1 /*- 2 * Copyright (c) 1987, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2005-2009 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94 32 */ 33 34 /* 35 * Kernel malloc(9) implementation -- general purpose kernel memory allocator 36 * based on memory types. Back end is implemented using the UMA(9) zone 37 * allocator. A set of fixed-size buckets are used for smaller allocations, 38 * and a special UMA allocation interface is used for larger allocations. 39 * Callers declare memory types, and statistics are maintained independently 40 * for each memory type. Statistics are maintained per-CPU for performance 41 * reasons. See malloc(9) and comments in malloc.h for a detailed 42 * description. 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include "opt_ddb.h" 49 #include "opt_kdtrace.h" 50 #include "opt_vm.h" 51 52 #include <sys/param.h> 53 #include <sys/systm.h> 54 #include <sys/kdb.h> 55 #include <sys/kernel.h> 56 #include <sys/lock.h> 57 #include <sys/malloc.h> 58 #include <sys/mbuf.h> 59 #include <sys/mutex.h> 60 #include <sys/vmmeter.h> 61 #include <sys/proc.h> 62 #include <sys/sbuf.h> 63 #include <sys/sysctl.h> 64 #include <sys/time.h> 65 66 #include <vm/vm.h> 67 #include <vm/pmap.h> 68 #include <vm/vm_param.h> 69 #include <vm/vm_kern.h> 70 #include <vm/vm_extern.h> 71 #include <vm/vm_map.h> 72 #include <vm/vm_page.h> 73 #include <vm/uma.h> 74 #include <vm/uma_int.h> 75 #include <vm/uma_dbg.h> 76 77 #ifdef DEBUG_MEMGUARD 78 #include <vm/memguard.h> 79 #endif 80 #ifdef DEBUG_REDZONE 81 #include <vm/redzone.h> 82 #endif 83 84 #if defined(INVARIANTS) && defined(__i386__) 85 #include <machine/cpu.h> 86 #endif 87 88 #include <ddb/ddb.h> 89 90 #ifdef KDTRACE_HOOKS 91 #include <sys/dtrace_bsd.h> 92 93 dtrace_malloc_probe_func_t dtrace_malloc_probe; 94 #endif 95 96 /* 97 * When realloc() is called, if the new size is sufficiently smaller than 98 * the old size, realloc() will allocate a new, smaller block to avoid 99 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 100 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 101 */ 102 #ifndef REALLOC_FRACTION 103 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 104 #endif 105 106 /* 107 * Centrally define some common malloc types. 108 */ 109 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 110 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 111 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 112 113 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); 114 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); 115 116 static void kmeminit(void *); 117 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL); 118 119 static MALLOC_DEFINE(M_FREE, "free", "should be on free list"); 120 121 static struct malloc_type *kmemstatistics; 122 static vm_offset_t kmembase; 123 static vm_offset_t kmemlimit; 124 static int kmemcount; 125 126 #define KMEM_ZSHIFT 4 127 #define KMEM_ZBASE 16 128 #define KMEM_ZMASK (KMEM_ZBASE - 1) 129 130 #define KMEM_ZMAX PAGE_SIZE 131 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 132 static uint8_t kmemsize[KMEM_ZSIZE + 1]; 133 134 #ifndef MALLOC_DEBUG_MAXZONES 135 #define MALLOC_DEBUG_MAXZONES 1 136 #endif 137 static int numzones = MALLOC_DEBUG_MAXZONES; 138 139 /* 140 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 141 * of various sizes. 142 * 143 * XXX: The comment here used to read "These won't be powers of two for 144 * long." It's possible that a significant amount of wasted memory could be 145 * recovered by tuning the sizes of these buckets. 146 */ 147 struct { 148 int kz_size; 149 char *kz_name; 150 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 151 } kmemzones[] = { 152 {16, "16", }, 153 {32, "32", }, 154 {64, "64", }, 155 {128, "128", }, 156 {256, "256", }, 157 {512, "512", }, 158 {1024, "1024", }, 159 {2048, "2048", }, 160 {4096, "4096", }, 161 #if PAGE_SIZE > 4096 162 {8192, "8192", }, 163 #if PAGE_SIZE > 8192 164 {16384, "16384", }, 165 #if PAGE_SIZE > 16384 166 {32768, "32768", }, 167 #if PAGE_SIZE > 32768 168 {65536, "65536", }, 169 #if PAGE_SIZE > 65536 170 #error "Unsupported PAGE_SIZE" 171 #endif /* 65536 */ 172 #endif /* 32768 */ 173 #endif /* 16384 */ 174 #endif /* 8192 */ 175 #endif /* 4096 */ 176 {0, NULL}, 177 }; 178 179 /* 180 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 181 * types are described by a data structure passed by the declaring code, but 182 * the malloc(9) implementation has its own data structure describing the 183 * type and statistics. This permits the malloc(9)-internal data structures 184 * to be modified without breaking binary-compiled kernel modules that 185 * declare malloc types. 186 */ 187 static uma_zone_t mt_zone; 188 189 u_long vm_kmem_size; 190 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 191 "Size of kernel memory"); 192 193 static u_long vm_kmem_size_min; 194 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 195 "Minimum size of kernel memory"); 196 197 static u_long vm_kmem_size_max; 198 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 199 "Maximum size of kernel memory"); 200 201 static u_int vm_kmem_size_scale; 202 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 203 "Scale factor for kernel memory size"); 204 205 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 206 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 207 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 208 sysctl_kmem_map_size, "LU", "Current kmem_map allocation size"); 209 210 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 211 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 212 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 213 sysctl_kmem_map_free, "LU", "Largest contiguous free range in kmem_map"); 214 215 /* 216 * The malloc_mtx protects the kmemstatistics linked list. 217 */ 218 struct mtx malloc_mtx; 219 220 #ifdef MALLOC_PROFILE 221 uint64_t krequests[KMEM_ZSIZE + 1]; 222 223 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 224 #endif 225 226 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 227 228 /* 229 * time_uptime of the last malloc(9) failure (induced or real). 230 */ 231 static time_t t_malloc_fail; 232 233 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 234 SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 235 "Kernel malloc debugging options"); 236 #endif 237 238 /* 239 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 240 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 241 */ 242 #ifdef MALLOC_MAKE_FAILURES 243 static int malloc_failure_rate; 244 static int malloc_nowait_count; 245 static int malloc_failure_count; 246 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW, 247 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 248 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate); 249 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 250 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 251 #endif 252 253 static int 254 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 255 { 256 u_long size; 257 258 size = kmem_map->size; 259 return (sysctl_handle_long(oidp, &size, 0, req)); 260 } 261 262 static int 263 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 264 { 265 u_long size; 266 267 vm_map_lock_read(kmem_map); 268 size = kmem_map->root != NULL ? 269 kmem_map->root->max_free : kmem_map->size; 270 vm_map_unlock_read(kmem_map); 271 return (sysctl_handle_long(oidp, &size, 0, req)); 272 } 273 274 /* 275 * malloc(9) uma zone separation -- sub-page buffer overruns in one 276 * malloc type will affect only a subset of other malloc types. 277 */ 278 #if MALLOC_DEBUG_MAXZONES > 1 279 static void 280 tunable_set_numzones(void) 281 { 282 283 TUNABLE_INT_FETCH("debug.malloc.numzones", 284 &numzones); 285 286 /* Sanity check the number of malloc uma zones. */ 287 if (numzones <= 0) 288 numzones = 1; 289 if (numzones > MALLOC_DEBUG_MAXZONES) 290 numzones = MALLOC_DEBUG_MAXZONES; 291 } 292 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 293 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN, 294 &numzones, 0, "Number of malloc uma subzones"); 295 296 /* 297 * Any number that changes regularly is an okay choice for the 298 * offset. Build numbers are pretty good of you have them. 299 */ 300 static u_int zone_offset = __FreeBSD_version; 301 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 302 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 303 &zone_offset, 0, "Separate malloc types by examining the " 304 "Nth character in the malloc type short description."); 305 306 static u_int 307 mtp_get_subzone(const char *desc) 308 { 309 size_t len; 310 u_int val; 311 312 if (desc == NULL || (len = strlen(desc)) == 0) 313 return (0); 314 val = desc[zone_offset % len]; 315 return (val % numzones); 316 } 317 #elif MALLOC_DEBUG_MAXZONES == 0 318 #error "MALLOC_DEBUG_MAXZONES must be positive." 319 #else 320 static inline u_int 321 mtp_get_subzone(const char *desc) 322 { 323 324 return (0); 325 } 326 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 327 328 int 329 malloc_last_fail(void) 330 { 331 332 return (time_uptime - t_malloc_fail); 333 } 334 335 /* 336 * An allocation has succeeded -- update malloc type statistics for the 337 * amount of bucket size. Occurs within a critical section so that the 338 * thread isn't preempted and doesn't migrate while updating per-PCU 339 * statistics. 340 */ 341 static void 342 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 343 int zindx) 344 { 345 struct malloc_type_internal *mtip; 346 struct malloc_type_stats *mtsp; 347 348 critical_enter(); 349 mtip = mtp->ks_handle; 350 mtsp = &mtip->mti_stats[curcpu]; 351 if (size > 0) { 352 mtsp->mts_memalloced += size; 353 mtsp->mts_numallocs++; 354 } 355 if (zindx != -1) 356 mtsp->mts_size |= 1 << zindx; 357 358 #ifdef KDTRACE_HOOKS 359 if (dtrace_malloc_probe != NULL) { 360 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 361 if (probe_id != 0) 362 (dtrace_malloc_probe)(probe_id, 363 (uintptr_t) mtp, (uintptr_t) mtip, 364 (uintptr_t) mtsp, size, zindx); 365 } 366 #endif 367 368 critical_exit(); 369 } 370 371 void 372 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 373 { 374 375 if (size > 0) 376 malloc_type_zone_allocated(mtp, size, -1); 377 } 378 379 /* 380 * A free operation has occurred -- update malloc type statistics for the 381 * amount of the bucket size. Occurs within a critical section so that the 382 * thread isn't preempted and doesn't migrate while updating per-CPU 383 * statistics. 384 */ 385 void 386 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 387 { 388 struct malloc_type_internal *mtip; 389 struct malloc_type_stats *mtsp; 390 391 critical_enter(); 392 mtip = mtp->ks_handle; 393 mtsp = &mtip->mti_stats[curcpu]; 394 mtsp->mts_memfreed += size; 395 mtsp->mts_numfrees++; 396 397 #ifdef KDTRACE_HOOKS 398 if (dtrace_malloc_probe != NULL) { 399 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 400 if (probe_id != 0) 401 (dtrace_malloc_probe)(probe_id, 402 (uintptr_t) mtp, (uintptr_t) mtip, 403 (uintptr_t) mtsp, size, 0); 404 } 405 #endif 406 407 critical_exit(); 408 } 409 410 /* 411 * malloc: 412 * 413 * Allocate a block of memory. 414 * 415 * If M_NOWAIT is set, this routine will not block and return NULL if 416 * the allocation fails. 417 */ 418 void * 419 malloc(unsigned long size, struct malloc_type *mtp, int flags) 420 { 421 int indx; 422 struct malloc_type_internal *mtip; 423 caddr_t va; 424 uma_zone_t zone; 425 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE) 426 unsigned long osize = size; 427 #endif 428 429 #ifdef INVARIANTS 430 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 431 /* 432 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 433 */ 434 indx = flags & (M_WAITOK | M_NOWAIT); 435 if (indx != M_NOWAIT && indx != M_WAITOK) { 436 static struct timeval lasterr; 437 static int curerr, once; 438 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 439 printf("Bad malloc flags: %x\n", indx); 440 kdb_backtrace(); 441 flags |= M_WAITOK; 442 once++; 443 } 444 } 445 #endif 446 #ifdef MALLOC_MAKE_FAILURES 447 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 448 atomic_add_int(&malloc_nowait_count, 1); 449 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 450 atomic_add_int(&malloc_failure_count, 1); 451 t_malloc_fail = time_uptime; 452 return (NULL); 453 } 454 } 455 #endif 456 if (flags & M_WAITOK) 457 KASSERT(curthread->td_intr_nesting_level == 0, 458 ("malloc(M_WAITOK) in interrupt context")); 459 460 #ifdef DEBUG_MEMGUARD 461 if (memguard_cmp(mtp, size)) { 462 va = memguard_alloc(size, flags); 463 if (va != NULL) 464 return (va); 465 /* This is unfortunate but should not be fatal. */ 466 } 467 #endif 468 469 #ifdef DEBUG_REDZONE 470 size = redzone_size_ntor(size); 471 #endif 472 473 if (size <= KMEM_ZMAX) { 474 mtip = mtp->ks_handle; 475 if (size & KMEM_ZMASK) 476 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 477 indx = kmemsize[size >> KMEM_ZSHIFT]; 478 KASSERT(mtip->mti_zone < numzones, 479 ("mti_zone %u out of range %d", 480 mtip->mti_zone, numzones)); 481 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 482 #ifdef MALLOC_PROFILE 483 krequests[size >> KMEM_ZSHIFT]++; 484 #endif 485 va = uma_zalloc(zone, flags); 486 if (va != NULL) 487 size = zone->uz_size; 488 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 489 } else { 490 size = roundup(size, PAGE_SIZE); 491 zone = NULL; 492 va = uma_large_malloc(size, flags); 493 malloc_type_allocated(mtp, va == NULL ? 0 : size); 494 } 495 if (flags & M_WAITOK) 496 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 497 else if (va == NULL) 498 t_malloc_fail = time_uptime; 499 #ifdef DIAGNOSTIC 500 if (va != NULL && !(flags & M_ZERO)) { 501 memset(va, 0x70, osize); 502 } 503 #endif 504 #ifdef DEBUG_REDZONE 505 if (va != NULL) 506 va = redzone_setup(va, osize); 507 #endif 508 return ((void *) va); 509 } 510 511 /* 512 * free: 513 * 514 * Free a block of memory allocated by malloc. 515 * 516 * This routine may not block. 517 */ 518 void 519 free(void *addr, struct malloc_type *mtp) 520 { 521 uma_slab_t slab; 522 u_long size; 523 524 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); 525 526 /* free(NULL, ...) does nothing */ 527 if (addr == NULL) 528 return; 529 530 #ifdef DEBUG_MEMGUARD 531 if (is_memguard_addr(addr)) { 532 memguard_free(addr); 533 return; 534 } 535 #endif 536 537 #ifdef DEBUG_REDZONE 538 redzone_check(addr); 539 addr = redzone_addr_ntor(addr); 540 #endif 541 542 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 543 544 if (slab == NULL) 545 panic("free: address %p(%p) has not been allocated.\n", 546 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 547 548 549 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 550 #ifdef INVARIANTS 551 struct malloc_type **mtpp = addr; 552 #endif 553 size = slab->us_keg->uk_size; 554 #ifdef INVARIANTS 555 /* 556 * Cache a pointer to the malloc_type that most recently freed 557 * this memory here. This way we know who is most likely to 558 * have stepped on it later. 559 * 560 * This code assumes that size is a multiple of 8 bytes for 561 * 64 bit machines 562 */ 563 mtpp = (struct malloc_type **) 564 ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 565 mtpp += (size - sizeof(struct malloc_type *)) / 566 sizeof(struct malloc_type *); 567 *mtpp = mtp; 568 #endif 569 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 570 } else { 571 size = slab->us_size; 572 uma_large_free(slab); 573 } 574 malloc_type_freed(mtp, size); 575 } 576 577 /* 578 * realloc: change the size of a memory block 579 */ 580 void * 581 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 582 { 583 uma_slab_t slab; 584 unsigned long alloc; 585 void *newaddr; 586 587 KASSERT(mtp->ks_magic == M_MAGIC, 588 ("realloc: bad malloc type magic")); 589 590 /* realloc(NULL, ...) is equivalent to malloc(...) */ 591 if (addr == NULL) 592 return (malloc(size, mtp, flags)); 593 594 /* 595 * XXX: Should report free of old memory and alloc of new memory to 596 * per-CPU stats. 597 */ 598 599 #ifdef DEBUG_MEMGUARD 600 if (is_memguard_addr(addr)) 601 return (memguard_realloc(addr, size, mtp, flags)); 602 #endif 603 604 #ifdef DEBUG_REDZONE 605 slab = NULL; 606 alloc = redzone_get_size(addr); 607 #else 608 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 609 610 /* Sanity check */ 611 KASSERT(slab != NULL, 612 ("realloc: address %p out of range", (void *)addr)); 613 614 /* Get the size of the original block */ 615 if (!(slab->us_flags & UMA_SLAB_MALLOC)) 616 alloc = slab->us_keg->uk_size; 617 else 618 alloc = slab->us_size; 619 620 /* Reuse the original block if appropriate */ 621 if (size <= alloc 622 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 623 return (addr); 624 #endif /* !DEBUG_REDZONE */ 625 626 /* Allocate a new, bigger (or smaller) block */ 627 if ((newaddr = malloc(size, mtp, flags)) == NULL) 628 return (NULL); 629 630 /* Copy over original contents */ 631 bcopy(addr, newaddr, min(size, alloc)); 632 free(addr, mtp); 633 return (newaddr); 634 } 635 636 /* 637 * reallocf: same as realloc() but free memory on failure. 638 */ 639 void * 640 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 641 { 642 void *mem; 643 644 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 645 free(addr, mtp); 646 return (mem); 647 } 648 649 /* 650 * Initialize the kernel memory allocator 651 */ 652 /* ARGSUSED*/ 653 static void 654 kmeminit(void *dummy) 655 { 656 uint8_t indx; 657 u_long mem_size, tmp; 658 int i; 659 660 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 661 662 /* 663 * Try to auto-tune the kernel memory size, so that it is 664 * more applicable for a wider range of machine sizes. 665 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while 666 * a VM_KMEM_SIZE of 12MB is a fair compromise. The 667 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space 668 * available, and on an X86 with a total KVA space of 256MB, 669 * try to keep VM_KMEM_SIZE_MAX at 80MB or below. 670 * 671 * Note that the kmem_map is also used by the zone allocator, 672 * so make sure that there is enough space. 673 */ 674 vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE; 675 mem_size = cnt.v_page_count; 676 677 #if defined(VM_KMEM_SIZE_SCALE) 678 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 679 #endif 680 TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale); 681 if (vm_kmem_size_scale > 0 && 682 (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE)) 683 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 684 685 #if defined(VM_KMEM_SIZE_MIN) 686 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 687 #endif 688 TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min); 689 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) { 690 vm_kmem_size = vm_kmem_size_min; 691 } 692 693 #if defined(VM_KMEM_SIZE_MAX) 694 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 695 #endif 696 TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max); 697 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 698 vm_kmem_size = vm_kmem_size_max; 699 700 /* Allow final override from the kernel environment */ 701 TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size); 702 703 /* 704 * Limit kmem virtual size to twice the physical memory. 705 * This allows for kmem map sparseness, but limits the size 706 * to something sane. Be careful to not overflow the 32bit 707 * ints while doing the check. 708 */ 709 if (((vm_kmem_size / 2) / PAGE_SIZE) > cnt.v_page_count) 710 vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE; 711 712 #ifdef DEBUG_MEMGUARD 713 tmp = memguard_fudge(vm_kmem_size, vm_kmem_size_max); 714 #else 715 tmp = vm_kmem_size; 716 #endif 717 kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit, 718 tmp, TRUE); 719 kmem_map->system_map = 1; 720 721 #ifdef DEBUG_MEMGUARD 722 /* 723 * Initialize MemGuard if support compiled in. MemGuard is a 724 * replacement allocator used for detecting tamper-after-free 725 * scenarios as they occur. It is only used for debugging. 726 */ 727 memguard_init(kmem_map); 728 #endif 729 730 uma_startup2(); 731 732 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 733 #ifdef INVARIANTS 734 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 735 #else 736 NULL, NULL, NULL, NULL, 737 #endif 738 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 739 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 740 int size = kmemzones[indx].kz_size; 741 char *name = kmemzones[indx].kz_name; 742 int subzone; 743 744 for (subzone = 0; subzone < numzones; subzone++) { 745 kmemzones[indx].kz_zone[subzone] = 746 uma_zcreate(name, size, 747 #ifdef INVARIANTS 748 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 749 #else 750 NULL, NULL, NULL, NULL, 751 #endif 752 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 753 } 754 for (;i <= size; i+= KMEM_ZBASE) 755 kmemsize[i >> KMEM_ZSHIFT] = indx; 756 757 } 758 } 759 760 void 761 malloc_init(void *data) 762 { 763 struct malloc_type_internal *mtip; 764 struct malloc_type *mtp; 765 766 KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init")); 767 768 mtp = data; 769 if (mtp->ks_magic != M_MAGIC) 770 panic("malloc_init: bad malloc type magic"); 771 772 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 773 mtp->ks_handle = mtip; 774 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); 775 776 mtx_lock(&malloc_mtx); 777 mtp->ks_next = kmemstatistics; 778 kmemstatistics = mtp; 779 kmemcount++; 780 mtx_unlock(&malloc_mtx); 781 } 782 783 void 784 malloc_uninit(void *data) 785 { 786 struct malloc_type_internal *mtip; 787 struct malloc_type_stats *mtsp; 788 struct malloc_type *mtp, *temp; 789 uma_slab_t slab; 790 long temp_allocs, temp_bytes; 791 int i; 792 793 mtp = data; 794 KASSERT(mtp->ks_magic == M_MAGIC, 795 ("malloc_uninit: bad malloc type magic")); 796 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 797 798 mtx_lock(&malloc_mtx); 799 mtip = mtp->ks_handle; 800 mtp->ks_handle = NULL; 801 if (mtp != kmemstatistics) { 802 for (temp = kmemstatistics; temp != NULL; 803 temp = temp->ks_next) { 804 if (temp->ks_next == mtp) { 805 temp->ks_next = mtp->ks_next; 806 break; 807 } 808 } 809 KASSERT(temp, 810 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 811 } else 812 kmemstatistics = mtp->ks_next; 813 kmemcount--; 814 mtx_unlock(&malloc_mtx); 815 816 /* 817 * Look for memory leaks. 818 */ 819 temp_allocs = temp_bytes = 0; 820 for (i = 0; i < MAXCPU; i++) { 821 mtsp = &mtip->mti_stats[i]; 822 temp_allocs += mtsp->mts_numallocs; 823 temp_allocs -= mtsp->mts_numfrees; 824 temp_bytes += mtsp->mts_memalloced; 825 temp_bytes -= mtsp->mts_memfreed; 826 } 827 if (temp_allocs > 0 || temp_bytes > 0) { 828 printf("Warning: memory type %s leaked memory on destroy " 829 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 830 temp_allocs, temp_bytes); 831 } 832 833 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 834 uma_zfree_arg(mt_zone, mtip, slab); 835 } 836 837 struct malloc_type * 838 malloc_desc2type(const char *desc) 839 { 840 struct malloc_type *mtp; 841 842 mtx_assert(&malloc_mtx, MA_OWNED); 843 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 844 if (strcmp(mtp->ks_shortdesc, desc) == 0) 845 return (mtp); 846 } 847 return (NULL); 848 } 849 850 static int 851 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 852 { 853 struct malloc_type_stream_header mtsh; 854 struct malloc_type_internal *mtip; 855 struct malloc_type_header mth; 856 struct malloc_type *mtp; 857 int error, i; 858 struct sbuf sbuf; 859 860 error = sysctl_wire_old_buffer(req, 0); 861 if (error != 0) 862 return (error); 863 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 864 mtx_lock(&malloc_mtx); 865 866 /* 867 * Insert stream header. 868 */ 869 bzero(&mtsh, sizeof(mtsh)); 870 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 871 mtsh.mtsh_maxcpus = MAXCPU; 872 mtsh.mtsh_count = kmemcount; 873 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 874 875 /* 876 * Insert alternating sequence of type headers and type statistics. 877 */ 878 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 879 mtip = (struct malloc_type_internal *)mtp->ks_handle; 880 881 /* 882 * Insert type header. 883 */ 884 bzero(&mth, sizeof(mth)); 885 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 886 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 887 888 /* 889 * Insert type statistics for each CPU. 890 */ 891 for (i = 0; i < MAXCPU; i++) { 892 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], 893 sizeof(mtip->mti_stats[i])); 894 } 895 } 896 mtx_unlock(&malloc_mtx); 897 error = sbuf_finish(&sbuf); 898 sbuf_delete(&sbuf); 899 return (error); 900 } 901 902 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 903 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 904 "Return malloc types"); 905 906 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 907 "Count of kernel malloc types"); 908 909 void 910 malloc_type_list(malloc_type_list_func_t *func, void *arg) 911 { 912 struct malloc_type *mtp, **bufmtp; 913 int count, i; 914 size_t buflen; 915 916 mtx_lock(&malloc_mtx); 917 restart: 918 mtx_assert(&malloc_mtx, MA_OWNED); 919 count = kmemcount; 920 mtx_unlock(&malloc_mtx); 921 922 buflen = sizeof(struct malloc_type *) * count; 923 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 924 925 mtx_lock(&malloc_mtx); 926 927 if (count < kmemcount) { 928 free(bufmtp, M_TEMP); 929 goto restart; 930 } 931 932 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 933 bufmtp[i] = mtp; 934 935 mtx_unlock(&malloc_mtx); 936 937 for (i = 0; i < count; i++) 938 (func)(bufmtp[i], arg); 939 940 free(bufmtp, M_TEMP); 941 } 942 943 #ifdef DDB 944 DB_SHOW_COMMAND(malloc, db_show_malloc) 945 { 946 struct malloc_type_internal *mtip; 947 struct malloc_type *mtp; 948 uint64_t allocs, frees; 949 uint64_t alloced, freed; 950 int i; 951 952 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 953 "Requests"); 954 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 955 mtip = (struct malloc_type_internal *)mtp->ks_handle; 956 allocs = 0; 957 frees = 0; 958 alloced = 0; 959 freed = 0; 960 for (i = 0; i < MAXCPU; i++) { 961 allocs += mtip->mti_stats[i].mts_numallocs; 962 frees += mtip->mti_stats[i].mts_numfrees; 963 alloced += mtip->mti_stats[i].mts_memalloced; 964 freed += mtip->mti_stats[i].mts_memfreed; 965 } 966 db_printf("%18s %12ju %12juK %12ju\n", 967 mtp->ks_shortdesc, allocs - frees, 968 (alloced - freed + 1023) / 1024, allocs); 969 } 970 } 971 972 #if MALLOC_DEBUG_MAXZONES > 1 973 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 974 { 975 struct malloc_type_internal *mtip; 976 struct malloc_type *mtp; 977 u_int subzone; 978 979 if (!have_addr) { 980 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 981 return; 982 } 983 mtp = (void *)addr; 984 if (mtp->ks_magic != M_MAGIC) { 985 db_printf("Magic %lx does not match expected %x\n", 986 mtp->ks_magic, M_MAGIC); 987 return; 988 } 989 990 mtip = mtp->ks_handle; 991 subzone = mtip->mti_zone; 992 993 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 994 mtip = mtp->ks_handle; 995 if (mtip->mti_zone != subzone) 996 continue; 997 db_printf("%s\n", mtp->ks_shortdesc); 998 } 999 } 1000 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1001 #endif /* DDB */ 1002 1003 #ifdef MALLOC_PROFILE 1004 1005 static int 1006 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1007 { 1008 struct sbuf sbuf; 1009 uint64_t count; 1010 uint64_t waste; 1011 uint64_t mem; 1012 int error; 1013 int rsize; 1014 int size; 1015 int i; 1016 1017 waste = 0; 1018 mem = 0; 1019 1020 error = sysctl_wire_old_buffer(req, 0); 1021 if (error != 0) 1022 return (error); 1023 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1024 sbuf_printf(&sbuf, 1025 "\n Size Requests Real Size\n"); 1026 for (i = 0; i < KMEM_ZSIZE; i++) { 1027 size = i << KMEM_ZSHIFT; 1028 rsize = kmemzones[kmemsize[i]].kz_size; 1029 count = (long long unsigned)krequests[i]; 1030 1031 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1032 (unsigned long long)count, rsize); 1033 1034 if ((rsize * count) > (size * count)) 1035 waste += (rsize * count) - (size * count); 1036 mem += (rsize * count); 1037 } 1038 sbuf_printf(&sbuf, 1039 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1040 (unsigned long long)mem, (unsigned long long)waste); 1041 error = sbuf_finish(&sbuf); 1042 sbuf_delete(&sbuf); 1043 return (error); 1044 } 1045 1046 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 1047 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 1048 #endif /* MALLOC_PROFILE */ 1049