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/time.h> 63 #include <sys/vmem.h> 64 65 #include <vm/vm.h> 66 #include <vm/pmap.h> 67 #include <vm/vm_pageout.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 struct malloc_type *kmemstatistics; 117 static int kmemcount; 118 119 #define KMEM_ZSHIFT 4 120 #define KMEM_ZBASE 16 121 #define KMEM_ZMASK (KMEM_ZBASE - 1) 122 123 #define KMEM_ZMAX 65536 124 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 125 static uint8_t kmemsize[KMEM_ZSIZE + 1]; 126 127 #ifndef MALLOC_DEBUG_MAXZONES 128 #define MALLOC_DEBUG_MAXZONES 1 129 #endif 130 static int numzones = MALLOC_DEBUG_MAXZONES; 131 132 /* 133 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 134 * of various sizes. 135 * 136 * XXX: The comment here used to read "These won't be powers of two for 137 * long." It's possible that a significant amount of wasted memory could be 138 * recovered by tuning the sizes of these buckets. 139 */ 140 struct { 141 int kz_size; 142 char *kz_name; 143 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 144 } kmemzones[] = { 145 {16, "16", }, 146 {32, "32", }, 147 {64, "64", }, 148 {128, "128", }, 149 {256, "256", }, 150 {512, "512", }, 151 {1024, "1024", }, 152 {2048, "2048", }, 153 {4096, "4096", }, 154 {8192, "8192", }, 155 {16384, "16384", }, 156 {32768, "32768", }, 157 {65536, "65536", }, 158 {0, NULL}, 159 }; 160 161 /* 162 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 163 * types are described by a data structure passed by the declaring code, but 164 * the malloc(9) implementation has its own data structure describing the 165 * type and statistics. This permits the malloc(9)-internal data structures 166 * to be modified without breaking binary-compiled kernel modules that 167 * declare malloc types. 168 */ 169 static uma_zone_t mt_zone; 170 171 u_long vm_kmem_size; 172 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 173 "Size of kernel memory"); 174 175 static u_long kmem_zmax = KMEM_ZMAX; 176 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 177 "Maximum allocation size that malloc(9) would use UMA as backend"); 178 179 static u_long vm_kmem_size_min; 180 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 181 "Minimum size of kernel memory"); 182 183 static u_long vm_kmem_size_max; 184 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 185 "Maximum size of kernel memory"); 186 187 static u_int vm_kmem_size_scale; 188 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 189 "Scale factor for kernel memory size"); 190 191 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 192 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 193 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 194 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 195 196 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 197 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 198 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 199 sysctl_kmem_map_free, "LU", "Free space in kmem"); 200 201 /* 202 * The malloc_mtx protects the kmemstatistics linked list. 203 */ 204 struct mtx malloc_mtx; 205 206 #ifdef MALLOC_PROFILE 207 uint64_t krequests[KMEM_ZSIZE + 1]; 208 209 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 210 #endif 211 212 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 213 214 /* 215 * time_uptime of the last malloc(9) failure (induced or real). 216 */ 217 static time_t t_malloc_fail; 218 219 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 220 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 221 "Kernel malloc debugging options"); 222 #endif 223 224 /* 225 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 226 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 227 */ 228 #ifdef MALLOC_MAKE_FAILURES 229 static int malloc_failure_rate; 230 static int malloc_nowait_count; 231 static int malloc_failure_count; 232 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, 233 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 234 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 235 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 236 #endif 237 238 static int 239 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 240 { 241 u_long size; 242 243 size = vmem_size(kmem_arena, VMEM_ALLOC); 244 return (sysctl_handle_long(oidp, &size, 0, req)); 245 } 246 247 static int 248 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 249 { 250 u_long size; 251 252 size = vmem_size(kmem_arena, VMEM_FREE); 253 return (sysctl_handle_long(oidp, &size, 0, req)); 254 } 255 256 /* 257 * malloc(9) uma zone separation -- sub-page buffer overruns in one 258 * malloc type will affect only a subset of other malloc types. 259 */ 260 #if MALLOC_DEBUG_MAXZONES > 1 261 static void 262 tunable_set_numzones(void) 263 { 264 265 TUNABLE_INT_FETCH("debug.malloc.numzones", 266 &numzones); 267 268 /* Sanity check the number of malloc uma zones. */ 269 if (numzones <= 0) 270 numzones = 1; 271 if (numzones > MALLOC_DEBUG_MAXZONES) 272 numzones = MALLOC_DEBUG_MAXZONES; 273 } 274 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 275 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 276 &numzones, 0, "Number of malloc uma subzones"); 277 278 /* 279 * Any number that changes regularly is an okay choice for the 280 * offset. Build numbers are pretty good of you have them. 281 */ 282 static u_int zone_offset = __FreeBSD_version; 283 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 284 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 285 &zone_offset, 0, "Separate malloc types by examining the " 286 "Nth character in the malloc type short description."); 287 288 static u_int 289 mtp_get_subzone(const char *desc) 290 { 291 size_t len; 292 u_int val; 293 294 if (desc == NULL || (len = strlen(desc)) == 0) 295 return (0); 296 val = desc[zone_offset % len]; 297 return (val % numzones); 298 } 299 #elif MALLOC_DEBUG_MAXZONES == 0 300 #error "MALLOC_DEBUG_MAXZONES must be positive." 301 #else 302 static inline u_int 303 mtp_get_subzone(const char *desc) 304 { 305 306 return (0); 307 } 308 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 309 310 int 311 malloc_last_fail(void) 312 { 313 314 return (time_uptime - t_malloc_fail); 315 } 316 317 /* 318 * An allocation has succeeded -- update malloc type statistics for the 319 * amount of bucket size. Occurs within a critical section so that the 320 * thread isn't preempted and doesn't migrate while updating per-PCU 321 * statistics. 322 */ 323 static void 324 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 325 int zindx) 326 { 327 struct malloc_type_internal *mtip; 328 struct malloc_type_stats *mtsp; 329 330 critical_enter(); 331 mtip = mtp->ks_handle; 332 mtsp = &mtip->mti_stats[curcpu]; 333 if (size > 0) { 334 mtsp->mts_memalloced += size; 335 mtsp->mts_numallocs++; 336 } 337 if (zindx != -1) 338 mtsp->mts_size |= 1 << zindx; 339 340 #ifdef KDTRACE_HOOKS 341 if (dtrace_malloc_probe != NULL) { 342 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 343 if (probe_id != 0) 344 (dtrace_malloc_probe)(probe_id, 345 (uintptr_t) mtp, (uintptr_t) mtip, 346 (uintptr_t) mtsp, size, zindx); 347 } 348 #endif 349 350 critical_exit(); 351 } 352 353 void 354 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 355 { 356 357 if (size > 0) 358 malloc_type_zone_allocated(mtp, size, -1); 359 } 360 361 /* 362 * A free operation has occurred -- update malloc type statistics for the 363 * amount of the bucket size. Occurs within a critical section so that the 364 * thread isn't preempted and doesn't migrate while updating per-CPU 365 * statistics. 366 */ 367 void 368 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 369 { 370 struct malloc_type_internal *mtip; 371 struct malloc_type_stats *mtsp; 372 373 critical_enter(); 374 mtip = mtp->ks_handle; 375 mtsp = &mtip->mti_stats[curcpu]; 376 mtsp->mts_memfreed += size; 377 mtsp->mts_numfrees++; 378 379 #ifdef KDTRACE_HOOKS 380 if (dtrace_malloc_probe != NULL) { 381 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 382 if (probe_id != 0) 383 (dtrace_malloc_probe)(probe_id, 384 (uintptr_t) mtp, (uintptr_t) mtip, 385 (uintptr_t) mtsp, size, 0); 386 } 387 #endif 388 389 critical_exit(); 390 } 391 392 /* 393 * contigmalloc: 394 * 395 * Allocate a block of physically contiguous memory. 396 * 397 * If M_NOWAIT is set, this routine will not block and return NULL if 398 * the allocation fails. 399 */ 400 void * 401 contigmalloc(unsigned long size, struct malloc_type *type, int flags, 402 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 403 vm_paddr_t boundary) 404 { 405 void *ret; 406 407 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high, 408 alignment, boundary, VM_MEMATTR_DEFAULT); 409 if (ret != NULL) 410 malloc_type_allocated(type, round_page(size)); 411 return (ret); 412 } 413 414 /* 415 * contigfree: 416 * 417 * Free a block of memory allocated by contigmalloc. 418 * 419 * This routine may not block. 420 */ 421 void 422 contigfree(void *addr, unsigned long size, struct malloc_type *type) 423 { 424 425 kmem_free(kernel_arena, (vm_offset_t)addr, size); 426 malloc_type_freed(type, round_page(size)); 427 } 428 429 /* 430 * malloc: 431 * 432 * Allocate a block of memory. 433 * 434 * If M_NOWAIT is set, this routine will not block and return NULL if 435 * the allocation fails. 436 */ 437 void * 438 malloc(unsigned long size, struct malloc_type *mtp, int flags) 439 { 440 int indx; 441 struct malloc_type_internal *mtip; 442 caddr_t va; 443 uma_zone_t zone; 444 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE) 445 unsigned long osize = size; 446 #endif 447 448 #ifdef INVARIANTS 449 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 450 /* 451 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 452 */ 453 indx = flags & (M_WAITOK | M_NOWAIT); 454 if (indx != M_NOWAIT && indx != M_WAITOK) { 455 static struct timeval lasterr; 456 static int curerr, once; 457 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 458 printf("Bad malloc flags: %x\n", indx); 459 kdb_backtrace(); 460 flags |= M_WAITOK; 461 once++; 462 } 463 } 464 #endif 465 #ifdef MALLOC_MAKE_FAILURES 466 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 467 atomic_add_int(&malloc_nowait_count, 1); 468 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 469 atomic_add_int(&malloc_failure_count, 1); 470 t_malloc_fail = time_uptime; 471 return (NULL); 472 } 473 } 474 #endif 475 if (flags & M_WAITOK) 476 KASSERT(curthread->td_intr_nesting_level == 0, 477 ("malloc(M_WAITOK) in interrupt context")); 478 479 KASSERT(curthread->td_critnest == 0, 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")); 547 548 KASSERT(curthread->td_critnest == 0, 549 ("free: called with spinlock or critical section held")); 550 551 /* free(NULL, ...) does nothing */ 552 if (addr == NULL) 553 return; 554 555 #ifdef DEBUG_MEMGUARD 556 if (is_memguard_addr(addr)) { 557 memguard_free(addr); 558 return; 559 } 560 #endif 561 562 #ifdef DEBUG_REDZONE 563 redzone_check(addr); 564 addr = redzone_addr_ntor(addr); 565 #endif 566 567 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 568 569 if (slab == NULL) 570 panic("free: address %p(%p) has not been allocated.\n", 571 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 572 573 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 574 #ifdef INVARIANTS 575 struct malloc_type **mtpp = addr; 576 #endif 577 size = slab->us_keg->uk_size; 578 #ifdef INVARIANTS 579 /* 580 * Cache a pointer to the malloc_type that most recently freed 581 * this memory here. This way we know who is most likely to 582 * have stepped on it later. 583 * 584 * This code assumes that size is a multiple of 8 bytes for 585 * 64 bit machines 586 */ 587 mtpp = (struct malloc_type **) 588 ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 589 mtpp += (size - sizeof(struct malloc_type *)) / 590 sizeof(struct malloc_type *); 591 *mtpp = mtp; 592 #endif 593 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 594 } else { 595 size = slab->us_size; 596 uma_large_free(slab); 597 } 598 malloc_type_freed(mtp, size); 599 } 600 601 /* 602 * realloc: change the size of a memory block 603 */ 604 void * 605 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 606 { 607 uma_slab_t slab; 608 unsigned long alloc; 609 void *newaddr; 610 611 KASSERT(mtp->ks_magic == M_MAGIC, 612 ("realloc: bad malloc type magic")); 613 614 KASSERT(curthread->td_critnest == 0, 615 ("realloc: called with spinlock or critical section held")); 616 617 /* realloc(NULL, ...) is equivalent to malloc(...) */ 618 if (addr == NULL) 619 return (malloc(size, mtp, flags)); 620 621 /* 622 * XXX: Should report free of old memory and alloc of new memory to 623 * per-CPU stats. 624 */ 625 626 #ifdef DEBUG_MEMGUARD 627 if (is_memguard_addr(addr)) 628 return (memguard_realloc(addr, size, mtp, flags)); 629 #endif 630 631 #ifdef DEBUG_REDZONE 632 slab = NULL; 633 alloc = redzone_get_size(addr); 634 #else 635 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 636 637 /* Sanity check */ 638 KASSERT(slab != NULL, 639 ("realloc: address %p out of range", (void *)addr)); 640 641 /* Get the size of the original block */ 642 if (!(slab->us_flags & UMA_SLAB_MALLOC)) 643 alloc = slab->us_keg->uk_size; 644 else 645 alloc = slab->us_size; 646 647 /* Reuse the original block if appropriate */ 648 if (size <= alloc 649 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 650 return (addr); 651 #endif /* !DEBUG_REDZONE */ 652 653 /* Allocate a new, bigger (or smaller) block */ 654 if ((newaddr = malloc(size, mtp, flags)) == NULL) 655 return (NULL); 656 657 /* Copy over original contents */ 658 bcopy(addr, newaddr, min(size, alloc)); 659 free(addr, mtp); 660 return (newaddr); 661 } 662 663 /* 664 * reallocf: same as realloc() but free memory on failure. 665 */ 666 void * 667 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 668 { 669 void *mem; 670 671 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 672 free(addr, mtp); 673 return (mem); 674 } 675 676 /* 677 * Wake the uma reclamation pagedaemon thread when we exhaust KVA. It 678 * will call the lowmem handler and uma_reclaim() callbacks in a 679 * context that is safe. 680 */ 681 static void 682 kmem_reclaim(vmem_t *vm, int flags) 683 { 684 685 uma_reclaim_wakeup(); 686 pagedaemon_wakeup(); 687 } 688 689 #ifndef __sparc64__ 690 CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 691 #endif 692 693 /* 694 * Initialize the kernel memory (kmem) arena. 695 */ 696 void 697 kmeminit(void) 698 { 699 u_long mem_size; 700 u_long tmp; 701 702 #ifdef VM_KMEM_SIZE 703 if (vm_kmem_size == 0) 704 vm_kmem_size = VM_KMEM_SIZE; 705 #endif 706 #ifdef VM_KMEM_SIZE_MIN 707 if (vm_kmem_size_min == 0) 708 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 709 #endif 710 #ifdef VM_KMEM_SIZE_MAX 711 if (vm_kmem_size_max == 0) 712 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 713 #endif 714 /* 715 * Calculate the amount of kernel virtual address (KVA) space that is 716 * preallocated to the kmem arena. In order to support a wide range 717 * of machines, it is a function of the physical memory size, 718 * specifically, 719 * 720 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 721 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 722 * 723 * Every architecture must define an integral value for 724 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 725 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 726 * ceiling on this preallocation, are optional. Typically, 727 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 728 * a given architecture. 729 */ 730 mem_size = vm_cnt.v_page_count; 731 if (mem_size <= 32768) /* delphij XXX 128MB */ 732 kmem_zmax = PAGE_SIZE; 733 734 if (vm_kmem_size_scale < 1) 735 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 736 737 /* 738 * Check if we should use defaults for the "vm_kmem_size" 739 * variable: 740 */ 741 if (vm_kmem_size == 0) { 742 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 743 744 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 745 vm_kmem_size = vm_kmem_size_min; 746 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 747 vm_kmem_size = vm_kmem_size_max; 748 } 749 750 /* 751 * The amount of KVA space that is preallocated to the 752 * kmem arena can be set statically at compile-time or manually 753 * through the kernel environment. However, it is still limited to 754 * twice the physical memory size, which has been sufficient to handle 755 * the most severe cases of external fragmentation in the kmem arena. 756 */ 757 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 758 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 759 760 vm_kmem_size = round_page(vm_kmem_size); 761 #ifdef DEBUG_MEMGUARD 762 tmp = memguard_fudge(vm_kmem_size, kernel_map); 763 #else 764 tmp = vm_kmem_size; 765 #endif 766 vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE, 767 0, 0); 768 vmem_set_reclaim(kmem_arena, kmem_reclaim); 769 770 #ifdef DEBUG_MEMGUARD 771 /* 772 * Initialize MemGuard if support compiled in. MemGuard is a 773 * replacement allocator used for detecting tamper-after-free 774 * scenarios as they occur. It is only used for debugging. 775 */ 776 memguard_init(kmem_arena); 777 #endif 778 } 779 780 /* 781 * Initialize the kernel memory allocator 782 */ 783 /* ARGSUSED*/ 784 static void 785 mallocinit(void *dummy) 786 { 787 int i; 788 uint8_t indx; 789 790 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 791 792 kmeminit(); 793 794 uma_startup2(); 795 796 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 797 kmem_zmax = KMEM_ZMAX; 798 799 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 800 #ifdef INVARIANTS 801 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 802 #else 803 NULL, NULL, NULL, NULL, 804 #endif 805 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 806 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 807 int size = kmemzones[indx].kz_size; 808 char *name = kmemzones[indx].kz_name; 809 int subzone; 810 811 for (subzone = 0; subzone < numzones; subzone++) { 812 kmemzones[indx].kz_zone[subzone] = 813 uma_zcreate(name, size, 814 #ifdef INVARIANTS 815 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 816 #else 817 NULL, NULL, NULL, NULL, 818 #endif 819 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 820 } 821 for (;i <= size; i+= KMEM_ZBASE) 822 kmemsize[i >> KMEM_ZSHIFT] = indx; 823 824 } 825 } 826 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); 827 828 void 829 malloc_init(void *data) 830 { 831 struct malloc_type_internal *mtip; 832 struct malloc_type *mtp; 833 834 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init")); 835 836 mtp = data; 837 if (mtp->ks_magic != M_MAGIC) 838 panic("malloc_init: bad malloc type magic"); 839 840 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 841 mtp->ks_handle = mtip; 842 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); 843 844 mtx_lock(&malloc_mtx); 845 mtp->ks_next = kmemstatistics; 846 kmemstatistics = mtp; 847 kmemcount++; 848 mtx_unlock(&malloc_mtx); 849 } 850 851 void 852 malloc_uninit(void *data) 853 { 854 struct malloc_type_internal *mtip; 855 struct malloc_type_stats *mtsp; 856 struct malloc_type *mtp, *temp; 857 uma_slab_t slab; 858 long temp_allocs, temp_bytes; 859 int i; 860 861 mtp = data; 862 KASSERT(mtp->ks_magic == M_MAGIC, 863 ("malloc_uninit: bad malloc type magic")); 864 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 865 866 mtx_lock(&malloc_mtx); 867 mtip = mtp->ks_handle; 868 mtp->ks_handle = NULL; 869 if (mtp != kmemstatistics) { 870 for (temp = kmemstatistics; temp != NULL; 871 temp = temp->ks_next) { 872 if (temp->ks_next == mtp) { 873 temp->ks_next = mtp->ks_next; 874 break; 875 } 876 } 877 KASSERT(temp, 878 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 879 } else 880 kmemstatistics = mtp->ks_next; 881 kmemcount--; 882 mtx_unlock(&malloc_mtx); 883 884 /* 885 * Look for memory leaks. 886 */ 887 temp_allocs = temp_bytes = 0; 888 for (i = 0; i < MAXCPU; i++) { 889 mtsp = &mtip->mti_stats[i]; 890 temp_allocs += mtsp->mts_numallocs; 891 temp_allocs -= mtsp->mts_numfrees; 892 temp_bytes += mtsp->mts_memalloced; 893 temp_bytes -= mtsp->mts_memfreed; 894 } 895 if (temp_allocs > 0 || temp_bytes > 0) { 896 printf("Warning: memory type %s leaked memory on destroy " 897 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 898 temp_allocs, temp_bytes); 899 } 900 901 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 902 uma_zfree_arg(mt_zone, mtip, slab); 903 } 904 905 struct malloc_type * 906 malloc_desc2type(const char *desc) 907 { 908 struct malloc_type *mtp; 909 910 mtx_assert(&malloc_mtx, MA_OWNED); 911 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 912 if (strcmp(mtp->ks_shortdesc, desc) == 0) 913 return (mtp); 914 } 915 return (NULL); 916 } 917 918 static int 919 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 920 { 921 struct malloc_type_stream_header mtsh; 922 struct malloc_type_internal *mtip; 923 struct malloc_type_header mth; 924 struct malloc_type *mtp; 925 int error, i; 926 struct sbuf sbuf; 927 928 error = sysctl_wire_old_buffer(req, 0); 929 if (error != 0) 930 return (error); 931 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 932 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); 933 mtx_lock(&malloc_mtx); 934 935 /* 936 * Insert stream header. 937 */ 938 bzero(&mtsh, sizeof(mtsh)); 939 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 940 mtsh.mtsh_maxcpus = MAXCPU; 941 mtsh.mtsh_count = kmemcount; 942 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 943 944 /* 945 * Insert alternating sequence of type headers and type statistics. 946 */ 947 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 948 mtip = (struct malloc_type_internal *)mtp->ks_handle; 949 950 /* 951 * Insert type header. 952 */ 953 bzero(&mth, sizeof(mth)); 954 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 955 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 956 957 /* 958 * Insert type statistics for each CPU. 959 */ 960 for (i = 0; i < MAXCPU; i++) { 961 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], 962 sizeof(mtip->mti_stats[i])); 963 } 964 } 965 mtx_unlock(&malloc_mtx); 966 error = sbuf_finish(&sbuf); 967 sbuf_delete(&sbuf); 968 return (error); 969 } 970 971 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 972 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 973 "Return malloc types"); 974 975 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 976 "Count of kernel malloc types"); 977 978 void 979 malloc_type_list(malloc_type_list_func_t *func, void *arg) 980 { 981 struct malloc_type *mtp, **bufmtp; 982 int count, i; 983 size_t buflen; 984 985 mtx_lock(&malloc_mtx); 986 restart: 987 mtx_assert(&malloc_mtx, MA_OWNED); 988 count = kmemcount; 989 mtx_unlock(&malloc_mtx); 990 991 buflen = sizeof(struct malloc_type *) * count; 992 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 993 994 mtx_lock(&malloc_mtx); 995 996 if (count < kmemcount) { 997 free(bufmtp, M_TEMP); 998 goto restart; 999 } 1000 1001 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 1002 bufmtp[i] = mtp; 1003 1004 mtx_unlock(&malloc_mtx); 1005 1006 for (i = 0; i < count; i++) 1007 (func)(bufmtp[i], arg); 1008 1009 free(bufmtp, M_TEMP); 1010 } 1011 1012 #ifdef DDB 1013 DB_SHOW_COMMAND(malloc, db_show_malloc) 1014 { 1015 struct malloc_type_internal *mtip; 1016 struct malloc_type *mtp; 1017 uint64_t allocs, frees; 1018 uint64_t alloced, freed; 1019 int i; 1020 1021 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 1022 "Requests"); 1023 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1024 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1025 allocs = 0; 1026 frees = 0; 1027 alloced = 0; 1028 freed = 0; 1029 for (i = 0; i < MAXCPU; i++) { 1030 allocs += mtip->mti_stats[i].mts_numallocs; 1031 frees += mtip->mti_stats[i].mts_numfrees; 1032 alloced += mtip->mti_stats[i].mts_memalloced; 1033 freed += mtip->mti_stats[i].mts_memfreed; 1034 } 1035 db_printf("%18s %12ju %12juK %12ju\n", 1036 mtp->ks_shortdesc, allocs - frees, 1037 (alloced - freed + 1023) / 1024, allocs); 1038 if (db_pager_quit) 1039 break; 1040 } 1041 } 1042 1043 #if MALLOC_DEBUG_MAXZONES > 1 1044 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1045 { 1046 struct malloc_type_internal *mtip; 1047 struct malloc_type *mtp; 1048 u_int subzone; 1049 1050 if (!have_addr) { 1051 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1052 return; 1053 } 1054 mtp = (void *)addr; 1055 if (mtp->ks_magic != M_MAGIC) { 1056 db_printf("Magic %lx does not match expected %x\n", 1057 mtp->ks_magic, M_MAGIC); 1058 return; 1059 } 1060 1061 mtip = mtp->ks_handle; 1062 subzone = mtip->mti_zone; 1063 1064 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1065 mtip = mtp->ks_handle; 1066 if (mtip->mti_zone != subzone) 1067 continue; 1068 db_printf("%s\n", mtp->ks_shortdesc); 1069 if (db_pager_quit) 1070 break; 1071 } 1072 } 1073 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1074 #endif /* DDB */ 1075 1076 #ifdef MALLOC_PROFILE 1077 1078 static int 1079 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1080 { 1081 struct sbuf sbuf; 1082 uint64_t count; 1083 uint64_t waste; 1084 uint64_t mem; 1085 int error; 1086 int rsize; 1087 int size; 1088 int i; 1089 1090 waste = 0; 1091 mem = 0; 1092 1093 error = sysctl_wire_old_buffer(req, 0); 1094 if (error != 0) 1095 return (error); 1096 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1097 sbuf_printf(&sbuf, 1098 "\n Size Requests Real Size\n"); 1099 for (i = 0; i < KMEM_ZSIZE; i++) { 1100 size = i << KMEM_ZSHIFT; 1101 rsize = kmemzones[kmemsize[i]].kz_size; 1102 count = (long long unsigned)krequests[i]; 1103 1104 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1105 (unsigned long long)count, rsize); 1106 1107 if ((rsize * count) > (size * count)) 1108 waste += (rsize * count) - (size * count); 1109 mem += (rsize * count); 1110 } 1111 sbuf_printf(&sbuf, 1112 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1113 (unsigned long long)mem, (unsigned long long)waste); 1114 error = sbuf_finish(&sbuf); 1115 sbuf_delete(&sbuf); 1116 return (error); 1117 } 1118 1119 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 1120 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 1121 #endif /* MALLOC_PROFILE */ 1122