1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1987, 1991, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2005-2009 Robert N. M. Watson 7 * Copyright (c) 2008 Otto Moerbeek <otto@drijf.net> (mallocarray) 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94 35 */ 36 37 /* 38 * Kernel malloc(9) implementation -- general purpose kernel memory allocator 39 * based on memory types. Back end is implemented using the UMA(9) zone 40 * allocator. A set of fixed-size buckets are used for smaller allocations, 41 * and a special UMA allocation interface is used for larger allocations. 42 * Callers declare memory types, and statistics are maintained independently 43 * for each memory type. Statistics are maintained per-CPU for performance 44 * reasons. See malloc(9) and comments in malloc.h for a detailed 45 * description. 46 */ 47 48 #include <sys/cdefs.h> 49 __FBSDID("$FreeBSD$"); 50 51 #include "opt_ddb.h" 52 #include "opt_vm.h" 53 54 #include <sys/param.h> 55 #include <sys/systm.h> 56 #include <sys/kdb.h> 57 #include <sys/kernel.h> 58 #include <sys/lock.h> 59 #include <sys/malloc.h> 60 #include <sys/mutex.h> 61 #include <sys/vmmeter.h> 62 #include <sys/proc.h> 63 #include <sys/queue.h> 64 #include <sys/sbuf.h> 65 #include <sys/smp.h> 66 #include <sys/sysctl.h> 67 #include <sys/time.h> 68 #include <sys/vmem.h> 69 #ifdef EPOCH_TRACE 70 #include <sys/epoch.h> 71 #endif 72 73 #include <vm/vm.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_domainset.h> 76 #include <vm/vm_pageout.h> 77 #include <vm/vm_param.h> 78 #include <vm/vm_kern.h> 79 #include <vm/vm_extern.h> 80 #include <vm/vm_map.h> 81 #include <vm/vm_page.h> 82 #include <vm/vm_phys.h> 83 #include <vm/vm_pagequeue.h> 84 #include <vm/uma.h> 85 #include <vm/uma_int.h> 86 #include <vm/uma_dbg.h> 87 88 #ifdef DEBUG_MEMGUARD 89 #include <vm/memguard.h> 90 #endif 91 #ifdef DEBUG_REDZONE 92 #include <vm/redzone.h> 93 #endif 94 95 #if defined(INVARIANTS) && defined(__i386__) 96 #include <machine/cpu.h> 97 #endif 98 99 #include <ddb/ddb.h> 100 101 #ifdef KDTRACE_HOOKS 102 #include <sys/dtrace_bsd.h> 103 104 bool __read_frequently dtrace_malloc_enabled; 105 dtrace_malloc_probe_func_t __read_mostly dtrace_malloc_probe; 106 #endif 107 108 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) || \ 109 defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE) 110 #define MALLOC_DEBUG 1 111 #endif 112 113 /* 114 * When realloc() is called, if the new size is sufficiently smaller than 115 * the old size, realloc() will allocate a new, smaller block to avoid 116 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 117 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 118 */ 119 #ifndef REALLOC_FRACTION 120 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 121 #endif 122 123 /* 124 * Centrally define some common malloc types. 125 */ 126 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 127 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 128 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 129 130 static struct malloc_type *kmemstatistics; 131 static int kmemcount; 132 133 #define KMEM_ZSHIFT 4 134 #define KMEM_ZBASE 16 135 #define KMEM_ZMASK (KMEM_ZBASE - 1) 136 137 #define KMEM_ZMAX 65536 138 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 139 static uint8_t kmemsize[KMEM_ZSIZE + 1]; 140 141 #ifndef MALLOC_DEBUG_MAXZONES 142 #define MALLOC_DEBUG_MAXZONES 1 143 #endif 144 static int numzones = MALLOC_DEBUG_MAXZONES; 145 146 /* 147 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 148 * of various sizes. 149 * 150 * XXX: The comment here used to read "These won't be powers of two for 151 * long." It's possible that a significant amount of wasted memory could be 152 * recovered by tuning the sizes of these buckets. 153 */ 154 struct { 155 int kz_size; 156 const char *kz_name; 157 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 158 } kmemzones[] = { 159 {16, "16", }, 160 {32, "32", }, 161 {64, "64", }, 162 {128, "128", }, 163 {256, "256", }, 164 {512, "512", }, 165 {1024, "1024", }, 166 {2048, "2048", }, 167 {4096, "4096", }, 168 {8192, "8192", }, 169 {16384, "16384", }, 170 {32768, "32768", }, 171 {65536, "65536", }, 172 {0, NULL}, 173 }; 174 175 /* 176 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 177 * types are described by a data structure passed by the declaring code, but 178 * the malloc(9) implementation has its own data structure describing the 179 * type and statistics. This permits the malloc(9)-internal data structures 180 * to be modified without breaking binary-compiled kernel modules that 181 * declare malloc types. 182 */ 183 static uma_zone_t mt_zone; 184 static uma_zone_t mt_stats_zone; 185 186 u_long vm_kmem_size; 187 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 188 "Size of kernel memory"); 189 190 static u_long kmem_zmax = KMEM_ZMAX; 191 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 192 "Maximum allocation size that malloc(9) would use UMA as backend"); 193 194 static u_long vm_kmem_size_min; 195 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 196 "Minimum size of kernel memory"); 197 198 static u_long vm_kmem_size_max; 199 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 200 "Maximum size of kernel memory"); 201 202 static u_int vm_kmem_size_scale; 203 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 204 "Scale factor for kernel memory size"); 205 206 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 207 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 208 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 209 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 210 211 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 212 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 213 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 214 sysctl_kmem_map_free, "LU", "Free space in kmem"); 215 216 /* 217 * The malloc_mtx protects the kmemstatistics linked list. 218 */ 219 struct mtx malloc_mtx; 220 221 #ifdef MALLOC_PROFILE 222 uint64_t krequests[KMEM_ZSIZE + 1]; 223 224 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 225 #endif 226 227 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 228 229 /* 230 * time_uptime of the last malloc(9) failure (induced or real). 231 */ 232 static time_t t_malloc_fail; 233 234 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 235 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 236 "Kernel malloc debugging options"); 237 #endif 238 239 /* 240 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 241 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 242 */ 243 #ifdef MALLOC_MAKE_FAILURES 244 static int malloc_failure_rate; 245 static int malloc_nowait_count; 246 static int malloc_failure_count; 247 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, 248 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 249 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 250 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 251 #endif 252 253 static int 254 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 255 { 256 u_long size; 257 258 size = uma_size(); 259 return (sysctl_handle_long(oidp, &size, 0, req)); 260 } 261 262 static int 263 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 264 { 265 u_long size, limit; 266 267 /* The sysctl is unsigned, implement as a saturation value. */ 268 size = uma_size(); 269 limit = uma_limit(); 270 if (size > limit) 271 size = 0; 272 else 273 size = limit - size; 274 return (sysctl_handle_long(oidp, &size, 0, req)); 275 } 276 277 /* 278 * malloc(9) uma zone separation -- sub-page buffer overruns in one 279 * malloc type will affect only a subset of other malloc types. 280 */ 281 #if MALLOC_DEBUG_MAXZONES > 1 282 static void 283 tunable_set_numzones(void) 284 { 285 286 TUNABLE_INT_FETCH("debug.malloc.numzones", 287 &numzones); 288 289 /* Sanity check the number of malloc uma zones. */ 290 if (numzones <= 0) 291 numzones = 1; 292 if (numzones > MALLOC_DEBUG_MAXZONES) 293 numzones = MALLOC_DEBUG_MAXZONES; 294 } 295 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 296 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 297 &numzones, 0, "Number of malloc uma subzones"); 298 299 /* 300 * Any number that changes regularly is an okay choice for the 301 * offset. Build numbers are pretty good of you have them. 302 */ 303 static u_int zone_offset = __FreeBSD_version; 304 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 305 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 306 &zone_offset, 0, "Separate malloc types by examining the " 307 "Nth character in the malloc type short description."); 308 309 static void 310 mtp_set_subzone(struct malloc_type *mtp) 311 { 312 struct malloc_type_internal *mtip; 313 const char *desc; 314 size_t len; 315 u_int val; 316 317 mtip = mtp->ks_handle; 318 desc = mtp->ks_shortdesc; 319 if (desc == NULL || (len = strlen(desc)) == 0) 320 val = 0; 321 else 322 val = desc[zone_offset % len]; 323 mtip->mti_zone = (val % numzones); 324 } 325 326 static inline u_int 327 mtp_get_subzone(struct malloc_type *mtp) 328 { 329 struct malloc_type_internal *mtip; 330 331 mtip = mtp->ks_handle; 332 333 KASSERT(mtip->mti_zone < numzones, 334 ("mti_zone %u out of range %d", 335 mtip->mti_zone, numzones)); 336 return (mtip->mti_zone); 337 } 338 #elif MALLOC_DEBUG_MAXZONES == 0 339 #error "MALLOC_DEBUG_MAXZONES must be positive." 340 #else 341 static void 342 mtp_set_subzone(struct malloc_type *mtp) 343 { 344 struct malloc_type_internal *mtip; 345 346 mtip = mtp->ks_handle; 347 mtip->mti_zone = 0; 348 } 349 350 static inline u_int 351 mtp_get_subzone(struct malloc_type *mtp) 352 { 353 354 return (0); 355 } 356 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 357 358 int 359 malloc_last_fail(void) 360 { 361 362 return (time_uptime - t_malloc_fail); 363 } 364 365 /* 366 * An allocation has succeeded -- update malloc type statistics for the 367 * amount of bucket size. Occurs within a critical section so that the 368 * thread isn't preempted and doesn't migrate while updating per-PCU 369 * statistics. 370 */ 371 static void 372 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 373 int zindx) 374 { 375 struct malloc_type_internal *mtip; 376 struct malloc_type_stats *mtsp; 377 378 critical_enter(); 379 mtip = mtp->ks_handle; 380 mtsp = zpcpu_get(mtip->mti_stats); 381 if (size > 0) { 382 mtsp->mts_memalloced += size; 383 mtsp->mts_numallocs++; 384 } 385 if (zindx != -1) 386 mtsp->mts_size |= 1 << zindx; 387 388 #ifdef KDTRACE_HOOKS 389 if (__predict_false(dtrace_malloc_enabled)) { 390 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 391 if (probe_id != 0) 392 (dtrace_malloc_probe)(probe_id, 393 (uintptr_t) mtp, (uintptr_t) mtip, 394 (uintptr_t) mtsp, size, zindx); 395 } 396 #endif 397 398 critical_exit(); 399 } 400 401 void 402 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 403 { 404 405 if (size > 0) 406 malloc_type_zone_allocated(mtp, size, -1); 407 } 408 409 /* 410 * A free operation has occurred -- update malloc type statistics for the 411 * amount of the bucket size. Occurs within a critical section so that the 412 * thread isn't preempted and doesn't migrate while updating per-CPU 413 * statistics. 414 */ 415 void 416 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 417 { 418 struct malloc_type_internal *mtip; 419 struct malloc_type_stats *mtsp; 420 421 critical_enter(); 422 mtip = mtp->ks_handle; 423 mtsp = zpcpu_get(mtip->mti_stats); 424 mtsp->mts_memfreed += size; 425 mtsp->mts_numfrees++; 426 427 #ifdef KDTRACE_HOOKS 428 if (__predict_false(dtrace_malloc_enabled)) { 429 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 430 if (probe_id != 0) 431 (dtrace_malloc_probe)(probe_id, 432 (uintptr_t) mtp, (uintptr_t) mtip, 433 (uintptr_t) mtsp, size, 0); 434 } 435 #endif 436 437 critical_exit(); 438 } 439 440 /* 441 * contigmalloc: 442 * 443 * Allocate a block of physically contiguous memory. 444 * 445 * If M_NOWAIT is set, this routine will not block and return NULL if 446 * the allocation fails. 447 */ 448 void * 449 contigmalloc(unsigned long size, struct malloc_type *type, int flags, 450 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 451 vm_paddr_t boundary) 452 { 453 void *ret; 454 455 ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment, 456 boundary, VM_MEMATTR_DEFAULT); 457 if (ret != NULL) 458 malloc_type_allocated(type, round_page(size)); 459 return (ret); 460 } 461 462 void * 463 contigmalloc_domainset(unsigned long size, struct malloc_type *type, 464 struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high, 465 unsigned long alignment, vm_paddr_t boundary) 466 { 467 void *ret; 468 469 ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high, 470 alignment, boundary, VM_MEMATTR_DEFAULT); 471 if (ret != NULL) 472 malloc_type_allocated(type, round_page(size)); 473 return (ret); 474 } 475 476 /* 477 * contigfree: 478 * 479 * Free a block of memory allocated by contigmalloc. 480 * 481 * This routine may not block. 482 */ 483 void 484 contigfree(void *addr, unsigned long size, struct malloc_type *type) 485 { 486 487 kmem_free((vm_offset_t)addr, size); 488 malloc_type_freed(type, round_page(size)); 489 } 490 491 #ifdef MALLOC_DEBUG 492 static int 493 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp, 494 int flags) 495 { 496 #ifdef INVARIANTS 497 int indx; 498 499 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 500 /* 501 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 502 */ 503 indx = flags & (M_WAITOK | M_NOWAIT); 504 if (indx != M_NOWAIT && indx != M_WAITOK) { 505 static struct timeval lasterr; 506 static int curerr, once; 507 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 508 printf("Bad malloc flags: %x\n", indx); 509 kdb_backtrace(); 510 flags |= M_WAITOK; 511 once++; 512 } 513 } 514 #endif 515 #ifdef MALLOC_MAKE_FAILURES 516 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 517 atomic_add_int(&malloc_nowait_count, 1); 518 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 519 atomic_add_int(&malloc_failure_count, 1); 520 t_malloc_fail = time_uptime; 521 *vap = NULL; 522 return (EJUSTRETURN); 523 } 524 } 525 #endif 526 if (flags & M_WAITOK) { 527 KASSERT(curthread->td_intr_nesting_level == 0, 528 ("malloc(M_WAITOK) in interrupt context")); 529 if (__predict_false(!THREAD_CAN_SLEEP())) { 530 #ifdef EPOCH_TRACE 531 epoch_trace_list(curthread); 532 #endif 533 KASSERT(1, 534 ("malloc(M_WAITOK) with sleeping prohibited")); 535 } 536 } 537 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 538 ("malloc: called with spinlock or critical section held")); 539 540 #ifdef DEBUG_MEMGUARD 541 if (memguard_cmp_mtp(mtp, *sizep)) { 542 *vap = memguard_alloc(*sizep, flags); 543 if (*vap != NULL) 544 return (EJUSTRETURN); 545 /* This is unfortunate but should not be fatal. */ 546 } 547 #endif 548 549 #ifdef DEBUG_REDZONE 550 *sizep = redzone_size_ntor(*sizep); 551 #endif 552 553 return (0); 554 } 555 #endif 556 557 /* 558 * Handle large allocations and frees by using kmem_malloc directly. 559 */ 560 static inline bool 561 malloc_large_slab(uma_slab_t slab) 562 { 563 uintptr_t va; 564 565 va = (uintptr_t)slab; 566 return ((va & 1) != 0); 567 } 568 569 static inline size_t 570 malloc_large_size(uma_slab_t slab) 571 { 572 uintptr_t va; 573 574 va = (uintptr_t)slab; 575 return (va >> 1); 576 } 577 578 static caddr_t 579 malloc_large(size_t *size, struct domainset *policy, int flags) 580 { 581 vm_offset_t va; 582 size_t sz; 583 584 sz = roundup(*size, PAGE_SIZE); 585 va = kmem_malloc_domainset(policy, sz, flags); 586 if (va != 0) { 587 /* The low bit is unused for slab pointers. */ 588 vsetzoneslab(va, NULL, (void *)((sz << 1) | 1)); 589 uma_total_inc(sz); 590 *size = sz; 591 } 592 return ((caddr_t)va); 593 } 594 595 static void 596 free_large(void *addr, size_t size) 597 { 598 599 kmem_free((vm_offset_t)addr, size); 600 uma_total_dec(size); 601 } 602 603 /* 604 * malloc: 605 * 606 * Allocate a block of memory. 607 * 608 * If M_NOWAIT is set, this routine will not block and return NULL if 609 * the allocation fails. 610 */ 611 void * 612 (malloc)(size_t size, struct malloc_type *mtp, int flags) 613 { 614 int indx; 615 caddr_t va; 616 uma_zone_t zone; 617 #if defined(DEBUG_REDZONE) 618 unsigned long osize = size; 619 #endif 620 621 #ifdef MALLOC_DEBUG 622 va = NULL; 623 if (malloc_dbg(&va, &size, mtp, flags) != 0) 624 return (va); 625 #endif 626 627 if (size <= kmem_zmax && (flags & M_EXEC) == 0) { 628 if (size & KMEM_ZMASK) 629 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 630 indx = kmemsize[size >> KMEM_ZSHIFT]; 631 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)]; 632 #ifdef MALLOC_PROFILE 633 krequests[size >> KMEM_ZSHIFT]++; 634 #endif 635 va = uma_zalloc(zone, flags); 636 if (va != NULL) 637 size = zone->uz_size; 638 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 639 } else { 640 va = malloc_large(&size, DOMAINSET_RR(), flags); 641 malloc_type_allocated(mtp, va == NULL ? 0 : size); 642 } 643 if (flags & M_WAITOK) 644 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 645 else if (va == NULL) 646 t_malloc_fail = time_uptime; 647 #ifdef DEBUG_REDZONE 648 if (va != NULL) 649 va = redzone_setup(va, osize); 650 #endif 651 return ((void *) va); 652 } 653 654 static void * 655 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain, 656 int flags) 657 { 658 uma_zone_t zone; 659 caddr_t va; 660 size_t size; 661 int indx; 662 663 size = *sizep; 664 KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0, 665 ("malloc_domain: Called with bad flag / size combination.")); 666 if (size & KMEM_ZMASK) 667 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 668 indx = kmemsize[size >> KMEM_ZSHIFT]; 669 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)]; 670 #ifdef MALLOC_PROFILE 671 krequests[size >> KMEM_ZSHIFT]++; 672 #endif 673 va = uma_zalloc_domain(zone, NULL, domain, flags); 674 if (va != NULL) 675 *sizep = zone->uz_size; 676 *indxp = indx; 677 return ((void *)va); 678 } 679 680 void * 681 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds, 682 int flags) 683 { 684 struct vm_domainset_iter di; 685 caddr_t ret; 686 int domain; 687 int indx; 688 689 #if defined(DEBUG_REDZONE) 690 unsigned long osize = size; 691 #endif 692 #ifdef MALLOC_DEBUG 693 ret= NULL; 694 if (malloc_dbg(&ret, &size, mtp, flags) != 0) 695 return (ret); 696 #endif 697 if (size <= kmem_zmax && (flags & M_EXEC) == 0) { 698 vm_domainset_iter_policy_init(&di, ds, &domain, &flags); 699 do { 700 ret = malloc_domain(&size, &indx, mtp, domain, flags); 701 } while (ret == NULL && 702 vm_domainset_iter_policy(&di, &domain) == 0); 703 malloc_type_zone_allocated(mtp, ret == NULL ? 0 : size, indx); 704 } else { 705 /* Policy is handled by kmem. */ 706 ret = malloc_large(&size, ds, flags); 707 malloc_type_allocated(mtp, ret == NULL ? 0 : size); 708 } 709 710 if (flags & M_WAITOK) 711 KASSERT(ret != NULL, ("malloc(M_WAITOK) returned NULL")); 712 else if (ret == NULL) 713 t_malloc_fail = time_uptime; 714 #ifdef DEBUG_REDZONE 715 if (ret != NULL) 716 ret = redzone_setup(ret, osize); 717 #endif 718 return (ret); 719 } 720 721 void * 722 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags) 723 { 724 725 if (WOULD_OVERFLOW(nmemb, size)) 726 panic("mallocarray: %zu * %zu overflowed", nmemb, size); 727 728 return (malloc(size * nmemb, type, flags)); 729 } 730 731 #ifdef INVARIANTS 732 static void 733 free_save_type(void *addr, struct malloc_type *mtp, u_long size) 734 { 735 struct malloc_type **mtpp = addr; 736 737 /* 738 * Cache a pointer to the malloc_type that most recently freed 739 * this memory here. This way we know who is most likely to 740 * have stepped on it later. 741 * 742 * This code assumes that size is a multiple of 8 bytes for 743 * 64 bit machines 744 */ 745 mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 746 mtpp += (size - sizeof(struct malloc_type *)) / 747 sizeof(struct malloc_type *); 748 *mtpp = mtp; 749 } 750 #endif 751 752 #ifdef MALLOC_DEBUG 753 static int 754 free_dbg(void **addrp, struct malloc_type *mtp) 755 { 756 void *addr; 757 758 addr = *addrp; 759 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); 760 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 761 ("free: called with spinlock or critical section held")); 762 763 /* free(NULL, ...) does nothing */ 764 if (addr == NULL) 765 return (EJUSTRETURN); 766 767 #ifdef DEBUG_MEMGUARD 768 if (is_memguard_addr(addr)) { 769 memguard_free(addr); 770 return (EJUSTRETURN); 771 } 772 #endif 773 774 #ifdef DEBUG_REDZONE 775 redzone_check(addr); 776 *addrp = redzone_addr_ntor(addr); 777 #endif 778 779 return (0); 780 } 781 #endif 782 783 /* 784 * free: 785 * 786 * Free a block of memory allocated by malloc. 787 * 788 * This routine may not block. 789 */ 790 void 791 free(void *addr, struct malloc_type *mtp) 792 { 793 uma_zone_t zone; 794 uma_slab_t slab; 795 u_long size; 796 797 #ifdef MALLOC_DEBUG 798 if (free_dbg(&addr, mtp) != 0) 799 return; 800 #endif 801 /* free(NULL, ...) does nothing */ 802 if (addr == NULL) 803 return; 804 805 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 806 if (slab == NULL) 807 panic("free: address %p(%p) has not been allocated.\n", 808 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 809 810 if (__predict_true(!malloc_large_slab(slab))) { 811 size = zone->uz_size; 812 #ifdef INVARIANTS 813 free_save_type(addr, mtp, size); 814 #endif 815 uma_zfree_arg(zone, addr, slab); 816 } else { 817 size = malloc_large_size(slab); 818 free_large(addr, size); 819 } 820 malloc_type_freed(mtp, size); 821 } 822 823 /* 824 * zfree: 825 * 826 * Zero then free a block of memory allocated by malloc. 827 * 828 * This routine may not block. 829 */ 830 void 831 zfree(void *addr, struct malloc_type *mtp) 832 { 833 uma_zone_t zone; 834 uma_slab_t slab; 835 u_long size; 836 837 #ifdef MALLOC_DEBUG 838 if (free_dbg(&addr, mtp) != 0) 839 return; 840 #endif 841 /* free(NULL, ...) does nothing */ 842 if (addr == NULL) 843 return; 844 845 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 846 if (slab == NULL) 847 panic("free: address %p(%p) has not been allocated.\n", 848 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 849 850 if (__predict_true(!malloc_large_slab(slab))) { 851 size = zone->uz_size; 852 #ifdef INVARIANTS 853 free_save_type(addr, mtp, size); 854 #endif 855 explicit_bzero(addr, size); 856 uma_zfree_arg(zone, addr, slab); 857 } else { 858 size = malloc_large_size(slab); 859 explicit_bzero(addr, size); 860 free_large(addr, size); 861 } 862 malloc_type_freed(mtp, size); 863 } 864 865 /* 866 * realloc: change the size of a memory block 867 */ 868 void * 869 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags) 870 { 871 uma_zone_t zone; 872 uma_slab_t slab; 873 unsigned long alloc; 874 void *newaddr; 875 876 KASSERT(mtp->ks_magic == M_MAGIC, 877 ("realloc: bad malloc type magic")); 878 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 879 ("realloc: called with spinlock or critical section held")); 880 881 /* realloc(NULL, ...) is equivalent to malloc(...) */ 882 if (addr == NULL) 883 return (malloc(size, mtp, flags)); 884 885 /* 886 * XXX: Should report free of old memory and alloc of new memory to 887 * per-CPU stats. 888 */ 889 890 #ifdef DEBUG_MEMGUARD 891 if (is_memguard_addr(addr)) 892 return (memguard_realloc(addr, size, mtp, flags)); 893 #endif 894 895 #ifdef DEBUG_REDZONE 896 slab = NULL; 897 zone = NULL; 898 alloc = redzone_get_size(addr); 899 #else 900 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 901 902 /* Sanity check */ 903 KASSERT(slab != NULL, 904 ("realloc: address %p out of range", (void *)addr)); 905 906 /* Get the size of the original block */ 907 if (!malloc_large_slab(slab)) 908 alloc = zone->uz_size; 909 else 910 alloc = malloc_large_size(slab); 911 912 /* Reuse the original block if appropriate */ 913 if (size <= alloc 914 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 915 return (addr); 916 #endif /* !DEBUG_REDZONE */ 917 918 /* Allocate a new, bigger (or smaller) block */ 919 if ((newaddr = malloc(size, mtp, flags)) == NULL) 920 return (NULL); 921 922 /* Copy over original contents */ 923 bcopy(addr, newaddr, min(size, alloc)); 924 free(addr, mtp); 925 return (newaddr); 926 } 927 928 /* 929 * reallocf: same as realloc() but free memory on failure. 930 */ 931 void * 932 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags) 933 { 934 void *mem; 935 936 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 937 free(addr, mtp); 938 return (mem); 939 } 940 941 /* 942 * malloc_usable_size: returns the usable size of the allocation. 943 */ 944 size_t 945 malloc_usable_size(const void *addr) 946 { 947 #ifndef DEBUG_REDZONE 948 uma_zone_t zone; 949 uma_slab_t slab; 950 #endif 951 u_long size; 952 953 if (addr == NULL) 954 return (0); 955 956 #ifdef DEBUG_MEMGUARD 957 if (is_memguard_addr(__DECONST(void *, addr))) 958 return (memguard_get_req_size(addr)); 959 #endif 960 961 #ifdef DEBUG_REDZONE 962 size = redzone_get_size(__DECONST(void *, addr)); 963 #else 964 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 965 if (slab == NULL) 966 panic("malloc_usable_size: address %p(%p) is not allocated.\n", 967 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 968 969 if (!malloc_large_slab(slab)) 970 size = zone->uz_size; 971 else 972 size = malloc_large_size(slab); 973 #endif 974 return (size); 975 } 976 977 CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 978 979 /* 980 * Initialize the kernel memory (kmem) arena. 981 */ 982 void 983 kmeminit(void) 984 { 985 u_long mem_size; 986 u_long tmp; 987 988 #ifdef VM_KMEM_SIZE 989 if (vm_kmem_size == 0) 990 vm_kmem_size = VM_KMEM_SIZE; 991 #endif 992 #ifdef VM_KMEM_SIZE_MIN 993 if (vm_kmem_size_min == 0) 994 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 995 #endif 996 #ifdef VM_KMEM_SIZE_MAX 997 if (vm_kmem_size_max == 0) 998 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 999 #endif 1000 /* 1001 * Calculate the amount of kernel virtual address (KVA) space that is 1002 * preallocated to the kmem arena. In order to support a wide range 1003 * of machines, it is a function of the physical memory size, 1004 * specifically, 1005 * 1006 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 1007 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 1008 * 1009 * Every architecture must define an integral value for 1010 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 1011 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 1012 * ceiling on this preallocation, are optional. Typically, 1013 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 1014 * a given architecture. 1015 */ 1016 mem_size = vm_cnt.v_page_count; 1017 if (mem_size <= 32768) /* delphij XXX 128MB */ 1018 kmem_zmax = PAGE_SIZE; 1019 1020 if (vm_kmem_size_scale < 1) 1021 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 1022 1023 /* 1024 * Check if we should use defaults for the "vm_kmem_size" 1025 * variable: 1026 */ 1027 if (vm_kmem_size == 0) { 1028 vm_kmem_size = mem_size / vm_kmem_size_scale; 1029 vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ? 1030 vm_kmem_size_max : vm_kmem_size * PAGE_SIZE; 1031 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 1032 vm_kmem_size = vm_kmem_size_min; 1033 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 1034 vm_kmem_size = vm_kmem_size_max; 1035 } 1036 if (vm_kmem_size == 0) 1037 panic("Tune VM_KMEM_SIZE_* for the platform"); 1038 1039 /* 1040 * The amount of KVA space that is preallocated to the 1041 * kmem arena can be set statically at compile-time or manually 1042 * through the kernel environment. However, it is still limited to 1043 * twice the physical memory size, which has been sufficient to handle 1044 * the most severe cases of external fragmentation in the kmem arena. 1045 */ 1046 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 1047 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 1048 1049 vm_kmem_size = round_page(vm_kmem_size); 1050 #ifdef DEBUG_MEMGUARD 1051 tmp = memguard_fudge(vm_kmem_size, kernel_map); 1052 #else 1053 tmp = vm_kmem_size; 1054 #endif 1055 uma_set_limit(tmp); 1056 1057 #ifdef DEBUG_MEMGUARD 1058 /* 1059 * Initialize MemGuard if support compiled in. MemGuard is a 1060 * replacement allocator used for detecting tamper-after-free 1061 * scenarios as they occur. It is only used for debugging. 1062 */ 1063 memguard_init(kernel_arena); 1064 #endif 1065 } 1066 1067 /* 1068 * Initialize the kernel memory allocator 1069 */ 1070 /* ARGSUSED*/ 1071 static void 1072 mallocinit(void *dummy) 1073 { 1074 int i; 1075 uint8_t indx; 1076 1077 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 1078 1079 kmeminit(); 1080 1081 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 1082 kmem_zmax = KMEM_ZMAX; 1083 1084 mt_stats_zone = uma_zcreate("mt_stats_zone", 1085 sizeof(struct malloc_type_stats), NULL, NULL, NULL, NULL, 1086 UMA_ALIGN_PTR, UMA_ZONE_PCPU); 1087 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 1088 #ifdef INVARIANTS 1089 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 1090 #else 1091 NULL, NULL, NULL, NULL, 1092 #endif 1093 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 1094 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 1095 int size = kmemzones[indx].kz_size; 1096 const char *name = kmemzones[indx].kz_name; 1097 int subzone; 1098 1099 for (subzone = 0; subzone < numzones; subzone++) { 1100 kmemzones[indx].kz_zone[subzone] = 1101 uma_zcreate(name, size, 1102 #ifdef INVARIANTS 1103 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 1104 #else 1105 NULL, NULL, NULL, NULL, 1106 #endif 1107 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 1108 } 1109 for (;i <= size; i+= KMEM_ZBASE) 1110 kmemsize[i >> KMEM_ZSHIFT] = indx; 1111 } 1112 } 1113 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); 1114 1115 void 1116 malloc_init(void *data) 1117 { 1118 struct malloc_type_internal *mtip; 1119 struct malloc_type *mtp; 1120 1121 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init")); 1122 1123 mtp = data; 1124 if (mtp->ks_magic != M_MAGIC) 1125 panic("malloc_init: bad malloc type magic"); 1126 1127 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 1128 mtip->mti_stats = uma_zalloc_pcpu(mt_stats_zone, M_WAITOK | M_ZERO); 1129 mtp->ks_handle = mtip; 1130 mtp_set_subzone(mtp); 1131 1132 mtx_lock(&malloc_mtx); 1133 mtp->ks_next = kmemstatistics; 1134 kmemstatistics = mtp; 1135 kmemcount++; 1136 mtx_unlock(&malloc_mtx); 1137 } 1138 1139 void 1140 malloc_uninit(void *data) 1141 { 1142 struct malloc_type_internal *mtip; 1143 struct malloc_type_stats *mtsp; 1144 struct malloc_type *mtp, *temp; 1145 uma_slab_t slab; 1146 long temp_allocs, temp_bytes; 1147 int i; 1148 1149 mtp = data; 1150 KASSERT(mtp->ks_magic == M_MAGIC, 1151 ("malloc_uninit: bad malloc type magic")); 1152 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 1153 1154 mtx_lock(&malloc_mtx); 1155 mtip = mtp->ks_handle; 1156 mtp->ks_handle = NULL; 1157 if (mtp != kmemstatistics) { 1158 for (temp = kmemstatistics; temp != NULL; 1159 temp = temp->ks_next) { 1160 if (temp->ks_next == mtp) { 1161 temp->ks_next = mtp->ks_next; 1162 break; 1163 } 1164 } 1165 KASSERT(temp, 1166 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 1167 } else 1168 kmemstatistics = mtp->ks_next; 1169 kmemcount--; 1170 mtx_unlock(&malloc_mtx); 1171 1172 /* 1173 * Look for memory leaks. 1174 */ 1175 temp_allocs = temp_bytes = 0; 1176 for (i = 0; i <= mp_maxid; i++) { 1177 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1178 temp_allocs += mtsp->mts_numallocs; 1179 temp_allocs -= mtsp->mts_numfrees; 1180 temp_bytes += mtsp->mts_memalloced; 1181 temp_bytes -= mtsp->mts_memfreed; 1182 } 1183 if (temp_allocs > 0 || temp_bytes > 0) { 1184 printf("Warning: memory type %s leaked memory on destroy " 1185 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 1186 temp_allocs, temp_bytes); 1187 } 1188 1189 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 1190 uma_zfree_pcpu(mt_stats_zone, mtip->mti_stats); 1191 uma_zfree_arg(mt_zone, mtip, slab); 1192 } 1193 1194 struct malloc_type * 1195 malloc_desc2type(const char *desc) 1196 { 1197 struct malloc_type *mtp; 1198 1199 mtx_assert(&malloc_mtx, MA_OWNED); 1200 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1201 if (strcmp(mtp->ks_shortdesc, desc) == 0) 1202 return (mtp); 1203 } 1204 return (NULL); 1205 } 1206 1207 static int 1208 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 1209 { 1210 struct malloc_type_stream_header mtsh; 1211 struct malloc_type_internal *mtip; 1212 struct malloc_type_stats *mtsp, zeromts; 1213 struct malloc_type_header mth; 1214 struct malloc_type *mtp; 1215 int error, i; 1216 struct sbuf sbuf; 1217 1218 error = sysctl_wire_old_buffer(req, 0); 1219 if (error != 0) 1220 return (error); 1221 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1222 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); 1223 mtx_lock(&malloc_mtx); 1224 1225 bzero(&zeromts, sizeof(zeromts)); 1226 1227 /* 1228 * Insert stream header. 1229 */ 1230 bzero(&mtsh, sizeof(mtsh)); 1231 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 1232 mtsh.mtsh_maxcpus = MAXCPU; 1233 mtsh.mtsh_count = kmemcount; 1234 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 1235 1236 /* 1237 * Insert alternating sequence of type headers and type statistics. 1238 */ 1239 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1240 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1241 1242 /* 1243 * Insert type header. 1244 */ 1245 bzero(&mth, sizeof(mth)); 1246 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 1247 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 1248 1249 /* 1250 * Insert type statistics for each CPU. 1251 */ 1252 for (i = 0; i <= mp_maxid; i++) { 1253 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1254 (void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp)); 1255 } 1256 /* 1257 * Fill in the missing CPUs. 1258 */ 1259 for (; i < MAXCPU; i++) { 1260 (void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts)); 1261 } 1262 } 1263 mtx_unlock(&malloc_mtx); 1264 error = sbuf_finish(&sbuf); 1265 sbuf_delete(&sbuf); 1266 return (error); 1267 } 1268 1269 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, 1270 CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0, 1271 sysctl_kern_malloc_stats, "s,malloc_type_ustats", 1272 "Return malloc types"); 1273 1274 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 1275 "Count of kernel malloc types"); 1276 1277 void 1278 malloc_type_list(malloc_type_list_func_t *func, void *arg) 1279 { 1280 struct malloc_type *mtp, **bufmtp; 1281 int count, i; 1282 size_t buflen; 1283 1284 mtx_lock(&malloc_mtx); 1285 restart: 1286 mtx_assert(&malloc_mtx, MA_OWNED); 1287 count = kmemcount; 1288 mtx_unlock(&malloc_mtx); 1289 1290 buflen = sizeof(struct malloc_type *) * count; 1291 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 1292 1293 mtx_lock(&malloc_mtx); 1294 1295 if (count < kmemcount) { 1296 free(bufmtp, M_TEMP); 1297 goto restart; 1298 } 1299 1300 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 1301 bufmtp[i] = mtp; 1302 1303 mtx_unlock(&malloc_mtx); 1304 1305 for (i = 0; i < count; i++) 1306 (func)(bufmtp[i], arg); 1307 1308 free(bufmtp, M_TEMP); 1309 } 1310 1311 #ifdef DDB 1312 static int64_t 1313 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs, 1314 uint64_t *inuse) 1315 { 1316 const struct malloc_type_stats *mtsp; 1317 uint64_t frees, alloced, freed; 1318 int i; 1319 1320 *allocs = 0; 1321 frees = 0; 1322 alloced = 0; 1323 freed = 0; 1324 for (i = 0; i <= mp_maxid; i++) { 1325 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1326 1327 *allocs += mtsp->mts_numallocs; 1328 frees += mtsp->mts_numfrees; 1329 alloced += mtsp->mts_memalloced; 1330 freed += mtsp->mts_memfreed; 1331 } 1332 *inuse = *allocs - frees; 1333 return (alloced - freed); 1334 } 1335 1336 DB_SHOW_COMMAND(malloc, db_show_malloc) 1337 { 1338 const char *fmt_hdr, *fmt_entry; 1339 struct malloc_type *mtp; 1340 uint64_t allocs, inuse; 1341 int64_t size; 1342 /* variables for sorting */ 1343 struct malloc_type *last_mtype, *cur_mtype; 1344 int64_t cur_size, last_size; 1345 int ties; 1346 1347 if (modif[0] == 'i') { 1348 fmt_hdr = "%s,%s,%s,%s\n"; 1349 fmt_entry = "\"%s\",%ju,%jdK,%ju\n"; 1350 } else { 1351 fmt_hdr = "%18s %12s %12s %12s\n"; 1352 fmt_entry = "%18s %12ju %12jdK %12ju\n"; 1353 } 1354 1355 db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests"); 1356 1357 /* Select sort, largest size first. */ 1358 last_mtype = NULL; 1359 last_size = INT64_MAX; 1360 for (;;) { 1361 cur_mtype = NULL; 1362 cur_size = -1; 1363 ties = 0; 1364 1365 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1366 /* 1367 * In the case of size ties, print out mtypes 1368 * in the order they are encountered. That is, 1369 * when we encounter the most recently output 1370 * mtype, we have already printed all preceding 1371 * ties, and we must print all following ties. 1372 */ 1373 if (mtp == last_mtype) { 1374 ties = 1; 1375 continue; 1376 } 1377 size = get_malloc_stats(mtp->ks_handle, &allocs, 1378 &inuse); 1379 if (size > cur_size && size < last_size + ties) { 1380 cur_size = size; 1381 cur_mtype = mtp; 1382 } 1383 } 1384 if (cur_mtype == NULL) 1385 break; 1386 1387 size = get_malloc_stats(cur_mtype->ks_handle, &allocs, &inuse); 1388 db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse, 1389 howmany(size, 1024), allocs); 1390 1391 if (db_pager_quit) 1392 break; 1393 1394 last_mtype = cur_mtype; 1395 last_size = cur_size; 1396 } 1397 } 1398 1399 #if MALLOC_DEBUG_MAXZONES > 1 1400 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1401 { 1402 struct malloc_type_internal *mtip; 1403 struct malloc_type *mtp; 1404 u_int subzone; 1405 1406 if (!have_addr) { 1407 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1408 return; 1409 } 1410 mtp = (void *)addr; 1411 if (mtp->ks_magic != M_MAGIC) { 1412 db_printf("Magic %lx does not match expected %x\n", 1413 mtp->ks_magic, M_MAGIC); 1414 return; 1415 } 1416 1417 mtip = mtp->ks_handle; 1418 subzone = mtip->mti_zone; 1419 1420 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1421 mtip = mtp->ks_handle; 1422 if (mtip->mti_zone != subzone) 1423 continue; 1424 db_printf("%s\n", mtp->ks_shortdesc); 1425 if (db_pager_quit) 1426 break; 1427 } 1428 } 1429 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1430 #endif /* DDB */ 1431 1432 #ifdef MALLOC_PROFILE 1433 1434 static int 1435 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1436 { 1437 struct sbuf sbuf; 1438 uint64_t count; 1439 uint64_t waste; 1440 uint64_t mem; 1441 int error; 1442 int rsize; 1443 int size; 1444 int i; 1445 1446 waste = 0; 1447 mem = 0; 1448 1449 error = sysctl_wire_old_buffer(req, 0); 1450 if (error != 0) 1451 return (error); 1452 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1453 sbuf_printf(&sbuf, 1454 "\n Size Requests Real Size\n"); 1455 for (i = 0; i < KMEM_ZSIZE; i++) { 1456 size = i << KMEM_ZSHIFT; 1457 rsize = kmemzones[kmemsize[i]].kz_size; 1458 count = (long long unsigned)krequests[i]; 1459 1460 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1461 (unsigned long long)count, rsize); 1462 1463 if ((rsize * count) > (size * count)) 1464 waste += (rsize * count) - (size * count); 1465 mem += (rsize * count); 1466 } 1467 sbuf_printf(&sbuf, 1468 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1469 (unsigned long long)mem, (unsigned long long)waste); 1470 error = sbuf_finish(&sbuf); 1471 sbuf_delete(&sbuf); 1472 return (error); 1473 } 1474 1475 SYSCTL_OID(_kern, OID_AUTO, mprof, 1476 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0, 1477 sysctl_kern_mprof, "A", 1478 "Malloc Profiling"); 1479 #endif /* MALLOC_PROFILE */ 1480