Initial import from FreeBSD RELENG_4:
[dragonfly.git] / sys / kern / kern_malloc.c
CommitLineData
984263bc
MD
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: src/sys/kern/kern_malloc.c,v 1.64.2.5 2002/03/16 02:19:51 archie Exp $
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/malloc.h>
43#include <sys/mbuf.h>
44#include <sys/vmmeter.h>
45#include <sys/lock.h>
46
47#include <vm/vm.h>
48#include <vm/vm_param.h>
49#include <vm/vm_kern.h>
50#include <vm/vm_extern.h>
51#include <vm/pmap.h>
52#include <vm/vm_map.h>
53
54#if defined(INVARIANTS) && defined(__i386__)
55#include <machine/cpu.h>
56#endif
57
58/*
59 * When realloc() is called, if the new size is sufficiently smaller than
60 * the old size, realloc() will allocate a new, smaller block to avoid
61 * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
62 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
63 */
64#ifndef REALLOC_FRACTION
65#define REALLOC_FRACTION 1 /* new block if <= half the size */
66#endif
67
68MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
69MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
70MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
71
72MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
73MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
74
75static void kmeminit __P((void *));
76SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL)
77
78static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
79
80static struct malloc_type *kmemstatistics;
81static struct kmembuckets bucket[MINBUCKET + 16];
82static struct kmemusage *kmemusage;
83static char *kmembase;
84static char *kmemlimit;
85
86u_int vm_kmem_size;
87
88#ifdef INVARIANTS
89/*
90 * This structure provides a set of masks to catch unaligned frees.
91 */
92static long addrmask[] = { 0,
93 0x00000001, 0x00000003, 0x00000007, 0x0000000f,
94 0x0000001f, 0x0000003f, 0x0000007f, 0x000000ff,
95 0x000001ff, 0x000003ff, 0x000007ff, 0x00000fff,
96 0x00001fff, 0x00003fff, 0x00007fff, 0x0000ffff,
97};
98
99/*
100 * The WEIRD_ADDR is used as known text to copy into free objects so
101 * that modifications after frees can be detected.
102 */
103#define WEIRD_ADDR 0xdeadc0de
104#define MAX_COPY 64
105
106/*
107 * Normally the first word of the structure is used to hold the list
108 * pointer for free objects. However, when running with diagnostics,
109 * we use the third and fourth fields, so as to catch modifications
110 * in the most commonly trashed first two words.
111 */
112struct freelist {
113 long spare0;
114 struct malloc_type *type;
115 long spare1;
116 caddr_t next;
117};
118#else /* !INVARIANTS */
119struct freelist {
120 caddr_t next;
121};
122#endif /* INVARIANTS */
123
124/*
125 * malloc:
126 *
127 * Allocate a block of memory.
128 *
129 * If M_NOWAIT is set, this routine will not block and return NULL if
130 * the allocation fails.
131 *
132 * If M_ASLEEP is set (M_NOWAIT must also be set), this routine
133 * will have the side effect of calling asleep() if it returns NULL,
134 * allowing the parent to await() at some future time.
135 */
136void *
137malloc(size, type, flags)
138 unsigned long size;
139 struct malloc_type *type;
140 int flags;
141{
142 register struct kmembuckets *kbp;
143 register struct kmemusage *kup;
144 register struct freelist *freep;
145 long indx, npg, allocsize;
146 int s;
147 caddr_t va, cp, savedlist;
148#ifdef INVARIANTS
149 long *end, *lp;
150 int copysize;
151 const char *savedtype;
152#endif
153 register struct malloc_type *ksp = type;
154
155#if defined(INVARIANTS) && defined(__i386__)
156 if (flags == M_WAITOK)
157 KASSERT(intr_nesting_level == 0,
158 ("malloc(M_WAITOK) in interrupt context"));
159#endif
160 /*
161 * Must be at splmem() prior to initializing segment to handle
162 * potential initialization race.
163 */
164
165 s = splmem();
166
167 if (type->ks_limit == 0)
168 malloc_init(type);
169
170 indx = BUCKETINDX(size);
171 kbp = &bucket[indx];
172
173 while (ksp->ks_memuse >= ksp->ks_limit) {
174 if (flags & M_ASLEEP) {
175 if (ksp->ks_limblocks < 65535)
176 ksp->ks_limblocks++;
177 asleep((caddr_t)ksp, PSWP+2, type->ks_shortdesc, 0);
178 }
179 if (flags & M_NOWAIT) {
180 splx(s);
181 return ((void *) NULL);
182 }
183 if (ksp->ks_limblocks < 65535)
184 ksp->ks_limblocks++;
185 tsleep((caddr_t)ksp, PSWP+2, type->ks_shortdesc, 0);
186 }
187 ksp->ks_size |= 1 << indx;
188#ifdef INVARIANTS
189 copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY;
190#endif
191 if (kbp->kb_next == NULL) {
192 kbp->kb_last = NULL;
193 if (size > MAXALLOCSAVE)
194 allocsize = roundup(size, PAGE_SIZE);
195 else
196 allocsize = 1 << indx;
197 npg = btoc(allocsize);
198 va = (caddr_t) kmem_malloc(kmem_map, (vm_size_t)ctob(npg), flags);
199 if (va == NULL) {
200 splx(s);
201 return ((void *) NULL);
202 }
203 kbp->kb_total += kbp->kb_elmpercl;
204 kup = btokup(va);
205 kup->ku_indx = indx;
206 if (allocsize > MAXALLOCSAVE) {
207 if (npg > 65535)
208 panic("malloc: allocation too large");
209 kup->ku_pagecnt = npg;
210 ksp->ks_memuse += allocsize;
211 goto out;
212 }
213 kup->ku_freecnt = kbp->kb_elmpercl;
214 kbp->kb_totalfree += kbp->kb_elmpercl;
215 /*
216 * Just in case we blocked while allocating memory,
217 * and someone else also allocated memory for this
218 * bucket, don't assume the list is still empty.
219 */
220 savedlist = kbp->kb_next;
221 kbp->kb_next = cp = va + (npg * PAGE_SIZE) - allocsize;
222 for (;;) {
223 freep = (struct freelist *)cp;
224#ifdef INVARIANTS
225 /*
226 * Copy in known text to detect modification
227 * after freeing.
228 */
229 end = (long *)&cp[copysize];
230 for (lp = (long *)cp; lp < end; lp++)
231 *lp = WEIRD_ADDR;
232 freep->type = M_FREE;
233#endif /* INVARIANTS */
234 if (cp <= va)
235 break;
236 cp -= allocsize;
237 freep->next = cp;
238 }
239 freep->next = savedlist;
240 if (kbp->kb_last == NULL)
241 kbp->kb_last = (caddr_t)freep;
242 }
243 va = kbp->kb_next;
244 kbp->kb_next = ((struct freelist *)va)->next;
245#ifdef INVARIANTS
246 freep = (struct freelist *)va;
247 savedtype = (const char *) freep->type->ks_shortdesc;
248#if BYTE_ORDER == BIG_ENDIAN
249 freep->type = (struct malloc_type *)WEIRD_ADDR >> 16;
250#endif
251#if BYTE_ORDER == LITTLE_ENDIAN
252 freep->type = (struct malloc_type *)WEIRD_ADDR;
253#endif
254 if ((intptr_t)(void *)&freep->next & 0x2)
255 freep->next = (caddr_t)((WEIRD_ADDR >> 16)|(WEIRD_ADDR << 16));
256 else
257 freep->next = (caddr_t)WEIRD_ADDR;
258 end = (long *)&va[copysize];
259 for (lp = (long *)va; lp < end; lp++) {
260 if (*lp == WEIRD_ADDR)
261 continue;
262 printf("%s %ld of object %p size %lu %s %s (0x%lx != 0x%lx)\n",
263 "Data modified on freelist: word",
264 (long)(lp - (long *)va), (void *)va, size,
265 "previous type", savedtype, *lp, (u_long)WEIRD_ADDR);
266 break;
267 }
268 freep->spare0 = 0;
269#endif /* INVARIANTS */
270 kup = btokup(va);
271 if (kup->ku_indx != indx)
272 panic("malloc: wrong bucket");
273 if (kup->ku_freecnt == 0)
274 panic("malloc: lost data");
275 kup->ku_freecnt--;
276 kbp->kb_totalfree--;
277 ksp->ks_memuse += 1 << indx;
278out:
279 kbp->kb_calls++;
280 ksp->ks_inuse++;
281 ksp->ks_calls++;
282 if (ksp->ks_memuse > ksp->ks_maxused)
283 ksp->ks_maxused = ksp->ks_memuse;
284 splx(s);
285 /* XXX: Do idle pre-zeroing. */
286 if (va != NULL && (flags & M_ZERO))
287 bzero(va, size);
288 return ((void *) va);
289}
290
291/*
292 * free:
293 *
294 * Free a block of memory allocated by malloc.
295 *
296 * This routine may not block.
297 */
298void
299free(addr, type)
300 void *addr;
301 struct malloc_type *type;
302{
303 register struct kmembuckets *kbp;
304 register struct kmemusage *kup;
305 register struct freelist *freep;
306 long size;
307 int s;
308#ifdef INVARIANTS
309 struct freelist *fp;
310 long *end, *lp, alloc, copysize;
311#endif
312 register struct malloc_type *ksp = type;
313
314 if (type->ks_limit == 0)
315 panic("freeing with unknown type (%s)", type->ks_shortdesc);
316
317 /* free(NULL, ...) does nothing */
318 if (addr == NULL)
319 return;
320
321 KASSERT(kmembase <= (char *)addr && (char *)addr < kmemlimit,
322 ("free: address %p out of range", (void *)addr));
323 kup = btokup(addr);
324 size = 1 << kup->ku_indx;
325 kbp = &bucket[kup->ku_indx];
326 s = splmem();
327#ifdef INVARIANTS
328 /*
329 * Check for returns of data that do not point to the
330 * beginning of the allocation.
331 */
332 if (size > PAGE_SIZE)
333 alloc = addrmask[BUCKETINDX(PAGE_SIZE)];
334 else
335 alloc = addrmask[kup->ku_indx];
336 if (((uintptr_t)(void *)addr & alloc) != 0)
337 panic("free: unaligned addr %p, size %ld, type %s, mask %ld",
338 (void *)addr, size, type->ks_shortdesc, alloc);
339#endif /* INVARIANTS */
340 if (size > MAXALLOCSAVE) {
341 kmem_free(kmem_map, (vm_offset_t)addr, ctob(kup->ku_pagecnt));
342 size = kup->ku_pagecnt << PAGE_SHIFT;
343 ksp->ks_memuse -= size;
344 kup->ku_indx = 0;
345 kup->ku_pagecnt = 0;
346 if (ksp->ks_memuse + size >= ksp->ks_limit &&
347 ksp->ks_memuse < ksp->ks_limit)
348 wakeup((caddr_t)ksp);
349 ksp->ks_inuse--;
350 kbp->kb_total -= 1;
351 splx(s);
352 return;
353 }
354 freep = (struct freelist *)addr;
355#ifdef INVARIANTS
356 /*
357 * Check for multiple frees. Use a quick check to see if
358 * it looks free before laboriously searching the freelist.
359 */
360 if (freep->spare0 == WEIRD_ADDR) {
361 fp = (struct freelist *)kbp->kb_next;
362 while (fp) {
363 if (fp->spare0 != WEIRD_ADDR)
364 panic("free: free item %p modified", fp);
365 else if (addr == (caddr_t)fp)
366 panic("free: multiple freed item %p", addr);
367 fp = (struct freelist *)fp->next;
368 }
369 }
370 /*
371 * Copy in known text to detect modification after freeing
372 * and to make it look free. Also, save the type being freed
373 * so we can list likely culprit if modification is detected
374 * when the object is reallocated.
375 */
376 copysize = size < MAX_COPY ? size : MAX_COPY;
377 end = (long *)&((caddr_t)addr)[copysize];
378 for (lp = (long *)addr; lp < end; lp++)
379 *lp = WEIRD_ADDR;
380 freep->type = type;
381#endif /* INVARIANTS */
382 kup->ku_freecnt++;
383 if (kup->ku_freecnt >= kbp->kb_elmpercl) {
384 if (kup->ku_freecnt > kbp->kb_elmpercl)
385 panic("free: multiple frees");
386 else if (kbp->kb_totalfree > kbp->kb_highwat)
387 kbp->kb_couldfree++;
388 }
389 kbp->kb_totalfree++;
390 ksp->ks_memuse -= size;
391 if (ksp->ks_memuse + size >= ksp->ks_limit &&
392 ksp->ks_memuse < ksp->ks_limit)
393 wakeup((caddr_t)ksp);
394 ksp->ks_inuse--;
395#ifdef OLD_MALLOC_MEMORY_POLICY
396 if (kbp->kb_next == NULL)
397 kbp->kb_next = addr;
398 else
399 ((struct freelist *)kbp->kb_last)->next = addr;
400 freep->next = NULL;
401 kbp->kb_last = addr;
402#else
403 /*
404 * Return memory to the head of the queue for quick reuse. This
405 * can improve performance by improving the probability of the
406 * item being in the cache when it is reused.
407 */
408 if (kbp->kb_next == NULL) {
409 kbp->kb_next = addr;
410 kbp->kb_last = addr;
411 freep->next = NULL;
412 } else {
413 freep->next = kbp->kb_next;
414 kbp->kb_next = addr;
415 }
416#endif
417 splx(s);
418}
419
420/*
421 * realloc: change the size of a memory block
422 */
423void *
424realloc(addr, size, type, flags)
425 void *addr;
426 unsigned long size;
427 struct malloc_type *type;
428 int flags;
429{
430 struct kmemusage *kup;
431 unsigned long alloc;
432 void *newaddr;
433
434 /* realloc(NULL, ...) is equivalent to malloc(...) */
435 if (addr == NULL)
436 return (malloc(size, type, flags));
437
438 /* Sanity check */
439 KASSERT(kmembase <= (char *)addr && (char *)addr < kmemlimit,
440 ("realloc: address %p out of range", (void *)addr));
441
442 /* Get the size of the original block */
443 kup = btokup(addr);
444 alloc = 1 << kup->ku_indx;
445 if (alloc > MAXALLOCSAVE)
446 alloc = kup->ku_pagecnt << PAGE_SHIFT;
447
448 /* Reuse the original block if appropriate */
449 if (size <= alloc
450 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
451 return (addr);
452
453 /* Allocate a new, bigger (or smaller) block */
454 if ((newaddr = malloc(size, type, flags)) == NULL)
455 return (NULL);
456
457 /* Copy over original contents */
458 bcopy(addr, newaddr, min(size, alloc));
459 free(addr, type);
460 return (newaddr);
461}
462
463/*
464 * reallocf: same as realloc() but free memory on failure.
465 */
466void *
467reallocf(addr, size, type, flags)
468 void *addr;
469 unsigned long size;
470 struct malloc_type *type;
471 int flags;
472{
473 void *mem;
474
475 if ((mem = realloc(addr, size, type, flags)) == NULL)
476 free(addr, type);
477 return (mem);
478}
479
480/*
481 * Initialize the kernel memory allocator
482 */
483/* ARGSUSED*/
484static void
485kmeminit(dummy)
486 void *dummy;
487{
488 register long indx;
489 u_long npg;
490 u_long mem_size;
491
492#if ((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0)
493#error "kmeminit: MAXALLOCSAVE not power of 2"
494#endif
495#if (MAXALLOCSAVE > MINALLOCSIZE * 32768)
496#error "kmeminit: MAXALLOCSAVE too big"
497#endif
498#if (MAXALLOCSAVE < PAGE_SIZE)
499#error "kmeminit: MAXALLOCSAVE too small"
500#endif
501
502 /*
503 * Try to auto-tune the kernel memory size, so that it is
504 * more applicable for a wider range of machine sizes.
505 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while
506 * a VM_KMEM_SIZE of 12MB is a fair compromise. The
507 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
508 * available, and on an X86 with a total KVA space of 256MB,
509 * try to keep VM_KMEM_SIZE_MAX at 80MB or below.
510 *
511 * Note that the kmem_map is also used by the zone allocator,
512 * so make sure that there is enough space.
513 */
514 vm_kmem_size = VM_KMEM_SIZE;
515 mem_size = cnt.v_page_count * PAGE_SIZE;
516
517#if defined(VM_KMEM_SIZE_SCALE)
518 if ((mem_size / VM_KMEM_SIZE_SCALE) > vm_kmem_size)
519 vm_kmem_size = mem_size / VM_KMEM_SIZE_SCALE;
520#endif
521
522#if defined(VM_KMEM_SIZE_MAX)
523 if (vm_kmem_size >= VM_KMEM_SIZE_MAX)
524 vm_kmem_size = VM_KMEM_SIZE_MAX;
525#endif
526
527 /* Allow final override from the kernel environment */
528 TUNABLE_INT_FETCH("kern.vm.kmem.size", &vm_kmem_size);
529
530 /*
531 * Limit kmem virtual size to twice the physical memory.
532 * This allows for kmem map sparseness, but limits the size
533 * to something sane. Be careful to not overflow the 32bit
534 * ints while doing the check.
535 */
536 if ((vm_kmem_size / 2) > (cnt.v_page_count * PAGE_SIZE))
537 vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE;
538
539 npg = (nmbufs * MSIZE + nmbclusters * MCLBYTES + vm_kmem_size)
540 / PAGE_SIZE;
541
542 kmemusage = (struct kmemusage *) kmem_alloc(kernel_map,
543 (vm_size_t)(npg * sizeof(struct kmemusage)));
544 kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase,
545 (vm_offset_t *)&kmemlimit, (vm_size_t)(npg * PAGE_SIZE));
546 kmem_map->system_map = 1;
547 for (indx = 0; indx < MINBUCKET + 16; indx++) {
548 if (1 << indx >= PAGE_SIZE)
549 bucket[indx].kb_elmpercl = 1;
550 else
551 bucket[indx].kb_elmpercl = PAGE_SIZE / (1 << indx);
552 bucket[indx].kb_highwat = 5 * bucket[indx].kb_elmpercl;
553 }
554}
555
556void
557malloc_init(data)
558 void *data;
559{
560 struct malloc_type *type = (struct malloc_type *)data;
561
562 if (type->ks_magic != M_MAGIC)
563 panic("malloc type lacks magic");
564
565 if (type->ks_limit != 0)
566 return;
567
568 if (cnt.v_page_count == 0)
569 panic("malloc_init not allowed before vm init");
570
571 /*
572 * The default limits for each malloc region is 1/2 of the
573 * malloc portion of the kmem map size.
574 */
575 type->ks_limit = vm_kmem_size / 2;
576 type->ks_next = kmemstatistics;
577 kmemstatistics = type;
578}
579
580void
581malloc_uninit(data)
582 void *data;
583{
584 struct malloc_type *type = (struct malloc_type *)data;
585 struct malloc_type *t;
586#ifdef INVARIANTS
587 struct kmembuckets *kbp;
588 struct freelist *freep;
589 long indx;
590 int s;
591#endif
592
593 if (type->ks_magic != M_MAGIC)
594 panic("malloc type lacks magic");
595
596 if (cnt.v_page_count == 0)
597 panic("malloc_uninit not allowed before vm init");
598
599 if (type->ks_limit == 0)
600 panic("malloc_uninit on uninitialized type");
601
602#ifdef INVARIANTS
603 s = splmem();
604 for (indx = 0; indx < MINBUCKET + 16; indx++) {
605 kbp = bucket + indx;
606 freep = (struct freelist*)kbp->kb_next;
607 while (freep) {
608 if (freep->type == type)
609 freep->type = M_FREE;
610 freep = (struct freelist*)freep->next;
611 }
612 }
613 splx(s);
614
615 if (type->ks_memuse != 0)
616 printf("malloc_uninit: %ld bytes of '%s' still allocated\n",
617 type->ks_memuse, type->ks_shortdesc);
618#endif
619
620 if (type == kmemstatistics)
621 kmemstatistics = type->ks_next;
622 else {
623 for (t = kmemstatistics; t->ks_next != NULL; t = t->ks_next) {
624 if (t->ks_next == type) {
625 t->ks_next = type->ks_next;
626 break;
627 }
628 }
629 }
630 type->ks_next = NULL;
631 type->ks_limit = 0;
632}