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 * Copyright (c) 2008 Otto Moerbeek <otto@drijf.net> (mallocarray) 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94 33 */ 34 35/* 36 * Kernel malloc(9) implementation -- general purpose kernel memory allocator 37 * based on memory types. Back end is implemented using the UMA(9) zone 38 * allocator. A set of fixed-size buckets are used for smaller allocations, 39 * and a special UMA allocation interface is used for larger allocations. 40 * Callers declare memory types, and statistics are maintained independently 41 * for each memory type. Statistics are maintained per-CPU for performance 42 * reasons. See malloc(9) and comments in malloc.h for a detailed 43 * description. 44 */ 45 46#include <sys/cdefs.h> 47__FBSDID("$FreeBSD: stable/11/sys/kern/kern_malloc.c 328210 2018-01-20 23:46:03Z kp $"); 48 49#include "opt_ddb.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 114static struct malloc_type *kmemstatistics; 115static int kmemcount; 116 117#define KMEM_ZSHIFT 4 118#define KMEM_ZBASE 16 119#define KMEM_ZMASK (KMEM_ZBASE - 1) 120 121#define KMEM_ZMAX 65536 122#define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 123static uint8_t kmemsize[KMEM_ZSIZE + 1]; 124 125#ifndef MALLOC_DEBUG_MAXZONES 126#define MALLOC_DEBUG_MAXZONES 1 127#endif 128static int numzones = MALLOC_DEBUG_MAXZONES; 129 130/* 131 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 132 * of various sizes. 133 * 134 * XXX: The comment here used to read "These won't be powers of two for 135 * long." It's possible that a significant amount of wasted memory could be 136 * recovered by tuning the sizes of these buckets. 137 */ 138struct { 139 int kz_size; 140 char *kz_name; 141 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 142} kmemzones[] = { 143 {16, "16", }, 144 {32, "32", }, 145 {64, "64", }, 146 {128, "128", }, 147 {256, "256", }, 148 {512, "512", }, 149 {1024, "1024", }, 150 {2048, "2048", }, 151 {4096, "4096", }, 152 {8192, "8192", }, 153 {16384, "16384", }, 154 {32768, "32768", }, 155 {65536, "65536", }, 156 {0, NULL}, 157}; 158 159/* 160 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 161 * types are described by a data structure passed by the declaring code, but 162 * the malloc(9) implementation has its own data structure describing the 163 * type and statistics. This permits the malloc(9)-internal data structures 164 * to be modified without breaking binary-compiled kernel modules that 165 * declare malloc types. 166 */ 167static uma_zone_t mt_zone; 168 169u_long vm_kmem_size; 170SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 171 "Size of kernel memory"); 172 173static u_long kmem_zmax = KMEM_ZMAX; 174SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 175 "Maximum allocation size that malloc(9) would use UMA as backend"); 176 177static u_long vm_kmem_size_min; 178SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 179 "Minimum size of kernel memory"); 180 181static u_long vm_kmem_size_max; 182SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 183 "Maximum size of kernel memory"); 184 185static u_int vm_kmem_size_scale; 186SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 187 "Scale factor for kernel memory size"); 188 189static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 190SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 191 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 192 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 193 194static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 195SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 196 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 197 sysctl_kmem_map_free, "LU", "Free space in kmem"); 198 199/* 200 * The malloc_mtx protects the kmemstatistics linked list. 201 */ 202struct mtx malloc_mtx; 203 204#ifdef MALLOC_PROFILE 205uint64_t krequests[KMEM_ZSIZE + 1]; 206 207static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 208#endif 209 210static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 211 212/* 213 * time_uptime of the last malloc(9) failure (induced or real). 214 */ 215static time_t t_malloc_fail; 216 217#if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 218static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 219 "Kernel malloc debugging options"); 220#endif 221 222/* 223 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 224 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 225 */ 226#ifdef MALLOC_MAKE_FAILURES 227static int malloc_failure_rate; 228static int malloc_nowait_count; 229static int malloc_failure_count; 230SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, 231 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 232SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 233 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 234#endif 235 236static int 237sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 238{ 239 u_long size; 240 241 size = vmem_size(kmem_arena, VMEM_ALLOC); 242 return (sysctl_handle_long(oidp, &size, 0, req)); 243} 244 245static int 246sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 247{ 248 u_long size; 249 250 size = vmem_size(kmem_arena, VMEM_FREE); 251 return (sysctl_handle_long(oidp, &size, 0, req)); 252} 253 254/* 255 * malloc(9) uma zone separation -- sub-page buffer overruns in one 256 * malloc type will affect only a subset of other malloc types. 257 */ 258#if MALLOC_DEBUG_MAXZONES > 1 259static void 260tunable_set_numzones(void) 261{ 262 263 TUNABLE_INT_FETCH("debug.malloc.numzones", 264 &numzones); 265 266 /* Sanity check the number of malloc uma zones. */ 267 if (numzones <= 0) 268 numzones = 1; 269 if (numzones > MALLOC_DEBUG_MAXZONES) 270 numzones = MALLOC_DEBUG_MAXZONES; 271} 272SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 273SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 274 &numzones, 0, "Number of malloc uma subzones"); 275 276/* 277 * Any number that changes regularly is an okay choice for the 278 * offset. Build numbers are pretty good of you have them. 279 */ 280static u_int zone_offset = __FreeBSD_version; 281TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 282SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 283 &zone_offset, 0, "Separate malloc types by examining the " 284 "Nth character in the malloc type short description."); 285 286static u_int 287mtp_get_subzone(const char *desc) 288{ 289 size_t len; 290 u_int val; 291 292 if (desc == NULL || (len = strlen(desc)) == 0) 293 return (0); 294 val = desc[zone_offset % len]; 295 return (val % numzones); 296} 297#elif MALLOC_DEBUG_MAXZONES == 0 298#error "MALLOC_DEBUG_MAXZONES must be positive." 299#else 300static inline u_int 301mtp_get_subzone(const char *desc) 302{ 303 304 return (0); 305} 306#endif /* MALLOC_DEBUG_MAXZONES > 1 */ 307 308int 309malloc_last_fail(void) 310{ 311 312 return (time_uptime - t_malloc_fail); 313} 314 315/* 316 * An allocation has succeeded -- update malloc type statistics for the 317 * amount of bucket size. Occurs within a critical section so that the 318 * thread isn't preempted and doesn't migrate while updating per-PCU 319 * statistics. 320 */ 321static void 322malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 323 int zindx) 324{ 325 struct malloc_type_internal *mtip; 326 struct malloc_type_stats *mtsp; 327 328 critical_enter(); 329 mtip = mtp->ks_handle; 330 mtsp = &mtip->mti_stats[curcpu]; 331 if (size > 0) { 332 mtsp->mts_memalloced += size; 333 mtsp->mts_numallocs++; 334 } 335 if (zindx != -1) 336 mtsp->mts_size |= 1 << zindx; 337 338#ifdef KDTRACE_HOOKS 339 if (dtrace_malloc_probe != NULL) { 340 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 341 if (probe_id != 0) 342 (dtrace_malloc_probe)(probe_id, 343 (uintptr_t) mtp, (uintptr_t) mtip, 344 (uintptr_t) mtsp, size, zindx); 345 } 346#endif 347 348 critical_exit(); 349} 350 351void 352malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 353{ 354 355 if (size > 0) 356 malloc_type_zone_allocated(mtp, size, -1); 357} 358 359/* 360 * A free operation has occurred -- update malloc type statistics for the 361 * amount of the bucket size. Occurs within a critical section so that the 362 * thread isn't preempted and doesn't migrate while updating per-CPU 363 * statistics. 364 */ 365void 366malloc_type_freed(struct malloc_type *mtp, unsigned long size) 367{ 368 struct malloc_type_internal *mtip; 369 struct malloc_type_stats *mtsp; 370 371 critical_enter(); 372 mtip = mtp->ks_handle; 373 mtsp = &mtip->mti_stats[curcpu]; 374 mtsp->mts_memfreed += size; 375 mtsp->mts_numfrees++; 376 377#ifdef KDTRACE_HOOKS 378 if (dtrace_malloc_probe != NULL) { 379 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 380 if (probe_id != 0) 381 (dtrace_malloc_probe)(probe_id, 382 (uintptr_t) mtp, (uintptr_t) mtip, 383 (uintptr_t) mtsp, size, 0); 384 } 385#endif 386 387 critical_exit(); 388} 389 390/* 391 * contigmalloc: 392 * 393 * Allocate a block of physically contiguous memory. 394 * 395 * If M_NOWAIT is set, this routine will not block and return NULL if 396 * the allocation fails. 397 */ 398void * 399contigmalloc(unsigned long size, struct malloc_type *type, int flags, 400 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 401 vm_paddr_t boundary) 402{ 403 void *ret; 404 405 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high, 406 alignment, boundary, VM_MEMATTR_DEFAULT); 407 if (ret != NULL) 408 malloc_type_allocated(type, round_page(size)); 409 return (ret); 410} 411 412/* 413 * contigfree: 414 * 415 * Free a block of memory allocated by contigmalloc. 416 * 417 * This routine may not block. 418 */ 419void 420contigfree(void *addr, unsigned long size, struct malloc_type *type) 421{ 422 423 kmem_free(kernel_arena, (vm_offset_t)addr, size); 424 malloc_type_freed(type, round_page(size)); 425} 426 427/* 428 * malloc: 429 * 430 * Allocate a block of memory. 431 * 432 * If M_NOWAIT is set, this routine will not block and return NULL if 433 * the allocation fails. 434 */ 435void * 436malloc(unsigned long size, struct malloc_type *mtp, int flags) 437{ 438 int indx; 439 struct malloc_type_internal *mtip; 440 caddr_t va; 441 uma_zone_t zone; 442#if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE) 443 unsigned long osize = size; 444#endif 445 446#ifdef INVARIANTS 447 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 448 /* 449 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 450 */ 451 indx = flags & (M_WAITOK | M_NOWAIT); 452 if (indx != M_NOWAIT && indx != M_WAITOK) { 453 static struct timeval lasterr; 454 static int curerr, once; 455 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 456 printf("Bad malloc flags: %x\n", indx); 457 kdb_backtrace(); 458 flags |= M_WAITOK; 459 once++; 460 } 461 } 462#endif 463#ifdef MALLOC_MAKE_FAILURES 464 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 465 atomic_add_int(&malloc_nowait_count, 1); 466 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 467 atomic_add_int(&malloc_failure_count, 1); 468 t_malloc_fail = time_uptime; 469 return (NULL); 470 } 471 } 472#endif 473 if (flags & M_WAITOK) 474 KASSERT(curthread->td_intr_nesting_level == 0, 475 ("malloc(M_WAITOK) in interrupt context")); 476 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 477 ("malloc: called with spinlock or critical section held")); 478 479#ifdef DEBUG_MEMGUARD 480 if (memguard_cmp_mtp(mtp, size)) { 481 va = memguard_alloc(size, flags); 482 if (va != NULL) 483 return (va); 484 /* This is unfortunate but should not be fatal. */ 485 } 486#endif 487 488#ifdef DEBUG_REDZONE 489 size = redzone_size_ntor(size); 490#endif 491 492 if (size <= kmem_zmax) { 493 mtip = mtp->ks_handle; 494 if (size & KMEM_ZMASK) 495 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 496 indx = kmemsize[size >> KMEM_ZSHIFT]; 497 KASSERT(mtip->mti_zone < numzones, 498 ("mti_zone %u out of range %d", 499 mtip->mti_zone, numzones)); 500 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 501#ifdef MALLOC_PROFILE 502 krequests[size >> KMEM_ZSHIFT]++; 503#endif 504 va = uma_zalloc(zone, flags); 505 if (va != NULL) 506 size = zone->uz_size; 507 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 508 } else { 509 size = roundup(size, PAGE_SIZE); 510 zone = NULL; 511 va = uma_large_malloc(size, flags); 512 malloc_type_allocated(mtp, va == NULL ? 0 : size); 513 } 514 if (flags & M_WAITOK) 515 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 516 else if (va == NULL) 517 t_malloc_fail = time_uptime; 518#ifdef DIAGNOSTIC 519 if (va != NULL && !(flags & M_ZERO)) { 520 memset(va, 0x70, osize); 521 } 522#endif 523#ifdef DEBUG_REDZONE 524 if (va != NULL) 525 va = redzone_setup(va, osize); 526#endif 527 return ((void *) va); 528} 529 530void * 531mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags) 532{ 533 534 if (WOULD_OVERFLOW(nmemb, size)) 535 panic("mallocarray: %zu * %zu overflowed", nmemb, size); 536 537 return (malloc(size * nmemb, type, flags)); 538} 539 540/* 541 * free: 542 * 543 * Free a block of memory allocated by malloc. 544 * 545 * This routine may not block. 546 */ 547void 548free(void *addr, struct malloc_type *mtp) 549{ 550 uma_slab_t slab; 551 u_long size; 552 553 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); 554 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 555 ("free: called with spinlock or critical section held")); 556 557 /* free(NULL, ...) does nothing */ 558 if (addr == NULL) 559 return; 560 561#ifdef DEBUG_MEMGUARD 562 if (is_memguard_addr(addr)) { 563 memguard_free(addr); 564 return; 565 } 566#endif 567 568#ifdef DEBUG_REDZONE 569 redzone_check(addr); 570 addr = redzone_addr_ntor(addr); 571#endif 572 573 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 574 575 if (slab == NULL) 576 panic("free: address %p(%p) has not been allocated.\n", 577 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 578 579 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 580#ifdef INVARIANTS 581 struct malloc_type **mtpp = addr; 582#endif 583 size = slab->us_keg->uk_size; 584#ifdef INVARIANTS 585 /* 586 * Cache a pointer to the malloc_type that most recently freed 587 * this memory here. This way we know who is most likely to 588 * have stepped on it later. 589 * 590 * This code assumes that size is a multiple of 8 bytes for 591 * 64 bit machines 592 */ 593 mtpp = (struct malloc_type **) 594 ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 595 mtpp += (size - sizeof(struct malloc_type *)) / 596 sizeof(struct malloc_type *); 597 *mtpp = mtp; 598#endif 599 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 600 } else { 601 size = slab->us_size; 602 uma_large_free(slab); 603 } 604 malloc_type_freed(mtp, size); 605} 606 607/* 608 * realloc: change the size of a memory block 609 */ 610void * 611realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 612{ 613 uma_slab_t slab; 614 unsigned long alloc; 615 void *newaddr; 616 617 KASSERT(mtp->ks_magic == M_MAGIC, 618 ("realloc: bad malloc type magic")); 619 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 620 ("realloc: called with spinlock or critical section held")); 621 622 /* realloc(NULL, ...) is equivalent to malloc(...) */ 623 if (addr == NULL) 624 return (malloc(size, mtp, flags)); 625 626 /* 627 * XXX: Should report free of old memory and alloc of new memory to 628 * per-CPU stats. 629 */ 630 631#ifdef DEBUG_MEMGUARD 632 if (is_memguard_addr(addr)) 633 return (memguard_realloc(addr, size, mtp, flags)); 634#endif 635 636#ifdef DEBUG_REDZONE 637 slab = NULL; 638 alloc = redzone_get_size(addr); 639#else 640 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 641 642 /* Sanity check */ 643 KASSERT(slab != NULL, 644 ("realloc: address %p out of range", (void *)addr)); 645 646 /* Get the size of the original block */ 647 if (!(slab->us_flags & UMA_SLAB_MALLOC)) 648 alloc = slab->us_keg->uk_size; 649 else 650 alloc = slab->us_size; 651 652 /* Reuse the original block if appropriate */ 653 if (size <= alloc 654 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 655 return (addr); 656#endif /* !DEBUG_REDZONE */ 657 658 /* Allocate a new, bigger (or smaller) block */ 659 if ((newaddr = malloc(size, mtp, flags)) == NULL) 660 return (NULL); 661 662 /* Copy over original contents */ 663 bcopy(addr, newaddr, min(size, alloc)); 664 free(addr, mtp); 665 return (newaddr); 666} 667 668/* 669 * reallocf: same as realloc() but free memory on failure. 670 */ 671void * 672reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 673{ 674 void *mem; 675 676 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 677 free(addr, mtp); 678 return (mem); 679} 680 681/* 682 * Wake the uma reclamation pagedaemon thread when we exhaust KVA. It 683 * will call the lowmem handler and uma_reclaim() callbacks in a 684 * context that is safe. 685 */ 686static void 687kmem_reclaim(vmem_t *vm, int flags) 688{ 689 690 uma_reclaim_wakeup(); 691 pagedaemon_wakeup(); 692} 693 694#ifndef __sparc64__ 695CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 696#endif 697 698/* 699 * Initialize the kernel memory (kmem) arena. 700 */ 701void 702kmeminit(void) 703{ 704 u_long mem_size; 705 u_long tmp; 706 707#ifdef VM_KMEM_SIZE 708 if (vm_kmem_size == 0) 709 vm_kmem_size = VM_KMEM_SIZE; 710#endif 711#ifdef VM_KMEM_SIZE_MIN 712 if (vm_kmem_size_min == 0) 713 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 714#endif 715#ifdef VM_KMEM_SIZE_MAX 716 if (vm_kmem_size_max == 0) 717 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 718#endif 719 /* 720 * Calculate the amount of kernel virtual address (KVA) space that is 721 * preallocated to the kmem arena. In order to support a wide range 722 * of machines, it is a function of the physical memory size, 723 * specifically, 724 * 725 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 726 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 727 * 728 * Every architecture must define an integral value for 729 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 730 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 731 * ceiling on this preallocation, are optional. Typically, 732 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 733 * a given architecture. 734 */ 735 mem_size = vm_cnt.v_page_count; 736 if (mem_size <= 32768) /* delphij XXX 128MB */ 737 kmem_zmax = PAGE_SIZE; 738 739 if (vm_kmem_size_scale < 1) 740 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 741 742 /* 743 * Check if we should use defaults for the "vm_kmem_size" 744 * variable: 745 */ 746 if (vm_kmem_size == 0) { 747 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 748 749 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 750 vm_kmem_size = vm_kmem_size_min; 751 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 752 vm_kmem_size = vm_kmem_size_max; 753 } 754 755 /* 756 * The amount of KVA space that is preallocated to the 757 * kmem arena can be set statically at compile-time or manually 758 * through the kernel environment. However, it is still limited to 759 * twice the physical memory size, which has been sufficient to handle 760 * the most severe cases of external fragmentation in the kmem arena. 761 */ 762 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 763 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 764 765 vm_kmem_size = round_page(vm_kmem_size); 766#ifdef DEBUG_MEMGUARD 767 tmp = memguard_fudge(vm_kmem_size, kernel_map); 768#else 769 tmp = vm_kmem_size; 770#endif 771 vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE, 772 0, 0); 773 vmem_set_reclaim(kmem_arena, kmem_reclaim); 774 775#ifdef DEBUG_MEMGUARD 776 /* 777 * Initialize MemGuard if support compiled in. MemGuard is a 778 * replacement allocator used for detecting tamper-after-free 779 * scenarios as they occur. It is only used for debugging. 780 */ 781 memguard_init(kmem_arena); 782#endif 783} 784 785/* 786 * Initialize the kernel memory allocator 787 */ 788/* ARGSUSED*/ 789static void 790mallocinit(void *dummy) 791{ 792 int i; 793 uint8_t indx; 794 795 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 796 797 kmeminit(); 798 799 uma_startup2(); 800 801 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 802 kmem_zmax = KMEM_ZMAX; 803 804 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 805#ifdef INVARIANTS 806 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 807#else 808 NULL, NULL, NULL, NULL, 809#endif 810 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 811 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 812 int size = kmemzones[indx].kz_size; 813 char *name = kmemzones[indx].kz_name; 814 int subzone; 815 816 for (subzone = 0; subzone < numzones; subzone++) { 817 kmemzones[indx].kz_zone[subzone] = 818 uma_zcreate(name, size, 819#ifdef INVARIANTS 820 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 821#else 822 NULL, NULL, NULL, NULL, 823#endif 824 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 825 } 826 for (;i <= size; i+= KMEM_ZBASE) 827 kmemsize[i >> KMEM_ZSHIFT] = indx; 828 829 } 830} 831SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); 832 833void 834malloc_init(void *data) 835{ 836 struct malloc_type_internal *mtip; 837 struct malloc_type *mtp; 838 839 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init")); 840 841 mtp = data; 842 if (mtp->ks_magic != M_MAGIC) 843 panic("malloc_init: bad malloc type magic"); 844 845 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 846 mtp->ks_handle = mtip; 847 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); 848 849 mtx_lock(&malloc_mtx); 850 mtp->ks_next = kmemstatistics; 851 kmemstatistics = mtp; 852 kmemcount++; 853 mtx_unlock(&malloc_mtx); 854} 855 856void 857malloc_uninit(void *data) 858{ 859 struct malloc_type_internal *mtip; 860 struct malloc_type_stats *mtsp; 861 struct malloc_type *mtp, *temp; 862 uma_slab_t slab; 863 long temp_allocs, temp_bytes; 864 int i; 865 866 mtp = data; 867 KASSERT(mtp->ks_magic == M_MAGIC, 868 ("malloc_uninit: bad malloc type magic")); 869 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 870 871 mtx_lock(&malloc_mtx); 872 mtip = mtp->ks_handle; 873 mtp->ks_handle = NULL; 874 if (mtp != kmemstatistics) { 875 for (temp = kmemstatistics; temp != NULL; 876 temp = temp->ks_next) { 877 if (temp->ks_next == mtp) { 878 temp->ks_next = mtp->ks_next; 879 break; 880 } 881 } 882 KASSERT(temp, 883 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 884 } else 885 kmemstatistics = mtp->ks_next; 886 kmemcount--; 887 mtx_unlock(&malloc_mtx); 888 889 /* 890 * Look for memory leaks. 891 */ 892 temp_allocs = temp_bytes = 0; 893 for (i = 0; i < MAXCPU; i++) { 894 mtsp = &mtip->mti_stats[i]; 895 temp_allocs += mtsp->mts_numallocs; 896 temp_allocs -= mtsp->mts_numfrees; 897 temp_bytes += mtsp->mts_memalloced; 898 temp_bytes -= mtsp->mts_memfreed; 899 } 900 if (temp_allocs > 0 || temp_bytes > 0) { 901 printf("Warning: memory type %s leaked memory on destroy " 902 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 903 temp_allocs, temp_bytes); 904 } 905 906 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 907 uma_zfree_arg(mt_zone, mtip, slab); 908} 909 910struct malloc_type * 911malloc_desc2type(const char *desc) 912{ 913 struct malloc_type *mtp; 914 915 mtx_assert(&malloc_mtx, MA_OWNED); 916 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 917 if (strcmp(mtp->ks_shortdesc, desc) == 0) 918 return (mtp); 919 } 920 return (NULL); 921} 922 923static int 924sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 925{ 926 struct malloc_type_stream_header mtsh; 927 struct malloc_type_internal *mtip; 928 struct malloc_type_header mth; 929 struct malloc_type *mtp; 930 int error, i; 931 struct sbuf sbuf; 932 933 error = sysctl_wire_old_buffer(req, 0); 934 if (error != 0) 935 return (error); 936 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 937 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); 938 mtx_lock(&malloc_mtx); 939 940 /* 941 * Insert stream header. 942 */ 943 bzero(&mtsh, sizeof(mtsh)); 944 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 945 mtsh.mtsh_maxcpus = MAXCPU; 946 mtsh.mtsh_count = kmemcount; 947 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 948 949 /* 950 * Insert alternating sequence of type headers and type statistics. 951 */ 952 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 953 mtip = (struct malloc_type_internal *)mtp->ks_handle; 954 955 /* 956 * Insert type header. 957 */ 958 bzero(&mth, sizeof(mth)); 959 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 960 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 961 962 /* 963 * Insert type statistics for each CPU. 964 */ 965 for (i = 0; i < MAXCPU; i++) { 966 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], 967 sizeof(mtip->mti_stats[i])); 968 } 969 } 970 mtx_unlock(&malloc_mtx); 971 error = sbuf_finish(&sbuf); 972 sbuf_delete(&sbuf); 973 return (error); 974} 975 976SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 977 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 978 "Return malloc types"); 979 980SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 981 "Count of kernel malloc types"); 982 983void 984malloc_type_list(malloc_type_list_func_t *func, void *arg) 985{ 986 struct malloc_type *mtp, **bufmtp; 987 int count, i; 988 size_t buflen; 989 990 mtx_lock(&malloc_mtx); 991restart: 992 mtx_assert(&malloc_mtx, MA_OWNED); 993 count = kmemcount; 994 mtx_unlock(&malloc_mtx); 995 996 buflen = sizeof(struct malloc_type *) * count; 997 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 998 999 mtx_lock(&malloc_mtx); 1000 1001 if (count < kmemcount) { 1002 free(bufmtp, M_TEMP); 1003 goto restart; 1004 } 1005 1006 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 1007 bufmtp[i] = mtp; 1008 1009 mtx_unlock(&malloc_mtx); 1010 1011 for (i = 0; i < count; i++) 1012 (func)(bufmtp[i], arg); 1013 1014 free(bufmtp, M_TEMP); 1015} 1016 1017#ifdef DDB 1018DB_SHOW_COMMAND(malloc, db_show_malloc) 1019{ 1020 struct malloc_type_internal *mtip; 1021 struct malloc_type *mtp; 1022 uint64_t allocs, frees; 1023 uint64_t alloced, freed; 1024 int i; 1025 1026 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 1027 "Requests"); 1028 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1029 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1030 allocs = 0; 1031 frees = 0; 1032 alloced = 0; 1033 freed = 0; 1034 for (i = 0; i < MAXCPU; i++) { 1035 allocs += mtip->mti_stats[i].mts_numallocs; 1036 frees += mtip->mti_stats[i].mts_numfrees; 1037 alloced += mtip->mti_stats[i].mts_memalloced; 1038 freed += mtip->mti_stats[i].mts_memfreed; 1039 } 1040 db_printf("%18s %12ju %12juK %12ju\n", 1041 mtp->ks_shortdesc, allocs - frees, 1042 (alloced - freed + 1023) / 1024, allocs); 1043 if (db_pager_quit) 1044 break; 1045 } 1046} 1047 1048#if MALLOC_DEBUG_MAXZONES > 1 1049DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1050{ 1051 struct malloc_type_internal *mtip; 1052 struct malloc_type *mtp; 1053 u_int subzone; 1054 1055 if (!have_addr) { 1056 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1057 return; 1058 } 1059 mtp = (void *)addr; 1060 if (mtp->ks_magic != M_MAGIC) { 1061 db_printf("Magic %lx does not match expected %x\n", 1062 mtp->ks_magic, M_MAGIC); 1063 return; 1064 } 1065 1066 mtip = mtp->ks_handle; 1067 subzone = mtip->mti_zone; 1068 1069 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1070 mtip = mtp->ks_handle; 1071 if (mtip->mti_zone != subzone) 1072 continue; 1073 db_printf("%s\n", mtp->ks_shortdesc); 1074 if (db_pager_quit) 1075 break; 1076 } 1077} 1078#endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1079#endif /* DDB */ 1080 1081#ifdef MALLOC_PROFILE 1082 1083static int 1084sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1085{ 1086 struct sbuf sbuf; 1087 uint64_t count; 1088 uint64_t waste; 1089 uint64_t mem; 1090 int error; 1091 int rsize; 1092 int size; 1093 int i; 1094 1095 waste = 0; 1096 mem = 0; 1097 1098 error = sysctl_wire_old_buffer(req, 0); 1099 if (error != 0) 1100 return (error); 1101 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1102 sbuf_printf(&sbuf, 1103 "\n Size Requests Real Size\n"); 1104 for (i = 0; i < KMEM_ZSIZE; i++) { 1105 size = i << KMEM_ZSHIFT; 1106 rsize = kmemzones[kmemsize[i]].kz_size; 1107 count = (long long unsigned)krequests[i]; 1108 1109 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1110 (unsigned long long)count, rsize); 1111 1112 if ((rsize * count) > (size * count)) 1113 waste += (rsize * count) - (size * count); 1114 mem += (rsize * count); 1115 } 1116 sbuf_printf(&sbuf, 1117 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1118 (unsigned long long)mem, (unsigned long long)waste); 1119 error = sbuf_finish(&sbuf); 1120 sbuf_delete(&sbuf); 1121 return (error); 1122} 1123 1124SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 1125 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 1126#endif /* MALLOC_PROFILE */ 1127