kern_malloc.c revision 266204
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: stable/10/sys/kern/kern_malloc.c 266204 2014-05-16 01:30:30Z ian $"); 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/mutex.h> 59#include <sys/vmmeter.h> 60#include <sys/proc.h> 61#include <sys/sbuf.h> 62#include <sys/sysctl.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 94dtrace_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 */ 110MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 111MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 112MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 113 114MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); 115MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); 116 117static struct malloc_type *kmemstatistics; 118static 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 PAGE_SIZE 125#define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 126static uint8_t kmemsize[KMEM_ZSIZE + 1]; 127 128#ifndef MALLOC_DEBUG_MAXZONES 129#define MALLOC_DEBUG_MAXZONES 1 130#endif 131static 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 */ 141struct { 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#if PAGE_SIZE > 4096 156 {8192, "8192", }, 157#if PAGE_SIZE > 8192 158 {16384, "16384", }, 159#if PAGE_SIZE > 16384 160 {32768, "32768", }, 161#if PAGE_SIZE > 32768 162 {65536, "65536", }, 163#if PAGE_SIZE > 65536 164#error "Unsupported PAGE_SIZE" 165#endif /* 65536 */ 166#endif /* 32768 */ 167#endif /* 16384 */ 168#endif /* 8192 */ 169#endif /* 4096 */ 170 {0, NULL}, 171}; 172 173/* 174 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 175 * types are described by a data structure passed by the declaring code, but 176 * the malloc(9) implementation has its own data structure describing the 177 * type and statistics. This permits the malloc(9)-internal data structures 178 * to be modified without breaking binary-compiled kernel modules that 179 * declare malloc types. 180 */ 181static uma_zone_t mt_zone; 182 183u_long vm_kmem_size; 184SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 185 "Size of kernel memory"); 186 187static u_long vm_kmem_size_min; 188SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 189 "Minimum size of kernel memory"); 190 191static u_long vm_kmem_size_max; 192SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 193 "Maximum size of kernel memory"); 194 195static u_int vm_kmem_size_scale; 196SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 197 "Scale factor for kernel memory size"); 198 199static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 200SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 201 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 202 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 203 204static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 205SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 206 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 207 sysctl_kmem_map_free, "LU", "Free space in kmem"); 208 209/* 210 * The malloc_mtx protects the kmemstatistics linked list. 211 */ 212struct mtx malloc_mtx; 213 214#ifdef MALLOC_PROFILE 215uint64_t krequests[KMEM_ZSIZE + 1]; 216 217static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 218#endif 219 220static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 221 222/* 223 * time_uptime of the last malloc(9) failure (induced or real). 224 */ 225static time_t t_malloc_fail; 226 227#if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 228static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 229 "Kernel malloc debugging options"); 230#endif 231 232/* 233 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 234 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 235 */ 236#ifdef MALLOC_MAKE_FAILURES 237static int malloc_failure_rate; 238static int malloc_nowait_count; 239static int malloc_failure_count; 240SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW, 241 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 242TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate); 243SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 244 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 245#endif 246 247static int 248sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 249{ 250 u_long size; 251 252 size = vmem_size(kmem_arena, VMEM_ALLOC); 253 return (sysctl_handle_long(oidp, &size, 0, req)); 254} 255 256static int 257sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 258{ 259 u_long size; 260 261 size = vmem_size(kmem_arena, VMEM_FREE); 262 return (sysctl_handle_long(oidp, &size, 0, req)); 263} 264 265/* 266 * malloc(9) uma zone separation -- sub-page buffer overruns in one 267 * malloc type will affect only a subset of other malloc types. 268 */ 269#if MALLOC_DEBUG_MAXZONES > 1 270static void 271tunable_set_numzones(void) 272{ 273 274 TUNABLE_INT_FETCH("debug.malloc.numzones", 275 &numzones); 276 277 /* Sanity check the number of malloc uma zones. */ 278 if (numzones <= 0) 279 numzones = 1; 280 if (numzones > MALLOC_DEBUG_MAXZONES) 281 numzones = MALLOC_DEBUG_MAXZONES; 282} 283SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 284SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN, 285 &numzones, 0, "Number of malloc uma subzones"); 286 287/* 288 * Any number that changes regularly is an okay choice for the 289 * offset. Build numbers are pretty good of you have them. 290 */ 291static u_int zone_offset = __FreeBSD_version; 292TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 293SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 294 &zone_offset, 0, "Separate malloc types by examining the " 295 "Nth character in the malloc type short description."); 296 297static u_int 298mtp_get_subzone(const char *desc) 299{ 300 size_t len; 301 u_int val; 302 303 if (desc == NULL || (len = strlen(desc)) == 0) 304 return (0); 305 val = desc[zone_offset % len]; 306 return (val % numzones); 307} 308#elif MALLOC_DEBUG_MAXZONES == 0 309#error "MALLOC_DEBUG_MAXZONES must be positive." 310#else 311static inline u_int 312mtp_get_subzone(const char *desc) 313{ 314 315 return (0); 316} 317#endif /* MALLOC_DEBUG_MAXZONES > 1 */ 318 319int 320malloc_last_fail(void) 321{ 322 323 return (time_uptime - t_malloc_fail); 324} 325 326/* 327 * An allocation has succeeded -- update malloc type statistics for the 328 * amount of bucket size. Occurs within a critical section so that the 329 * thread isn't preempted and doesn't migrate while updating per-PCU 330 * statistics. 331 */ 332static void 333malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 334 int zindx) 335{ 336 struct malloc_type_internal *mtip; 337 struct malloc_type_stats *mtsp; 338 339 critical_enter(); 340 mtip = mtp->ks_handle; 341 mtsp = &mtip->mti_stats[curcpu]; 342 if (size > 0) { 343 mtsp->mts_memalloced += size; 344 mtsp->mts_numallocs++; 345 } 346 if (zindx != -1) 347 mtsp->mts_size |= 1 << zindx; 348 349#ifdef KDTRACE_HOOKS 350 if (dtrace_malloc_probe != NULL) { 351 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 352 if (probe_id != 0) 353 (dtrace_malloc_probe)(probe_id, 354 (uintptr_t) mtp, (uintptr_t) mtip, 355 (uintptr_t) mtsp, size, zindx); 356 } 357#endif 358 359 critical_exit(); 360} 361 362void 363malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 364{ 365 366 if (size > 0) 367 malloc_type_zone_allocated(mtp, size, -1); 368} 369 370/* 371 * A free operation has occurred -- update malloc type statistics for the 372 * amount of the bucket size. Occurs within a critical section so that the 373 * thread isn't preempted and doesn't migrate while updating per-CPU 374 * statistics. 375 */ 376void 377malloc_type_freed(struct malloc_type *mtp, unsigned long size) 378{ 379 struct malloc_type_internal *mtip; 380 struct malloc_type_stats *mtsp; 381 382 critical_enter(); 383 mtip = mtp->ks_handle; 384 mtsp = &mtip->mti_stats[curcpu]; 385 mtsp->mts_memfreed += size; 386 mtsp->mts_numfrees++; 387 388#ifdef KDTRACE_HOOKS 389 if (dtrace_malloc_probe != NULL) { 390 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 391 if (probe_id != 0) 392 (dtrace_malloc_probe)(probe_id, 393 (uintptr_t) mtp, (uintptr_t) mtip, 394 (uintptr_t) mtsp, size, 0); 395 } 396#endif 397 398 critical_exit(); 399} 400 401/* 402 * contigmalloc: 403 * 404 * Allocate a block of physically contiguous memory. 405 * 406 * If M_NOWAIT is set, this routine will not block and return NULL if 407 * the allocation fails. 408 */ 409void * 410contigmalloc(unsigned long size, struct malloc_type *type, int flags, 411 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 412 vm_paddr_t boundary) 413{ 414 void *ret; 415 416 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high, 417 alignment, boundary, VM_MEMATTR_DEFAULT); 418 if (ret != NULL) 419 malloc_type_allocated(type, round_page(size)); 420 return (ret); 421} 422 423/* 424 * contigfree: 425 * 426 * Free a block of memory allocated by contigmalloc. 427 * 428 * This routine may not block. 429 */ 430void 431contigfree(void *addr, unsigned long size, struct malloc_type *type) 432{ 433 434 kmem_free(kernel_arena, (vm_offset_t)addr, size); 435 malloc_type_freed(type, round_page(size)); 436} 437 438/* 439 * malloc: 440 * 441 * Allocate a block of memory. 442 * 443 * If M_NOWAIT is set, this routine will not block and return NULL if 444 * the allocation fails. 445 */ 446void * 447malloc(unsigned long size, struct malloc_type *mtp, int flags) 448{ 449 int indx; 450 struct malloc_type_internal *mtip; 451 caddr_t va; 452 uma_zone_t zone; 453#if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE) 454 unsigned long osize = size; 455#endif 456 457#ifdef INVARIANTS 458 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 459 /* 460 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 461 */ 462 indx = flags & (M_WAITOK | M_NOWAIT); 463 if (indx != M_NOWAIT && indx != M_WAITOK) { 464 static struct timeval lasterr; 465 static int curerr, once; 466 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 467 printf("Bad malloc flags: %x\n", indx); 468 kdb_backtrace(); 469 flags |= M_WAITOK; 470 once++; 471 } 472 } 473#endif 474#ifdef MALLOC_MAKE_FAILURES 475 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 476 atomic_add_int(&malloc_nowait_count, 1); 477 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 478 atomic_add_int(&malloc_failure_count, 1); 479 t_malloc_fail = time_uptime; 480 return (NULL); 481 } 482 } 483#endif 484 if (flags & M_WAITOK) 485 KASSERT(curthread->td_intr_nesting_level == 0, 486 ("malloc(M_WAITOK) in interrupt context")); 487 488#ifdef DEBUG_MEMGUARD 489 if (memguard_cmp_mtp(mtp, size)) { 490 va = memguard_alloc(size, flags); 491 if (va != NULL) 492 return (va); 493 /* This is unfortunate but should not be fatal. */ 494 } 495#endif 496 497#ifdef DEBUG_REDZONE 498 size = redzone_size_ntor(size); 499#endif 500 501 if (size <= KMEM_ZMAX) { 502 mtip = mtp->ks_handle; 503 if (size & KMEM_ZMASK) 504 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 505 indx = kmemsize[size >> KMEM_ZSHIFT]; 506 KASSERT(mtip->mti_zone < numzones, 507 ("mti_zone %u out of range %d", 508 mtip->mti_zone, numzones)); 509 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 510#ifdef MALLOC_PROFILE 511 krequests[size >> KMEM_ZSHIFT]++; 512#endif 513 va = uma_zalloc(zone, flags); 514 if (va != NULL) 515 size = zone->uz_size; 516 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 517 } else { 518 size = roundup(size, PAGE_SIZE); 519 zone = NULL; 520 va = uma_large_malloc(size, flags); 521 malloc_type_allocated(mtp, va == NULL ? 0 : size); 522 } 523 if (flags & M_WAITOK) 524 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 525 else if (va == NULL) 526 t_malloc_fail = time_uptime; 527#ifdef DIAGNOSTIC 528 if (va != NULL && !(flags & M_ZERO)) { 529 memset(va, 0x70, osize); 530 } 531#endif 532#ifdef DEBUG_REDZONE 533 if (va != NULL) 534 va = redzone_setup(va, osize); 535#endif 536 return ((void *) va); 537} 538 539/* 540 * free: 541 * 542 * Free a block of memory allocated by malloc. 543 * 544 * This routine may not block. 545 */ 546void 547free(void *addr, struct malloc_type *mtp) 548{ 549 uma_slab_t slab; 550 u_long size; 551 552 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); 553 554 /* free(NULL, ...) does nothing */ 555 if (addr == NULL) 556 return; 557 558#ifdef DEBUG_MEMGUARD 559 if (is_memguard_addr(addr)) { 560 memguard_free(addr); 561 return; 562 } 563#endif 564 565#ifdef DEBUG_REDZONE 566 redzone_check(addr); 567 addr = redzone_addr_ntor(addr); 568#endif 569 570 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 571 572 if (slab == NULL) 573 panic("free: address %p(%p) has not been allocated.\n", 574 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 575 576 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 577#ifdef INVARIANTS 578 struct malloc_type **mtpp = addr; 579#endif 580 size = slab->us_keg->uk_size; 581#ifdef INVARIANTS 582 /* 583 * Cache a pointer to the malloc_type that most recently freed 584 * this memory here. This way we know who is most likely to 585 * have stepped on it later. 586 * 587 * This code assumes that size is a multiple of 8 bytes for 588 * 64 bit machines 589 */ 590 mtpp = (struct malloc_type **) 591 ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 592 mtpp += (size - sizeof(struct malloc_type *)) / 593 sizeof(struct malloc_type *); 594 *mtpp = mtp; 595#endif 596 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 597 } else { 598 size = slab->us_size; 599 uma_large_free(slab); 600 } 601 malloc_type_freed(mtp, size); 602} 603 604/* 605 * realloc: change the size of a memory block 606 */ 607void * 608realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 609{ 610 uma_slab_t slab; 611 unsigned long alloc; 612 void *newaddr; 613 614 KASSERT(mtp->ks_magic == M_MAGIC, 615 ("realloc: bad malloc type magic")); 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 */ 666void * 667reallocf(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 page daemon when we exhaust KVA. It will call the lowmem handler 678 * and uma_reclaim() callbacks in a context that is safe. 679 */ 680static void 681kmem_reclaim(vmem_t *vm, int flags) 682{ 683 684 pagedaemon_wakeup(); 685} 686 687CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 688 689/* 690 * Initialize the kernel memory (kmem) arena. 691 */ 692void 693kmeminit(void) 694{ 695 u_long mem_size, tmp; 696 697 /* 698 * Calculate the amount of kernel virtual address (KVA) space that is 699 * preallocated to the kmem arena. In order to support a wide range 700 * of machines, it is a function of the physical memory size, 701 * specifically, 702 * 703 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 704 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 705 * 706 * Every architecture must define an integral value for 707 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 708 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 709 * ceiling on this preallocation, are optional. Typically, 710 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 711 * a given architecture. 712 */ 713 mem_size = cnt.v_page_count; 714 715 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 716 TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale); 717 if (vm_kmem_size_scale < 1) 718 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 719 720 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 721 722#if defined(VM_KMEM_SIZE_MIN) 723 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 724#endif 725 TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min); 726 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 727 vm_kmem_size = vm_kmem_size_min; 728 729#if defined(VM_KMEM_SIZE_MAX) 730 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 731#endif 732 TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max); 733 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 734 vm_kmem_size = vm_kmem_size_max; 735 736 /* 737 * Alternatively, the amount of KVA space that is preallocated to the 738 * kmem arena can be set statically at compile-time or manually 739 * through the kernel environment. However, it is still limited to 740 * twice the physical memory size, which has been sufficient to handle 741 * the most severe cases of external fragmentation in the kmem arena. 742 */ 743#if defined(VM_KMEM_SIZE) 744 vm_kmem_size = VM_KMEM_SIZE; 745#endif 746 TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size); 747 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 748 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 749 750 vm_kmem_size = round_page(vm_kmem_size); 751#ifdef DEBUG_MEMGUARD 752 tmp = memguard_fudge(vm_kmem_size, kernel_map); 753#else 754 tmp = vm_kmem_size; 755#endif 756 vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE, 757 0, 0); 758 vmem_set_reclaim(kmem_arena, kmem_reclaim); 759 760#ifdef DEBUG_MEMGUARD 761 /* 762 * Initialize MemGuard if support compiled in. MemGuard is a 763 * replacement allocator used for detecting tamper-after-free 764 * scenarios as they occur. It is only used for debugging. 765 */ 766 memguard_init(kmem_arena); 767#endif 768} 769 770/* 771 * Initialize the kernel memory allocator 772 */ 773/* ARGSUSED*/ 774static void 775mallocinit(void *dummy) 776{ 777 int i; 778 uint8_t indx; 779 780 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 781 782 kmeminit(); 783 784 uma_startup2(); 785 786 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 787#ifdef INVARIANTS 788 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 789#else 790 NULL, NULL, NULL, NULL, 791#endif 792 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 793 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 794 int size = kmemzones[indx].kz_size; 795 char *name = kmemzones[indx].kz_name; 796 int subzone; 797 798 for (subzone = 0; subzone < numzones; subzone++) { 799 kmemzones[indx].kz_zone[subzone] = 800 uma_zcreate(name, size, 801#ifdef INVARIANTS 802 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 803#else 804 NULL, NULL, NULL, NULL, 805#endif 806 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 807 } 808 for (;i <= size; i+= KMEM_ZBASE) 809 kmemsize[i >> KMEM_ZSHIFT] = indx; 810 811 } 812} 813SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, mallocinit, NULL); 814 815void 816malloc_init(void *data) 817{ 818 struct malloc_type_internal *mtip; 819 struct malloc_type *mtp; 820 821 KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init")); 822 823 mtp = data; 824 if (mtp->ks_magic != M_MAGIC) 825 panic("malloc_init: bad malloc type magic"); 826 827 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 828 mtp->ks_handle = mtip; 829 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); 830 831 mtx_lock(&malloc_mtx); 832 mtp->ks_next = kmemstatistics; 833 kmemstatistics = mtp; 834 kmemcount++; 835 mtx_unlock(&malloc_mtx); 836} 837 838void 839malloc_uninit(void *data) 840{ 841 struct malloc_type_internal *mtip; 842 struct malloc_type_stats *mtsp; 843 struct malloc_type *mtp, *temp; 844 uma_slab_t slab; 845 long temp_allocs, temp_bytes; 846 int i; 847 848 mtp = data; 849 KASSERT(mtp->ks_magic == M_MAGIC, 850 ("malloc_uninit: bad malloc type magic")); 851 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 852 853 mtx_lock(&malloc_mtx); 854 mtip = mtp->ks_handle; 855 mtp->ks_handle = NULL; 856 if (mtp != kmemstatistics) { 857 for (temp = kmemstatistics; temp != NULL; 858 temp = temp->ks_next) { 859 if (temp->ks_next == mtp) { 860 temp->ks_next = mtp->ks_next; 861 break; 862 } 863 } 864 KASSERT(temp, 865 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 866 } else 867 kmemstatistics = mtp->ks_next; 868 kmemcount--; 869 mtx_unlock(&malloc_mtx); 870 871 /* 872 * Look for memory leaks. 873 */ 874 temp_allocs = temp_bytes = 0; 875 for (i = 0; i < MAXCPU; i++) { 876 mtsp = &mtip->mti_stats[i]; 877 temp_allocs += mtsp->mts_numallocs; 878 temp_allocs -= mtsp->mts_numfrees; 879 temp_bytes += mtsp->mts_memalloced; 880 temp_bytes -= mtsp->mts_memfreed; 881 } 882 if (temp_allocs > 0 || temp_bytes > 0) { 883 printf("Warning: memory type %s leaked memory on destroy " 884 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 885 temp_allocs, temp_bytes); 886 } 887 888 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 889 uma_zfree_arg(mt_zone, mtip, slab); 890} 891 892struct malloc_type * 893malloc_desc2type(const char *desc) 894{ 895 struct malloc_type *mtp; 896 897 mtx_assert(&malloc_mtx, MA_OWNED); 898 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 899 if (strcmp(mtp->ks_shortdesc, desc) == 0) 900 return (mtp); 901 } 902 return (NULL); 903} 904 905static int 906sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 907{ 908 struct malloc_type_stream_header mtsh; 909 struct malloc_type_internal *mtip; 910 struct malloc_type_header mth; 911 struct malloc_type *mtp; 912 int error, i; 913 struct sbuf sbuf; 914 915 error = sysctl_wire_old_buffer(req, 0); 916 if (error != 0) 917 return (error); 918 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 919 mtx_lock(&malloc_mtx); 920 921 /* 922 * Insert stream header. 923 */ 924 bzero(&mtsh, sizeof(mtsh)); 925 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 926 mtsh.mtsh_maxcpus = MAXCPU; 927 mtsh.mtsh_count = kmemcount; 928 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 929 930 /* 931 * Insert alternating sequence of type headers and type statistics. 932 */ 933 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 934 mtip = (struct malloc_type_internal *)mtp->ks_handle; 935 936 /* 937 * Insert type header. 938 */ 939 bzero(&mth, sizeof(mth)); 940 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 941 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 942 943 /* 944 * Insert type statistics for each CPU. 945 */ 946 for (i = 0; i < MAXCPU; i++) { 947 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], 948 sizeof(mtip->mti_stats[i])); 949 } 950 } 951 mtx_unlock(&malloc_mtx); 952 error = sbuf_finish(&sbuf); 953 sbuf_delete(&sbuf); 954 return (error); 955} 956 957SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 958 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 959 "Return malloc types"); 960 961SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 962 "Count of kernel malloc types"); 963 964void 965malloc_type_list(malloc_type_list_func_t *func, void *arg) 966{ 967 struct malloc_type *mtp, **bufmtp; 968 int count, i; 969 size_t buflen; 970 971 mtx_lock(&malloc_mtx); 972restart: 973 mtx_assert(&malloc_mtx, MA_OWNED); 974 count = kmemcount; 975 mtx_unlock(&malloc_mtx); 976 977 buflen = sizeof(struct malloc_type *) * count; 978 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 979 980 mtx_lock(&malloc_mtx); 981 982 if (count < kmemcount) { 983 free(bufmtp, M_TEMP); 984 goto restart; 985 } 986 987 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 988 bufmtp[i] = mtp; 989 990 mtx_unlock(&malloc_mtx); 991 992 for (i = 0; i < count; i++) 993 (func)(bufmtp[i], arg); 994 995 free(bufmtp, M_TEMP); 996} 997 998#ifdef DDB 999DB_SHOW_COMMAND(malloc, db_show_malloc) 1000{ 1001 struct malloc_type_internal *mtip; 1002 struct malloc_type *mtp; 1003 uint64_t allocs, frees; 1004 uint64_t alloced, freed; 1005 int i; 1006 1007 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 1008 "Requests"); 1009 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1010 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1011 allocs = 0; 1012 frees = 0; 1013 alloced = 0; 1014 freed = 0; 1015 for (i = 0; i < MAXCPU; i++) { 1016 allocs += mtip->mti_stats[i].mts_numallocs; 1017 frees += mtip->mti_stats[i].mts_numfrees; 1018 alloced += mtip->mti_stats[i].mts_memalloced; 1019 freed += mtip->mti_stats[i].mts_memfreed; 1020 } 1021 db_printf("%18s %12ju %12juK %12ju\n", 1022 mtp->ks_shortdesc, allocs - frees, 1023 (alloced - freed + 1023) / 1024, allocs); 1024 if (db_pager_quit) 1025 break; 1026 } 1027} 1028 1029#if MALLOC_DEBUG_MAXZONES > 1 1030DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1031{ 1032 struct malloc_type_internal *mtip; 1033 struct malloc_type *mtp; 1034 u_int subzone; 1035 1036 if (!have_addr) { 1037 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1038 return; 1039 } 1040 mtp = (void *)addr; 1041 if (mtp->ks_magic != M_MAGIC) { 1042 db_printf("Magic %lx does not match expected %x\n", 1043 mtp->ks_magic, M_MAGIC); 1044 return; 1045 } 1046 1047 mtip = mtp->ks_handle; 1048 subzone = mtip->mti_zone; 1049 1050 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1051 mtip = mtp->ks_handle; 1052 if (mtip->mti_zone != subzone) 1053 continue; 1054 db_printf("%s\n", mtp->ks_shortdesc); 1055 if (db_pager_quit) 1056 break; 1057 } 1058} 1059#endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1060#endif /* DDB */ 1061 1062#ifdef MALLOC_PROFILE 1063 1064static int 1065sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1066{ 1067 struct sbuf sbuf; 1068 uint64_t count; 1069 uint64_t waste; 1070 uint64_t mem; 1071 int error; 1072 int rsize; 1073 int size; 1074 int i; 1075 1076 waste = 0; 1077 mem = 0; 1078 1079 error = sysctl_wire_old_buffer(req, 0); 1080 if (error != 0) 1081 return (error); 1082 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1083 sbuf_printf(&sbuf, 1084 "\n Size Requests Real Size\n"); 1085 for (i = 0; i < KMEM_ZSIZE; i++) { 1086 size = i << KMEM_ZSHIFT; 1087 rsize = kmemzones[kmemsize[i]].kz_size; 1088 count = (long long unsigned)krequests[i]; 1089 1090 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1091 (unsigned long long)count, rsize); 1092 1093 if ((rsize * count) > (size * count)) 1094 waste += (rsize * count) - (size * count); 1095 mem += (rsize * count); 1096 } 1097 sbuf_printf(&sbuf, 1098 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1099 (unsigned long long)mem, (unsigned long long)waste); 1100 error = sbuf_finish(&sbuf); 1101 sbuf_delete(&sbuf); 1102 return (error); 1103} 1104 1105SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 1106 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 1107#endif /* MALLOC_PROFILE */ 1108