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_vm.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/kdb.h> 54 #include <sys/kernel.h> 55 #include <sys/lock.h> 56 #include <sys/malloc.h> 57 #include <sys/mutex.h> 58 #include <sys/vmmeter.h> 59 #include <sys/proc.h> 60 #include <sys/sbuf.h> 61 #include <sys/sysctl.h> 62 #include <sys/taskqueue.h> 63 #include <sys/time.h> 64 #include <sys/vmem.h> 65 66 #include <vm/vm.h> 67 #include <vm/pmap.h> 68 #include <vm/vm_pageout.h> 69 #include <vm/vm_param.h> 70 #include <vm/vm_kern.h> 71 #include <vm/vm_extern.h> 72 #include <vm/vm_map.h> 73 #include <vm/vm_page.h> 74 #include <vm/uma.h> 75 #include <vm/uma_int.h> 76 #include <vm/uma_dbg.h> 77 78 #ifdef DEBUG_MEMGUARD 79 #include <vm/memguard.h> 80 #endif 81 #ifdef DEBUG_REDZONE 82 #include <vm/redzone.h> 83 #endif 84 85 #if defined(INVARIANTS) && defined(__i386__) 86 #include <machine/cpu.h> 87 #endif 88 89 #include <ddb/ddb.h> 90 91 #ifdef KDTRACE_HOOKS 92 #include <sys/dtrace_bsd.h> 93 94 dtrace_malloc_probe_func_t dtrace_malloc_probe; 95 #endif 96 97 /* 98 * When realloc() is called, if the new size is sufficiently smaller than 99 * the old size, realloc() will allocate a new, smaller block to avoid 100 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 101 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 102 */ 103 #ifndef REALLOC_FRACTION 104 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 105 #endif 106 107 /* 108 * Centrally define some common malloc types. 109 */ 110 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 111 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 112 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 113 114 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); 115 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); 116 117 static struct malloc_type *kmemstatistics; 118 static int kmemcount; 119 120 #define KMEM_ZSHIFT 4 121 #define KMEM_ZBASE 16 122 #define KMEM_ZMASK (KMEM_ZBASE - 1) 123 124 #define KMEM_ZMAX 65536 125 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 126 static uint8_t kmemsize[KMEM_ZSIZE + 1]; 127 128 #ifndef MALLOC_DEBUG_MAXZONES 129 #define MALLOC_DEBUG_MAXZONES 1 130 #endif 131 static int numzones = MALLOC_DEBUG_MAXZONES; 132 133 /* 134 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 135 * of various sizes. 136 * 137 * XXX: The comment here used to read "These won't be powers of two for 138 * long." It's possible that a significant amount of wasted memory could be 139 * recovered by tuning the sizes of these buckets. 140 */ 141 struct { 142 int kz_size; 143 char *kz_name; 144 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 145 } kmemzones[] = { 146 {16, "16", }, 147 {32, "32", }, 148 {64, "64", }, 149 {128, "128", }, 150 {256, "256", }, 151 {512, "512", }, 152 {1024, "1024", }, 153 {2048, "2048", }, 154 {4096, "4096", }, 155 {8192, "8192", }, 156 {16384, "16384", }, 157 {32768, "32768", }, 158 {65536, "65536", }, 159 {0, NULL}, 160 }; 161 162 /* 163 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 164 * types are described by a data structure passed by the declaring code, but 165 * the malloc(9) implementation has its own data structure describing the 166 * type and statistics. This permits the malloc(9)-internal data structures 167 * to be modified without breaking binary-compiled kernel modules that 168 * declare malloc types. 169 */ 170 static uma_zone_t mt_zone; 171 172 u_long vm_kmem_size; 173 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 174 "Size of kernel memory"); 175 176 static u_long kmem_zmax = KMEM_ZMAX; 177 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 178 "Maximum allocation size that malloc(9) would use UMA as backend"); 179 180 static u_long vm_kmem_size_min; 181 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 182 "Minimum size of kernel memory"); 183 184 static u_long vm_kmem_size_max; 185 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 186 "Maximum size of kernel memory"); 187 188 static u_int vm_kmem_size_scale; 189 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 190 "Scale factor for kernel memory size"); 191 192 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 193 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 194 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 195 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 196 197 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 198 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 199 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 200 sysctl_kmem_map_free, "LU", "Free space in kmem"); 201 202 /* 203 * The malloc_mtx protects the kmemstatistics linked list. 204 */ 205 struct mtx malloc_mtx; 206 207 #ifdef MALLOC_PROFILE 208 uint64_t krequests[KMEM_ZSIZE + 1]; 209 210 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 211 #endif 212 213 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 214 215 /* 216 * time_uptime of the last malloc(9) failure (induced or real). 217 */ 218 static time_t t_malloc_fail; 219 220 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 221 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 222 "Kernel malloc debugging options"); 223 #endif 224 225 /* 226 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 227 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 228 */ 229 #ifdef MALLOC_MAKE_FAILURES 230 static int malloc_failure_rate; 231 static int malloc_nowait_count; 232 static int malloc_failure_count; 233 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, 234 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 235 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 236 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 237 #endif 238 239 static int 240 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 241 { 242 u_long size; 243 244 size = vmem_size(kmem_arena, VMEM_ALLOC); 245 return (sysctl_handle_long(oidp, &size, 0, req)); 246 } 247 248 static int 249 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 250 { 251 u_long size; 252 253 size = vmem_size(kmem_arena, VMEM_FREE); 254 return (sysctl_handle_long(oidp, &size, 0, req)); 255 } 256 257 /* 258 * malloc(9) uma zone separation -- sub-page buffer overruns in one 259 * malloc type will affect only a subset of other malloc types. 260 */ 261 #if MALLOC_DEBUG_MAXZONES > 1 262 static void 263 tunable_set_numzones(void) 264 { 265 266 TUNABLE_INT_FETCH("debug.malloc.numzones", 267 &numzones); 268 269 /* Sanity check the number of malloc uma zones. */ 270 if (numzones <= 0) 271 numzones = 1; 272 if (numzones > MALLOC_DEBUG_MAXZONES) 273 numzones = MALLOC_DEBUG_MAXZONES; 274 } 275 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 276 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 277 &numzones, 0, "Number of malloc uma subzones"); 278 279 /* 280 * Any number that changes regularly is an okay choice for the 281 * offset. Build numbers are pretty good of you have them. 282 */ 283 static u_int zone_offset = __FreeBSD_version; 284 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 285 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 286 &zone_offset, 0, "Separate malloc types by examining the " 287 "Nth character in the malloc type short description."); 288 289 static u_int 290 mtp_get_subzone(const char *desc) 291 { 292 size_t len; 293 u_int val; 294 295 if (desc == NULL || (len = strlen(desc)) == 0) 296 return (0); 297 val = desc[zone_offset % len]; 298 return (val % numzones); 299 } 300 #elif MALLOC_DEBUG_MAXZONES == 0 301 #error "MALLOC_DEBUG_MAXZONES must be positive." 302 #else 303 static inline u_int 304 mtp_get_subzone(const char *desc) 305 { 306 307 return (0); 308 } 309 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 310 311 int 312 malloc_last_fail(void) 313 { 314 315 return (time_uptime - t_malloc_fail); 316 } 317 318 /* 319 * An allocation has succeeded -- update malloc type statistics for the 320 * amount of bucket size. Occurs within a critical section so that the 321 * thread isn't preempted and doesn't migrate while updating per-PCU 322 * statistics. 323 */ 324 static void 325 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 326 int zindx) 327 { 328 struct malloc_type_internal *mtip; 329 struct malloc_type_stats *mtsp; 330 331 critical_enter(); 332 mtip = mtp->ks_handle; 333 mtsp = &mtip->mti_stats[curcpu]; 334 if (size > 0) { 335 mtsp->mts_memalloced += size; 336 mtsp->mts_numallocs++; 337 } 338 if (zindx != -1) 339 mtsp->mts_size |= 1 << zindx; 340 341 #ifdef KDTRACE_HOOKS 342 if (dtrace_malloc_probe != NULL) { 343 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 344 if (probe_id != 0) 345 (dtrace_malloc_probe)(probe_id, 346 (uintptr_t) mtp, (uintptr_t) mtip, 347 (uintptr_t) mtsp, size, zindx); 348 } 349 #endif 350 351 critical_exit(); 352 } 353 354 void 355 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 356 { 357 358 if (size > 0) 359 malloc_type_zone_allocated(mtp, size, -1); 360 } 361 362 /* 363 * A free operation has occurred -- update malloc type statistics for the 364 * amount of the bucket size. Occurs within a critical section so that the 365 * thread isn't preempted and doesn't migrate while updating per-CPU 366 * statistics. 367 */ 368 void 369 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 370 { 371 struct malloc_type_internal *mtip; 372 struct malloc_type_stats *mtsp; 373 374 critical_enter(); 375 mtip = mtp->ks_handle; 376 mtsp = &mtip->mti_stats[curcpu]; 377 mtsp->mts_memfreed += size; 378 mtsp->mts_numfrees++; 379 380 #ifdef KDTRACE_HOOKS 381 if (dtrace_malloc_probe != NULL) { 382 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 383 if (probe_id != 0) 384 (dtrace_malloc_probe)(probe_id, 385 (uintptr_t) mtp, (uintptr_t) mtip, 386 (uintptr_t) mtsp, size, 0); 387 } 388 #endif 389 390 critical_exit(); 391 } 392 393 /* 394 * contigmalloc: 395 * 396 * Allocate a block of physically contiguous memory. 397 * 398 * If M_NOWAIT is set, this routine will not block and return NULL if 399 * the allocation fails. 400 */ 401 void * 402 contigmalloc(unsigned long size, struct malloc_type *type, int flags, 403 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 404 vm_paddr_t boundary) 405 { 406 void *ret; 407 408 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high, 409 alignment, boundary, VM_MEMATTR_DEFAULT); 410 if (ret != NULL) 411 malloc_type_allocated(type, round_page(size)); 412 return (ret); 413 } 414 415 /* 416 * contigfree: 417 * 418 * Free a block of memory allocated by contigmalloc. 419 * 420 * This routine may not block. 421 */ 422 void 423 contigfree(void *addr, unsigned long size, struct malloc_type *type) 424 { 425 426 kmem_free(kernel_arena, (vm_offset_t)addr, size); 427 malloc_type_freed(type, round_page(size)); 428 } 429 430 /* 431 * malloc: 432 * 433 * Allocate a block of memory. 434 * 435 * If M_NOWAIT is set, this routine will not block and return NULL if 436 * the allocation fails. 437 */ 438 void * 439 malloc(unsigned long size, struct malloc_type *mtp, int flags) 440 { 441 int indx; 442 struct malloc_type_internal *mtip; 443 caddr_t va; 444 uma_zone_t zone; 445 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE) 446 unsigned long osize = size; 447 #endif 448 449 #ifdef INVARIANTS 450 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 451 /* 452 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 453 */ 454 indx = flags & (M_WAITOK | M_NOWAIT); 455 if (indx != M_NOWAIT && indx != M_WAITOK) { 456 static struct timeval lasterr; 457 static int curerr, once; 458 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 459 printf("Bad malloc flags: %x\n", indx); 460 kdb_backtrace(); 461 flags |= M_WAITOK; 462 once++; 463 } 464 } 465 #endif 466 #ifdef MALLOC_MAKE_FAILURES 467 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 468 atomic_add_int(&malloc_nowait_count, 1); 469 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 470 atomic_add_int(&malloc_failure_count, 1); 471 t_malloc_fail = time_uptime; 472 return (NULL); 473 } 474 } 475 #endif 476 if (flags & M_WAITOK) 477 KASSERT(curthread->td_intr_nesting_level == 0, 478 ("malloc(M_WAITOK) in interrupt context")); 479 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 480 ("malloc: called with spinlock or critical section held")); 481 482 #ifdef DEBUG_MEMGUARD 483 if (memguard_cmp_mtp(mtp, size)) { 484 va = memguard_alloc(size, flags); 485 if (va != NULL) 486 return (va); 487 /* This is unfortunate but should not be fatal. */ 488 } 489 #endif 490 491 #ifdef DEBUG_REDZONE 492 size = redzone_size_ntor(size); 493 #endif 494 495 if (size <= kmem_zmax) { 496 mtip = mtp->ks_handle; 497 if (size & KMEM_ZMASK) 498 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 499 indx = kmemsize[size >> KMEM_ZSHIFT]; 500 KASSERT(mtip->mti_zone < numzones, 501 ("mti_zone %u out of range %d", 502 mtip->mti_zone, numzones)); 503 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 504 #ifdef MALLOC_PROFILE 505 krequests[size >> KMEM_ZSHIFT]++; 506 #endif 507 va = uma_zalloc(zone, flags); 508 if (va != NULL) 509 size = zone->uz_size; 510 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 511 } else { 512 size = roundup(size, PAGE_SIZE); 513 zone = NULL; 514 va = uma_large_malloc(size, flags); 515 malloc_type_allocated(mtp, va == NULL ? 0 : size); 516 } 517 if (flags & M_WAITOK) 518 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 519 else if (va == NULL) 520 t_malloc_fail = time_uptime; 521 #ifdef DIAGNOSTIC 522 if (va != NULL && !(flags & M_ZERO)) { 523 memset(va, 0x70, osize); 524 } 525 #endif 526 #ifdef DEBUG_REDZONE 527 if (va != NULL) 528 va = redzone_setup(va, osize); 529 #endif 530 return ((void *) va); 531 } 532 533 /* 534 * free: 535 * 536 * Free a block of memory allocated by malloc. 537 * 538 * This routine may not block. 539 */ 540 void 541 free(void *addr, struct malloc_type *mtp) 542 { 543 uma_slab_t slab; 544 u_long size; 545 546 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); 547 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 548 ("free: called with spinlock or critical section held")); 549 550 /* free(NULL, ...) does nothing */ 551 if (addr == NULL) 552 return; 553 554 #ifdef DEBUG_MEMGUARD 555 if (is_memguard_addr(addr)) { 556 memguard_free(addr); 557 return; 558 } 559 #endif 560 561 #ifdef DEBUG_REDZONE 562 redzone_check(addr); 563 addr = redzone_addr_ntor(addr); 564 #endif 565 566 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 567 568 if (slab == NULL) 569 panic("free: address %p(%p) has not been allocated.\n", 570 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 571 572 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 573 #ifdef INVARIANTS 574 struct malloc_type **mtpp = addr; 575 #endif 576 size = slab->us_keg->uk_size; 577 #ifdef INVARIANTS 578 /* 579 * Cache a pointer to the malloc_type that most recently freed 580 * this memory here. This way we know who is most likely to 581 * have stepped on it later. 582 * 583 * This code assumes that size is a multiple of 8 bytes for 584 * 64 bit machines 585 */ 586 mtpp = (struct malloc_type **) 587 ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 588 mtpp += (size - sizeof(struct malloc_type *)) / 589 sizeof(struct malloc_type *); 590 *mtpp = mtp; 591 #endif 592 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 593 } else { 594 size = slab->us_size; 595 uma_large_free(slab); 596 } 597 malloc_type_freed(mtp, size); 598 } 599 600 /* 601 * realloc: change the size of a memory block 602 */ 603 void * 604 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 605 { 606 uma_slab_t slab; 607 unsigned long alloc; 608 void *newaddr; 609 610 KASSERT(mtp->ks_magic == M_MAGIC, 611 ("realloc: bad malloc type magic")); 612 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 613 ("realloc: called with spinlock or critical section held")); 614 615 /* realloc(NULL, ...) is equivalent to malloc(...) */ 616 if (addr == NULL) 617 return (malloc(size, mtp, flags)); 618 619 /* 620 * XXX: Should report free of old memory and alloc of new memory to 621 * per-CPU stats. 622 */ 623 624 #ifdef DEBUG_MEMGUARD 625 if (is_memguard_addr(addr)) 626 return (memguard_realloc(addr, size, mtp, flags)); 627 #endif 628 629 #ifdef DEBUG_REDZONE 630 slab = NULL; 631 alloc = redzone_get_size(addr); 632 #else 633 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 634 635 /* Sanity check */ 636 KASSERT(slab != NULL, 637 ("realloc: address %p out of range", (void *)addr)); 638 639 /* Get the size of the original block */ 640 if (!(slab->us_flags & UMA_SLAB_MALLOC)) 641 alloc = slab->us_keg->uk_size; 642 else 643 alloc = slab->us_size; 644 645 /* Reuse the original block if appropriate */ 646 if (size <= alloc 647 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 648 return (addr); 649 #endif /* !DEBUG_REDZONE */ 650 651 /* Allocate a new, bigger (or smaller) block */ 652 if ((newaddr = malloc(size, mtp, flags)) == NULL) 653 return (NULL); 654 655 /* Copy over original contents */ 656 bcopy(addr, newaddr, min(size, alloc)); 657 free(addr, mtp); 658 return (newaddr); 659 } 660 661 /* 662 * reallocf: same as realloc() but free memory on failure. 663 */ 664 void * 665 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 666 { 667 void *mem; 668 669 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 670 free(addr, mtp); 671 return (mem); 672 } 673 674 /* 675 * Wake the uma reclamation pagedaemon thread when we exhaust KVA. It 676 * will call the lowmem handler and uma_reclaim() callbacks in a 677 * context that is safe. 678 */ 679 static void 680 kmem_reclaim(vmem_t *vm, int flags) 681 { 682 683 uma_reclaim_wakeup(); 684 pagedaemon_wakeup(); 685 } 686 687 #ifndef __sparc64__ 688 CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 689 #endif 690 691 /* 692 * Initialize the kernel memory (kmem) arena. 693 */ 694 void 695 kmeminit(void) 696 { 697 u_long mem_size; 698 u_long tmp; 699 700 #ifdef VM_KMEM_SIZE 701 if (vm_kmem_size == 0) 702 vm_kmem_size = VM_KMEM_SIZE; 703 #endif 704 #ifdef VM_KMEM_SIZE_MIN 705 if (vm_kmem_size_min == 0) 706 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 707 #endif 708 #ifdef VM_KMEM_SIZE_MAX 709 if (vm_kmem_size_max == 0) 710 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 711 #endif 712 /* 713 * Calculate the amount of kernel virtual address (KVA) space that is 714 * preallocated to the kmem arena. In order to support a wide range 715 * of machines, it is a function of the physical memory size, 716 * specifically, 717 * 718 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 719 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 720 * 721 * Every architecture must define an integral value for 722 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 723 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 724 * ceiling on this preallocation, are optional. Typically, 725 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 726 * a given architecture. 727 */ 728 mem_size = vm_cnt.v_page_count; 729 if (mem_size <= 32768) /* delphij XXX 128MB */ 730 kmem_zmax = PAGE_SIZE; 731 732 if (vm_kmem_size_scale < 1) 733 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 734 735 /* 736 * Check if we should use defaults for the "vm_kmem_size" 737 * variable: 738 */ 739 if (vm_kmem_size == 0) { 740 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 741 742 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 743 vm_kmem_size = vm_kmem_size_min; 744 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 745 vm_kmem_size = vm_kmem_size_max; 746 } 747 748 /* 749 * The amount of KVA space that is preallocated to the 750 * kmem arena can be set statically at compile-time or manually 751 * through the kernel environment. However, it is still limited to 752 * twice the physical memory size, which has been sufficient to handle 753 * the most severe cases of external fragmentation in the kmem arena. 754 */ 755 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 756 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 757 758 vm_kmem_size = round_page(vm_kmem_size); 759 #ifdef DEBUG_MEMGUARD 760 tmp = memguard_fudge(vm_kmem_size, kernel_map); 761 #else 762 tmp = vm_kmem_size; 763 #endif 764 vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE, 765 0, 0); 766 vmem_set_reclaim(kmem_arena, kmem_reclaim); 767 768 #ifdef DEBUG_MEMGUARD 769 /* 770 * Initialize MemGuard if support compiled in. MemGuard is a 771 * replacement allocator used for detecting tamper-after-free 772 * scenarios as they occur. It is only used for debugging. 773 */ 774 memguard_init(kmem_arena); 775 #endif 776 } 777 778 /* 779 * Initialize the kernel memory allocator 780 */ 781 /* ARGSUSED*/ 782 static void 783 mallocinit(void *dummy) 784 { 785 int i; 786 uint8_t indx; 787 788 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 789 790 kmeminit(); 791 792 uma_startup2(); 793 794 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 795 kmem_zmax = KMEM_ZMAX; 796 797 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 798 #ifdef INVARIANTS 799 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 800 #else 801 NULL, NULL, NULL, NULL, 802 #endif 803 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 804 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 805 int size = kmemzones[indx].kz_size; 806 char *name = kmemzones[indx].kz_name; 807 int subzone; 808 809 for (subzone = 0; subzone < numzones; subzone++) { 810 kmemzones[indx].kz_zone[subzone] = 811 uma_zcreate(name, size, 812 #ifdef INVARIANTS 813 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 814 #else 815 NULL, NULL, NULL, NULL, 816 #endif 817 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 818 } 819 for (;i <= size; i+= KMEM_ZBASE) 820 kmemsize[i >> KMEM_ZSHIFT] = indx; 821 822 } 823 } 824 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); 825 826 void 827 malloc_init(void *data) 828 { 829 struct malloc_type_internal *mtip; 830 struct malloc_type *mtp; 831 832 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init")); 833 834 mtp = data; 835 if (mtp->ks_magic != M_MAGIC) 836 panic("malloc_init: bad malloc type magic"); 837 838 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 839 mtp->ks_handle = mtip; 840 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); 841 842 mtx_lock(&malloc_mtx); 843 mtp->ks_next = kmemstatistics; 844 kmemstatistics = mtp; 845 kmemcount++; 846 mtx_unlock(&malloc_mtx); 847 } 848 849 void 850 malloc_uninit(void *data) 851 { 852 struct malloc_type_internal *mtip; 853 struct malloc_type_stats *mtsp; 854 struct malloc_type *mtp, *temp; 855 uma_slab_t slab; 856 long temp_allocs, temp_bytes; 857 int i; 858 859 mtp = data; 860 KASSERT(mtp->ks_magic == M_MAGIC, 861 ("malloc_uninit: bad malloc type magic")); 862 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 863 864 mtx_lock(&malloc_mtx); 865 mtip = mtp->ks_handle; 866 mtp->ks_handle = NULL; 867 if (mtp != kmemstatistics) { 868 for (temp = kmemstatistics; temp != NULL; 869 temp = temp->ks_next) { 870 if (temp->ks_next == mtp) { 871 temp->ks_next = mtp->ks_next; 872 break; 873 } 874 } 875 KASSERT(temp, 876 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 877 } else 878 kmemstatistics = mtp->ks_next; 879 kmemcount--; 880 mtx_unlock(&malloc_mtx); 881 882 /* 883 * Look for memory leaks. 884 */ 885 temp_allocs = temp_bytes = 0; 886 for (i = 0; i < MAXCPU; i++) { 887 mtsp = &mtip->mti_stats[i]; 888 temp_allocs += mtsp->mts_numallocs; 889 temp_allocs -= mtsp->mts_numfrees; 890 temp_bytes += mtsp->mts_memalloced; 891 temp_bytes -= mtsp->mts_memfreed; 892 } 893 if (temp_allocs > 0 || temp_bytes > 0) { 894 printf("Warning: memory type %s leaked memory on destroy " 895 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 896 temp_allocs, temp_bytes); 897 } 898 899 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 900 uma_zfree_arg(mt_zone, mtip, slab); 901 } 902 903 struct malloc_type * 904 malloc_desc2type(const char *desc) 905 { 906 struct malloc_type *mtp; 907 908 mtx_assert(&malloc_mtx, MA_OWNED); 909 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 910 if (strcmp(mtp->ks_shortdesc, desc) == 0) 911 return (mtp); 912 } 913 return (NULL); 914 } 915 916 static int 917 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 918 { 919 struct malloc_type_stream_header mtsh; 920 struct malloc_type_internal *mtip; 921 struct malloc_type_header mth; 922 struct malloc_type *mtp; 923 int error, i; 924 struct sbuf sbuf; 925 926 error = sysctl_wire_old_buffer(req, 0); 927 if (error != 0) 928 return (error); 929 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 930 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); 931 mtx_lock(&malloc_mtx); 932 933 /* 934 * Insert stream header. 935 */ 936 bzero(&mtsh, sizeof(mtsh)); 937 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 938 mtsh.mtsh_maxcpus = MAXCPU; 939 mtsh.mtsh_count = kmemcount; 940 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 941 942 /* 943 * Insert alternating sequence of type headers and type statistics. 944 */ 945 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 946 mtip = (struct malloc_type_internal *)mtp->ks_handle; 947 948 /* 949 * Insert type header. 950 */ 951 bzero(&mth, sizeof(mth)); 952 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 953 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 954 955 /* 956 * Insert type statistics for each CPU. 957 */ 958 for (i = 0; i < MAXCPU; i++) { 959 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], 960 sizeof(mtip->mti_stats[i])); 961 } 962 } 963 mtx_unlock(&malloc_mtx); 964 error = sbuf_finish(&sbuf); 965 sbuf_delete(&sbuf); 966 return (error); 967 } 968 969 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 970 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 971 "Return malloc types"); 972 973 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 974 "Count of kernel malloc types"); 975 976 void 977 malloc_type_list(malloc_type_list_func_t *func, void *arg) 978 { 979 struct malloc_type *mtp, **bufmtp; 980 int count, i; 981 size_t buflen; 982 983 mtx_lock(&malloc_mtx); 984 restart: 985 mtx_assert(&malloc_mtx, MA_OWNED); 986 count = kmemcount; 987 mtx_unlock(&malloc_mtx); 988 989 buflen = sizeof(struct malloc_type *) * count; 990 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 991 992 mtx_lock(&malloc_mtx); 993 994 if (count < kmemcount) { 995 free(bufmtp, M_TEMP); 996 goto restart; 997 } 998 999 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 1000 bufmtp[i] = mtp; 1001 1002 mtx_unlock(&malloc_mtx); 1003 1004 for (i = 0; i < count; i++) 1005 (func)(bufmtp[i], arg); 1006 1007 free(bufmtp, M_TEMP); 1008 } 1009 1010 #ifdef DDB 1011 DB_SHOW_COMMAND(malloc, db_show_malloc) 1012 { 1013 struct malloc_type_internal *mtip; 1014 struct malloc_type *mtp; 1015 uint64_t allocs, frees; 1016 uint64_t alloced, freed; 1017 int i; 1018 1019 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 1020 "Requests"); 1021 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1022 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1023 allocs = 0; 1024 frees = 0; 1025 alloced = 0; 1026 freed = 0; 1027 for (i = 0; i < MAXCPU; i++) { 1028 allocs += mtip->mti_stats[i].mts_numallocs; 1029 frees += mtip->mti_stats[i].mts_numfrees; 1030 alloced += mtip->mti_stats[i].mts_memalloced; 1031 freed += mtip->mti_stats[i].mts_memfreed; 1032 } 1033 db_printf("%18s %12ju %12juK %12ju\n", 1034 mtp->ks_shortdesc, allocs - frees, 1035 (alloced - freed + 1023) / 1024, allocs); 1036 if (db_pager_quit) 1037 break; 1038 } 1039 } 1040 1041 #if MALLOC_DEBUG_MAXZONES > 1 1042 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1043 { 1044 struct malloc_type_internal *mtip; 1045 struct malloc_type *mtp; 1046 u_int subzone; 1047 1048 if (!have_addr) { 1049 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1050 return; 1051 } 1052 mtp = (void *)addr; 1053 if (mtp->ks_magic != M_MAGIC) { 1054 db_printf("Magic %lx does not match expected %x\n", 1055 mtp->ks_magic, M_MAGIC); 1056 return; 1057 } 1058 1059 mtip = mtp->ks_handle; 1060 subzone = mtip->mti_zone; 1061 1062 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1063 mtip = mtp->ks_handle; 1064 if (mtip->mti_zone != subzone) 1065 continue; 1066 db_printf("%s\n", mtp->ks_shortdesc); 1067 if (db_pager_quit) 1068 break; 1069 } 1070 } 1071 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1072 #endif /* DDB */ 1073 1074 #ifdef MALLOC_PROFILE 1075 1076 static int 1077 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1078 { 1079 struct sbuf sbuf; 1080 uint64_t count; 1081 uint64_t waste; 1082 uint64_t mem; 1083 int error; 1084 int rsize; 1085 int size; 1086 int i; 1087 1088 waste = 0; 1089 mem = 0; 1090 1091 error = sysctl_wire_old_buffer(req, 0); 1092 if (error != 0) 1093 return (error); 1094 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1095 sbuf_printf(&sbuf, 1096 "\n Size Requests Real Size\n"); 1097 for (i = 0; i < KMEM_ZSIZE; i++) { 1098 size = i << KMEM_ZSHIFT; 1099 rsize = kmemzones[kmemsize[i]].kz_size; 1100 count = (long long unsigned)krequests[i]; 1101 1102 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1103 (unsigned long long)count, rsize); 1104 1105 if ((rsize * count) > (size * count)) 1106 waste += (rsize * count) - (size * count); 1107 mem += (rsize * count); 1108 } 1109 sbuf_printf(&sbuf, 1110 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1111 (unsigned long long)mem, (unsigned long long)waste); 1112 error = sbuf_finish(&sbuf); 1113 sbuf_delete(&sbuf); 1114 return (error); 1115 } 1116 1117 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 1118 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 1119 #endif /* MALLOC_PROFILE */ 1120