jemalloc.c revision 1.4 1 1.4 ad /* $NetBSD: jemalloc.c,v 1.4 2007/10/09 00:59:52 ad Exp $ */
2 1.2 ad
3 1.1 ad /*-
4 1.1 ad * Copyright (C) 2006,2007 Jason Evans <jasone (at) FreeBSD.org>.
5 1.1 ad * All rights reserved.
6 1.1 ad *
7 1.1 ad * Redistribution and use in source and binary forms, with or without
8 1.1 ad * modification, are permitted provided that the following conditions
9 1.1 ad * are met:
10 1.1 ad * 1. Redistributions of source code must retain the above copyright
11 1.1 ad * notice(s), this list of conditions and the following disclaimer as
12 1.1 ad * the first lines of this file unmodified other than the possible
13 1.1 ad * addition of one or more copyright notices.
14 1.1 ad * 2. Redistributions in binary form must reproduce the above copyright
15 1.1 ad * notice(s), this list of conditions and the following disclaimer in
16 1.1 ad * the documentation and/or other materials provided with the
17 1.1 ad * distribution.
18 1.1 ad *
19 1.1 ad * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
20 1.1 ad * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 1.1 ad * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 1.1 ad * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE
23 1.1 ad * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 1.1 ad * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 1.1 ad * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
26 1.1 ad * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27 1.1 ad * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
28 1.1 ad * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
29 1.1 ad * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 1.1 ad *
31 1.1 ad *******************************************************************************
32 1.1 ad *
33 1.1 ad * This allocator implementation is designed to provide scalable performance
34 1.1 ad * for multi-threaded programs on multi-processor systems. The following
35 1.1 ad * features are included for this purpose:
36 1.1 ad *
37 1.1 ad * + Multiple arenas are used if there are multiple CPUs, which reduces lock
38 1.1 ad * contention and cache sloshing.
39 1.1 ad *
40 1.1 ad * + Cache line sharing between arenas is avoided for internal data
41 1.1 ad * structures.
42 1.1 ad *
43 1.1 ad * + Memory is managed in chunks and runs (chunks can be split into runs),
44 1.1 ad * rather than as individual pages. This provides a constant-time
45 1.1 ad * mechanism for associating allocations with particular arenas.
46 1.1 ad *
47 1.1 ad * Allocation requests are rounded up to the nearest size class, and no record
48 1.1 ad * of the original request size is maintained. Allocations are broken into
49 1.1 ad * categories according to size class. Assuming runtime defaults, 4 kB pages
50 1.1 ad * and a 16 byte quantum, the size classes in each category are as follows:
51 1.1 ad *
52 1.1 ad * |=====================================|
53 1.1 ad * | Category | Subcategory | Size |
54 1.1 ad * |=====================================|
55 1.1 ad * | Small | Tiny | 2 |
56 1.1 ad * | | | 4 |
57 1.1 ad * | | | 8 |
58 1.1 ad * | |----------------+---------|
59 1.1 ad * | | Quantum-spaced | 16 |
60 1.1 ad * | | | 32 |
61 1.1 ad * | | | 48 |
62 1.1 ad * | | | ... |
63 1.1 ad * | | | 480 |
64 1.1 ad * | | | 496 |
65 1.1 ad * | | | 512 |
66 1.1 ad * | |----------------+---------|
67 1.1 ad * | | Sub-page | 1 kB |
68 1.1 ad * | | | 2 kB |
69 1.1 ad * |=====================================|
70 1.1 ad * | Large | 4 kB |
71 1.1 ad * | | 8 kB |
72 1.1 ad * | | 12 kB |
73 1.1 ad * | | ... |
74 1.1 ad * | | 1012 kB |
75 1.1 ad * | | 1016 kB |
76 1.1 ad * | | 1020 kB |
77 1.1 ad * |=====================================|
78 1.1 ad * | Huge | 1 MB |
79 1.1 ad * | | 2 MB |
80 1.1 ad * | | 3 MB |
81 1.1 ad * | | ... |
82 1.1 ad * |=====================================|
83 1.1 ad *
84 1.1 ad * A different mechanism is used for each category:
85 1.1 ad *
86 1.1 ad * Small : Each size class is segregated into its own set of runs. Each run
87 1.1 ad * maintains a bitmap of which regions are free/allocated.
88 1.1 ad *
89 1.1 ad * Large : Each allocation is backed by a dedicated run. Metadata are stored
90 1.1 ad * in the associated arena chunk header maps.
91 1.1 ad *
92 1.1 ad * Huge : Each allocation is backed by a dedicated contiguous set of chunks.
93 1.1 ad * Metadata are stored in a separate red-black tree.
94 1.1 ad *
95 1.1 ad *******************************************************************************
96 1.1 ad */
97 1.1 ad
98 1.2 ad /* LINTLIBRARY */
99 1.2 ad
100 1.2 ad #ifdef __NetBSD__
101 1.2 ad # define xutrace(a, b) utrace("malloc", (a), (b))
102 1.2 ad # define __DECONST(x, y) ((x)__UNCONST(y))
103 1.2 ad # define NO_TLS
104 1.2 ad #else
105 1.2 ad # define xutrace(a, b) utrace((a), (b))
106 1.2 ad #endif /* __NetBSD__ */
107 1.2 ad
108 1.1 ad /*
109 1.1 ad * MALLOC_PRODUCTION disables assertions and statistics gathering. It also
110 1.1 ad * defaults the A and J runtime options to off. These settings are appropriate
111 1.1 ad * for production systems.
112 1.1 ad */
113 1.2 ad #define MALLOC_PRODUCTION
114 1.1 ad
115 1.1 ad #ifndef MALLOC_PRODUCTION
116 1.1 ad # define MALLOC_DEBUG
117 1.1 ad #endif
118 1.1 ad
119 1.1 ad #include <sys/cdefs.h>
120 1.2 ad /* __FBSDID("$FreeBSD: src/lib/libc/stdlib/malloc.c,v 1.147 2007/06/15 22:00:16 jasone Exp $"); */
121 1.4 ad __RCSID("$NetBSD: jemalloc.c,v 1.4 2007/10/09 00:59:52 ad Exp $");
122 1.1 ad
123 1.2 ad #ifdef __FreeBSD__
124 1.1 ad #include "libc_private.h"
125 1.1 ad #ifdef MALLOC_DEBUG
126 1.1 ad # define _LOCK_DEBUG
127 1.1 ad #endif
128 1.1 ad #include "spinlock.h"
129 1.2 ad #endif
130 1.1 ad #include "namespace.h"
131 1.1 ad #include <sys/mman.h>
132 1.1 ad #include <sys/param.h>
133 1.2 ad #ifdef __FreeBSD__
134 1.1 ad #include <sys/stddef.h>
135 1.2 ad #endif
136 1.1 ad #include <sys/time.h>
137 1.1 ad #include <sys/types.h>
138 1.1 ad #include <sys/sysctl.h>
139 1.1 ad #include <sys/tree.h>
140 1.1 ad #include <sys/uio.h>
141 1.1 ad #include <sys/ktrace.h> /* Must come after several other sys/ includes. */
142 1.1 ad
143 1.2 ad #ifdef __FreeBSD__
144 1.1 ad #include <machine/atomic.h>
145 1.1 ad #include <machine/cpufunc.h>
146 1.2 ad #endif
147 1.1 ad #include <machine/vmparam.h>
148 1.1 ad
149 1.1 ad #include <errno.h>
150 1.1 ad #include <limits.h>
151 1.1 ad #include <pthread.h>
152 1.1 ad #include <sched.h>
153 1.1 ad #include <stdarg.h>
154 1.1 ad #include <stdbool.h>
155 1.1 ad #include <stdio.h>
156 1.1 ad #include <stdint.h>
157 1.1 ad #include <stdlib.h>
158 1.1 ad #include <string.h>
159 1.1 ad #include <strings.h>
160 1.1 ad #include <unistd.h>
161 1.1 ad
162 1.2 ad #ifdef __NetBSD__
163 1.2 ad # include <reentrant.h>
164 1.2 ad void _malloc_prefork(void);
165 1.2 ad void _malloc_postfork(void);
166 1.2 ad ssize_t _write(int, const void *, size_t);
167 1.2 ad const char *_getprogname(void);
168 1.2 ad #endif
169 1.2 ad
170 1.2 ad #ifdef __FreeBSD__
171 1.1 ad #include "un-namespace.h"
172 1.2 ad #endif
173 1.1 ad
174 1.1 ad /* MALLOC_STATS enables statistics calculation. */
175 1.1 ad #ifndef MALLOC_PRODUCTION
176 1.1 ad # define MALLOC_STATS
177 1.1 ad #endif
178 1.1 ad
179 1.1 ad #ifdef MALLOC_DEBUG
180 1.1 ad # ifdef NDEBUG
181 1.1 ad # undef NDEBUG
182 1.1 ad # endif
183 1.1 ad #else
184 1.1 ad # ifndef NDEBUG
185 1.1 ad # define NDEBUG
186 1.1 ad # endif
187 1.1 ad #endif
188 1.1 ad #include <assert.h>
189 1.1 ad
190 1.1 ad #ifdef MALLOC_DEBUG
191 1.1 ad /* Disable inlining to make debugging easier. */
192 1.1 ad # define inline
193 1.1 ad #endif
194 1.1 ad
195 1.1 ad /* Size of stack-allocated buffer passed to strerror_r(). */
196 1.1 ad #define STRERROR_BUF 64
197 1.1 ad
198 1.1 ad /* Minimum alignment of allocations is 2^QUANTUM_2POW_MIN bytes. */
199 1.1 ad #ifdef __i386__
200 1.1 ad # define QUANTUM_2POW_MIN 4
201 1.1 ad # define SIZEOF_PTR_2POW 2
202 1.1 ad # define USE_BRK
203 1.1 ad #endif
204 1.1 ad #ifdef __ia64__
205 1.1 ad # define QUANTUM_2POW_MIN 4
206 1.1 ad # define SIZEOF_PTR_2POW 3
207 1.1 ad #endif
208 1.1 ad #ifdef __alpha__
209 1.1 ad # define QUANTUM_2POW_MIN 4
210 1.1 ad # define SIZEOF_PTR_2POW 3
211 1.1 ad # define NO_TLS
212 1.1 ad #endif
213 1.1 ad #ifdef __sparc64__
214 1.1 ad # define QUANTUM_2POW_MIN 4
215 1.1 ad # define SIZEOF_PTR_2POW 3
216 1.1 ad # define NO_TLS
217 1.1 ad #endif
218 1.1 ad #ifdef __amd64__
219 1.1 ad # define QUANTUM_2POW_MIN 4
220 1.1 ad # define SIZEOF_PTR_2POW 3
221 1.1 ad #endif
222 1.1 ad #ifdef __arm__
223 1.1 ad # define QUANTUM_2POW_MIN 3
224 1.1 ad # define SIZEOF_PTR_2POW 2
225 1.1 ad # define USE_BRK
226 1.1 ad # define NO_TLS
227 1.1 ad #endif
228 1.1 ad #ifdef __powerpc__
229 1.1 ad # define QUANTUM_2POW_MIN 4
230 1.1 ad # define SIZEOF_PTR_2POW 2
231 1.1 ad # define USE_BRK
232 1.1 ad #endif
233 1.3 he #if defined(__sparc__) && !defined(__sparc64__)
234 1.2 ad # define QUANTUM_2POW_MIN 4
235 1.2 ad # define SIZEOF_PTR_2POW 2
236 1.2 ad # define USE_BRK
237 1.2 ad #endif
238 1.2 ad #ifdef __vax__
239 1.2 ad # define QUANTUM_2POW_MIN 4
240 1.2 ad # define SIZEOF_PTR_2POW 2
241 1.2 ad # define USE_BRK
242 1.2 ad #endif
243 1.2 ad #ifdef __sh__
244 1.2 ad # define QUANTUM_2POW_MIN 4
245 1.2 ad # define SIZEOF_PTR_2POW 2
246 1.2 ad # define USE_BRK
247 1.2 ad #endif
248 1.2 ad #ifdef __m68k__
249 1.2 ad # define QUANTUM_2POW_MIN 4
250 1.2 ad # define SIZEOF_PTR_2POW 2
251 1.2 ad # define USE_BRK
252 1.2 ad #endif
253 1.2 ad #ifdef __mips__
254 1.2 ad # define QUANTUM_2POW_MIN 4
255 1.2 ad # define SIZEOF_PTR_2POW 2
256 1.2 ad # define USE_BRK
257 1.2 ad #endif
258 1.4 ad #ifdef __hppa__
259 1.4 ad # define QUANTUM_2POW_MIN 4
260 1.4 ad # define SIZEOF_PTR_2POW 2
261 1.4 ad # define USE_BRK
262 1.4 ad #endif
263 1.1 ad
264 1.1 ad #define SIZEOF_PTR (1 << SIZEOF_PTR_2POW)
265 1.1 ad
266 1.1 ad /* sizeof(int) == (1 << SIZEOF_INT_2POW). */
267 1.1 ad #ifndef SIZEOF_INT_2POW
268 1.1 ad # define SIZEOF_INT_2POW 2
269 1.1 ad #endif
270 1.1 ad
271 1.1 ad /* We can't use TLS in non-PIC programs, since TLS relies on loader magic. */
272 1.1 ad #if (!defined(PIC) && !defined(NO_TLS))
273 1.1 ad # define NO_TLS
274 1.1 ad #endif
275 1.1 ad
276 1.1 ad /*
277 1.1 ad * Size and alignment of memory chunks that are allocated by the OS's virtual
278 1.1 ad * memory system.
279 1.1 ad */
280 1.1 ad #define CHUNK_2POW_DEFAULT 20
281 1.1 ad
282 1.1 ad /*
283 1.1 ad * Maximum size of L1 cache line. This is used to avoid cache line aliasing,
284 1.1 ad * so over-estimates are okay (up to a point), but under-estimates will
285 1.1 ad * negatively affect performance.
286 1.1 ad */
287 1.1 ad #define CACHELINE_2POW 6
288 1.1 ad #define CACHELINE ((size_t)(1 << CACHELINE_2POW))
289 1.1 ad
290 1.1 ad /* Smallest size class to support. */
291 1.1 ad #define TINY_MIN_2POW 1
292 1.1 ad
293 1.1 ad /*
294 1.1 ad * Maximum size class that is a multiple of the quantum, but not (necessarily)
295 1.1 ad * a power of 2. Above this size, allocations are rounded up to the nearest
296 1.1 ad * power of 2.
297 1.1 ad */
298 1.1 ad #define SMALL_MAX_2POW_DEFAULT 9
299 1.1 ad #define SMALL_MAX_DEFAULT (1 << SMALL_MAX_2POW_DEFAULT)
300 1.1 ad
301 1.1 ad /*
302 1.1 ad * Maximum desired run header overhead. Runs are sized as small as possible
303 1.1 ad * such that this setting is still honored, without violating other constraints.
304 1.1 ad * The goal is to make runs as small as possible without exceeding a per run
305 1.1 ad * external fragmentation threshold.
306 1.1 ad *
307 1.1 ad * Note that it is possible to set this low enough that it cannot be honored
308 1.1 ad * for some/all object sizes, since there is one bit of header overhead per
309 1.1 ad * object (plus a constant). In such cases, this constraint is relaxed.
310 1.1 ad *
311 1.1 ad * RUN_MAX_OVRHD_RELAX specifies the maximum number of bits per region of
312 1.1 ad * overhead for which RUN_MAX_OVRHD is relaxed.
313 1.1 ad */
314 1.1 ad #define RUN_MAX_OVRHD 0.015
315 1.1 ad #define RUN_MAX_OVRHD_RELAX 1.5
316 1.1 ad
317 1.1 ad /* Put a cap on small object run size. This overrides RUN_MAX_OVRHD. */
318 1.1 ad #define RUN_MAX_SMALL_2POW 15
319 1.1 ad #define RUN_MAX_SMALL (1 << RUN_MAX_SMALL_2POW)
320 1.1 ad
321 1.1 ad /******************************************************************************/
322 1.1 ad
323 1.2 ad #ifdef __FreeBSD__
324 1.1 ad /*
325 1.1 ad * Mutexes based on spinlocks. We can't use normal pthread mutexes, because
326 1.1 ad * they require malloc()ed memory.
327 1.1 ad */
328 1.1 ad typedef struct {
329 1.1 ad spinlock_t lock;
330 1.1 ad } malloc_mutex_t;
331 1.1 ad
332 1.1 ad /* Set to true once the allocator has been initialized. */
333 1.1 ad static bool malloc_initialized = false;
334 1.1 ad
335 1.1 ad /* Used to avoid initialization races. */
336 1.1 ad static malloc_mutex_t init_lock = {_SPINLOCK_INITIALIZER};
337 1.2 ad #else
338 1.2 ad #define malloc_mutex_t mutex_t
339 1.2 ad
340 1.2 ad /* Set to true once the allocator has been initialized. */
341 1.2 ad static bool malloc_initialized = false;
342 1.2 ad
343 1.2 ad /* Used to avoid initialization races. */
344 1.2 ad static mutex_t init_lock = MUTEX_INITIALIZER;
345 1.2 ad #endif
346 1.1 ad
347 1.1 ad /******************************************************************************/
348 1.1 ad /*
349 1.1 ad * Statistics data structures.
350 1.1 ad */
351 1.1 ad
352 1.1 ad #ifdef MALLOC_STATS
353 1.1 ad
354 1.1 ad typedef struct malloc_bin_stats_s malloc_bin_stats_t;
355 1.1 ad struct malloc_bin_stats_s {
356 1.1 ad /*
357 1.1 ad * Number of allocation requests that corresponded to the size of this
358 1.1 ad * bin.
359 1.1 ad */
360 1.1 ad uint64_t nrequests;
361 1.1 ad
362 1.1 ad /* Total number of runs created for this bin's size class. */
363 1.1 ad uint64_t nruns;
364 1.1 ad
365 1.1 ad /*
366 1.1 ad * Total number of runs reused by extracting them from the runs tree for
367 1.1 ad * this bin's size class.
368 1.1 ad */
369 1.1 ad uint64_t reruns;
370 1.1 ad
371 1.1 ad /* High-water mark for this bin. */
372 1.1 ad unsigned long highruns;
373 1.1 ad
374 1.1 ad /* Current number of runs in this bin. */
375 1.1 ad unsigned long curruns;
376 1.1 ad };
377 1.1 ad
378 1.1 ad typedef struct arena_stats_s arena_stats_t;
379 1.1 ad struct arena_stats_s {
380 1.1 ad /* Number of bytes currently mapped. */
381 1.1 ad size_t mapped;
382 1.1 ad
383 1.1 ad /* Per-size-category statistics. */
384 1.1 ad size_t allocated_small;
385 1.1 ad uint64_t nmalloc_small;
386 1.1 ad uint64_t ndalloc_small;
387 1.1 ad
388 1.1 ad size_t allocated_large;
389 1.1 ad uint64_t nmalloc_large;
390 1.1 ad uint64_t ndalloc_large;
391 1.1 ad };
392 1.1 ad
393 1.1 ad typedef struct chunk_stats_s chunk_stats_t;
394 1.1 ad struct chunk_stats_s {
395 1.1 ad /* Number of chunks that were allocated. */
396 1.1 ad uint64_t nchunks;
397 1.1 ad
398 1.1 ad /* High-water mark for number of chunks allocated. */
399 1.1 ad unsigned long highchunks;
400 1.1 ad
401 1.1 ad /*
402 1.1 ad * Current number of chunks allocated. This value isn't maintained for
403 1.1 ad * any other purpose, so keep track of it in order to be able to set
404 1.1 ad * highchunks.
405 1.1 ad */
406 1.1 ad unsigned long curchunks;
407 1.1 ad };
408 1.1 ad
409 1.1 ad #endif /* #ifdef MALLOC_STATS */
410 1.1 ad
411 1.1 ad /******************************************************************************/
412 1.1 ad /*
413 1.1 ad * Chunk data structures.
414 1.1 ad */
415 1.1 ad
416 1.1 ad /* Tree of chunks. */
417 1.1 ad typedef struct chunk_node_s chunk_node_t;
418 1.1 ad struct chunk_node_s {
419 1.1 ad /* Linkage for the chunk tree. */
420 1.1 ad RB_ENTRY(chunk_node_s) link;
421 1.1 ad
422 1.1 ad /*
423 1.1 ad * Pointer to the chunk that this tree node is responsible for. In some
424 1.1 ad * (but certainly not all) cases, this data structure is placed at the
425 1.1 ad * beginning of the corresponding chunk, so this field may point to this
426 1.1 ad * node.
427 1.1 ad */
428 1.1 ad void *chunk;
429 1.1 ad
430 1.1 ad /* Total chunk size. */
431 1.1 ad size_t size;
432 1.1 ad };
433 1.1 ad typedef struct chunk_tree_s chunk_tree_t;
434 1.1 ad RB_HEAD(chunk_tree_s, chunk_node_s);
435 1.1 ad
436 1.1 ad /******************************************************************************/
437 1.1 ad /*
438 1.1 ad * Arena data structures.
439 1.1 ad */
440 1.1 ad
441 1.1 ad typedef struct arena_s arena_t;
442 1.1 ad typedef struct arena_bin_s arena_bin_t;
443 1.1 ad
444 1.1 ad typedef struct arena_chunk_map_s arena_chunk_map_t;
445 1.1 ad struct arena_chunk_map_s {
446 1.1 ad /* Number of pages in run. */
447 1.1 ad uint32_t npages;
448 1.1 ad /*
449 1.1 ad * Position within run. For a free run, this is POS_FREE for the first
450 1.1 ad * and last pages. The POS_FREE special value makes it possible to
451 1.1 ad * quickly coalesce free runs.
452 1.1 ad *
453 1.1 ad * This is the limiting factor for chunksize; there can be at most 2^31
454 1.1 ad * pages in a run.
455 1.1 ad */
456 1.1 ad #define POS_FREE ((uint32_t)0xffffffffU)
457 1.1 ad uint32_t pos;
458 1.1 ad };
459 1.1 ad
460 1.1 ad /* Arena chunk header. */
461 1.1 ad typedef struct arena_chunk_s arena_chunk_t;
462 1.1 ad struct arena_chunk_s {
463 1.1 ad /* Arena that owns the chunk. */
464 1.1 ad arena_t *arena;
465 1.1 ad
466 1.1 ad /* Linkage for the arena's chunk tree. */
467 1.1 ad RB_ENTRY(arena_chunk_s) link;
468 1.1 ad
469 1.1 ad /*
470 1.1 ad * Number of pages in use. This is maintained in order to make
471 1.1 ad * detection of empty chunks fast.
472 1.1 ad */
473 1.1 ad uint32_t pages_used;
474 1.1 ad
475 1.1 ad /*
476 1.1 ad * Every time a free run larger than this value is created/coalesced,
477 1.1 ad * this value is increased. The only way that the value decreases is if
478 1.1 ad * arena_run_alloc() fails to find a free run as large as advertised by
479 1.1 ad * this value.
480 1.1 ad */
481 1.1 ad uint32_t max_frun_npages;
482 1.1 ad
483 1.1 ad /*
484 1.1 ad * Every time a free run that starts at an earlier page than this value
485 1.1 ad * is created/coalesced, this value is decreased. It is reset in a
486 1.1 ad * similar fashion to max_frun_npages.
487 1.1 ad */
488 1.1 ad uint32_t min_frun_ind;
489 1.1 ad
490 1.1 ad /*
491 1.1 ad * Map of pages within chunk that keeps track of free/large/small. For
492 1.1 ad * free runs, only the map entries for the first and last pages are
493 1.1 ad * kept up to date, so that free runs can be quickly coalesced.
494 1.1 ad */
495 1.1 ad arena_chunk_map_t map[1]; /* Dynamically sized. */
496 1.1 ad };
497 1.1 ad typedef struct arena_chunk_tree_s arena_chunk_tree_t;
498 1.1 ad RB_HEAD(arena_chunk_tree_s, arena_chunk_s);
499 1.1 ad
500 1.1 ad typedef struct arena_run_s arena_run_t;
501 1.1 ad struct arena_run_s {
502 1.1 ad /* Linkage for run trees. */
503 1.1 ad RB_ENTRY(arena_run_s) link;
504 1.1 ad
505 1.1 ad #ifdef MALLOC_DEBUG
506 1.1 ad uint32_t magic;
507 1.1 ad # define ARENA_RUN_MAGIC 0x384adf93
508 1.1 ad #endif
509 1.1 ad
510 1.1 ad /* Bin this run is associated with. */
511 1.1 ad arena_bin_t *bin;
512 1.1 ad
513 1.1 ad /* Index of first element that might have a free region. */
514 1.1 ad unsigned regs_minelm;
515 1.1 ad
516 1.1 ad /* Number of free regions in run. */
517 1.1 ad unsigned nfree;
518 1.1 ad
519 1.1 ad /* Bitmask of in-use regions (0: in use, 1: free). */
520 1.1 ad unsigned regs_mask[1]; /* Dynamically sized. */
521 1.1 ad };
522 1.1 ad typedef struct arena_run_tree_s arena_run_tree_t;
523 1.1 ad RB_HEAD(arena_run_tree_s, arena_run_s);
524 1.1 ad
525 1.1 ad struct arena_bin_s {
526 1.1 ad /*
527 1.1 ad * Current run being used to service allocations of this bin's size
528 1.1 ad * class.
529 1.1 ad */
530 1.1 ad arena_run_t *runcur;
531 1.1 ad
532 1.1 ad /*
533 1.1 ad * Tree of non-full runs. This tree is used when looking for an
534 1.1 ad * existing run when runcur is no longer usable. We choose the
535 1.1 ad * non-full run that is lowest in memory; this policy tends to keep
536 1.1 ad * objects packed well, and it can also help reduce the number of
537 1.1 ad * almost-empty chunks.
538 1.1 ad */
539 1.1 ad arena_run_tree_t runs;
540 1.1 ad
541 1.1 ad /* Size of regions in a run for this bin's size class. */
542 1.1 ad size_t reg_size;
543 1.1 ad
544 1.1 ad /* Total size of a run for this bin's size class. */
545 1.1 ad size_t run_size;
546 1.1 ad
547 1.1 ad /* Total number of regions in a run for this bin's size class. */
548 1.1 ad uint32_t nregs;
549 1.1 ad
550 1.1 ad /* Number of elements in a run's regs_mask for this bin's size class. */
551 1.1 ad uint32_t regs_mask_nelms;
552 1.1 ad
553 1.1 ad /* Offset of first region in a run for this bin's size class. */
554 1.1 ad uint32_t reg0_offset;
555 1.1 ad
556 1.1 ad #ifdef MALLOC_STATS
557 1.1 ad /* Bin statistics. */
558 1.1 ad malloc_bin_stats_t stats;
559 1.1 ad #endif
560 1.1 ad };
561 1.1 ad
562 1.1 ad struct arena_s {
563 1.1 ad #ifdef MALLOC_DEBUG
564 1.1 ad uint32_t magic;
565 1.1 ad # define ARENA_MAGIC 0x947d3d24
566 1.1 ad #endif
567 1.1 ad
568 1.1 ad /* All operations on this arena require that mtx be locked. */
569 1.1 ad malloc_mutex_t mtx;
570 1.1 ad
571 1.1 ad #ifdef MALLOC_STATS
572 1.1 ad arena_stats_t stats;
573 1.1 ad #endif
574 1.1 ad
575 1.1 ad /*
576 1.1 ad * Tree of chunks this arena manages.
577 1.1 ad */
578 1.1 ad arena_chunk_tree_t chunks;
579 1.1 ad
580 1.1 ad /*
581 1.1 ad * In order to avoid rapid chunk allocation/deallocation when an arena
582 1.1 ad * oscillates right on the cusp of needing a new chunk, cache the most
583 1.1 ad * recently freed chunk. This caching is disabled by opt_hint.
584 1.1 ad *
585 1.1 ad * There is one spare chunk per arena, rather than one spare total, in
586 1.1 ad * order to avoid interactions between multiple threads that could make
587 1.1 ad * a single spare inadequate.
588 1.1 ad */
589 1.1 ad arena_chunk_t *spare;
590 1.1 ad
591 1.1 ad /*
592 1.1 ad * bins is used to store rings of free regions of the following sizes,
593 1.1 ad * assuming a 16-byte quantum, 4kB pagesize, and default MALLOC_OPTIONS.
594 1.1 ad *
595 1.1 ad * bins[i] | size |
596 1.1 ad * --------+------+
597 1.1 ad * 0 | 2 |
598 1.1 ad * 1 | 4 |
599 1.1 ad * 2 | 8 |
600 1.1 ad * --------+------+
601 1.1 ad * 3 | 16 |
602 1.1 ad * 4 | 32 |
603 1.1 ad * 5 | 48 |
604 1.1 ad * 6 | 64 |
605 1.1 ad * : :
606 1.1 ad * : :
607 1.1 ad * 33 | 496 |
608 1.1 ad * 34 | 512 |
609 1.1 ad * --------+------+
610 1.1 ad * 35 | 1024 |
611 1.1 ad * 36 | 2048 |
612 1.1 ad * --------+------+
613 1.1 ad */
614 1.1 ad arena_bin_t bins[1]; /* Dynamically sized. */
615 1.1 ad };
616 1.1 ad
617 1.1 ad /******************************************************************************/
618 1.1 ad /*
619 1.1 ad * Data.
620 1.1 ad */
621 1.1 ad
622 1.1 ad /* Number of CPUs. */
623 1.1 ad static unsigned ncpus;
624 1.1 ad
625 1.1 ad /* VM page size. */
626 1.1 ad static size_t pagesize;
627 1.1 ad static size_t pagesize_mask;
628 1.1 ad static size_t pagesize_2pow;
629 1.1 ad
630 1.1 ad /* Various bin-related settings. */
631 1.1 ad static size_t bin_maxclass; /* Max size class for bins. */
632 1.1 ad static unsigned ntbins; /* Number of (2^n)-spaced tiny bins. */
633 1.1 ad static unsigned nqbins; /* Number of quantum-spaced bins. */
634 1.1 ad static unsigned nsbins; /* Number of (2^n)-spaced sub-page bins. */
635 1.1 ad static size_t small_min;
636 1.1 ad static size_t small_max;
637 1.1 ad
638 1.1 ad /* Various quantum-related settings. */
639 1.1 ad static size_t quantum;
640 1.1 ad static size_t quantum_mask; /* (quantum - 1). */
641 1.1 ad
642 1.1 ad /* Various chunk-related settings. */
643 1.1 ad static size_t chunksize;
644 1.1 ad static size_t chunksize_mask; /* (chunksize - 1). */
645 1.1 ad static unsigned chunk_npages;
646 1.1 ad static unsigned arena_chunk_header_npages;
647 1.1 ad static size_t arena_maxclass; /* Max size class for arenas. */
648 1.1 ad
649 1.1 ad /********/
650 1.1 ad /*
651 1.1 ad * Chunks.
652 1.1 ad */
653 1.1 ad
654 1.1 ad /* Protects chunk-related data structures. */
655 1.1 ad static malloc_mutex_t chunks_mtx;
656 1.1 ad
657 1.1 ad /* Tree of chunks that are stand-alone huge allocations. */
658 1.1 ad static chunk_tree_t huge;
659 1.1 ad
660 1.1 ad #ifdef USE_BRK
661 1.1 ad /*
662 1.1 ad * Try to use brk for chunk-size allocations, due to address space constraints.
663 1.1 ad */
664 1.1 ad /*
665 1.1 ad * Protects sbrk() calls. This must be separate from chunks_mtx, since
666 1.1 ad * base_pages_alloc() also uses sbrk(), but cannot lock chunks_mtx (doing so
667 1.1 ad * could cause recursive lock acquisition).
668 1.1 ad */
669 1.1 ad static malloc_mutex_t brk_mtx;
670 1.1 ad /* Result of first sbrk(0) call. */
671 1.1 ad static void *brk_base;
672 1.1 ad /* Current end of brk, or ((void *)-1) if brk is exhausted. */
673 1.1 ad static void *brk_prev;
674 1.1 ad /* Current upper limit on brk addresses. */
675 1.1 ad static void *brk_max;
676 1.1 ad #endif
677 1.1 ad
678 1.1 ad #ifdef MALLOC_STATS
679 1.1 ad /* Huge allocation statistics. */
680 1.1 ad static uint64_t huge_nmalloc;
681 1.1 ad static uint64_t huge_ndalloc;
682 1.1 ad static size_t huge_allocated;
683 1.1 ad #endif
684 1.1 ad
685 1.1 ad /*
686 1.1 ad * Tree of chunks that were previously allocated. This is used when allocating
687 1.1 ad * chunks, in an attempt to re-use address space.
688 1.1 ad */
689 1.1 ad static chunk_tree_t old_chunks;
690 1.1 ad
691 1.1 ad /****************************/
692 1.1 ad /*
693 1.1 ad * base (internal allocation).
694 1.1 ad */
695 1.1 ad
696 1.1 ad /*
697 1.1 ad * Current pages that are being used for internal memory allocations. These
698 1.1 ad * pages are carved up in cacheline-size quanta, so that there is no chance of
699 1.1 ad * false cache line sharing.
700 1.1 ad */
701 1.1 ad static void *base_pages;
702 1.1 ad static void *base_next_addr;
703 1.1 ad static void *base_past_addr; /* Addr immediately past base_pages. */
704 1.1 ad static chunk_node_t *base_chunk_nodes; /* LIFO cache of chunk nodes. */
705 1.1 ad static malloc_mutex_t base_mtx;
706 1.1 ad #ifdef MALLOC_STATS
707 1.1 ad static size_t base_mapped;
708 1.1 ad #endif
709 1.1 ad
710 1.1 ad /********/
711 1.1 ad /*
712 1.1 ad * Arenas.
713 1.1 ad */
714 1.1 ad
715 1.1 ad /*
716 1.1 ad * Arenas that are used to service external requests. Not all elements of the
717 1.1 ad * arenas array are necessarily used; arenas are created lazily as needed.
718 1.1 ad */
719 1.1 ad static arena_t **arenas;
720 1.1 ad static unsigned narenas;
721 1.1 ad static unsigned next_arena;
722 1.1 ad static malloc_mutex_t arenas_mtx; /* Protects arenas initialization. */
723 1.1 ad
724 1.1 ad #ifndef NO_TLS
725 1.1 ad /*
726 1.1 ad * Map of pthread_self() --> arenas[???], used for selecting an arena to use
727 1.1 ad * for allocations.
728 1.1 ad */
729 1.1 ad static __thread arena_t *arenas_map;
730 1.2 ad #define get_arenas_map() (arenas_map)
731 1.2 ad #define set_arenas_map(x) (arenas_map = x)
732 1.2 ad #else
733 1.2 ad static thread_key_t arenas_map_key;
734 1.2 ad #define get_arenas_map() thr_getspecific(arenas_map_key)
735 1.2 ad #define set_arenas_map(x) thr_setspecific(arenas_map_key, x)
736 1.1 ad #endif
737 1.1 ad
738 1.1 ad #ifdef MALLOC_STATS
739 1.1 ad /* Chunk statistics. */
740 1.1 ad static chunk_stats_t stats_chunks;
741 1.1 ad #endif
742 1.1 ad
743 1.1 ad /*******************************/
744 1.1 ad /*
745 1.1 ad * Runtime configuration options.
746 1.1 ad */
747 1.1 ad const char *_malloc_options;
748 1.1 ad
749 1.1 ad #ifndef MALLOC_PRODUCTION
750 1.1 ad static bool opt_abort = true;
751 1.1 ad static bool opt_junk = true;
752 1.1 ad #else
753 1.1 ad static bool opt_abort = false;
754 1.1 ad static bool opt_junk = false;
755 1.1 ad #endif
756 1.1 ad static bool opt_hint = false;
757 1.1 ad static bool opt_print_stats = false;
758 1.1 ad static size_t opt_quantum_2pow = QUANTUM_2POW_MIN;
759 1.1 ad static size_t opt_small_max_2pow = SMALL_MAX_2POW_DEFAULT;
760 1.1 ad static size_t opt_chunk_2pow = CHUNK_2POW_DEFAULT;
761 1.1 ad static bool opt_utrace = false;
762 1.1 ad static bool opt_sysv = false;
763 1.1 ad static bool opt_xmalloc = false;
764 1.1 ad static bool opt_zero = false;
765 1.1 ad static int32_t opt_narenas_lshift = 0;
766 1.1 ad
767 1.1 ad typedef struct {
768 1.1 ad void *p;
769 1.1 ad size_t s;
770 1.1 ad void *r;
771 1.1 ad } malloc_utrace_t;
772 1.1 ad
773 1.1 ad #define UTRACE(a, b, c) \
774 1.1 ad if (opt_utrace) { \
775 1.2 ad malloc_utrace_t ut; \
776 1.2 ad ut.p = a; \
777 1.2 ad ut.s = b; \
778 1.2 ad ut.r = c; \
779 1.2 ad xutrace(&ut, sizeof(ut)); \
780 1.1 ad }
781 1.1 ad
782 1.1 ad /******************************************************************************/
783 1.1 ad /*
784 1.1 ad * Begin function prototypes for non-inline static functions.
785 1.1 ad */
786 1.1 ad
787 1.1 ad static void wrtmessage(const char *p1, const char *p2, const char *p3,
788 1.1 ad const char *p4);
789 1.1 ad #ifdef MALLOC_STATS
790 1.1 ad static void malloc_printf(const char *format, ...);
791 1.1 ad #endif
792 1.1 ad static char *umax2s(uintmax_t x, char *s);
793 1.1 ad static bool base_pages_alloc(size_t minsize);
794 1.1 ad static void *base_alloc(size_t size);
795 1.1 ad static chunk_node_t *base_chunk_node_alloc(void);
796 1.1 ad static void base_chunk_node_dealloc(chunk_node_t *node);
797 1.1 ad #ifdef MALLOC_STATS
798 1.1 ad static void stats_print(arena_t *arena);
799 1.1 ad #endif
800 1.1 ad static void *pages_map(void *addr, size_t size);
801 1.1 ad static void pages_unmap(void *addr, size_t size);
802 1.1 ad static void *chunk_alloc(size_t size);
803 1.1 ad static void chunk_dealloc(void *chunk, size_t size);
804 1.1 ad static arena_t *choose_arena_hard(void);
805 1.1 ad static void arena_run_split(arena_t *arena, arena_run_t *run, size_t size);
806 1.1 ad static arena_chunk_t *arena_chunk_alloc(arena_t *arena);
807 1.1 ad static void arena_chunk_dealloc(arena_t *arena, arena_chunk_t *chunk);
808 1.1 ad static arena_run_t *arena_run_alloc(arena_t *arena, size_t size);
809 1.1 ad static void arena_run_dalloc(arena_t *arena, arena_run_t *run, size_t size);
810 1.1 ad static arena_run_t *arena_bin_nonfull_run_get(arena_t *arena, arena_bin_t *bin);
811 1.1 ad static void *arena_bin_malloc_hard(arena_t *arena, arena_bin_t *bin);
812 1.1 ad static size_t arena_bin_run_size_calc(arena_bin_t *bin, size_t min_run_size);
813 1.1 ad static void *arena_malloc(arena_t *arena, size_t size);
814 1.1 ad static void *arena_palloc(arena_t *arena, size_t alignment, size_t size,
815 1.1 ad size_t alloc_size);
816 1.1 ad static size_t arena_salloc(const void *ptr);
817 1.1 ad static void *arena_ralloc(void *ptr, size_t size, size_t oldsize);
818 1.1 ad static void arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr);
819 1.1 ad static bool arena_new(arena_t *arena);
820 1.1 ad static arena_t *arenas_extend(unsigned ind);
821 1.1 ad static void *huge_malloc(size_t size);
822 1.1 ad static void *huge_palloc(size_t alignment, size_t size);
823 1.1 ad static void *huge_ralloc(void *ptr, size_t size, size_t oldsize);
824 1.1 ad static void huge_dalloc(void *ptr);
825 1.1 ad static void *imalloc(size_t size);
826 1.1 ad static void *ipalloc(size_t alignment, size_t size);
827 1.1 ad static void *icalloc(size_t size);
828 1.1 ad static size_t isalloc(const void *ptr);
829 1.1 ad static void *iralloc(void *ptr, size_t size);
830 1.1 ad static void idalloc(void *ptr);
831 1.1 ad static void malloc_print_stats(void);
832 1.1 ad static bool malloc_init_hard(void);
833 1.1 ad
834 1.1 ad /*
835 1.1 ad * End function prototypes.
836 1.1 ad */
837 1.1 ad /******************************************************************************/
838 1.1 ad /*
839 1.1 ad * Begin mutex.
840 1.1 ad */
841 1.1 ad
842 1.2 ad #ifdef __NetBSD__
843 1.2 ad #define malloc_mutex_init(m) mutex_init(m, NULL)
844 1.2 ad #define malloc_mutex_lock(m) mutex_lock(m)
845 1.2 ad #define malloc_mutex_unlock(m) mutex_unlock(m)
846 1.2 ad #else /* __NetBSD__ */
847 1.2 ad static inline void
848 1.1 ad malloc_mutex_init(malloc_mutex_t *a_mutex)
849 1.1 ad {
850 1.1 ad static const spinlock_t lock = _SPINLOCK_INITIALIZER;
851 1.1 ad
852 1.1 ad a_mutex->lock = lock;
853 1.1 ad }
854 1.1 ad
855 1.1 ad static inline void
856 1.1 ad malloc_mutex_lock(malloc_mutex_t *a_mutex)
857 1.1 ad {
858 1.1 ad
859 1.1 ad if (__isthreaded)
860 1.1 ad _SPINLOCK(&a_mutex->lock);
861 1.1 ad }
862 1.1 ad
863 1.1 ad static inline void
864 1.1 ad malloc_mutex_unlock(malloc_mutex_t *a_mutex)
865 1.1 ad {
866 1.1 ad
867 1.1 ad if (__isthreaded)
868 1.1 ad _SPINUNLOCK(&a_mutex->lock);
869 1.1 ad }
870 1.2 ad #endif /* __NetBSD__ */
871 1.1 ad
872 1.1 ad /*
873 1.1 ad * End mutex.
874 1.1 ad */
875 1.1 ad /******************************************************************************/
876 1.1 ad /*
877 1.1 ad * Begin Utility functions/macros.
878 1.1 ad */
879 1.1 ad
880 1.1 ad /* Return the chunk address for allocation address a. */
881 1.1 ad #define CHUNK_ADDR2BASE(a) \
882 1.1 ad ((void *)((uintptr_t)(a) & ~chunksize_mask))
883 1.1 ad
884 1.1 ad /* Return the chunk offset of address a. */
885 1.1 ad #define CHUNK_ADDR2OFFSET(a) \
886 1.1 ad ((size_t)((uintptr_t)(a) & chunksize_mask))
887 1.1 ad
888 1.1 ad /* Return the smallest chunk multiple that is >= s. */
889 1.1 ad #define CHUNK_CEILING(s) \
890 1.1 ad (((s) + chunksize_mask) & ~chunksize_mask)
891 1.1 ad
892 1.1 ad /* Return the smallest cacheline multiple that is >= s. */
893 1.1 ad #define CACHELINE_CEILING(s) \
894 1.1 ad (((s) + (CACHELINE - 1)) & ~(CACHELINE - 1))
895 1.1 ad
896 1.1 ad /* Return the smallest quantum multiple that is >= a. */
897 1.1 ad #define QUANTUM_CEILING(a) \
898 1.1 ad (((a) + quantum_mask) & ~quantum_mask)
899 1.1 ad
900 1.1 ad /* Return the smallest pagesize multiple that is >= s. */
901 1.1 ad #define PAGE_CEILING(s) \
902 1.1 ad (((s) + pagesize_mask) & ~pagesize_mask)
903 1.1 ad
904 1.1 ad /* Compute the smallest power of 2 that is >= x. */
905 1.1 ad static inline size_t
906 1.1 ad pow2_ceil(size_t x)
907 1.1 ad {
908 1.1 ad
909 1.1 ad x--;
910 1.1 ad x |= x >> 1;
911 1.1 ad x |= x >> 2;
912 1.1 ad x |= x >> 4;
913 1.1 ad x |= x >> 8;
914 1.1 ad x |= x >> 16;
915 1.1 ad #if (SIZEOF_PTR == 8)
916 1.1 ad x |= x >> 32;
917 1.1 ad #endif
918 1.1 ad x++;
919 1.1 ad return (x);
920 1.1 ad }
921 1.1 ad
922 1.1 ad static void
923 1.1 ad wrtmessage(const char *p1, const char *p2, const char *p3, const char *p4)
924 1.1 ad {
925 1.1 ad
926 1.1 ad _write(STDERR_FILENO, p1, strlen(p1));
927 1.1 ad _write(STDERR_FILENO, p2, strlen(p2));
928 1.1 ad _write(STDERR_FILENO, p3, strlen(p3));
929 1.1 ad _write(STDERR_FILENO, p4, strlen(p4));
930 1.1 ad }
931 1.1 ad
932 1.1 ad void (*_malloc_message)(const char *p1, const char *p2, const char *p3,
933 1.1 ad const char *p4) = wrtmessage;
934 1.1 ad
935 1.1 ad #ifdef MALLOC_STATS
936 1.1 ad /*
937 1.1 ad * Print to stderr in such a way as to (hopefully) avoid memory allocation.
938 1.1 ad */
939 1.1 ad static void
940 1.1 ad malloc_printf(const char *format, ...)
941 1.1 ad {
942 1.1 ad char buf[4096];
943 1.1 ad va_list ap;
944 1.1 ad
945 1.1 ad va_start(ap, format);
946 1.1 ad vsnprintf(buf, sizeof(buf), format, ap);
947 1.1 ad va_end(ap);
948 1.1 ad _malloc_message(buf, "", "", "");
949 1.1 ad }
950 1.1 ad #endif
951 1.1 ad
952 1.1 ad /*
953 1.1 ad * We don't want to depend on vsnprintf() for production builds, since that can
954 1.1 ad * cause unnecessary bloat for static binaries. umax2s() provides minimal
955 1.1 ad * integer printing functionality, so that malloc_printf() use can be limited to
956 1.1 ad * MALLOC_STATS code.
957 1.1 ad */
958 1.1 ad #define UMAX2S_BUFSIZE 21
959 1.1 ad static char *
960 1.1 ad umax2s(uintmax_t x, char *s)
961 1.1 ad {
962 1.1 ad unsigned i;
963 1.1 ad
964 1.1 ad /* Make sure UMAX2S_BUFSIZE is large enough. */
965 1.1 ad assert(sizeof(uintmax_t) <= 8);
966 1.1 ad
967 1.1 ad i = UMAX2S_BUFSIZE - 1;
968 1.1 ad s[i] = '\0';
969 1.1 ad do {
970 1.1 ad i--;
971 1.2 ad s[i] = "0123456789"[(int)x % 10];
972 1.2 ad x /= (uintmax_t)10LL;
973 1.1 ad } while (x > 0);
974 1.1 ad
975 1.1 ad return (&s[i]);
976 1.1 ad }
977 1.1 ad
978 1.1 ad /******************************************************************************/
979 1.1 ad
980 1.1 ad static bool
981 1.1 ad base_pages_alloc(size_t minsize)
982 1.1 ad {
983 1.2 ad size_t csize = 0;
984 1.1 ad
985 1.1 ad #ifdef USE_BRK
986 1.1 ad /*
987 1.1 ad * Do special brk allocation here, since base allocations don't need to
988 1.1 ad * be chunk-aligned.
989 1.1 ad */
990 1.1 ad if (brk_prev != (void *)-1) {
991 1.1 ad void *brk_cur;
992 1.1 ad intptr_t incr;
993 1.1 ad
994 1.1 ad if (minsize != 0)
995 1.1 ad csize = CHUNK_CEILING(minsize);
996 1.1 ad
997 1.1 ad malloc_mutex_lock(&brk_mtx);
998 1.1 ad do {
999 1.1 ad /* Get the current end of brk. */
1000 1.1 ad brk_cur = sbrk(0);
1001 1.1 ad
1002 1.1 ad /*
1003 1.1 ad * Calculate how much padding is necessary to
1004 1.1 ad * chunk-align the end of brk. Don't worry about
1005 1.1 ad * brk_cur not being chunk-aligned though.
1006 1.1 ad */
1007 1.1 ad incr = (intptr_t)chunksize
1008 1.1 ad - (intptr_t)CHUNK_ADDR2OFFSET(brk_cur);
1009 1.1 ad if (incr < minsize)
1010 1.1 ad incr += csize;
1011 1.1 ad
1012 1.1 ad brk_prev = sbrk(incr);
1013 1.1 ad if (brk_prev == brk_cur) {
1014 1.1 ad /* Success. */
1015 1.1 ad malloc_mutex_unlock(&brk_mtx);
1016 1.1 ad base_pages = brk_cur;
1017 1.1 ad base_next_addr = base_pages;
1018 1.1 ad base_past_addr = (void *)((uintptr_t)base_pages
1019 1.1 ad + incr);
1020 1.1 ad #ifdef MALLOC_STATS
1021 1.1 ad base_mapped += incr;
1022 1.1 ad #endif
1023 1.1 ad return (false);
1024 1.1 ad }
1025 1.1 ad } while (brk_prev != (void *)-1);
1026 1.1 ad malloc_mutex_unlock(&brk_mtx);
1027 1.1 ad }
1028 1.1 ad if (minsize == 0) {
1029 1.1 ad /*
1030 1.1 ad * Failure during initialization doesn't matter, so avoid
1031 1.1 ad * falling through to the mmap-based page mapping code.
1032 1.1 ad */
1033 1.1 ad return (true);
1034 1.1 ad }
1035 1.1 ad #endif
1036 1.1 ad assert(minsize != 0);
1037 1.1 ad csize = PAGE_CEILING(minsize);
1038 1.1 ad base_pages = pages_map(NULL, csize);
1039 1.1 ad if (base_pages == NULL)
1040 1.1 ad return (true);
1041 1.1 ad base_next_addr = base_pages;
1042 1.1 ad base_past_addr = (void *)((uintptr_t)base_pages + csize);
1043 1.1 ad #ifdef MALLOC_STATS
1044 1.1 ad base_mapped += csize;
1045 1.1 ad #endif
1046 1.1 ad return (false);
1047 1.1 ad }
1048 1.1 ad
1049 1.1 ad static void *
1050 1.1 ad base_alloc(size_t size)
1051 1.1 ad {
1052 1.1 ad void *ret;
1053 1.1 ad size_t csize;
1054 1.1 ad
1055 1.1 ad /* Round size up to nearest multiple of the cacheline size. */
1056 1.1 ad csize = CACHELINE_CEILING(size);
1057 1.1 ad
1058 1.1 ad malloc_mutex_lock(&base_mtx);
1059 1.1 ad
1060 1.1 ad /* Make sure there's enough space for the allocation. */
1061 1.1 ad if ((uintptr_t)base_next_addr + csize > (uintptr_t)base_past_addr) {
1062 1.1 ad if (base_pages_alloc(csize)) {
1063 1.1 ad ret = NULL;
1064 1.1 ad goto RETURN;
1065 1.1 ad }
1066 1.1 ad }
1067 1.1 ad
1068 1.1 ad /* Allocate. */
1069 1.1 ad ret = base_next_addr;
1070 1.1 ad base_next_addr = (void *)((uintptr_t)base_next_addr + csize);
1071 1.1 ad
1072 1.1 ad RETURN:
1073 1.1 ad malloc_mutex_unlock(&base_mtx);
1074 1.1 ad return (ret);
1075 1.1 ad }
1076 1.1 ad
1077 1.1 ad static chunk_node_t *
1078 1.1 ad base_chunk_node_alloc(void)
1079 1.1 ad {
1080 1.1 ad chunk_node_t *ret;
1081 1.1 ad
1082 1.1 ad malloc_mutex_lock(&base_mtx);
1083 1.1 ad if (base_chunk_nodes != NULL) {
1084 1.1 ad ret = base_chunk_nodes;
1085 1.2 ad /* LINTED */
1086 1.1 ad base_chunk_nodes = *(chunk_node_t **)ret;
1087 1.1 ad malloc_mutex_unlock(&base_mtx);
1088 1.1 ad } else {
1089 1.1 ad malloc_mutex_unlock(&base_mtx);
1090 1.1 ad ret = (chunk_node_t *)base_alloc(sizeof(chunk_node_t));
1091 1.1 ad }
1092 1.1 ad
1093 1.1 ad return (ret);
1094 1.1 ad }
1095 1.1 ad
1096 1.1 ad static void
1097 1.1 ad base_chunk_node_dealloc(chunk_node_t *node)
1098 1.1 ad {
1099 1.1 ad
1100 1.1 ad malloc_mutex_lock(&base_mtx);
1101 1.2 ad /* LINTED */
1102 1.1 ad *(chunk_node_t **)node = base_chunk_nodes;
1103 1.1 ad base_chunk_nodes = node;
1104 1.1 ad malloc_mutex_unlock(&base_mtx);
1105 1.1 ad }
1106 1.1 ad
1107 1.1 ad /******************************************************************************/
1108 1.1 ad
1109 1.1 ad #ifdef MALLOC_STATS
1110 1.1 ad static void
1111 1.1 ad stats_print(arena_t *arena)
1112 1.1 ad {
1113 1.1 ad unsigned i;
1114 1.1 ad int gap_start;
1115 1.1 ad
1116 1.1 ad malloc_printf(
1117 1.1 ad " allocated/mapped nmalloc ndalloc\n");
1118 1.2 ad
1119 1.2 ad malloc_printf("small: %12zu %-12s %12llu %12llu\n",
1120 1.1 ad arena->stats.allocated_small, "", arena->stats.nmalloc_small,
1121 1.1 ad arena->stats.ndalloc_small);
1122 1.2 ad malloc_printf("large: %12zu %-12s %12llu %12llu\n",
1123 1.1 ad arena->stats.allocated_large, "", arena->stats.nmalloc_large,
1124 1.1 ad arena->stats.ndalloc_large);
1125 1.2 ad malloc_printf("total: %12zu/%-12zu %12llu %12llu\n",
1126 1.1 ad arena->stats.allocated_small + arena->stats.allocated_large,
1127 1.1 ad arena->stats.mapped,
1128 1.1 ad arena->stats.nmalloc_small + arena->stats.nmalloc_large,
1129 1.1 ad arena->stats.ndalloc_small + arena->stats.ndalloc_large);
1130 1.1 ad
1131 1.1 ad malloc_printf("bins: bin size regs pgs requests newruns"
1132 1.1 ad " reruns maxruns curruns\n");
1133 1.1 ad for (i = 0, gap_start = -1; i < ntbins + nqbins + nsbins; i++) {
1134 1.1 ad if (arena->bins[i].stats.nrequests == 0) {
1135 1.1 ad if (gap_start == -1)
1136 1.1 ad gap_start = i;
1137 1.1 ad } else {
1138 1.1 ad if (gap_start != -1) {
1139 1.1 ad if (i > gap_start + 1) {
1140 1.1 ad /* Gap of more than one size class. */
1141 1.1 ad malloc_printf("[%u..%u]\n",
1142 1.1 ad gap_start, i - 1);
1143 1.1 ad } else {
1144 1.1 ad /* Gap of one size class. */
1145 1.1 ad malloc_printf("[%u]\n", gap_start);
1146 1.1 ad }
1147 1.1 ad gap_start = -1;
1148 1.1 ad }
1149 1.1 ad malloc_printf(
1150 1.1 ad "%13u %1s %4u %4u %3u %9llu %9llu"
1151 1.1 ad " %9llu %7lu %7lu\n",
1152 1.1 ad i,
1153 1.1 ad i < ntbins ? "T" : i < ntbins + nqbins ? "Q" : "S",
1154 1.1 ad arena->bins[i].reg_size,
1155 1.1 ad arena->bins[i].nregs,
1156 1.1 ad arena->bins[i].run_size >> pagesize_2pow,
1157 1.1 ad arena->bins[i].stats.nrequests,
1158 1.1 ad arena->bins[i].stats.nruns,
1159 1.1 ad arena->bins[i].stats.reruns,
1160 1.1 ad arena->bins[i].stats.highruns,
1161 1.1 ad arena->bins[i].stats.curruns);
1162 1.1 ad }
1163 1.1 ad }
1164 1.1 ad if (gap_start != -1) {
1165 1.1 ad if (i > gap_start + 1) {
1166 1.1 ad /* Gap of more than one size class. */
1167 1.1 ad malloc_printf("[%u..%u]\n", gap_start, i - 1);
1168 1.1 ad } else {
1169 1.1 ad /* Gap of one size class. */
1170 1.1 ad malloc_printf("[%u]\n", gap_start);
1171 1.1 ad }
1172 1.1 ad }
1173 1.1 ad }
1174 1.1 ad #endif
1175 1.1 ad
1176 1.1 ad /*
1177 1.1 ad * End Utility functions/macros.
1178 1.1 ad */
1179 1.1 ad /******************************************************************************/
1180 1.1 ad /*
1181 1.1 ad * Begin chunk management functions.
1182 1.1 ad */
1183 1.1 ad
1184 1.1 ad static inline int
1185 1.1 ad chunk_comp(chunk_node_t *a, chunk_node_t *b)
1186 1.1 ad {
1187 1.1 ad
1188 1.1 ad assert(a != NULL);
1189 1.1 ad assert(b != NULL);
1190 1.1 ad
1191 1.1 ad if ((uintptr_t)a->chunk < (uintptr_t)b->chunk)
1192 1.1 ad return (-1);
1193 1.1 ad else if (a->chunk == b->chunk)
1194 1.1 ad return (0);
1195 1.1 ad else
1196 1.1 ad return (1);
1197 1.1 ad }
1198 1.1 ad
1199 1.1 ad /* Generate red-black tree code for chunks. */
1200 1.2 ad #ifndef lint
1201 1.1 ad RB_GENERATE_STATIC(chunk_tree_s, chunk_node_s, link, chunk_comp);
1202 1.2 ad #endif
1203 1.1 ad
1204 1.1 ad static void *
1205 1.1 ad pages_map(void *addr, size_t size)
1206 1.1 ad {
1207 1.1 ad void *ret;
1208 1.1 ad
1209 1.1 ad /*
1210 1.1 ad * We don't use MAP_FIXED here, because it can cause the *replacement*
1211 1.1 ad * of existing mappings, and we only want to create new mappings.
1212 1.1 ad */
1213 1.1 ad ret = mmap(addr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON,
1214 1.1 ad -1, 0);
1215 1.1 ad assert(ret != NULL);
1216 1.1 ad
1217 1.1 ad if (ret == MAP_FAILED)
1218 1.1 ad ret = NULL;
1219 1.1 ad else if (addr != NULL && ret != addr) {
1220 1.1 ad /*
1221 1.1 ad * We succeeded in mapping memory, but not in the right place.
1222 1.1 ad */
1223 1.1 ad if (munmap(ret, size) == -1) {
1224 1.1 ad char buf[STRERROR_BUF];
1225 1.1 ad
1226 1.1 ad strerror_r(errno, buf, sizeof(buf));
1227 1.1 ad _malloc_message(_getprogname(),
1228 1.1 ad ": (malloc) Error in munmap(): ", buf, "\n");
1229 1.1 ad if (opt_abort)
1230 1.1 ad abort();
1231 1.1 ad }
1232 1.1 ad ret = NULL;
1233 1.1 ad }
1234 1.1 ad
1235 1.1 ad assert(ret == NULL || (addr == NULL && ret != addr)
1236 1.1 ad || (addr != NULL && ret == addr));
1237 1.1 ad return (ret);
1238 1.1 ad }
1239 1.1 ad
1240 1.1 ad static void
1241 1.1 ad pages_unmap(void *addr, size_t size)
1242 1.1 ad {
1243 1.1 ad
1244 1.1 ad if (munmap(addr, size) == -1) {
1245 1.1 ad char buf[STRERROR_BUF];
1246 1.1 ad
1247 1.1 ad strerror_r(errno, buf, sizeof(buf));
1248 1.1 ad _malloc_message(_getprogname(),
1249 1.1 ad ": (malloc) Error in munmap(): ", buf, "\n");
1250 1.1 ad if (opt_abort)
1251 1.1 ad abort();
1252 1.1 ad }
1253 1.1 ad }
1254 1.1 ad
1255 1.1 ad static void *
1256 1.1 ad chunk_alloc(size_t size)
1257 1.1 ad {
1258 1.1 ad void *ret, *chunk;
1259 1.1 ad chunk_node_t *tchunk, *delchunk;
1260 1.1 ad
1261 1.1 ad assert(size != 0);
1262 1.1 ad assert((size & chunksize_mask) == 0);
1263 1.1 ad
1264 1.1 ad malloc_mutex_lock(&chunks_mtx);
1265 1.1 ad
1266 1.1 ad if (size == chunksize) {
1267 1.1 ad /*
1268 1.1 ad * Check for address ranges that were previously chunks and try
1269 1.1 ad * to use them.
1270 1.1 ad */
1271 1.1 ad
1272 1.2 ad /* LINTED */
1273 1.1 ad tchunk = RB_MIN(chunk_tree_s, &old_chunks);
1274 1.1 ad while (tchunk != NULL) {
1275 1.1 ad /* Found an address range. Try to recycle it. */
1276 1.1 ad
1277 1.1 ad chunk = tchunk->chunk;
1278 1.1 ad delchunk = tchunk;
1279 1.2 ad /* LINTED */
1280 1.1 ad tchunk = RB_NEXT(chunk_tree_s, &old_chunks, delchunk);
1281 1.1 ad
1282 1.1 ad /* Remove delchunk from the tree. */
1283 1.2 ad /* LINTED */
1284 1.1 ad RB_REMOVE(chunk_tree_s, &old_chunks, delchunk);
1285 1.1 ad base_chunk_node_dealloc(delchunk);
1286 1.1 ad
1287 1.1 ad #ifdef USE_BRK
1288 1.1 ad if ((uintptr_t)chunk >= (uintptr_t)brk_base
1289 1.1 ad && (uintptr_t)chunk < (uintptr_t)brk_max) {
1290 1.1 ad /* Re-use a previously freed brk chunk. */
1291 1.1 ad ret = chunk;
1292 1.1 ad goto RETURN;
1293 1.1 ad }
1294 1.1 ad #endif
1295 1.1 ad if ((ret = pages_map(chunk, size)) != NULL) {
1296 1.1 ad /* Success. */
1297 1.1 ad goto RETURN;
1298 1.1 ad }
1299 1.1 ad }
1300 1.1 ad }
1301 1.1 ad
1302 1.1 ad /*
1303 1.1 ad * Try to over-allocate, but allow the OS to place the allocation
1304 1.1 ad * anywhere. Beware of size_t wrap-around.
1305 1.1 ad */
1306 1.1 ad if (size + chunksize > size) {
1307 1.1 ad if ((ret = pages_map(NULL, size + chunksize)) != NULL) {
1308 1.1 ad size_t offset = CHUNK_ADDR2OFFSET(ret);
1309 1.1 ad
1310 1.1 ad /*
1311 1.1 ad * Success. Clean up unneeded leading/trailing space.
1312 1.1 ad */
1313 1.1 ad if (offset != 0) {
1314 1.1 ad /* Leading space. */
1315 1.1 ad pages_unmap(ret, chunksize - offset);
1316 1.1 ad
1317 1.1 ad ret = (void *)((uintptr_t)ret + (chunksize -
1318 1.1 ad offset));
1319 1.1 ad
1320 1.1 ad /* Trailing space. */
1321 1.1 ad pages_unmap((void *)((uintptr_t)ret + size),
1322 1.1 ad offset);
1323 1.1 ad } else {
1324 1.1 ad /* Trailing space only. */
1325 1.1 ad pages_unmap((void *)((uintptr_t)ret + size),
1326 1.1 ad chunksize);
1327 1.1 ad }
1328 1.1 ad goto RETURN;
1329 1.1 ad }
1330 1.1 ad }
1331 1.1 ad
1332 1.1 ad #ifdef USE_BRK
1333 1.1 ad /*
1334 1.1 ad * Try to create allocations in brk, in order to make full use of
1335 1.1 ad * limited address space.
1336 1.1 ad */
1337 1.1 ad if (brk_prev != (void *)-1) {
1338 1.1 ad void *brk_cur;
1339 1.1 ad intptr_t incr;
1340 1.1 ad
1341 1.1 ad /*
1342 1.1 ad * The loop is necessary to recover from races with other
1343 1.1 ad * threads that are using brk for something other than malloc.
1344 1.1 ad */
1345 1.1 ad malloc_mutex_lock(&brk_mtx);
1346 1.1 ad do {
1347 1.1 ad /* Get the current end of brk. */
1348 1.1 ad brk_cur = sbrk(0);
1349 1.1 ad
1350 1.1 ad /*
1351 1.1 ad * Calculate how much padding is necessary to
1352 1.1 ad * chunk-align the end of brk.
1353 1.1 ad */
1354 1.1 ad incr = (intptr_t)size
1355 1.1 ad - (intptr_t)CHUNK_ADDR2OFFSET(brk_cur);
1356 1.1 ad if (incr == size) {
1357 1.1 ad ret = brk_cur;
1358 1.1 ad } else {
1359 1.1 ad ret = (void *)((intptr_t)brk_cur + incr);
1360 1.1 ad incr += size;
1361 1.1 ad }
1362 1.1 ad
1363 1.1 ad brk_prev = sbrk(incr);
1364 1.1 ad if (brk_prev == brk_cur) {
1365 1.1 ad /* Success. */
1366 1.1 ad malloc_mutex_unlock(&brk_mtx);
1367 1.1 ad brk_max = (void *)((intptr_t)ret + size);
1368 1.1 ad goto RETURN;
1369 1.1 ad }
1370 1.1 ad } while (brk_prev != (void *)-1);
1371 1.1 ad malloc_mutex_unlock(&brk_mtx);
1372 1.1 ad }
1373 1.1 ad #endif
1374 1.1 ad
1375 1.1 ad /* All strategies for allocation failed. */
1376 1.1 ad ret = NULL;
1377 1.1 ad RETURN:
1378 1.1 ad if (ret != NULL) {
1379 1.1 ad chunk_node_t key;
1380 1.1 ad /*
1381 1.1 ad * Clean out any entries in old_chunks that overlap with the
1382 1.1 ad * memory we just allocated.
1383 1.1 ad */
1384 1.1 ad key.chunk = ret;
1385 1.2 ad /* LINTED */
1386 1.1 ad tchunk = RB_NFIND(chunk_tree_s, &old_chunks, &key);
1387 1.1 ad while (tchunk != NULL
1388 1.1 ad && (uintptr_t)tchunk->chunk >= (uintptr_t)ret
1389 1.1 ad && (uintptr_t)tchunk->chunk < (uintptr_t)ret + size) {
1390 1.1 ad delchunk = tchunk;
1391 1.2 ad /* LINTED */
1392 1.1 ad tchunk = RB_NEXT(chunk_tree_s, &old_chunks, delchunk);
1393 1.2 ad /* LINTED */
1394 1.1 ad RB_REMOVE(chunk_tree_s, &old_chunks, delchunk);
1395 1.1 ad base_chunk_node_dealloc(delchunk);
1396 1.1 ad }
1397 1.1 ad
1398 1.1 ad }
1399 1.1 ad #ifdef MALLOC_STATS
1400 1.1 ad if (ret != NULL) {
1401 1.1 ad stats_chunks.nchunks += (size / chunksize);
1402 1.1 ad stats_chunks.curchunks += (size / chunksize);
1403 1.1 ad }
1404 1.1 ad if (stats_chunks.curchunks > stats_chunks.highchunks)
1405 1.1 ad stats_chunks.highchunks = stats_chunks.curchunks;
1406 1.1 ad #endif
1407 1.1 ad malloc_mutex_unlock(&chunks_mtx);
1408 1.1 ad
1409 1.1 ad assert(CHUNK_ADDR2BASE(ret) == ret);
1410 1.1 ad return (ret);
1411 1.1 ad }
1412 1.1 ad
1413 1.1 ad static void
1414 1.1 ad chunk_dealloc(void *chunk, size_t size)
1415 1.1 ad {
1416 1.1 ad chunk_node_t *node;
1417 1.1 ad
1418 1.1 ad assert(chunk != NULL);
1419 1.1 ad assert(CHUNK_ADDR2BASE(chunk) == chunk);
1420 1.1 ad assert(size != 0);
1421 1.1 ad assert((size & chunksize_mask) == 0);
1422 1.1 ad
1423 1.1 ad malloc_mutex_lock(&chunks_mtx);
1424 1.1 ad
1425 1.1 ad #ifdef USE_BRK
1426 1.1 ad if ((uintptr_t)chunk >= (uintptr_t)brk_base
1427 1.1 ad && (uintptr_t)chunk < (uintptr_t)brk_max) {
1428 1.1 ad void *brk_cur;
1429 1.1 ad
1430 1.1 ad malloc_mutex_lock(&brk_mtx);
1431 1.1 ad /* Get the current end of brk. */
1432 1.1 ad brk_cur = sbrk(0);
1433 1.1 ad
1434 1.1 ad /*
1435 1.1 ad * Try to shrink the data segment if this chunk is at the end
1436 1.1 ad * of the data segment. The sbrk() call here is subject to a
1437 1.1 ad * race condition with threads that use brk(2) or sbrk(2)
1438 1.1 ad * directly, but the alternative would be to leak memory for
1439 1.1 ad * the sake of poorly designed multi-threaded programs.
1440 1.1 ad */
1441 1.1 ad if (brk_cur == brk_max
1442 1.1 ad && (void *)((uintptr_t)chunk + size) == brk_max
1443 1.1 ad && sbrk(-(intptr_t)size) == brk_max) {
1444 1.1 ad malloc_mutex_unlock(&brk_mtx);
1445 1.1 ad if (brk_prev == brk_max) {
1446 1.1 ad /* Success. */
1447 1.1 ad brk_prev = (void *)((intptr_t)brk_max
1448 1.1 ad - (intptr_t)size);
1449 1.1 ad brk_max = brk_prev;
1450 1.1 ad }
1451 1.1 ad } else {
1452 1.1 ad size_t offset;
1453 1.1 ad
1454 1.1 ad malloc_mutex_unlock(&brk_mtx);
1455 1.1 ad madvise(chunk, size, MADV_FREE);
1456 1.1 ad
1457 1.1 ad /*
1458 1.1 ad * Iteratively create records of each chunk-sized
1459 1.1 ad * memory region that 'chunk' is comprised of, so that
1460 1.1 ad * the address range can be recycled if memory usage
1461 1.1 ad * increases later on.
1462 1.1 ad */
1463 1.1 ad for (offset = 0; offset < size; offset += chunksize) {
1464 1.1 ad node = base_chunk_node_alloc();
1465 1.1 ad if (node == NULL)
1466 1.1 ad break;
1467 1.1 ad
1468 1.1 ad node->chunk = (void *)((uintptr_t)chunk
1469 1.1 ad + (uintptr_t)offset);
1470 1.1 ad node->size = chunksize;
1471 1.2 ad /* LINTED */
1472 1.1 ad RB_INSERT(chunk_tree_s, &old_chunks, node);
1473 1.1 ad }
1474 1.1 ad }
1475 1.1 ad } else {
1476 1.1 ad #endif
1477 1.1 ad pages_unmap(chunk, size);
1478 1.1 ad
1479 1.1 ad /*
1480 1.1 ad * Make a record of the chunk's address, so that the address
1481 1.1 ad * range can be recycled if memory usage increases later on.
1482 1.1 ad * Don't bother to create entries if (size > chunksize), since
1483 1.1 ad * doing so could cause scalability issues for truly gargantuan
1484 1.1 ad * objects (many gigabytes or larger).
1485 1.1 ad */
1486 1.1 ad if (size == chunksize) {
1487 1.1 ad node = base_chunk_node_alloc();
1488 1.1 ad if (node != NULL) {
1489 1.1 ad node->chunk = (void *)(uintptr_t)chunk;
1490 1.1 ad node->size = chunksize;
1491 1.2 ad /* LINTED */
1492 1.1 ad RB_INSERT(chunk_tree_s, &old_chunks, node);
1493 1.1 ad }
1494 1.1 ad }
1495 1.1 ad #ifdef USE_BRK
1496 1.1 ad }
1497 1.1 ad #endif
1498 1.1 ad
1499 1.1 ad #ifdef MALLOC_STATS
1500 1.1 ad stats_chunks.curchunks -= (size / chunksize);
1501 1.1 ad #endif
1502 1.1 ad malloc_mutex_unlock(&chunks_mtx);
1503 1.1 ad }
1504 1.1 ad
1505 1.1 ad /*
1506 1.1 ad * End chunk management functions.
1507 1.1 ad */
1508 1.1 ad /******************************************************************************/
1509 1.1 ad /*
1510 1.1 ad * Begin arena.
1511 1.1 ad */
1512 1.1 ad
1513 1.1 ad /*
1514 1.1 ad * Choose an arena based on a per-thread value (fast-path code, calls slow-path
1515 1.1 ad * code if necessary).
1516 1.1 ad */
1517 1.1 ad static inline arena_t *
1518 1.1 ad choose_arena(void)
1519 1.1 ad {
1520 1.1 ad arena_t *ret;
1521 1.1 ad
1522 1.1 ad /*
1523 1.1 ad * We can only use TLS if this is a PIC library, since for the static
1524 1.1 ad * library version, libc's malloc is used by TLS allocation, which
1525 1.1 ad * introduces a bootstrapping issue.
1526 1.1 ad */
1527 1.1 ad if (__isthreaded == false) {
1528 1.1 ad /*
1529 1.1 ad * Avoid the overhead of TLS for single-threaded operation. If the
1530 1.1 ad * app switches to threaded mode, the initial thread may end up
1531 1.1 ad * being assigned to some other arena, but this one-time switch
1532 1.1 ad * shouldn't cause significant issues.
1533 1.1 ad */
1534 1.1 ad return (arenas[0]);
1535 1.1 ad }
1536 1.1 ad
1537 1.2 ad ret = get_arenas_map();
1538 1.1 ad if (ret == NULL)
1539 1.1 ad ret = choose_arena_hard();
1540 1.1 ad
1541 1.1 ad assert(ret != NULL);
1542 1.1 ad return (ret);
1543 1.1 ad }
1544 1.1 ad
1545 1.1 ad /*
1546 1.1 ad * Choose an arena based on a per-thread value (slow-path code only, called
1547 1.1 ad * only by choose_arena()).
1548 1.1 ad */
1549 1.1 ad static arena_t *
1550 1.1 ad choose_arena_hard(void)
1551 1.1 ad {
1552 1.1 ad arena_t *ret;
1553 1.1 ad
1554 1.1 ad assert(__isthreaded);
1555 1.1 ad
1556 1.1 ad /* Assign one of the arenas to this thread, in a round-robin fashion. */
1557 1.1 ad malloc_mutex_lock(&arenas_mtx);
1558 1.1 ad ret = arenas[next_arena];
1559 1.1 ad if (ret == NULL)
1560 1.1 ad ret = arenas_extend(next_arena);
1561 1.1 ad if (ret == NULL) {
1562 1.1 ad /*
1563 1.1 ad * Make sure that this function never returns NULL, so that
1564 1.1 ad * choose_arena() doesn't have to check for a NULL return
1565 1.1 ad * value.
1566 1.1 ad */
1567 1.1 ad ret = arenas[0];
1568 1.1 ad }
1569 1.1 ad next_arena = (next_arena + 1) % narenas;
1570 1.1 ad malloc_mutex_unlock(&arenas_mtx);
1571 1.2 ad set_arenas_map(ret);
1572 1.1 ad
1573 1.1 ad return (ret);
1574 1.1 ad }
1575 1.1 ad
1576 1.1 ad static inline int
1577 1.1 ad arena_chunk_comp(arena_chunk_t *a, arena_chunk_t *b)
1578 1.1 ad {
1579 1.1 ad
1580 1.1 ad assert(a != NULL);
1581 1.1 ad assert(b != NULL);
1582 1.1 ad
1583 1.1 ad if ((uintptr_t)a < (uintptr_t)b)
1584 1.1 ad return (-1);
1585 1.1 ad else if (a == b)
1586 1.1 ad return (0);
1587 1.1 ad else
1588 1.1 ad return (1);
1589 1.1 ad }
1590 1.1 ad
1591 1.1 ad /* Generate red-black tree code for arena chunks. */
1592 1.2 ad #ifndef lint
1593 1.1 ad RB_GENERATE_STATIC(arena_chunk_tree_s, arena_chunk_s, link, arena_chunk_comp);
1594 1.2 ad #endif
1595 1.1 ad
1596 1.1 ad static inline int
1597 1.1 ad arena_run_comp(arena_run_t *a, arena_run_t *b)
1598 1.1 ad {
1599 1.1 ad
1600 1.1 ad assert(a != NULL);
1601 1.1 ad assert(b != NULL);
1602 1.1 ad
1603 1.1 ad if ((uintptr_t)a < (uintptr_t)b)
1604 1.1 ad return (-1);
1605 1.1 ad else if (a == b)
1606 1.1 ad return (0);
1607 1.1 ad else
1608 1.1 ad return (1);
1609 1.1 ad }
1610 1.1 ad
1611 1.1 ad /* Generate red-black tree code for arena runs. */
1612 1.2 ad #ifndef lint
1613 1.1 ad RB_GENERATE_STATIC(arena_run_tree_s, arena_run_s, link, arena_run_comp);
1614 1.2 ad #endif
1615 1.1 ad
1616 1.1 ad static inline void *
1617 1.1 ad arena_run_reg_alloc(arena_run_t *run, arena_bin_t *bin)
1618 1.1 ad {
1619 1.1 ad void *ret;
1620 1.1 ad unsigned i, mask, bit, regind;
1621 1.1 ad
1622 1.1 ad assert(run->magic == ARENA_RUN_MAGIC);
1623 1.1 ad assert(run->regs_minelm < bin->regs_mask_nelms);
1624 1.1 ad
1625 1.1 ad /*
1626 1.1 ad * Move the first check outside the loop, so that run->regs_minelm can
1627 1.1 ad * be updated unconditionally, without the possibility of updating it
1628 1.1 ad * multiple times.
1629 1.1 ad */
1630 1.1 ad i = run->regs_minelm;
1631 1.1 ad mask = run->regs_mask[i];
1632 1.1 ad if (mask != 0) {
1633 1.1 ad /* Usable allocation found. */
1634 1.1 ad bit = ffs((int)mask) - 1;
1635 1.1 ad
1636 1.1 ad regind = ((i << (SIZEOF_INT_2POW + 3)) + bit);
1637 1.1 ad ret = (void *)(((uintptr_t)run) + bin->reg0_offset
1638 1.1 ad + (bin->reg_size * regind));
1639 1.1 ad
1640 1.1 ad /* Clear bit. */
1641 1.1 ad mask ^= (1 << bit);
1642 1.1 ad run->regs_mask[i] = mask;
1643 1.1 ad
1644 1.1 ad return (ret);
1645 1.1 ad }
1646 1.1 ad
1647 1.1 ad for (i++; i < bin->regs_mask_nelms; i++) {
1648 1.1 ad mask = run->regs_mask[i];
1649 1.1 ad if (mask != 0) {
1650 1.1 ad /* Usable allocation found. */
1651 1.1 ad bit = ffs((int)mask) - 1;
1652 1.1 ad
1653 1.1 ad regind = ((i << (SIZEOF_INT_2POW + 3)) + bit);
1654 1.1 ad ret = (void *)(((uintptr_t)run) + bin->reg0_offset
1655 1.1 ad + (bin->reg_size * regind));
1656 1.1 ad
1657 1.1 ad /* Clear bit. */
1658 1.1 ad mask ^= (1 << bit);
1659 1.1 ad run->regs_mask[i] = mask;
1660 1.1 ad
1661 1.1 ad /*
1662 1.1 ad * Make a note that nothing before this element
1663 1.1 ad * contains a free region.
1664 1.1 ad */
1665 1.1 ad run->regs_minelm = i; /* Low payoff: + (mask == 0); */
1666 1.1 ad
1667 1.1 ad return (ret);
1668 1.1 ad }
1669 1.1 ad }
1670 1.1 ad /* Not reached. */
1671 1.1 ad assert(0);
1672 1.1 ad return (NULL);
1673 1.1 ad }
1674 1.1 ad
1675 1.1 ad static inline void
1676 1.1 ad arena_run_reg_dalloc(arena_run_t *run, arena_bin_t *bin, void *ptr, size_t size)
1677 1.1 ad {
1678 1.1 ad /*
1679 1.1 ad * To divide by a number D that is not a power of two we multiply
1680 1.1 ad * by (2^21 / D) and then right shift by 21 positions.
1681 1.1 ad *
1682 1.1 ad * X / D
1683 1.1 ad *
1684 1.1 ad * becomes
1685 1.1 ad *
1686 1.1 ad * (X * size_invs[(D >> QUANTUM_2POW_MIN) - 3]) >> SIZE_INV_SHIFT
1687 1.1 ad */
1688 1.1 ad #define SIZE_INV_SHIFT 21
1689 1.1 ad #define SIZE_INV(s) (((1 << SIZE_INV_SHIFT) / (s << QUANTUM_2POW_MIN)) + 1)
1690 1.1 ad static const unsigned size_invs[] = {
1691 1.1 ad SIZE_INV(3),
1692 1.1 ad SIZE_INV(4), SIZE_INV(5), SIZE_INV(6), SIZE_INV(7),
1693 1.1 ad SIZE_INV(8), SIZE_INV(9), SIZE_INV(10), SIZE_INV(11),
1694 1.1 ad SIZE_INV(12),SIZE_INV(13), SIZE_INV(14), SIZE_INV(15),
1695 1.1 ad SIZE_INV(16),SIZE_INV(17), SIZE_INV(18), SIZE_INV(19),
1696 1.1 ad SIZE_INV(20),SIZE_INV(21), SIZE_INV(22), SIZE_INV(23),
1697 1.1 ad SIZE_INV(24),SIZE_INV(25), SIZE_INV(26), SIZE_INV(27),
1698 1.1 ad SIZE_INV(28),SIZE_INV(29), SIZE_INV(30), SIZE_INV(31)
1699 1.1 ad #if (QUANTUM_2POW_MIN < 4)
1700 1.1 ad ,
1701 1.1 ad SIZE_INV(32), SIZE_INV(33), SIZE_INV(34), SIZE_INV(35),
1702 1.1 ad SIZE_INV(36), SIZE_INV(37), SIZE_INV(38), SIZE_INV(39),
1703 1.1 ad SIZE_INV(40), SIZE_INV(41), SIZE_INV(42), SIZE_INV(43),
1704 1.1 ad SIZE_INV(44), SIZE_INV(45), SIZE_INV(46), SIZE_INV(47),
1705 1.1 ad SIZE_INV(48), SIZE_INV(49), SIZE_INV(50), SIZE_INV(51),
1706 1.1 ad SIZE_INV(52), SIZE_INV(53), SIZE_INV(54), SIZE_INV(55),
1707 1.1 ad SIZE_INV(56), SIZE_INV(57), SIZE_INV(58), SIZE_INV(59),
1708 1.1 ad SIZE_INV(60), SIZE_INV(61), SIZE_INV(62), SIZE_INV(63)
1709 1.1 ad #endif
1710 1.1 ad };
1711 1.1 ad unsigned diff, regind, elm, bit;
1712 1.1 ad
1713 1.1 ad assert(run->magic == ARENA_RUN_MAGIC);
1714 1.1 ad assert(((sizeof(size_invs)) / sizeof(unsigned)) + 3
1715 1.1 ad >= (SMALL_MAX_DEFAULT >> QUANTUM_2POW_MIN));
1716 1.1 ad
1717 1.1 ad /*
1718 1.1 ad * Avoid doing division with a variable divisor if possible. Using
1719 1.1 ad * actual division here can reduce allocator throughput by over 20%!
1720 1.1 ad */
1721 1.1 ad diff = (unsigned)((uintptr_t)ptr - (uintptr_t)run - bin->reg0_offset);
1722 1.1 ad if ((size & (size - 1)) == 0) {
1723 1.1 ad /*
1724 1.1 ad * log2_table allows fast division of a power of two in the
1725 1.1 ad * [1..128] range.
1726 1.1 ad *
1727 1.1 ad * (x / divisor) becomes (x >> log2_table[divisor - 1]).
1728 1.1 ad */
1729 1.1 ad static const unsigned char log2_table[] = {
1730 1.1 ad 0, 1, 0, 2, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 4,
1731 1.1 ad 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 5,
1732 1.1 ad 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1733 1.1 ad 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6,
1734 1.1 ad 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1735 1.1 ad 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1736 1.1 ad 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1737 1.1 ad 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 7
1738 1.1 ad };
1739 1.1 ad
1740 1.1 ad if (size <= 128)
1741 1.1 ad regind = (diff >> log2_table[size - 1]);
1742 1.1 ad else if (size <= 32768)
1743 1.1 ad regind = diff >> (8 + log2_table[(size >> 8) - 1]);
1744 1.1 ad else {
1745 1.1 ad /*
1746 1.1 ad * The page size is too large for us to use the lookup
1747 1.1 ad * table. Use real division.
1748 1.1 ad */
1749 1.1 ad regind = diff / size;
1750 1.1 ad }
1751 1.1 ad } else if (size <= ((sizeof(size_invs) / sizeof(unsigned))
1752 1.1 ad << QUANTUM_2POW_MIN) + 2) {
1753 1.1 ad regind = size_invs[(size >> QUANTUM_2POW_MIN) - 3] * diff;
1754 1.1 ad regind >>= SIZE_INV_SHIFT;
1755 1.1 ad } else {
1756 1.1 ad /*
1757 1.1 ad * size_invs isn't large enough to handle this size class, so
1758 1.1 ad * calculate regind using actual division. This only happens
1759 1.1 ad * if the user increases small_max via the 'S' runtime
1760 1.1 ad * configuration option.
1761 1.1 ad */
1762 1.1 ad regind = diff / size;
1763 1.1 ad };
1764 1.1 ad assert(diff == regind * size);
1765 1.1 ad assert(regind < bin->nregs);
1766 1.1 ad
1767 1.1 ad elm = regind >> (SIZEOF_INT_2POW + 3);
1768 1.1 ad if (elm < run->regs_minelm)
1769 1.1 ad run->regs_minelm = elm;
1770 1.1 ad bit = regind - (elm << (SIZEOF_INT_2POW + 3));
1771 1.1 ad assert((run->regs_mask[elm] & (1 << bit)) == 0);
1772 1.1 ad run->regs_mask[elm] |= (1 << bit);
1773 1.1 ad #undef SIZE_INV
1774 1.1 ad #undef SIZE_INV_SHIFT
1775 1.1 ad }
1776 1.1 ad
1777 1.1 ad static void
1778 1.1 ad arena_run_split(arena_t *arena, arena_run_t *run, size_t size)
1779 1.1 ad {
1780 1.1 ad arena_chunk_t *chunk;
1781 1.1 ad unsigned run_ind, map_offset, total_pages, need_pages, rem_pages;
1782 1.1 ad unsigned i;
1783 1.1 ad
1784 1.1 ad chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(run);
1785 1.1 ad run_ind = (unsigned)(((uintptr_t)run - (uintptr_t)chunk)
1786 1.1 ad >> pagesize_2pow);
1787 1.1 ad total_pages = chunk->map[run_ind].npages;
1788 1.1 ad need_pages = (size >> pagesize_2pow);
1789 1.1 ad assert(need_pages <= total_pages);
1790 1.1 ad rem_pages = total_pages - need_pages;
1791 1.1 ad
1792 1.1 ad /* Split enough pages from the front of run to fit allocation size. */
1793 1.1 ad map_offset = run_ind;
1794 1.1 ad for (i = 0; i < need_pages; i++) {
1795 1.1 ad chunk->map[map_offset + i].npages = need_pages;
1796 1.1 ad chunk->map[map_offset + i].pos = i;
1797 1.1 ad }
1798 1.1 ad
1799 1.1 ad /* Keep track of trailing unused pages for later use. */
1800 1.1 ad if (rem_pages > 0) {
1801 1.1 ad /* Update map for trailing pages. */
1802 1.1 ad map_offset += need_pages;
1803 1.1 ad chunk->map[map_offset].npages = rem_pages;
1804 1.1 ad chunk->map[map_offset].pos = POS_FREE;
1805 1.1 ad chunk->map[map_offset + rem_pages - 1].npages = rem_pages;
1806 1.1 ad chunk->map[map_offset + rem_pages - 1].pos = POS_FREE;
1807 1.1 ad }
1808 1.1 ad
1809 1.1 ad chunk->pages_used += need_pages;
1810 1.1 ad }
1811 1.1 ad
1812 1.1 ad static arena_chunk_t *
1813 1.1 ad arena_chunk_alloc(arena_t *arena)
1814 1.1 ad {
1815 1.1 ad arena_chunk_t *chunk;
1816 1.1 ad
1817 1.1 ad if (arena->spare != NULL) {
1818 1.1 ad chunk = arena->spare;
1819 1.1 ad arena->spare = NULL;
1820 1.1 ad
1821 1.2 ad /* LINTED */
1822 1.1 ad RB_INSERT(arena_chunk_tree_s, &arena->chunks, chunk);
1823 1.1 ad } else {
1824 1.1 ad chunk = (arena_chunk_t *)chunk_alloc(chunksize);
1825 1.1 ad if (chunk == NULL)
1826 1.1 ad return (NULL);
1827 1.1 ad #ifdef MALLOC_STATS
1828 1.1 ad arena->stats.mapped += chunksize;
1829 1.1 ad #endif
1830 1.1 ad
1831 1.1 ad chunk->arena = arena;
1832 1.1 ad
1833 1.2 ad /* LINTED */
1834 1.1 ad RB_INSERT(arena_chunk_tree_s, &arena->chunks, chunk);
1835 1.1 ad
1836 1.1 ad /*
1837 1.1 ad * Claim that no pages are in use, since the header is merely
1838 1.1 ad * overhead.
1839 1.1 ad */
1840 1.1 ad chunk->pages_used = 0;
1841 1.1 ad
1842 1.1 ad chunk->max_frun_npages = chunk_npages -
1843 1.1 ad arena_chunk_header_npages;
1844 1.1 ad chunk->min_frun_ind = arena_chunk_header_npages;
1845 1.1 ad
1846 1.1 ad /*
1847 1.1 ad * Initialize enough of the map to support one maximal free run.
1848 1.1 ad */
1849 1.1 ad chunk->map[arena_chunk_header_npages].npages = chunk_npages -
1850 1.1 ad arena_chunk_header_npages;
1851 1.1 ad chunk->map[arena_chunk_header_npages].pos = POS_FREE;
1852 1.1 ad chunk->map[chunk_npages - 1].npages = chunk_npages -
1853 1.1 ad arena_chunk_header_npages;
1854 1.1 ad chunk->map[chunk_npages - 1].pos = POS_FREE;
1855 1.1 ad }
1856 1.1 ad
1857 1.1 ad return (chunk);
1858 1.1 ad }
1859 1.1 ad
1860 1.1 ad static void
1861 1.1 ad arena_chunk_dealloc(arena_t *arena, arena_chunk_t *chunk)
1862 1.1 ad {
1863 1.1 ad
1864 1.1 ad /*
1865 1.1 ad * Remove chunk from the chunk tree, regardless of whether this chunk
1866 1.1 ad * will be cached, so that the arena does not use it.
1867 1.1 ad */
1868 1.2 ad /* LINTED */
1869 1.1 ad RB_REMOVE(arena_chunk_tree_s, &chunk->arena->chunks, chunk);
1870 1.1 ad
1871 1.1 ad if (opt_hint == false) {
1872 1.1 ad if (arena->spare != NULL) {
1873 1.1 ad chunk_dealloc((void *)arena->spare, chunksize);
1874 1.1 ad #ifdef MALLOC_STATS
1875 1.1 ad arena->stats.mapped -= chunksize;
1876 1.1 ad #endif
1877 1.1 ad }
1878 1.1 ad arena->spare = chunk;
1879 1.1 ad } else {
1880 1.1 ad assert(arena->spare == NULL);
1881 1.1 ad chunk_dealloc((void *)chunk, chunksize);
1882 1.1 ad #ifdef MALLOC_STATS
1883 1.1 ad arena->stats.mapped -= chunksize;
1884 1.1 ad #endif
1885 1.1 ad }
1886 1.1 ad }
1887 1.1 ad
1888 1.1 ad static arena_run_t *
1889 1.1 ad arena_run_alloc(arena_t *arena, size_t size)
1890 1.1 ad {
1891 1.1 ad arena_chunk_t *chunk;
1892 1.1 ad arena_run_t *run;
1893 1.1 ad unsigned need_npages, limit_pages, compl_need_npages;
1894 1.1 ad
1895 1.1 ad assert(size <= (chunksize - (arena_chunk_header_npages <<
1896 1.1 ad pagesize_2pow)));
1897 1.1 ad assert((size & pagesize_mask) == 0);
1898 1.1 ad
1899 1.1 ad /*
1900 1.1 ad * Search through arena's chunks in address order for a free run that is
1901 1.1 ad * large enough. Look for the first fit.
1902 1.1 ad */
1903 1.1 ad need_npages = (size >> pagesize_2pow);
1904 1.1 ad limit_pages = chunk_npages - arena_chunk_header_npages;
1905 1.1 ad compl_need_npages = limit_pages - need_npages;
1906 1.2 ad /* LINTED */
1907 1.1 ad RB_FOREACH(chunk, arena_chunk_tree_s, &arena->chunks) {
1908 1.1 ad /*
1909 1.1 ad * Avoid searching this chunk if there are not enough
1910 1.1 ad * contiguous free pages for there to possibly be a large
1911 1.1 ad * enough free run.
1912 1.1 ad */
1913 1.1 ad if (chunk->pages_used <= compl_need_npages &&
1914 1.1 ad need_npages <= chunk->max_frun_npages) {
1915 1.1 ad arena_chunk_map_t *mapelm;
1916 1.1 ad unsigned i;
1917 1.1 ad unsigned max_frun_npages = 0;
1918 1.1 ad unsigned min_frun_ind = chunk_npages;
1919 1.1 ad
1920 1.1 ad assert(chunk->min_frun_ind >=
1921 1.1 ad arena_chunk_header_npages);
1922 1.1 ad for (i = chunk->min_frun_ind; i < chunk_npages;) {
1923 1.1 ad mapelm = &chunk->map[i];
1924 1.1 ad if (mapelm->pos == POS_FREE) {
1925 1.1 ad if (mapelm->npages >= need_npages) {
1926 1.1 ad run = (arena_run_t *)
1927 1.1 ad ((uintptr_t)chunk + (i <<
1928 1.1 ad pagesize_2pow));
1929 1.1 ad /* Update page map. */
1930 1.1 ad arena_run_split(arena, run,
1931 1.1 ad size);
1932 1.1 ad return (run);
1933 1.1 ad }
1934 1.1 ad if (mapelm->npages >
1935 1.1 ad max_frun_npages) {
1936 1.1 ad max_frun_npages =
1937 1.1 ad mapelm->npages;
1938 1.1 ad }
1939 1.1 ad if (i < min_frun_ind) {
1940 1.1 ad min_frun_ind = i;
1941 1.1 ad if (i < chunk->min_frun_ind)
1942 1.1 ad chunk->min_frun_ind = i;
1943 1.1 ad }
1944 1.1 ad }
1945 1.1 ad i += mapelm->npages;
1946 1.1 ad }
1947 1.1 ad /*
1948 1.1 ad * Search failure. Reset cached chunk->max_frun_npages.
1949 1.1 ad * chunk->min_frun_ind was already reset above (if
1950 1.1 ad * necessary).
1951 1.1 ad */
1952 1.1 ad chunk->max_frun_npages = max_frun_npages;
1953 1.1 ad }
1954 1.1 ad }
1955 1.1 ad
1956 1.1 ad /*
1957 1.1 ad * No usable runs. Create a new chunk from which to allocate the run.
1958 1.1 ad */
1959 1.1 ad chunk = arena_chunk_alloc(arena);
1960 1.1 ad if (chunk == NULL)
1961 1.1 ad return (NULL);
1962 1.1 ad run = (arena_run_t *)((uintptr_t)chunk + (arena_chunk_header_npages <<
1963 1.1 ad pagesize_2pow));
1964 1.1 ad /* Update page map. */
1965 1.1 ad arena_run_split(arena, run, size);
1966 1.1 ad return (run);
1967 1.1 ad }
1968 1.1 ad
1969 1.1 ad static void
1970 1.1 ad arena_run_dalloc(arena_t *arena, arena_run_t *run, size_t size)
1971 1.1 ad {
1972 1.1 ad arena_chunk_t *chunk;
1973 1.1 ad unsigned run_ind, run_pages;
1974 1.1 ad
1975 1.1 ad chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(run);
1976 1.1 ad
1977 1.1 ad run_ind = (unsigned)(((uintptr_t)run - (uintptr_t)chunk)
1978 1.1 ad >> pagesize_2pow);
1979 1.1 ad assert(run_ind >= arena_chunk_header_npages);
1980 1.1 ad assert(run_ind < (chunksize >> pagesize_2pow));
1981 1.1 ad run_pages = (size >> pagesize_2pow);
1982 1.1 ad assert(run_pages == chunk->map[run_ind].npages);
1983 1.1 ad
1984 1.1 ad /* Subtract pages from count of pages used in chunk. */
1985 1.1 ad chunk->pages_used -= run_pages;
1986 1.1 ad
1987 1.1 ad /* Mark run as deallocated. */
1988 1.1 ad assert(chunk->map[run_ind].npages == run_pages);
1989 1.1 ad chunk->map[run_ind].pos = POS_FREE;
1990 1.1 ad assert(chunk->map[run_ind + run_pages - 1].npages == run_pages);
1991 1.1 ad chunk->map[run_ind + run_pages - 1].pos = POS_FREE;
1992 1.1 ad
1993 1.1 ad /*
1994 1.1 ad * Tell the kernel that we don't need the data in this run, but only if
1995 1.1 ad * requested via runtime configuration.
1996 1.1 ad */
1997 1.1 ad if (opt_hint)
1998 1.1 ad madvise(run, size, MADV_FREE);
1999 1.1 ad
2000 1.1 ad /* Try to coalesce with neighboring runs. */
2001 1.1 ad if (run_ind > arena_chunk_header_npages &&
2002 1.1 ad chunk->map[run_ind - 1].pos == POS_FREE) {
2003 1.1 ad unsigned prev_npages;
2004 1.1 ad
2005 1.1 ad /* Coalesce with previous run. */
2006 1.1 ad prev_npages = chunk->map[run_ind - 1].npages;
2007 1.1 ad run_ind -= prev_npages;
2008 1.1 ad assert(chunk->map[run_ind].npages == prev_npages);
2009 1.1 ad assert(chunk->map[run_ind].pos == POS_FREE);
2010 1.1 ad run_pages += prev_npages;
2011 1.1 ad
2012 1.1 ad chunk->map[run_ind].npages = run_pages;
2013 1.1 ad assert(chunk->map[run_ind].pos == POS_FREE);
2014 1.1 ad chunk->map[run_ind + run_pages - 1].npages = run_pages;
2015 1.1 ad assert(chunk->map[run_ind + run_pages - 1].pos == POS_FREE);
2016 1.1 ad }
2017 1.1 ad
2018 1.1 ad if (run_ind + run_pages < chunk_npages &&
2019 1.1 ad chunk->map[run_ind + run_pages].pos == POS_FREE) {
2020 1.1 ad unsigned next_npages;
2021 1.1 ad
2022 1.1 ad /* Coalesce with next run. */
2023 1.1 ad next_npages = chunk->map[run_ind + run_pages].npages;
2024 1.1 ad run_pages += next_npages;
2025 1.1 ad assert(chunk->map[run_ind + run_pages - 1].npages ==
2026 1.1 ad next_npages);
2027 1.1 ad assert(chunk->map[run_ind + run_pages - 1].pos == POS_FREE);
2028 1.1 ad
2029 1.1 ad chunk->map[run_ind].npages = run_pages;
2030 1.1 ad chunk->map[run_ind].pos = POS_FREE;
2031 1.1 ad chunk->map[run_ind + run_pages - 1].npages = run_pages;
2032 1.1 ad assert(chunk->map[run_ind + run_pages - 1].pos == POS_FREE);
2033 1.1 ad }
2034 1.1 ad
2035 1.1 ad if (chunk->map[run_ind].npages > chunk->max_frun_npages)
2036 1.1 ad chunk->max_frun_npages = chunk->map[run_ind].npages;
2037 1.1 ad if (run_ind < chunk->min_frun_ind)
2038 1.1 ad chunk->min_frun_ind = run_ind;
2039 1.1 ad
2040 1.1 ad /* Deallocate chunk if it is now completely unused. */
2041 1.1 ad if (chunk->pages_used == 0)
2042 1.1 ad arena_chunk_dealloc(arena, chunk);
2043 1.1 ad }
2044 1.1 ad
2045 1.1 ad static arena_run_t *
2046 1.1 ad arena_bin_nonfull_run_get(arena_t *arena, arena_bin_t *bin)
2047 1.1 ad {
2048 1.1 ad arena_run_t *run;
2049 1.1 ad unsigned i, remainder;
2050 1.1 ad
2051 1.1 ad /* Look for a usable run. */
2052 1.2 ad /* LINTED */
2053 1.1 ad if ((run = RB_MIN(arena_run_tree_s, &bin->runs)) != NULL) {
2054 1.1 ad /* run is guaranteed to have available space. */
2055 1.2 ad /* LINTED */
2056 1.1 ad RB_REMOVE(arena_run_tree_s, &bin->runs, run);
2057 1.1 ad #ifdef MALLOC_STATS
2058 1.1 ad bin->stats.reruns++;
2059 1.1 ad #endif
2060 1.1 ad return (run);
2061 1.1 ad }
2062 1.1 ad /* No existing runs have any space available. */
2063 1.1 ad
2064 1.1 ad /* Allocate a new run. */
2065 1.1 ad run = arena_run_alloc(arena, bin->run_size);
2066 1.1 ad if (run == NULL)
2067 1.1 ad return (NULL);
2068 1.1 ad
2069 1.1 ad /* Initialize run internals. */
2070 1.1 ad run->bin = bin;
2071 1.1 ad
2072 1.1 ad for (i = 0; i < bin->regs_mask_nelms; i++)
2073 1.1 ad run->regs_mask[i] = UINT_MAX;
2074 1.1 ad remainder = bin->nregs & ((1 << (SIZEOF_INT_2POW + 3)) - 1);
2075 1.1 ad if (remainder != 0) {
2076 1.1 ad /* The last element has spare bits that need to be unset. */
2077 1.1 ad run->regs_mask[i] = (UINT_MAX >> ((1 << (SIZEOF_INT_2POW + 3))
2078 1.1 ad - remainder));
2079 1.1 ad }
2080 1.1 ad
2081 1.1 ad run->regs_minelm = 0;
2082 1.1 ad
2083 1.1 ad run->nfree = bin->nregs;
2084 1.1 ad #ifdef MALLOC_DEBUG
2085 1.1 ad run->magic = ARENA_RUN_MAGIC;
2086 1.1 ad #endif
2087 1.1 ad
2088 1.1 ad #ifdef MALLOC_STATS
2089 1.1 ad bin->stats.nruns++;
2090 1.1 ad bin->stats.curruns++;
2091 1.1 ad if (bin->stats.curruns > bin->stats.highruns)
2092 1.1 ad bin->stats.highruns = bin->stats.curruns;
2093 1.1 ad #endif
2094 1.1 ad return (run);
2095 1.1 ad }
2096 1.1 ad
2097 1.1 ad /* bin->runcur must have space available before this function is called. */
2098 1.1 ad static inline void *
2099 1.1 ad arena_bin_malloc_easy(arena_t *arena, arena_bin_t *bin, arena_run_t *run)
2100 1.1 ad {
2101 1.1 ad void *ret;
2102 1.1 ad
2103 1.1 ad assert(run->magic == ARENA_RUN_MAGIC);
2104 1.1 ad assert(run->nfree > 0);
2105 1.1 ad
2106 1.1 ad ret = arena_run_reg_alloc(run, bin);
2107 1.1 ad assert(ret != NULL);
2108 1.1 ad run->nfree--;
2109 1.1 ad
2110 1.1 ad return (ret);
2111 1.1 ad }
2112 1.1 ad
2113 1.1 ad /* Re-fill bin->runcur, then call arena_bin_malloc_easy(). */
2114 1.1 ad static void *
2115 1.1 ad arena_bin_malloc_hard(arena_t *arena, arena_bin_t *bin)
2116 1.1 ad {
2117 1.1 ad
2118 1.1 ad bin->runcur = arena_bin_nonfull_run_get(arena, bin);
2119 1.1 ad if (bin->runcur == NULL)
2120 1.1 ad return (NULL);
2121 1.1 ad assert(bin->runcur->magic == ARENA_RUN_MAGIC);
2122 1.1 ad assert(bin->runcur->nfree > 0);
2123 1.1 ad
2124 1.1 ad return (arena_bin_malloc_easy(arena, bin, bin->runcur));
2125 1.1 ad }
2126 1.1 ad
2127 1.1 ad /*
2128 1.1 ad * Calculate bin->run_size such that it meets the following constraints:
2129 1.1 ad *
2130 1.1 ad * *) bin->run_size >= min_run_size
2131 1.1 ad * *) bin->run_size <= arena_maxclass
2132 1.1 ad * *) bin->run_size <= RUN_MAX_SMALL
2133 1.1 ad * *) run header overhead <= RUN_MAX_OVRHD (or header overhead relaxed).
2134 1.1 ad *
2135 1.1 ad * bin->nregs, bin->regs_mask_nelms, and bin->reg0_offset are
2136 1.1 ad * also calculated here, since these settings are all interdependent.
2137 1.1 ad */
2138 1.1 ad static size_t
2139 1.1 ad arena_bin_run_size_calc(arena_bin_t *bin, size_t min_run_size)
2140 1.1 ad {
2141 1.1 ad size_t try_run_size, good_run_size;
2142 1.1 ad unsigned good_nregs, good_mask_nelms, good_reg0_offset;
2143 1.1 ad unsigned try_nregs, try_mask_nelms, try_reg0_offset;
2144 1.1 ad float max_ovrhd = RUN_MAX_OVRHD;
2145 1.1 ad
2146 1.1 ad assert(min_run_size >= pagesize);
2147 1.1 ad assert(min_run_size <= arena_maxclass);
2148 1.1 ad assert(min_run_size <= RUN_MAX_SMALL);
2149 1.1 ad
2150 1.1 ad /*
2151 1.1 ad * Calculate known-valid settings before entering the run_size
2152 1.1 ad * expansion loop, so that the first part of the loop always copies
2153 1.1 ad * valid settings.
2154 1.1 ad *
2155 1.1 ad * The do..while loop iteratively reduces the number of regions until
2156 1.1 ad * the run header and the regions no longer overlap. A closed formula
2157 1.1 ad * would be quite messy, since there is an interdependency between the
2158 1.1 ad * header's mask length and the number of regions.
2159 1.1 ad */
2160 1.1 ad try_run_size = min_run_size;
2161 1.1 ad try_nregs = ((try_run_size - sizeof(arena_run_t)) / bin->reg_size)
2162 1.1 ad + 1; /* Counter-act the first line of the loop. */
2163 1.1 ad do {
2164 1.1 ad try_nregs--;
2165 1.1 ad try_mask_nelms = (try_nregs >> (SIZEOF_INT_2POW + 3)) +
2166 1.1 ad ((try_nregs & ((1 << (SIZEOF_INT_2POW + 3)) - 1)) ? 1 : 0);
2167 1.1 ad try_reg0_offset = try_run_size - (try_nregs * bin->reg_size);
2168 1.1 ad } while (sizeof(arena_run_t) + (sizeof(unsigned) * (try_mask_nelms - 1))
2169 1.1 ad > try_reg0_offset);
2170 1.1 ad
2171 1.1 ad /* run_size expansion loop. */
2172 1.1 ad do {
2173 1.1 ad /*
2174 1.1 ad * Copy valid settings before trying more aggressive settings.
2175 1.1 ad */
2176 1.1 ad good_run_size = try_run_size;
2177 1.1 ad good_nregs = try_nregs;
2178 1.1 ad good_mask_nelms = try_mask_nelms;
2179 1.1 ad good_reg0_offset = try_reg0_offset;
2180 1.1 ad
2181 1.1 ad /* Try more aggressive settings. */
2182 1.1 ad try_run_size += pagesize;
2183 1.1 ad try_nregs = ((try_run_size - sizeof(arena_run_t)) /
2184 1.1 ad bin->reg_size) + 1; /* Counter-act try_nregs-- in loop. */
2185 1.1 ad do {
2186 1.1 ad try_nregs--;
2187 1.1 ad try_mask_nelms = (try_nregs >> (SIZEOF_INT_2POW + 3)) +
2188 1.1 ad ((try_nregs & ((1 << (SIZEOF_INT_2POW + 3)) - 1)) ?
2189 1.1 ad 1 : 0);
2190 1.1 ad try_reg0_offset = try_run_size - (try_nregs *
2191 1.1 ad bin->reg_size);
2192 1.1 ad } while (sizeof(arena_run_t) + (sizeof(unsigned) *
2193 1.1 ad (try_mask_nelms - 1)) > try_reg0_offset);
2194 1.1 ad } while (try_run_size <= arena_maxclass && try_run_size <= RUN_MAX_SMALL
2195 1.1 ad && max_ovrhd > RUN_MAX_OVRHD_RELAX / ((float)(bin->reg_size << 3))
2196 1.1 ad && ((float)(try_reg0_offset)) / ((float)(try_run_size)) >
2197 1.1 ad max_ovrhd);
2198 1.1 ad
2199 1.1 ad assert(sizeof(arena_run_t) + (sizeof(unsigned) * (good_mask_nelms - 1))
2200 1.1 ad <= good_reg0_offset);
2201 1.1 ad assert((good_mask_nelms << (SIZEOF_INT_2POW + 3)) >= good_nregs);
2202 1.1 ad
2203 1.1 ad /* Copy final settings. */
2204 1.1 ad bin->run_size = good_run_size;
2205 1.1 ad bin->nregs = good_nregs;
2206 1.1 ad bin->regs_mask_nelms = good_mask_nelms;
2207 1.1 ad bin->reg0_offset = good_reg0_offset;
2208 1.1 ad
2209 1.1 ad return (good_run_size);
2210 1.1 ad }
2211 1.1 ad
2212 1.1 ad static void *
2213 1.1 ad arena_malloc(arena_t *arena, size_t size)
2214 1.1 ad {
2215 1.1 ad void *ret;
2216 1.1 ad
2217 1.1 ad assert(arena != NULL);
2218 1.1 ad assert(arena->magic == ARENA_MAGIC);
2219 1.1 ad assert(size != 0);
2220 1.1 ad assert(QUANTUM_CEILING(size) <= arena_maxclass);
2221 1.1 ad
2222 1.1 ad if (size <= bin_maxclass) {
2223 1.1 ad arena_bin_t *bin;
2224 1.1 ad arena_run_t *run;
2225 1.1 ad
2226 1.1 ad /* Small allocation. */
2227 1.1 ad
2228 1.1 ad if (size < small_min) {
2229 1.1 ad /* Tiny. */
2230 1.1 ad size = pow2_ceil(size);
2231 1.1 ad bin = &arena->bins[ffs((int)(size >> (TINY_MIN_2POW +
2232 1.1 ad 1)))];
2233 1.1 ad #if (!defined(NDEBUG) || defined(MALLOC_STATS))
2234 1.1 ad /*
2235 1.1 ad * Bin calculation is always correct, but we may need
2236 1.1 ad * to fix size for the purposes of assertions and/or
2237 1.1 ad * stats accuracy.
2238 1.1 ad */
2239 1.1 ad if (size < (1 << TINY_MIN_2POW))
2240 1.1 ad size = (1 << TINY_MIN_2POW);
2241 1.1 ad #endif
2242 1.1 ad } else if (size <= small_max) {
2243 1.1 ad /* Quantum-spaced. */
2244 1.1 ad size = QUANTUM_CEILING(size);
2245 1.1 ad bin = &arena->bins[ntbins + (size >> opt_quantum_2pow)
2246 1.1 ad - 1];
2247 1.1 ad } else {
2248 1.1 ad /* Sub-page. */
2249 1.1 ad size = pow2_ceil(size);
2250 1.1 ad bin = &arena->bins[ntbins + nqbins
2251 1.1 ad + (ffs((int)(size >> opt_small_max_2pow)) - 2)];
2252 1.1 ad }
2253 1.1 ad assert(size == bin->reg_size);
2254 1.1 ad
2255 1.1 ad malloc_mutex_lock(&arena->mtx);
2256 1.1 ad if ((run = bin->runcur) != NULL && run->nfree > 0)
2257 1.1 ad ret = arena_bin_malloc_easy(arena, bin, run);
2258 1.1 ad else
2259 1.1 ad ret = arena_bin_malloc_hard(arena, bin);
2260 1.1 ad
2261 1.1 ad if (ret == NULL) {
2262 1.1 ad malloc_mutex_unlock(&arena->mtx);
2263 1.1 ad return (NULL);
2264 1.1 ad }
2265 1.1 ad
2266 1.1 ad #ifdef MALLOC_STATS
2267 1.1 ad bin->stats.nrequests++;
2268 1.1 ad arena->stats.nmalloc_small++;
2269 1.1 ad arena->stats.allocated_small += size;
2270 1.1 ad #endif
2271 1.1 ad } else {
2272 1.1 ad /* Large allocation. */
2273 1.1 ad size = PAGE_CEILING(size);
2274 1.1 ad malloc_mutex_lock(&arena->mtx);
2275 1.1 ad ret = (void *)arena_run_alloc(arena, size);
2276 1.1 ad if (ret == NULL) {
2277 1.1 ad malloc_mutex_unlock(&arena->mtx);
2278 1.1 ad return (NULL);
2279 1.1 ad }
2280 1.1 ad #ifdef MALLOC_STATS
2281 1.1 ad arena->stats.nmalloc_large++;
2282 1.1 ad arena->stats.allocated_large += size;
2283 1.1 ad #endif
2284 1.1 ad }
2285 1.1 ad
2286 1.1 ad malloc_mutex_unlock(&arena->mtx);
2287 1.1 ad
2288 1.1 ad if (opt_junk)
2289 1.1 ad memset(ret, 0xa5, size);
2290 1.1 ad else if (opt_zero)
2291 1.1 ad memset(ret, 0, size);
2292 1.1 ad return (ret);
2293 1.1 ad }
2294 1.1 ad
2295 1.1 ad static inline void
2296 1.1 ad arena_palloc_trim(arena_t *arena, arena_chunk_t *chunk, unsigned pageind,
2297 1.1 ad unsigned npages)
2298 1.1 ad {
2299 1.1 ad unsigned i;
2300 1.1 ad
2301 1.1 ad assert(npages > 0);
2302 1.1 ad
2303 1.1 ad /*
2304 1.1 ad * Modifiy the map such that arena_run_dalloc() sees the run as
2305 1.1 ad * separately allocated.
2306 1.1 ad */
2307 1.1 ad for (i = 0; i < npages; i++) {
2308 1.1 ad chunk->map[pageind + i].npages = npages;
2309 1.1 ad chunk->map[pageind + i].pos = i;
2310 1.1 ad }
2311 1.1 ad arena_run_dalloc(arena, (arena_run_t *)((uintptr_t)chunk + (pageind <<
2312 1.1 ad pagesize_2pow)), npages << pagesize_2pow);
2313 1.1 ad }
2314 1.1 ad
2315 1.1 ad /* Only handles large allocations that require more than page alignment. */
2316 1.1 ad static void *
2317 1.1 ad arena_palloc(arena_t *arena, size_t alignment, size_t size, size_t alloc_size)
2318 1.1 ad {
2319 1.1 ad void *ret;
2320 1.1 ad size_t offset;
2321 1.1 ad arena_chunk_t *chunk;
2322 1.1 ad unsigned pageind, i, npages;
2323 1.1 ad
2324 1.1 ad assert((size & pagesize_mask) == 0);
2325 1.1 ad assert((alignment & pagesize_mask) == 0);
2326 1.1 ad
2327 1.1 ad npages = size >> pagesize_2pow;
2328 1.1 ad
2329 1.1 ad malloc_mutex_lock(&arena->mtx);
2330 1.1 ad ret = (void *)arena_run_alloc(arena, alloc_size);
2331 1.1 ad if (ret == NULL) {
2332 1.1 ad malloc_mutex_unlock(&arena->mtx);
2333 1.1 ad return (NULL);
2334 1.1 ad }
2335 1.1 ad
2336 1.1 ad chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ret);
2337 1.1 ad
2338 1.1 ad offset = (uintptr_t)ret & (alignment - 1);
2339 1.1 ad assert((offset & pagesize_mask) == 0);
2340 1.1 ad assert(offset < alloc_size);
2341 1.1 ad if (offset == 0) {
2342 1.1 ad pageind = (((uintptr_t)ret - (uintptr_t)chunk) >>
2343 1.1 ad pagesize_2pow);
2344 1.1 ad
2345 1.1 ad /* Update the map for the run to be kept. */
2346 1.1 ad for (i = 0; i < npages; i++) {
2347 1.1 ad chunk->map[pageind + i].npages = npages;
2348 1.1 ad assert(chunk->map[pageind + i].pos == i);
2349 1.1 ad }
2350 1.1 ad
2351 1.1 ad /* Trim trailing space. */
2352 1.1 ad arena_palloc_trim(arena, chunk, pageind + npages,
2353 1.1 ad (alloc_size - size) >> pagesize_2pow);
2354 1.1 ad } else {
2355 1.1 ad size_t leadsize, trailsize;
2356 1.1 ad
2357 1.1 ad leadsize = alignment - offset;
2358 1.1 ad ret = (void *)((uintptr_t)ret + leadsize);
2359 1.1 ad pageind = (((uintptr_t)ret - (uintptr_t)chunk) >>
2360 1.1 ad pagesize_2pow);
2361 1.1 ad
2362 1.1 ad /* Update the map for the run to be kept. */
2363 1.1 ad for (i = 0; i < npages; i++) {
2364 1.1 ad chunk->map[pageind + i].npages = npages;
2365 1.1 ad chunk->map[pageind + i].pos = i;
2366 1.1 ad }
2367 1.1 ad
2368 1.1 ad /* Trim leading space. */
2369 1.1 ad arena_palloc_trim(arena, chunk, pageind - (leadsize >>
2370 1.1 ad pagesize_2pow), leadsize >> pagesize_2pow);
2371 1.1 ad
2372 1.1 ad trailsize = alloc_size - leadsize - size;
2373 1.1 ad if (trailsize != 0) {
2374 1.1 ad /* Trim trailing space. */
2375 1.1 ad assert(trailsize < alloc_size);
2376 1.1 ad arena_palloc_trim(arena, chunk, pageind + npages,
2377 1.1 ad trailsize >> pagesize_2pow);
2378 1.1 ad }
2379 1.1 ad }
2380 1.1 ad
2381 1.1 ad #ifdef MALLOC_STATS
2382 1.1 ad arena->stats.nmalloc_large++;
2383 1.1 ad arena->stats.allocated_large += size;
2384 1.1 ad #endif
2385 1.1 ad malloc_mutex_unlock(&arena->mtx);
2386 1.1 ad
2387 1.1 ad if (opt_junk)
2388 1.1 ad memset(ret, 0xa5, size);
2389 1.1 ad else if (opt_zero)
2390 1.1 ad memset(ret, 0, size);
2391 1.1 ad return (ret);
2392 1.1 ad }
2393 1.1 ad
2394 1.1 ad /* Return the size of the allocation pointed to by ptr. */
2395 1.1 ad static size_t
2396 1.1 ad arena_salloc(const void *ptr)
2397 1.1 ad {
2398 1.1 ad size_t ret;
2399 1.1 ad arena_chunk_t *chunk;
2400 1.1 ad arena_chunk_map_t *mapelm;
2401 1.1 ad unsigned pageind;
2402 1.1 ad
2403 1.1 ad assert(ptr != NULL);
2404 1.1 ad assert(CHUNK_ADDR2BASE(ptr) != ptr);
2405 1.1 ad
2406 1.1 ad /*
2407 1.1 ad * No arena data structures that we query here can change in a way that
2408 1.1 ad * affects this function, so we don't need to lock.
2409 1.1 ad */
2410 1.1 ad chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr);
2411 1.1 ad pageind = (((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow);
2412 1.1 ad mapelm = &chunk->map[pageind];
2413 1.2 ad if (mapelm->pos != 0 || ptr != (char *)((uintptr_t)chunk) + (pageind <<
2414 1.1 ad pagesize_2pow)) {
2415 1.1 ad arena_run_t *run;
2416 1.1 ad
2417 1.1 ad pageind -= mapelm->pos;
2418 1.1 ad
2419 1.1 ad run = (arena_run_t *)((uintptr_t)chunk + (pageind <<
2420 1.1 ad pagesize_2pow));
2421 1.1 ad assert(run->magic == ARENA_RUN_MAGIC);
2422 1.1 ad ret = run->bin->reg_size;
2423 1.1 ad } else
2424 1.1 ad ret = mapelm->npages << pagesize_2pow;
2425 1.1 ad
2426 1.1 ad return (ret);
2427 1.1 ad }
2428 1.1 ad
2429 1.1 ad static void *
2430 1.1 ad arena_ralloc(void *ptr, size_t size, size_t oldsize)
2431 1.1 ad {
2432 1.1 ad void *ret;
2433 1.1 ad
2434 1.1 ad /* Avoid moving the allocation if the size class would not change. */
2435 1.1 ad if (size < small_min) {
2436 1.1 ad if (oldsize < small_min &&
2437 1.1 ad ffs((int)(pow2_ceil(size) >> (TINY_MIN_2POW + 1)))
2438 1.1 ad == ffs((int)(pow2_ceil(oldsize) >> (TINY_MIN_2POW + 1))))
2439 1.1 ad goto IN_PLACE;
2440 1.1 ad } else if (size <= small_max) {
2441 1.1 ad if (oldsize >= small_min && oldsize <= small_max &&
2442 1.1 ad (QUANTUM_CEILING(size) >> opt_quantum_2pow)
2443 1.1 ad == (QUANTUM_CEILING(oldsize) >> opt_quantum_2pow))
2444 1.1 ad goto IN_PLACE;
2445 1.1 ad } else {
2446 1.1 ad /*
2447 1.1 ad * We make no attempt to resize runs here, though it would be
2448 1.1 ad * possible to do so.
2449 1.1 ad */
2450 1.1 ad if (oldsize > small_max && PAGE_CEILING(size) == oldsize)
2451 1.1 ad goto IN_PLACE;
2452 1.1 ad }
2453 1.1 ad
2454 1.1 ad /*
2455 1.1 ad * If we get here, then size and oldsize are different enough that we
2456 1.1 ad * need to use a different size class. In that case, fall back to
2457 1.1 ad * allocating new space and copying.
2458 1.1 ad */
2459 1.1 ad ret = arena_malloc(choose_arena(), size);
2460 1.1 ad if (ret == NULL)
2461 1.1 ad return (NULL);
2462 1.1 ad
2463 1.1 ad /* Junk/zero-filling were already done by arena_malloc(). */
2464 1.1 ad if (size < oldsize)
2465 1.1 ad memcpy(ret, ptr, size);
2466 1.1 ad else
2467 1.1 ad memcpy(ret, ptr, oldsize);
2468 1.1 ad idalloc(ptr);
2469 1.1 ad return (ret);
2470 1.1 ad IN_PLACE:
2471 1.1 ad if (opt_junk && size < oldsize)
2472 1.1 ad memset((void *)((uintptr_t)ptr + size), 0x5a, oldsize - size);
2473 1.1 ad else if (opt_zero && size > oldsize)
2474 1.1 ad memset((void *)((uintptr_t)ptr + oldsize), 0, size - oldsize);
2475 1.1 ad return (ptr);
2476 1.1 ad }
2477 1.1 ad
2478 1.1 ad static void
2479 1.1 ad arena_dalloc(arena_t *arena, arena_chunk_t *chunk, void *ptr)
2480 1.1 ad {
2481 1.1 ad unsigned pageind;
2482 1.1 ad arena_chunk_map_t *mapelm;
2483 1.1 ad size_t size;
2484 1.1 ad
2485 1.1 ad assert(arena != NULL);
2486 1.1 ad assert(arena->magic == ARENA_MAGIC);
2487 1.1 ad assert(chunk->arena == arena);
2488 1.1 ad assert(ptr != NULL);
2489 1.1 ad assert(CHUNK_ADDR2BASE(ptr) != ptr);
2490 1.1 ad
2491 1.1 ad pageind = (((uintptr_t)ptr - (uintptr_t)chunk) >> pagesize_2pow);
2492 1.1 ad mapelm = &chunk->map[pageind];
2493 1.2 ad if (mapelm->pos != 0 || ptr != (char *)((uintptr_t)chunk) + (pageind <<
2494 1.1 ad pagesize_2pow)) {
2495 1.1 ad arena_run_t *run;
2496 1.1 ad arena_bin_t *bin;
2497 1.1 ad
2498 1.1 ad /* Small allocation. */
2499 1.1 ad
2500 1.1 ad pageind -= mapelm->pos;
2501 1.1 ad
2502 1.1 ad run = (arena_run_t *)((uintptr_t)chunk + (pageind <<
2503 1.1 ad pagesize_2pow));
2504 1.1 ad assert(run->magic == ARENA_RUN_MAGIC);
2505 1.1 ad bin = run->bin;
2506 1.1 ad size = bin->reg_size;
2507 1.1 ad
2508 1.1 ad if (opt_junk)
2509 1.1 ad memset(ptr, 0x5a, size);
2510 1.1 ad
2511 1.1 ad malloc_mutex_lock(&arena->mtx);
2512 1.1 ad arena_run_reg_dalloc(run, bin, ptr, size);
2513 1.1 ad run->nfree++;
2514 1.1 ad
2515 1.1 ad if (run->nfree == bin->nregs) {
2516 1.1 ad /* Deallocate run. */
2517 1.1 ad if (run == bin->runcur)
2518 1.1 ad bin->runcur = NULL;
2519 1.1 ad else if (bin->nregs != 1) {
2520 1.1 ad /*
2521 1.1 ad * This block's conditional is necessary because
2522 1.1 ad * if the run only contains one region, then it
2523 1.1 ad * never gets inserted into the non-full runs
2524 1.1 ad * tree.
2525 1.1 ad */
2526 1.2 ad /* LINTED */
2527 1.1 ad RB_REMOVE(arena_run_tree_s, &bin->runs, run);
2528 1.1 ad }
2529 1.1 ad #ifdef MALLOC_DEBUG
2530 1.1 ad run->magic = 0;
2531 1.1 ad #endif
2532 1.1 ad arena_run_dalloc(arena, run, bin->run_size);
2533 1.1 ad #ifdef MALLOC_STATS
2534 1.1 ad bin->stats.curruns--;
2535 1.1 ad #endif
2536 1.1 ad } else if (run->nfree == 1 && run != bin->runcur) {
2537 1.1 ad /*
2538 1.1 ad * Make sure that bin->runcur always refers to the
2539 1.1 ad * lowest non-full run, if one exists.
2540 1.1 ad */
2541 1.1 ad if (bin->runcur == NULL)
2542 1.1 ad bin->runcur = run;
2543 1.1 ad else if ((uintptr_t)run < (uintptr_t)bin->runcur) {
2544 1.1 ad /* Switch runcur. */
2545 1.1 ad if (bin->runcur->nfree > 0) {
2546 1.1 ad /* Insert runcur. */
2547 1.2 ad /* LINTED */
2548 1.1 ad RB_INSERT(arena_run_tree_s, &bin->runs,
2549 1.1 ad bin->runcur);
2550 1.1 ad }
2551 1.1 ad bin->runcur = run;
2552 1.2 ad } else {
2553 1.2 ad /* LINTED */
2554 1.1 ad RB_INSERT(arena_run_tree_s, &bin->runs, run);
2555 1.2 ad }
2556 1.1 ad }
2557 1.1 ad #ifdef MALLOC_STATS
2558 1.1 ad arena->stats.allocated_small -= size;
2559 1.1 ad arena->stats.ndalloc_small++;
2560 1.1 ad #endif
2561 1.1 ad } else {
2562 1.1 ad /* Large allocation. */
2563 1.1 ad
2564 1.1 ad size = mapelm->npages << pagesize_2pow;
2565 1.1 ad assert((((uintptr_t)ptr) & pagesize_mask) == 0);
2566 1.1 ad
2567 1.1 ad if (opt_junk)
2568 1.1 ad memset(ptr, 0x5a, size);
2569 1.1 ad
2570 1.1 ad malloc_mutex_lock(&arena->mtx);
2571 1.1 ad arena_run_dalloc(arena, (arena_run_t *)ptr, size);
2572 1.1 ad #ifdef MALLOC_STATS
2573 1.1 ad arena->stats.allocated_large -= size;
2574 1.1 ad arena->stats.ndalloc_large++;
2575 1.1 ad #endif
2576 1.1 ad }
2577 1.1 ad
2578 1.1 ad malloc_mutex_unlock(&arena->mtx);
2579 1.1 ad }
2580 1.1 ad
2581 1.1 ad static bool
2582 1.1 ad arena_new(arena_t *arena)
2583 1.1 ad {
2584 1.1 ad unsigned i;
2585 1.1 ad arena_bin_t *bin;
2586 1.2 ad size_t prev_run_size;
2587 1.1 ad
2588 1.1 ad malloc_mutex_init(&arena->mtx);
2589 1.1 ad
2590 1.1 ad #ifdef MALLOC_STATS
2591 1.1 ad memset(&arena->stats, 0, sizeof(arena_stats_t));
2592 1.1 ad #endif
2593 1.1 ad
2594 1.1 ad /* Initialize chunks. */
2595 1.1 ad RB_INIT(&arena->chunks);
2596 1.1 ad arena->spare = NULL;
2597 1.1 ad
2598 1.1 ad /* Initialize bins. */
2599 1.1 ad prev_run_size = pagesize;
2600 1.1 ad
2601 1.1 ad /* (2^n)-spaced tiny bins. */
2602 1.1 ad for (i = 0; i < ntbins; i++) {
2603 1.1 ad bin = &arena->bins[i];
2604 1.1 ad bin->runcur = NULL;
2605 1.1 ad RB_INIT(&bin->runs);
2606 1.1 ad
2607 1.1 ad bin->reg_size = (1 << (TINY_MIN_2POW + i));
2608 1.1 ad prev_run_size = arena_bin_run_size_calc(bin, prev_run_size);
2609 1.1 ad
2610 1.1 ad #ifdef MALLOC_STATS
2611 1.1 ad memset(&bin->stats, 0, sizeof(malloc_bin_stats_t));
2612 1.1 ad #endif
2613 1.1 ad }
2614 1.1 ad
2615 1.1 ad /* Quantum-spaced bins. */
2616 1.1 ad for (; i < ntbins + nqbins; i++) {
2617 1.1 ad bin = &arena->bins[i];
2618 1.1 ad bin->runcur = NULL;
2619 1.1 ad RB_INIT(&bin->runs);
2620 1.1 ad
2621 1.1 ad bin->reg_size = quantum * (i - ntbins + 1);
2622 1.2 ad /*
2623 1.1 ad pow2_size = pow2_ceil(quantum * (i - ntbins + 1));
2624 1.2 ad */
2625 1.1 ad prev_run_size = arena_bin_run_size_calc(bin, prev_run_size);
2626 1.1 ad
2627 1.1 ad #ifdef MALLOC_STATS
2628 1.1 ad memset(&bin->stats, 0, sizeof(malloc_bin_stats_t));
2629 1.1 ad #endif
2630 1.1 ad }
2631 1.1 ad
2632 1.1 ad /* (2^n)-spaced sub-page bins. */
2633 1.1 ad for (; i < ntbins + nqbins + nsbins; i++) {
2634 1.1 ad bin = &arena->bins[i];
2635 1.1 ad bin->runcur = NULL;
2636 1.1 ad RB_INIT(&bin->runs);
2637 1.1 ad
2638 1.1 ad bin->reg_size = (small_max << (i - (ntbins + nqbins) + 1));
2639 1.1 ad
2640 1.1 ad prev_run_size = arena_bin_run_size_calc(bin, prev_run_size);
2641 1.1 ad
2642 1.1 ad #ifdef MALLOC_STATS
2643 1.1 ad memset(&bin->stats, 0, sizeof(malloc_bin_stats_t));
2644 1.1 ad #endif
2645 1.1 ad }
2646 1.1 ad
2647 1.1 ad #ifdef MALLOC_DEBUG
2648 1.1 ad arena->magic = ARENA_MAGIC;
2649 1.1 ad #endif
2650 1.1 ad
2651 1.1 ad return (false);
2652 1.1 ad }
2653 1.1 ad
2654 1.1 ad /* Create a new arena and insert it into the arenas array at index ind. */
2655 1.1 ad static arena_t *
2656 1.1 ad arenas_extend(unsigned ind)
2657 1.1 ad {
2658 1.1 ad arena_t *ret;
2659 1.1 ad
2660 1.1 ad /* Allocate enough space for trailing bins. */
2661 1.1 ad ret = (arena_t *)base_alloc(sizeof(arena_t)
2662 1.1 ad + (sizeof(arena_bin_t) * (ntbins + nqbins + nsbins - 1)));
2663 1.1 ad if (ret != NULL && arena_new(ret) == false) {
2664 1.1 ad arenas[ind] = ret;
2665 1.1 ad return (ret);
2666 1.1 ad }
2667 1.1 ad /* Only reached if there is an OOM error. */
2668 1.1 ad
2669 1.1 ad /*
2670 1.1 ad * OOM here is quite inconvenient to propagate, since dealing with it
2671 1.1 ad * would require a check for failure in the fast path. Instead, punt
2672 1.1 ad * by using arenas[0]. In practice, this is an extremely unlikely
2673 1.1 ad * failure.
2674 1.1 ad */
2675 1.1 ad _malloc_message(_getprogname(),
2676 1.1 ad ": (malloc) Error initializing arena\n", "", "");
2677 1.1 ad if (opt_abort)
2678 1.1 ad abort();
2679 1.1 ad
2680 1.1 ad return (arenas[0]);
2681 1.1 ad }
2682 1.1 ad
2683 1.1 ad /*
2684 1.1 ad * End arena.
2685 1.1 ad */
2686 1.1 ad /******************************************************************************/
2687 1.1 ad /*
2688 1.1 ad * Begin general internal functions.
2689 1.1 ad */
2690 1.1 ad
2691 1.1 ad static void *
2692 1.1 ad huge_malloc(size_t size)
2693 1.1 ad {
2694 1.1 ad void *ret;
2695 1.1 ad size_t csize;
2696 1.1 ad chunk_node_t *node;
2697 1.1 ad
2698 1.1 ad /* Allocate one or more contiguous chunks for this request. */
2699 1.1 ad
2700 1.1 ad csize = CHUNK_CEILING(size);
2701 1.1 ad if (csize == 0) {
2702 1.1 ad /* size is large enough to cause size_t wrap-around. */
2703 1.1 ad return (NULL);
2704 1.1 ad }
2705 1.1 ad
2706 1.1 ad /* Allocate a chunk node with which to track the chunk. */
2707 1.1 ad node = base_chunk_node_alloc();
2708 1.1 ad if (node == NULL)
2709 1.1 ad return (NULL);
2710 1.1 ad
2711 1.1 ad ret = chunk_alloc(csize);
2712 1.1 ad if (ret == NULL) {
2713 1.1 ad base_chunk_node_dealloc(node);
2714 1.1 ad return (NULL);
2715 1.1 ad }
2716 1.1 ad
2717 1.1 ad /* Insert node into huge. */
2718 1.1 ad node->chunk = ret;
2719 1.1 ad node->size = csize;
2720 1.1 ad
2721 1.1 ad malloc_mutex_lock(&chunks_mtx);
2722 1.1 ad RB_INSERT(chunk_tree_s, &huge, node);
2723 1.1 ad #ifdef MALLOC_STATS
2724 1.1 ad huge_nmalloc++;
2725 1.1 ad huge_allocated += csize;
2726 1.1 ad #endif
2727 1.1 ad malloc_mutex_unlock(&chunks_mtx);
2728 1.1 ad
2729 1.1 ad if (opt_junk)
2730 1.1 ad memset(ret, 0xa5, csize);
2731 1.1 ad else if (opt_zero)
2732 1.1 ad memset(ret, 0, csize);
2733 1.1 ad
2734 1.1 ad return (ret);
2735 1.1 ad }
2736 1.1 ad
2737 1.1 ad /* Only handles large allocations that require more than chunk alignment. */
2738 1.1 ad static void *
2739 1.1 ad huge_palloc(size_t alignment, size_t size)
2740 1.1 ad {
2741 1.1 ad void *ret;
2742 1.1 ad size_t alloc_size, chunk_size, offset;
2743 1.1 ad chunk_node_t *node;
2744 1.1 ad
2745 1.1 ad /*
2746 1.1 ad * This allocation requires alignment that is even larger than chunk
2747 1.1 ad * alignment. This means that huge_malloc() isn't good enough.
2748 1.1 ad *
2749 1.1 ad * Allocate almost twice as many chunks as are demanded by the size or
2750 1.1 ad * alignment, in order to assure the alignment can be achieved, then
2751 1.1 ad * unmap leading and trailing chunks.
2752 1.1 ad */
2753 1.1 ad assert(alignment >= chunksize);
2754 1.1 ad
2755 1.1 ad chunk_size = CHUNK_CEILING(size);
2756 1.1 ad
2757 1.1 ad if (size >= alignment)
2758 1.1 ad alloc_size = chunk_size + alignment - chunksize;
2759 1.1 ad else
2760 1.1 ad alloc_size = (alignment << 1) - chunksize;
2761 1.1 ad
2762 1.1 ad /* Allocate a chunk node with which to track the chunk. */
2763 1.1 ad node = base_chunk_node_alloc();
2764 1.1 ad if (node == NULL)
2765 1.1 ad return (NULL);
2766 1.1 ad
2767 1.1 ad ret = chunk_alloc(alloc_size);
2768 1.1 ad if (ret == NULL) {
2769 1.1 ad base_chunk_node_dealloc(node);
2770 1.1 ad return (NULL);
2771 1.1 ad }
2772 1.1 ad
2773 1.1 ad offset = (uintptr_t)ret & (alignment - 1);
2774 1.1 ad assert((offset & chunksize_mask) == 0);
2775 1.1 ad assert(offset < alloc_size);
2776 1.1 ad if (offset == 0) {
2777 1.1 ad /* Trim trailing space. */
2778 1.1 ad chunk_dealloc((void *)((uintptr_t)ret + chunk_size), alloc_size
2779 1.1 ad - chunk_size);
2780 1.1 ad } else {
2781 1.1 ad size_t trailsize;
2782 1.1 ad
2783 1.1 ad /* Trim leading space. */
2784 1.1 ad chunk_dealloc(ret, alignment - offset);
2785 1.1 ad
2786 1.1 ad ret = (void *)((uintptr_t)ret + (alignment - offset));
2787 1.1 ad
2788 1.1 ad trailsize = alloc_size - (alignment - offset) - chunk_size;
2789 1.1 ad if (trailsize != 0) {
2790 1.1 ad /* Trim trailing space. */
2791 1.1 ad assert(trailsize < alloc_size);
2792 1.1 ad chunk_dealloc((void *)((uintptr_t)ret + chunk_size),
2793 1.1 ad trailsize);
2794 1.1 ad }
2795 1.1 ad }
2796 1.1 ad
2797 1.1 ad /* Insert node into huge. */
2798 1.1 ad node->chunk = ret;
2799 1.1 ad node->size = chunk_size;
2800 1.1 ad
2801 1.1 ad malloc_mutex_lock(&chunks_mtx);
2802 1.1 ad RB_INSERT(chunk_tree_s, &huge, node);
2803 1.1 ad #ifdef MALLOC_STATS
2804 1.1 ad huge_nmalloc++;
2805 1.1 ad huge_allocated += chunk_size;
2806 1.1 ad #endif
2807 1.1 ad malloc_mutex_unlock(&chunks_mtx);
2808 1.1 ad
2809 1.1 ad if (opt_junk)
2810 1.1 ad memset(ret, 0xa5, chunk_size);
2811 1.1 ad else if (opt_zero)
2812 1.1 ad memset(ret, 0, chunk_size);
2813 1.1 ad
2814 1.1 ad return (ret);
2815 1.1 ad }
2816 1.1 ad
2817 1.1 ad static void *
2818 1.1 ad huge_ralloc(void *ptr, size_t size, size_t oldsize)
2819 1.1 ad {
2820 1.1 ad void *ret;
2821 1.1 ad
2822 1.1 ad /* Avoid moving the allocation if the size class would not change. */
2823 1.1 ad if (oldsize > arena_maxclass &&
2824 1.1 ad CHUNK_CEILING(size) == CHUNK_CEILING(oldsize)) {
2825 1.1 ad if (opt_junk && size < oldsize) {
2826 1.1 ad memset((void *)((uintptr_t)ptr + size), 0x5a, oldsize
2827 1.1 ad - size);
2828 1.1 ad } else if (opt_zero && size > oldsize) {
2829 1.1 ad memset((void *)((uintptr_t)ptr + oldsize), 0, size
2830 1.1 ad - oldsize);
2831 1.1 ad }
2832 1.1 ad return (ptr);
2833 1.1 ad }
2834 1.1 ad
2835 1.1 ad /*
2836 1.1 ad * If we get here, then size and oldsize are different enough that we
2837 1.1 ad * need to use a different size class. In that case, fall back to
2838 1.1 ad * allocating new space and copying.
2839 1.1 ad */
2840 1.1 ad ret = huge_malloc(size);
2841 1.1 ad if (ret == NULL)
2842 1.1 ad return (NULL);
2843 1.1 ad
2844 1.1 ad if (CHUNK_ADDR2BASE(ptr) == ptr) {
2845 1.1 ad /* The old allocation is a chunk. */
2846 1.1 ad if (size < oldsize)
2847 1.1 ad memcpy(ret, ptr, size);
2848 1.1 ad else
2849 1.1 ad memcpy(ret, ptr, oldsize);
2850 1.1 ad } else {
2851 1.1 ad /* The old allocation is a region. */
2852 1.1 ad assert(oldsize < size);
2853 1.1 ad memcpy(ret, ptr, oldsize);
2854 1.1 ad }
2855 1.1 ad idalloc(ptr);
2856 1.1 ad return (ret);
2857 1.1 ad }
2858 1.1 ad
2859 1.1 ad static void
2860 1.1 ad huge_dalloc(void *ptr)
2861 1.1 ad {
2862 1.1 ad chunk_node_t key;
2863 1.1 ad chunk_node_t *node;
2864 1.1 ad
2865 1.1 ad malloc_mutex_lock(&chunks_mtx);
2866 1.1 ad
2867 1.1 ad /* Extract from tree of huge allocations. */
2868 1.1 ad key.chunk = ptr;
2869 1.2 ad /* LINTED */
2870 1.1 ad node = RB_FIND(chunk_tree_s, &huge, &key);
2871 1.1 ad assert(node != NULL);
2872 1.1 ad assert(node->chunk == ptr);
2873 1.2 ad /* LINTED */
2874 1.1 ad RB_REMOVE(chunk_tree_s, &huge, node);
2875 1.1 ad
2876 1.1 ad #ifdef MALLOC_STATS
2877 1.1 ad huge_ndalloc++;
2878 1.1 ad huge_allocated -= node->size;
2879 1.1 ad #endif
2880 1.1 ad
2881 1.1 ad malloc_mutex_unlock(&chunks_mtx);
2882 1.1 ad
2883 1.1 ad /* Unmap chunk. */
2884 1.1 ad #ifdef USE_BRK
2885 1.1 ad if (opt_junk)
2886 1.1 ad memset(node->chunk, 0x5a, node->size);
2887 1.1 ad #endif
2888 1.1 ad chunk_dealloc(node->chunk, node->size);
2889 1.1 ad
2890 1.1 ad base_chunk_node_dealloc(node);
2891 1.1 ad }
2892 1.1 ad
2893 1.1 ad static void *
2894 1.1 ad imalloc(size_t size)
2895 1.1 ad {
2896 1.1 ad void *ret;
2897 1.1 ad
2898 1.1 ad assert(size != 0);
2899 1.1 ad
2900 1.1 ad if (size <= arena_maxclass)
2901 1.1 ad ret = arena_malloc(choose_arena(), size);
2902 1.1 ad else
2903 1.1 ad ret = huge_malloc(size);
2904 1.1 ad
2905 1.1 ad return (ret);
2906 1.1 ad }
2907 1.1 ad
2908 1.1 ad static void *
2909 1.1 ad ipalloc(size_t alignment, size_t size)
2910 1.1 ad {
2911 1.1 ad void *ret;
2912 1.1 ad size_t ceil_size;
2913 1.1 ad
2914 1.1 ad /*
2915 1.1 ad * Round size up to the nearest multiple of alignment.
2916 1.1 ad *
2917 1.1 ad * This done, we can take advantage of the fact that for each small
2918 1.1 ad * size class, every object is aligned at the smallest power of two
2919 1.1 ad * that is non-zero in the base two representation of the size. For
2920 1.1 ad * example:
2921 1.1 ad *
2922 1.1 ad * Size | Base 2 | Minimum alignment
2923 1.1 ad * -----+----------+------------------
2924 1.1 ad * 96 | 1100000 | 32
2925 1.1 ad * 144 | 10100000 | 32
2926 1.1 ad * 192 | 11000000 | 64
2927 1.1 ad *
2928 1.1 ad * Depending on runtime settings, it is possible that arena_malloc()
2929 1.1 ad * will further round up to a power of two, but that never causes
2930 1.1 ad * correctness issues.
2931 1.1 ad */
2932 1.1 ad ceil_size = (size + (alignment - 1)) & (-alignment);
2933 1.1 ad /*
2934 1.1 ad * (ceil_size < size) protects against the combination of maximal
2935 1.1 ad * alignment and size greater than maximal alignment.
2936 1.1 ad */
2937 1.1 ad if (ceil_size < size) {
2938 1.1 ad /* size_t overflow. */
2939 1.1 ad return (NULL);
2940 1.1 ad }
2941 1.1 ad
2942 1.1 ad if (ceil_size <= pagesize || (alignment <= pagesize
2943 1.1 ad && ceil_size <= arena_maxclass))
2944 1.1 ad ret = arena_malloc(choose_arena(), ceil_size);
2945 1.1 ad else {
2946 1.1 ad size_t run_size;
2947 1.1 ad
2948 1.1 ad /*
2949 1.1 ad * We can't achieve sub-page alignment, so round up alignment
2950 1.1 ad * permanently; it makes later calculations simpler.
2951 1.1 ad */
2952 1.1 ad alignment = PAGE_CEILING(alignment);
2953 1.1 ad ceil_size = PAGE_CEILING(size);
2954 1.1 ad /*
2955 1.1 ad * (ceil_size < size) protects against very large sizes within
2956 1.1 ad * pagesize of SIZE_T_MAX.
2957 1.1 ad *
2958 1.1 ad * (ceil_size + alignment < ceil_size) protects against the
2959 1.1 ad * combination of maximal alignment and ceil_size large enough
2960 1.1 ad * to cause overflow. This is similar to the first overflow
2961 1.1 ad * check above, but it needs to be repeated due to the new
2962 1.1 ad * ceil_size value, which may now be *equal* to maximal
2963 1.1 ad * alignment, whereas before we only detected overflow if the
2964 1.1 ad * original size was *greater* than maximal alignment.
2965 1.1 ad */
2966 1.1 ad if (ceil_size < size || ceil_size + alignment < ceil_size) {
2967 1.1 ad /* size_t overflow. */
2968 1.1 ad return (NULL);
2969 1.1 ad }
2970 1.1 ad
2971 1.1 ad /*
2972 1.1 ad * Calculate the size of the over-size run that arena_palloc()
2973 1.1 ad * would need to allocate in order to guarantee the alignment.
2974 1.1 ad */
2975 1.1 ad if (ceil_size >= alignment)
2976 1.1 ad run_size = ceil_size + alignment - pagesize;
2977 1.1 ad else {
2978 1.1 ad /*
2979 1.1 ad * It is possible that (alignment << 1) will cause
2980 1.1 ad * overflow, but it doesn't matter because we also
2981 1.1 ad * subtract pagesize, which in the case of overflow
2982 1.1 ad * leaves us with a very large run_size. That causes
2983 1.1 ad * the first conditional below to fail, which means
2984 1.1 ad * that the bogus run_size value never gets used for
2985 1.1 ad * anything important.
2986 1.1 ad */
2987 1.1 ad run_size = (alignment << 1) - pagesize;
2988 1.1 ad }
2989 1.1 ad
2990 1.1 ad if (run_size <= arena_maxclass) {
2991 1.1 ad ret = arena_palloc(choose_arena(), alignment, ceil_size,
2992 1.1 ad run_size);
2993 1.1 ad } else if (alignment <= chunksize)
2994 1.1 ad ret = huge_malloc(ceil_size);
2995 1.1 ad else
2996 1.1 ad ret = huge_palloc(alignment, ceil_size);
2997 1.1 ad }
2998 1.1 ad
2999 1.1 ad assert(((uintptr_t)ret & (alignment - 1)) == 0);
3000 1.1 ad return (ret);
3001 1.1 ad }
3002 1.1 ad
3003 1.1 ad static void *
3004 1.1 ad icalloc(size_t size)
3005 1.1 ad {
3006 1.1 ad void *ret;
3007 1.1 ad
3008 1.1 ad if (size <= arena_maxclass) {
3009 1.1 ad ret = arena_malloc(choose_arena(), size);
3010 1.1 ad if (ret == NULL)
3011 1.1 ad return (NULL);
3012 1.1 ad memset(ret, 0, size);
3013 1.1 ad } else {
3014 1.1 ad /*
3015 1.1 ad * The virtual memory system provides zero-filled pages, so
3016 1.1 ad * there is no need to do so manually, unless opt_junk is
3017 1.1 ad * enabled, in which case huge_malloc() fills huge allocations
3018 1.1 ad * with junk.
3019 1.1 ad */
3020 1.1 ad ret = huge_malloc(size);
3021 1.1 ad if (ret == NULL)
3022 1.1 ad return (NULL);
3023 1.1 ad
3024 1.1 ad if (opt_junk)
3025 1.1 ad memset(ret, 0, size);
3026 1.1 ad #ifdef USE_BRK
3027 1.1 ad else if ((uintptr_t)ret >= (uintptr_t)brk_base
3028 1.1 ad && (uintptr_t)ret < (uintptr_t)brk_max) {
3029 1.1 ad /*
3030 1.1 ad * This may be a re-used brk chunk. Therefore, zero
3031 1.1 ad * the memory.
3032 1.1 ad */
3033 1.1 ad memset(ret, 0, size);
3034 1.1 ad }
3035 1.1 ad #endif
3036 1.1 ad }
3037 1.1 ad
3038 1.1 ad return (ret);
3039 1.1 ad }
3040 1.1 ad
3041 1.1 ad static size_t
3042 1.1 ad isalloc(const void *ptr)
3043 1.1 ad {
3044 1.1 ad size_t ret;
3045 1.1 ad arena_chunk_t *chunk;
3046 1.1 ad
3047 1.1 ad assert(ptr != NULL);
3048 1.1 ad
3049 1.1 ad chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr);
3050 1.1 ad if (chunk != ptr) {
3051 1.1 ad /* Region. */
3052 1.1 ad assert(chunk->arena->magic == ARENA_MAGIC);
3053 1.1 ad
3054 1.1 ad ret = arena_salloc(ptr);
3055 1.1 ad } else {
3056 1.1 ad chunk_node_t *node, key;
3057 1.1 ad
3058 1.1 ad /* Chunk (huge allocation). */
3059 1.1 ad
3060 1.1 ad malloc_mutex_lock(&chunks_mtx);
3061 1.1 ad
3062 1.1 ad /* Extract from tree of huge allocations. */
3063 1.1 ad key.chunk = __DECONST(void *, ptr);
3064 1.2 ad /* LINTED */
3065 1.1 ad node = RB_FIND(chunk_tree_s, &huge, &key);
3066 1.1 ad assert(node != NULL);
3067 1.1 ad
3068 1.1 ad ret = node->size;
3069 1.1 ad
3070 1.1 ad malloc_mutex_unlock(&chunks_mtx);
3071 1.1 ad }
3072 1.1 ad
3073 1.1 ad return (ret);
3074 1.1 ad }
3075 1.1 ad
3076 1.1 ad static void *
3077 1.1 ad iralloc(void *ptr, size_t size)
3078 1.1 ad {
3079 1.1 ad void *ret;
3080 1.1 ad size_t oldsize;
3081 1.1 ad
3082 1.1 ad assert(ptr != NULL);
3083 1.1 ad assert(size != 0);
3084 1.1 ad
3085 1.1 ad oldsize = isalloc(ptr);
3086 1.1 ad
3087 1.1 ad if (size <= arena_maxclass)
3088 1.1 ad ret = arena_ralloc(ptr, size, oldsize);
3089 1.1 ad else
3090 1.1 ad ret = huge_ralloc(ptr, size, oldsize);
3091 1.1 ad
3092 1.1 ad return (ret);
3093 1.1 ad }
3094 1.1 ad
3095 1.1 ad static void
3096 1.1 ad idalloc(void *ptr)
3097 1.1 ad {
3098 1.1 ad arena_chunk_t *chunk;
3099 1.1 ad
3100 1.1 ad assert(ptr != NULL);
3101 1.1 ad
3102 1.1 ad chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr);
3103 1.1 ad if (chunk != ptr) {
3104 1.1 ad /* Region. */
3105 1.1 ad arena_dalloc(chunk->arena, chunk, ptr);
3106 1.1 ad } else
3107 1.1 ad huge_dalloc(ptr);
3108 1.1 ad }
3109 1.1 ad
3110 1.1 ad static void
3111 1.1 ad malloc_print_stats(void)
3112 1.1 ad {
3113 1.1 ad
3114 1.1 ad if (opt_print_stats) {
3115 1.1 ad char s[UMAX2S_BUFSIZE];
3116 1.1 ad _malloc_message("___ Begin malloc statistics ___\n", "", "",
3117 1.1 ad "");
3118 1.1 ad _malloc_message("Assertions ",
3119 1.1 ad #ifdef NDEBUG
3120 1.1 ad "disabled",
3121 1.1 ad #else
3122 1.1 ad "enabled",
3123 1.1 ad #endif
3124 1.1 ad "\n", "");
3125 1.1 ad _malloc_message("Boolean MALLOC_OPTIONS: ",
3126 1.1 ad opt_abort ? "A" : "a",
3127 1.1 ad opt_junk ? "J" : "j",
3128 1.1 ad opt_hint ? "H" : "h");
3129 1.1 ad _malloc_message(opt_utrace ? "PU" : "Pu",
3130 1.1 ad opt_sysv ? "V" : "v",
3131 1.1 ad opt_xmalloc ? "X" : "x",
3132 1.1 ad opt_zero ? "Z\n" : "z\n");
3133 1.1 ad
3134 1.1 ad _malloc_message("CPUs: ", umax2s(ncpus, s), "\n", "");
3135 1.1 ad _malloc_message("Max arenas: ", umax2s(narenas, s), "\n", "");
3136 1.1 ad _malloc_message("Pointer size: ", umax2s(sizeof(void *), s),
3137 1.1 ad "\n", "");
3138 1.1 ad _malloc_message("Quantum size: ", umax2s(quantum, s), "\n", "");
3139 1.1 ad _malloc_message("Max small size: ", umax2s(small_max, s), "\n",
3140 1.1 ad "");
3141 1.1 ad
3142 1.1 ad _malloc_message("Chunk size: ", umax2s(chunksize, s), "", "");
3143 1.1 ad _malloc_message(" (2^", umax2s(opt_chunk_2pow, s), ")\n", "");
3144 1.1 ad
3145 1.1 ad #ifdef MALLOC_STATS
3146 1.1 ad {
3147 1.1 ad size_t allocated, mapped;
3148 1.1 ad unsigned i;
3149 1.1 ad arena_t *arena;
3150 1.1 ad
3151 1.1 ad /* Calculate and print allocated/mapped stats. */
3152 1.1 ad
3153 1.1 ad /* arenas. */
3154 1.1 ad for (i = 0, allocated = 0; i < narenas; i++) {
3155 1.1 ad if (arenas[i] != NULL) {
3156 1.1 ad malloc_mutex_lock(&arenas[i]->mtx);
3157 1.1 ad allocated +=
3158 1.1 ad arenas[i]->stats.allocated_small;
3159 1.1 ad allocated +=
3160 1.1 ad arenas[i]->stats.allocated_large;
3161 1.1 ad malloc_mutex_unlock(&arenas[i]->mtx);
3162 1.1 ad }
3163 1.1 ad }
3164 1.1 ad
3165 1.1 ad /* huge/base. */
3166 1.1 ad malloc_mutex_lock(&chunks_mtx);
3167 1.1 ad allocated += huge_allocated;
3168 1.1 ad mapped = stats_chunks.curchunks * chunksize;
3169 1.1 ad malloc_mutex_unlock(&chunks_mtx);
3170 1.1 ad
3171 1.1 ad malloc_mutex_lock(&base_mtx);
3172 1.1 ad mapped += base_mapped;
3173 1.1 ad malloc_mutex_unlock(&base_mtx);
3174 1.1 ad
3175 1.1 ad malloc_printf("Allocated: %zu, mapped: %zu\n",
3176 1.1 ad allocated, mapped);
3177 1.1 ad
3178 1.1 ad /* Print chunk stats. */
3179 1.1 ad {
3180 1.1 ad chunk_stats_t chunks_stats;
3181 1.1 ad
3182 1.1 ad malloc_mutex_lock(&chunks_mtx);
3183 1.1 ad chunks_stats = stats_chunks;
3184 1.1 ad malloc_mutex_unlock(&chunks_mtx);
3185 1.1 ad
3186 1.1 ad malloc_printf("chunks: nchunks "
3187 1.1 ad "highchunks curchunks\n");
3188 1.1 ad malloc_printf(" %13llu%13lu%13lu\n",
3189 1.1 ad chunks_stats.nchunks,
3190 1.1 ad chunks_stats.highchunks,
3191 1.1 ad chunks_stats.curchunks);
3192 1.1 ad }
3193 1.1 ad
3194 1.1 ad /* Print chunk stats. */
3195 1.1 ad malloc_printf(
3196 1.1 ad "huge: nmalloc ndalloc allocated\n");
3197 1.1 ad malloc_printf(" %12llu %12llu %12zu\n",
3198 1.1 ad huge_nmalloc, huge_ndalloc, huge_allocated
3199 1.1 ad * chunksize);
3200 1.1 ad
3201 1.1 ad /* Print stats for each arena. */
3202 1.1 ad for (i = 0; i < narenas; i++) {
3203 1.1 ad arena = arenas[i];
3204 1.1 ad if (arena != NULL) {
3205 1.1 ad malloc_printf(
3206 1.2 ad "\narenas[%u] @ %p\n", i, arena);
3207 1.1 ad malloc_mutex_lock(&arena->mtx);
3208 1.1 ad stats_print(arena);
3209 1.1 ad malloc_mutex_unlock(&arena->mtx);
3210 1.1 ad }
3211 1.1 ad }
3212 1.1 ad }
3213 1.1 ad #endif /* #ifdef MALLOC_STATS */
3214 1.1 ad _malloc_message("--- End malloc statistics ---\n", "", "", "");
3215 1.1 ad }
3216 1.1 ad }
3217 1.1 ad
3218 1.1 ad /*
3219 1.1 ad * FreeBSD's pthreads implementation calls malloc(3), so the malloc
3220 1.1 ad * implementation has to take pains to avoid infinite recursion during
3221 1.1 ad * initialization.
3222 1.1 ad */
3223 1.1 ad static inline bool
3224 1.1 ad malloc_init(void)
3225 1.1 ad {
3226 1.1 ad
3227 1.1 ad if (malloc_initialized == false)
3228 1.1 ad return (malloc_init_hard());
3229 1.1 ad
3230 1.1 ad return (false);
3231 1.1 ad }
3232 1.1 ad
3233 1.1 ad static bool
3234 1.1 ad malloc_init_hard(void)
3235 1.1 ad {
3236 1.1 ad unsigned i, j;
3237 1.1 ad int linklen;
3238 1.1 ad char buf[PATH_MAX + 1];
3239 1.2 ad const char *opts = "";
3240 1.1 ad
3241 1.1 ad malloc_mutex_lock(&init_lock);
3242 1.1 ad if (malloc_initialized) {
3243 1.1 ad /*
3244 1.1 ad * Another thread initialized the allocator before this one
3245 1.1 ad * acquired init_lock.
3246 1.1 ad */
3247 1.1 ad malloc_mutex_unlock(&init_lock);
3248 1.1 ad return (false);
3249 1.1 ad }
3250 1.1 ad
3251 1.1 ad /* Get number of CPUs. */
3252 1.1 ad {
3253 1.1 ad int mib[2];
3254 1.1 ad size_t len;
3255 1.1 ad
3256 1.1 ad mib[0] = CTL_HW;
3257 1.1 ad mib[1] = HW_NCPU;
3258 1.1 ad len = sizeof(ncpus);
3259 1.1 ad if (sysctl(mib, 2, &ncpus, &len, (void *) 0, 0) == -1) {
3260 1.1 ad /* Error. */
3261 1.1 ad ncpus = 1;
3262 1.1 ad }
3263 1.1 ad }
3264 1.1 ad
3265 1.1 ad /* Get page size. */
3266 1.1 ad {
3267 1.1 ad long result;
3268 1.1 ad
3269 1.1 ad result = sysconf(_SC_PAGESIZE);
3270 1.1 ad assert(result != -1);
3271 1.1 ad pagesize = (unsigned) result;
3272 1.1 ad
3273 1.1 ad /*
3274 1.1 ad * We assume that pagesize is a power of 2 when calculating
3275 1.1 ad * pagesize_mask and pagesize_2pow.
3276 1.1 ad */
3277 1.1 ad assert(((result - 1) & result) == 0);
3278 1.1 ad pagesize_mask = result - 1;
3279 1.1 ad pagesize_2pow = ffs((int)result) - 1;
3280 1.1 ad }
3281 1.1 ad
3282 1.1 ad for (i = 0; i < 3; i++) {
3283 1.1 ad /* Get runtime configuration. */
3284 1.1 ad switch (i) {
3285 1.1 ad case 0:
3286 1.1 ad if ((linklen = readlink("/etc/malloc.conf", buf,
3287 1.1 ad sizeof(buf) - 1)) != -1) {
3288 1.1 ad /*
3289 1.1 ad * Use the contents of the "/etc/malloc.conf"
3290 1.1 ad * symbolic link's name.
3291 1.1 ad */
3292 1.1 ad buf[linklen] = '\0';
3293 1.1 ad opts = buf;
3294 1.1 ad } else {
3295 1.1 ad /* No configuration specified. */
3296 1.1 ad buf[0] = '\0';
3297 1.1 ad opts = buf;
3298 1.1 ad }
3299 1.1 ad break;
3300 1.1 ad case 1:
3301 1.1 ad if (issetugid() == 0 && (opts =
3302 1.1 ad getenv("MALLOC_OPTIONS")) != NULL) {
3303 1.1 ad /*
3304 1.1 ad * Do nothing; opts is already initialized to
3305 1.1 ad * the value of the MALLOC_OPTIONS environment
3306 1.1 ad * variable.
3307 1.1 ad */
3308 1.1 ad } else {
3309 1.1 ad /* No configuration specified. */
3310 1.1 ad buf[0] = '\0';
3311 1.1 ad opts = buf;
3312 1.1 ad }
3313 1.1 ad break;
3314 1.1 ad case 2:
3315 1.1 ad if (_malloc_options != NULL) {
3316 1.1 ad /*
3317 1.1 ad * Use options that were compiled into the program.
3318 1.1 ad */
3319 1.1 ad opts = _malloc_options;
3320 1.1 ad } else {
3321 1.1 ad /* No configuration specified. */
3322 1.1 ad buf[0] = '\0';
3323 1.1 ad opts = buf;
3324 1.1 ad }
3325 1.1 ad break;
3326 1.1 ad default:
3327 1.1 ad /* NOTREACHED */
3328 1.1 ad assert(false);
3329 1.1 ad }
3330 1.1 ad
3331 1.1 ad for (j = 0; opts[j] != '\0'; j++) {
3332 1.1 ad switch (opts[j]) {
3333 1.1 ad case 'a':
3334 1.1 ad opt_abort = false;
3335 1.1 ad break;
3336 1.1 ad case 'A':
3337 1.1 ad opt_abort = true;
3338 1.1 ad break;
3339 1.1 ad case 'h':
3340 1.1 ad opt_hint = false;
3341 1.1 ad break;
3342 1.1 ad case 'H':
3343 1.1 ad opt_hint = true;
3344 1.1 ad break;
3345 1.1 ad case 'j':
3346 1.1 ad opt_junk = false;
3347 1.1 ad break;
3348 1.1 ad case 'J':
3349 1.1 ad opt_junk = true;
3350 1.1 ad break;
3351 1.1 ad case 'k':
3352 1.1 ad /*
3353 1.1 ad * Chunks always require at least one header
3354 1.1 ad * page, so chunks can never be smaller than
3355 1.1 ad * two pages.
3356 1.1 ad */
3357 1.1 ad if (opt_chunk_2pow > pagesize_2pow + 1)
3358 1.1 ad opt_chunk_2pow--;
3359 1.1 ad break;
3360 1.1 ad case 'K':
3361 1.1 ad /*
3362 1.1 ad * There must be fewer pages in a chunk than
3363 1.1 ad * can be recorded by the pos field of
3364 1.1 ad * arena_chunk_map_t, in order to make POS_FREE
3365 1.1 ad * special.
3366 1.1 ad */
3367 1.1 ad if (opt_chunk_2pow - pagesize_2pow
3368 1.1 ad < (sizeof(uint32_t) << 3) - 1)
3369 1.1 ad opt_chunk_2pow++;
3370 1.1 ad break;
3371 1.1 ad case 'n':
3372 1.1 ad opt_narenas_lshift--;
3373 1.1 ad break;
3374 1.1 ad case 'N':
3375 1.1 ad opt_narenas_lshift++;
3376 1.1 ad break;
3377 1.1 ad case 'p':
3378 1.1 ad opt_print_stats = false;
3379 1.1 ad break;
3380 1.1 ad case 'P':
3381 1.1 ad opt_print_stats = true;
3382 1.1 ad break;
3383 1.1 ad case 'q':
3384 1.1 ad if (opt_quantum_2pow > QUANTUM_2POW_MIN)
3385 1.1 ad opt_quantum_2pow--;
3386 1.1 ad break;
3387 1.1 ad case 'Q':
3388 1.1 ad if (opt_quantum_2pow < pagesize_2pow - 1)
3389 1.1 ad opt_quantum_2pow++;
3390 1.1 ad break;
3391 1.1 ad case 's':
3392 1.1 ad if (opt_small_max_2pow > QUANTUM_2POW_MIN)
3393 1.1 ad opt_small_max_2pow--;
3394 1.1 ad break;
3395 1.1 ad case 'S':
3396 1.1 ad if (opt_small_max_2pow < pagesize_2pow - 1)
3397 1.1 ad opt_small_max_2pow++;
3398 1.1 ad break;
3399 1.1 ad case 'u':
3400 1.1 ad opt_utrace = false;
3401 1.1 ad break;
3402 1.1 ad case 'U':
3403 1.1 ad opt_utrace = true;
3404 1.1 ad break;
3405 1.1 ad case 'v':
3406 1.1 ad opt_sysv = false;
3407 1.1 ad break;
3408 1.1 ad case 'V':
3409 1.1 ad opt_sysv = true;
3410 1.1 ad break;
3411 1.1 ad case 'x':
3412 1.1 ad opt_xmalloc = false;
3413 1.1 ad break;
3414 1.1 ad case 'X':
3415 1.1 ad opt_xmalloc = true;
3416 1.1 ad break;
3417 1.1 ad case 'z':
3418 1.1 ad opt_zero = false;
3419 1.1 ad break;
3420 1.1 ad case 'Z':
3421 1.1 ad opt_zero = true;
3422 1.1 ad break;
3423 1.1 ad default: {
3424 1.1 ad char cbuf[2];
3425 1.1 ad
3426 1.1 ad cbuf[0] = opts[j];
3427 1.1 ad cbuf[1] = '\0';
3428 1.1 ad _malloc_message(_getprogname(),
3429 1.1 ad ": (malloc) Unsupported character in "
3430 1.1 ad "malloc options: '", cbuf, "'\n");
3431 1.1 ad }
3432 1.1 ad }
3433 1.1 ad }
3434 1.1 ad }
3435 1.1 ad
3436 1.1 ad /* Take care to call atexit() only once. */
3437 1.1 ad if (opt_print_stats) {
3438 1.1 ad /* Print statistics at exit. */
3439 1.1 ad atexit(malloc_print_stats);
3440 1.1 ad }
3441 1.1 ad
3442 1.1 ad /* Set variables according to the value of opt_small_max_2pow. */
3443 1.1 ad if (opt_small_max_2pow < opt_quantum_2pow)
3444 1.1 ad opt_small_max_2pow = opt_quantum_2pow;
3445 1.1 ad small_max = (1 << opt_small_max_2pow);
3446 1.1 ad
3447 1.1 ad /* Set bin-related variables. */
3448 1.1 ad bin_maxclass = (pagesize >> 1);
3449 1.1 ad assert(opt_quantum_2pow >= TINY_MIN_2POW);
3450 1.1 ad ntbins = opt_quantum_2pow - TINY_MIN_2POW;
3451 1.1 ad assert(ntbins <= opt_quantum_2pow);
3452 1.1 ad nqbins = (small_max >> opt_quantum_2pow);
3453 1.1 ad nsbins = pagesize_2pow - opt_small_max_2pow - 1;
3454 1.1 ad
3455 1.1 ad /* Set variables according to the value of opt_quantum_2pow. */
3456 1.1 ad quantum = (1 << opt_quantum_2pow);
3457 1.1 ad quantum_mask = quantum - 1;
3458 1.1 ad if (ntbins > 0)
3459 1.1 ad small_min = (quantum >> 1) + 1;
3460 1.1 ad else
3461 1.1 ad small_min = 1;
3462 1.1 ad assert(small_min <= quantum);
3463 1.1 ad
3464 1.1 ad /* Set variables according to the value of opt_chunk_2pow. */
3465 1.1 ad chunksize = (1LU << opt_chunk_2pow);
3466 1.1 ad chunksize_mask = chunksize - 1;
3467 1.1 ad chunk_npages = (chunksize >> pagesize_2pow);
3468 1.1 ad {
3469 1.1 ad unsigned header_size;
3470 1.1 ad
3471 1.1 ad header_size = sizeof(arena_chunk_t) + (sizeof(arena_chunk_map_t)
3472 1.1 ad * (chunk_npages - 1));
3473 1.1 ad arena_chunk_header_npages = (header_size >> pagesize_2pow);
3474 1.1 ad if ((header_size & pagesize_mask) != 0)
3475 1.1 ad arena_chunk_header_npages++;
3476 1.1 ad }
3477 1.1 ad arena_maxclass = chunksize - (arena_chunk_header_npages <<
3478 1.1 ad pagesize_2pow);
3479 1.1 ad
3480 1.1 ad UTRACE(0, 0, 0);
3481 1.1 ad
3482 1.1 ad #ifdef MALLOC_STATS
3483 1.1 ad memset(&stats_chunks, 0, sizeof(chunk_stats_t));
3484 1.1 ad #endif
3485 1.1 ad
3486 1.1 ad /* Various sanity checks that regard configuration. */
3487 1.1 ad assert(quantum >= sizeof(void *));
3488 1.1 ad assert(quantum <= pagesize);
3489 1.1 ad assert(chunksize >= pagesize);
3490 1.1 ad assert(quantum * 4 <= chunksize);
3491 1.1 ad
3492 1.1 ad /* Initialize chunks data. */
3493 1.1 ad malloc_mutex_init(&chunks_mtx);
3494 1.1 ad RB_INIT(&huge);
3495 1.1 ad #ifdef USE_BRK
3496 1.1 ad malloc_mutex_init(&brk_mtx);
3497 1.1 ad brk_base = sbrk(0);
3498 1.1 ad brk_prev = brk_base;
3499 1.1 ad brk_max = brk_base;
3500 1.1 ad #endif
3501 1.1 ad #ifdef MALLOC_STATS
3502 1.1 ad huge_nmalloc = 0;
3503 1.1 ad huge_ndalloc = 0;
3504 1.1 ad huge_allocated = 0;
3505 1.1 ad #endif
3506 1.1 ad RB_INIT(&old_chunks);
3507 1.1 ad
3508 1.1 ad /* Initialize base allocation data structures. */
3509 1.1 ad #ifdef MALLOC_STATS
3510 1.1 ad base_mapped = 0;
3511 1.1 ad #endif
3512 1.1 ad #ifdef USE_BRK
3513 1.1 ad /*
3514 1.1 ad * Allocate a base chunk here, since it doesn't actually have to be
3515 1.1 ad * chunk-aligned. Doing this before allocating any other chunks allows
3516 1.1 ad * the use of space that would otherwise be wasted.
3517 1.1 ad */
3518 1.1 ad base_pages_alloc(0);
3519 1.1 ad #endif
3520 1.1 ad base_chunk_nodes = NULL;
3521 1.1 ad malloc_mutex_init(&base_mtx);
3522 1.1 ad
3523 1.1 ad if (ncpus > 1) {
3524 1.1 ad /*
3525 1.1 ad * For SMP systems, create four times as many arenas as there
3526 1.1 ad * are CPUs by default.
3527 1.1 ad */
3528 1.1 ad opt_narenas_lshift += 2;
3529 1.1 ad }
3530 1.1 ad
3531 1.2 ad #ifdef NO_TLS
3532 1.2 ad /* Initialize arena key. */
3533 1.2 ad (void)thr_keycreate(&arenas_map_key, NULL);
3534 1.2 ad #endif
3535 1.2 ad
3536 1.1 ad /* Determine how many arenas to use. */
3537 1.1 ad narenas = ncpus;
3538 1.1 ad if (opt_narenas_lshift > 0) {
3539 1.1 ad if ((narenas << opt_narenas_lshift) > narenas)
3540 1.1 ad narenas <<= opt_narenas_lshift;
3541 1.1 ad /*
3542 1.1 ad * Make sure not to exceed the limits of what base_malloc()
3543 1.1 ad * can handle.
3544 1.1 ad */
3545 1.1 ad if (narenas * sizeof(arena_t *) > chunksize)
3546 1.1 ad narenas = chunksize / sizeof(arena_t *);
3547 1.1 ad } else if (opt_narenas_lshift < 0) {
3548 1.1 ad if ((narenas << opt_narenas_lshift) < narenas)
3549 1.1 ad narenas <<= opt_narenas_lshift;
3550 1.1 ad /* Make sure there is at least one arena. */
3551 1.1 ad if (narenas == 0)
3552 1.1 ad narenas = 1;
3553 1.1 ad }
3554 1.1 ad
3555 1.1 ad next_arena = 0;
3556 1.1 ad
3557 1.1 ad /* Allocate and initialize arenas. */
3558 1.1 ad arenas = (arena_t **)base_alloc(sizeof(arena_t *) * narenas);
3559 1.1 ad if (arenas == NULL) {
3560 1.1 ad malloc_mutex_unlock(&init_lock);
3561 1.1 ad return (true);
3562 1.1 ad }
3563 1.1 ad /*
3564 1.1 ad * Zero the array. In practice, this should always be pre-zeroed,
3565 1.1 ad * since it was just mmap()ed, but let's be sure.
3566 1.1 ad */
3567 1.1 ad memset(arenas, 0, sizeof(arena_t *) * narenas);
3568 1.1 ad
3569 1.1 ad /*
3570 1.1 ad * Initialize one arena here. The rest are lazily created in
3571 1.1 ad * arena_choose_hard().
3572 1.1 ad */
3573 1.1 ad arenas_extend(0);
3574 1.1 ad if (arenas[0] == NULL) {
3575 1.1 ad malloc_mutex_unlock(&init_lock);
3576 1.1 ad return (true);
3577 1.1 ad }
3578 1.1 ad
3579 1.1 ad malloc_mutex_init(&arenas_mtx);
3580 1.1 ad
3581 1.1 ad malloc_initialized = true;
3582 1.1 ad malloc_mutex_unlock(&init_lock);
3583 1.1 ad return (false);
3584 1.1 ad }
3585 1.1 ad
3586 1.1 ad /*
3587 1.1 ad * End general internal functions.
3588 1.1 ad */
3589 1.1 ad /******************************************************************************/
3590 1.1 ad /*
3591 1.1 ad * Begin malloc(3)-compatible functions.
3592 1.1 ad */
3593 1.1 ad
3594 1.1 ad void *
3595 1.1 ad malloc(size_t size)
3596 1.1 ad {
3597 1.1 ad void *ret;
3598 1.1 ad
3599 1.1 ad if (malloc_init()) {
3600 1.1 ad ret = NULL;
3601 1.1 ad goto RETURN;
3602 1.1 ad }
3603 1.1 ad
3604 1.1 ad if (size == 0) {
3605 1.1 ad if (opt_sysv == false)
3606 1.1 ad size = 1;
3607 1.1 ad else {
3608 1.1 ad ret = NULL;
3609 1.1 ad goto RETURN;
3610 1.1 ad }
3611 1.1 ad }
3612 1.1 ad
3613 1.1 ad ret = imalloc(size);
3614 1.1 ad
3615 1.1 ad RETURN:
3616 1.1 ad if (ret == NULL) {
3617 1.1 ad if (opt_xmalloc) {
3618 1.1 ad _malloc_message(_getprogname(),
3619 1.1 ad ": (malloc) Error in malloc(): out of memory\n", "",
3620 1.1 ad "");
3621 1.1 ad abort();
3622 1.1 ad }
3623 1.1 ad errno = ENOMEM;
3624 1.1 ad }
3625 1.1 ad
3626 1.1 ad UTRACE(0, size, ret);
3627 1.1 ad return (ret);
3628 1.1 ad }
3629 1.1 ad
3630 1.2 ad /* XXXAD */
3631 1.2 ad int posix_memalign(void **memptr, size_t alignment, size_t size);
3632 1.2 ad
3633 1.1 ad int
3634 1.1 ad posix_memalign(void **memptr, size_t alignment, size_t size)
3635 1.1 ad {
3636 1.1 ad int ret;
3637 1.1 ad void *result;
3638 1.1 ad
3639 1.1 ad if (malloc_init())
3640 1.1 ad result = NULL;
3641 1.1 ad else {
3642 1.1 ad /* Make sure that alignment is a large enough power of 2. */
3643 1.1 ad if (((alignment - 1) & alignment) != 0
3644 1.1 ad || alignment < sizeof(void *)) {
3645 1.1 ad if (opt_xmalloc) {
3646 1.1 ad _malloc_message(_getprogname(),
3647 1.1 ad ": (malloc) Error in posix_memalign(): "
3648 1.1 ad "invalid alignment\n", "", "");
3649 1.1 ad abort();
3650 1.1 ad }
3651 1.1 ad result = NULL;
3652 1.1 ad ret = EINVAL;
3653 1.1 ad goto RETURN;
3654 1.1 ad }
3655 1.1 ad
3656 1.1 ad result = ipalloc(alignment, size);
3657 1.1 ad }
3658 1.1 ad
3659 1.1 ad if (result == NULL) {
3660 1.1 ad if (opt_xmalloc) {
3661 1.1 ad _malloc_message(_getprogname(),
3662 1.1 ad ": (malloc) Error in posix_memalign(): out of memory\n",
3663 1.1 ad "", "");
3664 1.1 ad abort();
3665 1.1 ad }
3666 1.1 ad ret = ENOMEM;
3667 1.1 ad goto RETURN;
3668 1.1 ad }
3669 1.1 ad
3670 1.1 ad *memptr = result;
3671 1.1 ad ret = 0;
3672 1.1 ad
3673 1.1 ad RETURN:
3674 1.1 ad UTRACE(0, size, result);
3675 1.1 ad return (ret);
3676 1.1 ad }
3677 1.1 ad
3678 1.1 ad void *
3679 1.1 ad calloc(size_t num, size_t size)
3680 1.1 ad {
3681 1.1 ad void *ret;
3682 1.1 ad size_t num_size;
3683 1.1 ad
3684 1.1 ad if (malloc_init()) {
3685 1.1 ad num_size = 0;
3686 1.1 ad ret = NULL;
3687 1.1 ad goto RETURN;
3688 1.1 ad }
3689 1.1 ad
3690 1.1 ad num_size = num * size;
3691 1.1 ad if (num_size == 0) {
3692 1.1 ad if ((opt_sysv == false) && ((num == 0) || (size == 0)))
3693 1.1 ad num_size = 1;
3694 1.1 ad else {
3695 1.1 ad ret = NULL;
3696 1.1 ad goto RETURN;
3697 1.1 ad }
3698 1.1 ad /*
3699 1.1 ad * Try to avoid division here. We know that it isn't possible to
3700 1.1 ad * overflow during multiplication if neither operand uses any of the
3701 1.1 ad * most significant half of the bits in a size_t.
3702 1.1 ad */
3703 1.2 ad } else if ((unsigned long long)((num | size) &
3704 1.2 ad ((unsigned long long)SIZE_T_MAX << (sizeof(size_t) << 2))) &&
3705 1.2 ad (num_size / size != num)) {
3706 1.1 ad /* size_t overflow. */
3707 1.1 ad ret = NULL;
3708 1.1 ad goto RETURN;
3709 1.1 ad }
3710 1.1 ad
3711 1.1 ad ret = icalloc(num_size);
3712 1.1 ad
3713 1.1 ad RETURN:
3714 1.1 ad if (ret == NULL) {
3715 1.1 ad if (opt_xmalloc) {
3716 1.1 ad _malloc_message(_getprogname(),
3717 1.1 ad ": (malloc) Error in calloc(): out of memory\n", "",
3718 1.1 ad "");
3719 1.1 ad abort();
3720 1.1 ad }
3721 1.1 ad errno = ENOMEM;
3722 1.1 ad }
3723 1.1 ad
3724 1.1 ad UTRACE(0, num_size, ret);
3725 1.1 ad return (ret);
3726 1.1 ad }
3727 1.1 ad
3728 1.1 ad void *
3729 1.1 ad realloc(void *ptr, size_t size)
3730 1.1 ad {
3731 1.1 ad void *ret;
3732 1.1 ad
3733 1.1 ad if (size == 0) {
3734 1.1 ad if (opt_sysv == false)
3735 1.1 ad size = 1;
3736 1.1 ad else {
3737 1.1 ad if (ptr != NULL)
3738 1.1 ad idalloc(ptr);
3739 1.1 ad ret = NULL;
3740 1.1 ad goto RETURN;
3741 1.1 ad }
3742 1.1 ad }
3743 1.1 ad
3744 1.1 ad if (ptr != NULL) {
3745 1.1 ad assert(malloc_initialized);
3746 1.1 ad
3747 1.1 ad ret = iralloc(ptr, size);
3748 1.1 ad
3749 1.1 ad if (ret == NULL) {
3750 1.1 ad if (opt_xmalloc) {
3751 1.1 ad _malloc_message(_getprogname(),
3752 1.1 ad ": (malloc) Error in realloc(): out of "
3753 1.1 ad "memory\n", "", "");
3754 1.1 ad abort();
3755 1.1 ad }
3756 1.1 ad errno = ENOMEM;
3757 1.1 ad }
3758 1.1 ad } else {
3759 1.1 ad if (malloc_init())
3760 1.1 ad ret = NULL;
3761 1.1 ad else
3762 1.1 ad ret = imalloc(size);
3763 1.1 ad
3764 1.1 ad if (ret == NULL) {
3765 1.1 ad if (opt_xmalloc) {
3766 1.1 ad _malloc_message(_getprogname(),
3767 1.1 ad ": (malloc) Error in realloc(): out of "
3768 1.1 ad "memory\n", "", "");
3769 1.1 ad abort();
3770 1.1 ad }
3771 1.1 ad errno = ENOMEM;
3772 1.1 ad }
3773 1.1 ad }
3774 1.1 ad
3775 1.1 ad RETURN:
3776 1.1 ad UTRACE(ptr, size, ret);
3777 1.1 ad return (ret);
3778 1.1 ad }
3779 1.1 ad
3780 1.1 ad void
3781 1.1 ad free(void *ptr)
3782 1.1 ad {
3783 1.1 ad
3784 1.1 ad UTRACE(ptr, 0, 0);
3785 1.1 ad if (ptr != NULL) {
3786 1.1 ad assert(malloc_initialized);
3787 1.1 ad
3788 1.1 ad idalloc(ptr);
3789 1.1 ad }
3790 1.1 ad }
3791 1.1 ad
3792 1.1 ad /*
3793 1.1 ad * End malloc(3)-compatible functions.
3794 1.1 ad */
3795 1.1 ad /******************************************************************************/
3796 1.1 ad /*
3797 1.1 ad * Begin non-standard functions.
3798 1.1 ad */
3799 1.2 ad #ifndef __NetBSD__
3800 1.1 ad size_t
3801 1.1 ad malloc_usable_size(const void *ptr)
3802 1.1 ad {
3803 1.1 ad
3804 1.1 ad assert(ptr != NULL);
3805 1.1 ad
3806 1.1 ad return (isalloc(ptr));
3807 1.1 ad }
3808 1.2 ad #endif
3809 1.1 ad
3810 1.1 ad /*
3811 1.1 ad * End non-standard functions.
3812 1.1 ad */
3813 1.1 ad /******************************************************************************/
3814 1.1 ad /*
3815 1.1 ad * Begin library-private functions, used by threading libraries for protection
3816 1.1 ad * of malloc during fork(). These functions are only called if the program is
3817 1.1 ad * running in threaded mode, so there is no need to check whether the program
3818 1.1 ad * is threaded here.
3819 1.1 ad */
3820 1.1 ad
3821 1.1 ad void
3822 1.1 ad _malloc_prefork(void)
3823 1.1 ad {
3824 1.1 ad unsigned i;
3825 1.1 ad
3826 1.1 ad /* Acquire all mutexes in a safe order. */
3827 1.1 ad
3828 1.1 ad malloc_mutex_lock(&arenas_mtx);
3829 1.1 ad for (i = 0; i < narenas; i++) {
3830 1.1 ad if (arenas[i] != NULL)
3831 1.1 ad malloc_mutex_lock(&arenas[i]->mtx);
3832 1.1 ad }
3833 1.1 ad malloc_mutex_unlock(&arenas_mtx);
3834 1.1 ad
3835 1.1 ad malloc_mutex_lock(&base_mtx);
3836 1.1 ad
3837 1.1 ad malloc_mutex_lock(&chunks_mtx);
3838 1.1 ad }
3839 1.1 ad
3840 1.1 ad void
3841 1.1 ad _malloc_postfork(void)
3842 1.1 ad {
3843 1.1 ad unsigned i;
3844 1.1 ad
3845 1.1 ad /* Release all mutexes, now that fork() has completed. */
3846 1.1 ad
3847 1.1 ad malloc_mutex_unlock(&chunks_mtx);
3848 1.1 ad
3849 1.1 ad malloc_mutex_unlock(&base_mtx);
3850 1.1 ad
3851 1.1 ad malloc_mutex_lock(&arenas_mtx);
3852 1.1 ad for (i = 0; i < narenas; i++) {
3853 1.1 ad if (arenas[i] != NULL)
3854 1.1 ad malloc_mutex_unlock(&arenas[i]->mtx);
3855 1.1 ad }
3856 1.1 ad malloc_mutex_unlock(&arenas_mtx);
3857 1.1 ad }
3858 1.1 ad
3859 1.1 ad /*
3860 1.1 ad * End library-private functions.
3861 1.1 ad */
3862 1.1 ad /******************************************************************************/
3863