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