1 /*- 2 * Copyright (c) 1987, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2005-2006 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94 32 */ 33 34 /* 35 * Kernel malloc(9) implementation -- general purpose kernel memory allocator 36 * based on memory types. Back end is implemented using the UMA(9) zone 37 * allocator. A set of fixed-size buckets are used for smaller allocations, 38 * and a special UMA allocation interface is used for larger allocations. 39 * Callers declare memory types, and statistics are maintained independently 40 * for each memory type. Statistics are maintained per-CPU for performance 41 * reasons. See malloc(9) and comments in malloc.h for a detailed 42 * description. 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include "opt_ddb.h" 49 #include "opt_vm.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/kdb.h> 54 #include <sys/kernel.h> 55 #include <sys/lock.h> 56 #include <sys/malloc.h> 57 #include <sys/mbuf.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 65 #include <vm/vm.h> 66 #include <vm/pmap.h> 67 #include <vm/vm_param.h> 68 #include <vm/vm_kern.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_map.h> 71 #include <vm/vm_page.h> 72 #include <vm/uma.h> 73 #include <vm/uma_int.h> 74 #include <vm/uma_dbg.h> 75 76 #ifdef DEBUG_MEMGUARD 77 #include <vm/memguard.h> 78 #endif 79 #ifdef DEBUG_REDZONE 80 #include <vm/redzone.h> 81 #endif 82 83 #if defined(INVARIANTS) && defined(__i386__) 84 #include <machine/cpu.h> 85 #endif 86 87 #include <ddb/ddb.h> 88 89 /* 90 * When realloc() is called, if the new size is sufficiently smaller than 91 * the old size, realloc() will allocate a new, smaller block to avoid 92 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 93 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 94 */ 95 #ifndef REALLOC_FRACTION 96 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 97 #endif 98 99 /* 100 * Centrally define some common malloc types. 101 */ 102 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 103 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 104 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 105 106 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); 107 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); 108 109 static void kmeminit(void *); 110 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL) 111 112 static MALLOC_DEFINE(M_FREE, "free", "should be on free list"); 113 114 static struct malloc_type *kmemstatistics; 115 static char *kmembase; 116 static char *kmemlimit; 117 static int kmemcount; 118 119 #define KMEM_ZSHIFT 4 120 #define KMEM_ZBASE 16 121 #define KMEM_ZMASK (KMEM_ZBASE - 1) 122 123 #define KMEM_ZMAX PAGE_SIZE 124 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 125 static u_int8_t kmemsize[KMEM_ZSIZE + 1]; 126 127 /* 128 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 129 * of various sizes. 130 * 131 * XXX: The comment here used to read "These won't be powers of two for 132 * long." It's possible that a significant amount of wasted memory could be 133 * recovered by tuning the sizes of these buckets. 134 */ 135 struct { 136 int kz_size; 137 char *kz_name; 138 uma_zone_t kz_zone; 139 } kmemzones[] = { 140 {16, "16", NULL}, 141 {32, "32", NULL}, 142 {64, "64", NULL}, 143 {128, "128", NULL}, 144 {256, "256", NULL}, 145 {512, "512", NULL}, 146 {1024, "1024", NULL}, 147 {2048, "2048", NULL}, 148 {4096, "4096", NULL}, 149 #if PAGE_SIZE > 4096 150 {8192, "8192", NULL}, 151 #if PAGE_SIZE > 8192 152 {16384, "16384", NULL}, 153 #if PAGE_SIZE > 16384 154 {32768, "32768", NULL}, 155 #if PAGE_SIZE > 32768 156 {65536, "65536", NULL}, 157 #if PAGE_SIZE > 65536 158 #error "Unsupported PAGE_SIZE" 159 #endif /* 65536 */ 160 #endif /* 32768 */ 161 #endif /* 16384 */ 162 #endif /* 8192 */ 163 #endif /* 4096 */ 164 {0, NULL}, 165 }; 166 167 /* 168 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 169 * types are described by a data structure passed by the declaring code, but 170 * the malloc(9) implementation has its own data structure describing the 171 * type and statistics. This permits the malloc(9)-internal data structures 172 * to be modified without breaking binary-compiled kernel modules that 173 * declare malloc types. 174 */ 175 static uma_zone_t mt_zone; 176 177 u_int vm_kmem_size; 178 SYSCTL_UINT(_vm, OID_AUTO, kmem_size, CTLFLAG_RD, &vm_kmem_size, 0, 179 "Size of kernel memory"); 180 181 u_int vm_kmem_size_max; 182 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RD, &vm_kmem_size_max, 0, 183 "Maximum size of kernel memory"); 184 185 u_int vm_kmem_size_scale; 186 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RD, &vm_kmem_size_scale, 0, 187 "Scale factor for kernel memory size"); 188 189 /* 190 * The malloc_mtx protects the kmemstatistics linked list. 191 */ 192 struct mtx malloc_mtx; 193 194 #ifdef MALLOC_PROFILE 195 uint64_t krequests[KMEM_ZSIZE + 1]; 196 197 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 198 #endif 199 200 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 201 202 /* 203 * time_uptime of the last malloc(9) failure (induced or real). 204 */ 205 static time_t t_malloc_fail; 206 207 /* 208 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 209 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 210 */ 211 #ifdef MALLOC_MAKE_FAILURES 212 SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 213 "Kernel malloc debugging options"); 214 215 static int malloc_failure_rate; 216 static int malloc_nowait_count; 217 static int malloc_failure_count; 218 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW, 219 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 220 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate); 221 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 222 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 223 #endif 224 225 int 226 malloc_last_fail(void) 227 { 228 229 return (time_uptime - t_malloc_fail); 230 } 231 232 /* 233 * An allocation has succeeded -- update malloc type statistics for the 234 * amount of bucket size. Occurs within a critical section so that the 235 * thread isn't preempted and doesn't migrate while updating per-PCU 236 * statistics. 237 */ 238 static void 239 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 240 int zindx) 241 { 242 struct malloc_type_internal *mtip; 243 struct malloc_type_stats *mtsp; 244 245 critical_enter(); 246 mtip = mtp->ks_handle; 247 mtsp = &mtip->mti_stats[curcpu]; 248 if (size > 0) { 249 mtsp->mts_memalloced += size; 250 mtsp->mts_numallocs++; 251 } 252 if (zindx != -1) 253 mtsp->mts_size |= 1 << zindx; 254 critical_exit(); 255 } 256 257 void 258 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 259 { 260 261 if (size > 0) 262 malloc_type_zone_allocated(mtp, size, -1); 263 } 264 265 /* 266 * A free operation has occurred -- update malloc type statistis for the 267 * amount of the bucket size. Occurs within a critical section so that the 268 * thread isn't preempted and doesn't migrate while updating per-CPU 269 * statistics. 270 */ 271 void 272 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 273 { 274 struct malloc_type_internal *mtip; 275 struct malloc_type_stats *mtsp; 276 277 critical_enter(); 278 mtip = mtp->ks_handle; 279 mtsp = &mtip->mti_stats[curcpu]; 280 mtsp->mts_memfreed += size; 281 mtsp->mts_numfrees++; 282 critical_exit(); 283 } 284 285 /* 286 * malloc: 287 * 288 * Allocate a block of memory. 289 * 290 * If M_NOWAIT is set, this routine will not block and return NULL if 291 * the allocation fails. 292 */ 293 void * 294 malloc(unsigned long size, struct malloc_type *mtp, int flags) 295 { 296 int indx; 297 caddr_t va; 298 uma_zone_t zone; 299 uma_keg_t keg; 300 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE) 301 unsigned long osize = size; 302 #endif 303 304 #ifdef INVARIANTS 305 /* 306 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 307 */ 308 indx = flags & (M_WAITOK | M_NOWAIT); 309 if (indx != M_NOWAIT && indx != M_WAITOK) { 310 static struct timeval lasterr; 311 static int curerr, once; 312 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 313 printf("Bad malloc flags: %x\n", indx); 314 kdb_backtrace(); 315 flags |= M_WAITOK; 316 once++; 317 } 318 } 319 #endif 320 #if 0 321 if (size == 0) 322 kdb_enter("zero size malloc"); 323 #endif 324 #ifdef MALLOC_MAKE_FAILURES 325 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 326 atomic_add_int(&malloc_nowait_count, 1); 327 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 328 atomic_add_int(&malloc_failure_count, 1); 329 t_malloc_fail = time_uptime; 330 return (NULL); 331 } 332 } 333 #endif 334 if (flags & M_WAITOK) 335 KASSERT(curthread->td_intr_nesting_level == 0, 336 ("malloc(M_WAITOK) in interrupt context")); 337 338 #ifdef DEBUG_MEMGUARD 339 if (memguard_cmp(mtp)) 340 return memguard_alloc(size, flags); 341 #endif 342 343 #ifdef DEBUG_REDZONE 344 size = redzone_size_ntor(size); 345 #endif 346 347 if (size <= KMEM_ZMAX) { 348 if (size & KMEM_ZMASK) 349 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 350 indx = kmemsize[size >> KMEM_ZSHIFT]; 351 zone = kmemzones[indx].kz_zone; 352 keg = zone->uz_keg; 353 #ifdef MALLOC_PROFILE 354 krequests[size >> KMEM_ZSHIFT]++; 355 #endif 356 va = uma_zalloc(zone, flags); 357 if (va != NULL) 358 size = keg->uk_size; 359 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 360 } else { 361 size = roundup(size, PAGE_SIZE); 362 zone = NULL; 363 keg = NULL; 364 va = uma_large_malloc(size, flags); 365 malloc_type_allocated(mtp, va == NULL ? 0 : size); 366 } 367 if (flags & M_WAITOK) 368 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 369 else if (va == NULL) 370 t_malloc_fail = time_uptime; 371 #ifdef DIAGNOSTIC 372 if (va != NULL && !(flags & M_ZERO)) { 373 memset(va, 0x70, osize); 374 } 375 #endif 376 #ifdef DEBUG_REDZONE 377 if (va != NULL) 378 va = redzone_setup(va, osize); 379 #endif 380 return ((void *) va); 381 } 382 383 /* 384 * free: 385 * 386 * Free a block of memory allocated by malloc. 387 * 388 * This routine may not block. 389 */ 390 void 391 free(void *addr, struct malloc_type *mtp) 392 { 393 uma_slab_t slab; 394 u_long size; 395 396 /* free(NULL, ...) does nothing */ 397 if (addr == NULL) 398 return; 399 400 #ifdef DEBUG_MEMGUARD 401 if (memguard_cmp(mtp)) { 402 memguard_free(addr); 403 return; 404 } 405 #endif 406 407 #ifdef DEBUG_REDZONE 408 redzone_check(addr); 409 addr = redzone_addr_ntor(addr); 410 #endif 411 412 size = 0; 413 414 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 415 416 if (slab == NULL) 417 panic("free: address %p(%p) has not been allocated.\n", 418 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 419 420 421 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 422 #ifdef INVARIANTS 423 struct malloc_type **mtpp = addr; 424 #endif 425 size = slab->us_keg->uk_size; 426 #ifdef INVARIANTS 427 /* 428 * Cache a pointer to the malloc_type that most recently freed 429 * this memory here. This way we know who is most likely to 430 * have stepped on it later. 431 * 432 * This code assumes that size is a multiple of 8 bytes for 433 * 64 bit machines 434 */ 435 mtpp = (struct malloc_type **) 436 ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 437 mtpp += (size - sizeof(struct malloc_type *)) / 438 sizeof(struct malloc_type *); 439 *mtpp = mtp; 440 #endif 441 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 442 } else { 443 size = slab->us_size; 444 uma_large_free(slab); 445 } 446 malloc_type_freed(mtp, size); 447 } 448 449 /* 450 * realloc: change the size of a memory block 451 */ 452 void * 453 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 454 { 455 uma_slab_t slab; 456 unsigned long alloc; 457 void *newaddr; 458 459 /* realloc(NULL, ...) is equivalent to malloc(...) */ 460 if (addr == NULL) 461 return (malloc(size, mtp, flags)); 462 463 /* 464 * XXX: Should report free of old memory and alloc of new memory to 465 * per-CPU stats. 466 */ 467 468 #ifdef DEBUG_MEMGUARD 469 if (memguard_cmp(mtp)) { 470 slab = NULL; 471 alloc = size; 472 } else { 473 #endif 474 475 #ifdef DEBUG_REDZONE 476 slab = NULL; 477 alloc = redzone_get_size(addr); 478 #else 479 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 480 481 /* Sanity check */ 482 KASSERT(slab != NULL, 483 ("realloc: address %p out of range", (void *)addr)); 484 485 /* Get the size of the original block */ 486 if (!(slab->us_flags & UMA_SLAB_MALLOC)) 487 alloc = slab->us_keg->uk_size; 488 else 489 alloc = slab->us_size; 490 491 /* Reuse the original block if appropriate */ 492 if (size <= alloc 493 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 494 return (addr); 495 #endif /* !DEBUG_REDZONE */ 496 497 #ifdef DEBUG_MEMGUARD 498 } 499 #endif 500 501 /* Allocate a new, bigger (or smaller) block */ 502 if ((newaddr = malloc(size, mtp, flags)) == NULL) 503 return (NULL); 504 505 /* Copy over original contents */ 506 bcopy(addr, newaddr, min(size, alloc)); 507 free(addr, mtp); 508 return (newaddr); 509 } 510 511 /* 512 * reallocf: same as realloc() but free memory on failure. 513 */ 514 void * 515 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 516 { 517 void *mem; 518 519 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 520 free(addr, mtp); 521 return (mem); 522 } 523 524 /* 525 * Initialize the kernel memory allocator 526 */ 527 /* ARGSUSED*/ 528 static void 529 kmeminit(void *dummy) 530 { 531 u_int8_t indx; 532 u_long mem_size; 533 int i; 534 535 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 536 537 /* 538 * Try to auto-tune the kernel memory size, so that it is 539 * more applicable for a wider range of machine sizes. 540 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while 541 * a VM_KMEM_SIZE of 12MB is a fair compromise. The 542 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space 543 * available, and on an X86 with a total KVA space of 256MB, 544 * try to keep VM_KMEM_SIZE_MAX at 80MB or below. 545 * 546 * Note that the kmem_map is also used by the zone allocator, 547 * so make sure that there is enough space. 548 */ 549 vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE; 550 mem_size = cnt.v_page_count; 551 552 #if defined(VM_KMEM_SIZE_SCALE) 553 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 554 #endif 555 TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale); 556 if (vm_kmem_size_scale > 0 && 557 (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE)) 558 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 559 560 #if defined(VM_KMEM_SIZE_MAX) 561 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 562 #endif 563 TUNABLE_INT_FETCH("vm.kmem_size_max", &vm_kmem_size_max); 564 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 565 vm_kmem_size = vm_kmem_size_max; 566 567 /* Allow final override from the kernel environment */ 568 #ifndef BURN_BRIDGES 569 if (TUNABLE_INT_FETCH("kern.vm.kmem.size", &vm_kmem_size) != 0) 570 printf("kern.vm.kmem.size is now called vm.kmem_size!\n"); 571 #endif 572 TUNABLE_INT_FETCH("vm.kmem_size", &vm_kmem_size); 573 574 /* 575 * Limit kmem virtual size to twice the physical memory. 576 * This allows for kmem map sparseness, but limits the size 577 * to something sane. Be careful to not overflow the 32bit 578 * ints while doing the check. 579 */ 580 if (((vm_kmem_size / 2) / PAGE_SIZE) > cnt.v_page_count) 581 vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE; 582 583 /* 584 * Tune settings based on the kernel map's size at this time. 585 */ 586 init_param3(vm_kmem_size / PAGE_SIZE); 587 588 kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase, 589 (vm_offset_t *)&kmemlimit, vm_kmem_size); 590 kmem_map->system_map = 1; 591 592 #ifdef DEBUG_MEMGUARD 593 /* 594 * Initialize MemGuard if support compiled in. MemGuard is a 595 * replacement allocator used for detecting tamper-after-free 596 * scenarios as they occur. It is only used for debugging. 597 */ 598 vm_memguard_divisor = 10; 599 TUNABLE_INT_FETCH("vm.memguard.divisor", &vm_memguard_divisor); 600 601 /* Pick a conservative value if provided value sucks. */ 602 if ((vm_memguard_divisor <= 0) || 603 ((vm_kmem_size / vm_memguard_divisor) == 0)) 604 vm_memguard_divisor = 10; 605 memguard_init(kmem_map, vm_kmem_size / vm_memguard_divisor); 606 #endif 607 608 uma_startup2(); 609 610 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 611 #ifdef INVARIANTS 612 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 613 #else 614 NULL, NULL, NULL, NULL, 615 #endif 616 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 617 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 618 int size = kmemzones[indx].kz_size; 619 char *name = kmemzones[indx].kz_name; 620 621 kmemzones[indx].kz_zone = uma_zcreate(name, size, 622 #ifdef INVARIANTS 623 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 624 #else 625 NULL, NULL, NULL, NULL, 626 #endif 627 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 628 629 for (;i <= size; i+= KMEM_ZBASE) 630 kmemsize[i >> KMEM_ZSHIFT] = indx; 631 632 } 633 } 634 635 void 636 malloc_init(void *data) 637 { 638 struct malloc_type_internal *mtip; 639 struct malloc_type *mtp; 640 641 KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init")); 642 643 mtp = data; 644 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 645 mtp->ks_handle = mtip; 646 647 mtx_lock(&malloc_mtx); 648 mtp->ks_next = kmemstatistics; 649 kmemstatistics = mtp; 650 kmemcount++; 651 mtx_unlock(&malloc_mtx); 652 } 653 654 void 655 malloc_uninit(void *data) 656 { 657 struct malloc_type_internal *mtip; 658 struct malloc_type_stats *mtsp; 659 struct malloc_type *mtp, *temp; 660 uma_slab_t slab; 661 long temp_allocs, temp_bytes; 662 int i; 663 664 mtp = data; 665 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 666 mtx_lock(&malloc_mtx); 667 mtip = mtp->ks_handle; 668 mtp->ks_handle = NULL; 669 if (mtp != kmemstatistics) { 670 for (temp = kmemstatistics; temp != NULL; 671 temp = temp->ks_next) { 672 if (temp->ks_next == mtp) 673 temp->ks_next = mtp->ks_next; 674 } 675 } else 676 kmemstatistics = mtp->ks_next; 677 kmemcount--; 678 mtx_unlock(&malloc_mtx); 679 680 /* 681 * Look for memory leaks. 682 */ 683 temp_allocs = temp_bytes = 0; 684 for (i = 0; i < MAXCPU; i++) { 685 mtsp = &mtip->mti_stats[i]; 686 temp_allocs += mtsp->mts_numallocs; 687 temp_allocs -= mtsp->mts_numfrees; 688 temp_bytes += mtsp->mts_memalloced; 689 temp_bytes -= mtsp->mts_memfreed; 690 } 691 if (temp_allocs > 0 || temp_bytes > 0) { 692 printf("Warning: memory type %s leaked memory on destroy " 693 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 694 temp_allocs, temp_bytes); 695 } 696 697 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 698 uma_zfree_arg(mt_zone, mtip, slab); 699 } 700 701 struct malloc_type * 702 malloc_desc2type(const char *desc) 703 { 704 struct malloc_type *mtp; 705 706 mtx_assert(&malloc_mtx, MA_OWNED); 707 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 708 if (strcmp(mtp->ks_shortdesc, desc) == 0) 709 return (mtp); 710 } 711 return (NULL); 712 } 713 714 static int 715 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 716 { 717 struct malloc_type_stream_header mtsh; 718 struct malloc_type_internal *mtip; 719 struct malloc_type_header mth; 720 struct malloc_type *mtp; 721 int buflen, count, error, i; 722 struct sbuf sbuf; 723 char *buffer; 724 725 mtx_lock(&malloc_mtx); 726 restart: 727 mtx_assert(&malloc_mtx, MA_OWNED); 728 count = kmemcount; 729 mtx_unlock(&malloc_mtx); 730 buflen = sizeof(mtsh) + count * (sizeof(mth) + 731 sizeof(struct malloc_type_stats) * MAXCPU) + 1; 732 buffer = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 733 mtx_lock(&malloc_mtx); 734 if (count < kmemcount) { 735 free(buffer, M_TEMP); 736 goto restart; 737 } 738 739 sbuf_new(&sbuf, buffer, buflen, SBUF_FIXEDLEN); 740 741 /* 742 * Insert stream header. 743 */ 744 bzero(&mtsh, sizeof(mtsh)); 745 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 746 mtsh.mtsh_maxcpus = MAXCPU; 747 mtsh.mtsh_count = kmemcount; 748 if (sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)) < 0) { 749 mtx_unlock(&malloc_mtx); 750 error = ENOMEM; 751 goto out; 752 } 753 754 /* 755 * Insert alternating sequence of type headers and type statistics. 756 */ 757 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 758 mtip = (struct malloc_type_internal *)mtp->ks_handle; 759 760 /* 761 * Insert type header. 762 */ 763 bzero(&mth, sizeof(mth)); 764 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 765 if (sbuf_bcat(&sbuf, &mth, sizeof(mth)) < 0) { 766 mtx_unlock(&malloc_mtx); 767 error = ENOMEM; 768 goto out; 769 } 770 771 /* 772 * Insert type statistics for each CPU. 773 */ 774 for (i = 0; i < MAXCPU; i++) { 775 if (sbuf_bcat(&sbuf, &mtip->mti_stats[i], 776 sizeof(mtip->mti_stats[i])) < 0) { 777 mtx_unlock(&malloc_mtx); 778 error = ENOMEM; 779 goto out; 780 } 781 } 782 } 783 mtx_unlock(&malloc_mtx); 784 sbuf_finish(&sbuf); 785 error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf)); 786 out: 787 sbuf_delete(&sbuf); 788 free(buffer, M_TEMP); 789 return (error); 790 } 791 792 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 793 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 794 "Return malloc types"); 795 796 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 797 "Count of kernel malloc types"); 798 799 #ifdef DDB 800 DB_SHOW_COMMAND(malloc, db_show_malloc) 801 { 802 struct malloc_type_internal *mtip; 803 struct malloc_type *mtp; 804 u_int64_t allocs, frees; 805 u_int64_t alloced, freed; 806 int i; 807 808 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 809 "Requests"); 810 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 811 mtip = (struct malloc_type_internal *)mtp->ks_handle; 812 allocs = 0; 813 frees = 0; 814 alloced = 0; 815 freed = 0; 816 for (i = 0; i < MAXCPU; i++) { 817 allocs += mtip->mti_stats[i].mts_numallocs; 818 frees += mtip->mti_stats[i].mts_numfrees; 819 alloced += mtip->mti_stats[i].mts_memalloced; 820 freed += mtip->mti_stats[i].mts_memfreed; 821 } 822 db_printf("%18s %12ju %12juK %12ju\n", 823 mtp->ks_shortdesc, allocs - frees, 824 (alloced - freed + 1023) / 1024, allocs); 825 } 826 } 827 #endif 828 829 #ifdef MALLOC_PROFILE 830 831 static int 832 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 833 { 834 int linesize = 64; 835 struct sbuf sbuf; 836 uint64_t count; 837 uint64_t waste; 838 uint64_t mem; 839 int bufsize; 840 int error; 841 char *buf; 842 int rsize; 843 int size; 844 int i; 845 846 bufsize = linesize * (KMEM_ZSIZE + 1); 847 bufsize += 128; /* For the stats line */ 848 bufsize += 128; /* For the banner line */ 849 waste = 0; 850 mem = 0; 851 852 buf = malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO); 853 sbuf_new(&sbuf, buf, bufsize, SBUF_FIXEDLEN); 854 sbuf_printf(&sbuf, 855 "\n Size Requests Real Size\n"); 856 for (i = 0; i < KMEM_ZSIZE; i++) { 857 size = i << KMEM_ZSHIFT; 858 rsize = kmemzones[kmemsize[i]].kz_size; 859 count = (long long unsigned)krequests[i]; 860 861 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 862 (unsigned long long)count, rsize); 863 864 if ((rsize * count) > (size * count)) 865 waste += (rsize * count) - (size * count); 866 mem += (rsize * count); 867 } 868 sbuf_printf(&sbuf, 869 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 870 (unsigned long long)mem, (unsigned long long)waste); 871 sbuf_finish(&sbuf); 872 873 error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf)); 874 875 sbuf_delete(&sbuf); 876 free(buf, M_TEMP); 877 return (error); 878 } 879 880 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 881 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 882 #endif /* MALLOC_PROFILE */ 883