2 * DMALLOC.C - Dillon's malloc
4 * Copyright (c) 2011,2017 The DragonFly Project. All rights reserved.
6 * This code is derived from software contributed to The DragonFly Project
7 * by Matthew Dillon <dillon@backplane.com>.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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
17 * the documentation and/or other materials provided with the
19 * 3. Neither the name of The DragonFly Project nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific, prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * This module implements a modified slab allocator as a drop-in replacement
38 * for the libc malloc(). The slab algorithm has been adjusted to support
39 * dynamic sizing of slabs which effectively allows slabs to be used for
40 * allocations of any size. Because of this we neither have a small-block
41 * allocator or a big-block allocator and the code paths are simplified.
43 * To support dynamic slab sizing available user virtual memory is broken
44 * down into ~1024 regions. Each region has fixed slab size whos value is
45 * set when the region is opened up for use. The free() path simply applies
46 * a mask based on the region to the pointer to acquire the base of the
47 * governing slab structure.
49 * Regions[NREGIONS] (1024)
51 * Slab management and locking is done on a per-zone basis.
53 * Alloc Size Chunking Number of zones
64 * ... continues forever ... 4 zones
66 * For a 2^63 memory space each doubling >= 64K is broken down into
67 * 4 chunking zones, so we support 88 + (48 * 4) = 280 zones.
69 * API FEATURES AND SIDE EFFECTS
71 * + power-of-2 sized allocations up to a page will be power-of-2 aligned.
72 * Above that power-of-2 sized allocations are page-aligned. Non
73 * power-of-2 sized allocations are aligned the same as the chunk
74 * size for their zone.
75 * + ability to allocate arbitrarily large chunks of memory
76 * + realloc will reuse the passed pointer if possible, within the
77 * limitations of the zone chunking.
79 * On top of the slab allocator we also implement a 16-entry-per-thread
80 * magazine cache for allocations <= NOMSLABSIZE.
84 * + [better] garbage collection
85 * + better initial sizing.
89 * The value of the environment variable MALLOC_OPTIONS is a character string
90 * containing various flags to tune nmalloc. Upper case letters enabled
91 * or increase the feature, lower case disables or decreases the feature.
93 * U Enable UTRACE for all operations, observable with ktrace.
94 * Diasbled by default.
96 * Z Zero out allocations, otherwise allocations (except for
97 * calloc) will contain garbage.
98 * Disabled by default.
100 * H Pass a hint with madvise() about unused pages.
101 * Disabled by default.
102 * Not currently implemented.
104 * F Disable local per-thread caching.
105 * Disabled by default.
107 * C Increase (decrease) how much excess cache to retain.
108 * Set to 4 by default.
111 /* cc -shared -fPIC -g -O -I/usr/src/lib/libc/include -o dmalloc.so dmalloc.c */
113 #ifndef STANDALONE_DEBUG
114 #include "libc_private.h"
117 #include <sys/param.h>
118 #include <sys/types.h>
119 #include <sys/mman.h>
120 #include <sys/queue.h>
121 #include <sys/ktrace.h>
134 #include <machine/atomic.h>
135 #include <machine/cpufunc.h>
137 #ifdef STANDALONE_DEBUG
138 void _nmalloc_thr_init(void);
140 #include "spinlock.h"
141 #include "un-namespace.h"
144 #ifndef MAP_SIZEALIGN
145 #define MAP_SIZEALIGN 0
148 #if SSIZE_MAX == 0x7FFFFFFF
150 #define UVM_BITS 32 /* worst case */
153 #define UVM_BITS 48 /* worst case XXX */
156 #if LONG_MAX == 0x7FFFFFFF
158 #define LONG_BITS_SHIFT 5
161 #define LONG_BITS_SHIFT 6
164 #define LOCKEDPTR ((void *)(intptr_t)-1)
169 #define NREGIONS_BITS 10
170 #define NREGIONS (1 << NREGIONS_BITS)
171 #define NREGIONS_MASK (NREGIONS - 1)
172 #define NREGIONS_SHIFT (UVM_BITS - NREGIONS_BITS)
173 #define NREGIONS_SIZE (1LU << NREGIONS_SHIFT)
175 typedef struct region *region_t;
176 typedef struct slglobaldata *slglobaldata_t;
177 typedef struct slab *slab_t;
181 slab_t slab; /* conditional out of band slab */
184 static struct region Regions[NREGIONS];
187 * Number of chunking zones available
189 #define CHUNKFACTOR 8
191 #define NZONES (16 + 9 * CHUNKFACTOR + 16 * CHUNKFACTOR)
193 #define NZONES (16 + 9 * CHUNKFACTOR + 48 * CHUNKFACTOR)
196 static int MaxChunks[NZONES];
198 #define NDEPOTS 8 /* must be power of 2 */
201 * Maximum number of chunks per slab, governed by the allocation bitmap in
202 * each slab. The maximum is reduced for large chunk sizes.
204 #define MAXCHUNKS (LONG_BITS * LONG_BITS)
205 #define MAXCHUNKS_BITS (LONG_BITS_SHIFT * LONG_BITS_SHIFT)
206 #define LITSLABSIZE (32 * 1024)
207 #define NOMSLABSIZE (2 * 1024 * 1024)
208 #define BIGSLABSIZE (128 * 1024 * 1024)
210 #define ZALLOC_SLAB_MAGIC 0x736c6162 /* magic sanity */
212 TAILQ_HEAD(slab_list, slab);
218 struct slab *next; /* slabs with available space */
219 TAILQ_ENTRY(slab) entry;
220 int32_t magic; /* magic number for sanity check */
221 u_int navail; /* number of free elements available */
223 u_int free_bit; /* free hint bitno */
224 u_int free_index; /* free hint index */
225 u_long bitmap[LONG_BITS]; /* free chunks */
226 size_t slab_size; /* size of entire slab */
227 size_t chunk_size; /* chunk size for validation */
229 enum { UNKNOWN, AVAIL, EMPTY, FULL } state;
231 region_t region; /* related region */
232 char *chunks; /* chunk base */
233 slglobaldata_t slgd; /* localized to thread else NULL */
237 * per-thread data + global depot
239 * NOTE: The magazine shortcut is only used for per-thread data.
241 #define NMAGSHORTCUT 16
243 struct slglobaldata {
244 spinlock_t lock; /* only used by slglobaldepot */
252 void *mag_shortcut[NMAGSHORTCUT];
254 struct slab_list full_zones; /* via entry */
260 #define SLAB_ZEROD 0x0001
263 * Misc constants. Note that allocations that are exact multiples of
264 * PAGE_SIZE, or exceed the zone limit, fall through to the kmem module.
265 * IN_SAME_PAGE_MASK is used to sanity-check the per-page free lists.
267 #define MIN_CHUNK_SIZE 8 /* in bytes */
268 #define MIN_CHUNK_MASK (MIN_CHUNK_SIZE - 1)
270 #define SAFLAG_ZERO 0x00000001
273 * The WEIRD_ADDR is used as known text to copy into free objects to
274 * try to create deterministic failure cases if the data is accessed after
277 * WARNING: A limited number of spinlocks are available, BIGXSIZE should
278 * not be larger then 64.
281 #define WEIRD_ADDR 0xdeadc0de
288 #define MASSERT(exp) do { if (__predict_false(!(exp))) \
289 _mpanic("assertion: %s in %s", \
294 * With this attribute set, do not require a function call for accessing
295 * this variable when the code is compiled -fPIC.
297 * Must be empty for libc_rtld (similar to __thread)
299 #if defined(__LIBC_RTLD)
300 #define TLS_ATTRIBUTE
302 #define TLS_ATTRIBUTE __attribute__ ((tls_model ("initial-exec")));
305 static __thread struct slglobaldata slglobal TLS_ATTRIBUTE;
306 static pthread_key_t thread_malloc_key;
307 static pthread_once_t thread_malloc_once = PTHREAD_ONCE_INIT;
308 static struct slglobaldata slglobaldepot;
310 static int opt_madvise = 0;
311 static int opt_free = 0;
312 static int opt_cache = 4;
313 static int opt_utrace = 0;
314 static int g_malloc_flags = 0;
315 static int malloc_panic;
318 static const int32_t weirdary[16] = {
319 WEIRD_ADDR, WEIRD_ADDR, WEIRD_ADDR, WEIRD_ADDR,
320 WEIRD_ADDR, WEIRD_ADDR, WEIRD_ADDR, WEIRD_ADDR,
321 WEIRD_ADDR, WEIRD_ADDR, WEIRD_ADDR, WEIRD_ADDR,
322 WEIRD_ADDR, WEIRD_ADDR, WEIRD_ADDR, WEIRD_ADDR
326 static void *memalloc(size_t size, int flags);
327 static void *memrealloc(void *ptr, size_t size);
328 static void memfree(void *ptr, int);
329 static int memalign(void **memptr, size_t alignment, size_t size);
330 static slab_t slaballoc(int zi, size_t chunking, size_t chunk_size);
331 static void slabfree(slab_t slab);
332 static void slabterm(slglobaldata_t slgd, slab_t slab);
333 static void *_vmem_alloc(int ri, size_t slab_size);
334 static void _vmem_free(void *ptr, size_t slab_size);
335 static void _mpanic(const char *ctl, ...) __printflike(1, 2);
336 #ifndef STANDALONE_DEBUG
337 static void malloc_init(void) __constructor(101);
339 static void malloc_init(void) __constructor(101);
343 struct nmalloc_utrace {
349 #define UTRACE(a, b, c) \
351 struct nmalloc_utrace ut = { \
356 utrace(&ut, sizeof(ut)); \
361 * If enabled any memory allocated without M_ZERO is initialized to -1.
363 static int use_malloc_pattern;
369 const char *p = NULL;
371 TAILQ_INIT(&slglobal.full_zones);
373 Regions[0].mask = -1; /* disallow activity in lowest region */
375 if (issetugid() == 0)
376 p = getenv("MALLOC_OPTIONS");
378 for (; p != NULL && *p != '\0'; p++) {
409 g_malloc_flags = SAFLAG_ZERO;
416 UTRACE((void *) -1, 0, NULL);
420 * We have to install a handler for nmalloc thread teardowns when
421 * the thread is created. We cannot delay this because destructors in
422 * sophisticated userland programs can call malloc() for the first time
423 * during their thread exit.
425 * This routine is called directly from pthreads.
427 static void _nmalloc_thr_init_once(void);
428 static void _nmalloc_thr_destructor(void *thrp);
431 _nmalloc_thr_init(void)
435 TAILQ_INIT(&slglobal.full_zones);
443 pthread_once(&thread_malloc_once, _nmalloc_thr_init_once);
445 pthread_setspecific(thread_malloc_key, &slglobal);
450 _nmalloc_thr_prepfork(void)
453 _SPINLOCK(&slglobaldepot.lock);
457 _nmalloc_thr_parentfork(void)
460 _SPINUNLOCK(&slglobaldepot.lock);
464 _nmalloc_thr_childfork(void)
467 _SPINUNLOCK(&slglobaldepot.lock);
474 _nmalloc_thr_init_once(void)
478 error = pthread_key_create(&thread_malloc_key, _nmalloc_thr_destructor);
484 * Called for each thread undergoing exit
486 * Move all of the thread's slabs into a depot.
489 _nmalloc_thr_destructor(void *thrp)
491 slglobaldata_t slgd = thrp;
492 struct zoneinfo *zinfo;
500 for (i = 0; i <= slgd->biggest_index; i++) {
501 zinfo = &slgd->zone[i];
503 while ((j = zinfo->mag_index) > 0) {
505 ptr = zinfo->mag_shortcut[j];
506 zinfo->mag_shortcut[j] = NULL; /* SAFETY */
507 zinfo->mag_index = j;
511 while ((slab = zinfo->empty_base) != NULL) {
512 zinfo->empty_base = slab->next;
513 --zinfo->empty_count;
514 slabterm(slgd, slab);
517 while ((slab = zinfo->avail_base) != NULL) {
518 zinfo->avail_base = slab->next;
519 --zinfo->avail_count;
520 slabterm(slgd, slab);
523 while ((slab = TAILQ_FIRST(&slgd->full_zones)) != NULL) {
524 TAILQ_REMOVE(&slgd->full_zones, slab, entry);
525 slabterm(slgd, slab);
531 * Calculate the zone index for the allocation request size and set the
532 * allocation request size to that particular zone's chunk size.
534 * Minimum alignment is 16 bytes for allocations >= 16 bytes to conform
535 * with malloc requirements for intel/amd.
538 zoneindex(size_t *bytes, size_t *chunking)
540 size_t n = (size_t)*bytes;
547 *bytes = n = (n + 7) & ~7;
549 return(n / 8 - 1); /* 8 byte chunks, 2 zones */
551 *bytes = n = (n + 15) & ~15;
553 return(n / 16 + 2); /* 16 byte chunks, 8 zones */
558 c = x / (CHUNKFACTOR * 2);
562 c = x / (CHUNKFACTOR * 2);
563 i = 16 + CHUNKFACTOR * 5; /* 256->512,1024,2048,4096,8192 */
570 _mpanic("slaballoc: byte value too high");
572 *bytes = n = roundup2(n, c);
574 return (i + n / c - CHUNKFACTOR);
576 *bytes = n = (n + c - 1) & ~(c - 1);
581 *bytes = n = (n + 15) & ~15;
583 return(n / (CHUNKINGLO*2) + CHUNKINGLO*1 - 1);
587 *bytes = n = (n + 31) & ~31;
589 return(n / (CHUNKINGLO*4) + CHUNKINGLO*2 - 1);
592 *bytes = n = (n + 63) & ~63;
594 return(n / (CHUNKINGLO*8) + CHUNKINGLO*3 - 1);
597 *bytes = n = (n + 127) & ~127;
599 return(n / (CHUNKINGLO*16) + CHUNKINGLO*4 - 1);
602 *bytes = n = (n + 255) & ~255;
604 return(n / (CHUNKINGLO*32) + CHUNKINGLO*5 - 1);
606 *bytes = n = (n + 511) & ~511;
608 return(n / (CHUNKINGLO*64) + CHUNKINGLO*6 - 1);
611 *bytes = n = (n + 1023) & ~1023;
613 return(n / (CHUNKINGLO*128) + CHUNKINGLO*7 - 1);
615 if (n < 32768) { /* 16384-32767 */
616 *bytes = n = (n + 2047) & ~2047;
618 return(n / (CHUNKINGLO*256) + CHUNKINGLO*8 - 1);
621 *bytes = n = (n + 4095) & ~4095; /* 32768-65535 */
623 return(n / (CHUNKINGLO*512) + CHUNKINGLO*9 - 1);
628 i = CHUNKINGLO*10 - 1;
635 _mpanic("slaballoc: byte value too high");
637 *bytes = n = (n + c - 1) & ~(c - 1);
644 * malloc() - call internal slab allocator
647 __malloc(size_t size)
651 ptr = memalloc(size, 0);
655 UTRACE(0, size, ptr);
660 * calloc() - call internal slab allocator
663 __calloc(size_t number, size_t size)
667 ptr = memalloc(number * size, SAFLAG_ZERO);
671 UTRACE(0, number * size, ptr);
676 * realloc() (SLAB ALLOCATOR)
678 * We do not attempt to optimize this routine beyond reusing the same
679 * pointer if the new size fits within the chunking of the old pointer's
683 __realloc(void *ptr, size_t size)
688 ret = memalloc(size, 0);
690 ret = memrealloc(ptr, size);
694 UTRACE(ptr, size, ret);
701 * Allocate (size) bytes with a alignment of (alignment).
704 __aligned_alloc(size_t alignment, size_t size)
710 rc = memalign(&ptr, alignment, size);
720 * Allocate (size) bytes with a alignment of (alignment), where (alignment)
721 * is a power of 2 >= sizeof(void *).
724 __posix_memalign(void **memptr, size_t alignment, size_t size)
729 * OpenGroup spec issue 6 check
731 if (alignment < sizeof(void *)) {
736 rc = memalign(memptr, alignment, size);
742 * The slab allocator will allocate on power-of-2 boundaries up to at least
743 * PAGE_SIZE. Otherwise we use the zoneindex mechanic to find a zone
744 * matching the requirements.
747 memalign(void **memptr, size_t alignment, size_t size)
756 * OpenGroup spec issue 6 check
758 if ((alignment | (alignment - 1)) + 1 != (alignment << 1)) {
764 * XXX for now just find the nearest power of 2 >= size and also
765 * >= alignment and allocate that.
767 while (alignment < size) {
770 _mpanic("posix_memalign: byte value too high");
772 *memptr = memalloc(alignment, 0);
773 return(*memptr ? 0 : ENOMEM);
777 * free() (SLAB ALLOCATOR) - do the obvious
789 * memalloc() (SLAB ALLOCATOR)
791 * Allocate memory via the slab allocator.
794 memalloc(size_t size, int flags)
797 struct zoneinfo *zinfo;
811 * If 0 bytes is requested we have to return a unique pointer, allocate
817 /* Capture global flags */
818 flags |= g_malloc_flags;
820 /* Compute allocation zone; zoneindex will panic on excessive sizes */
821 zi = zoneindex(&size, &chunking);
822 MASSERT(zi < NZONES);
827 * Try magazine shortcut first
830 zinfo = &slgd->zone[zi];
832 if ((j = zinfo->mag_index) != 0) {
833 zinfo->mag_index = --j;
834 obj = zinfo->mag_shortcut[j];
835 zinfo->mag_shortcut[j] = NULL; /* SAFETY */
836 if (flags & SAFLAG_ZERO)
842 * Locate a slab with available space. If no slabs are available
843 * back-off to the empty list and if we still come up dry allocate
844 * a new slab (which will try the depot first).
847 if ((slab = zinfo->avail_base) == NULL) {
848 if ((slab = zinfo->empty_base) == NULL) {
852 slab = slaballoc(zi, chunking, size);
855 slab->next = zinfo->avail_base;
856 zinfo->avail_base = slab;
857 ++zinfo->avail_count;
859 if (slgd->biggest_index < zi)
860 slgd->biggest_index = zi;
864 * Pulled from empty list
866 zinfo->empty_base = slab->next;
867 slab->next = zinfo->avail_base;
868 zinfo->avail_base = slab;
869 ++zinfo->avail_count;
871 --zinfo->empty_count;
876 * Allocate a chunk out of the slab. HOT PATH
878 * Only the thread owning the slab can allocate out of it.
880 * NOTE: The last bit in the bitmap is always marked allocated so
881 * we cannot overflow here.
883 bno = slab->free_bit;
884 bmi = slab->free_index;
885 bmp = &slab->bitmap[bmi];
886 if (*bmp & (1LU << bno)) {
887 atomic_clear_long(bmp, 1LU << bno);
888 obj = slab->chunks + ((bmi << LONG_BITS_SHIFT) + bno) * size;
889 slab->free_bit = (bno + 1) & (LONG_BITS - 1);
890 atomic_add_int(&slab->navail, -1);
891 if (flags & SAFLAG_ZERO)
897 * Allocate a chunk out of a slab. COLD PATH
899 if (slab->navail == 0) {
900 zinfo->avail_base = slab->next;
901 --zinfo->avail_count;
903 TAILQ_INSERT_TAIL(&slgd->full_zones, slab, entry);
907 while (bmi < LONG_BITS) {
908 bmp = &slab->bitmap[bmi];
911 atomic_clear_long(bmp, 1LU << bno);
912 obj = slab->chunks + ((bmi << LONG_BITS_SHIFT) + bno) *
914 slab->free_index = bmi;
915 slab->free_bit = (bno + 1) & (LONG_BITS - 1);
916 atomic_add_int(&slab->navail, -1);
917 if (flags & SAFLAG_ZERO)
924 while (bmi < LONG_BITS) {
925 bmp = &slab->bitmap[bmi];
928 atomic_clear_long(bmp, 1LU << bno);
929 obj = slab->chunks + ((bmi << LONG_BITS_SHIFT) + bno) *
931 slab->free_index = bmi;
932 slab->free_bit = (bno + 1) & (LONG_BITS - 1);
933 atomic_add_int(&slab->navail, -1);
934 if (flags & SAFLAG_ZERO)
940 _mpanic("slaballoc: corrupted zone: navail %d", slab->navail);
946 * Reallocate memory within the chunk
949 memrealloc(void *ptr, size_t nsize)
958 * If 0 bytes is requested we have to return a unique pointer, allocate
964 /* Capture global flags */
965 flags |= g_malloc_flags;
968 * Locate the zone by looking up the dynamic slab size mask based
969 * on the memory region the allocation resides in.
971 region = &Regions[((uintptr_t)ptr >> NREGIONS_SHIFT) & NREGIONS_MASK];
972 if ((slab = region->slab) == NULL)
973 slab = (void *)((uintptr_t)ptr & region->mask);
974 MASSERT(slab->magic == ZALLOC_SLAB_MAGIC);
975 osize = slab->chunk_size;
976 if (nsize <= osize) {
977 if (osize < 32 || nsize >= osize / 2) {
979 if ((flags & SAFLAG_ZERO) && nsize < osize)
980 bzero(obj + nsize, osize - nsize);
986 * Otherwise resize the object
988 obj = memalloc(nsize, 0);
992 bcopy(ptr, obj, nsize);
999 * free (SLAB ALLOCATOR)
1001 * Free a memory block previously allocated by malloc.
1006 memfree(void *ptr, int flags)
1009 slglobaldata_t slgd;
1019 * Locate the zone by looking up the dynamic slab size mask based
1020 * on the memory region the allocation resides in.
1022 * WARNING! The slab may be owned by another thread!
1024 region = &Regions[((uintptr_t)ptr >> NREGIONS_SHIFT) & NREGIONS_MASK];
1025 if ((slab = region->slab) == NULL)
1026 slab = (void *)((uintptr_t)ptr & region->mask);
1027 MASSERT(slab != NULL);
1028 MASSERT(slab->magic == ZALLOC_SLAB_MAGIC);
1032 * Put weird data into the memory to detect modifications after
1033 * freeing, illegal pointer use after freeing (we should fault on
1034 * the odd address), and so forth.
1036 if (slab->chunk_size < sizeof(weirdary))
1037 bcopy(weirdary, ptr, slab->chunk_size);
1039 bcopy(weirdary, ptr, sizeof(weirdary));
1044 * Use mag_shortcut[] when possible
1046 if (slgd->masked == 0 && slab->chunk_size <= NOMSLABSIZE) {
1047 struct zoneinfo *zinfo;
1049 zinfo = &slgd->zone[slab->zone_index];
1050 j = zinfo->mag_index;
1051 if (j < NMAGSHORTCUT) {
1052 zinfo->mag_shortcut[j] = ptr;
1053 zinfo->mag_index = j + 1;
1059 * Free to slab and increment navail. We can delay incrementing
1060 * navail to prevent the slab from being destroyed out from under
1061 * us while we do other optimizations.
1063 bno = ((uintptr_t)ptr - (uintptr_t)slab->chunks) / slab->chunk_size;
1064 bmi = bno >> LONG_BITS_SHIFT;
1065 bno &= (LONG_BITS - 1);
1066 bmp = &slab->bitmap[bmi];
1068 MASSERT(bmi >= 0 && bmi < slab->nmax);
1069 MASSERT((*bmp & (1LU << bno)) == 0);
1070 atomic_set_long(bmp, 1LU << bno);
1072 if (slab->slgd == slgd) {
1074 * We can only do the following if we own the slab. Note
1075 * that navail can be incremented by any thread even if
1078 struct zoneinfo *zinfo;
1080 atomic_add_int(&slab->navail, 1);
1081 if (slab->free_index > bmi) {
1082 slab->free_index = bmi;
1083 slab->free_bit = bno;
1084 } else if (slab->free_index == bmi &&
1085 slab->free_bit > bno) {
1086 slab->free_bit = bno;
1088 zinfo = &slgd->zone[slab->zone_index];
1091 * Freeing an object from a full slab makes it less than
1092 * full. The slab must be moved to the available list.
1094 * If the available list has too many slabs, release some
1097 if (slab->state == FULL) {
1098 TAILQ_REMOVE(&slgd->full_zones, slab, entry);
1099 slab->state = AVAIL;
1100 stmp = zinfo->avail_base;
1102 zinfo->avail_base = slab;
1103 ++zinfo->avail_count;
1104 while (zinfo->avail_count > opt_cache) {
1105 slab = zinfo->avail_base;
1106 zinfo->avail_base = slab->next;
1107 --zinfo->avail_count;
1108 slabterm(slgd, slab);
1114 * If the slab becomes completely empty dispose of it in
1115 * some manner. By default each thread caches up to 4
1116 * empty slabs. Only small slabs are cached.
1118 if (slab->navail == slab->nmax && slab->state == AVAIL) {
1120 * Remove slab from available queue
1122 slabp = &zinfo->avail_base;
1123 while ((stmp = *slabp) != slab)
1124 slabp = &stmp->next;
1125 *slabp = slab->next;
1126 --zinfo->avail_count;
1128 if (opt_free || opt_cache == 0) {
1130 * If local caching is disabled cache the
1131 * slab in the depot (or free it).
1133 slabterm(slgd, slab);
1134 } else if (slab->slab_size > BIGSLABSIZE) {
1136 * We do not try to retain large slabs
1137 * in per-thread caches.
1139 slabterm(slgd, slab);
1140 } else if (zinfo->empty_count > opt_cache) {
1142 * We have too many slabs cached, but
1143 * instead of freeing this one free
1144 * an empty slab that's been idle longer.
1146 * (empty_count does not change)
1148 stmp = zinfo->empty_base;
1149 slab->state = EMPTY;
1150 slab->next = stmp->next;
1151 zinfo->empty_base = slab;
1152 slabterm(slgd, stmp);
1155 * Cache the empty slab in our thread local
1158 ++zinfo->empty_count;
1159 slab->state = EMPTY;
1160 slab->next = zinfo->empty_base;
1161 zinfo->empty_base = slab;
1164 } else if (slab->slgd == NULL && slab->navail + 1 == slab->nmax) {
1165 slglobaldata_t sldepot;
1168 * If freeing to a slab owned by the global depot, and
1169 * the slab becomes completely EMPTY, try to move it to
1172 sldepot = &slglobaldepot;
1174 _SPINLOCK(&sldepot->lock);
1175 if (slab->slgd == NULL && slab->navail + 1 == slab->nmax) {
1176 struct zoneinfo *zinfo;
1179 * Move the slab to the empty list
1181 MASSERT(slab->state == AVAIL);
1182 atomic_add_int(&slab->navail, 1);
1183 zinfo = &sldepot->zone[slab->zone_index];
1184 slabp = &zinfo->avail_base;
1185 while (slab != *slabp)
1186 slabp = &(*slabp)->next;
1187 *slabp = slab->next;
1188 --zinfo->avail_count;
1191 * Clean out excessive empty entries from the
1194 slab->state = EMPTY;
1195 slab->next = zinfo->empty_base;
1196 zinfo->empty_base = slab;
1197 ++zinfo->empty_count;
1198 while (zinfo->empty_count > opt_cache) {
1199 slab = zinfo->empty_base;
1200 zinfo->empty_base = slab->next;
1201 --zinfo->empty_count;
1202 slab->state = UNKNOWN;
1204 _SPINUNLOCK(&sldepot->lock);
1207 _SPINLOCK(&sldepot->lock);
1210 atomic_add_int(&slab->navail, 1);
1213 _SPINUNLOCK(&sldepot->lock);
1216 * We can't act on the slab other than by adjusting navail
1217 * (and the bitmap which we did in the common code at the
1220 atomic_add_int(&slab->navail, 1);
1227 * Allocate a new slab holding objects of size chunk_size.
1230 slaballoc(int zi, size_t chunking, size_t chunk_size)
1232 slglobaldata_t slgd;
1233 slglobaldata_t sldepot;
1234 struct zoneinfo *zinfo;
1241 uintptr_t chunk_offset;
1252 * First look in the depot. Any given zone in the depot may be
1253 * locked by being set to -1. We have to do this instead of simply
1254 * removing the entire chain because removing the entire chain can
1255 * cause racing threads to allocate local slabs for large objects,
1256 * resulting in a large VSZ.
1259 sldepot = &slglobaldepot;
1260 zinfo = &sldepot->zone[zi];
1262 if (zinfo->avail_base) {
1264 _SPINLOCK(&sldepot->lock);
1265 slab = zinfo->avail_base;
1267 MASSERT(slab->slgd == NULL);
1269 zinfo->avail_base = slab->next;
1270 --zinfo->avail_count;
1272 _SPINUNLOCK(&sldepot->lock);
1276 _SPINUNLOCK(&sldepot->lock);
1280 * Nothing in the depot, allocate a new slab by locating or assigning
1281 * a region and then using the system virtual memory allocator.
1286 * Calculate the start of the data chunks relative to the start
1287 * of the slab. If chunk_size is a power of 2 we guarantee
1288 * power of 2 alignment. If it is not we guarantee alignment
1289 * to the chunk size.
1291 if ((chunk_size ^ (chunk_size - 1)) == (chunk_size << 1) - 1) {
1293 chunk_offset = roundup2(sizeof(*slab), chunk_size);
1296 chunk_offset = sizeof(*slab) + chunking - 1;
1297 chunk_offset -= chunk_offset % chunking;
1301 * Calculate a reasonable number of chunks for the slab.
1303 * Once initialized the MaxChunks[] array can only ever be
1304 * reinitialized to the same value.
1306 maxchunks = MaxChunks[zi];
1307 if (maxchunks == 0) {
1309 * First calculate how many chunks would fit in 1/1024
1310 * available memory. This is around 2MB on a 32 bit
1311 * system and 128G on a 64-bit (48-bits available) system.
1313 maxchunks = (ssize_t)(NREGIONS_SIZE - chunk_offset) /
1314 (ssize_t)chunk_size;
1319 * A slab cannot handle more than MAXCHUNKS chunks, but
1320 * limit us to approximately MAXCHUNKS / 2 here because
1321 * we may have to expand maxchunks when we calculate the
1322 * actual power-of-2 slab.
1324 if (maxchunks > MAXCHUNKS / 2)
1325 maxchunks = MAXCHUNKS / 2;
1328 * Try to limit the slabs to BIGSLABSIZE (~128MB). Larger
1329 * slabs will be created if the allocation does not fit.
1331 if (chunk_offset + chunk_size * maxchunks > BIGSLABSIZE) {
1332 tmpchunks = (ssize_t)(BIGSLABSIZE - chunk_offset) /
1333 (ssize_t)chunk_size;
1336 if (maxchunks > tmpchunks)
1337 maxchunks = tmpchunks;
1341 * If the slab calculates to greater than 2MB see if we
1342 * can cut it down to ~2MB. This controls VSZ but has
1343 * no effect on run-time size or performance.
1345 * This is very important in case you core dump and also
1346 * important to reduce unnecessary region allocations.
1348 if (chunk_offset + chunk_size * maxchunks > NOMSLABSIZE) {
1349 tmpchunks = (ssize_t)(NOMSLABSIZE - chunk_offset) /
1350 (ssize_t)chunk_size;
1353 if (maxchunks > tmpchunks)
1354 maxchunks = tmpchunks;
1358 * If the slab calculates to greater than 128K see if we
1359 * can cut it down to ~128K while still maintaining a
1360 * reasonably large number of chunks in each slab. This
1361 * controls VSZ but has no effect on run-time size or
1364 * This is very important in case you core dump and also
1365 * important to reduce unnecessary region allocations.
1367 if (chunk_offset + chunk_size * maxchunks > LITSLABSIZE) {
1368 tmpchunks = (ssize_t)(LITSLABSIZE - chunk_offset) /
1369 (ssize_t)chunk_size;
1372 if (maxchunks > tmpchunks)
1373 maxchunks = tmpchunks;
1376 MaxChunks[zi] = maxchunks;
1378 MASSERT(maxchunks > 0 && maxchunks <= MAXCHUNKS);
1381 * Calculate the actual slab size. maxchunks will be recalculated
1384 slab_desire = chunk_offset + chunk_size * maxchunks;
1385 slab_size = 8 * MAXCHUNKS;
1386 power = 3 + MAXCHUNKS_BITS;
1387 while (slab_size < slab_desire) {
1393 * Do a quick recalculation based on the actual slab size but not
1394 * yet dealing with whether the slab header is in-band or out-of-band.
1395 * The purpose here is to see if we can reasonably reduce slab_size
1396 * to a power of 4 to allow more chunk sizes to use the same slab
1399 if ((power & 1) && slab_size > 32768) {
1400 maxchunks = (slab_size - chunk_offset) / chunk_size;
1401 if (maxchunks >= MAXCHUNKS / 8) {
1406 if ((power & 2) && slab_size > 32768 * 4) {
1407 maxchunks = (slab_size - chunk_offset) / chunk_size;
1408 if (maxchunks >= MAXCHUNKS / 4) {
1414 * This case occurs when the slab_size is larger than 1/1024 available
1417 nswath = slab_size / NREGIONS_SIZE;
1418 if (nswath > NREGIONS)
1423 * Try to allocate from our current best region for this zi
1425 region_mask = ~(slab_size - 1);
1426 ri = slgd->zone[zi].best_region;
1427 if (Regions[ri].mask == region_mask) {
1428 if ((slab = _vmem_alloc(ri, slab_size)) != NULL)
1433 * Try to find an existing region to allocate from. The normal
1434 * case will be for allocations that are less than 1/1024 available
1435 * UVM, which fit into a single Regions[] entry.
1437 while (slab_size <= NREGIONS_SIZE) {
1439 for (ri = 0; ri < NREGIONS; ++ri) {
1440 if (rx < 0 && Regions[ri].mask == 0)
1442 if (Regions[ri].mask == region_mask) {
1443 slab = _vmem_alloc(ri, slab_size);
1445 slgd->zone[zi].best_region = ri;
1455 * This can fail, retry either way
1457 atomic_cmpset_ptr((void **)&Regions[rx].mask,
1459 (void *)region_mask);
1464 for (ri = 0; ri < NREGIONS; ri += nswath) {
1465 if (Regions[ri].mask == region_mask) {
1466 slab = _vmem_alloc(ri, slab_size);
1468 slgd->zone[zi].best_region = ri;
1473 for (j = nswath - 1; j >= 0; --j) {
1474 if (Regions[ri+j].mask != 0)
1483 * We found a candidate, try to allocate it backwards so
1484 * another thread racing a slaballoc() does not see the
1485 * mask in the base index position until we are done.
1487 * We can safely zero-out any partial allocations because
1488 * the mask is only accessed from the base index. Any other
1489 * threads racing us will fail prior to us clearing the mask.
1493 for (j = nswath - 1; j >= 0; --j) {
1494 if (!atomic_cmpset_ptr((void **)&Regions[rx+j].mask,
1495 NULL, (void *)region_mask)) {
1496 while (++j < nswath)
1497 Regions[rx+j].mask = 0;
1505 * Fill in the new slab in region ri. If the slab_size completely
1506 * fills one or more region slots we move the slab structure out of
1507 * band which should optimize the chunking (particularly for a power
1511 region = &Regions[ri];
1512 MASSERT(region->slab == NULL);
1513 if (slab_size >= NREGIONS_SIZE) {
1515 slab = memalloc(sizeof(*slab), 0);
1516 bzero(slab, sizeof(*slab));
1517 slab->chunks = save;
1518 for (j = 0; j < nswath; ++j)
1519 region[j].slab = slab;
1522 bzero(slab, sizeof(*slab));
1523 slab->chunks = (char *)slab + chunk_offset;
1527 * Calculate the start of the chunks memory and recalculate the
1528 * actual number of chunks the slab can hold.
1530 maxchunks = (slab_size - chunk_offset) / chunk_size;
1531 if (maxchunks > MAXCHUNKS)
1532 maxchunks = MAXCHUNKS;
1535 * And fill in the rest
1537 slab->magic = ZALLOC_SLAB_MAGIC;
1538 slab->navail = maxchunks;
1539 slab->nmax = maxchunks;
1540 slab->slab_size = slab_size;
1541 slab->chunk_size = chunk_size;
1542 slab->zone_index = zi;
1544 slab->state = UNKNOWN;
1545 slab->region = region;
1547 for (ri = 0; ri < maxchunks; ri += LONG_BITS) {
1548 if (ri + LONG_BITS <= maxchunks)
1549 slab->bitmap[ri >> LONG_BITS_SHIFT] = ULONG_MAX;
1551 slab->bitmap[ri >> LONG_BITS_SHIFT] =
1552 (1LU << (maxchunks - ri)) - 1;
1561 slabfree(slab_t slab)
1566 if (slab->region->slab == slab) {
1570 nswath = slab->slab_size / NREGIONS_SIZE;
1571 for (j = 0; j < nswath; ++j)
1572 slab->region[j].slab = NULL;
1574 _vmem_free(slab->chunks, slab->slab_size);
1581 _vmem_free(slab, slab->slab_size);
1586 * Terminate a slab's use in the current thread. The slab may still have
1587 * outstanding allocations and thus not be deallocatable.
1590 slabterm(slglobaldata_t slgd, slab_t slab)
1592 slglobaldata_t sldepot;
1593 struct zoneinfo *zinfo;
1594 int zi = slab->zone_index;
1598 sldepot = &slglobaldepot;
1599 zinfo = &sldepot->zone[zi];
1602 * Move the slab to the avail list or the empty list.
1605 _SPINLOCK(&sldepot->lock);
1606 if (slab->navail == slab->nmax) {
1607 slab->state = EMPTY;
1608 slab->next = zinfo->empty_base;
1609 zinfo->empty_base = slab;
1610 ++zinfo->empty_count;
1612 slab->state = AVAIL;
1613 slab->next = zinfo->avail_base;
1614 zinfo->avail_base = slab;
1615 ++zinfo->avail_count;
1619 * Clean extra slabs out of the empty list
1621 while (zinfo->empty_count > opt_cache) {
1622 slab = zinfo->empty_base;
1623 zinfo->empty_base = slab->next;
1624 --zinfo->empty_count;
1625 slab->state = UNKNOWN;
1627 _SPINUNLOCK(&sldepot->lock);
1630 _SPINLOCK(&sldepot->lock);
1633 _SPINUNLOCK(&sldepot->lock);
1639 * Directly map memory in PAGE_SIZE'd chunks with the specified
1642 * Alignment must be a multiple of PAGE_SIZE.
1644 * Size must be >= alignment.
1647 _vmem_alloc(int ri, size_t slab_size)
1649 char *baddr = (void *)((uintptr_t)ri << NREGIONS_SHIFT);
1655 if (slab_size < NREGIONS_SIZE)
1656 eaddr = baddr + NREGIONS_SIZE;
1658 eaddr = baddr + slab_size;
1661 * This usually just works but might not.
1663 addr = mmap(baddr, slab_size, PROT_READ|PROT_WRITE,
1664 MAP_PRIVATE | MAP_ANON | MAP_SIZEALIGN, -1, 0);
1665 if (addr == MAP_FAILED) {
1668 if (addr < baddr || addr + slab_size > eaddr) {
1669 munmap(addr, slab_size);
1674 * Check alignment. The misaligned offset is also the excess
1675 * amount. If misaligned unmap the excess so we have a chance of
1676 * mapping at the next alignment point and recursively try again.
1678 * BBBBBBBBBBB BBBBBBBBBBB BBBBBBBBBBB block alignment
1679 * aaaaaaaaa aaaaaaaaaaa aa mis-aligned allocation
1680 * xxxxxxxxx final excess calculation
1681 * ^ returned address
1683 excess = (uintptr_t)addr & (slab_size - 1);
1685 excess = slab_size - excess;
1688 munmap(save + excess, slab_size - excess);
1689 addr = _vmem_alloc(ri, slab_size);
1690 munmap(save, excess);
1693 if (addr < baddr || addr + slab_size > eaddr) {
1694 munmap(addr, slab_size);
1697 excess = (uintptr_t)addr & (slab_size - 1);
1705 * Free a chunk of memory allocated with _vmem_alloc()
1708 _vmem_free(void *ptr, size_t size)
1714 * Panic on fatal conditions
1717 _mpanic(const char *ctl, ...)
1721 if (malloc_panic == 0) {
1724 vfprintf(stderr, ctl, va);
1725 fprintf(stderr, "\n");
1732 __weak_reference(__aligned_alloc, aligned_alloc);
1733 __weak_reference(__malloc, malloc);
1734 __weak_reference(__calloc, calloc);
1735 __weak_reference(__posix_memalign, posix_memalign);
1736 __weak_reference(__realloc, realloc);
1737 __weak_reference(__free, free);