kern_malloc.c revision 62231
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 62231 2000-06-29 03:41:30Z bp $ 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 58MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 59MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 60MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 61 62MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); 63MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); 64 65static void kmeminit __P((void *)); 66SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL) 67 68static MALLOC_DEFINE(M_FREE, "free", "should be on free list"); 69 70static struct malloc_type *kmemstatistics; 71static struct kmembuckets bucket[MINBUCKET + 16]; 72static struct kmemusage *kmemusage; 73static char *kmembase; 74static char *kmemlimit; 75 76u_int vm_kmem_size; 77 78#ifdef INVARIANTS 79/* 80 * This structure provides a set of masks to catch unaligned frees. 81 */ 82static long addrmask[] = { 0, 83 0x00000001, 0x00000003, 0x00000007, 0x0000000f, 84 0x0000001f, 0x0000003f, 0x0000007f, 0x000000ff, 85 0x000001ff, 0x000003ff, 0x000007ff, 0x00000fff, 86 0x00001fff, 0x00003fff, 0x00007fff, 0x0000ffff, 87}; 88 89/* 90 * The WEIRD_ADDR is used as known text to copy into free objects so 91 * that modifications after frees can be detected. 92 */ 93#define WEIRD_ADDR 0xdeadc0de 94#define MAX_COPY 64 95 96/* 97 * Normally the first word of the structure is used to hold the list 98 * pointer for free objects. However, when running with diagnostics, 99 * we use the third and fourth fields, so as to catch modifications 100 * in the most commonly trashed first two words. 101 */ 102struct freelist { 103 long spare0; 104 struct malloc_type *type; 105 long spare1; 106 caddr_t next; 107}; 108#else /* !INVARIANTS */ 109struct freelist { 110 caddr_t next; 111}; 112#endif /* INVARIANTS */ 113 114/* 115 * malloc: 116 * 117 * Allocate a block of memory. 118 * 119 * If M_NOWAIT is set, this routine will not block and return NULL if 120 * the allocation fails. 121 * 122 * If M_ASLEEP is set (M_NOWAIT must also be set), this routine 123 * will have the side effect of calling asleep() if it returns NULL, 124 * allowing the parent to await() at some future time. 125 */ 126void * 127malloc(size, type, flags) 128 unsigned long size; 129 struct malloc_type *type; 130 int flags; 131{ 132 register struct kmembuckets *kbp; 133 register struct kmemusage *kup; 134 register struct freelist *freep; 135 long indx, npg, allocsize; 136 int s; 137 caddr_t va, cp, savedlist; 138#ifdef INVARIANTS 139 long *end, *lp; 140 int copysize; 141 const char *savedtype; 142#endif 143 register struct malloc_type *ksp = type; 144 145#if defined(INVARIANTS) && defined(__i386__) 146 if (flags == M_WAITOK) 147 KASSERT(intr_nesting_level == 0, 148 ("malloc(M_WAITOK) in interrupt context")); 149#endif 150 indx = BUCKETINDX(size); 151 kbp = &bucket[indx]; 152 s = splmem(); 153 while (ksp->ks_memuse >= ksp->ks_limit) { 154 if (flags & M_ASLEEP) { 155 if (ksp->ks_limblocks < 65535) 156 ksp->ks_limblocks++; 157 asleep((caddr_t)ksp, PSWP+2, type->ks_shortdesc, 0); 158 } 159 if (flags & M_NOWAIT) { 160 splx(s); 161 return ((void *) NULL); 162 } 163 if (ksp->ks_limblocks < 65535) 164 ksp->ks_limblocks++; 165 tsleep((caddr_t)ksp, PSWP+2, type->ks_shortdesc, 0); 166 } 167 ksp->ks_size |= 1 << indx; 168#ifdef INVARIANTS 169 copysize = 1 << indx < MAX_COPY ? 1 << indx : MAX_COPY; 170#endif 171 if (kbp->kb_next == NULL) { 172 kbp->kb_last = NULL; 173 if (size > MAXALLOCSAVE) 174 allocsize = roundup(size, PAGE_SIZE); 175 else 176 allocsize = 1 << indx; 177 npg = btoc(allocsize); 178 va = (caddr_t) kmem_malloc(kmem_map, (vm_size_t)ctob(npg), flags); 179 if (va == NULL) { 180 splx(s); 181 return ((void *) NULL); 182 } 183 kbp->kb_total += kbp->kb_elmpercl; 184 kup = btokup(va); 185 kup->ku_indx = indx; 186 if (allocsize > MAXALLOCSAVE) { 187 if (npg > 65535) 188 panic("malloc: allocation too large"); 189 kup->ku_pagecnt = npg; 190 ksp->ks_memuse += allocsize; 191 goto out; 192 } 193 kup->ku_freecnt = kbp->kb_elmpercl; 194 kbp->kb_totalfree += kbp->kb_elmpercl; 195 /* 196 * Just in case we blocked while allocating memory, 197 * and someone else also allocated memory for this 198 * bucket, don't assume the list is still empty. 199 */ 200 savedlist = kbp->kb_next; 201 kbp->kb_next = cp = va + (npg * PAGE_SIZE) - allocsize; 202 for (;;) { 203 freep = (struct freelist *)cp; 204#ifdef INVARIANTS 205 /* 206 * Copy in known text to detect modification 207 * after freeing. 208 */ 209 end = (long *)&cp[copysize]; 210 for (lp = (long *)cp; lp < end; lp++) 211 *lp = WEIRD_ADDR; 212 freep->type = M_FREE; 213#endif /* INVARIANTS */ 214 if (cp <= va) 215 break; 216 cp -= allocsize; 217 freep->next = cp; 218 } 219 freep->next = savedlist; 220 if (kbp->kb_last == NULL) 221 kbp->kb_last = (caddr_t)freep; 222 } 223 va = kbp->kb_next; 224 kbp->kb_next = ((struct freelist *)va)->next; 225#ifdef INVARIANTS 226 freep = (struct freelist *)va; 227 savedtype = (const char *) freep->type->ks_shortdesc; 228#if BYTE_ORDER == BIG_ENDIAN 229 freep->type = (struct malloc_type *)WEIRD_ADDR >> 16; 230#endif 231#if BYTE_ORDER == LITTLE_ENDIAN 232 freep->type = (struct malloc_type *)WEIRD_ADDR; 233#endif 234 if ((intptr_t)(void *)&freep->next & 0x2) 235 freep->next = (caddr_t)((WEIRD_ADDR >> 16)|(WEIRD_ADDR << 16)); 236 else 237 freep->next = (caddr_t)WEIRD_ADDR; 238 end = (long *)&va[copysize]; 239 for (lp = (long *)va; lp < end; lp++) { 240 if (*lp == WEIRD_ADDR) 241 continue; 242 printf("%s %ld of object %p size %lu %s %s (0x%lx != 0x%lx)\n", 243 "Data modified on freelist: word", 244 (long)(lp - (long *)va), (void *)va, size, 245 "previous type", savedtype, *lp, (u_long)WEIRD_ADDR); 246 break; 247 } 248 freep->spare0 = 0; 249#endif /* INVARIANTS */ 250 kup = btokup(va); 251 if (kup->ku_indx != indx) 252 panic("malloc: wrong bucket"); 253 if (kup->ku_freecnt == 0) 254 panic("malloc: lost data"); 255 kup->ku_freecnt--; 256 kbp->kb_totalfree--; 257 ksp->ks_memuse += 1 << indx; 258out: 259 kbp->kb_calls++; 260 ksp->ks_inuse++; 261 ksp->ks_calls++; 262 if (ksp->ks_memuse > ksp->ks_maxused) 263 ksp->ks_maxused = ksp->ks_memuse; 264 splx(s); 265 return ((void *) va); 266} 267 268/* 269 * free: 270 * 271 * Free a block of memory allocated by malloc. 272 * 273 * This routine may not block. 274 */ 275void 276free(addr, type) 277 void *addr; 278 struct malloc_type *type; 279{ 280 register struct kmembuckets *kbp; 281 register struct kmemusage *kup; 282 register struct freelist *freep; 283 long size; 284 int s; 285#ifdef INVARIANTS 286 struct freelist *fp; 287 long *end, *lp, alloc, copysize; 288#endif 289 register struct malloc_type *ksp = type; 290 291 KASSERT(kmembase <= (char *)addr && (char *)addr < kmemlimit, 292 ("free: address %p out of range", (void *)addr)); 293 kup = btokup(addr); 294 size = 1 << kup->ku_indx; 295 kbp = &bucket[kup->ku_indx]; 296 s = splmem(); 297#ifdef INVARIANTS 298 /* 299 * Check for returns of data that do not point to the 300 * beginning of the allocation. 301 */ 302 if (size > PAGE_SIZE) 303 alloc = addrmask[BUCKETINDX(PAGE_SIZE)]; 304 else 305 alloc = addrmask[kup->ku_indx]; 306 if (((uintptr_t)(void *)addr & alloc) != 0) 307 panic("free: unaligned addr %p, size %ld, type %s, mask %ld", 308 (void *)addr, size, type->ks_shortdesc, alloc); 309#endif /* INVARIANTS */ 310 if (size > MAXALLOCSAVE) { 311 kmem_free(kmem_map, (vm_offset_t)addr, ctob(kup->ku_pagecnt)); 312 size = kup->ku_pagecnt << PAGE_SHIFT; 313 ksp->ks_memuse -= size; 314 kup->ku_indx = 0; 315 kup->ku_pagecnt = 0; 316 if (ksp->ks_memuse + size >= ksp->ks_limit && 317 ksp->ks_memuse < ksp->ks_limit) 318 wakeup((caddr_t)ksp); 319 ksp->ks_inuse--; 320 kbp->kb_total -= 1; 321 splx(s); 322 return; 323 } 324 freep = (struct freelist *)addr; 325#ifdef INVARIANTS 326 /* 327 * Check for multiple frees. Use a quick check to see if 328 * it looks free before laboriously searching the freelist. 329 */ 330 if (freep->spare0 == WEIRD_ADDR) { 331 fp = (struct freelist *)kbp->kb_next; 332 while (fp) { 333 if (fp->spare0 != WEIRD_ADDR) 334 panic("free: free item %p modified", fp); 335 else if (addr == (caddr_t)fp) 336 panic("free: multiple freed item %p", addr); 337 fp = (struct freelist *)fp->next; 338 } 339 } 340 /* 341 * Copy in known text to detect modification after freeing 342 * and to make it look free. Also, save the type being freed 343 * so we can list likely culprit if modification is detected 344 * when the object is reallocated. 345 */ 346 copysize = size < MAX_COPY ? size : MAX_COPY; 347 end = (long *)&((caddr_t)addr)[copysize]; 348 for (lp = (long *)addr; lp < end; lp++) 349 *lp = WEIRD_ADDR; 350 freep->type = type; 351#endif /* INVARIANTS */ 352 kup->ku_freecnt++; 353 if (kup->ku_freecnt >= kbp->kb_elmpercl) { 354 if (kup->ku_freecnt > kbp->kb_elmpercl) 355 panic("free: multiple frees"); 356 else if (kbp->kb_totalfree > kbp->kb_highwat) 357 kbp->kb_couldfree++; 358 } 359 kbp->kb_totalfree++; 360 ksp->ks_memuse -= size; 361 if (ksp->ks_memuse + size >= ksp->ks_limit && 362 ksp->ks_memuse < ksp->ks_limit) 363 wakeup((caddr_t)ksp); 364 ksp->ks_inuse--; 365#ifdef OLD_MALLOC_MEMORY_POLICY 366 if (kbp->kb_next == NULL) 367 kbp->kb_next = addr; 368 else 369 ((struct freelist *)kbp->kb_last)->next = addr; 370 freep->next = NULL; 371 kbp->kb_last = addr; 372#else 373 /* 374 * Return memory to the head of the queue for quick reuse. This 375 * can improve performance by improving the probability of the 376 * item being in the cache when it is reused. 377 */ 378 if (kbp->kb_next == NULL) { 379 kbp->kb_next = addr; 380 kbp->kb_last = addr; 381 freep->next = NULL; 382 } else { 383 freep->next = kbp->kb_next; 384 kbp->kb_next = addr; 385 } 386#endif 387 splx(s); 388} 389 390/* 391 * Initialize the kernel memory allocator 392 */ 393/* ARGSUSED*/ 394static void 395kmeminit(dummy) 396 void *dummy; 397{ 398 register long indx; 399 u_long npg; 400 u_long mem_size; 401 u_long xvm_kmem_size; 402 403#if ((MAXALLOCSAVE & (MAXALLOCSAVE - 1)) != 0) 404#error "kmeminit: MAXALLOCSAVE not power of 2" 405#endif 406#if (MAXALLOCSAVE > MINALLOCSIZE * 32768) 407#error "kmeminit: MAXALLOCSAVE too big" 408#endif 409#if (MAXALLOCSAVE < PAGE_SIZE) 410#error "kmeminit: MAXALLOCSAVE too small" 411#endif 412 413 /* 414 * Try to auto-tune the kernel memory size, so that it is 415 * more applicable for a wider range of machine sizes. 416 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while 417 * a VM_KMEM_SIZE of 12MB is a fair compromise. The 418 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space 419 * available, and on an X86 with a total KVA space of 256MB, 420 * try to keep VM_KMEM_SIZE_MAX at 80MB or below. 421 * 422 * Note that the kmem_map is also used by the zone allocator, 423 * so make sure that there is enough space. 424 */ 425 xvm_kmem_size = VM_KMEM_SIZE; 426 mem_size = cnt.v_page_count * PAGE_SIZE; 427 428#if defined(VM_KMEM_SIZE_SCALE) 429 if ((mem_size / VM_KMEM_SIZE_SCALE) > xvm_kmem_size) 430 xvm_kmem_size = mem_size / VM_KMEM_SIZE_SCALE; 431#endif 432 433#if defined(VM_KMEM_SIZE_MAX) 434 if (xvm_kmem_size >= VM_KMEM_SIZE_MAX) 435 xvm_kmem_size = VM_KMEM_SIZE_MAX; 436#endif 437 438 /* Allow final override from the kernel environment */ 439 TUNABLE_INT_FETCH("kern.vm.kmem.size", xvm_kmem_size, vm_kmem_size); 440 441 /* 442 * Limit kmem virtual size to twice the physical memory. 443 * This allows for kmem map sparseness, but limits the size 444 * to something sane. Be careful to not overflow the 32bit 445 * ints while doing the check. 446 */ 447 if ((vm_kmem_size / 2) > (cnt.v_page_count * PAGE_SIZE)) 448 vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE; 449 450 npg = (nmbufs * MSIZE + nmbclusters * MCLBYTES + vm_kmem_size) 451 / PAGE_SIZE; 452 453 kmemusage = (struct kmemusage *) kmem_alloc(kernel_map, 454 (vm_size_t)(npg * sizeof(struct kmemusage))); 455 kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase, 456 (vm_offset_t *)&kmemlimit, (vm_size_t)(npg * PAGE_SIZE)); 457 kmem_map->system_map = 1; 458 for (indx = 0; indx < MINBUCKET + 16; indx++) { 459 if (1 << indx >= PAGE_SIZE) 460 bucket[indx].kb_elmpercl = 1; 461 else 462 bucket[indx].kb_elmpercl = PAGE_SIZE / (1 << indx); 463 bucket[indx].kb_highwat = 5 * bucket[indx].kb_elmpercl; 464 } 465} 466 467void 468malloc_init(data) 469 void *data; 470{ 471 struct malloc_type *type = (struct malloc_type *)data; 472 473 if (type->ks_magic != M_MAGIC) 474 panic("malloc type lacks magic"); 475 476 if (type->ks_limit != 0) 477 return; 478 479 if (cnt.v_page_count == 0) 480 panic("malloc_init not allowed before vm init"); 481 482 /* 483 * The default limits for each malloc region is 1/2 of the 484 * malloc portion of the kmem map size. 485 */ 486 type->ks_limit = vm_kmem_size / 2; 487 type->ks_next = kmemstatistics; 488 kmemstatistics = type; 489} 490 491void 492malloc_uninit(data) 493 void *data; 494{ 495 struct malloc_type *type = (struct malloc_type *)data; 496 struct malloc_type *t; 497 struct kmembuckets *kbp; 498 struct freelist *freep; 499 long indx; 500#ifdef INVARIANTS 501 int s; 502#endif 503 504 if (type->ks_magic != M_MAGIC) 505 panic("malloc type lacks magic"); 506 507 if (cnt.v_page_count == 0) 508 panic("malloc_uninit not allowed before vm init"); 509 510 if (type->ks_limit == 0) 511 panic("malloc_uninit on uninitialized type"); 512 513#ifdef INVARIANTS 514 s = splmem(); 515 for (indx = 0; indx < MINBUCKET + 16; indx++) { 516 kbp = bucket + indx; 517 freep = (struct freelist*)kbp->kb_next; 518 while (freep) { 519 if (freep->type == type) 520 freep->type = M_FREE; 521 freep = (struct freelist*)freep->next; 522 } 523 } 524 splx(s); 525 526 if (type->ks_memuse != 0) 527 printf("malloc_uninit: %ld bytes of '%s' still allocated\n", 528 type->ks_memuse, type->ks_shortdesc); 529#endif 530 531 if (type == kmemstatistics) 532 kmemstatistics = type->ks_next; 533 else { 534 for (t = kmemstatistics; t->ks_next != NULL; t = t->ks_next) { 535 if (t->ks_next == type) { 536 t->ks_next = type->ks_next; 537 break; 538 } 539 } 540 } 541 type->ks_next = NULL; 542 type->ks_limit = 0; 543} 544