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
|
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 59#if defined(INVARIANTS) && defined(__i386__) 60#include <machine/cpu.h> 61#endif 62 63/* 64 * When realloc() is called, if the new size is sufficiently smaller than 65 * the old size, realloc() will allocate a new, smaller block to avoid 66 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 67 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 68 */ 69#ifndef REALLOC_FRACTION 70#define REALLOC_FRACTION 1 /* new block if <= half the size */ 71#endif 72 73MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 74MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 75MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 76 77MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); 78MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); 79 80static void kmeminit(void *); 81SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL) 82 83static MALLOC_DEFINE(M_FREE, "free", "should be on free list"); 84 85static struct malloc_type *kmemstatistics; 86static char *kmembase; 87static char *kmemlimit; 88 89#define KMEM_ZSHIFT 4 90#define KMEM_ZBASE 16 91#define KMEM_ZMASK (KMEM_ZBASE - 1) 92 93#define KMEM_ZMAX 65536 94#define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 95static u_int8_t kmemsize[KMEM_ZSIZE + 1]; 96 97/* These won't be powers of two for long */ 98struct { 99 int kz_size; 100 char *kz_name; 101 uma_zone_t kz_zone; 102} kmemzones[] = { 103 {16, "16", NULL}, 104 {32, "32", NULL}, 105 {64, "64", NULL}, 106 {128, "128", NULL}, 107 {256, "256", NULL}, 108 {512, "512", NULL}, 109 {1024, "1024", NULL}, 110 {2048, "2048", NULL}, 111 {4096, "4096", NULL}, 112 {8192, "8192", NULL}, 113 {16384, "16384", NULL}, 114 {32768, "32768", NULL}, 115 {65536, "65536", NULL}, 116 {0, NULL}, 117}; 118 119u_int vm_kmem_size; 120static struct mtx malloc_mtx; 121 122#ifdef MALLOC_PROFILE 123uint64_t krequests[KMEM_ZSIZE + 1]; 124 125static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 126#endif 127 128static int sysctl_kern_malloc(SYSCTL_HANDLER_ARGS); 129 130/* 131 * malloc: 132 * 133 * Allocate a block of memory. 134 * 135 * If M_NOWAIT is set, this routine will not block and return NULL if 136 * the allocation fails. 137 */ 138void * 139malloc(size, type, flags) 140 unsigned long size; 141 struct malloc_type *type; 142 int flags; 143{ 144 int indx; 145 caddr_t va; 146 uma_zone_t zone; 147 register struct malloc_type *ksp = type; 148 149#if 0 150 if (size == 0) 151 Debugger("zero size malloc"); 152#endif 153#if defined(INVARIANTS) 154 if (flags == M_WAITOK) 155 KASSERT(curthread->td_intr_nesting_level == 0, 156 ("malloc(M_WAITOK) in interrupt context")); 157#endif 158 if (size <= KMEM_ZMAX) { 159 if (size & KMEM_ZMASK) 160 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 161 indx = kmemsize[size >> KMEM_ZSHIFT]; 162 zone = kmemzones[indx].kz_zone; 163#ifdef MALLOC_PROFILE 164 krequests[size >> KMEM_ZSHIFT]++; 165#endif 166 va = uma_zalloc(zone, flags); 167 if (va == NULL) 168 goto out; 169 170 ksp->ks_size |= 1 << indx; 171 size = zone->uz_size; 172 } else { 173 size = roundup(size, PAGE_SIZE); 174 zone = NULL; 175 va = uma_large_malloc(size, flags); 176 if (va == NULL) 177 goto out; 178 } 179 ksp->ks_memuse += size; 180 ksp->ks_inuse++; 181out: 182 ksp->ks_calls++; 183 if (ksp->ks_memuse > ksp->ks_maxused) 184 ksp->ks_maxused = ksp->ks_memuse; 185
|
186 return ((void *) va); 187} 188 189/* 190 * free: 191 * 192 * Free a block of memory allocated by malloc. 193 * 194 * This routine may not block. 195 */ 196void 197free(addr, type) 198 void *addr; 199 struct malloc_type *type; 200{ 201 uma_slab_t slab; 202 void *mem; 203 u_long size; 204 register struct malloc_type *ksp = type; 205 206 /* free(NULL, ...) does nothing */ 207 if (addr == NULL) 208 return; 209 210 if ((u_long)addr & 3) { /* XXX: Jeff: find better value for 3 */ 211 printf("free(9)'ing unaligned pointer %p\n", addr); 212 Debugger("Don't do that..."); 213 return; 214 } 215 216 size = 0; 217 218 mem = (void *)((u_long)addr & (~UMA_SLAB_MASK)); 219 slab = hash_sfind(mallochash, mem); 220 221 if (slab == NULL) 222 panic("free: address %p(%p) has not been allocated.\n", 223 addr, mem); 224 225 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 226 size = slab->us_zone->uz_size; 227 uma_zfree_arg(slab->us_zone, addr, slab); 228 } else { 229 size = slab->us_size; 230 uma_large_free(slab); 231 } 232 ksp->ks_memuse -= size; 233 ksp->ks_inuse--; 234} 235 236/* 237 * realloc: change the size of a memory block 238 */ 239void * 240realloc(addr, size, type, flags) 241 void *addr; 242 unsigned long size; 243 struct malloc_type *type; 244 int flags; 245{ 246 uma_slab_t slab; 247 unsigned long alloc; 248 void *newaddr; 249 250 /* realloc(NULL, ...) is equivalent to malloc(...) */ 251 if (addr == NULL) 252 return (malloc(size, type, flags)); 253 254 slab = hash_sfind(mallochash, 255 (void *)((u_long)addr & ~(UMA_SLAB_MASK))); 256 257 /* Sanity check */ 258 KASSERT(slab != NULL, 259 ("realloc: address %p out of range", (void *)addr)); 260 261 /* Get the size of the original block */ 262 if (slab->us_zone) 263 alloc = slab->us_zone->uz_size; 264 else 265 alloc = slab->us_size; 266 267 /* Reuse the original block if appropriate */ 268 if (size <= alloc 269 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 270 return (addr); 271 272 /* Allocate a new, bigger (or smaller) block */ 273 if ((newaddr = malloc(size, type, flags)) == NULL) 274 return (NULL); 275 276 /* Copy over original contents */ 277 bcopy(addr, newaddr, min(size, alloc)); 278 free(addr, type); 279 return (newaddr); 280} 281 282/* 283 * reallocf: same as realloc() but free memory on failure. 284 */ 285void * 286reallocf(addr, size, type, flags) 287 void *addr; 288 unsigned long size; 289 struct malloc_type *type; 290 int flags; 291{ 292 void *mem; 293 294 if ((mem = realloc(addr, size, type, flags)) == NULL) 295 free(addr, type); 296 return (mem); 297} 298 299/* 300 * Initialize the kernel memory allocator 301 */ 302/* ARGSUSED*/ 303static void 304kmeminit(dummy) 305 void *dummy; 306{ 307 u_int8_t indx; 308 u_long npg; 309 u_long mem_size; 310 void *hashmem; 311 u_long hashsize; 312 int highbit; 313 int bits; 314 int i; 315 316 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 317 318 /* 319 * Try to auto-tune the kernel memory size, so that it is 320 * more applicable for a wider range of machine sizes. 321 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while 322 * a VM_KMEM_SIZE of 12MB is a fair compromise. The 323 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space 324 * available, and on an X86 with a total KVA space of 256MB, 325 * try to keep VM_KMEM_SIZE_MAX at 80MB or below. 326 * 327 * Note that the kmem_map is also used by the zone allocator, 328 * so make sure that there is enough space. 329 */ 330 vm_kmem_size = VM_KMEM_SIZE; 331 mem_size = cnt.v_page_count * PAGE_SIZE; 332 333#if defined(VM_KMEM_SIZE_SCALE) 334 if ((mem_size / VM_KMEM_SIZE_SCALE) > vm_kmem_size) 335 vm_kmem_size = mem_size / VM_KMEM_SIZE_SCALE; 336#endif 337 338#if defined(VM_KMEM_SIZE_MAX) 339 if (vm_kmem_size >= VM_KMEM_SIZE_MAX) 340 vm_kmem_size = VM_KMEM_SIZE_MAX; 341#endif 342 343 /* Allow final override from the kernel environment */ 344 TUNABLE_INT_FETCH("kern.vm.kmem.size", &vm_kmem_size); 345 346 /* 347 * Limit kmem virtual size to twice the physical memory. 348 * This allows for kmem map sparseness, but limits the size 349 * to something sane. Be careful to not overflow the 32bit 350 * ints while doing the check. 351 */ 352 if ((vm_kmem_size / 2) > (cnt.v_page_count * PAGE_SIZE)) 353 vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE; 354 355 /* 356 * In mbuf_init(), we set up submaps for mbufs and clusters, in which 357 * case we rounddown() (nmbufs * MSIZE) and (nmbclusters * MCLBYTES), 358 * respectively. Mathematically, this means that what we do here may 359 * amount to slightly more address space than we need for the submaps, 360 * but it never hurts to have an extra page in kmem_map. 361 */ 362 npg = (nmbufs * MSIZE + nmbclusters * MCLBYTES + nmbcnt * 363 sizeof(u_int) + vm_kmem_size) / PAGE_SIZE; 364 365 kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase, 366 (vm_offset_t *)&kmemlimit, (vm_size_t)(npg * PAGE_SIZE)); 367 kmem_map->system_map = 1; 368 369 hashsize = npg * sizeof(void *); 370 371 highbit = 0; 372 bits = 0; 373 /* The hash size must be a power of two */ 374 for (i = 0; i < 8 * sizeof(hashsize); i++) 375 if (hashsize & (1 << i)) { 376 highbit = i; 377 bits++; 378 } 379 if (bits > 1) 380 hashsize = 1 << (highbit); 381 382 hashmem = (void *)kmem_alloc(kernel_map, (vm_size_t)hashsize); 383 uma_startup2(hashmem, hashsize / sizeof(void *)); 384 385 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 386 int size = kmemzones[indx].kz_size; 387 char *name = kmemzones[indx].kz_name; 388 389 kmemzones[indx].kz_zone = uma_zcreate(name, size, NULL, NULL, 390 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 391 392 for (;i <= size; i+= KMEM_ZBASE) 393 kmemsize[i >> KMEM_ZSHIFT] = indx; 394 395 } 396} 397 398void 399malloc_init(data) 400 void *data; 401{ 402 struct malloc_type *type = (struct malloc_type *)data; 403 404 mtx_lock(&malloc_mtx); 405 if (type->ks_magic != M_MAGIC) 406 panic("malloc type lacks magic"); 407 408 if (cnt.v_page_count == 0) 409 panic("malloc_init not allowed before vm init"); 410 411 if (type->ks_next != NULL) 412 return; 413 414 type->ks_next = kmemstatistics; 415 kmemstatistics = type; 416 mtx_unlock(&malloc_mtx); 417} 418 419void 420malloc_uninit(data) 421 void *data; 422{ 423 struct malloc_type *type = (struct malloc_type *)data; 424 struct malloc_type *t; 425 426 mtx_lock(&malloc_mtx); 427 if (type->ks_magic != M_MAGIC) 428 panic("malloc type lacks magic"); 429 430 if (cnt.v_page_count == 0) 431 panic("malloc_uninit not allowed before vm init"); 432 433 if (type == kmemstatistics) 434 kmemstatistics = type->ks_next; 435 else { 436 for (t = kmemstatistics; t->ks_next != NULL; t = t->ks_next) { 437 if (t->ks_next == type) { 438 t->ks_next = type->ks_next; 439 break; 440 } 441 } 442 } 443 type->ks_next = NULL; 444 mtx_unlock(&malloc_mtx); 445} 446 447static int 448sysctl_kern_malloc(SYSCTL_HANDLER_ARGS) 449{ 450 struct malloc_type *type; 451 int linesize = 128; 452 int curline; 453 int bufsize; 454 int first; 455 int error; 456 char *buf; 457 char *p; 458 int cnt; 459 int len; 460 int i; 461 462 cnt = 0; 463 464 mtx_lock(&malloc_mtx); 465 for (type = kmemstatistics; type != NULL; type = type->ks_next) 466 cnt++; 467 468 bufsize = linesize * (cnt + 1); 469 p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO); 470 471 len = snprintf(p, linesize, 472 "\n Type InUse MemUse HighUse Requests Size(s)\n"); 473 p += len; 474 475 for (type = kmemstatistics; cnt != 0 && type != NULL; 476 type = type->ks_next, cnt--) { 477 if (type->ks_calls == 0) 478 continue; 479 480 curline = linesize - 2; /* Leave room for the \n */ 481 len = snprintf(p, curline, "%13s%6ld%6ldK%7ldK%9llu", 482 type->ks_shortdesc, 483 type->ks_inuse, 484 (type->ks_memuse + 1023) / 1024, 485 (type->ks_maxused + 1023) / 1024, 486 (long long unsigned)type->ks_calls); 487 curline -= len; 488 p += len; 489 490 first = 1; 491 for (i = 0; i < 8 * sizeof(type->ks_size); i++) 492 if (type->ks_size & (1 << i)) { 493 if (first) 494 len = snprintf(p, curline, " "); 495 else 496 len = snprintf(p, curline, ","); 497 curline -= len; 498 p += len; 499 500 len = snprintf(p, curline, 501 "%s", kmemzones[i].kz_name); 502 curline -= len; 503 p += len; 504 505 first = 0; 506 } 507 508 len = snprintf(p, 2, "\n"); 509 p += len; 510 } 511 512 mtx_unlock(&malloc_mtx); 513 error = SYSCTL_OUT(req, buf, p - buf); 514 515 free(buf, M_TEMP); 516 return (error); 517} 518 519SYSCTL_OID(_kern, OID_AUTO, malloc, CTLTYPE_STRING|CTLFLAG_RD, 520 NULL, 0, sysctl_kern_malloc, "A", "Malloc Stats"); 521 522#ifdef MALLOC_PROFILE 523 524static int 525sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 526{ 527 int linesize = 64; 528 uint64_t count; 529 uint64_t waste; 530 uint64_t mem; 531 int bufsize; 532 int error; 533 char *buf; 534 int rsize; 535 int size; 536 char *p; 537 int len; 538 int i; 539 540 bufsize = linesize * (KMEM_ZSIZE + 1); 541 bufsize += 128; /* For the stats line */ 542 bufsize += 128; /* For the banner line */ 543 waste = 0; 544 mem = 0; 545 546 p = buf = (char *)malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO); 547 len = snprintf(p, bufsize, 548 "\n Size Requests Real Size\n"); 549 bufsize -= len; 550 p += len; 551 552 for (i = 0; i < KMEM_ZSIZE; i++) { 553 size = i << KMEM_ZSHIFT; 554 rsize = kmemzones[kmemsize[i]].kz_size; 555 count = (long long unsigned)krequests[i]; 556 557 len = snprintf(p, bufsize, "%6d%28llu%11d\n", 558 size, (unsigned long long)count, rsize); 559 bufsize -= len; 560 p += len; 561 562 if ((rsize * count) > (size * count)) 563 waste += (rsize * count) - (size * count); 564 mem += (rsize * count); 565 } 566 567 len = snprintf(p, bufsize, 568 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 569 (unsigned long long)mem, (unsigned long long)waste); 570 p += len; 571 572 error = SYSCTL_OUT(req, buf, p - buf); 573 574 free(buf, M_TEMP); 575 return (error); 576} 577 578SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 579 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 580#endif /* MALLOC_PROFILE */
|