocsize;
    324 		for (;;) {
    325 			freep = (struct freelist *)cp;
    326 #ifdef DIAGNOSTIC
    327 			/*
    328 			 * Copy in known text to detect modification
    329 			 * after freeing.
    330 			 */
    331 			end = (int32_t *)&cp[copysize];
    332 			for (lp = (int32_t *)cp; lp < end; lp++)
    333 				*lp = WEIRD_ADDR;
    334 			freep->type = M_FREE;
    335 #endif /* DIAGNOSTIC */
    336 			if (cp <= va)
    337 				break;
    338 			cp -= allocsize;
    339 			freep->next = cp;
    340 		}
    341 		freep->next = savedlist;
    342 		if (kbp->kb_last == NULL)
    343 			kbp->kb_last = (caddr_t)freep;
    344 	}
    345 	va = kbp->kb_next;
    346 	kbp->kb_next = ((struct freelist *)va)->next;
    347 #ifdef DIAGNOSTIC
    348 	freep = (struct freelist *)va;
    349 	/* XXX potential to get garbage pointer here. */
    350 	savedtype = freep->type->ks_shortdesc;
    351 	if (kbp->kb_next) {
    352 		int rv;
    353 		vaddr_t addr = (vaddr_t)kbp->kb_next;
    354 
    355 		vm_map_lock(kmem_map);
    356 		rv = uvm_map_checkprot(kmem_map, addr,
    357 		    addr + sizeof(struct freelist), VM_PROT_WRITE);
    358 		vm_map_unlock(kmem_map);
    359 
    360 		if (__predict_false(rv == 0)) {
    361 			printf("Data modified on freelist: "
    362 			    "word %ld of object %p size %ld previous type %s "
    363 			    "(invalid addr %p)\n",
    364 			    (long)((int32_t *)&kbp->kb_next - (int32_t *)kbp),
    365 			    va, size, savedtype, kbp->kb_next);
    366 #ifdef MALLOCLOG
    367 			hitmlog(va);
    368 #endif
    369 			kbp->kb_next = NULL;
    370 		}
    371 	}
    372 
    373 	/* Fill the fields that we've used with WEIRD_ADDR */
    374 #ifdef _LP64
    375 	freep->type = (struct malloc_type *)
    376 	    (WEIRD_ADDR | (((u_long) WEIRD_ADDR) << 32));
    377 #else
    378 	freep->type = (struct malloc_type *) WEIRD_ADDR;
    379 #endif
    380 	end = (int32_t *)&freep->next +
    381 	    (sizeof(freep->next) / sizeof(int32_t));
    382 	for (lp = (int32_t *)&freep->next; lp < end; lp++)
    383 		*lp = WEIRD_ADDR;
    384 
    385 	/* and check that the data hasn't been modified. */
    386 	end = (uint32_t *)&va[copysize];
    387 	for (lp = (int32_t *)va; lp < end; lp++) {
    388 		if (__predict_true(*lp == WEIRD_ADDR))
    389 			continue;
    390 		printf("Data modified on freelist: "
    391 		    "word %ld of object %p size %ld previous type %s "
    392 		    "(0x%x != 0x%x)\n",
    393 		    (long)(lp - (uint32_t *)va), va, size,
    394 		    savedtype, *lp, WEIRD_ADDR);
    395 #ifdef MALLOCLOG
    396 		hitmlog(va);
    397 #endif
    398 		break;
    399 	}
    400 
    401 	freep->spare0 = 0;
    402 #endif /* DIAGNOSTIC */
    403 #ifdef KMEMSTATS
    404 	kup = btokup(va);
    405 	if (kup->ku_indx != indx)
    406 		panic("malloc: wrong bucket");
    407 	if (kup->ku_freecnt == 0)
    408 		panic("malloc: lost data");
    409 	kup->ku_freecnt--;
    410 	kbp->kb_totalfree--;
    411 	ksp->ks_memuse += 1 << indx;
    412 out:
    413 	kbp->kb_calls++;
    414 	ksp->ks_inuse++;
    415 	ksp->ks_calls++;
    416 	if (ksp->ks_memuse > ksp->ks_maxused)
    417 		ksp->ks_maxused = ksp->ks_memuse;
    418 #else
    419 out:
    420 #endif
    421 #ifdef MALLOCLOG
    422 	domlog(va, size, type, 1, file, line);
    423 #endif
    424 	simple_unlock(&malloc_slock);
    425 	splx(s);
    426 	if ((flags & M_ZERO) != 0)
    427 		memset(va, 0, size);
    428 	return ((void *) va);
    429 }
    430 
    431 /*
    432  * Free a block of memory allocated by malloc.
    433  */
    434 #ifdef MALLOCLOG
    435 void
    436 _free(void *addr, struct malloc_type *type, const char *file, long line)
    437 #else
    438 void
    439 free(void *addr, struct malloc_type *ksp)
    440 #endif /* MALLOCLOG */
    441 {
    442 	struct kmembuckets *kbp;
    443 	struct kmemusage *kup;
    444 	struct freelist *freep;
    445 	long size;
    446 	int s;
    447 #ifdef DIAGNOSTIC
    448 	caddr_t cp;
    449 	int32_t *end, *lp;
    450 	long alloc, copysize;
    451 #endif
    452 
    453 #ifdef MALLOC_DEBUG
    454 	if (debug_free(addr, ksp))
    455 		return;
    456 #endif
    457 
    458 #ifdef DIAGNOSTIC
    459 	/*
    460 	 * Ensure that we're free'ing something that we could
    461 	 * have allocated in the first place.  That is, check
    462 	 * to see that the address is within kmem_map.
    463 	 */
    464 	if (__predict_false((vaddr_t)addr < kmem_map->header.start ||
    465 	    (vaddr_t)addr >= kmem_map->header.end))
    466 		panic("free: addr %p not within kmem_map", addr);
    467 #endif
    468 
    469 	kup = btokup(addr);
    470 	size = 1 << kup->ku_indx;
    471 	kbp = &bucket[kup->ku_indx];
    472 	s = splvm();
    473 	simple_lock(&malloc_slock);
    474 #ifdef MALLOCLOG
    475 	domlog(addr, 0, type, 2, file, line);
    476 #endif
    477 #ifdef DIAGNOSTIC
    478 	/*
    479 	 * Check for returns of data that do not point to the
    480 	 * beginning of the allocation.
    481 	 */
    482 	if (size > PAGE_SIZE)
    483 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
    484 	else
    485 		alloc = addrmask[kup->ku_indx];
    486 	if (((u_long)addr & alloc) != 0)
    487 		panic("free: unaligned addr %p, size %ld, type %s, mask %ld",
    488 		    addr, size, ksp->ks_shortdesc, alloc);
    489 #endif /* DIAGNOSTIC */
    490 	if (size > MAXALLOCSAVE) {
    491 		uvm_km_free(kmem_map, (vaddr_t)addr, ctob(kup->ku_pagecnt));
    492 #ifdef KMEMSTATS
    493 		size = kup->ku_pagecnt << PGSHIFT;
    494 		ksp->ks_memuse -= size;
    495 		kup->ku_indx = 0;
    496 		kup->ku_pagecnt = 0;
    497 		if (ksp->ks_memuse + size >= ksp->ks_limit &&
    498 		    ksp->ks_memuse < ksp->ks_limit)
    499 			wakeup((caddr_t)ksp);
    500 		ksp->ks_inuse--;
    501 		kbp->kb_total -= 1;
    502 #endif
    503 		simple_unlock(&malloc_slock);
    504 		splx(s);
    505 		return;
    506 	}
    507 	freep = (struct freelist *)addr;
    508 #ifdef DIAGNOSTIC
    509 	/*
    510 	 * Check for multiple frees. Use a quick check to see if
    511 	 * it looks free before laboriously searching the freelist.
    512 	 */
    513 	if (__predict_false(freep->spare0 == WEIRD_ADDR)) {
    514 		for (cp = kbp->kb_next; cp;
    515 		    cp = ((struct freelist *)cp)->next) {
    516 			if (addr != cp)
    517 				continue;
    518 			printf("multiply freed item %p\n", addr);
    519 #ifdef MALLOCLOG
    520 			hitmlog(addr);
    521 #endif
    522 			panic("free: duplicated free");
    523 		}
    524 	}
    525 #ifdef LOCKDEBUG
    526 	/*
    527 	 * Check if we're freeing a locked simple lock.
    528 	 */
    529 	simple_lock_freecheck(addr, (char *)addr + size);
    530 #endif
    531 	/*
    532 	 * Copy in known text to detect modification after freeing
    533 	 * and to make it look free. Also, save the type being freed
    534 	 * so we can list likely culprit if modification is detected
    535 	 * when the object is reallocated.
    536 	 */
    537 	copysize = size < MAX_COPY ? size : MAX_COPY;
    538 	end = (int32_t *)&((caddr_t)addr)[copysize];
    539 	for (lp = (int32_t *)addr; lp < end; lp++)
    540 		*lp = WEIRD_ADDR;
    541 	freep->type = ksp;
    542 #endif /* DIAGNOSTIC */
    543 #ifdef KMEMSTATS
    544 	kup->ku_freecnt++;
    545 	if (kup->ku_freecnt >= kbp->kb_elmpercl) {
    546 		if (kup->ku_freecnt > kbp->kb_elmpercl)
    547 			panic("free: multiple frees");
    548 		else if (kbp->kb_totalfree > kbp->kb_highwat)
    549 			kbp->kb_couldfree++;
    550 	}
    551 	kbp->kb_totalfree++;
    552 	ksp->ks_memuse -= size;
    553 	if (ksp->ks_memuse + size >= ksp->ks_limit &&
    554 	    ksp->ks_memuse < ksp->ks_limit)
    555 		wakeup((caddr_t)ksp);
    556 	ksp->ks_inuse--;
    557 #endif
    558 	if (kbp->kb_next == NULL)
    559 		kbp->kb_next = addr;
    560 	else
    561 		((struct freelist *)kbp->kb_last)->next = addr;
    562 	freep->next = NULL;
    563 	kbp->kb_last = addr;
    564 	simple_unlock(&malloc_slock);
    565 	splx(s);
    566 }
    567 
    568 /*
    569  * Change the size of a block of memory.
    570  */
    571 void *
    572 realloc(void *curaddr, unsigned long newsize, struct malloc_type *ksp,
    573     int flags)
    574 {
    575 	struct kmemusage *kup;
    576 	unsigned long cursize;
    577 	void *newaddr;
    578 #ifdef DIAGNOSTIC
    579 	long alloc;
    580 #endif
    581 
    582 	/*
    583 	 * realloc() with a NULL pointer is the same as malloc().
    584 	 */
    585 	if (curaddr == NULL)
    586 		return (malloc(newsize, ksp, flags));
    587 
    588 	/*
    589 	 * realloc() with zero size is the same as free().
    590 	 */
    591 	if (newsize == 0) {
    592 		free(curaddr, ksp);
    593 		return (NULL);
    594 	}
    595 
    596 #ifdef LOCKDEBUG
    597 	if ((flags & M_NOWAIT) == 0)
    598 		simple_lock_only_held(NULL, "realloc");
    599 #endif
    600 
    601 	/*
    602 	 * Find out how large the old allocation was (and do some
    603 	 * sanity checking).
    604 	 */
    605 	kup = btokup(curaddr);
    606 	cursize = 1 << kup->ku_indx;
    607 
    608 #ifdef DIAGNOSTIC
    609 	/*
    610 	 * Check for returns of data that do not point to the
    611 	 * beginning of the allocation.
    612 	 */
    613 	if (cursize > PAGE_SIZE)
    614 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
    615 	else
    616 		alloc = addrmask[kup->ku_indx];
    617 	if (((u_long)curaddr & alloc) != 0)
    618 		panic("realloc: "
    619 		    "unaligned addr %p, size %ld, type %s, mask %ld\n",
    620 		    curaddr, cursize, ksp->ks_shortdesc, alloc);
    621 #endif /* DIAGNOSTIC */
    622 
    623 	if (cursize > MAXALLOCSAVE)
    624 		cursize = ctob(kup->ku_pagecnt);
    625 
    626 	/*
    627 	 * If we already actually have as much as they want, we're done.
    628 	 */
    629 	if (newsize <= cursize)
    630 		return (curaddr);
    631 
    632 	/*
    633 	 * Can't satisfy the allocation with the existing block.
    634 	 * Allocate a new one and copy the data.
    635 	 */
    636 	newaddr = malloc(newsize, ksp, flags);
    637 	if (__predict_false(newaddr == NULL)) {
    638 		/*
    639 		 * malloc() failed, because flags included M_NOWAIT.
    640 		 * Return NULL to indicate that failure.  The old
    641 		 * pointer is still valid.
    642 		 */
    643 		return (NULL);
    644 	}
    645 	memcpy(newaddr, curaddr, cursize);
    646 
    647 	/*
    648 	 * We were successful: free the old allocation and return
    649 	 * the new one.
    650 	 */
    651 	free(curaddr, ksp);
    652 	return (newaddr);
    653 }
    654 
    655 /*
    656  * Roundup size to the actual allocation size.
    657  */
    658 unsigned long
    659 malloc_roundup(unsigned long size)
    660 {
    661 
    662 	if (size > MAXALLOCSAVE)
    663 		return (roundup(size, PAGE_SIZE));
    664 	else
    665 		return (1 << BUCKETINDX(size));
    666 }
    667 
    668 /*
    669  * Add a malloc type to the system.
    670  */
    671 void
    672 malloc_type_attach(struct malloc_type *type)
    673 {
    674 
    675 	if (nkmempages == 0)
    676 		panic("malloc_type_attach: nkmempages == 0");
    677 
    678 	if (type->ks_magic != M_MAGIC)
    679 		panic("malloc_type_attach: bad magic");
    680 
    681 #ifdef DIAGNOSTIC
    682 	{
    683 		struct malloc_type *ksp;
    684 		for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
    685 			if (ksp == type)
    686 				panic("malloc_type_attach: already on list");
    687 		}
    688 	}
    689 #endif
    690 
    691 #ifdef KMEMSTATS
    692 	if (type->ks_limit == 0)
    693 		type->ks_limit = ((u_long)nkmempages << PAGE_SHIFT) * 6U / 10U;
    694 #else
    695 	type->ks_limit = 0;
    696 #endif
    697 
    698 	type->ks_next = kmemstatistics;
    699 	kmemstatistics = type;
    700 }
    701 
    702 /*
    703  * Remove a malloc type from the system..
    704  */
    705 void
    706 malloc_type_detach(struct malloc_type *type)
    707 {
    708 	struct malloc_type *ksp;
    709 
    710 #ifdef DIAGNOSTIC
    711 	if (type->ks_magic != M_MAGIC)
    712 		panic("malloc_type_detach: bad magic");
    713 #endif
    714 
    715 	if (type == kmemstatistics)
    716 		kmemstatistics = type->ks_next;
    717 	else {
    718 		for (ksp = kmemstatistics; ksp->ks_next != NULL;
    719 		     ksp = ksp->ks_next) {
    720 			if (ksp->ks_next == type) {
    721 				ksp->ks_next = type->ks_next;
    722 				break;
    723 			}
    724 		}
    725 #ifdef DIAGNOSTIC
    726 		if (ksp->ks_next == NULL)
    727 			panic("malloc_type_detach: not on list");
    728 #endif
    729 	}
    730 	type->ks_next = NULL;
    731 }
    732 
    733 /*
    734  * Set the limit on a malloc type.
    735  */
    736 void
    737 malloc_type_setlimit(struct malloc_type *type, u_long limit)
    738 {
    739 #ifdef KMEMSTATS
    740 	int s;
    741 
    742 	s = splvm();
    743 	type->ks_limit = limit;
    744 	splx(s);
    745 #endif
    746 }
    747 
    748 /*
    749  * Compute the number of pages that kmem_map will map, that is,
    750  * the size of the kernel malloc arena.
    751  */
    752 void
    753 kmeminit_nkmempages(void)
    754 {
    755 	int npages;
    756 
    757 	if (nkmempages != 0) {
    758 		/*
    759 		 * It's already been set (by us being here before, or
    760 		 * by patching or kernel config options), bail out now.
    761 		 */
    762 		return;
    763 	}
    764 
    765 	/*
    766 	 * We use the following (simple) formula:
    767 	 *
    768 	 *	- Starting point is physical memory / 4.
    769 	 *
    770 	 *	- Clamp it down to NKMEMPAGES_MAX.
    771 	 *
    772 	 *	- Round it up to NKMEMPAGES_MIN.
    773 	 */
    774 	npages = physmem / 4;
    775 
    776 	if (npages > NKMEMPAGES_MAX)
    777 		npages = NKMEMPAGES_MAX;
    778 
    779 	if (npages < NKMEMPAGES_MIN)
    780 		npages = NKMEMPAGES_MIN;
    781 
    782 	nkmempages = npages;
    783 }
    784 
    785 /*
    786  * Initialize the kernel memory allocator
    787  */
    788 void
    789 kmeminit(void)
    790 {
    791 	__link_set_decl(malloc_types, struct malloc_type);
    792 	struct malloc_type * const *ksp;
    793 #ifdef KMEMSTATS
    794 	long indx;
    795 #endif
    796 
    797 #if	((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
    798 		ERROR!_kmeminit:_MAXALLOCSAVE_not_power_of_2
    799 #endif
    800 #if	(MAXALLOCSAVE > MINALLOCSIZE * 32768)
    801 		ERROR!_kmeminit:_MAXALLOCSAVE_too_big
    802 #endif
    803 #if	(MAXALLOCSAVE < NBPG)
    804 		ERROR!_kmeminit:_MAXALLOCSAVE_too_small
    805 #endif
    806 
    807 	if (sizeof(struct freelist) > (1 << MINBUCKET))
    808 		panic("minbucket too small/struct freelist too big");
    809 
    810 	/*
    811 	 * Compute the number of kmem_map pages, if we have not
    812 	 * done so already.
    813 	 */
    814 	kmeminit_nkmempages();
    815 
    816 	kmemusage = (struct kmemusage *) uvm_km_zalloc(kernel_map,
    817 	    (vsize_t)(nkmempages * sizeof(struct kmemusage)));
    818 	kmem_map = uvm_km_suballoc(kernel_map, (void *)&kmembase,
    819 	    (void *)&kmemlimit, (vsize_t)(nkmempages << PAGE_SHIFT),
    820 	    VM_MAP_INTRSAFE, FALSE, &kmem_map_store);
    821 #ifdef KMEMSTATS
    822 	for (indx = 0; indx < MINBUCKET + 16; indx++) {
    823 		if (1 << indx >= PAGE_SIZE)
    824 			bucket[indx].kb_elmpercl = 1;
    825 		else
    826 			bucket[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
    827 		bucket[indx].kb_highwat = 5 * bucket[indx].kb_elmpercl;
    828 	}
    829 #endif
    830 
    831 	/* Attach all of the statically-linked malloc types. */
    832 	__link_set_foreach(ksp, malloc_types)
    833 		malloc_type_attach(*ksp);
    834 
    835 #ifdef MALLOC_DEBUG
    836 	debug_malloc_init();
    837 #endif
    838 }
    839 
    840 #ifdef DDB
    841 #include <ddb/db_output.h>
    842 
    843 /*
    844  * Dump kmem statistics from ddb.
    845  *
    846  * usage: call dump_kmemstats
    847  */
    848 void	dump_kmemstats(void);
    849 
    850 void
    851 dump_kmemstats(void)
    852 {
    853 #ifdef KMEMSTATS
    854 	struct malloc_type *ksp;
    855 
    856 	for (ksp = kmemstatistics; ksp != NULL; ksp = ksp->ks_next) {
    857 		if (ksp->ks_memuse == 0)
    858 			continue;
    859 		db_printf("%s%.*s %ld\n", ksp->ks_shortdesc,
    860 		    (int)(20 - strlen(ksp->ks_shortdesc)),
    861 		    "                    ",
    862 		    ksp->ks_memuse);
    863 	}
    864 #else
    865 	db_printf("Kmem stats are not being collected.\n");
    866 #endif /* KMEMSTATS */
    867 }
    868 #endif /* DDB */
    869