Home | History | Annotate | Line # | Download | only in kern
kern_malloc.c revision 1.70.4.1
      1  1.70.4.1   thorpej /*	$NetBSD: kern_malloc.c,v 1.70.4.1 2002/03/22 18:59:31 thorpej Exp $	*/
      2       1.9       cgd 
      3       1.1       cgd /*
      4      1.37  christos  * Copyright (c) 1996 Christopher G. Demetriou.  All rights reserved.
      5       1.8       cgd  * Copyright (c) 1987, 1991, 1993
      6       1.8       cgd  *	The Regents of the University of California.  All rights reserved.
      7       1.1       cgd  *
      8       1.1       cgd  * Redistribution and use in source and binary forms, with or without
      9       1.1       cgd  * modification, are permitted provided that the following conditions
     10       1.1       cgd  * are met:
     11       1.1       cgd  * 1. Redistributions of source code must retain the above copyright
     12       1.1       cgd  *    notice, this list of conditions and the following disclaimer.
     13       1.1       cgd  * 2. Redistributions in binary form must reproduce the above copyright
     14       1.1       cgd  *    notice, this list of conditions and the following disclaimer in the
     15       1.1       cgd  *    documentation and/or other materials provided with the distribution.
     16       1.1       cgd  * 3. All advertising materials mentioning features or use of this software
     17       1.1       cgd  *    must display the following acknowledgement:
     18       1.1       cgd  *	This product includes software developed by the University of
     19       1.1       cgd  *	California, Berkeley and its contributors.
     20       1.1       cgd  * 4. Neither the name of the University nor the names of its contributors
     21       1.1       cgd  *    may be used to endorse or promote products derived from this software
     22       1.1       cgd  *    without specific prior written permission.
     23       1.1       cgd  *
     24       1.1       cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25       1.1       cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26       1.1       cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27       1.1       cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28       1.1       cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29       1.1       cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30       1.1       cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31       1.1       cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32       1.1       cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33       1.1       cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34       1.1       cgd  * SUCH DAMAGE.
     35       1.1       cgd  *
     36      1.32      fvdl  *	@(#)kern_malloc.c	8.4 (Berkeley) 5/20/95
     37       1.1       cgd  */
     38      1.64     lukem 
     39      1.64     lukem #include <sys/cdefs.h>
     40  1.70.4.1   thorpej __KERNEL_RCSID(0, "$NetBSD: kern_malloc.c,v 1.70.4.1 2002/03/22 18:59:31 thorpej Exp $");
     41      1.31       mrg 
     42      1.33   thorpej #include "opt_lockdebug.h"
     43       1.1       cgd 
     44       1.7   mycroft #include <sys/param.h>
     45       1.7   mycroft #include <sys/proc.h>
     46       1.8       cgd #include <sys/map.h>
     47       1.7   mycroft #include <sys/kernel.h>
     48       1.7   mycroft #include <sys/malloc.h>
     49      1.12  christos #include <sys/systm.h>
     50      1.24   thorpej 
     51      1.28       mrg #include <uvm/uvm_extern.h>
     52      1.28       mrg 
     53      1.61   thorpej static struct vm_map kmem_map_store;
     54      1.58       chs struct vm_map *kmem_map = NULL;
     55      1.28       mrg 
     56  1.70.4.1   thorpej kmutex_t malloc_mutex;
     57  1.70.4.1   thorpej 
     58      1.49   thorpej #include "opt_kmempages.h"
     59      1.49   thorpej 
     60      1.49   thorpej #ifdef NKMEMCLUSTERS
     61      1.52  sommerfe #error NKMEMCLUSTERS is obsolete; remove it from your kernel config file and use NKMEMPAGES instead or let the kernel auto-size
     62      1.49   thorpej #endif
     63      1.49   thorpej 
     64      1.49   thorpej /*
     65      1.49   thorpej  * Default number of pages in kmem_map.  We attempt to calculate this
     66      1.49   thorpej  * at run-time, but allow it to be either patched or set in the kernel
     67      1.49   thorpej  * config file.
     68      1.49   thorpej  */
     69      1.49   thorpej #ifndef NKMEMPAGES
     70      1.49   thorpej #define	NKMEMPAGES	0
     71      1.49   thorpej #endif
     72      1.49   thorpej int	nkmempages = NKMEMPAGES;
     73      1.49   thorpej 
     74      1.49   thorpej /*
     75      1.49   thorpej  * Defaults for lower- and upper-bounds for the kmem_map page count.
     76      1.49   thorpej  * Can be overridden by kernel config options.
     77      1.49   thorpej  */
     78      1.49   thorpej #ifndef	NKMEMPAGES_MIN
     79      1.49   thorpej #define	NKMEMPAGES_MIN	NKMEMPAGES_MIN_DEFAULT
     80      1.49   thorpej #endif
     81      1.49   thorpej 
     82      1.49   thorpej #ifndef NKMEMPAGES_MAX
     83      1.49   thorpej #define	NKMEMPAGES_MAX	NKMEMPAGES_MAX_DEFAULT
     84      1.49   thorpej #endif
     85      1.49   thorpej 
     86      1.24   thorpej #include "opt_kmemstats.h"
     87      1.27   thorpej #include "opt_malloclog.h"
     88      1.12  christos 
     89       1.1       cgd struct kmembuckets bucket[MINBUCKET + 16];
     90       1.8       cgd struct kmemstats kmemstats[M_LAST];
     91       1.1       cgd struct kmemusage *kmemusage;
     92       1.1       cgd char *kmembase, *kmemlimit;
     93      1.57  jdolecek const char * const memname[] = INITKMEMNAMES;
     94       1.1       cgd 
     95      1.27   thorpej #ifdef MALLOCLOG
     96      1.27   thorpej #ifndef MALLOCLOGSIZE
     97      1.27   thorpej #define	MALLOCLOGSIZE	100000
     98      1.27   thorpej #endif
     99      1.27   thorpej 
    100      1.27   thorpej struct malloclog {
    101      1.27   thorpej 	void *addr;
    102      1.27   thorpej 	long size;
    103      1.27   thorpej 	int type;
    104      1.27   thorpej 	int action;
    105      1.27   thorpej 	const char *file;
    106      1.27   thorpej 	long line;
    107      1.27   thorpej } malloclog[MALLOCLOGSIZE];
    108      1.27   thorpej 
    109      1.27   thorpej long	malloclogptr;
    110      1.27   thorpej 
    111      1.69     enami static void domlog(void *, long, int, int, const char *, long);
    112      1.69     enami static void hitmlog(void *);
    113      1.27   thorpej 
    114      1.27   thorpej static void
    115      1.69     enami domlog(void *a, long size, int type, int action, const char *file, long line)
    116      1.27   thorpej {
    117      1.27   thorpej 
    118      1.27   thorpej 	malloclog[malloclogptr].addr = a;
    119      1.27   thorpej 	malloclog[malloclogptr].size = size;
    120      1.27   thorpej 	malloclog[malloclogptr].type = type;
    121      1.27   thorpej 	malloclog[malloclogptr].action = action;
    122      1.27   thorpej 	malloclog[malloclogptr].file = file;
    123      1.27   thorpej 	malloclog[malloclogptr].line = line;
    124      1.27   thorpej 	malloclogptr++;
    125      1.27   thorpej 	if (malloclogptr >= MALLOCLOGSIZE)
    126      1.27   thorpej 		malloclogptr = 0;
    127      1.27   thorpej }
    128      1.27   thorpej 
    129      1.27   thorpej static void
    130      1.69     enami hitmlog(void *a)
    131      1.27   thorpej {
    132      1.27   thorpej 	struct malloclog *lp;
    133      1.27   thorpej 	long l;
    134      1.27   thorpej 
    135      1.69     enami #define	PRT do { \
    136      1.27   thorpej 	if (malloclog[l].addr == a && malloclog[l].action) { \
    137      1.27   thorpej 		lp = &malloclog[l]; \
    138      1.27   thorpej 		printf("malloc log entry %ld:\n", l); \
    139      1.27   thorpej 		printf("\taddr = %p\n", lp->addr); \
    140      1.27   thorpej 		printf("\tsize = %ld\n", lp->size); \
    141      1.27   thorpej 		printf("\ttype = %s\n", memname[lp->type]); \
    142      1.27   thorpej 		printf("\taction = %s\n", lp->action == 1 ? "alloc" : "free"); \
    143      1.27   thorpej 		printf("\tfile = %s\n", lp->file); \
    144      1.27   thorpej 		printf("\tline = %ld\n", lp->line); \
    145      1.69     enami 	} \
    146      1.69     enami } while (/* CONSTCOND */0)
    147      1.27   thorpej 
    148      1.27   thorpej 	for (l = malloclogptr; l < MALLOCLOGSIZE; l++)
    149      1.69     enami 		PRT;
    150      1.27   thorpej 
    151      1.27   thorpej 	for (l = 0; l < malloclogptr; l++)
    152      1.69     enami 		PRT;
    153      1.27   thorpej }
    154      1.27   thorpej #endif /* MALLOCLOG */
    155      1.27   thorpej 
    156       1.8       cgd #ifdef DIAGNOSTIC
    157       1.8       cgd /*
    158       1.8       cgd  * This structure provides a set of masks to catch unaligned frees.
    159       1.8       cgd  */
    160      1.57  jdolecek const long addrmask[] = { 0,
    161       1.8       cgd 	0x00000001, 0x00000003, 0x00000007, 0x0000000f,
    162       1.8       cgd 	0x0000001f, 0x0000003f, 0x0000007f, 0x000000ff,
    163       1.8       cgd 	0x000001ff, 0x000003ff, 0x000007ff, 0x00000fff,
    164       1.8       cgd 	0x00001fff, 0x00003fff, 0x00007fff, 0x0000ffff,
    165       1.8       cgd };
    166       1.8       cgd 
    167       1.8       cgd /*
    168       1.8       cgd  * The WEIRD_ADDR is used as known text to copy into free objects so
    169       1.8       cgd  * that modifications after frees can be detected.
    170       1.8       cgd  */
    171      1.69     enami #define	WEIRD_ADDR	((unsigned) 0xdeadbeef)
    172      1.55       chs #ifdef DEBUG
    173      1.69     enami #define	MAX_COPY	PAGE_SIZE
    174      1.55       chs #else
    175      1.69     enami #define	MAX_COPY	32
    176      1.55       chs #endif
    177       1.8       cgd 
    178       1.8       cgd /*
    179      1.11       cgd  * Normally the freelist structure is used only to hold the list pointer
    180      1.11       cgd  * for free objects.  However, when running with diagnostics, the first
    181      1.11       cgd  * 8 bytes of the structure is unused except for diagnostic information,
    182      1.11       cgd  * and the free list pointer is at offst 8 in the structure.  Since the
    183      1.11       cgd  * first 8 bytes is the portion of the structure most often modified, this
    184      1.11       cgd  * helps to detect memory reuse problems and avoid free list corruption.
    185       1.8       cgd  */
    186       1.8       cgd struct freelist {
    187      1.11       cgd 	int32_t	spare0;
    188      1.11       cgd 	int16_t	type;
    189      1.11       cgd 	int16_t	spare1;
    190       1.8       cgd 	caddr_t	next;
    191       1.8       cgd };
    192       1.8       cgd #else /* !DIAGNOSTIC */
    193       1.8       cgd struct freelist {
    194       1.8       cgd 	caddr_t	next;
    195       1.8       cgd };
    196       1.8       cgd #endif /* DIAGNOSTIC */
    197       1.8       cgd 
    198       1.1       cgd /*
    199       1.1       cgd  * Allocate a block of memory
    200       1.1       cgd  */
    201      1.27   thorpej #ifdef MALLOCLOG
    202      1.27   thorpej void *
    203      1.69     enami _malloc(unsigned long size, int type, int flags, const char *file, long line)
    204      1.27   thorpej #else
    205       1.1       cgd void *
    206      1.69     enami malloc(unsigned long size, int type, int flags)
    207      1.27   thorpej #endif /* MALLOCLOG */
    208       1.1       cgd {
    209      1.50  augustss 	struct kmembuckets *kbp;
    210      1.50  augustss 	struct kmemusage *kup;
    211      1.50  augustss 	struct freelist *freep;
    212       1.5    andrew 	long indx, npg, allocsize;
    213       1.1       cgd 	caddr_t va, cp, savedlist;
    214       1.8       cgd #ifdef DIAGNOSTIC
    215      1.11       cgd 	int32_t *end, *lp;
    216       1.8       cgd 	int copysize;
    217      1.26   mycroft 	const char *savedtype;
    218       1.8       cgd #endif
    219       1.1       cgd #ifdef KMEMSTATS
    220      1.50  augustss 	struct kmemstats *ksp = &kmemstats[type];
    221       1.1       cgd 
    222      1.51   thorpej 	if (__predict_false(((unsigned long)type) > M_LAST))
    223       1.1       cgd 		panic("malloc - bogus type");
    224       1.1       cgd #endif
    225      1.59   thorpej #ifdef LOCKDEBUG
    226      1.59   thorpej 	if ((flags & M_NOWAIT) == 0)
    227      1.59   thorpej 		simple_lock_only_held(NULL, "malloc");
    228      1.59   thorpej #endif
    229      1.62   thorpej #ifdef MALLOC_DEBUG
    230      1.62   thorpej 	if (debug_malloc(size, type, flags, (void **) &va))
    231      1.62   thorpej 		return ((void *) va);
    232      1.62   thorpej #endif
    233       1.1       cgd 	indx = BUCKETINDX(size);
    234       1.1       cgd 	kbp = &bucket[indx];
    235  1.70.4.1   thorpej 
    236  1.70.4.1   thorpej 	mutex_enter(&malloc_mutex);
    237  1.70.4.1   thorpej 
    238       1.1       cgd #ifdef KMEMSTATS
    239       1.1       cgd 	while (ksp->ks_memuse >= ksp->ks_limit) {
    240       1.1       cgd 		if (flags & M_NOWAIT) {
    241  1.70.4.1   thorpej 			mutex_exit(&malloc_mutex);
    242       1.1       cgd 			return ((void *) NULL);
    243       1.1       cgd 		}
    244       1.1       cgd 		if (ksp->ks_limblocks < 65535)
    245       1.1       cgd 			ksp->ks_limblocks++;
    246       1.1       cgd 		tsleep((caddr_t)ksp, PSWP+2, memname[type], 0);
    247       1.1       cgd 	}
    248       1.8       cgd 	ksp->ks_size |= 1 << indx;
    249       1.8       cgd #endif
    250       1.8       cgd #ifdef DIAGNOSTIC
    251       1.8       cgd 	copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
    252       1.1       cgd #endif
    253       1.1       cgd 	if (kbp->kb_next == NULL) {
    254       1.8       cgd 		kbp->kb_last = NULL;
    255       1.1       cgd 		if (size > MAXALLOCSAVE)
    256      1.66     enami 			allocsize = round_page(size);
    257       1.1       cgd 		else
    258       1.1       cgd 			allocsize = 1 << indx;
    259      1.47     ragge 		npg = btoc(allocsize);
    260      1.63       chs 		va = (caddr_t) uvm_km_kmemalloc(kmem_map, NULL,
    261      1.69     enami 		    (vsize_t)ctob(npg),
    262      1.69     enami 		    (flags & M_NOWAIT) ? UVM_KMF_NOWAIT : 0);
    263      1.51   thorpej 		if (__predict_false(va == NULL)) {
    264      1.17       cgd 			/*
    265      1.17       cgd 			 * Kmem_malloc() can return NULL, even if it can
    266      1.17       cgd 			 * wait, if there is no map space avaiable, because
    267      1.17       cgd 			 * it can't fix that problem.  Neither can we,
    268      1.17       cgd 			 * right now.  (We should release pages which
    269      1.17       cgd 			 * are completely free and which are in buckets
    270      1.17       cgd 			 * with too many free elements.)
    271      1.17       cgd 			 */
    272      1.68  jdolecek 			if ((flags & (M_NOWAIT|M_CANFAIL)) == 0)
    273      1.17       cgd 				panic("malloc: out of space in kmem_map");
    274  1.70.4.1   thorpej 			mutex_exit(&malloc_mutex);
    275       1.6       cgd 			return ((void *) NULL);
    276       1.1       cgd 		}
    277       1.1       cgd #ifdef KMEMSTATS
    278       1.1       cgd 		kbp->kb_total += kbp->kb_elmpercl;
    279       1.1       cgd #endif
    280       1.1       cgd 		kup = btokup(va);
    281       1.1       cgd 		kup->ku_indx = indx;
    282       1.1       cgd 		if (allocsize > MAXALLOCSAVE) {
    283       1.1       cgd 			if (npg > 65535)
    284       1.1       cgd 				panic("malloc: allocation too large");
    285       1.1       cgd 			kup->ku_pagecnt = npg;
    286       1.1       cgd #ifdef KMEMSTATS
    287       1.1       cgd 			ksp->ks_memuse += allocsize;
    288       1.1       cgd #endif
    289       1.1       cgd 			goto out;
    290       1.1       cgd 		}
    291       1.1       cgd #ifdef KMEMSTATS
    292       1.1       cgd 		kup->ku_freecnt = kbp->kb_elmpercl;
    293       1.1       cgd 		kbp->kb_totalfree += kbp->kb_elmpercl;
    294       1.1       cgd #endif
    295       1.1       cgd 		/*
    296       1.1       cgd 		 * Just in case we blocked while allocating memory,
    297       1.1       cgd 		 * and someone else also allocated memory for this
    298       1.1       cgd 		 * bucket, don't assume the list is still empty.
    299       1.1       cgd 		 */
    300       1.1       cgd 		savedlist = kbp->kb_next;
    301      1.49   thorpej 		kbp->kb_next = cp = va + (npg << PAGE_SHIFT) - allocsize;
    302       1.8       cgd 		for (;;) {
    303       1.8       cgd 			freep = (struct freelist *)cp;
    304       1.8       cgd #ifdef DIAGNOSTIC
    305       1.8       cgd 			/*
    306       1.8       cgd 			 * Copy in known text to detect modification
    307       1.8       cgd 			 * after freeing.
    308       1.8       cgd 			 */
    309      1.11       cgd 			end = (int32_t *)&cp[copysize];
    310      1.11       cgd 			for (lp = (int32_t *)cp; lp < end; lp++)
    311       1.8       cgd 				*lp = WEIRD_ADDR;
    312       1.8       cgd 			freep->type = M_FREE;
    313       1.8       cgd #endif /* DIAGNOSTIC */
    314       1.8       cgd 			if (cp <= va)
    315       1.8       cgd 				break;
    316       1.8       cgd 			cp -= allocsize;
    317       1.8       cgd 			freep->next = cp;
    318       1.8       cgd 		}
    319       1.8       cgd 		freep->next = savedlist;
    320       1.8       cgd 		if (kbp->kb_last == NULL)
    321       1.8       cgd 			kbp->kb_last = (caddr_t)freep;
    322       1.1       cgd 	}
    323       1.1       cgd 	va = kbp->kb_next;
    324       1.8       cgd 	kbp->kb_next = ((struct freelist *)va)->next;
    325       1.8       cgd #ifdef DIAGNOSTIC
    326       1.8       cgd 	freep = (struct freelist *)va;
    327       1.8       cgd 	savedtype = (unsigned)freep->type < M_LAST ?
    328       1.8       cgd 		memname[freep->type] : "???";
    329      1.29       chs 	if (kbp->kb_next) {
    330      1.29       chs 		int rv;
    331      1.35       eeh 		vaddr_t addr = (vaddr_t)kbp->kb_next;
    332      1.29       chs 
    333      1.43   thorpej 		vm_map_lock(kmem_map);
    334      1.29       chs 		rv = uvm_map_checkprot(kmem_map, addr,
    335      1.69     enami 		    addr + sizeof(struct freelist), VM_PROT_WRITE);
    336      1.43   thorpej 		vm_map_unlock(kmem_map);
    337      1.29       chs 
    338      1.51   thorpej 		if (__predict_false(rv == 0)) {
    339      1.69     enami 			printf("Data modified on freelist: "
    340      1.69     enami 			    "word %ld of object %p size %ld previous type %s "
    341      1.69     enami 			    "(invalid addr %p)\n",
    342      1.41       mrg 			    (long)((int32_t *)&kbp->kb_next - (int32_t *)kbp),
    343      1.69     enami 			    va, size, savedtype, kbp->kb_next);
    344      1.27   thorpej #ifdef MALLOCLOG
    345      1.41       mrg 			hitmlog(va);
    346      1.27   thorpej #endif
    347      1.41       mrg 			kbp->kb_next = NULL;
    348      1.29       chs 		}
    349       1.8       cgd 	}
    350      1.11       cgd 
    351      1.11       cgd 	/* Fill the fields that we've used with WEIRD_ADDR */
    352       1.8       cgd #if BYTE_ORDER == BIG_ENDIAN
    353       1.8       cgd 	freep->type = WEIRD_ADDR >> 16;
    354       1.8       cgd #endif
    355       1.8       cgd #if BYTE_ORDER == LITTLE_ENDIAN
    356       1.8       cgd 	freep->type = (short)WEIRD_ADDR;
    357       1.8       cgd #endif
    358      1.11       cgd 	end = (int32_t *)&freep->next +
    359      1.11       cgd 	    (sizeof(freep->next) / sizeof(int32_t));
    360      1.11       cgd 	for (lp = (int32_t *)&freep->next; lp < end; lp++)
    361      1.11       cgd 		*lp = WEIRD_ADDR;
    362      1.11       cgd 
    363      1.11       cgd 	/* and check that the data hasn't been modified. */
    364      1.11       cgd 	end = (int32_t *)&va[copysize];
    365      1.11       cgd 	for (lp = (int32_t *)va; lp < end; lp++) {
    366      1.51   thorpej 		if (__predict_true(*lp == WEIRD_ADDR))
    367       1.8       cgd 			continue;
    368      1.69     enami 		printf("Data modified on freelist: "
    369      1.69     enami 		    "word %ld of object %p size %ld previous type %s "
    370      1.69     enami 		    "(0x%x != 0x%x)\n",
    371      1.69     enami 		    (long)(lp - (int32_t *)va), va, size,
    372      1.21  christos 		    savedtype, *lp, WEIRD_ADDR);
    373      1.27   thorpej #ifdef MALLOCLOG
    374      1.27   thorpej 		hitmlog(va);
    375      1.27   thorpej #endif
    376       1.8       cgd 		break;
    377       1.8       cgd 	}
    378      1.11       cgd 
    379       1.8       cgd 	freep->spare0 = 0;
    380       1.8       cgd #endif /* DIAGNOSTIC */
    381       1.1       cgd #ifdef KMEMSTATS
    382       1.1       cgd 	kup = btokup(va);
    383       1.1       cgd 	if (kup->ku_indx != indx)
    384       1.1       cgd 		panic("malloc: wrong bucket");
    385       1.1       cgd 	if (kup->ku_freecnt == 0)
    386       1.1       cgd 		panic("malloc: lost data");
    387       1.1       cgd 	kup->ku_freecnt--;
    388       1.1       cgd 	kbp->kb_totalfree--;
    389       1.1       cgd 	ksp->ks_memuse += 1 << indx;
    390       1.1       cgd out:
    391       1.1       cgd 	kbp->kb_calls++;
    392       1.1       cgd 	ksp->ks_inuse++;
    393       1.1       cgd 	ksp->ks_calls++;
    394       1.1       cgd 	if (ksp->ks_memuse > ksp->ks_maxused)
    395       1.1       cgd 		ksp->ks_maxused = ksp->ks_memuse;
    396       1.1       cgd #else
    397       1.1       cgd out:
    398       1.1       cgd #endif
    399      1.27   thorpej #ifdef MALLOCLOG
    400      1.27   thorpej 	domlog(va, size, type, 1, file, line);
    401      1.27   thorpej #endif
    402  1.70.4.1   thorpej 	mutex_exit(&malloc_mutex);
    403      1.67     enami 	if ((flags & M_ZERO) != 0)
    404      1.65     lukem 		memset(va, 0, size);
    405       1.1       cgd 	return ((void *) va);
    406       1.1       cgd }
    407       1.1       cgd 
    408       1.1       cgd /*
    409       1.1       cgd  * Free a block of memory allocated by malloc.
    410       1.1       cgd  */
    411      1.27   thorpej #ifdef MALLOCLOG
    412      1.27   thorpej void
    413      1.69     enami _free(void *addr, int type, const char *file, long line)
    414      1.27   thorpej #else
    415       1.1       cgd void
    416      1.69     enami free(void *addr, int type)
    417      1.27   thorpej #endif /* MALLOCLOG */
    418       1.1       cgd {
    419      1.50  augustss 	struct kmembuckets *kbp;
    420      1.50  augustss 	struct kmemusage *kup;
    421      1.50  augustss 	struct freelist *freep;
    422       1.8       cgd 	long size;
    423       1.5    andrew #ifdef DIAGNOSTIC
    424       1.8       cgd 	caddr_t cp;
    425      1.11       cgd 	int32_t *end, *lp;
    426      1.11       cgd 	long alloc, copysize;
    427       1.5    andrew #endif
    428       1.1       cgd #ifdef KMEMSTATS
    429      1.50  augustss 	struct kmemstats *ksp = &kmemstats[type];
    430      1.48   thorpej #endif
    431      1.48   thorpej 
    432      1.62   thorpej #ifdef MALLOC_DEBUG
    433      1.62   thorpej 	if (debug_free(addr, type))
    434      1.62   thorpej 		return;
    435      1.62   thorpej #endif
    436      1.62   thorpej 
    437      1.48   thorpej #ifdef DIAGNOSTIC
    438      1.48   thorpej 	/*
    439      1.48   thorpej 	 * Ensure that we're free'ing something that we could
    440      1.48   thorpej 	 * have allocated in the first place.  That is, check
    441      1.48   thorpej 	 * to see that the address is within kmem_map.
    442      1.48   thorpej 	 */
    443      1.51   thorpej 	if (__predict_false((vaddr_t)addr < kmem_map->header.start ||
    444      1.69     enami 	    (vaddr_t)addr >= kmem_map->header.end))
    445      1.48   thorpej 		panic("free: addr %p not within kmem_map", addr);
    446       1.1       cgd #endif
    447       1.1       cgd 
    448       1.1       cgd 	kup = btokup(addr);
    449       1.1       cgd 	size = 1 << kup->ku_indx;
    450       1.8       cgd 	kbp = &bucket[kup->ku_indx];
    451  1.70.4.1   thorpej 
    452  1.70.4.1   thorpej 	mutex_enter(&malloc_mutex);
    453  1.70.4.1   thorpej 
    454      1.27   thorpej #ifdef MALLOCLOG
    455      1.27   thorpej 	domlog(addr, 0, type, 2, file, line);
    456      1.27   thorpej #endif
    457       1.1       cgd #ifdef DIAGNOSTIC
    458       1.8       cgd 	/*
    459       1.8       cgd 	 * Check for returns of data that do not point to the
    460       1.8       cgd 	 * beginning of the allocation.
    461       1.8       cgd 	 */
    462      1.49   thorpej 	if (size > PAGE_SIZE)
    463      1.49   thorpej 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
    464       1.1       cgd 	else
    465       1.1       cgd 		alloc = addrmask[kup->ku_indx];
    466       1.8       cgd 	if (((u_long)addr & alloc) != 0)
    467      1.15  christos 		panic("free: unaligned addr %p, size %ld, type %s, mask %ld\n",
    468      1.69     enami 		    addr, size, memname[type], alloc);
    469       1.1       cgd #endif /* DIAGNOSTIC */
    470       1.1       cgd 	if (size > MAXALLOCSAVE) {
    471      1.35       eeh 		uvm_km_free(kmem_map, (vaddr_t)addr, ctob(kup->ku_pagecnt));
    472       1.1       cgd #ifdef KMEMSTATS
    473       1.1       cgd 		size = kup->ku_pagecnt << PGSHIFT;
    474       1.1       cgd 		ksp->ks_memuse -= size;
    475       1.1       cgd 		kup->ku_indx = 0;
    476       1.1       cgd 		kup->ku_pagecnt = 0;
    477       1.1       cgd 		if (ksp->ks_memuse + size >= ksp->ks_limit &&
    478       1.1       cgd 		    ksp->ks_memuse < ksp->ks_limit)
    479       1.1       cgd 			wakeup((caddr_t)ksp);
    480       1.1       cgd 		ksp->ks_inuse--;
    481       1.1       cgd 		kbp->kb_total -= 1;
    482       1.1       cgd #endif
    483  1.70.4.1   thorpej 		mutex_exit(&malloc_mutex);
    484       1.1       cgd 		return;
    485       1.1       cgd 	}
    486       1.8       cgd 	freep = (struct freelist *)addr;
    487       1.8       cgd #ifdef DIAGNOSTIC
    488       1.8       cgd 	/*
    489       1.8       cgd 	 * Check for multiple frees. Use a quick check to see if
    490       1.8       cgd 	 * it looks free before laboriously searching the freelist.
    491       1.8       cgd 	 */
    492      1.51   thorpej 	if (__predict_false(freep->spare0 == WEIRD_ADDR)) {
    493      1.16       cgd 		for (cp = kbp->kb_next; cp;
    494      1.16       cgd 		    cp = ((struct freelist *)cp)->next) {
    495       1.8       cgd 			if (addr != cp)
    496       1.8       cgd 				continue;
    497      1.22  christos 			printf("multiply freed item %p\n", addr);
    498      1.27   thorpej #ifdef MALLOCLOG
    499      1.27   thorpej 			hitmlog(addr);
    500      1.27   thorpej #endif
    501       1.8       cgd 			panic("free: duplicated free");
    502       1.8       cgd 		}
    503       1.8       cgd 	}
    504      1.38       chs #ifdef LOCKDEBUG
    505      1.38       chs 	/*
    506      1.38       chs 	 * Check if we're freeing a locked simple lock.
    507      1.38       chs 	 */
    508      1.40       chs 	simple_lock_freecheck(addr, (char *)addr + size);
    509      1.38       chs #endif
    510       1.8       cgd 	/*
    511       1.8       cgd 	 * Copy in known text to detect modification after freeing
    512       1.8       cgd 	 * and to make it look free. Also, save the type being freed
    513       1.8       cgd 	 * so we can list likely culprit if modification is detected
    514       1.8       cgd 	 * when the object is reallocated.
    515       1.8       cgd 	 */
    516       1.8       cgd 	copysize = size < MAX_COPY ? size : MAX_COPY;
    517      1.11       cgd 	end = (int32_t *)&((caddr_t)addr)[copysize];
    518      1.11       cgd 	for (lp = (int32_t *)addr; lp < end; lp++)
    519       1.8       cgd 		*lp = WEIRD_ADDR;
    520       1.8       cgd 	freep->type = type;
    521       1.8       cgd #endif /* DIAGNOSTIC */
    522       1.1       cgd #ifdef KMEMSTATS
    523       1.1       cgd 	kup->ku_freecnt++;
    524      1.36   thorpej 	if (kup->ku_freecnt >= kbp->kb_elmpercl) {
    525       1.1       cgd 		if (kup->ku_freecnt > kbp->kb_elmpercl)
    526       1.1       cgd 			panic("free: multiple frees");
    527       1.1       cgd 		else if (kbp->kb_totalfree > kbp->kb_highwat)
    528       1.1       cgd 			kbp->kb_couldfree++;
    529      1.36   thorpej 	}
    530       1.1       cgd 	kbp->kb_totalfree++;
    531       1.1       cgd 	ksp->ks_memuse -= size;
    532       1.1       cgd 	if (ksp->ks_memuse + size >= ksp->ks_limit &&
    533       1.1       cgd 	    ksp->ks_memuse < ksp->ks_limit)
    534       1.1       cgd 		wakeup((caddr_t)ksp);
    535       1.1       cgd 	ksp->ks_inuse--;
    536       1.1       cgd #endif
    537       1.8       cgd 	if (kbp->kb_next == NULL)
    538       1.8       cgd 		kbp->kb_next = addr;
    539       1.8       cgd 	else
    540       1.8       cgd 		((struct freelist *)kbp->kb_last)->next = addr;
    541       1.8       cgd 	freep->next = NULL;
    542       1.8       cgd 	kbp->kb_last = addr;
    543  1.70.4.1   thorpej 	mutex_exit(&malloc_mutex);
    544      1.20       cgd }
    545      1.20       cgd 
    546      1.20       cgd /*
    547      1.20       cgd  * Change the size of a block of memory.
    548      1.20       cgd  */
    549      1.20       cgd void *
    550      1.69     enami realloc(void *curaddr, unsigned long newsize, int type, int flags)
    551      1.20       cgd {
    552      1.50  augustss 	struct kmemusage *kup;
    553      1.20       cgd 	long cursize;
    554      1.20       cgd 	void *newaddr;
    555      1.20       cgd #ifdef DIAGNOSTIC
    556      1.20       cgd 	long alloc;
    557      1.20       cgd #endif
    558      1.20       cgd 
    559      1.20       cgd 	/*
    560      1.69     enami 	 * realloc() with a NULL pointer is the same as malloc().
    561      1.20       cgd 	 */
    562      1.20       cgd 	if (curaddr == NULL)
    563      1.20       cgd 		return (malloc(newsize, type, flags));
    564      1.20       cgd 
    565      1.20       cgd 	/*
    566      1.69     enami 	 * realloc() with zero size is the same as free().
    567      1.20       cgd 	 */
    568      1.20       cgd 	if (newsize == 0) {
    569      1.20       cgd 		free(curaddr, type);
    570      1.20       cgd 		return (NULL);
    571      1.20       cgd 	}
    572      1.59   thorpej 
    573      1.59   thorpej #ifdef LOCKDEBUG
    574      1.59   thorpej 	if ((flags & M_NOWAIT) == 0)
    575      1.59   thorpej 		simple_lock_only_held(NULL, "realloc");
    576      1.59   thorpej #endif
    577      1.20       cgd 
    578      1.20       cgd 	/*
    579      1.20       cgd 	 * Find out how large the old allocation was (and do some
    580      1.20       cgd 	 * sanity checking).
    581      1.20       cgd 	 */
    582      1.20       cgd 	kup = btokup(curaddr);
    583      1.20       cgd 	cursize = 1 << kup->ku_indx;
    584      1.20       cgd 
    585      1.20       cgd #ifdef DIAGNOSTIC
    586      1.20       cgd 	/*
    587      1.20       cgd 	 * Check for returns of data that do not point to the
    588      1.20       cgd 	 * beginning of the allocation.
    589      1.20       cgd 	 */
    590      1.49   thorpej 	if (cursize > PAGE_SIZE)
    591      1.49   thorpej 		alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
    592      1.20       cgd 	else
    593      1.20       cgd 		alloc = addrmask[kup->ku_indx];
    594      1.20       cgd 	if (((u_long)curaddr & alloc) != 0)
    595      1.69     enami 		panic("realloc: "
    596      1.69     enami 		    "unaligned addr %p, size %ld, type %s, mask %ld\n",
    597      1.69     enami 		    curaddr, cursize, memname[type], alloc);
    598      1.20       cgd #endif /* DIAGNOSTIC */
    599      1.20       cgd 
    600      1.20       cgd 	if (cursize > MAXALLOCSAVE)
    601      1.20       cgd 		cursize = ctob(kup->ku_pagecnt);
    602      1.20       cgd 
    603      1.20       cgd 	/*
    604      1.20       cgd 	 * If we already actually have as much as they want, we're done.
    605      1.20       cgd 	 */
    606      1.20       cgd 	if (newsize <= cursize)
    607      1.20       cgd 		return (curaddr);
    608      1.20       cgd 
    609      1.20       cgd 	/*
    610      1.20       cgd 	 * Can't satisfy the allocation with the existing block.
    611      1.20       cgd 	 * Allocate a new one and copy the data.
    612      1.20       cgd 	 */
    613      1.20       cgd 	newaddr = malloc(newsize, type, flags);
    614      1.51   thorpej 	if (__predict_false(newaddr == NULL)) {
    615      1.20       cgd 		/*
    616      1.69     enami 		 * malloc() failed, because flags included M_NOWAIT.
    617      1.20       cgd 		 * Return NULL to indicate that failure.  The old
    618      1.20       cgd 		 * pointer is still valid.
    619      1.20       cgd 		 */
    620      1.69     enami 		return (NULL);
    621      1.20       cgd 	}
    622      1.34     perry 	memcpy(newaddr, curaddr, cursize);
    623      1.20       cgd 
    624      1.20       cgd 	/*
    625      1.20       cgd 	 * We were successful: free the old allocation and return
    626      1.20       cgd 	 * the new one.
    627      1.20       cgd 	 */
    628      1.20       cgd 	free(curaddr, type);
    629      1.20       cgd 	return (newaddr);
    630      1.70     enami }
    631      1.70     enami 
    632      1.70     enami /*
    633      1.70     enami  * Roundup size to the actual allocation size.
    634      1.70     enami  */
    635      1.70     enami unsigned long
    636      1.70     enami malloc_roundup(unsigned long size)
    637      1.70     enami {
    638      1.70     enami 
    639      1.70     enami 	if (size > MAXALLOCSAVE)
    640      1.70     enami 		return (roundup(size, PAGE_SIZE));
    641      1.70     enami 	else
    642      1.70     enami 		return (1 << BUCKETINDX(size));
    643       1.1       cgd }
    644       1.1       cgd 
    645       1.1       cgd /*
    646      1.49   thorpej  * Compute the number of pages that kmem_map will map, that is,
    647      1.49   thorpej  * the size of the kernel malloc arena.
    648      1.49   thorpej  */
    649      1.49   thorpej void
    650      1.69     enami kmeminit_nkmempages(void)
    651      1.49   thorpej {
    652      1.49   thorpej 	int npages;
    653      1.49   thorpej 
    654      1.49   thorpej 	if (nkmempages != 0) {
    655      1.49   thorpej 		/*
    656      1.49   thorpej 		 * It's already been set (by us being here before, or
    657      1.49   thorpej 		 * by patching or kernel config options), bail out now.
    658      1.49   thorpej 		 */
    659      1.49   thorpej 		return;
    660      1.49   thorpej 	}
    661      1.49   thorpej 
    662      1.49   thorpej 	/*
    663      1.49   thorpej 	 * We use the following (simple) formula:
    664      1.49   thorpej 	 *
    665      1.49   thorpej 	 *	- Starting point is physical memory / 4.
    666      1.49   thorpej 	 *
    667      1.49   thorpej 	 *	- Clamp it down to NKMEMPAGES_MAX.
    668      1.49   thorpej 	 *
    669      1.49   thorpej 	 *	- Round it up to NKMEMPAGES_MIN.
    670      1.49   thorpej 	 */
    671      1.49   thorpej 	npages = physmem / 4;
    672      1.49   thorpej 
    673      1.49   thorpej 	if (npages > NKMEMPAGES_MAX)
    674      1.49   thorpej 		npages = NKMEMPAGES_MAX;
    675      1.49   thorpej 
    676      1.49   thorpej 	if (npages < NKMEMPAGES_MIN)
    677      1.49   thorpej 		npages = NKMEMPAGES_MIN;
    678      1.49   thorpej 
    679      1.49   thorpej 	nkmempages = npages;
    680      1.49   thorpej }
    681      1.49   thorpej 
    682      1.49   thorpej /*
    683       1.1       cgd  * Initialize the kernel memory allocator
    684       1.1       cgd  */
    685      1.12  christos void
    686      1.69     enami kmeminit(void)
    687       1.1       cgd {
    688      1.23       tls #ifdef KMEMSTATS
    689      1.50  augustss 	long indx;
    690      1.23       tls #endif
    691       1.1       cgd 
    692       1.1       cgd #if	((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
    693       1.1       cgd 		ERROR!_kmeminit:_MAXALLOCSAVE_not_power_of_2
    694       1.1       cgd #endif
    695       1.1       cgd #if	(MAXALLOCSAVE > MINALLOCSIZE * 32768)
    696       1.1       cgd 		ERROR!_kmeminit:_MAXALLOCSAVE_too_big
    697       1.1       cgd #endif
    698      1.47     ragge #if	(MAXALLOCSAVE < NBPG)
    699       1.1       cgd 		ERROR!_kmeminit:_MAXALLOCSAVE_too_small
    700       1.1       cgd #endif
    701      1.11       cgd 
    702      1.11       cgd 	if (sizeof(struct freelist) > (1 << MINBUCKET))
    703      1.11       cgd 		panic("minbucket too small/struct freelist too big");
    704      1.11       cgd 
    705      1.49   thorpej 	/*
    706      1.49   thorpej 	 * Compute the number of kmem_map pages, if we have not
    707      1.49   thorpej 	 * done so already.
    708      1.49   thorpej 	 */
    709      1.49   thorpej 	kmeminit_nkmempages();
    710      1.49   thorpej 
    711      1.28       mrg 	kmemusage = (struct kmemusage *) uvm_km_zalloc(kernel_map,
    712      1.69     enami 	    (vsize_t)(nkmempages * sizeof(struct kmemusage)));
    713      1.35       eeh 	kmem_map = uvm_km_suballoc(kernel_map, (vaddr_t *)&kmembase,
    714      1.69     enami 	    (vaddr_t *)&kmemlimit, (vsize_t)(nkmempages << PAGE_SHIFT),
    715      1.69     enami 	    VM_MAP_INTRSAFE, FALSE, &kmem_map_store);
    716       1.1       cgd #ifdef KMEMSTATS
    717       1.1       cgd 	for (indx = 0; indx < MINBUCKET + 16; indx++) {
    718      1.49   thorpej 		if (1 << indx >= PAGE_SIZE)
    719       1.1       cgd 			bucket[indx].kb_elmpercl = 1;
    720       1.1       cgd 		else
    721      1.49   thorpej 			bucket[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
    722       1.1       cgd 		bucket[indx].kb_highwat = 5 * bucket[indx].kb_elmpercl;
    723       1.1       cgd 	}
    724       1.8       cgd 	for (indx = 0; indx < M_LAST; indx++)
    725      1.60   thorpej 		kmemstats[indx].ks_limit =
    726      1.60   thorpej 		    ((u_long)nkmempages << PAGE_SHIFT) * 6U / 10U;
    727      1.62   thorpej #endif
    728      1.62   thorpej #ifdef MALLOC_DEBUG
    729      1.62   thorpej 	debug_malloc_init();
    730       1.1       cgd #endif
    731  1.70.4.1   thorpej 
    732  1.70.4.1   thorpej 	mutex_init(&malloc_mutex, MUTEX_SPIN, IPL_VM);
    733       1.1       cgd }
    734      1.39   thorpej 
    735      1.39   thorpej #ifdef DDB
    736      1.39   thorpej #include <ddb/db_output.h>
    737      1.39   thorpej 
    738      1.39   thorpej /*
    739      1.39   thorpej  * Dump kmem statistics from ddb.
    740      1.39   thorpej  *
    741      1.39   thorpej  * usage: call dump_kmemstats
    742      1.39   thorpej  */
    743      1.69     enami void	dump_kmemstats(void);
    744      1.39   thorpej 
    745      1.39   thorpej void
    746      1.69     enami dump_kmemstats(void)
    747      1.39   thorpej {
    748      1.39   thorpej #ifdef KMEMSTATS
    749      1.39   thorpej 	const char *name;
    750      1.39   thorpej 	int i;
    751      1.39   thorpej 
    752      1.39   thorpej 	for (i = 0; i < M_LAST; i++) {
    753      1.39   thorpej 		name = memname[i] ? memname[i] : "";
    754      1.39   thorpej 
    755      1.39   thorpej 		db_printf("%2d %s%.*s %ld\n", i, name,
    756      1.39   thorpej 		    (int)(20 - strlen(name)), "                    ",
    757      1.39   thorpej 		    kmemstats[i].ks_memuse);
    758      1.39   thorpej 	}
    759      1.39   thorpej #else
    760      1.39   thorpej 	db_printf("Kmem stats are not being collected.\n");
    761      1.39   thorpej #endif /* KMEMSTATS */
    762      1.39   thorpej }
    763      1.39   thorpej #endif /* DDB */
    764