kern_malloc.c revision 95923 
1/*
2 * Copyright (c) 1987, 1991, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 *    must display the following acknowledgement:
15 *	This product includes software developed by the University of
16 *	California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 *    may be used to endorse or promote products derived from this software
19 *    without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 *	@(#)kern_malloc.c	8.3 (Berkeley) 1/4/94
34 * $FreeBSD: head/sys/kern/kern_malloc.c 95923 2002-05-02 07:22:19Z jeff $
35 */
36
37#include "opt_vm.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/kernel.h>
42#include <sys/lock.h>
43#include <sys/malloc.h>
44#include <sys/mbuf.h>
45#include <sys/mutex.h>
46#include <sys/vmmeter.h>
47#include <sys/proc.h>
48#include <sys/sysctl.h>
49
50#include <vm/vm.h>
51#include <vm/vm_param.h>
52#include <vm/vm_kern.h>
53#include <vm/vm_extern.h>
54#include <vm/pmap.h>
55#include <vm/vm_map.h>
56#include <vm/uma.h>
57#include <vm/uma_int.h>
58#include <vm/uma_dbg.h>
59
60#if defined(INVARIANTS) && defined(__i386__)
61#include <machine/cpu.h>
62#endif
63
64/*
65 * When realloc() is called, if the new size is sufficiently smaller than
66 * the old size, realloc() will allocate a new, smaller block to avoid
67 * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
68 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
69 */
70#ifndef REALLOC_FRACTION
71#define	REALLOC_FRACTION	1	/* new block if <= half the size */
72#endif
73
74MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
75MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
76MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
77
78MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
79MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
80
81static void kmeminit(void *);
82SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL)
83
84static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
85
86static struct malloc_type *kmemstatistics;
87static char *kmembase;
88static char *kmemlimit;
89
90#define KMEM_ZSHIFT	4
91#define KMEM_ZBASE	16
92#define KMEM_ZMASK	(KMEM_ZBASE - 1)
93
94#define KMEM_ZMAX	65536
95#define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
96static u_int8_t kmemsize[KMEM_ZSIZE + 1];
97
98/* These won't be powers of two for long */
99struct {
100	int kz_size;
101	char *kz_name;
102	uma_zone_t kz_zone;
103} kmemzones[] = {
104	{16, "16", NULL},
105	{32, "32", NULL},
106	{64, "64", NULL},
107	{128, "128", NULL},
108	{256, "256", NULL},
109	{512, "512", NULL},
110	{1024, "1024", NULL},
111	{2048, "2048", NULL},
112	{4096, "4096", NULL},
113	{8192, "8192", NULL},
114	{16384, "16384", NULL},
115	{32768, "32768", NULL},
116	{65536, "65536", NULL},
117	{0, NULL},
118};
119
120u_int vm_kmem_size;
121
122/*
123 * The malloc_mtx protects the kmemstatistics linked list as well as the
124 * mallochash.
125 */
126
127struct mtx malloc_mtx;
128
129#ifdef MALLOC_PROFILE
130uint64_t krequests[KMEM_ZSIZE + 1];
131
132static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
133#endif
134
135static int sysctl_kern_malloc(SYSCTL_HANDLER_ARGS);
136
137/*
138 *	malloc:
139 *
140 *	Allocate a block of memory.
141 *
142 *	If M_NOWAIT is set, this routine will not block and return NULL if
143 *	the allocation fails.
144 */
145void *
146malloc(size, type, flags)
147	unsigned long size;
148	struct malloc_type *type;
149	int flags;
150{
151	int indx;
152	caddr_t va;
153	uma_zone_t zone;
154	register struct malloc_type *ksp = type;
155
156#if 0
157	if (size == 0)
158		Debugger("zero size malloc");
159#endif
160#if defined(INVARIANTS)
161	if (flags == M_WAITOK)
162		KASSERT(curthread->td_intr_nesting_level == 0,
163		   ("malloc(M_WAITOK) in interrupt context"));
164#endif
165	if (size <= KMEM_ZMAX) {
166		if (size & KMEM_ZMASK)
167			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
168		indx = kmemsize[size >> KMEM_ZSHIFT];
169		zone = kmemzones[indx].kz_zone;
170#ifdef MALLOC_PROFILE
171		krequests[size >> KMEM_ZSHIFT]++;
172#endif
173		va = uma_zalloc(zone, flags);
174		mtx_lock(&ksp->ks_mtx);
175		if (va == NULL)
176			goto out;
177
178		ksp->ks_size |= 1 << indx;
179		size = zone->uz_size;
180	} else {
181		size = roundup(size, PAGE_SIZE);
182		zone = NULL;
183		va = uma_large_malloc(size, flags);
184		mtx_lock(&ksp->ks_mtx);
185		if (va == NULL)
186			goto out;
187	}
188	ksp->ks_memuse += size;
189	ksp->ks_inuse++;
190out:
191	ksp->ks_calls++;
192	if (ksp->ks_memuse > ksp->ks_maxused)
193		ksp->ks_maxused = ksp->ks_memuse;
194
195	mtx_unlock(&ksp->ks_mtx);
196	return ((void *) va);
197}
198
199/*
200 *	free:
201 *
202 *	Free a block of memory allocated by malloc.
203 *
204 *	This routine may not block.
205 */
206void
207free(addr, type)
208	void *addr;
209	struct malloc_type *type;
210{
211	uma_slab_t slab;
212	void *mem;
213	u_long size;
214	register struct malloc_type *ksp = type;
215
216	/* free(NULL, ...) does nothing */
217	if (addr == NULL)
218		return;
219
220	size = 0;
221
222	mem = (void *)((u_long)addr & (~UMA_SLAB_MASK));
223	mtx_lock(&malloc_mtx);
224	slab = hash_sfind(mallochash, mem);
225	mtx_unlock(&malloc_mtx);
226
227	if (slab == NULL)
228		panic("free: address %p(%p) has not been allocated.\n",
229		    addr, mem);
230
231	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
232		size = slab->us_zone->uz_size;
233		uma_zfree_arg(slab->us_zone, addr, slab);
234	} else {
235		size = slab->us_size;
236		uma_large_free(slab);
237	}
238	mtx_lock(&ksp->ks_mtx);
239	ksp->ks_memuse -= size;
240	ksp->ks_inuse--;
241	mtx_unlock(&ksp->ks_mtx);
242}
243
244/*
245 *	realloc: change the size of a memory block
246 */
247void *
248realloc(addr, size, type, flags)
249	void *addr;
250	unsigned long size;
251	struct malloc_type *type;
252	int flags;
253{
254	uma_slab_t slab;
255	unsigned long alloc;
256	void *newaddr;
257
258	/* realloc(NULL, ...) is equivalent to malloc(...) */
259	if (addr == NULL)
260		return (malloc(size, type, flags));
261
262	mtx_lock(&malloc_mtx);
263	slab = hash_sfind(mallochash,
264	    (void *)((u_long)addr & ~(UMA_SLAB_MASK)));
265	mtx_unlock(&malloc_mtx);
266
267	/* Sanity check */
268	KASSERT(slab != NULL,
269	    ("realloc: address %p out of range", (void *)addr));
270
271	/* Get the size of the original block */
272	if (slab->us_zone)
273		alloc = slab->us_zone->uz_size;
274	else
275		alloc = slab->us_size;
276
277	/* Reuse the original block if appropriate */
278	if (size <= alloc
279	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
280		return (addr);
281
282	/* Allocate a new, bigger (or smaller) block */
283	if ((newaddr = malloc(size, type, flags)) == NULL)
284		return (NULL);
285
286	/* Copy over original contents */
287	bcopy(addr, newaddr, min(size, alloc));
288	free(addr, type);
289	return (newaddr);
290}
291
292/*
293 *	reallocf: same as realloc() but free memory on failure.
294 */
295void *
296reallocf(addr, size, type, flags)
297	void *addr;
298	unsigned long size;
299	struct malloc_type *type;
300	int flags;
301{
302	void *mem;
303
304	if ((mem = realloc(addr, size, type, flags)) == NULL)
305		free(addr, type);
306	return (mem);
307}
308
309/*
310 * Initialize the kernel memory allocator
311 */
312/* ARGSUSED*/
313static void
314kmeminit(dummy)
315	void *dummy;
316{
317	u_int8_t indx;
318	u_long npg;
319	u_long mem_size;
320	void *hashmem;
321	u_long hashsize;
322	int highbit;
323	int bits;
324	int i;
325
326	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
327
328	/*
329	 * Try to auto-tune the kernel memory size, so that it is
330	 * more applicable for a wider range of machine sizes.
331	 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while
332	 * a VM_KMEM_SIZE of 12MB is a fair compromise.  The
333	 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
334	 * available, and on an X86 with a total KVA space of 256MB,
335	 * try to keep VM_KMEM_SIZE_MAX at 80MB or below.
336	 *
337	 * Note that the kmem_map is also used by the zone allocator,
338	 * so make sure that there is enough space.
339	 */
340	vm_kmem_size = VM_KMEM_SIZE;
341	mem_size = cnt.v_page_count * PAGE_SIZE;
342
343#if defined(VM_KMEM_SIZE_SCALE)
344	if ((mem_size / VM_KMEM_SIZE_SCALE) > vm_kmem_size)
345		vm_kmem_size = mem_size / VM_KMEM_SIZE_SCALE;
346#endif
347
348#if defined(VM_KMEM_SIZE_MAX)
349	if (vm_kmem_size >= VM_KMEM_SIZE_MAX)
350		vm_kmem_size = VM_KMEM_SIZE_MAX;
351#endif
352
353	/* Allow final override from the kernel environment */
354	TUNABLE_INT_FETCH("kern.vm.kmem.size", &vm_kmem_size);
355
356	/*
357	 * Limit kmem virtual size to twice the physical memory.
358	 * This allows for kmem map sparseness, but limits the size
359	 * to something sane. Be careful to not overflow the 32bit
360	 * ints while doing the check.
361	 */
362	if ((vm_kmem_size / 2) > (cnt.v_page_count * PAGE_SIZE))
363		vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE;
364
365	/*
366	 * In mbuf_init(), we set up submaps for mbufs and clusters, in which
367	 * case we rounddown() (nmbufs * MSIZE) and (nmbclusters * MCLBYTES),
368	 * respectively. Mathematically, this means that what we do here may
369	 * amount to slightly more address space than we need for the submaps,
370	 * but it never hurts to have an extra page in kmem_map.
371	 */
372	npg = (nmbufs * MSIZE + nmbclusters * MCLBYTES + nmbcnt *
373	    sizeof(u_int) + vm_kmem_size) / PAGE_SIZE;
374
375	kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase,
376		(vm_offset_t *)&kmemlimit, (vm_size_t)(npg * PAGE_SIZE));
377	kmem_map->system_map = 1;
378
379	hashsize = npg * sizeof(void *);
380
381	highbit = 0;
382	bits = 0;
383	/* The hash size must be a power of two */
384	for (i = 0; i < 8 * sizeof(hashsize); i++)
385		if (hashsize & (1 << i)) {
386			highbit = i;
387			bits++;
388		}
389	if (bits > 1)
390		hashsize = 1 << (highbit);
391
392	hashmem = (void *)kmem_alloc(kernel_map, (vm_size_t)hashsize);
393	uma_startup2(hashmem, hashsize / sizeof(void *));
394
395	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
396		int size = kmemzones[indx].kz_size;
397		char *name = kmemzones[indx].kz_name;
398
399		kmemzones[indx].kz_zone = uma_zcreate(name, size,
400#ifdef INVARIANTS
401		    trash_ctor, trash_dtor, trash_init, trash_fini,
402#else
403		    NULL, NULL, NULL, NULL,
404#endif
405		    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
406
407		for (;i <= size; i+= KMEM_ZBASE)
408			kmemsize[i >> KMEM_ZSHIFT] = indx;
409
410	}
411}
412
413void
414malloc_init(data)
415	void *data;
416{
417	struct malloc_type *type = (struct malloc_type *)data;
418
419	mtx_lock(&malloc_mtx);
420	if (type->ks_magic != M_MAGIC)
421		panic("malloc type lacks magic");
422
423	if (cnt.v_page_count == 0)
424		panic("malloc_init not allowed before vm init");
425
426	if (type->ks_next != NULL)
427		return;
428
429	type->ks_next = kmemstatistics;
430	kmemstatistics = type;
431	mtx_init(&type->ks_mtx, type->ks_shortdesc, "Malloc Stats", MTX_DEF);
432	mtx_unlock(&malloc_mtx);
433}
434
435void
436malloc_uninit(data)
437	void *data;
438{
439	struct malloc_type *type = (struct malloc_type *)data;
440	struct malloc_type *t;
441
442	mtx_lock(&malloc_mtx);
443	mtx_lock(&type->ks_mtx);
444	if (type->ks_magic != M_MAGIC)
445		panic("malloc type lacks magic");
446
447	if (cnt.v_page_count == 0)
448		panic("malloc_uninit not allowed before vm init");
449
450	if (type == kmemstatistics)
451		kmemstatistics = type->ks_next;
452	else {
453		for (t = kmemstatistics; t->ks_next != NULL; t = t->ks_next) {
454			if (t->ks_next == type) {
455				t->ks_next = type->ks_next;
456				break;
457			}
458		}
459	}
460	type->ks_next = NULL;
461	mtx_destroy(&type->ks_mtx);
462	mtx_unlock(&malloc_mtx);
463}
464
465static int
466sysctl_kern_malloc(SYSCTL_HANDLER_ARGS)
467{
468	struct malloc_type *type;
469	int linesize = 128;
470	int curline;
471	int bufsize;
472	int first;
473	int error;
474	char *buf;
475	char *p;
476	int cnt;
477	int len;
478	int i;
479
480	cnt = 0;
481
482	mtx_lock(&malloc_mtx);
483	for (type = kmemstatistics; type != NULL; type = type->ks_next)
484		cnt++;
485
486	mtx_unlock(&malloc_mtx);
487	bufsize = linesize * (cnt + 1);
488	p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
489	mtx_lock(&malloc_mtx);
490
491	len = snprintf(p, linesize,
492	    "\n        Type  InUse MemUse HighUse Requests  Size(s)\n");
493	p += len;
494
495	for (type = kmemstatistics; cnt != 0 && type != NULL;
496	    type = type->ks_next, cnt--) {
497		if (type->ks_calls == 0)
498			continue;
499
500		curline = linesize - 2;	/* Leave room for the \n */
501		len = snprintf(p, curline, "%13s%6lu%6luK%7luK%9llu",
502			type->ks_shortdesc,
503			type->ks_inuse,
504			(type->ks_memuse + 1023) / 1024,
505			(type->ks_maxused + 1023) / 1024,
506			(long long unsigned)type->ks_calls);
507		curline -= len;
508		p += len;
509
510		first = 1;
511		for (i = 0; i < 8 * sizeof(type->ks_size); i++)
512			if (type->ks_size & (1 << i)) {
513				if (first)
514					len = snprintf(p, curline, "  ");
515				else
516					len = snprintf(p, curline, ",");
517				curline -= len;
518				p += len;
519
520				len = snprintf(p, curline,
521				    "%s", kmemzones[i].kz_name);
522				curline -= len;
523				p += len;
524
525				first = 0;
526			}
527
528		len = snprintf(p, 2, "\n");
529		p += len;
530	}
531
532	mtx_unlock(&malloc_mtx);
533	error = SYSCTL_OUT(req, buf, p - buf);
534
535	free(buf, M_TEMP);
536	return (error);
537}
538
539SYSCTL_OID(_kern, OID_AUTO, malloc, CTLTYPE_STRING|CTLFLAG_RD,
540    NULL, 0, sysctl_kern_malloc, "A", "Malloc Stats");
541
542#ifdef MALLOC_PROFILE
543
544static int
545sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
546{
547	int linesize = 64;
548	uint64_t count;
549	uint64_t waste;
550	uint64_t mem;
551	int bufsize;
552	int error;
553	char *buf;
554	int rsize;
555	int size;
556	char *p;
557	int len;
558	int i;
559
560	bufsize = linesize * (KMEM_ZSIZE + 1);
561	bufsize += 128; 	/* For the stats line */
562	bufsize += 128; 	/* For the banner line */
563	waste = 0;
564	mem = 0;
565
566	p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
567	len = snprintf(p, bufsize,
568	    "\n  Size                    Requests  Real Size\n");
569	bufsize -= len;
570	p += len;
571
572	for (i = 0; i < KMEM_ZSIZE; i++) {
573		size = i << KMEM_ZSHIFT;
574		rsize = kmemzones[kmemsize[i]].kz_size;
575		count = (long long unsigned)krequests[i];
576
577		len = snprintf(p, bufsize, "%6d%28llu%11d\n",
578		    size, (unsigned long long)count, rsize);
579		bufsize -= len;
580		p += len;
581
582		if ((rsize * count) > (size * count))
583			waste += (rsize * count) - (size * count);
584		mem += (rsize * count);
585	}
586
587	len = snprintf(p, bufsize,
588	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
589	    (unsigned long long)mem, (unsigned long long)waste);
590	p += len;
591
592	error = SYSCTL_OUT(req, buf, p - buf);
593
594	free(buf, M_TEMP);
595	return (error);
596}
597
598SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
599    NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
600#endif /* MALLOC_PROFILE */
601