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 * $FreeBSD$ 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 #include <sys/time.h> 50 51 #include <vm/vm.h> 52 #include <vm/pmap.h> 53 #include <vm/vm_param.h> 54 #include <vm/vm_kern.h> 55 #include <vm/vm_extern.h> 56 #include <vm/vm_map.h> 57 #include <vm/vm_page.h> 58 #include <vm/uma.h> 59 #include <vm/uma_int.h> 60 #include <vm/uma_dbg.h> 61 62 #if defined(INVARIANTS) && defined(__i386__) 63 #include <machine/cpu.h> 64 #endif 65 66 /* 67 * When realloc() is called, if the new size is sufficiently smaller than 68 * the old size, realloc() will allocate a new, smaller block to avoid 69 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 70 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 71 */ 72 #ifndef REALLOC_FRACTION 73 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 74 #endif 75 76 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 77 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 78 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 79 80 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); 81 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); 82 83 static void kmeminit(void *); 84 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL) 85 86 static MALLOC_DEFINE(M_FREE, "free", "should be on free list"); 87 88 static struct malloc_type *kmemstatistics; 89 static char *kmembase; 90 static char *kmemlimit; 91 92 #define KMEM_ZSHIFT 4 93 #define KMEM_ZBASE 16 94 #define KMEM_ZMASK (KMEM_ZBASE - 1) 95 96 #define KMEM_ZMAX 65536 97 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 98 static u_int8_t kmemsize[KMEM_ZSIZE + 1]; 99 100 /* These won't be powers of two for long */ 101 struct { 102 int kz_size; 103 char *kz_name; 104 uma_zone_t kz_zone; 105 } kmemzones[] = { 106 {16, "16", NULL}, 107 {32, "32", NULL}, 108 {64, "64", NULL}, 109 {128, "128", NULL}, 110 {256, "256", NULL}, 111 {512, "512", NULL}, 112 {1024, "1024", NULL}, 113 {2048, "2048", NULL}, 114 {4096, "4096", NULL}, 115 {8192, "8192", NULL}, 116 {16384, "16384", NULL}, 117 {32768, "32768", NULL}, 118 {65536, "65536", NULL}, 119 {0, NULL}, 120 }; 121 122 u_int vm_kmem_size; 123 124 /* 125 * The malloc_mtx protects the kmemstatistics linked list. 126 */ 127 128 struct mtx malloc_mtx; 129 130 #ifdef MALLOC_PROFILE 131 uint64_t krequests[KMEM_ZSIZE + 1]; 132 133 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 134 #endif 135 136 static int sysctl_kern_malloc(SYSCTL_HANDLER_ARGS); 137 138 /* time_uptime of last malloc(9) failure */ 139 static time_t t_malloc_fail; 140 141 int 142 malloc_last_fail(void) 143 { 144 145 return (time_uptime - t_malloc_fail); 146 } 147 148 /* 149 * malloc: 150 * 151 * Allocate a block of memory. 152 * 153 * If M_NOWAIT is set, this routine will not block and return NULL if 154 * the allocation fails. 155 */ 156 void * 157 malloc(size, type, flags) 158 unsigned long size; 159 struct malloc_type *type; 160 int flags; 161 { 162 int indx; 163 caddr_t va; 164 uma_zone_t zone; 165 #ifdef DIAGNOSTIC 166 unsigned long osize = size; 167 #endif 168 register struct malloc_type *ksp = type; 169 170 #if 0 171 if (size == 0) 172 Debugger("zero size malloc"); 173 #endif 174 if (!(flags & M_NOWAIT)) 175 KASSERT(curthread->td_intr_nesting_level == 0, 176 ("malloc() without M_NOWAIT in interrupt context")); 177 if (size <= KMEM_ZMAX) { 178 if (size & KMEM_ZMASK) 179 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 180 indx = kmemsize[size >> KMEM_ZSHIFT]; 181 zone = kmemzones[indx].kz_zone; 182 #ifdef MALLOC_PROFILE 183 krequests[size >> KMEM_ZSHIFT]++; 184 #endif 185 va = uma_zalloc(zone, flags); 186 mtx_lock(&ksp->ks_mtx); 187 if (va == NULL) 188 goto out; 189 190 ksp->ks_size |= 1 << indx; 191 size = zone->uz_size; 192 } else { 193 size = roundup(size, PAGE_SIZE); 194 zone = NULL; 195 va = uma_large_malloc(size, flags); 196 mtx_lock(&ksp->ks_mtx); 197 if (va == NULL) 198 goto out; 199 } 200 ksp->ks_memuse += size; 201 ksp->ks_inuse++; 202 out: 203 ksp->ks_calls++; 204 if (ksp->ks_memuse > ksp->ks_maxused) 205 ksp->ks_maxused = ksp->ks_memuse; 206 207 mtx_unlock(&ksp->ks_mtx); 208 if (!(flags & M_NOWAIT)) 209 KASSERT(va != NULL, ("malloc() without M_NOWAIT returned NULL")); 210 if (va == NULL) { 211 t_malloc_fail = time_uptime; 212 } 213 #ifdef DIAGNOSTIC 214 if (!(flags & M_ZERO)) { 215 memset(va, 0x70, osize); 216 } 217 #endif 218 return ((void *) va); 219 } 220 221 /* 222 * free: 223 * 224 * Free a block of memory allocated by malloc. 225 * 226 * This routine may not block. 227 */ 228 void 229 free(addr, type) 230 void *addr; 231 struct malloc_type *type; 232 { 233 register struct malloc_type *ksp = type; 234 uma_slab_t slab; 235 u_long size; 236 237 /* free(NULL, ...) does nothing */ 238 if (addr == NULL) 239 return; 240 241 size = 0; 242 243 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 244 245 if (slab == NULL) 246 panic("free: address %p(%p) has not been allocated.\n", 247 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 248 249 250 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 251 #ifdef INVARIANTS 252 struct malloc_type **mtp = addr; 253 #endif 254 size = slab->us_zone->uz_size; 255 #ifdef INVARIANTS 256 /* 257 * Cache a pointer to the malloc_type that most recently freed 258 * this memory here. This way we know who is most likely to 259 * have stepped on it later. 260 * 261 * This code assumes that size is a multiple of 8 bytes for 262 * 64 bit machines 263 */ 264 mtp = (struct malloc_type **) 265 ((unsigned long)mtp & ~UMA_ALIGN_PTR); 266 mtp += (size - sizeof(struct malloc_type *)) / 267 sizeof(struct malloc_type *); 268 *mtp = type; 269 #endif 270 uma_zfree_arg(slab->us_zone, addr, slab); 271 } else { 272 size = slab->us_size; 273 uma_large_free(slab); 274 } 275 mtx_lock(&ksp->ks_mtx); 276 ksp->ks_memuse -= size; 277 ksp->ks_inuse--; 278 mtx_unlock(&ksp->ks_mtx); 279 } 280 281 /* 282 * realloc: change the size of a memory block 283 */ 284 void * 285 realloc(addr, size, type, flags) 286 void *addr; 287 unsigned long size; 288 struct malloc_type *type; 289 int flags; 290 { 291 uma_slab_t slab; 292 unsigned long alloc; 293 void *newaddr; 294 295 /* realloc(NULL, ...) is equivalent to malloc(...) */ 296 if (addr == NULL) 297 return (malloc(size, type, flags)); 298 299 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 300 301 /* Sanity check */ 302 KASSERT(slab != NULL, 303 ("realloc: address %p out of range", (void *)addr)); 304 305 /* Get the size of the original block */ 306 if (slab->us_zone) 307 alloc = slab->us_zone->uz_size; 308 else 309 alloc = slab->us_size; 310 311 /* Reuse the original block if appropriate */ 312 if (size <= alloc 313 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 314 return (addr); 315 316 /* Allocate a new, bigger (or smaller) block */ 317 if ((newaddr = malloc(size, type, flags)) == NULL) 318 return (NULL); 319 320 /* Copy over original contents */ 321 bcopy(addr, newaddr, min(size, alloc)); 322 free(addr, type); 323 return (newaddr); 324 } 325 326 /* 327 * reallocf: same as realloc() but free memory on failure. 328 */ 329 void * 330 reallocf(addr, size, type, flags) 331 void *addr; 332 unsigned long size; 333 struct malloc_type *type; 334 int flags; 335 { 336 void *mem; 337 338 if ((mem = realloc(addr, size, type, flags)) == NULL) 339 free(addr, type); 340 return (mem); 341 } 342 343 /* 344 * Initialize the kernel memory allocator 345 */ 346 /* ARGSUSED*/ 347 static void 348 kmeminit(dummy) 349 void *dummy; 350 { 351 u_int8_t indx; 352 u_long npg; 353 u_long mem_size; 354 int i; 355 356 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 357 358 /* 359 * Try to auto-tune the kernel memory size, so that it is 360 * more applicable for a wider range of machine sizes. 361 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while 362 * a VM_KMEM_SIZE of 12MB is a fair compromise. The 363 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space 364 * available, and on an X86 with a total KVA space of 256MB, 365 * try to keep VM_KMEM_SIZE_MAX at 80MB or below. 366 * 367 * Note that the kmem_map is also used by the zone allocator, 368 * so make sure that there is enough space. 369 */ 370 vm_kmem_size = VM_KMEM_SIZE; 371 mem_size = cnt.v_page_count * PAGE_SIZE; 372 373 #if defined(VM_KMEM_SIZE_SCALE) 374 if ((mem_size / VM_KMEM_SIZE_SCALE) > vm_kmem_size) 375 vm_kmem_size = mem_size / VM_KMEM_SIZE_SCALE; 376 #endif 377 378 #if defined(VM_KMEM_SIZE_MAX) 379 if (vm_kmem_size >= VM_KMEM_SIZE_MAX) 380 vm_kmem_size = VM_KMEM_SIZE_MAX; 381 #endif 382 383 /* Allow final override from the kernel environment */ 384 TUNABLE_INT_FETCH("kern.vm.kmem.size", &vm_kmem_size); 385 386 /* 387 * Limit kmem virtual size to twice the physical memory. 388 * This allows for kmem map sparseness, but limits the size 389 * to something sane. Be careful to not overflow the 32bit 390 * ints while doing the check. 391 */ 392 if ((vm_kmem_size / 2) > (cnt.v_page_count * PAGE_SIZE)) 393 vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE; 394 395 /* 396 * In mbuf_init(), we set up submaps for mbufs and clusters, in which 397 * case we rounddown() (nmbufs * MSIZE) and (nmbclusters * MCLBYTES), 398 * respectively. Mathematically, this means that what we do here may 399 * amount to slightly more address space than we need for the submaps, 400 * but it never hurts to have an extra page in kmem_map. 401 */ 402 npg = (nmbufs * MSIZE + nmbclusters * MCLBYTES + nmbcnt * 403 sizeof(u_int) + vm_kmem_size) / PAGE_SIZE; 404 405 kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase, 406 (vm_offset_t *)&kmemlimit, (vm_size_t)(npg * PAGE_SIZE)); 407 kmem_map->system_map = 1; 408 409 uma_startup2(); 410 411 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 412 int size = kmemzones[indx].kz_size; 413 char *name = kmemzones[indx].kz_name; 414 415 kmemzones[indx].kz_zone = uma_zcreate(name, size, 416 #ifdef INVARIANTS 417 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 418 #else 419 NULL, NULL, NULL, NULL, 420 #endif 421 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 422 423 for (;i <= size; i+= KMEM_ZBASE) 424 kmemsize[i >> KMEM_ZSHIFT] = indx; 425 426 } 427 } 428 429 void 430 malloc_init(data) 431 void *data; 432 { 433 struct malloc_type *type = (struct malloc_type *)data; 434 435 mtx_lock(&malloc_mtx); 436 if (type->ks_magic != M_MAGIC) 437 panic("malloc type lacks magic"); 438 439 if (cnt.v_page_count == 0) 440 panic("malloc_init not allowed before vm init"); 441 442 if (type->ks_next != NULL) 443 return; 444 445 type->ks_next = kmemstatistics; 446 kmemstatistics = type; 447 mtx_init(&type->ks_mtx, type->ks_shortdesc, "Malloc Stats", MTX_DEF); 448 mtx_unlock(&malloc_mtx); 449 } 450 451 void 452 malloc_uninit(data) 453 void *data; 454 { 455 struct malloc_type *type = (struct malloc_type *)data; 456 struct malloc_type *t; 457 458 mtx_lock(&malloc_mtx); 459 mtx_lock(&type->ks_mtx); 460 if (type->ks_magic != M_MAGIC) 461 panic("malloc type lacks magic"); 462 463 if (cnt.v_page_count == 0) 464 panic("malloc_uninit not allowed before vm init"); 465 466 if (type == kmemstatistics) 467 kmemstatistics = type->ks_next; 468 else { 469 for (t = kmemstatistics; t->ks_next != NULL; t = t->ks_next) { 470 if (t->ks_next == type) { 471 t->ks_next = type->ks_next; 472 break; 473 } 474 } 475 } 476 type->ks_next = NULL; 477 mtx_destroy(&type->ks_mtx); 478 mtx_unlock(&malloc_mtx); 479 } 480 481 static int 482 sysctl_kern_malloc(SYSCTL_HANDLER_ARGS) 483 { 484 struct malloc_type *type; 485 int linesize = 128; 486 int curline; 487 int bufsize; 488 int first; 489 int error; 490 char *buf; 491 char *p; 492 int cnt; 493 int len; 494 int i; 495 496 cnt = 0; 497 498 mtx_lock(&malloc_mtx); 499 for (type = kmemstatistics; type != NULL; type = type->ks_next) 500 cnt++; 501 502 mtx_unlock(&malloc_mtx); 503 bufsize = linesize * (cnt + 1); 504 p = buf = (char *)malloc(bufsize, M_TEMP, M_ZERO); 505 mtx_lock(&malloc_mtx); 506 507 len = snprintf(p, linesize, 508 "\n Type InUse MemUse HighUse Requests Size(s)\n"); 509 p += len; 510 511 for (type = kmemstatistics; cnt != 0 && type != NULL; 512 type = type->ks_next, cnt--) { 513 if (type->ks_calls == 0) 514 continue; 515 516 curline = linesize - 2; /* Leave room for the \n */ 517 len = snprintf(p, curline, "%13s%6lu%6luK%7luK%9llu", 518 type->ks_shortdesc, 519 type->ks_inuse, 520 (type->ks_memuse + 1023) / 1024, 521 (type->ks_maxused + 1023) / 1024, 522 (long long unsigned)type->ks_calls); 523 curline -= len; 524 p += len; 525 526 first = 1; 527 for (i = 0; i < sizeof(kmemzones) / sizeof(kmemzones[0]) - 1; 528 i++) { 529 if (type->ks_size & (1 << i)) { 530 if (first) 531 len = snprintf(p, curline, " "); 532 else 533 len = snprintf(p, curline, ","); 534 curline -= len; 535 p += len; 536 537 len = snprintf(p, curline, 538 "%s", kmemzones[i].kz_name); 539 curline -= len; 540 p += len; 541 542 first = 0; 543 } 544 } 545 546 len = snprintf(p, 2, "\n"); 547 p += len; 548 } 549 550 mtx_unlock(&malloc_mtx); 551 error = SYSCTL_OUT(req, buf, p - buf); 552 553 free(buf, M_TEMP); 554 return (error); 555 } 556 557 SYSCTL_OID(_kern, OID_AUTO, malloc, CTLTYPE_STRING|CTLFLAG_RD, 558 NULL, 0, sysctl_kern_malloc, "A", "Malloc Stats"); 559 560 #ifdef MALLOC_PROFILE 561 562 static int 563 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 564 { 565 int linesize = 64; 566 uint64_t count; 567 uint64_t waste; 568 uint64_t mem; 569 int bufsize; 570 int error; 571 char *buf; 572 int rsize; 573 int size; 574 char *p; 575 int len; 576 int i; 577 578 bufsize = linesize * (KMEM_ZSIZE + 1); 579 bufsize += 128; /* For the stats line */ 580 bufsize += 128; /* For the banner line */ 581 waste = 0; 582 mem = 0; 583 584 p = buf = (char *)malloc(bufsize, M_TEMP, M_ZERO); 585 len = snprintf(p, bufsize, 586 "\n Size Requests Real Size\n"); 587 bufsize -= len; 588 p += len; 589 590 for (i = 0; i < KMEM_ZSIZE; i++) { 591 size = i << KMEM_ZSHIFT; 592 rsize = kmemzones[kmemsize[i]].kz_size; 593 count = (long long unsigned)krequests[i]; 594 595 len = snprintf(p, bufsize, "%6d%28llu%11d\n", 596 size, (unsigned long long)count, rsize); 597 bufsize -= len; 598 p += len; 599 600 if ((rsize * count) > (size * count)) 601 waste += (rsize * count) - (size * count); 602 mem += (rsize * count); 603 } 604 605 len = snprintf(p, bufsize, 606 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 607 (unsigned long long)mem, (unsigned long long)waste); 608 p += len; 609 610 error = SYSCTL_OUT(req, buf, p - buf); 611 612 free(buf, M_TEMP); 613 return (error); 614 } 615 616 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 617 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 618 #endif /* MALLOC_PROFILE */ 619