]>
Commit | Line | Data |
---|---|---|
f65fd747 UD |
1 | /* Malloc implementation for multiple threads without lock contention. |
2 | Copyright (C) 1996 Free Software Foundation, Inc. | |
3 | This file is part of the GNU C Library. | |
4 | Contributed by Wolfram Gloger <wmglo@dent.med.uni-muenchen.de>, 1996. | |
5 | ||
6 | The GNU C Library is free software; you can redistribute it and/or | |
7 | modify it under the terms of the GNU Library General Public License as | |
8 | published by the Free Software Foundation; either version 2 of the | |
9 | License, or (at your option) any later version. | |
10 | ||
11 | The GNU C Library is distributed in the hope that it will be useful, | |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | Library General Public License for more details. | |
15 | ||
16 | You should have received a copy of the GNU Library General Public | |
17 | License along with the GNU C Library; see the file COPYING.LIB. If not, | |
18 | write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
19 | Boston, MA 02111-1307, USA. */ | |
20 | ||
21 | /* VERSION 2.6.4-pt Wed Dec 4 00:35:54 MET 1996 | |
22 | ||
23 | This work is mainly derived from malloc-2.6.4 by Doug Lea | |
24 | <dl@cs.oswego.edu>, which is available from: | |
25 | ||
26 | ftp://g.oswego.edu/pub/misc/malloc.c | |
27 | ||
28 | Most of the original comments are reproduced in the code below. | |
29 | ||
30 | * Why use this malloc? | |
31 | ||
32 | This is not the fastest, most space-conserving, most portable, or | |
33 | most tunable malloc ever written. However it is among the fastest | |
34 | while also being among the most space-conserving, portable and tunable. | |
35 | Consistent balance across these factors results in a good general-purpose | |
36 | allocator. For a high-level description, see | |
37 | http://g.oswego.edu/dl/html/malloc.html | |
38 | ||
39 | On many systems, the standard malloc implementation is by itself not | |
40 | thread-safe, and therefore wrapped with a single global lock around | |
41 | all malloc-related functions. In some applications, especially with | |
42 | multiple available processors, this can lead to contention problems | |
43 | and bad performance. This malloc version was designed with the goal | |
44 | to avoid waiting for locks as much as possible. Statistics indicate | |
45 | that this goal is achieved in many cases. | |
46 | ||
47 | * Synopsis of public routines | |
48 | ||
49 | (Much fuller descriptions are contained in the program documentation below.) | |
50 | ||
51 | ptmalloc_init(); | |
52 | Initialize global configuration. When compiled for multiple threads, | |
53 | this function must be called once before any other function in the | |
54 | package. It is not required otherwise. It is called automatically | |
55 | in the Linux/GNU C libray. | |
56 | malloc(size_t n); | |
57 | Return a pointer to a newly allocated chunk of at least n bytes, or null | |
58 | if no space is available. | |
59 | free(Void_t* p); | |
60 | Release the chunk of memory pointed to by p, or no effect if p is null. | |
61 | realloc(Void_t* p, size_t n); | |
62 | Return a pointer to a chunk of size n that contains the same data | |
63 | as does chunk p up to the minimum of (n, p's size) bytes, or null | |
64 | if no space is available. The returned pointer may or may not be | |
65 | the same as p. If p is null, equivalent to malloc. Unless the | |
66 | #define REALLOC_ZERO_BYTES_FREES below is set, realloc with a | |
67 | size argument of zero (re)allocates a minimum-sized chunk. | |
68 | memalign(size_t alignment, size_t n); | |
69 | Return a pointer to a newly allocated chunk of n bytes, aligned | |
70 | in accord with the alignment argument, which must be a power of | |
71 | two. | |
72 | valloc(size_t n); | |
73 | Equivalent to memalign(pagesize, n), where pagesize is the page | |
74 | size of the system (or as near to this as can be figured out from | |
75 | all the includes/defines below.) | |
76 | pvalloc(size_t n); | |
77 | Equivalent to valloc(minimum-page-that-holds(n)), that is, | |
78 | round up n to nearest pagesize. | |
79 | calloc(size_t unit, size_t quantity); | |
80 | Returns a pointer to quantity * unit bytes, with all locations | |
81 | set to zero. | |
82 | cfree(Void_t* p); | |
83 | Equivalent to free(p). | |
84 | malloc_trim(size_t pad); | |
85 | Release all but pad bytes of freed top-most memory back | |
86 | to the system. Return 1 if successful, else 0. | |
87 | malloc_usable_size(Void_t* p); | |
88 | Report the number usable allocated bytes associated with allocated | |
89 | chunk p. This may or may not report more bytes than were requested, | |
90 | due to alignment and minimum size constraints. | |
91 | malloc_stats(); | |
92 | Prints brief summary statistics on stderr. | |
93 | mallinfo() | |
94 | Returns (by copy) a struct containing various summary statistics. | |
95 | mallopt(int parameter_number, int parameter_value) | |
96 | Changes one of the tunable parameters described below. Returns | |
97 | 1 if successful in changing the parameter, else 0. | |
98 | ||
99 | * Vital statistics: | |
100 | ||
101 | Alignment: 8-byte | |
102 | 8 byte alignment is currently hardwired into the design. This | |
103 | seems to suffice for all current machines and C compilers. | |
104 | ||
105 | Assumed pointer representation: 4 or 8 bytes | |
106 | Code for 8-byte pointers is untested by me but has worked | |
107 | reliably by Wolfram Gloger, who contributed most of the | |
108 | changes supporting this. | |
109 | ||
110 | Assumed size_t representation: 4 or 8 bytes | |
111 | Note that size_t is allowed to be 4 bytes even if pointers are 8. | |
112 | ||
113 | Minimum overhead per allocated chunk: 4 or 8 bytes | |
114 | Each malloced chunk has a hidden overhead of 4 bytes holding size | |
115 | and status information. | |
116 | ||
117 | Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead) | |
118 | 8-byte ptrs: 24/32 bytes (including, 4/8 overhead) | |
119 | ||
120 | When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte | |
121 | ptrs but 4 byte size) or 24 (for 8/8) additional bytes are | |
122 | needed; 4 (8) for a trailing size field | |
123 | and 8 (16) bytes for free list pointers. Thus, the minimum | |
124 | allocatable size is 16/24/32 bytes. | |
125 | ||
126 | Even a request for zero bytes (i.e., malloc(0)) returns a | |
127 | pointer to something of the minimum allocatable size. | |
128 | ||
129 | Maximum allocated size: 4-byte size_t: 2^31 - 8 bytes | |
130 | 8-byte size_t: 2^63 - 16 bytes | |
131 | ||
132 | It is assumed that (possibly signed) size_t bit values suffice to | |
133 | represent chunk sizes. `Possibly signed' is due to the fact | |
134 | that `size_t' may be defined on a system as either a signed or | |
135 | an unsigned type. To be conservative, values that would appear | |
136 | as negative numbers are avoided. | |
137 | Requests for sizes with a negative sign bit will return a | |
138 | minimum-sized chunk. | |
139 | ||
140 | Maximum overhead wastage per allocated chunk: normally 15 bytes | |
141 | ||
142 | Alignnment demands, plus the minimum allocatable size restriction | |
143 | make the normal worst-case wastage 15 bytes (i.e., up to 15 | |
144 | more bytes will be allocated than were requested in malloc), with | |
145 | two exceptions: | |
146 | 1. Because requests for zero bytes allocate non-zero space, | |
147 | the worst case wastage for a request of zero bytes is 24 bytes. | |
148 | 2. For requests >= mmap_threshold that are serviced via | |
149 | mmap(), the worst case wastage is 8 bytes plus the remainder | |
150 | from a system page (the minimal mmap unit); typically 4096 bytes. | |
151 | ||
152 | * Limitations | |
153 | ||
154 | Here are some features that are NOT currently supported | |
155 | ||
156 | * No user-definable hooks for callbacks and the like. | |
157 | * No automated mechanism for fully checking that all accesses | |
158 | to malloced memory stay within their bounds. | |
159 | * No support for compaction. | |
160 | ||
161 | * Synopsis of compile-time options: | |
162 | ||
163 | People have reported using previous versions of this malloc on all | |
164 | versions of Unix, sometimes by tweaking some of the defines | |
165 | below. It has been tested most extensively on Solaris and | |
166 | Linux. People have also reported adapting this malloc for use in | |
167 | stand-alone embedded systems. | |
168 | ||
169 | The implementation is in straight, hand-tuned ANSI C. Among other | |
170 | consequences, it uses a lot of macros. Because of this, to be at | |
171 | all usable, this code should be compiled using an optimizing compiler | |
172 | (for example gcc -O2) that can simplify expressions and control | |
173 | paths. | |
174 | ||
175 | __STD_C (default: derived from C compiler defines) | |
176 | Nonzero if using ANSI-standard C compiler, a C++ compiler, or | |
177 | a C compiler sufficiently close to ANSI to get away with it. | |
178 | MALLOC_DEBUG (default: NOT defined) | |
179 | Define to enable debugging. Adds fairly extensive assertion-based | |
180 | checking to help track down memory errors, but noticeably slows down | |
181 | execution. | |
182 | REALLOC_ZERO_BYTES_FREES (default: NOT defined) | |
183 | Define this if you think that realloc(p, 0) should be equivalent | |
184 | to free(p). Otherwise, since malloc returns a unique pointer for | |
185 | malloc(0), so does realloc(p, 0). | |
186 | HAVE_MEMCPY (default: defined) | |
187 | Define if you are not otherwise using ANSI STD C, but still | |
188 | have memcpy and memset in your C library and want to use them. | |
189 | Otherwise, simple internal versions are supplied. | |
190 | USE_MEMCPY (default: 1 if HAVE_MEMCPY is defined, 0 otherwise) | |
191 | Define as 1 if you want the C library versions of memset and | |
192 | memcpy called in realloc and calloc (otherwise macro versions are used). | |
193 | At least on some platforms, the simple macro versions usually | |
194 | outperform libc versions. | |
195 | HAVE_MMAP (default: defined as 1) | |
196 | Define to non-zero to optionally make malloc() use mmap() to | |
197 | allocate very large blocks. | |
198 | HAVE_MREMAP (default: defined as 0 unless Linux libc set) | |
199 | Define to non-zero to optionally make realloc() use mremap() to | |
200 | reallocate very large blocks. | |
201 | malloc_getpagesize (default: derived from system #includes) | |
202 | Either a constant or routine call returning the system page size. | |
203 | HAVE_USR_INCLUDE_MALLOC_H (default: NOT defined) | |
204 | Optionally define if you are on a system with a /usr/include/malloc.h | |
205 | that declares struct mallinfo. It is not at all necessary to | |
206 | define this even if you do, but will ensure consistency. | |
207 | INTERNAL_SIZE_T (default: size_t) | |
208 | Define to a 32-bit type (probably `unsigned int') if you are on a | |
209 | 64-bit machine, yet do not want or need to allow malloc requests of | |
210 | greater than 2^31 to be handled. This saves space, especially for | |
211 | very small chunks. | |
212 | _LIBC (default: NOT defined) | |
213 | Defined only when compiled as part of the Linux libc/glibc. | |
214 | Also note that there is some odd internal name-mangling via defines | |
215 | (for example, internally, `malloc' is named `mALLOc') needed | |
216 | when compiling in this case. These look funny but don't otherwise | |
217 | affect anything. | |
218 | LACKS_UNISTD_H (default: undefined) | |
219 | Define this if your system does not have a <unistd.h>. | |
220 | MORECORE (default: sbrk) | |
221 | The name of the routine to call to obtain more memory from the system. | |
222 | MORECORE_FAILURE (default: -1) | |
223 | The value returned upon failure of MORECORE. | |
224 | MORECORE_CLEARS (default 1) | |
225 | True (1) if the routine mapped to MORECORE zeroes out memory (which | |
226 | holds for sbrk). | |
227 | DEFAULT_TRIM_THRESHOLD | |
228 | DEFAULT_TOP_PAD | |
229 | DEFAULT_MMAP_THRESHOLD | |
230 | DEFAULT_MMAP_MAX | |
231 | Default values of tunable parameters (described in detail below) | |
232 | controlling interaction with host system routines (sbrk, mmap, etc). | |
233 | These values may also be changed dynamically via mallopt(). The | |
234 | preset defaults are those that give best performance for typical | |
235 | programs/systems. | |
236 | ||
237 | ||
238 | */ | |
239 | ||
240 | /* | |
241 | ||
242 | * Compile-time options for multiple threads: | |
243 | ||
244 | USE_PTHREADS, USE_THR, USE_SPROC | |
245 | Define one of these as 1 to select the thread interface: | |
246 | POSIX threads, Solaris threads or SGI sproc's, respectively. | |
247 | If none of these is defined as non-zero, you get a `normal' | |
248 | malloc implementation which is not thread-safe. Support for | |
249 | multiple threads requires HAVE_MMAP=1. As an exception, when | |
250 | compiling for GNU libc, i.e. when _LIBC is defined, then none of | |
251 | the USE_... symbols have to be defined. | |
252 | ||
253 | HEAP_MIN_SIZE | |
254 | HEAP_MAX_SIZE | |
255 | When thread support is enabled, additional `heap's are created | |
256 | with mmap calls. These are limited in size; HEAP_MIN_SIZE should | |
257 | be a multiple of the page size, while HEAP_MAX_SIZE must be a power | |
258 | of two for alignment reasons. HEAP_MAX_SIZE should be at least | |
259 | twice as large as the mmap threshold. | |
260 | THREAD_STATS | |
261 | When this is defined as non-zero, some statistics on mutex locking | |
262 | are computed. | |
263 | ||
264 | */ | |
265 | ||
266 | \f | |
267 | ||
268 | ||
269 | /* Macros for handling mutexes and thread-specific data. This is | |
270 | included first, because some thread-related header files (such as | |
271 | pthread.h) should be included before any others. */ | |
272 | #include "thread-m.h" | |
273 | ||
274 | ||
275 | /* Preliminaries */ | |
276 | ||
277 | #ifndef __STD_C | |
278 | #if defined (__STDC__) | |
279 | #define __STD_C 1 | |
280 | #else | |
281 | #if __cplusplus | |
282 | #define __STD_C 1 | |
283 | #else | |
284 | #define __STD_C 0 | |
285 | #endif /*__cplusplus*/ | |
286 | #endif /*__STDC__*/ | |
287 | #endif /*__STD_C*/ | |
288 | ||
289 | #ifndef Void_t | |
290 | #if __STD_C | |
291 | #define Void_t void | |
292 | #else | |
293 | #define Void_t char | |
294 | #endif | |
295 | #endif /*Void_t*/ | |
296 | ||
297 | #if __STD_C | |
298 | #include <stddef.h> /* for size_t */ | |
299 | #else | |
300 | #include <sys/types.h> | |
301 | #endif | |
302 | ||
303 | #ifdef __cplusplus | |
304 | extern "C" { | |
305 | #endif | |
306 | ||
307 | #include <stdio.h> /* needed for malloc_stats */ | |
308 | ||
309 | ||
310 | /* | |
311 | Compile-time options | |
312 | */ | |
313 | ||
314 | ||
315 | /* | |
316 | Debugging: | |
317 | ||
318 | Because freed chunks may be overwritten with link fields, this | |
319 | malloc will often die when freed memory is overwritten by user | |
320 | programs. This can be very effective (albeit in an annoying way) | |
321 | in helping track down dangling pointers. | |
322 | ||
323 | If you compile with -DMALLOC_DEBUG, a number of assertion checks are | |
324 | enabled that will catch more memory errors. You probably won't be | |
325 | able to make much sense of the actual assertion errors, but they | |
326 | should help you locate incorrectly overwritten memory. The | |
327 | checking is fairly extensive, and will slow down execution | |
328 | noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set will | |
329 | attempt to check every non-mmapped allocated and free chunk in the | |
330 | course of computing the summmaries. (By nature, mmapped regions | |
331 | cannot be checked very much automatically.) | |
332 | ||
333 | Setting MALLOC_DEBUG may also be helpful if you are trying to modify | |
334 | this code. The assertions in the check routines spell out in more | |
335 | detail the assumptions and invariants underlying the algorithms. | |
336 | ||
337 | */ | |
338 | ||
339 | #if MALLOC_DEBUG | |
340 | #include <assert.h> | |
341 | #else | |
342 | #define assert(x) ((void)0) | |
343 | #endif | |
344 | ||
345 | ||
346 | /* | |
347 | INTERNAL_SIZE_T is the word-size used for internal bookkeeping | |
348 | of chunk sizes. On a 64-bit machine, you can reduce malloc | |
349 | overhead by defining INTERNAL_SIZE_T to be a 32 bit `unsigned int' | |
350 | at the expense of not being able to handle requests greater than | |
351 | 2^31. This limitation is hardly ever a concern; you are encouraged | |
352 | to set this. However, the default version is the same as size_t. | |
353 | */ | |
354 | ||
355 | #ifndef INTERNAL_SIZE_T | |
356 | #define INTERNAL_SIZE_T size_t | |
357 | #endif | |
358 | ||
359 | /* | |
360 | REALLOC_ZERO_BYTES_FREES should be set if a call to | |
361 | realloc with zero bytes should be the same as a call to free. | |
362 | Some people think it should. Otherwise, since this malloc | |
363 | returns a unique pointer for malloc(0), so does realloc(p, 0). | |
364 | */ | |
365 | ||
366 | ||
367 | /* #define REALLOC_ZERO_BYTES_FREES */ | |
368 | ||
369 | ||
370 | /* | |
371 | HAVE_MEMCPY should be defined if you are not otherwise using | |
372 | ANSI STD C, but still have memcpy and memset in your C library | |
373 | and want to use them in calloc and realloc. Otherwise simple | |
374 | macro versions are defined here. | |
375 | ||
376 | USE_MEMCPY should be defined as 1 if you actually want to | |
377 | have memset and memcpy called. People report that the macro | |
378 | versions are often enough faster than libc versions on many | |
379 | systems that it is better to use them. | |
380 | ||
381 | */ | |
382 | ||
383 | #define HAVE_MEMCPY | |
384 | ||
385 | #ifndef USE_MEMCPY | |
386 | #ifdef HAVE_MEMCPY | |
387 | #define USE_MEMCPY 1 | |
388 | #else | |
389 | #define USE_MEMCPY 0 | |
390 | #endif | |
391 | #endif | |
392 | ||
393 | #if (__STD_C || defined(HAVE_MEMCPY)) | |
394 | ||
395 | #if __STD_C | |
396 | void* memset(void*, int, size_t); | |
397 | void* memcpy(void*, const void*, size_t); | |
398 | #else | |
399 | Void_t* memset(); | |
400 | Void_t* memcpy(); | |
401 | #endif | |
402 | #endif | |
403 | ||
404 | #if USE_MEMCPY | |
405 | ||
406 | /* The following macros are only invoked with (2n+1)-multiples of | |
407 | INTERNAL_SIZE_T units, with a positive integer n. This is exploited | |
408 | for fast inline execution when n is small. */ | |
409 | ||
410 | #define MALLOC_ZERO(charp, nbytes) \ | |
411 | do { \ | |
412 | INTERNAL_SIZE_T mzsz = (nbytes); \ | |
413 | if(mzsz <= 9*sizeof(mzsz)) { \ | |
414 | INTERNAL_SIZE_T* mz = (INTERNAL_SIZE_T*) (charp); \ | |
415 | if(mzsz >= 5*sizeof(mzsz)) { *mz++ = 0; \ | |
416 | *mz++ = 0; \ | |
417 | if(mzsz >= 7*sizeof(mzsz)) { *mz++ = 0; \ | |
418 | *mz++ = 0; \ | |
419 | if(mzsz >= 9*sizeof(mzsz)) { *mz++ = 0; \ | |
420 | *mz++ = 0; }}} \ | |
421 | *mz++ = 0; \ | |
422 | *mz++ = 0; \ | |
423 | *mz = 0; \ | |
424 | } else memset((charp), 0, mzsz); \ | |
425 | } while(0) | |
426 | ||
427 | #define MALLOC_COPY(dest,src,nbytes) \ | |
428 | do { \ | |
429 | INTERNAL_SIZE_T mcsz = (nbytes); \ | |
430 | if(mcsz <= 9*sizeof(mcsz)) { \ | |
431 | INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) (src); \ | |
432 | INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) (dest); \ | |
433 | if(mcsz >= 5*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \ | |
434 | *mcdst++ = *mcsrc++; \ | |
435 | if(mcsz >= 7*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \ | |
436 | *mcdst++ = *mcsrc++; \ | |
437 | if(mcsz >= 9*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \ | |
438 | *mcdst++ = *mcsrc++; }}} \ | |
439 | *mcdst++ = *mcsrc++; \ | |
440 | *mcdst++ = *mcsrc++; \ | |
441 | *mcdst = *mcsrc ; \ | |
442 | } else memcpy(dest, src, mcsz); \ | |
443 | } while(0) | |
444 | ||
445 | #else /* !USE_MEMCPY */ | |
446 | ||
447 | /* Use Duff's device for good zeroing/copying performance. */ | |
448 | ||
449 | #define MALLOC_ZERO(charp, nbytes) \ | |
450 | do { \ | |
451 | INTERNAL_SIZE_T* mzp = (INTERNAL_SIZE_T*)(charp); \ | |
452 | long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn; \ | |
453 | if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \ | |
454 | switch (mctmp) { \ | |
455 | case 0: for(;;) { *mzp++ = 0; \ | |
456 | case 7: *mzp++ = 0; \ | |
457 | case 6: *mzp++ = 0; \ | |
458 | case 5: *mzp++ = 0; \ | |
459 | case 4: *mzp++ = 0; \ | |
460 | case 3: *mzp++ = 0; \ | |
461 | case 2: *mzp++ = 0; \ | |
462 | case 1: *mzp++ = 0; if(mcn <= 0) break; mcn--; } \ | |
463 | } \ | |
464 | } while(0) | |
465 | ||
466 | #define MALLOC_COPY(dest,src,nbytes) \ | |
467 | do { \ | |
468 | INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) src; \ | |
469 | INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) dest; \ | |
470 | long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn; \ | |
471 | if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \ | |
472 | switch (mctmp) { \ | |
473 | case 0: for(;;) { *mcdst++ = *mcsrc++; \ | |
474 | case 7: *mcdst++ = *mcsrc++; \ | |
475 | case 6: *mcdst++ = *mcsrc++; \ | |
476 | case 5: *mcdst++ = *mcsrc++; \ | |
477 | case 4: *mcdst++ = *mcsrc++; \ | |
478 | case 3: *mcdst++ = *mcsrc++; \ | |
479 | case 2: *mcdst++ = *mcsrc++; \ | |
480 | case 1: *mcdst++ = *mcsrc++; if(mcn <= 0) break; mcn--; } \ | |
481 | } \ | |
482 | } while(0) | |
483 | ||
484 | #endif | |
485 | ||
486 | ||
487 | /* | |
488 | Define HAVE_MMAP to optionally make malloc() use mmap() to | |
489 | allocate very large blocks. These will be returned to the | |
490 | operating system immediately after a free(). | |
491 | */ | |
492 | ||
493 | #ifndef HAVE_MMAP | |
494 | #define HAVE_MMAP 1 | |
495 | #endif | |
496 | ||
497 | /* | |
498 | Define HAVE_MREMAP to make realloc() use mremap() to re-allocate | |
499 | large blocks. This is currently only possible on Linux with | |
500 | kernel versions newer than 1.3.77. | |
501 | */ | |
502 | ||
503 | #ifndef HAVE_MREMAP | |
504 | #define HAVE_MREMAP defined(__linux__) | |
505 | #endif | |
506 | ||
507 | #if HAVE_MMAP | |
508 | ||
509 | #include <unistd.h> | |
510 | #include <fcntl.h> | |
511 | #include <sys/mman.h> | |
512 | ||
513 | #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON) | |
514 | #define MAP_ANONYMOUS MAP_ANON | |
515 | #endif | |
516 | ||
517 | #endif /* HAVE_MMAP */ | |
518 | ||
519 | /* | |
520 | Access to system page size. To the extent possible, this malloc | |
521 | manages memory from the system in page-size units. | |
522 | ||
523 | The following mechanics for getpagesize were adapted from | |
524 | bsd/gnu getpagesize.h | |
525 | */ | |
526 | ||
527 | #ifndef LACKS_UNISTD_H | |
528 | # include <unistd.h> | |
529 | #endif | |
530 | ||
531 | #ifndef malloc_getpagesize | |
532 | # ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */ | |
533 | # ifndef _SC_PAGE_SIZE | |
534 | # define _SC_PAGE_SIZE _SC_PAGESIZE | |
535 | # endif | |
536 | # endif | |
537 | # ifdef _SC_PAGE_SIZE | |
538 | # define malloc_getpagesize sysconf(_SC_PAGE_SIZE) | |
539 | # else | |
540 | # if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE) | |
541 | extern size_t getpagesize(); | |
542 | # define malloc_getpagesize getpagesize() | |
543 | # else | |
544 | # include <sys/param.h> | |
545 | # ifdef EXEC_PAGESIZE | |
546 | # define malloc_getpagesize EXEC_PAGESIZE | |
547 | # else | |
548 | # ifdef NBPG | |
549 | # ifndef CLSIZE | |
550 | # define malloc_getpagesize NBPG | |
551 | # else | |
552 | # define malloc_getpagesize (NBPG * CLSIZE) | |
553 | # endif | |
554 | # else | |
555 | # ifdef NBPC | |
556 | # define malloc_getpagesize NBPC | |
557 | # else | |
558 | # ifdef PAGESIZE | |
559 | # define malloc_getpagesize PAGESIZE | |
560 | # else | |
561 | # define malloc_getpagesize (4096) /* just guess */ | |
562 | # endif | |
563 | # endif | |
564 | # endif | |
565 | # endif | |
566 | # endif | |
567 | # endif | |
568 | #endif | |
569 | ||
570 | ||
571 | ||
572 | /* | |
573 | ||
574 | This version of malloc supports the standard SVID/XPG mallinfo | |
575 | routine that returns a struct containing the same kind of | |
576 | information you can get from malloc_stats. It should work on | |
577 | any SVID/XPG compliant system that has a /usr/include/malloc.h | |
578 | defining struct mallinfo. (If you'd like to install such a thing | |
579 | yourself, cut out the preliminary declarations as described above | |
580 | and below and save them in a malloc.h file. But there's no | |
581 | compelling reason to bother to do this.) | |
582 | ||
583 | The main declaration needed is the mallinfo struct that is returned | |
584 | (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a | |
585 | bunch of fields, most of which are not even meaningful in this | |
586 | version of malloc. Some of these fields are are instead filled by | |
587 | mallinfo() with other numbers that might possibly be of interest. | |
588 | ||
589 | HAVE_USR_INCLUDE_MALLOC_H should be set if you have a | |
590 | /usr/include/malloc.h file that includes a declaration of struct | |
591 | mallinfo. If so, it is included; else an SVID2/XPG2 compliant | |
592 | version is declared below. These must be precisely the same for | |
593 | mallinfo() to work. | |
594 | ||
595 | */ | |
596 | ||
597 | /* #define HAVE_USR_INCLUDE_MALLOC_H */ | |
598 | ||
599 | #if HAVE_USR_INCLUDE_MALLOC_H | |
600 | #include "/usr/include/malloc.h" | |
601 | #else | |
602 | #include "malloc.h" | |
603 | #endif | |
604 | ||
605 | ||
606 | ||
607 | #ifndef DEFAULT_TRIM_THRESHOLD | |
608 | #define DEFAULT_TRIM_THRESHOLD (128 * 1024) | |
609 | #endif | |
610 | ||
611 | /* | |
612 | M_TRIM_THRESHOLD is the maximum amount of unused top-most memory | |
613 | to keep before releasing via malloc_trim in free(). | |
614 | ||
615 | Automatic trimming is mainly useful in long-lived programs. | |
616 | Because trimming via sbrk can be slow on some systems, and can | |
617 | sometimes be wasteful (in cases where programs immediately | |
618 | afterward allocate more large chunks) the value should be high | |
619 | enough so that your overall system performance would improve by | |
620 | releasing. | |
621 | ||
622 | The trim threshold and the mmap control parameters (see below) | |
623 | can be traded off with one another. Trimming and mmapping are | |
624 | two different ways of releasing unused memory back to the | |
625 | system. Between these two, it is often possible to keep | |
626 | system-level demands of a long-lived program down to a bare | |
627 | minimum. For example, in one test suite of sessions measuring | |
628 | the XF86 X server on Linux, using a trim threshold of 128K and a | |
629 | mmap threshold of 192K led to near-minimal long term resource | |
630 | consumption. | |
631 | ||
632 | If you are using this malloc in a long-lived program, it should | |
633 | pay to experiment with these values. As a rough guide, you | |
634 | might set to a value close to the average size of a process | |
635 | (program) running on your system. Releasing this much memory | |
636 | would allow such a process to run in memory. Generally, it's | |
637 | worth it to tune for trimming rather tham memory mapping when a | |
638 | program undergoes phases where several large chunks are | |
639 | allocated and released in ways that can reuse each other's | |
640 | storage, perhaps mixed with phases where there are no such | |
641 | chunks at all. And in well-behaved long-lived programs, | |
642 | controlling release of large blocks via trimming versus mapping | |
643 | is usually faster. | |
644 | ||
645 | However, in most programs, these parameters serve mainly as | |
646 | protection against the system-level effects of carrying around | |
647 | massive amounts of unneeded memory. Since frequent calls to | |
648 | sbrk, mmap, and munmap otherwise degrade performance, the default | |
649 | parameters are set to relatively high values that serve only as | |
650 | safeguards. | |
651 | ||
652 | The default trim value is high enough to cause trimming only in | |
653 | fairly extreme (by current memory consumption standards) cases. | |
654 | It must be greater than page size to have any useful effect. To | |
655 | disable trimming completely, you can set to (unsigned long)(-1); | |
656 | ||
657 | ||
658 | */ | |
659 | ||
660 | ||
661 | #ifndef DEFAULT_TOP_PAD | |
662 | #define DEFAULT_TOP_PAD (0) | |
663 | #endif | |
664 | ||
665 | /* | |
666 | M_TOP_PAD is the amount of extra `padding' space to allocate or | |
667 | retain whenever sbrk is called. It is used in two ways internally: | |
668 | ||
669 | * When sbrk is called to extend the top of the arena to satisfy | |
670 | a new malloc request, this much padding is added to the sbrk | |
671 | request. | |
672 | ||
673 | * When malloc_trim is called automatically from free(), | |
674 | it is used as the `pad' argument. | |
675 | ||
676 | In both cases, the actual amount of padding is rounded | |
677 | so that the end of the arena is always a system page boundary. | |
678 | ||
679 | The main reason for using padding is to avoid calling sbrk so | |
680 | often. Having even a small pad greatly reduces the likelihood | |
681 | that nearly every malloc request during program start-up (or | |
682 | after trimming) will invoke sbrk, which needlessly wastes | |
683 | time. | |
684 | ||
685 | Automatic rounding-up to page-size units is normally sufficient | |
686 | to avoid measurable overhead, so the default is 0. However, in | |
687 | systems where sbrk is relatively slow, it can pay to increase | |
688 | this value, at the expense of carrying around more memory than | |
689 | the program needs. | |
690 | ||
691 | */ | |
692 | ||
693 | ||
694 | #ifndef DEFAULT_MMAP_THRESHOLD | |
695 | #define DEFAULT_MMAP_THRESHOLD (128 * 1024) | |
696 | #endif | |
697 | ||
698 | /* | |
699 | ||
700 | M_MMAP_THRESHOLD is the request size threshold for using mmap() | |
701 | to service a request. Requests of at least this size that cannot | |
702 | be allocated using already-existing space will be serviced via mmap. | |
703 | (If enough normal freed space already exists it is used instead.) | |
704 | ||
705 | Using mmap segregates relatively large chunks of memory so that | |
706 | they can be individually obtained and released from the host | |
707 | system. A request serviced through mmap is never reused by any | |
708 | other request (at least not directly; the system may just so | |
709 | happen to remap successive requests to the same locations). | |
710 | ||
711 | Segregating space in this way has the benefit that mmapped space | |
712 | can ALWAYS be individually released back to the system, which | |
713 | helps keep the system level memory demands of a long-lived | |
714 | program low. Mapped memory can never become `locked' between | |
715 | other chunks, as can happen with normally allocated chunks, which | |
716 | menas that even trimming via malloc_trim would not release them. | |
717 | ||
718 | However, it has the disadvantages that: | |
719 | ||
720 | 1. The space cannot be reclaimed, consolidated, and then | |
721 | used to service later requests, as happens with normal chunks. | |
722 | 2. It can lead to more wastage because of mmap page alignment | |
723 | requirements | |
724 | 3. It causes malloc performance to be more dependent on host | |
725 | system memory management support routines which may vary in | |
726 | implementation quality and may impose arbitrary | |
727 | limitations. Generally, servicing a request via normal | |
728 | malloc steps is faster than going through a system's mmap. | |
729 | ||
730 | All together, these considerations should lead you to use mmap | |
731 | only for relatively large requests. | |
732 | ||
733 | ||
734 | */ | |
735 | ||
736 | ||
737 | ||
738 | #ifndef DEFAULT_MMAP_MAX | |
739 | #if HAVE_MMAP | |
740 | #define DEFAULT_MMAP_MAX (1024) | |
741 | #else | |
742 | #define DEFAULT_MMAP_MAX (0) | |
743 | #endif | |
744 | #endif | |
745 | ||
746 | /* | |
747 | M_MMAP_MAX is the maximum number of requests to simultaneously | |
748 | service using mmap. This parameter exists because: | |
749 | ||
750 | 1. Some systems have a limited number of internal tables for | |
751 | use by mmap. | |
752 | 2. In most systems, overreliance on mmap can degrade overall | |
753 | performance. | |
754 | 3. If a program allocates many large regions, it is probably | |
755 | better off using normal sbrk-based allocation routines that | |
756 | can reclaim and reallocate normal heap memory. Using a | |
757 | small value allows transition into this mode after the | |
758 | first few allocations. | |
759 | ||
760 | Setting to 0 disables all use of mmap. If HAVE_MMAP is not set, | |
761 | the default value is 0, and attempts to set it to non-zero values | |
762 | in mallopt will fail. | |
763 | */ | |
764 | ||
765 | ||
766 | ||
767 | #define HEAP_MIN_SIZE (32*1024) | |
768 | #define HEAP_MAX_SIZE (1024*1024) /* must be a power of two */ | |
769 | ||
770 | /* HEAP_MIN_SIZE and HEAP_MAX_SIZE limit the size of mmap()ed heaps | |
771 | that are dynamically created for multi-threaded programs. The | |
772 | maximum size must be a power of two, for fast determination of | |
773 | which heap belongs to a chunk. It should be much larger than | |
774 | the mmap threshold, so that requests with a size just below that | |
775 | threshold can be fulfilled without creating too many heaps. | |
776 | */ | |
777 | ||
778 | ||
779 | ||
780 | #ifndef THREAD_STATS | |
781 | #define THREAD_STATS 0 | |
782 | #endif | |
783 | ||
784 | /* If THREAD_STATS is non-zero, some statistics on mutex locking are | |
785 | computed. */ | |
786 | ||
787 | ||
788 | /* | |
789 | ||
790 | Special defines for the Linux/GNU C library. | |
791 | ||
792 | */ | |
793 | ||
794 | ||
795 | #ifdef _LIBC | |
796 | ||
797 | #if __STD_C | |
798 | ||
799 | Void_t * __default_morecore (ptrdiff_t); | |
800 | static Void_t *(*__morecore)(ptrdiff_t) = __default_morecore; | |
801 | ||
802 | #else | |
803 | ||
804 | Void_t * __default_morecore (); | |
805 | static Void_t *(*__morecore)() = __default_morecore; | |
806 | ||
807 | #endif | |
808 | ||
809 | #define MORECORE (*__morecore) | |
810 | #define MORECORE_FAILURE 0 | |
811 | #define MORECORE_CLEARS 1 | |
812 | ||
813 | #else /* _LIBC */ | |
814 | ||
815 | #if __STD_C | |
816 | extern Void_t* sbrk(ptrdiff_t); | |
817 | #else | |
818 | extern Void_t* sbrk(); | |
819 | #endif | |
820 | ||
821 | #ifndef MORECORE | |
822 | #define MORECORE sbrk | |
823 | #endif | |
824 | ||
825 | #ifndef MORECORE_FAILURE | |
826 | #define MORECORE_FAILURE -1 | |
827 | #endif | |
828 | ||
829 | #ifndef MORECORE_CLEARS | |
830 | #define MORECORE_CLEARS 1 | |
831 | #endif | |
832 | ||
833 | #endif /* _LIBC */ | |
834 | ||
835 | #if 0 && defined(_LIBC) | |
836 | ||
837 | #define cALLOc __libc_calloc | |
838 | #define fREe __libc_free | |
839 | #define mALLOc __libc_malloc | |
840 | #define mEMALIGn __libc_memalign | |
841 | #define rEALLOc __libc_realloc | |
842 | #define vALLOc __libc_valloc | |
843 | #define pvALLOc __libc_pvalloc | |
844 | #define mALLINFo __libc_mallinfo | |
845 | #define mALLOPt __libc_mallopt | |
846 | ||
847 | #pragma weak calloc = __libc_calloc | |
848 | #pragma weak free = __libc_free | |
849 | #pragma weak cfree = __libc_free | |
850 | #pragma weak malloc = __libc_malloc | |
851 | #pragma weak memalign = __libc_memalign | |
852 | #pragma weak realloc = __libc_realloc | |
853 | #pragma weak valloc = __libc_valloc | |
854 | #pragma weak pvalloc = __libc_pvalloc | |
855 | #pragma weak mallinfo = __libc_mallinfo | |
856 | #pragma weak mallopt = __libc_mallopt | |
857 | ||
858 | #else | |
859 | ||
860 | #define cALLOc calloc | |
861 | #define fREe free | |
862 | #define mALLOc malloc | |
863 | #define mEMALIGn memalign | |
864 | #define rEALLOc realloc | |
865 | #define vALLOc valloc | |
866 | #define pvALLOc pvalloc | |
867 | #define mALLINFo mallinfo | |
868 | #define mALLOPt mallopt | |
869 | ||
870 | #endif | |
871 | ||
872 | /* Public routines */ | |
873 | ||
874 | #if __STD_C | |
875 | ||
876 | #ifndef _LIBC | |
877 | void ptmalloc_init(void); | |
878 | #endif | |
879 | Void_t* mALLOc(size_t); | |
880 | void fREe(Void_t*); | |
881 | Void_t* rEALLOc(Void_t*, size_t); | |
882 | Void_t* mEMALIGn(size_t, size_t); | |
883 | Void_t* vALLOc(size_t); | |
884 | Void_t* pvALLOc(size_t); | |
885 | Void_t* cALLOc(size_t, size_t); | |
886 | void cfree(Void_t*); | |
887 | int malloc_trim(size_t); | |
888 | size_t malloc_usable_size(Void_t*); | |
889 | void malloc_stats(void); | |
890 | int mALLOPt(int, int); | |
891 | struct mallinfo mALLINFo(void); | |
892 | #else | |
893 | #ifndef _LIBC | |
894 | void ptmalloc_init(); | |
895 | #endif | |
896 | Void_t* mALLOc(); | |
897 | void fREe(); | |
898 | Void_t* rEALLOc(); | |
899 | Void_t* mEMALIGn(); | |
900 | Void_t* vALLOc(); | |
901 | Void_t* pvALLOc(); | |
902 | Void_t* cALLOc(); | |
903 | void cfree(); | |
904 | int malloc_trim(); | |
905 | size_t malloc_usable_size(); | |
906 | void malloc_stats(); | |
907 | int mALLOPt(); | |
908 | struct mallinfo mALLINFo(); | |
909 | #endif | |
910 | ||
911 | ||
912 | #ifdef __cplusplus | |
913 | }; /* end of extern "C" */ | |
914 | #endif | |
915 | ||
916 | #if !defined(NO_THREADS) && !HAVE_MMAP | |
917 | "Can't have threads support without mmap" | |
918 | #endif | |
919 | ||
920 | ||
921 | /* | |
922 | Type declarations | |
923 | */ | |
924 | ||
925 | ||
926 | struct malloc_chunk | |
927 | { | |
928 | INTERNAL_SIZE_T prev_size; /* Size of previous chunk (if free). */ | |
929 | INTERNAL_SIZE_T size; /* Size in bytes, including overhead. */ | |
930 | struct malloc_chunk* fd; /* double links -- used only if free. */ | |
931 | struct malloc_chunk* bk; | |
932 | }; | |
933 | ||
934 | typedef struct malloc_chunk* mchunkptr; | |
935 | ||
936 | /* | |
937 | ||
938 | malloc_chunk details: | |
939 | ||
940 | (The following includes lightly edited explanations by Colin Plumb.) | |
941 | ||
942 | Chunks of memory are maintained using a `boundary tag' method as | |
943 | described in e.g., Knuth or Standish. (See the paper by Paul | |
944 | Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a | |
945 | survey of such techniques.) Sizes of free chunks are stored both | |
946 | in the front of each chunk and at the end. This makes | |
947 | consolidating fragmented chunks into bigger chunks very fast. The | |
948 | size fields also hold bits representing whether chunks are free or | |
949 | in use. | |
950 | ||
951 | An allocated chunk looks like this: | |
952 | ||
953 | ||
954 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
955 | | Size of previous chunk, if allocated | | | |
956 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
957 | | Size of chunk, in bytes |P| | |
958 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
959 | | User data starts here... . | |
960 | . . | |
961 | . (malloc_usable_space() bytes) . | |
962 | . | | |
963 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
964 | | Size of chunk | | |
965 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
966 | ||
967 | ||
968 | Where "chunk" is the front of the chunk for the purpose of most of | |
969 | the malloc code, but "mem" is the pointer that is returned to the | |
970 | user. "Nextchunk" is the beginning of the next contiguous chunk. | |
971 | ||
972 | Chunks always begin on even word boundries, so the mem portion | |
973 | (which is returned to the user) is also on an even word boundary, and | |
974 | thus double-word aligned. | |
975 | ||
976 | Free chunks are stored in circular doubly-linked lists, and look like this: | |
977 | ||
978 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
979 | | Size of previous chunk | | |
980 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
981 | `head:' | Size of chunk, in bytes |P| | |
982 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
983 | | Forward pointer to next chunk in list | | |
984 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
985 | | Back pointer to previous chunk in list | | |
986 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
987 | | Unused space (may be 0 bytes long) . | |
988 | . . | |
989 | . | | |
990 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
991 | `foot:' | Size of chunk, in bytes | | |
992 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
993 | ||
994 | The P (PREV_INUSE) bit, stored in the unused low-order bit of the | |
995 | chunk size (which is always a multiple of two words), is an in-use | |
996 | bit for the *previous* chunk. If that bit is *clear*, then the | |
997 | word before the current chunk size contains the previous chunk | |
998 | size, and can be used to find the front of the previous chunk. | |
999 | (The very first chunk allocated always has this bit set, | |
1000 | preventing access to non-existent (or non-owned) memory.) | |
1001 | ||
1002 | Note that the `foot' of the current chunk is actually represented | |
1003 | as the prev_size of the NEXT chunk. (This makes it easier to | |
1004 | deal with alignments etc). | |
1005 | ||
1006 | The two exceptions to all this are | |
1007 | ||
1008 | 1. The special chunk `top', which doesn't bother using the | |
1009 | trailing size field since there is no | |
1010 | next contiguous chunk that would have to index off it. (After | |
1011 | initialization, `top' is forced to always exist. If it would | |
1012 | become less than MINSIZE bytes long, it is replenished via | |
1013 | malloc_extend_top.) | |
1014 | ||
1015 | 2. Chunks allocated via mmap, which have the second-lowest-order | |
1016 | bit (IS_MMAPPED) set in their size fields. Because they are | |
1017 | never merged or traversed from any other chunk, they have no | |
1018 | foot size or inuse information. | |
1019 | ||
1020 | Available chunks are kept in any of several places (all declared below): | |
1021 | ||
1022 | * `av': An array of chunks serving as bin headers for consolidated | |
1023 | chunks. Each bin is doubly linked. The bins are approximately | |
1024 | proportionally (log) spaced. There are a lot of these bins | |
1025 | (128). This may look excessive, but works very well in | |
1026 | practice. All procedures maintain the invariant that no | |
1027 | consolidated chunk physically borders another one. Chunks in | |
1028 | bins are kept in size order, with ties going to the | |
1029 | approximately least recently used chunk. | |
1030 | ||
1031 | The chunks in each bin are maintained in decreasing sorted order by | |
1032 | size. This is irrelevant for the small bins, which all contain | |
1033 | the same-sized chunks, but facilitates best-fit allocation for | |
1034 | larger chunks. (These lists are just sequential. Keeping them in | |
1035 | order almost never requires enough traversal to warrant using | |
1036 | fancier ordered data structures.) Chunks of the same size are | |
1037 | linked with the most recently freed at the front, and allocations | |
1038 | are taken from the back. This results in LRU or FIFO allocation | |
1039 | order, which tends to give each chunk an equal opportunity to be | |
1040 | consolidated with adjacent freed chunks, resulting in larger free | |
1041 | chunks and less fragmentation. | |
1042 | ||
1043 | * `top': The top-most available chunk (i.e., the one bordering the | |
1044 | end of available memory) is treated specially. It is never | |
1045 | included in any bin, is used only if no other chunk is | |
1046 | available, and is released back to the system if it is very | |
1047 | large (see M_TRIM_THRESHOLD). | |
1048 | ||
1049 | * `last_remainder': A bin holding only the remainder of the | |
1050 | most recently split (non-top) chunk. This bin is checked | |
1051 | before other non-fitting chunks, so as to provide better | |
1052 | locality for runs of sequentially allocated chunks. | |
1053 | ||
1054 | * Implicitly, through the host system's memory mapping tables. | |
1055 | If supported, requests greater than a threshold are usually | |
1056 | serviced via calls to mmap, and then later released via munmap. | |
1057 | ||
1058 | */ | |
1059 | ||
1060 | /* | |
1061 | Bins | |
1062 | ||
1063 | The bins are an array of pairs of pointers serving as the | |
1064 | heads of (initially empty) doubly-linked lists of chunks, laid out | |
1065 | in a way so that each pair can be treated as if it were in a | |
1066 | malloc_chunk. (This way, the fd/bk offsets for linking bin heads | |
1067 | and chunks are the same). | |
1068 | ||
1069 | Bins for sizes < 512 bytes contain chunks of all the same size, spaced | |
1070 | 8 bytes apart. Larger bins are approximately logarithmically | |
1071 | spaced. (See the table below.) | |
1072 | ||
1073 | Bin layout: | |
1074 | ||
1075 | 64 bins of size 8 | |
1076 | 32 bins of size 64 | |
1077 | 16 bins of size 512 | |
1078 | 8 bins of size 4096 | |
1079 | 4 bins of size 32768 | |
1080 | 2 bins of size 262144 | |
1081 | 1 bin of size what's left | |
1082 | ||
1083 | There is actually a little bit of slop in the numbers in bin_index | |
1084 | for the sake of speed. This makes no difference elsewhere. | |
1085 | ||
1086 | The special chunks `top' and `last_remainder' get their own bins, | |
1087 | (this is implemented via yet more trickery with the av array), | |
1088 | although `top' is never properly linked to its bin since it is | |
1089 | always handled specially. | |
1090 | ||
1091 | */ | |
1092 | ||
1093 | #define NAV 128 /* number of bins */ | |
1094 | ||
1095 | typedef struct malloc_chunk* mbinptr; | |
1096 | ||
1097 | /* An arena is a configuration of malloc_chunks together with an array | |
1098 | of bins. With multiple threads, it must be locked via a mutex | |
1099 | before changing its data structures. One or more `heaps' are | |
1100 | associated with each arena, except for the main_arena, which is | |
1101 | associated only with the `main heap', i.e. the conventional free | |
1102 | store obtained with calls to MORECORE() (usually sbrk). The `av' | |
1103 | array is never mentioned directly in the code, but instead used via | |
1104 | bin access macros. */ | |
1105 | ||
1106 | typedef struct _arena { | |
1107 | mbinptr av[2*NAV + 2]; | |
1108 | struct _arena *next; | |
1109 | mutex_t mutex; | |
1110 | } arena; | |
1111 | ||
1112 | ||
1113 | /* A heap is a single contiguous memory region holding (coalescable) | |
1114 | malloc_chunks. It is allocated with mmap() and always starts at an | |
1115 | address aligned to HEAP_MAX_SIZE. Not used unless compiling for | |
1116 | multiple threads. */ | |
1117 | ||
1118 | typedef struct _heap_info { | |
1119 | arena *ar_ptr; | |
1120 | size_t size; | |
1121 | } heap_info; | |
1122 | ||
1123 | ||
1124 | /* | |
1125 | Static functions (forward declarations) | |
1126 | */ | |
1127 | ||
1128 | #if __STD_C | |
1129 | static void chunk_free(arena *ar_ptr, mchunkptr p); | |
1130 | static mchunkptr chunk_alloc(arena *ar_ptr, INTERNAL_SIZE_T size); | |
1131 | static int arena_trim(arena *ar_ptr, size_t pad); | |
1132 | #else | |
1133 | static void chunk_free(); | |
1134 | static mchunkptr chunk_alloc(); | |
1135 | static int arena_trim(); | |
1136 | #endif | |
1137 | ||
1138 | \f | |
1139 | ||
1140 | /* sizes, alignments */ | |
1141 | ||
1142 | #define SIZE_SZ (sizeof(INTERNAL_SIZE_T)) | |
1143 | #define MALLOC_ALIGNMENT (SIZE_SZ + SIZE_SZ) | |
1144 | #define MALLOC_ALIGN_MASK (MALLOC_ALIGNMENT - 1) | |
1145 | #define MINSIZE (sizeof(struct malloc_chunk)) | |
1146 | ||
1147 | /* conversion from malloc headers to user pointers, and back */ | |
1148 | ||
1149 | #define chunk2mem(p) ((Void_t*)((char*)(p) + 2*SIZE_SZ)) | |
1150 | #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ)) | |
1151 | ||
1152 | /* pad request bytes into a usable size */ | |
1153 | ||
1154 | #define request2size(req) \ | |
1155 | (((long)((req) + (SIZE_SZ + MALLOC_ALIGN_MASK)) < \ | |
1156 | (long)(MINSIZE + MALLOC_ALIGN_MASK)) ? MINSIZE : \ | |
1157 | (((req) + (SIZE_SZ + MALLOC_ALIGN_MASK)) & ~(MALLOC_ALIGN_MASK))) | |
1158 | ||
1159 | /* Check if m has acceptable alignment */ | |
1160 | ||
1161 | #define aligned_OK(m) (((unsigned long)((m)) & (MALLOC_ALIGN_MASK)) == 0) | |
1162 | ||
1163 | ||
1164 | \f | |
1165 | ||
1166 | /* | |
1167 | Physical chunk operations | |
1168 | */ | |
1169 | ||
1170 | ||
1171 | /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */ | |
1172 | ||
1173 | #define PREV_INUSE 0x1 | |
1174 | ||
1175 | /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */ | |
1176 | ||
1177 | #define IS_MMAPPED 0x2 | |
1178 | ||
1179 | /* Bits to mask off when extracting size */ | |
1180 | ||
1181 | #define SIZE_BITS (PREV_INUSE|IS_MMAPPED) | |
1182 | ||
1183 | ||
1184 | /* Ptr to next physical malloc_chunk. */ | |
1185 | ||
1186 | #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->size & ~PREV_INUSE) )) | |
1187 | ||
1188 | /* Ptr to previous physical malloc_chunk */ | |
1189 | ||
1190 | #define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_size) )) | |
1191 | ||
1192 | ||
1193 | /* Treat space at ptr + offset as a chunk */ | |
1194 | ||
1195 | #define chunk_at_offset(p, s) ((mchunkptr)(((char*)(p)) + (s))) | |
1196 | ||
1197 | ||
1198 | \f | |
1199 | ||
1200 | /* | |
1201 | Dealing with use bits | |
1202 | */ | |
1203 | ||
1204 | /* extract p's inuse bit */ | |
1205 | ||
1206 | #define inuse(p) \ | |
1207 | ((((mchunkptr)(((char*)(p))+((p)->size & ~PREV_INUSE)))->size) & PREV_INUSE) | |
1208 | ||
1209 | /* extract inuse bit of previous chunk */ | |
1210 | ||
1211 | #define prev_inuse(p) ((p)->size & PREV_INUSE) | |
1212 | ||
1213 | /* check for mmap()'ed chunk */ | |
1214 | ||
1215 | #define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED) | |
1216 | ||
1217 | /* set/clear chunk as in use without otherwise disturbing */ | |
1218 | ||
1219 | #define set_inuse(p) \ | |
1220 | ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size |= PREV_INUSE | |
1221 | ||
1222 | #define clear_inuse(p) \ | |
1223 | ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size &= ~(PREV_INUSE) | |
1224 | ||
1225 | /* check/set/clear inuse bits in known places */ | |
1226 | ||
1227 | #define inuse_bit_at_offset(p, s)\ | |
1228 | (((mchunkptr)(((char*)(p)) + (s)))->size & PREV_INUSE) | |
1229 | ||
1230 | #define set_inuse_bit_at_offset(p, s)\ | |
1231 | (((mchunkptr)(((char*)(p)) + (s)))->size |= PREV_INUSE) | |
1232 | ||
1233 | #define clear_inuse_bit_at_offset(p, s)\ | |
1234 | (((mchunkptr)(((char*)(p)) + (s)))->size &= ~(PREV_INUSE)) | |
1235 | ||
1236 | ||
1237 | \f | |
1238 | ||
1239 | /* | |
1240 | Dealing with size fields | |
1241 | */ | |
1242 | ||
1243 | /* Get size, ignoring use bits */ | |
1244 | ||
1245 | #define chunksize(p) ((p)->size & ~(SIZE_BITS)) | |
1246 | ||
1247 | /* Set size at head, without disturbing its use bit */ | |
1248 | ||
1249 | #define set_head_size(p, s) ((p)->size = (((p)->size & PREV_INUSE) | (s))) | |
1250 | ||
1251 | /* Set size/use ignoring previous bits in header */ | |
1252 | ||
1253 | #define set_head(p, s) ((p)->size = (s)) | |
1254 | ||
1255 | /* Set size at footer (only when chunk is not in use) */ | |
1256 | ||
1257 | #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_size = (s)) | |
1258 | ||
1259 | ||
1260 | \f | |
1261 | ||
1262 | ||
1263 | /* access macros */ | |
1264 | ||
1265 | #define bin_at(a, i) ((mbinptr)((char*)&(((a)->av)[2*(i) + 2]) - 2*SIZE_SZ)) | |
1266 | #define init_bin(a, i) ((a)->av[2*i+2] = (a)->av[2*i+3] = bin_at((a), i)) | |
1267 | #define next_bin(b) ((mbinptr)((char*)(b) + 2 * sizeof(mbinptr))) | |
1268 | #define prev_bin(b) ((mbinptr)((char*)(b) - 2 * sizeof(mbinptr))) | |
1269 | ||
1270 | /* | |
1271 | The first 2 bins are never indexed. The corresponding av cells are instead | |
1272 | used for bookkeeping. This is not to save space, but to simplify | |
1273 | indexing, maintain locality, and avoid some initialization tests. | |
1274 | */ | |
1275 | ||
1276 | #define binblocks(a) (bin_at(a,0)->size)/* bitvector of nonempty blocks */ | |
1277 | #define top(a) (bin_at(a,0)->fd) /* The topmost chunk */ | |
1278 | #define last_remainder(a) (bin_at(a,1)) /* remainder from last split */ | |
1279 | ||
1280 | /* | |
1281 | Because top initially points to its own bin with initial | |
1282 | zero size, thus forcing extension on the first malloc request, | |
1283 | we avoid having any special code in malloc to check whether | |
1284 | it even exists yet. But we still need to in malloc_extend_top. | |
1285 | */ | |
1286 | ||
1287 | #define initial_top(a) ((mchunkptr)bin_at(a, 0)) | |
1288 | ||
1289 | \f | |
1290 | ||
1291 | /* field-extraction macros */ | |
1292 | ||
1293 | #define first(b) ((b)->fd) | |
1294 | #define last(b) ((b)->bk) | |
1295 | ||
1296 | /* | |
1297 | Indexing into bins | |
1298 | */ | |
1299 | ||
1300 | #define bin_index(sz) \ | |
1301 | (((((unsigned long)(sz)) >> 9) == 0) ? (((unsigned long)(sz)) >> 3): \ | |
1302 | ((((unsigned long)(sz)) >> 9) <= 4) ? 56 + (((unsigned long)(sz)) >> 6): \ | |
1303 | ((((unsigned long)(sz)) >> 9) <= 20) ? 91 + (((unsigned long)(sz)) >> 9): \ | |
1304 | ((((unsigned long)(sz)) >> 9) <= 84) ? 110 + (((unsigned long)(sz)) >> 12): \ | |
1305 | ((((unsigned long)(sz)) >> 9) <= 340) ? 119 + (((unsigned long)(sz)) >> 15): \ | |
1306 | ((((unsigned long)(sz)) >> 9) <= 1364) ? 124 + (((unsigned long)(sz)) >> 18): \ | |
1307 | 126) | |
1308 | /* | |
1309 | bins for chunks < 512 are all spaced 8 bytes apart, and hold | |
1310 | identically sized chunks. This is exploited in malloc. | |
1311 | */ | |
1312 | ||
1313 | #define MAX_SMALLBIN 63 | |
1314 | #define MAX_SMALLBIN_SIZE 512 | |
1315 | #define SMALLBIN_WIDTH 8 | |
1316 | ||
1317 | #define smallbin_index(sz) (((unsigned long)(sz)) >> 3) | |
1318 | ||
1319 | /* | |
1320 | Requests are `small' if both the corresponding and the next bin are small | |
1321 | */ | |
1322 | ||
1323 | #define is_small_request(nb) ((nb) < MAX_SMALLBIN_SIZE - SMALLBIN_WIDTH) | |
1324 | ||
1325 | \f | |
1326 | ||
1327 | /* | |
1328 | To help compensate for the large number of bins, a one-level index | |
1329 | structure is used for bin-by-bin searching. `binblocks' is a | |
1330 | one-word bitvector recording whether groups of BINBLOCKWIDTH bins | |
1331 | have any (possibly) non-empty bins, so they can be skipped over | |
1332 | all at once during during traversals. The bits are NOT always | |
1333 | cleared as soon as all bins in a block are empty, but instead only | |
1334 | when all are noticed to be empty during traversal in malloc. | |
1335 | */ | |
1336 | ||
1337 | #define BINBLOCKWIDTH 4 /* bins per block */ | |
1338 | ||
1339 | /* bin<->block macros */ | |
1340 | ||
1341 | #define idx2binblock(ix) ((unsigned)1 << ((ix) / BINBLOCKWIDTH)) | |
1342 | #define mark_binblock(a, ii) (binblocks(a) |= idx2binblock(ii)) | |
1343 | #define clear_binblock(a, ii) (binblocks(a) &= ~(idx2binblock(ii))) | |
1344 | ||
1345 | ||
1346 | \f | |
1347 | ||
1348 | /* Static bookkeeping data */ | |
1349 | ||
1350 | /* Helper macro to initialize bins */ | |
1351 | #define IAV(i) bin_at(&main_arena, i), bin_at(&main_arena, i) | |
1352 | ||
1353 | static arena main_arena = { | |
1354 | { | |
1355 | 0, 0, | |
1356 | IAV(0), IAV(1), IAV(2), IAV(3), IAV(4), IAV(5), IAV(6), IAV(7), | |
1357 | IAV(8), IAV(9), IAV(10), IAV(11), IAV(12), IAV(13), IAV(14), IAV(15), | |
1358 | IAV(16), IAV(17), IAV(18), IAV(19), IAV(20), IAV(21), IAV(22), IAV(23), | |
1359 | IAV(24), IAV(25), IAV(26), IAV(27), IAV(28), IAV(29), IAV(30), IAV(31), | |
1360 | IAV(32), IAV(33), IAV(34), IAV(35), IAV(36), IAV(37), IAV(38), IAV(39), | |
1361 | IAV(40), IAV(41), IAV(42), IAV(43), IAV(44), IAV(45), IAV(46), IAV(47), | |
1362 | IAV(48), IAV(49), IAV(50), IAV(51), IAV(52), IAV(53), IAV(54), IAV(55), | |
1363 | IAV(56), IAV(57), IAV(58), IAV(59), IAV(60), IAV(61), IAV(62), IAV(63), | |
1364 | IAV(64), IAV(65), IAV(66), IAV(67), IAV(68), IAV(69), IAV(70), IAV(71), | |
1365 | IAV(72), IAV(73), IAV(74), IAV(75), IAV(76), IAV(77), IAV(78), IAV(79), | |
1366 | IAV(80), IAV(81), IAV(82), IAV(83), IAV(84), IAV(85), IAV(86), IAV(87), | |
1367 | IAV(88), IAV(89), IAV(90), IAV(91), IAV(92), IAV(93), IAV(94), IAV(95), | |
1368 | IAV(96), IAV(97), IAV(98), IAV(99), IAV(100), IAV(101), IAV(102), IAV(103), | |
1369 | IAV(104), IAV(105), IAV(106), IAV(107), IAV(108), IAV(109), IAV(110), IAV(111), | |
1370 | IAV(112), IAV(113), IAV(114), IAV(115), IAV(116), IAV(117), IAV(118), IAV(119), | |
1371 | IAV(120), IAV(121), IAV(122), IAV(123), IAV(124), IAV(125), IAV(126), IAV(127) | |
1372 | }, | |
1373 | NULL, /* next */ | |
1374 | MUTEX_INITIALIZER /* mutex */ | |
1375 | }; | |
1376 | ||
1377 | #undef IAV | |
1378 | ||
1379 | /* Thread specific data */ | |
1380 | ||
1381 | static tsd_key_t arena_key; | |
1382 | static mutex_t list_lock = MUTEX_INITIALIZER; | |
1383 | ||
1384 | #if THREAD_STATS | |
1385 | static int stat_n_arenas = 0; | |
1386 | static int stat_n_heaps = 0; | |
1387 | static long stat_lock_direct = 0; | |
1388 | static long stat_lock_loop = 0; | |
1389 | #define THREAD_STAT(x) x | |
1390 | #else | |
1391 | #define THREAD_STAT(x) do ; while(0) | |
1392 | #endif | |
1393 | ||
1394 | /* variables holding tunable values */ | |
1395 | ||
1396 | static unsigned long trim_threshold = DEFAULT_TRIM_THRESHOLD; | |
1397 | static unsigned long top_pad = DEFAULT_TOP_PAD; | |
1398 | static unsigned int n_mmaps_max = DEFAULT_MMAP_MAX; | |
1399 | static unsigned long mmap_threshold = DEFAULT_MMAP_THRESHOLD; | |
1400 | ||
1401 | /* The first value returned from sbrk */ | |
1402 | static char* sbrk_base = (char*)(-1); | |
1403 | ||
1404 | /* The maximum memory obtained from system via sbrk */ | |
1405 | static unsigned long max_sbrked_mem = 0; | |
1406 | ||
1407 | /* The maximum via either sbrk or mmap */ | |
1408 | static unsigned long max_total_mem = 0; | |
1409 | ||
1410 | /* internal working copy of mallinfo */ | |
1411 | static struct mallinfo current_mallinfo = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; | |
1412 | ||
1413 | /* The total memory obtained from system via sbrk */ | |
1414 | #define sbrked_mem (current_mallinfo.arena) | |
1415 | ||
1416 | /* Tracking mmaps */ | |
1417 | ||
1418 | static unsigned int n_mmaps = 0; | |
1419 | static unsigned int max_n_mmaps = 0; | |
1420 | static unsigned long mmapped_mem = 0; | |
1421 | static unsigned long max_mmapped_mem = 0; | |
1422 | ||
1423 | ||
1424 | \f | |
1425 | ||
1426 | ||
1427 | /* Initialization routine. */ | |
1428 | #if defined(_LIBC) | |
1429 | static void ptmalloc_init __MALLOC_P ((void)) __attribute__ ((constructor)); | |
1430 | ||
1431 | static void | |
1432 | ptmalloc_init __MALLOC_P((void)) | |
1433 | #else | |
1434 | void | |
1435 | ptmalloc_init __MALLOC_P((void)) | |
1436 | #endif | |
1437 | { | |
1438 | static int first = 1; | |
1439 | ||
1440 | #if defined(_LIBC) | |
1441 | /* Initialize the pthread. */ | |
1442 | if (__pthread_initialize != NULL) | |
1443 | __pthread_initialize (); | |
1444 | #endif | |
1445 | ||
1446 | if(first) { | |
1447 | first = 0; | |
1448 | mutex_init(&main_arena.mutex); | |
1449 | mutex_init(&list_lock); | |
1450 | tsd_key_create(&arena_key, NULL); | |
1451 | tsd_setspecific(arena_key, (Void_t *)&main_arena); | |
1452 | } | |
1453 | } | |
1454 | ||
1455 | ||
1456 | \f | |
1457 | ||
1458 | ||
1459 | /* Routines dealing with mmap(). */ | |
1460 | ||
1461 | #if HAVE_MMAP | |
1462 | ||
1463 | #ifndef MAP_ANONYMOUS | |
1464 | ||
1465 | static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */ | |
1466 | ||
1467 | #define MMAP(size, prot) ((dev_zero_fd < 0) ? \ | |
1468 | (dev_zero_fd = open("/dev/zero", O_RDWR), \ | |
1469 | mmap(0, (size), (prot), MAP_PRIVATE, dev_zero_fd, 0)) : \ | |
1470 | mmap(0, (size), (prot), MAP_PRIVATE, dev_zero_fd, 0)) | |
1471 | ||
1472 | #else | |
1473 | ||
1474 | #define MMAP(size, prot) \ | |
1475 | (mmap(0, (size), (prot), MAP_PRIVATE|MAP_ANONYMOUS, -1, 0)) | |
1476 | ||
1477 | #endif | |
1478 | ||
1479 | #if __STD_C | |
1480 | static mchunkptr mmap_chunk(size_t size) | |
1481 | #else | |
1482 | static mchunkptr mmap_chunk(size) size_t size; | |
1483 | #endif | |
1484 | { | |
1485 | size_t page_mask = malloc_getpagesize - 1; | |
1486 | mchunkptr p; | |
1487 | ||
1488 | if(n_mmaps >= n_mmaps_max) return 0; /* too many regions */ | |
1489 | ||
1490 | /* For mmapped chunks, the overhead is one SIZE_SZ unit larger, because | |
1491 | * there is no following chunk whose prev_size field could be used. | |
1492 | */ | |
1493 | size = (size + SIZE_SZ + page_mask) & ~page_mask; | |
1494 | ||
1495 | p = (mchunkptr)MMAP(size, PROT_READ|PROT_WRITE); | |
1496 | if(p == (mchunkptr)-1) return 0; | |
1497 | ||
1498 | n_mmaps++; | |
1499 | if (n_mmaps > max_n_mmaps) max_n_mmaps = n_mmaps; | |
1500 | ||
1501 | /* We demand that eight bytes into a page must be 8-byte aligned. */ | |
1502 | assert(aligned_OK(chunk2mem(p))); | |
1503 | ||
1504 | /* The offset to the start of the mmapped region is stored | |
1505 | * in the prev_size field of the chunk; normally it is zero, | |
1506 | * but that can be changed in memalign(). | |
1507 | */ | |
1508 | p->prev_size = 0; | |
1509 | set_head(p, size|IS_MMAPPED); | |
1510 | ||
1511 | mmapped_mem += size; | |
1512 | if ((unsigned long)mmapped_mem > (unsigned long)max_mmapped_mem) | |
1513 | max_mmapped_mem = mmapped_mem; | |
1514 | if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem) | |
1515 | max_total_mem = mmapped_mem + sbrked_mem; | |
1516 | return p; | |
1517 | } | |
1518 | ||
1519 | #if __STD_C | |
1520 | static void munmap_chunk(mchunkptr p) | |
1521 | #else | |
1522 | static void munmap_chunk(p) mchunkptr p; | |
1523 | #endif | |
1524 | { | |
1525 | INTERNAL_SIZE_T size = chunksize(p); | |
1526 | int ret; | |
1527 | ||
1528 | assert (chunk_is_mmapped(p)); | |
1529 | assert(! ((char*)p >= sbrk_base && (char*)p < sbrk_base + sbrked_mem)); | |
1530 | assert((n_mmaps > 0)); | |
1531 | assert(((p->prev_size + size) & (malloc_getpagesize-1)) == 0); | |
1532 | ||
1533 | n_mmaps--; | |
1534 | mmapped_mem -= (size + p->prev_size); | |
1535 | ||
1536 | ret = munmap((char *)p - p->prev_size, size + p->prev_size); | |
1537 | ||
1538 | /* munmap returns non-zero on failure */ | |
1539 | assert(ret == 0); | |
1540 | } | |
1541 | ||
1542 | #if HAVE_MREMAP | |
1543 | ||
1544 | #if __STD_C | |
1545 | static mchunkptr mremap_chunk(mchunkptr p, size_t new_size) | |
1546 | #else | |
1547 | static mchunkptr mremap_chunk(p, new_size) mchunkptr p; size_t new_size; | |
1548 | #endif | |
1549 | { | |
1550 | size_t page_mask = malloc_getpagesize - 1; | |
1551 | INTERNAL_SIZE_T offset = p->prev_size; | |
1552 | INTERNAL_SIZE_T size = chunksize(p); | |
1553 | char *cp; | |
1554 | ||
1555 | assert (chunk_is_mmapped(p)); | |
1556 | assert(! ((char*)p >= sbrk_base && (char*)p < sbrk_base + sbrked_mem)); | |
1557 | assert((n_mmaps > 0)); | |
1558 | assert(((size + offset) & (malloc_getpagesize-1)) == 0); | |
1559 | ||
1560 | /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */ | |
1561 | new_size = (new_size + offset + SIZE_SZ + page_mask) & ~page_mask; | |
1562 | ||
1563 | cp = (char *)mremap((char *)p - offset, size + offset, new_size, | |
1564 | MREMAP_MAYMOVE); | |
1565 | ||
1566 | if (cp == (char *)-1) return 0; | |
1567 | ||
1568 | p = (mchunkptr)(cp + offset); | |
1569 | ||
1570 | assert(aligned_OK(chunk2mem(p))); | |
1571 | ||
1572 | assert((p->prev_size == offset)); | |
1573 | set_head(p, (new_size - offset)|IS_MMAPPED); | |
1574 | ||
1575 | mmapped_mem -= size + offset; | |
1576 | mmapped_mem += new_size; | |
1577 | if ((unsigned long)mmapped_mem > (unsigned long)max_mmapped_mem) | |
1578 | max_mmapped_mem = mmapped_mem; | |
1579 | if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem) | |
1580 | max_total_mem = mmapped_mem + sbrked_mem; | |
1581 | return p; | |
1582 | } | |
1583 | ||
1584 | #endif /* HAVE_MREMAP */ | |
1585 | ||
1586 | #endif /* HAVE_MMAP */ | |
1587 | ||
1588 | \f | |
1589 | ||
1590 | /* Managing heaps and arenas (for concurrent threads) */ | |
1591 | ||
1592 | #ifndef NO_THREADS | |
1593 | ||
1594 | /* Create a new heap. size is automatically rounded up to a multiple | |
1595 | of the page size. */ | |
1596 | ||
1597 | static heap_info * | |
1598 | #if __STD_C | |
1599 | new_heap(size_t size) | |
1600 | #else | |
1601 | new_heap(size) size_t size; | |
1602 | #endif | |
1603 | { | |
1604 | size_t page_mask = malloc_getpagesize - 1; | |
1605 | char *p1, *p2; | |
1606 | unsigned long ul; | |
1607 | heap_info *h; | |
1608 | ||
1609 | if(size < HEAP_MIN_SIZE) | |
1610 | size = HEAP_MIN_SIZE; | |
1611 | size = (size + page_mask) & ~page_mask; | |
1612 | if(size > HEAP_MAX_SIZE) | |
1613 | return 0; | |
1614 | p1 = (char *)MMAP(HEAP_MAX_SIZE<<1, PROT_NONE); | |
1615 | if(p1 == (char *)-1) | |
1616 | return 0; | |
1617 | p2 = (char *)(((unsigned long)p1 + HEAP_MAX_SIZE) & ~(HEAP_MAX_SIZE-1)); | |
1618 | ul = p2 - p1; | |
1619 | munmap(p1, ul); | |
1620 | munmap(p2 + HEAP_MAX_SIZE, HEAP_MAX_SIZE - ul); | |
1621 | if(mprotect(p2, size, PROT_READ|PROT_WRITE) != 0) { | |
1622 | munmap(p2, HEAP_MAX_SIZE); | |
1623 | return 0; | |
1624 | } | |
1625 | h = (heap_info *)p2; | |
1626 | h->size = size; | |
1627 | THREAD_STAT(stat_n_heaps++); | |
1628 | return h; | |
1629 | } | |
1630 | ||
1631 | /* Grow or shrink a heap. size is automatically rounded up to a | |
1632 | multiple of the page size. */ | |
1633 | ||
1634 | static int | |
1635 | #if __STD_C | |
1636 | grow_heap(heap_info *h, long diff) | |
1637 | #else | |
1638 | grow_heap(h, diff) heap_info *h; long diff; | |
1639 | #endif | |
1640 | { | |
1641 | size_t page_mask = malloc_getpagesize - 1; | |
1642 | long new_size; | |
1643 | ||
1644 | if(diff >= 0) { | |
1645 | diff = (diff + page_mask) & ~page_mask; | |
1646 | new_size = (long)h->size + diff; | |
1647 | if(new_size > HEAP_MAX_SIZE) | |
1648 | return -1; | |
1649 | if(mprotect((char *)h + h->size, diff, PROT_READ|PROT_WRITE) != 0) | |
1650 | return -2; | |
1651 | } else { | |
1652 | new_size = (long)h->size + diff; | |
1653 | if(new_size < 0) | |
1654 | return -1; | |
1655 | if(mprotect((char *)h + new_size, -diff, PROT_NONE) != 0) | |
1656 | return -2; | |
1657 | } | |
1658 | h->size = new_size; | |
1659 | return 0; | |
1660 | } | |
1661 | ||
1662 | /* arena_get() acquires an arena and locks the corresponding mutex. | |
1663 | First, try the one last locked successfully by this thread. (This | |
1664 | is the common case and handled with a macro for speed.) Then, loop | |
1665 | over the singly linked list of arenas. If no arena is readily | |
1666 | available, create a new one. */ | |
1667 | ||
1668 | #define arena_get(ptr, size) do { \ | |
1669 | Void_t *vptr = NULL; \ | |
1670 | ptr = (arena *)tsd_getspecific(arena_key, vptr); \ | |
1671 | if(ptr && !mutex_trylock(&ptr->mutex)) { \ | |
1672 | THREAD_STAT(stat_lock_direct++); \ | |
1673 | } else { \ | |
1674 | ptr = arena_get2(ptr, (size)); \ | |
1675 | } \ | |
1676 | } while(0) | |
1677 | ||
1678 | static arena * | |
1679 | #if __STD_C | |
1680 | arena_get2(arena *a_tsd, size_t size) | |
1681 | #else | |
1682 | arena_get2(a_tsd, size) arena *a_tsd; size_t size; | |
1683 | #endif | |
1684 | { | |
1685 | arena *a; | |
1686 | heap_info *h; | |
1687 | char *ptr; | |
1688 | int i; | |
1689 | unsigned long misalign; | |
1690 | ||
1691 | /* Check the list for unlocked arenas. */ | |
1692 | if(a_tsd) { | |
1693 | for(a = a_tsd->next; a; a = a->next) { | |
1694 | if(!mutex_trylock(&a->mutex)) | |
1695 | goto done; | |
1696 | } | |
1697 | for(a = &main_arena; a != a_tsd; a = a->next) { | |
1698 | if(!mutex_trylock(&a->mutex)) | |
1699 | goto done; | |
1700 | } | |
1701 | } else { | |
1702 | for(a = &main_arena; a; a = a->next) { | |
1703 | if(!mutex_trylock(&a->mutex)) | |
1704 | goto done; | |
1705 | } | |
1706 | } | |
1707 | ||
1708 | /* Nothing immediately available, so generate a new arena. */ | |
1709 | h = new_heap(size + (sizeof(*h) + sizeof(*a) + MALLOC_ALIGNMENT)); | |
1710 | if(!h) | |
1711 | return 0; | |
1712 | a = h->ar_ptr = (arena *)(h+1); | |
1713 | for(i=0; i<NAV; i++) | |
1714 | init_bin(a, i); | |
1715 | mutex_init(&a->mutex); | |
1716 | i = mutex_lock(&a->mutex); /* remember result */ | |
1717 | ||
1718 | /* Set up the top chunk, with proper alignment. */ | |
1719 | ptr = (char *)(a + 1); | |
1720 | misalign = (unsigned long)chunk2mem(ptr) & MALLOC_ALIGN_MASK; | |
1721 | if (misalign > 0) | |
1722 | ptr += MALLOC_ALIGNMENT - misalign; | |
1723 | top(a) = (mchunkptr)ptr; | |
1724 | set_head(top(a), (h->size - (ptr-(char*)h)) | PREV_INUSE); | |
1725 | ||
1726 | /* Add the new arena to the list. */ | |
1727 | (void)mutex_lock(&list_lock); | |
1728 | a->next = main_arena.next; | |
1729 | main_arena.next = a; | |
1730 | THREAD_STAT(stat_n_arenas++); | |
1731 | (void)mutex_unlock(&list_lock); | |
1732 | ||
1733 | if(i) /* locking failed; keep arena for further attempts later */ | |
1734 | return 0; | |
1735 | ||
1736 | done: | |
1737 | THREAD_STAT(stat_lock_loop++); | |
1738 | tsd_setspecific(arena_key, (Void_t *)a); | |
1739 | return a; | |
1740 | } | |
1741 | ||
1742 | /* find the heap and corresponding arena for a given ptr */ | |
1743 | ||
1744 | #define heap_for_ptr(ptr) \ | |
1745 | ((heap_info *)((unsigned long)(ptr) & ~(HEAP_MAX_SIZE-1))) | |
1746 | #define arena_for_ptr(ptr) \ | |
1747 | (((mchunkptr)(ptr) < top(&main_arena) && (char *)(ptr) >= sbrk_base) ? \ | |
1748 | &main_arena : heap_for_ptr(ptr)->ar_ptr) | |
1749 | ||
1750 | #else /* defined(NO_THREADS) */ | |
1751 | ||
1752 | /* Without concurrent threads, there is only one arena. */ | |
1753 | ||
1754 | #define arena_get(ptr, sz) (ptr = &main_arena) | |
1755 | #define arena_for_ptr(ptr) (&main_arena) | |
1756 | ||
1757 | #endif /* !defined(NO_THREADS) */ | |
1758 | ||
1759 | \f | |
1760 | ||
1761 | /* | |
1762 | Debugging support | |
1763 | */ | |
1764 | ||
1765 | #if MALLOC_DEBUG | |
1766 | ||
1767 | ||
1768 | /* | |
1769 | These routines make a number of assertions about the states | |
1770 | of data structures that should be true at all times. If any | |
1771 | are not true, it's very likely that a user program has somehow | |
1772 | trashed memory. (It's also possible that there is a coding error | |
1773 | in malloc. In which case, please report it!) | |
1774 | */ | |
1775 | ||
1776 | #if __STD_C | |
1777 | static void do_check_chunk(arena *ar_ptr, mchunkptr p) | |
1778 | #else | |
1779 | static void do_check_chunk(ar_ptr, p) arena *ar_ptr; mchunkptr p; | |
1780 | #endif | |
1781 | { | |
1782 | INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE; | |
1783 | ||
1784 | /* No checkable chunk is mmapped */ | |
1785 | assert(!chunk_is_mmapped(p)); | |
1786 | ||
1787 | #ifndef NO_THREADS | |
1788 | if(ar_ptr != &main_arena) { | |
1789 | heap_info *heap = heap_for_ptr(p); | |
1790 | assert(heap->ar_ptr == ar_ptr); | |
1791 | assert((char *)p + sz <= (char *)heap + heap->size); | |
1792 | return; | |
1793 | } | |
1794 | #endif | |
1795 | ||
1796 | /* Check for legal address ... */ | |
1797 | assert((char*)p >= sbrk_base); | |
1798 | if (p != top(ar_ptr)) | |
1799 | assert((char*)p + sz <= (char*)top(ar_ptr)); | |
1800 | else | |
1801 | assert((char*)p + sz <= sbrk_base + sbrked_mem); | |
1802 | ||
1803 | } | |
1804 | ||
1805 | ||
1806 | #if __STD_C | |
1807 | static void do_check_free_chunk(arena *ar_ptr, mchunkptr p) | |
1808 | #else | |
1809 | static void do_check_free_chunk(ar_ptr, p) arena *ar_ptr; mchunkptr p; | |
1810 | #endif | |
1811 | { | |
1812 | INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE; | |
1813 | mchunkptr next = chunk_at_offset(p, sz); | |
1814 | ||
1815 | do_check_chunk(ar_ptr, p); | |
1816 | ||
1817 | /* Check whether it claims to be free ... */ | |
1818 | assert(!inuse(p)); | |
1819 | ||
1820 | /* Unless a special marker, must have OK fields */ | |
1821 | if ((long)sz >= (long)MINSIZE) | |
1822 | { | |
1823 | assert((sz & MALLOC_ALIGN_MASK) == 0); | |
1824 | assert(aligned_OK(chunk2mem(p))); | |
1825 | /* ... matching footer field */ | |
1826 | assert(next->prev_size == sz); | |
1827 | /* ... and is fully consolidated */ | |
1828 | assert(prev_inuse(p)); | |
1829 | assert (next == top(ar_ptr) || inuse(next)); | |
1830 | ||
1831 | /* ... and has minimally sane links */ | |
1832 | assert(p->fd->bk == p); | |
1833 | assert(p->bk->fd == p); | |
1834 | } | |
1835 | else /* markers are always of size SIZE_SZ */ | |
1836 | assert(sz == SIZE_SZ); | |
1837 | } | |
1838 | ||
1839 | #if __STD_C | |
1840 | static void do_check_inuse_chunk(arena *ar_ptr, mchunkptr p) | |
1841 | #else | |
1842 | static void do_check_inuse_chunk(ar_ptr, p) arena *ar_ptr; mchunkptr p; | |
1843 | #endif | |
1844 | { | |
1845 | mchunkptr next = next_chunk(p); | |
1846 | do_check_chunk(ar_ptr, p); | |
1847 | ||
1848 | /* Check whether it claims to be in use ... */ | |
1849 | assert(inuse(p)); | |
1850 | ||
1851 | /* ... and is surrounded by OK chunks. | |
1852 | Since more things can be checked with free chunks than inuse ones, | |
1853 | if an inuse chunk borders them and debug is on, it's worth doing them. | |
1854 | */ | |
1855 | if (!prev_inuse(p)) | |
1856 | { | |
1857 | mchunkptr prv = prev_chunk(p); | |
1858 | assert(next_chunk(prv) == p); | |
1859 | do_check_free_chunk(ar_ptr, prv); | |
1860 | } | |
1861 | if (next == top(ar_ptr)) | |
1862 | { | |
1863 | assert(prev_inuse(next)); | |
1864 | assert(chunksize(next) >= MINSIZE); | |
1865 | } | |
1866 | else if (!inuse(next)) | |
1867 | do_check_free_chunk(ar_ptr, next); | |
1868 | ||
1869 | } | |
1870 | ||
1871 | #if __STD_C | |
1872 | static void do_check_malloced_chunk(arena *ar_ptr, | |
1873 | mchunkptr p, INTERNAL_SIZE_T s) | |
1874 | #else | |
1875 | static void do_check_malloced_chunk(ar_ptr, p, s) | |
1876 | arena *ar_ptr; mchunkptr p; INTERNAL_SIZE_T s; | |
1877 | #endif | |
1878 | { | |
1879 | INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE; | |
1880 | long room = sz - s; | |
1881 | ||
1882 | do_check_inuse_chunk(ar_ptr, p); | |
1883 | ||
1884 | /* Legal size ... */ | |
1885 | assert((long)sz >= (long)MINSIZE); | |
1886 | assert((sz & MALLOC_ALIGN_MASK) == 0); | |
1887 | assert(room >= 0); | |
1888 | assert(room < (long)MINSIZE); | |
1889 | ||
1890 | /* ... and alignment */ | |
1891 | assert(aligned_OK(chunk2mem(p))); | |
1892 | ||
1893 | ||
1894 | /* ... and was allocated at front of an available chunk */ | |
1895 | assert(prev_inuse(p)); | |
1896 | ||
1897 | } | |
1898 | ||
1899 | ||
1900 | #define check_free_chunk(A,P) do_check_free_chunk(A,P) | |
1901 | #define check_inuse_chunk(A,P) do_check_inuse_chunk(A,P) | |
1902 | #define check_chunk(A,P) do_check_chunk(A,P) | |
1903 | #define check_malloced_chunk(A,P,N) do_check_malloced_chunk(A,P,N) | |
1904 | #else | |
1905 | #define check_free_chunk(A,P) | |
1906 | #define check_inuse_chunk(A,P) | |
1907 | #define check_chunk(A,P) | |
1908 | #define check_malloced_chunk(A,P,N) | |
1909 | #endif | |
1910 | ||
1911 | \f | |
1912 | ||
1913 | /* | |
1914 | Macro-based internal utilities | |
1915 | */ | |
1916 | ||
1917 | ||
1918 | /* | |
1919 | Linking chunks in bin lists. | |
1920 | Call these only with variables, not arbitrary expressions, as arguments. | |
1921 | */ | |
1922 | ||
1923 | /* | |
1924 | Place chunk p of size s in its bin, in size order, | |
1925 | putting it ahead of others of same size. | |
1926 | */ | |
1927 | ||
1928 | ||
1929 | #define frontlink(A, P, S, IDX, BK, FD) \ | |
1930 | { \ | |
1931 | if (S < MAX_SMALLBIN_SIZE) \ | |
1932 | { \ | |
1933 | IDX = smallbin_index(S); \ | |
1934 | mark_binblock(A, IDX); \ | |
1935 | BK = bin_at(A, IDX); \ | |
1936 | FD = BK->fd; \ | |
1937 | P->bk = BK; \ | |
1938 | P->fd = FD; \ | |
1939 | FD->bk = BK->fd = P; \ | |
1940 | } \ | |
1941 | else \ | |
1942 | { \ | |
1943 | IDX = bin_index(S); \ | |
1944 | BK = bin_at(A, IDX); \ | |
1945 | FD = BK->fd; \ | |
1946 | if (FD == BK) mark_binblock(A, IDX); \ | |
1947 | else \ | |
1948 | { \ | |
1949 | while (FD != BK && S < chunksize(FD)) FD = FD->fd; \ | |
1950 | BK = FD->bk; \ | |
1951 | } \ | |
1952 | P->bk = BK; \ | |
1953 | P->fd = FD; \ | |
1954 | FD->bk = BK->fd = P; \ | |
1955 | } \ | |
1956 | } | |
1957 | ||
1958 | ||
1959 | /* take a chunk off a list */ | |
1960 | ||
1961 | #define unlink(P, BK, FD) \ | |
1962 | { \ | |
1963 | BK = P->bk; \ | |
1964 | FD = P->fd; \ | |
1965 | FD->bk = BK; \ | |
1966 | BK->fd = FD; \ | |
1967 | } \ | |
1968 | ||
1969 | /* Place p as the last remainder */ | |
1970 | ||
1971 | #define link_last_remainder(A, P) \ | |
1972 | { \ | |
1973 | last_remainder(A)->fd = last_remainder(A)->bk = P; \ | |
1974 | P->fd = P->bk = last_remainder(A); \ | |
1975 | } | |
1976 | ||
1977 | /* Clear the last_remainder bin */ | |
1978 | ||
1979 | #define clear_last_remainder(A) \ | |
1980 | (last_remainder(A)->fd = last_remainder(A)->bk = last_remainder(A)) | |
1981 | ||
1982 | ||
1983 | ||
1984 | \f | |
1985 | ||
1986 | /* | |
1987 | Extend the top-most chunk by obtaining memory from system. | |
1988 | Main interface to sbrk (but see also malloc_trim). | |
1989 | */ | |
1990 | ||
1991 | #if __STD_C | |
1992 | static void malloc_extend_top(arena *ar_ptr, INTERNAL_SIZE_T nb) | |
1993 | #else | |
1994 | static void malloc_extend_top(ar_ptr, nb) arena *ar_ptr; INTERNAL_SIZE_T nb; | |
1995 | #endif | |
1996 | { | |
1997 | unsigned long pagesz = malloc_getpagesize; | |
1998 | mchunkptr old_top = top(ar_ptr); /* Record state of old top */ | |
1999 | INTERNAL_SIZE_T old_top_size = chunksize(old_top); | |
2000 | INTERNAL_SIZE_T top_size; /* new size of top chunk */ | |
2001 | ||
2002 | #ifndef NO_THREADS | |
2003 | if(ar_ptr == &main_arena) { | |
2004 | #endif | |
2005 | ||
2006 | char* brk; /* return value from sbrk */ | |
2007 | INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of sbrked space */ | |
2008 | INTERNAL_SIZE_T correction; /* bytes for 2nd sbrk call */ | |
2009 | char* new_brk; /* return of 2nd sbrk call */ | |
2010 | char* old_end = (char*)(chunk_at_offset(old_top, old_top_size)); | |
2011 | ||
2012 | /* Pad request with top_pad plus minimal overhead */ | |
2013 | INTERNAL_SIZE_T sbrk_size = nb + top_pad + MINSIZE; | |
2014 | ||
2015 | /* If not the first time through, round to preserve page boundary */ | |
2016 | /* Otherwise, we need to correct to a page size below anyway. */ | |
2017 | /* (We also correct below if an intervening foreign sbrk call.) */ | |
2018 | ||
2019 | if (sbrk_base != (char*)(-1)) | |
2020 | sbrk_size = (sbrk_size + (pagesz - 1)) & ~(pagesz - 1); | |
2021 | ||
2022 | brk = (char*)(MORECORE (sbrk_size)); | |
2023 | ||
2024 | /* Fail if sbrk failed or if a foreign sbrk call killed our space */ | |
2025 | if (brk == (char*)(MORECORE_FAILURE) || | |
2026 | (brk < old_end && old_top != initial_top(&main_arena))) | |
2027 | return; | |
2028 | ||
2029 | sbrked_mem += sbrk_size; | |
2030 | ||
2031 | if (brk == old_end) { /* can just add bytes to current top */ | |
2032 | top_size = sbrk_size + old_top_size; | |
2033 | set_head(old_top, top_size | PREV_INUSE); | |
2034 | old_top = 0; /* don't free below */ | |
2035 | } else { | |
2036 | if (sbrk_base == (char*)(-1)) /* First time through. Record base */ | |
2037 | sbrk_base = brk; | |
2038 | else | |
2039 | /* Someone else called sbrk(). Count those bytes as sbrked_mem. */ | |
2040 | sbrked_mem += brk - (char*)old_end; | |
2041 | ||
2042 | /* Guarantee alignment of first new chunk made from this space */ | |
2043 | front_misalign = (unsigned long)chunk2mem(brk) & MALLOC_ALIGN_MASK; | |
2044 | if (front_misalign > 0) { | |
2045 | correction = (MALLOC_ALIGNMENT) - front_misalign; | |
2046 | brk += correction; | |
2047 | } else | |
2048 | correction = 0; | |
2049 | ||
2050 | /* Guarantee the next brk will be at a page boundary */ | |
2051 | correction += pagesz - ((unsigned long)(brk + sbrk_size) & (pagesz - 1)); | |
2052 | ||
2053 | /* Allocate correction */ | |
2054 | new_brk = (char*)(MORECORE (correction)); | |
2055 | if (new_brk == (char*)(MORECORE_FAILURE)) return; | |
2056 | ||
2057 | sbrked_mem += correction; | |
2058 | ||
2059 | top(&main_arena) = (mchunkptr)brk; | |
2060 | top_size = new_brk - brk + correction; | |
2061 | set_head(top(&main_arena), top_size | PREV_INUSE); | |
2062 | ||
2063 | if (old_top == initial_top(&main_arena)) | |
2064 | old_top = 0; /* don't free below */ | |
2065 | } | |
2066 | ||
2067 | if ((unsigned long)sbrked_mem > (unsigned long)max_sbrked_mem) | |
2068 | max_sbrked_mem = sbrked_mem; | |
2069 | if ((unsigned long)(mmapped_mem + sbrked_mem) > | |
2070 | (unsigned long)max_total_mem) | |
2071 | max_total_mem = mmapped_mem + sbrked_mem; | |
2072 | ||
2073 | #ifndef NO_THREADS | |
2074 | } else { /* ar_ptr != &main_arena */ | |
2075 | ||
2076 | heap_info *heap; | |
2077 | ||
2078 | if(old_top_size < MINSIZE) /* this should never happen */ | |
2079 | return; | |
2080 | ||
2081 | /* First try to extend the current heap. */ | |
2082 | if(MINSIZE + nb <= old_top_size) | |
2083 | return; | |
2084 | heap = heap_for_ptr(old_top); | |
2085 | if(grow_heap(heap, MINSIZE + nb - old_top_size) == 0) { | |
2086 | top_size = heap->size - ((char *)old_top - (char *)heap); | |
2087 | set_head(old_top, top_size | PREV_INUSE); | |
2088 | return; | |
2089 | } | |
2090 | ||
2091 | /* A new heap must be created. */ | |
2092 | heap = new_heap(nb + top_pad + (MINSIZE + sizeof(*heap))); | |
2093 | if(!heap) | |
2094 | return; | |
2095 | heap->ar_ptr = ar_ptr; | |
2096 | ||
2097 | /* Set up the new top, so we can safely use chunk_free() below. */ | |
2098 | top(ar_ptr) = chunk_at_offset(heap, sizeof(*heap)); | |
2099 | top_size = heap->size - sizeof(*heap); | |
2100 | set_head(top(ar_ptr), top_size | PREV_INUSE); | |
2101 | } | |
2102 | #endif /* !defined(NO_THREADS) */ | |
2103 | ||
2104 | /* We always land on a page boundary */ | |
2105 | assert(((unsigned long)((char*)top(ar_ptr) + top_size) & (pagesz-1)) == 0); | |
2106 | ||
2107 | /* Setup fencepost and free the old top chunk. */ | |
2108 | if(old_top) { | |
2109 | /* Keep size a multiple of MALLOC_ALIGNMENT. */ | |
2110 | old_top_size = (old_top_size - 3*SIZE_SZ) & ~MALLOC_ALIGN_MASK; | |
2111 | /* If possible, release the rest. */ | |
2112 | if (old_top_size >= MINSIZE) { | |
2113 | set_head(chunk_at_offset(old_top, old_top_size ), | |
2114 | SIZE_SZ|PREV_INUSE); | |
2115 | set_head(chunk_at_offset(old_top, old_top_size+SIZE_SZ), | |
2116 | SIZE_SZ|PREV_INUSE); | |
2117 | set_head_size(old_top, old_top_size); | |
2118 | chunk_free(ar_ptr, old_top); | |
2119 | } else { | |
2120 | set_head(old_top, SIZE_SZ|PREV_INUSE); | |
2121 | set_head(chunk_at_offset(old_top, SIZE_SZ), SIZE_SZ|PREV_INUSE); | |
2122 | } | |
2123 | } | |
2124 | } | |
2125 | ||
2126 | ||
2127 | \f | |
2128 | ||
2129 | /* Main public routines */ | |
2130 | ||
2131 | ||
2132 | /* | |
2133 | Malloc Algorthim: | |
2134 | ||
2135 | The requested size is first converted into a usable form, `nb'. | |
2136 | This currently means to add 4 bytes overhead plus possibly more to | |
2137 | obtain 8-byte alignment and/or to obtain a size of at least | |
2138 | MINSIZE (currently 16 bytes), the smallest allocatable size. | |
2139 | (All fits are considered `exact' if they are within MINSIZE bytes.) | |
2140 | ||
2141 | From there, the first successful of the following steps is taken: | |
2142 | ||
2143 | 1. The bin corresponding to the request size is scanned, and if | |
2144 | a chunk of exactly the right size is found, it is taken. | |
2145 | ||
2146 | 2. The most recently remaindered chunk is used if it is big | |
2147 | enough. This is a form of (roving) first fit, used only in | |
2148 | the absence of exact fits. Runs of consecutive requests use | |
2149 | the remainder of the chunk used for the previous such request | |
2150 | whenever possible. This limited use of a first-fit style | |
2151 | allocation strategy tends to give contiguous chunks | |
2152 | coextensive lifetimes, which improves locality and can reduce | |
2153 | fragmentation in the long run. | |
2154 | ||
2155 | 3. Other bins are scanned in increasing size order, using a | |
2156 | chunk big enough to fulfill the request, and splitting off | |
2157 | any remainder. This search is strictly by best-fit; i.e., | |
2158 | the smallest (with ties going to approximately the least | |
2159 | recently used) chunk that fits is selected. | |
2160 | ||
2161 | 4. If large enough, the chunk bordering the end of memory | |
2162 | (`top') is split off. (This use of `top' is in accord with | |
2163 | the best-fit search rule. In effect, `top' is treated as | |
2164 | larger (and thus less well fitting) than any other available | |
2165 | chunk since it can be extended to be as large as necessary | |
2166 | (up to system limitations). | |
2167 | ||
2168 | 5. If the request size meets the mmap threshold and the | |
2169 | system supports mmap, and there are few enough currently | |
2170 | allocated mmapped regions, and a call to mmap succeeds, | |
2171 | the request is allocated via direct memory mapping. | |
2172 | ||
2173 | 6. Otherwise, the top of memory is extended by | |
2174 | obtaining more space from the system (normally using sbrk, | |
2175 | but definable to anything else via the MORECORE macro). | |
2176 | Memory is gathered from the system (in system page-sized | |
2177 | units) in a way that allows chunks obtained across different | |
2178 | sbrk calls to be consolidated, but does not require | |
2179 | contiguous memory. Thus, it should be safe to intersperse | |
2180 | mallocs with other sbrk calls. | |
2181 | ||
2182 | ||
2183 | All allocations are made from the the `lowest' part of any found | |
2184 | chunk. (The implementation invariant is that prev_inuse is | |
2185 | always true of any allocated chunk; i.e., that each allocated | |
2186 | chunk borders either a previously allocated and still in-use chunk, | |
2187 | or the base of its memory arena.) | |
2188 | ||
2189 | */ | |
2190 | ||
2191 | #if __STD_C | |
2192 | Void_t* mALLOc(size_t bytes) | |
2193 | #else | |
2194 | Void_t* mALLOc(bytes) size_t bytes; | |
2195 | #endif | |
2196 | { | |
2197 | arena *ar_ptr; | |
2198 | INTERNAL_SIZE_T nb = request2size(bytes); /* padded request size; */ | |
2199 | mchunkptr victim; | |
2200 | ||
2201 | arena_get(ar_ptr, nb + top_pad); | |
2202 | if(!ar_ptr) | |
2203 | return 0; | |
2204 | victim = chunk_alloc(ar_ptr, nb); | |
2205 | (void)mutex_unlock(&ar_ptr->mutex); | |
2206 | return victim ? chunk2mem(victim) : 0; | |
2207 | } | |
2208 | ||
2209 | static mchunkptr | |
2210 | #if __STD_C | |
2211 | chunk_alloc(arena *ar_ptr, INTERNAL_SIZE_T nb) | |
2212 | #else | |
2213 | chunk_alloc(ar_ptr, nb) arena *ar_ptr; INTERNAL_SIZE_T nb; | |
2214 | #endif | |
2215 | { | |
2216 | mchunkptr victim; /* inspected/selected chunk */ | |
2217 | INTERNAL_SIZE_T victim_size; /* its size */ | |
2218 | int idx; /* index for bin traversal */ | |
2219 | mbinptr bin; /* associated bin */ | |
2220 | mchunkptr remainder; /* remainder from a split */ | |
2221 | long remainder_size; /* its size */ | |
2222 | int remainder_index; /* its bin index */ | |
2223 | unsigned long block; /* block traverser bit */ | |
2224 | int startidx; /* first bin of a traversed block */ | |
2225 | mchunkptr fwd; /* misc temp for linking */ | |
2226 | mchunkptr bck; /* misc temp for linking */ | |
2227 | mbinptr q; /* misc temp */ | |
2228 | ||
2229 | ||
2230 | /* Check for exact match in a bin */ | |
2231 | ||
2232 | if (is_small_request(nb)) /* Faster version for small requests */ | |
2233 | { | |
2234 | idx = smallbin_index(nb); | |
2235 | ||
2236 | /* No traversal or size check necessary for small bins. */ | |
2237 | ||
2238 | q = bin_at(ar_ptr, idx); | |
2239 | victim = last(q); | |
2240 | ||
2241 | /* Also scan the next one, since it would have a remainder < MINSIZE */ | |
2242 | if (victim == q) | |
2243 | { | |
2244 | q = next_bin(q); | |
2245 | victim = last(q); | |
2246 | } | |
2247 | if (victim != q) | |
2248 | { | |
2249 | victim_size = chunksize(victim); | |
2250 | unlink(victim, bck, fwd); | |
2251 | set_inuse_bit_at_offset(victim, victim_size); | |
2252 | check_malloced_chunk(ar_ptr, victim, nb); | |
2253 | return victim; | |
2254 | } | |
2255 | ||
2256 | idx += 2; /* Set for bin scan below. We've already scanned 2 bins. */ | |
2257 | ||
2258 | } | |
2259 | else | |
2260 | { | |
2261 | idx = bin_index(nb); | |
2262 | bin = bin_at(ar_ptr, idx); | |
2263 | ||
2264 | for (victim = last(bin); victim != bin; victim = victim->bk) | |
2265 | { | |
2266 | victim_size = chunksize(victim); | |
2267 | remainder_size = victim_size - nb; | |
2268 | ||
2269 | if (remainder_size >= (long)MINSIZE) /* too big */ | |
2270 | { | |
2271 | --idx; /* adjust to rescan below after checking last remainder */ | |
2272 | break; | |
2273 | } | |
2274 | ||
2275 | else if (remainder_size >= 0) /* exact fit */ | |
2276 | { | |
2277 | unlink(victim, bck, fwd); | |
2278 | set_inuse_bit_at_offset(victim, victim_size); | |
2279 | check_malloced_chunk(ar_ptr, victim, nb); | |
2280 | return victim; | |
2281 | } | |
2282 | } | |
2283 | ||
2284 | ++idx; | |
2285 | ||
2286 | } | |
2287 | ||
2288 | /* Try to use the last split-off remainder */ | |
2289 | ||
2290 | if ( (victim = last_remainder(ar_ptr)->fd) != last_remainder(ar_ptr)) | |
2291 | { | |
2292 | victim_size = chunksize(victim); | |
2293 | remainder_size = victim_size - nb; | |
2294 | ||
2295 | if (remainder_size >= (long)MINSIZE) /* re-split */ | |
2296 | { | |
2297 | remainder = chunk_at_offset(victim, nb); | |
2298 | set_head(victim, nb | PREV_INUSE); | |
2299 | link_last_remainder(ar_ptr, remainder); | |
2300 | set_head(remainder, remainder_size | PREV_INUSE); | |
2301 | set_foot(remainder, remainder_size); | |
2302 | check_malloced_chunk(ar_ptr, victim, nb); | |
2303 | return victim; | |
2304 | } | |
2305 | ||
2306 | clear_last_remainder(ar_ptr); | |
2307 | ||
2308 | if (remainder_size >= 0) /* exhaust */ | |
2309 | { | |
2310 | set_inuse_bit_at_offset(victim, victim_size); | |
2311 | check_malloced_chunk(ar_ptr, victim, nb); | |
2312 | return victim; | |
2313 | } | |
2314 | ||
2315 | /* Else place in bin */ | |
2316 | ||
2317 | frontlink(ar_ptr, victim, victim_size, remainder_index, bck, fwd); | |
2318 | } | |
2319 | ||
2320 | /* | |
2321 | If there are any possibly nonempty big-enough blocks, | |
2322 | search for best fitting chunk by scanning bins in blockwidth units. | |
2323 | */ | |
2324 | ||
2325 | if ( (block = idx2binblock(idx)) <= binblocks(ar_ptr)) | |
2326 | { | |
2327 | ||
2328 | /* Get to the first marked block */ | |
2329 | ||
2330 | if ( (block & binblocks(ar_ptr)) == 0) | |
2331 | { | |
2332 | /* force to an even block boundary */ | |
2333 | idx = (idx & ~(BINBLOCKWIDTH - 1)) + BINBLOCKWIDTH; | |
2334 | block <<= 1; | |
2335 | while ((block & binblocks(ar_ptr)) == 0) | |
2336 | { | |
2337 | idx += BINBLOCKWIDTH; | |
2338 | block <<= 1; | |
2339 | } | |
2340 | } | |
2341 | ||
2342 | /* For each possibly nonempty block ... */ | |
2343 | for (;;) | |
2344 | { | |
2345 | startidx = idx; /* (track incomplete blocks) */ | |
2346 | q = bin = bin_at(ar_ptr, idx); | |
2347 | ||
2348 | /* For each bin in this block ... */ | |
2349 | do | |
2350 | { | |
2351 | /* Find and use first big enough chunk ... */ | |
2352 | ||
2353 | for (victim = last(bin); victim != bin; victim = victim->bk) | |
2354 | { | |
2355 | victim_size = chunksize(victim); | |
2356 | remainder_size = victim_size - nb; | |
2357 | ||
2358 | if (remainder_size >= (long)MINSIZE) /* split */ | |
2359 | { | |
2360 | remainder = chunk_at_offset(victim, nb); | |
2361 | set_head(victim, nb | PREV_INUSE); | |
2362 | unlink(victim, bck, fwd); | |
2363 | link_last_remainder(ar_ptr, remainder); | |
2364 | set_head(remainder, remainder_size | PREV_INUSE); | |
2365 | set_foot(remainder, remainder_size); | |
2366 | check_malloced_chunk(ar_ptr, victim, nb); | |
2367 | return victim; | |
2368 | } | |
2369 | ||
2370 | else if (remainder_size >= 0) /* take */ | |
2371 | { | |
2372 | set_inuse_bit_at_offset(victim, victim_size); | |
2373 | unlink(victim, bck, fwd); | |
2374 | check_malloced_chunk(ar_ptr, victim, nb); | |
2375 | return victim; | |
2376 | } | |
2377 | ||
2378 | } | |
2379 | ||
2380 | bin = next_bin(bin); | |
2381 | ||
2382 | } while ((++idx & (BINBLOCKWIDTH - 1)) != 0); | |
2383 | ||
2384 | /* Clear out the block bit. */ | |
2385 | ||
2386 | do /* Possibly backtrack to try to clear a partial block */ | |
2387 | { | |
2388 | if ((startidx & (BINBLOCKWIDTH - 1)) == 0) | |
2389 | { | |
2390 | binblocks(ar_ptr) &= ~block; | |
2391 | break; | |
2392 | } | |
2393 | --startidx; | |
2394 | q = prev_bin(q); | |
2395 | } while (first(q) == q); | |
2396 | ||
2397 | /* Get to the next possibly nonempty block */ | |
2398 | ||
2399 | if ( (block <<= 1) <= binblocks(ar_ptr) && (block != 0) ) | |
2400 | { | |
2401 | while ((block & binblocks(ar_ptr)) == 0) | |
2402 | { | |
2403 | idx += BINBLOCKWIDTH; | |
2404 | block <<= 1; | |
2405 | } | |
2406 | } | |
2407 | else | |
2408 | break; | |
2409 | } | |
2410 | } | |
2411 | ||
2412 | ||
2413 | /* Try to use top chunk */ | |
2414 | ||
2415 | /* Require that there be a remainder, ensuring top always exists */ | |
2416 | if ( (remainder_size = chunksize(top(ar_ptr)) - nb) < (long)MINSIZE) | |
2417 | { | |
2418 | ||
2419 | #if HAVE_MMAP | |
2420 | /* If big and would otherwise need to extend, try to use mmap instead */ | |
2421 | if ((unsigned long)nb >= (unsigned long)mmap_threshold && | |
2422 | (victim = mmap_chunk(nb)) != 0) | |
2423 | return victim; | |
2424 | #endif | |
2425 | ||
2426 | /* Try to extend */ | |
2427 | malloc_extend_top(ar_ptr, nb); | |
2428 | if ((remainder_size = chunksize(top(ar_ptr)) - nb) < (long)MINSIZE) | |
2429 | return 0; /* propagate failure */ | |
2430 | } | |
2431 | ||
2432 | victim = top(ar_ptr); | |
2433 | set_head(victim, nb | PREV_INUSE); | |
2434 | top(ar_ptr) = chunk_at_offset(victim, nb); | |
2435 | set_head(top(ar_ptr), remainder_size | PREV_INUSE); | |
2436 | check_malloced_chunk(ar_ptr, victim, nb); | |
2437 | return victim; | |
2438 | ||
2439 | } | |
2440 | ||
2441 | ||
2442 | \f | |
2443 | ||
2444 | /* | |
2445 | ||
2446 | free() algorithm : | |
2447 | ||
2448 | cases: | |
2449 | ||
2450 | 1. free(0) has no effect. | |
2451 | ||
2452 | 2. If the chunk was allocated via mmap, it is released via munmap(). | |
2453 | ||
2454 | 3. If a returned chunk borders the current high end of memory, | |
2455 | it is consolidated into the top, and if the total unused | |
2456 | topmost memory exceeds the trim threshold, malloc_trim is | |
2457 | called. | |
2458 | ||
2459 | 4. Other chunks are consolidated as they arrive, and | |
2460 | placed in corresponding bins. (This includes the case of | |
2461 | consolidating with the current `last_remainder'). | |
2462 | ||
2463 | */ | |
2464 | ||
2465 | ||
2466 | #if __STD_C | |
2467 | void fREe(Void_t* mem) | |
2468 | #else | |
2469 | void fREe(mem) Void_t* mem; | |
2470 | #endif | |
2471 | { | |
2472 | arena *ar_ptr; | |
2473 | mchunkptr p; /* chunk corresponding to mem */ | |
2474 | ||
2475 | if (mem == 0) /* free(0) has no effect */ | |
2476 | return; | |
2477 | ||
2478 | p = mem2chunk(mem); | |
2479 | ||
2480 | #if HAVE_MMAP | |
2481 | if (chunk_is_mmapped(p)) /* release mmapped memory. */ | |
2482 | { | |
2483 | munmap_chunk(p); | |
2484 | return; | |
2485 | } | |
2486 | #endif | |
2487 | ||
2488 | ar_ptr = arena_for_ptr(p); | |
2489 | (void)mutex_lock(&ar_ptr->mutex); | |
2490 | chunk_free(ar_ptr, p); | |
2491 | (void)mutex_unlock(&ar_ptr->mutex); | |
2492 | } | |
2493 | ||
2494 | static void | |
2495 | #if __STD_C | |
2496 | chunk_free(arena *ar_ptr, mchunkptr p) | |
2497 | #else | |
2498 | chunk_free(ar_ptr, p) arena *ar_ptr; mchunkptr p; | |
2499 | #endif | |
2500 | { | |
2501 | INTERNAL_SIZE_T hd = p->size; /* its head field */ | |
2502 | INTERNAL_SIZE_T sz; /* its size */ | |
2503 | int idx; /* its bin index */ | |
2504 | mchunkptr next; /* next contiguous chunk */ | |
2505 | INTERNAL_SIZE_T nextsz; /* its size */ | |
2506 | INTERNAL_SIZE_T prevsz; /* size of previous contiguous chunk */ | |
2507 | mchunkptr bck; /* misc temp for linking */ | |
2508 | mchunkptr fwd; /* misc temp for linking */ | |
2509 | int islr; /* track whether merging with last_remainder */ | |
2510 | ||
2511 | check_inuse_chunk(ar_ptr, p); | |
2512 | ||
2513 | sz = hd & ~PREV_INUSE; | |
2514 | next = chunk_at_offset(p, sz); | |
2515 | nextsz = chunksize(next); | |
2516 | ||
2517 | if (next == top(ar_ptr)) /* merge with top */ | |
2518 | { | |
2519 | sz += nextsz; | |
2520 | ||
2521 | if (!(hd & PREV_INUSE)) /* consolidate backward */ | |
2522 | { | |
2523 | prevsz = p->prev_size; | |
2524 | p = chunk_at_offset(p, -prevsz); | |
2525 | sz += prevsz; | |
2526 | unlink(p, bck, fwd); | |
2527 | } | |
2528 | ||
2529 | set_head(p, sz | PREV_INUSE); | |
2530 | top(ar_ptr) = p; | |
2531 | if ((unsigned long)(sz) >= (unsigned long)trim_threshold) | |
2532 | arena_trim(ar_ptr, top_pad); | |
2533 | return; | |
2534 | } | |
2535 | ||
2536 | set_head(next, nextsz); /* clear inuse bit */ | |
2537 | ||
2538 | islr = 0; | |
2539 | ||
2540 | if (!(hd & PREV_INUSE)) /* consolidate backward */ | |
2541 | { | |
2542 | prevsz = p->prev_size; | |
2543 | p = chunk_at_offset(p, -prevsz); | |
2544 | sz += prevsz; | |
2545 | ||
2546 | if (p->fd == last_remainder(ar_ptr)) /* keep as last_remainder */ | |
2547 | islr = 1; | |
2548 | else | |
2549 | unlink(p, bck, fwd); | |
2550 | } | |
2551 | ||
2552 | if (!(inuse_bit_at_offset(next, nextsz))) /* consolidate forward */ | |
2553 | { | |
2554 | sz += nextsz; | |
2555 | ||
2556 | if (!islr && next->fd == last_remainder(ar_ptr)) | |
2557 | /* re-insert last_remainder */ | |
2558 | { | |
2559 | islr = 1; | |
2560 | link_last_remainder(ar_ptr, p); | |
2561 | } | |
2562 | else | |
2563 | unlink(next, bck, fwd); | |
2564 | } | |
2565 | ||
2566 | set_head(p, sz | PREV_INUSE); | |
2567 | set_foot(p, sz); | |
2568 | if (!islr) | |
2569 | frontlink(ar_ptr, p, sz, idx, bck, fwd); | |
2570 | } | |
2571 | ||
2572 | ||
2573 | \f | |
2574 | ||
2575 | ||
2576 | /* | |
2577 | ||
2578 | Realloc algorithm: | |
2579 | ||
2580 | Chunks that were obtained via mmap cannot be extended or shrunk | |
2581 | unless HAVE_MREMAP is defined, in which case mremap is used. | |
2582 | Otherwise, if their reallocation is for additional space, they are | |
2583 | copied. If for less, they are just left alone. | |
2584 | ||
2585 | Otherwise, if the reallocation is for additional space, and the | |
2586 | chunk can be extended, it is, else a malloc-copy-free sequence is | |
2587 | taken. There are several different ways that a chunk could be | |
2588 | extended. All are tried: | |
2589 | ||
2590 | * Extending forward into following adjacent free chunk. | |
2591 | * Shifting backwards, joining preceding adjacent space | |
2592 | * Both shifting backwards and extending forward. | |
2593 | * Extending into newly sbrked space | |
2594 | ||
2595 | Unless the #define REALLOC_ZERO_BYTES_FREES is set, realloc with a | |
2596 | size argument of zero (re)allocates a minimum-sized chunk. | |
2597 | ||
2598 | If the reallocation is for less space, and the new request is for | |
2599 | a `small' (<512 bytes) size, then the newly unused space is lopped | |
2600 | off and freed. | |
2601 | ||
2602 | The old unix realloc convention of allowing the last-free'd chunk | |
2603 | to be used as an argument to realloc is no longer supported. | |
2604 | I don't know of any programs still relying on this feature, | |
2605 | and allowing it would also allow too many other incorrect | |
2606 | usages of realloc to be sensible. | |
2607 | ||
2608 | ||
2609 | */ | |
2610 | ||
2611 | ||
2612 | #if __STD_C | |
2613 | Void_t* rEALLOc(Void_t* oldmem, size_t bytes) | |
2614 | #else | |
2615 | Void_t* rEALLOc(oldmem, bytes) Void_t* oldmem; size_t bytes; | |
2616 | #endif | |
2617 | { | |
2618 | arena *ar_ptr; | |
2619 | INTERNAL_SIZE_T nb; /* padded request size */ | |
2620 | ||
2621 | mchunkptr oldp; /* chunk corresponding to oldmem */ | |
2622 | INTERNAL_SIZE_T oldsize; /* its size */ | |
2623 | ||
2624 | mchunkptr newp; /* chunk to return */ | |
2625 | INTERNAL_SIZE_T newsize; /* its size */ | |
2626 | Void_t* newmem; /* corresponding user mem */ | |
2627 | ||
2628 | mchunkptr next; /* next contiguous chunk after oldp */ | |
2629 | INTERNAL_SIZE_T nextsize; /* its size */ | |
2630 | ||
2631 | mchunkptr prev; /* previous contiguous chunk before oldp */ | |
2632 | INTERNAL_SIZE_T prevsize; /* its size */ | |
2633 | ||
2634 | mchunkptr remainder; /* holds split off extra space from newp */ | |
2635 | INTERNAL_SIZE_T remainder_size; /* its size */ | |
2636 | ||
2637 | mchunkptr bck; /* misc temp for linking */ | |
2638 | mchunkptr fwd; /* misc temp for linking */ | |
2639 | ||
2640 | #ifdef REALLOC_ZERO_BYTES_FREES | |
2641 | if (bytes == 0) { fREe(oldmem); return 0; } | |
2642 | #endif | |
2643 | ||
2644 | ||
2645 | /* realloc of null is supposed to be same as malloc */ | |
2646 | if (oldmem == 0) return mALLOc(bytes); | |
2647 | ||
2648 | newp = oldp = mem2chunk(oldmem); | |
2649 | newsize = oldsize = chunksize(oldp); | |
2650 | ||
2651 | ||
2652 | nb = request2size(bytes); | |
2653 | ||
2654 | #if HAVE_MMAP | |
2655 | if (chunk_is_mmapped(oldp)) | |
2656 | { | |
2657 | #if HAVE_MREMAP | |
2658 | newp = mremap_chunk(oldp, nb); | |
2659 | if(newp) return chunk2mem(newp); | |
2660 | #endif | |
2661 | /* Note the extra SIZE_SZ overhead. */ | |
2662 | if(oldsize - SIZE_SZ >= nb) return oldmem; /* do nothing */ | |
2663 | /* Must alloc, copy, free. */ | |
2664 | newmem = mALLOc(bytes); | |
2665 | if (newmem == 0) return 0; /* propagate failure */ | |
2666 | MALLOC_COPY(newmem, oldmem, oldsize - 2*SIZE_SZ); | |
2667 | munmap_chunk(oldp); | |
2668 | return newmem; | |
2669 | } | |
2670 | #endif | |
2671 | ||
2672 | ar_ptr = arena_for_ptr(oldp); | |
2673 | (void)mutex_lock(&ar_ptr->mutex); | |
2674 | /* As in malloc(), remember this arena for the next allocation. */ | |
2675 | tsd_setspecific(arena_key, (Void_t *)ar_ptr); | |
2676 | ||
2677 | check_inuse_chunk(ar_ptr, oldp); | |
2678 | ||
2679 | if ((long)(oldsize) < (long)(nb)) | |
2680 | { | |
2681 | ||
2682 | /* Try expanding forward */ | |
2683 | ||
2684 | next = chunk_at_offset(oldp, oldsize); | |
2685 | if (next == top(ar_ptr) || !inuse(next)) | |
2686 | { | |
2687 | nextsize = chunksize(next); | |
2688 | ||
2689 | /* Forward into top only if a remainder */ | |
2690 | if (next == top(ar_ptr)) | |
2691 | { | |
2692 | if ((long)(nextsize + newsize) >= (long)(nb + MINSIZE)) | |
2693 | { | |
2694 | newsize += nextsize; | |
2695 | top(ar_ptr) = chunk_at_offset(oldp, nb); | |
2696 | set_head(top(ar_ptr), (newsize - nb) | PREV_INUSE); | |
2697 | set_head_size(oldp, nb); | |
2698 | (void)mutex_unlock(&ar_ptr->mutex); | |
2699 | return chunk2mem(oldp); | |
2700 | } | |
2701 | } | |
2702 | ||
2703 | /* Forward into next chunk */ | |
2704 | else if (((long)(nextsize + newsize) >= (long)(nb))) | |
2705 | { | |
2706 | unlink(next, bck, fwd); | |
2707 | newsize += nextsize; | |
2708 | goto split; | |
2709 | } | |
2710 | } | |
2711 | else | |
2712 | { | |
2713 | next = 0; | |
2714 | nextsize = 0; | |
2715 | } | |
2716 | ||
2717 | /* Try shifting backwards. */ | |
2718 | ||
2719 | if (!prev_inuse(oldp)) | |
2720 | { | |
2721 | prev = prev_chunk(oldp); | |
2722 | prevsize = chunksize(prev); | |
2723 | ||
2724 | /* try forward + backward first to save a later consolidation */ | |
2725 | ||
2726 | if (next != 0) | |
2727 | { | |
2728 | /* into top */ | |
2729 | if (next == top(ar_ptr)) | |
2730 | { | |
2731 | if ((long)(nextsize + prevsize + newsize) >= (long)(nb + MINSIZE)) | |
2732 | { | |
2733 | unlink(prev, bck, fwd); | |
2734 | newp = prev; | |
2735 | newsize += prevsize + nextsize; | |
2736 | newmem = chunk2mem(newp); | |
2737 | MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ); | |
2738 | top(ar_ptr) = chunk_at_offset(newp, nb); | |
2739 | set_head(top(ar_ptr), (newsize - nb) | PREV_INUSE); | |
2740 | set_head_size(newp, nb); | |
2741 | (void)mutex_unlock(&ar_ptr->mutex); | |
2742 | return newmem; | |
2743 | } | |
2744 | } | |
2745 | ||
2746 | /* into next chunk */ | |
2747 | else if (((long)(nextsize + prevsize + newsize) >= (long)(nb))) | |
2748 | { | |
2749 | unlink(next, bck, fwd); | |
2750 | unlink(prev, bck, fwd); | |
2751 | newp = prev; | |
2752 | newsize += nextsize + prevsize; | |
2753 | newmem = chunk2mem(newp); | |
2754 | MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ); | |
2755 | goto split; | |
2756 | } | |
2757 | } | |
2758 | ||
2759 | /* backward only */ | |
2760 | if (prev != 0 && (long)(prevsize + newsize) >= (long)nb) | |
2761 | { | |
2762 | unlink(prev, bck, fwd); | |
2763 | newp = prev; | |
2764 | newsize += prevsize; | |
2765 | newmem = chunk2mem(newp); | |
2766 | MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ); | |
2767 | goto split; | |
2768 | } | |
2769 | } | |
2770 | ||
2771 | /* Must allocate */ | |
2772 | ||
2773 | newp = chunk_alloc (ar_ptr, nb); | |
2774 | ||
2775 | if (newp == 0) /* propagate failure */ | |
2776 | return 0; | |
2777 | ||
2778 | /* Avoid copy if newp is next chunk after oldp. */ | |
2779 | /* (This can only happen when new chunk is sbrk'ed.) */ | |
2780 | ||
2781 | if ( newp == next_chunk(oldp)) | |
2782 | { | |
2783 | newsize += chunksize(newp); | |
2784 | newp = oldp; | |
2785 | goto split; | |
2786 | } | |
2787 | ||
2788 | /* Otherwise copy, free, and exit */ | |
2789 | newmem = chunk2mem(newp); | |
2790 | MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ); | |
2791 | chunk_free(ar_ptr, oldp); | |
2792 | (void)mutex_unlock(&ar_ptr->mutex); | |
2793 | return newmem; | |
2794 | } | |
2795 | ||
2796 | ||
2797 | split: /* split off extra room in old or expanded chunk */ | |
2798 | ||
2799 | if (newsize - nb >= MINSIZE) /* split off remainder */ | |
2800 | { | |
2801 | remainder = chunk_at_offset(newp, nb); | |
2802 | remainder_size = newsize - nb; | |
2803 | set_head_size(newp, nb); | |
2804 | set_head(remainder, remainder_size | PREV_INUSE); | |
2805 | set_inuse_bit_at_offset(remainder, remainder_size); | |
2806 | chunk_free(ar_ptr, remainder); | |
2807 | } | |
2808 | else | |
2809 | { | |
2810 | set_head_size(newp, newsize); | |
2811 | set_inuse_bit_at_offset(newp, newsize); | |
2812 | } | |
2813 | ||
2814 | check_inuse_chunk(ar_ptr, newp); | |
2815 | (void)mutex_unlock(&ar_ptr->mutex); | |
2816 | return chunk2mem(newp); | |
2817 | } | |
2818 | ||
2819 | ||
2820 | \f | |
2821 | ||
2822 | /* | |
2823 | ||
2824 | memalign algorithm: | |
2825 | ||
2826 | memalign requests more than enough space from malloc, finds a spot | |
2827 | within that chunk that meets the alignment request, and then | |
2828 | possibly frees the leading and trailing space. | |
2829 | ||
2830 | The alignment argument must be a power of two. This property is not | |
2831 | checked by memalign, so misuse may result in random runtime errors. | |
2832 | ||
2833 | 8-byte alignment is guaranteed by normal malloc calls, so don't | |
2834 | bother calling memalign with an argument of 8 or less. | |
2835 | ||
2836 | Overreliance on memalign is a sure way to fragment space. | |
2837 | ||
2838 | */ | |
2839 | ||
2840 | ||
2841 | #if __STD_C | |
2842 | Void_t* mEMALIGn(size_t alignment, size_t bytes) | |
2843 | #else | |
2844 | Void_t* mEMALIGn(alignment, bytes) size_t alignment; size_t bytes; | |
2845 | #endif | |
2846 | { | |
2847 | arena *ar_ptr; | |
2848 | INTERNAL_SIZE_T nb; /* padded request size */ | |
2849 | char* m; /* memory returned by malloc call */ | |
2850 | mchunkptr p; /* corresponding chunk */ | |
2851 | char* brk; /* alignment point within p */ | |
2852 | mchunkptr newp; /* chunk to return */ | |
2853 | INTERNAL_SIZE_T newsize; /* its size */ | |
2854 | INTERNAL_SIZE_T leadsize; /* leading space befor alignment point */ | |
2855 | mchunkptr remainder; /* spare room at end to split off */ | |
2856 | long remainder_size; /* its size */ | |
2857 | ||
2858 | /* If need less alignment than we give anyway, just relay to malloc */ | |
2859 | ||
2860 | if (alignment <= MALLOC_ALIGNMENT) return mALLOc(bytes); | |
2861 | ||
2862 | /* Otherwise, ensure that it is at least a minimum chunk size */ | |
2863 | ||
2864 | if (alignment < MINSIZE) alignment = MINSIZE; | |
2865 | ||
2866 | /* Call malloc with worst case padding to hit alignment. */ | |
2867 | ||
2868 | nb = request2size(bytes); | |
2869 | arena_get(ar_ptr, nb + alignment + MINSIZE); | |
2870 | if(!ar_ptr) | |
2871 | return 0; | |
2872 | p = chunk_alloc(ar_ptr, nb + alignment + MINSIZE); | |
2873 | ||
2874 | if (p == 0) { | |
2875 | (void)mutex_unlock(&ar_ptr->mutex); | |
2876 | return 0; /* propagate failure */ | |
2877 | } | |
2878 | ||
2879 | m = chunk2mem(p); | |
2880 | ||
2881 | if ((((unsigned long)(m)) % alignment) == 0) /* aligned */ | |
2882 | { | |
2883 | #if HAVE_MMAP | |
2884 | if(chunk_is_mmapped(p)) { | |
2885 | (void)mutex_unlock(&ar_ptr->mutex); | |
2886 | return chunk2mem(p); /* nothing more to do */ | |
2887 | } | |
2888 | #endif | |
2889 | } | |
2890 | else /* misaligned */ | |
2891 | { | |
2892 | /* | |
2893 | Find an aligned spot inside chunk. | |
2894 | Since we need to give back leading space in a chunk of at | |
2895 | least MINSIZE, if the first calculation places us at | |
2896 | a spot with less than MINSIZE leader, we can move to the | |
2897 | next aligned spot -- we've allocated enough total room so that | |
2898 | this is always possible. | |
2899 | */ | |
2900 | ||
2901 | brk = (char*)mem2chunk(((unsigned long)(m + alignment - 1)) & -alignment); | |
2902 | if ((long)(brk - (char*)(p)) < (long) MINSIZE) brk = brk + alignment; | |
2903 | ||
2904 | newp = (mchunkptr)brk; | |
2905 | leadsize = brk - (char*)(p); | |
2906 | newsize = chunksize(p) - leadsize; | |
2907 | ||
2908 | #if HAVE_MMAP | |
2909 | if(chunk_is_mmapped(p)) | |
2910 | { | |
2911 | newp->prev_size = p->prev_size + leadsize; | |
2912 | set_head(newp, newsize|IS_MMAPPED); | |
2913 | (void)mutex_unlock(&ar_ptr->mutex); | |
2914 | return chunk2mem(newp); | |
2915 | } | |
2916 | #endif | |
2917 | ||
2918 | /* give back leader, use the rest */ | |
2919 | ||
2920 | set_head(newp, newsize | PREV_INUSE); | |
2921 | set_inuse_bit_at_offset(newp, newsize); | |
2922 | set_head_size(p, leadsize); | |
2923 | chunk_free(ar_ptr, p); | |
2924 | p = newp; | |
2925 | ||
2926 | assert (newsize>=nb && (((unsigned long)(chunk2mem(p))) % alignment) == 0); | |
2927 | } | |
2928 | ||
2929 | /* Also give back spare room at the end */ | |
2930 | ||
2931 | remainder_size = chunksize(p) - nb; | |
2932 | ||
2933 | if (remainder_size >= (long)MINSIZE) | |
2934 | { | |
2935 | remainder = chunk_at_offset(p, nb); | |
2936 | set_head(remainder, remainder_size | PREV_INUSE); | |
2937 | set_head_size(p, nb); | |
2938 | chunk_free(ar_ptr, remainder); | |
2939 | } | |
2940 | ||
2941 | check_inuse_chunk(ar_ptr, p); | |
2942 | (void)mutex_unlock(&ar_ptr->mutex); | |
2943 | return chunk2mem(p); | |
2944 | ||
2945 | } | |
2946 | ||
2947 | \f | |
2948 | ||
2949 | ||
2950 | /* | |
2951 | valloc just invokes memalign with alignment argument equal | |
2952 | to the page size of the system (or as near to this as can | |
2953 | be figured out from all the includes/defines above.) | |
2954 | */ | |
2955 | ||
2956 | #if __STD_C | |
2957 | Void_t* vALLOc(size_t bytes) | |
2958 | #else | |
2959 | Void_t* vALLOc(bytes) size_t bytes; | |
2960 | #endif | |
2961 | { | |
2962 | return mEMALIGn (malloc_getpagesize, bytes); | |
2963 | } | |
2964 | ||
2965 | /* | |
2966 | pvalloc just invokes valloc for the nearest pagesize | |
2967 | that will accommodate request | |
2968 | */ | |
2969 | ||
2970 | ||
2971 | #if __STD_C | |
2972 | Void_t* pvALLOc(size_t bytes) | |
2973 | #else | |
2974 | Void_t* pvALLOc(bytes) size_t bytes; | |
2975 | #endif | |
2976 | { | |
2977 | size_t pagesize = malloc_getpagesize; | |
2978 | return mEMALIGn (pagesize, (bytes + pagesize - 1) & ~(pagesize - 1)); | |
2979 | } | |
2980 | ||
2981 | /* | |
2982 | ||
2983 | calloc calls malloc, then zeroes out the allocated chunk. | |
2984 | ||
2985 | */ | |
2986 | ||
2987 | #if __STD_C | |
2988 | Void_t* cALLOc(size_t n, size_t elem_size) | |
2989 | #else | |
2990 | Void_t* cALLOc(n, elem_size) size_t n; size_t elem_size; | |
2991 | #endif | |
2992 | { | |
2993 | arena *ar_ptr; | |
2994 | mchunkptr p, oldtop; | |
2995 | INTERNAL_SIZE_T csz, oldtopsize; | |
2996 | Void_t* mem; | |
2997 | ||
2998 | INTERNAL_SIZE_T sz = request2size(n * elem_size); | |
2999 | ||
3000 | arena_get(ar_ptr, sz); | |
3001 | if(!ar_ptr) | |
3002 | return 0; | |
3003 | ||
3004 | /* check if expand_top called, in which case don't need to clear */ | |
3005 | #if MORECORE_CLEARS | |
3006 | oldtop = top(ar_ptr); | |
3007 | oldtopsize = chunksize(top(ar_ptr)); | |
3008 | #endif | |
3009 | p = chunk_alloc (ar_ptr, sz); | |
3010 | ||
3011 | /* Only clearing follows, so we can unlock early. */ | |
3012 | (void)mutex_unlock(&ar_ptr->mutex); | |
3013 | ||
3014 | if (p == 0) | |
3015 | return 0; | |
3016 | else | |
3017 | { | |
3018 | mem = chunk2mem(p); | |
3019 | ||
3020 | /* Two optional cases in which clearing not necessary */ | |
3021 | ||
3022 | #if HAVE_MMAP | |
3023 | if (chunk_is_mmapped(p)) return mem; | |
3024 | #endif | |
3025 | ||
3026 | csz = chunksize(p); | |
3027 | ||
3028 | #if MORECORE_CLEARS | |
3029 | if (p == oldtop && csz > oldtopsize) | |
3030 | { | |
3031 | /* clear only the bytes from non-freshly-sbrked memory */ | |
3032 | csz = oldtopsize; | |
3033 | } | |
3034 | #endif | |
3035 | ||
3036 | MALLOC_ZERO(mem, csz - SIZE_SZ); | |
3037 | return mem; | |
3038 | } | |
3039 | } | |
3040 | ||
3041 | /* | |
3042 | ||
3043 | cfree just calls free. It is needed/defined on some systems | |
3044 | that pair it with calloc, presumably for odd historical reasons. | |
3045 | ||
3046 | */ | |
3047 | ||
3048 | #if !defined(_LIBC) | |
3049 | #if __STD_C | |
3050 | void cfree(Void_t *mem) | |
3051 | #else | |
3052 | void cfree(mem) Void_t *mem; | |
3053 | #endif | |
3054 | { | |
3055 | free(mem); | |
3056 | } | |
3057 | #endif | |
3058 | ||
3059 | \f | |
3060 | ||
3061 | /* | |
3062 | ||
3063 | Malloc_trim gives memory back to the system (via negative | |
3064 | arguments to sbrk) if there is unused memory at the `high' end of | |
3065 | the malloc pool. You can call this after freeing large blocks of | |
3066 | memory to potentially reduce the system-level memory requirements | |
3067 | of a program. However, it cannot guarantee to reduce memory. Under | |
3068 | some allocation patterns, some large free blocks of memory will be | |
3069 | locked between two used chunks, so they cannot be given back to | |
3070 | the system. | |
3071 | ||
3072 | The `pad' argument to malloc_trim represents the amount of free | |
3073 | trailing space to leave untrimmed. If this argument is zero, | |
3074 | only the minimum amount of memory to maintain internal data | |
3075 | structures will be left (one page or less). Non-zero arguments | |
3076 | can be supplied to maintain enough trailing space to service | |
3077 | future expected allocations without having to re-obtain memory | |
3078 | from the system. | |
3079 | ||
3080 | Malloc_trim returns 1 if it actually released any memory, else 0. | |
3081 | ||
3082 | */ | |
3083 | ||
3084 | #if __STD_C | |
3085 | int malloc_trim(size_t pad) | |
3086 | #else | |
3087 | int malloc_trim(pad) size_t pad; | |
3088 | #endif | |
3089 | { | |
3090 | int res; | |
3091 | ||
3092 | (void)mutex_lock(&main_arena.mutex); | |
3093 | res = arena_trim(&main_arena, pad); | |
3094 | (void)mutex_unlock(&main_arena.mutex); | |
3095 | return res; | |
3096 | } | |
3097 | ||
3098 | static int | |
3099 | #if __STD_C | |
3100 | arena_trim(arena *ar_ptr, size_t pad) | |
3101 | #else | |
3102 | arena_trim(ar_ptr, pad) arena *ar_ptr; size_t pad; | |
3103 | #endif | |
3104 | { | |
3105 | mchunkptr top_chunk; /* The current top chunk */ | |
3106 | long top_size; /* Amount of top-most memory */ | |
3107 | long extra; /* Amount to release */ | |
3108 | char* current_brk; /* address returned by pre-check sbrk call */ | |
3109 | char* new_brk; /* address returned by negative sbrk call */ | |
3110 | ||
3111 | unsigned long pagesz = malloc_getpagesize; | |
3112 | ||
3113 | top_chunk = top(ar_ptr); | |
3114 | top_size = chunksize(top_chunk); | |
3115 | extra = ((top_size - pad - MINSIZE + (pagesz-1)) / pagesz - 1) * pagesz; | |
3116 | ||
3117 | if (extra < (long)pagesz) /* Not enough memory to release */ | |
3118 | return 0; | |
3119 | ||
3120 | #ifndef NO_THREADS | |
3121 | if(ar_ptr == &main_arena) { | |
3122 | #endif | |
3123 | ||
3124 | /* Test to make sure no one else called sbrk */ | |
3125 | current_brk = (char*)(MORECORE (0)); | |
3126 | if (current_brk != (char*)(top_chunk) + top_size) | |
3127 | return 0; /* Apparently we don't own memory; must fail */ | |
3128 | ||
3129 | new_brk = (char*)(MORECORE (-extra)); | |
3130 | ||
3131 | if (new_brk == (char*)(MORECORE_FAILURE)) { /* sbrk failed? */ | |
3132 | /* Try to figure out what we have */ | |
3133 | current_brk = (char*)(MORECORE (0)); | |
3134 | top_size = current_brk - (char*)top_chunk; | |
3135 | if (top_size >= (long)MINSIZE) /* if not, we are very very dead! */ | |
3136 | { | |
3137 | sbrked_mem = current_brk - sbrk_base; | |
3138 | set_head(top_chunk, top_size | PREV_INUSE); | |
3139 | } | |
3140 | check_chunk(ar_ptr, top_chunk); | |
3141 | return 0; | |
3142 | } | |
3143 | sbrked_mem -= extra; | |
3144 | ||
3145 | #ifndef NO_THREADS | |
3146 | } else { | |
3147 | if(grow_heap(heap_for_ptr(top_chunk), -extra) != 0) | |
3148 | return 0; | |
3149 | } | |
3150 | #endif | |
3151 | ||
3152 | /* Success. Adjust top accordingly. */ | |
3153 | set_head(top_chunk, (top_size - extra) | PREV_INUSE); | |
3154 | check_chunk(ar_ptr, top_chunk); | |
3155 | return 1; | |
3156 | } | |
3157 | ||
3158 | \f | |
3159 | ||
3160 | /* | |
3161 | malloc_usable_size: | |
3162 | ||
3163 | This routine tells you how many bytes you can actually use in an | |
3164 | allocated chunk, which may be more than you requested (although | |
3165 | often not). You can use this many bytes without worrying about | |
3166 | overwriting other allocated objects. Not a particularly great | |
3167 | programming practice, but still sometimes useful. | |
3168 | ||
3169 | */ | |
3170 | ||
3171 | #if __STD_C | |
3172 | size_t malloc_usable_size(Void_t* mem) | |
3173 | #else | |
3174 | size_t malloc_usable_size(mem) Void_t* mem; | |
3175 | #endif | |
3176 | { | |
3177 | mchunkptr p; | |
3178 | ||
3179 | if (mem == 0) | |
3180 | return 0; | |
3181 | else | |
3182 | { | |
3183 | p = mem2chunk(mem); | |
3184 | if(!chunk_is_mmapped(p)) | |
3185 | { | |
3186 | if (!inuse(p)) return 0; | |
3187 | check_inuse_chunk(arena_for_ptr(mem), p); | |
3188 | return chunksize(p) - SIZE_SZ; | |
3189 | } | |
3190 | return chunksize(p) - 2*SIZE_SZ; | |
3191 | } | |
3192 | } | |
3193 | ||
3194 | ||
3195 | \f | |
3196 | ||
3197 | /* Utility to update current_mallinfo for malloc_stats and mallinfo() */ | |
3198 | ||
3199 | static void malloc_update_mallinfo __MALLOC_P ((void)) | |
3200 | { | |
3201 | arena *ar_ptr = &main_arena; | |
3202 | int i, navail; | |
3203 | mbinptr b; | |
3204 | mchunkptr p; | |
3205 | #if MALLOC_DEBUG | |
3206 | mchunkptr q; | |
3207 | #endif | |
3208 | INTERNAL_SIZE_T avail; | |
3209 | ||
3210 | (void)mutex_lock(&ar_ptr->mutex); | |
3211 | avail = chunksize(top(ar_ptr)); | |
3212 | navail = ((long)(avail) >= (long)MINSIZE)? 1 : 0; | |
3213 | ||
3214 | for (i = 1; i < NAV; ++i) | |
3215 | { | |
3216 | b = bin_at(ar_ptr, i); | |
3217 | for (p = last(b); p != b; p = p->bk) | |
3218 | { | |
3219 | #if MALLOC_DEBUG | |
3220 | check_free_chunk(ar_ptr, p); | |
3221 | for (q = next_chunk(p); | |
3222 | q < top(ar_ptr) && inuse(q) && (long)chunksize(q) >= (long)MINSIZE; | |
3223 | q = next_chunk(q)) | |
3224 | check_inuse_chunk(ar_ptr, q); | |
3225 | #endif | |
3226 | avail += chunksize(p); | |
3227 | navail++; | |
3228 | } | |
3229 | } | |
3230 | ||
3231 | current_mallinfo.ordblks = navail; | |
3232 | current_mallinfo.uordblks = sbrked_mem - avail; | |
3233 | current_mallinfo.fordblks = avail; | |
3234 | current_mallinfo.hblks = n_mmaps; | |
3235 | current_mallinfo.hblkhd = mmapped_mem; | |
3236 | current_mallinfo.keepcost = chunksize(top(ar_ptr)); | |
3237 | ||
3238 | (void)mutex_unlock(&ar_ptr->mutex); | |
3239 | } | |
3240 | ||
3241 | \f | |
3242 | ||
3243 | /* | |
3244 | ||
3245 | malloc_stats: | |
3246 | ||
3247 | Prints on stderr the amount of space obtain from the system (both | |
3248 | via sbrk and mmap), the maximum amount (which may be more than | |
3249 | current if malloc_trim and/or munmap got called), the maximum | |
3250 | number of simultaneous mmap regions used, and the current number | |
3251 | of bytes allocated via malloc (or realloc, etc) but not yet | |
3252 | freed. (Note that this is the number of bytes allocated, not the | |
3253 | number requested. It will be larger than the number requested | |
3254 | because of alignment and bookkeeping overhead.) | |
3255 | ||
3256 | */ | |
3257 | ||
3258 | void malloc_stats() | |
3259 | { | |
3260 | malloc_update_mallinfo(); | |
3261 | fprintf(stderr, "max system bytes = %10u\n", | |
3262 | (unsigned int)(max_total_mem)); | |
3263 | fprintf(stderr, "system bytes = %10u\n", | |
3264 | (unsigned int)(sbrked_mem + mmapped_mem)); | |
3265 | fprintf(stderr, "in use bytes = %10u\n", | |
3266 | (unsigned int)(current_mallinfo.uordblks + mmapped_mem)); | |
3267 | #if HAVE_MMAP | |
3268 | fprintf(stderr, "max mmap regions = %10u\n", | |
3269 | (unsigned int)max_n_mmaps); | |
3270 | #endif | |
3271 | #if THREAD_STATS | |
3272 | fprintf(stderr, "arenas created = %10d\n", stat_n_arenas); | |
3273 | fprintf(stderr, "heaps created = %10d\n", stat_n_heaps); | |
3274 | fprintf(stderr, "locked directly = %10ld\n", stat_lock_direct); | |
3275 | fprintf(stderr, "locked in loop = %10ld\n", stat_lock_loop); | |
3276 | #endif | |
3277 | } | |
3278 | ||
3279 | /* | |
3280 | mallinfo returns a copy of updated current mallinfo. | |
3281 | */ | |
3282 | ||
3283 | struct mallinfo mALLINFo() | |
3284 | { | |
3285 | malloc_update_mallinfo(); | |
3286 | return current_mallinfo; | |
3287 | } | |
3288 | ||
3289 | ||
3290 | \f | |
3291 | ||
3292 | /* | |
3293 | mallopt: | |
3294 | ||
3295 | mallopt is the general SVID/XPG interface to tunable parameters. | |
3296 | The format is to provide a (parameter-number, parameter-value) pair. | |
3297 | mallopt then sets the corresponding parameter to the argument | |
3298 | value if it can (i.e., so long as the value is meaningful), | |
3299 | and returns 1 if successful else 0. | |
3300 | ||
3301 | See descriptions of tunable parameters above. | |
3302 | ||
3303 | */ | |
3304 | ||
3305 | #if __STD_C | |
3306 | int mALLOPt(int param_number, int value) | |
3307 | #else | |
3308 | int mALLOPt(param_number, value) int param_number; int value; | |
3309 | #endif | |
3310 | { | |
3311 | switch(param_number) | |
3312 | { | |
3313 | case M_TRIM_THRESHOLD: | |
3314 | trim_threshold = value; return 1; | |
3315 | case M_TOP_PAD: | |
3316 | top_pad = value; return 1; | |
3317 | case M_MMAP_THRESHOLD: | |
3318 | #ifndef NO_THREADS | |
3319 | /* Forbid setting the threshold too high. */ | |
3320 | if((unsigned long)value > HEAP_MAX_SIZE/2) return 0; | |
3321 | #endif | |
3322 | mmap_threshold = value; return 1; | |
3323 | case M_MMAP_MAX: | |
3324 | #if HAVE_MMAP | |
3325 | n_mmaps_max = value; return 1; | |
3326 | #else | |
3327 | if (value != 0) return 0; else n_mmaps_max = value; return 1; | |
3328 | #endif | |
3329 | ||
3330 | default: | |
3331 | return 0; | |
3332 | } | |
3333 | } | |
3334 | ||
3335 | #if 0 && defined(_LIBC) | |
3336 | weak_alias (__libc_calloc, calloc) | |
3337 | weak_alias (__libc_free, cfree) | |
3338 | weak_alias (__libc_free, free) | |
3339 | weak_alias (__libc_malloc, malloc) | |
3340 | weak_alias (__libc_memalign, memalign) | |
3341 | weak_alias (__libc_realloc, realloc) | |
3342 | weak_alias (__libc_valloc, valloc) | |
3343 | weak_alias (__libc_pvalloc, pvalloc) | |
3344 | weak_alias (__libc_mallinfo, mallinfo) | |
3345 | weak_alias (__libc_mallopt, mallopt) | |
3346 | #endif | |
3347 | ||
3348 | /* | |
3349 | ||
3350 | History: | |
3351 | ||
3352 | V2.6.4-pt Wed Dec 4 1996 Wolfram Gloger (wmglo@dent.med.uni-muenchen.de) | |
3353 | * Very minor updates from the released 2.6.4 version. | |
3354 | * Trimmed include file down to exported data structures. | |
3355 | * Changes from H.J. Lu for glibc-2.0. | |
3356 | ||
3357 | V2.6.3i-pt Sep 16 1996 Wolfram Gloger (wmglo@dent.med.uni-muenchen.de) | |
3358 | * Many changes for multiple threads | |
3359 | * Introduced arenas and heaps | |
3360 | ||
3361 | V2.6.3 Sun May 19 08:17:58 1996 Doug Lea (dl at gee) | |
3362 | * Added pvalloc, as recommended by H.J. Liu | |
3363 | * Added 64bit pointer support mainly from Wolfram Gloger | |
3364 | * Added anonymously donated WIN32 sbrk emulation | |
3365 | * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen | |
3366 | * malloc_extend_top: fix mask error that caused wastage after | |
3367 | foreign sbrks | |
3368 | * Add linux mremap support code from HJ Liu | |
3369 | ||
3370 | V2.6.2 Tue Dec 5 06:52:55 1995 Doug Lea (dl at gee) | |
3371 | * Integrated most documentation with the code. | |
3372 | * Add support for mmap, with help from | |
3373 | Wolfram Gloger (Gloger@lrz.uni-muenchen.de). | |
3374 | * Use last_remainder in more cases. | |
3375 | * Pack bins using idea from colin@nyx10.cs.du.edu | |
3376 | * Use ordered bins instead of best-fit threshhold | |
3377 | * Eliminate block-local decls to simplify tracing and debugging. | |
3378 | * Support another case of realloc via move into top | |
3379 | * Fix error occuring when initial sbrk_base not word-aligned. | |
3380 | * Rely on page size for units instead of SBRK_UNIT to | |
3381 | avoid surprises about sbrk alignment conventions. | |
3382 | * Add mallinfo, mallopt. Thanks to Raymond Nijssen | |
3383 | (raymond@es.ele.tue.nl) for the suggestion. | |
3384 | * Add `pad' argument to malloc_trim and top_pad mallopt parameter. | |
3385 | * More precautions for cases where other routines call sbrk, | |
3386 | courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de). | |
3387 | * Added macros etc., allowing use in linux libc from | |
3388 | H.J. Lu (hjl@gnu.ai.mit.edu) | |
3389 | * Inverted this history list | |
3390 | ||
3391 | V2.6.1 Sat Dec 2 14:10:57 1995 Doug Lea (dl at gee) | |
3392 | * Re-tuned and fixed to behave more nicely with V2.6.0 changes. | |
3393 | * Removed all preallocation code since under current scheme | |
3394 | the work required to undo bad preallocations exceeds | |
3395 | the work saved in good cases for most test programs. | |
3396 | * No longer use return list or unconsolidated bins since | |
3397 | no scheme using them consistently outperforms those that don't | |
3398 | given above changes. | |
3399 | * Use best fit for very large chunks to prevent some worst-cases. | |
3400 | * Added some support for debugging | |
3401 | ||
3402 | V2.6.0 Sat Nov 4 07:05:23 1995 Doug Lea (dl at gee) | |
3403 | * Removed footers when chunks are in use. Thanks to | |
3404 | Paul Wilson (wilson@cs.texas.edu) for the suggestion. | |
3405 | ||
3406 | V2.5.4 Wed Nov 1 07:54:51 1995 Doug Lea (dl at gee) | |
3407 | * Added malloc_trim, with help from Wolfram Gloger | |
3408 | (wmglo@Dent.MED.Uni-Muenchen.DE). | |
3409 | ||
3410 | V2.5.3 Tue Apr 26 10:16:01 1994 Doug Lea (dl at g) | |
3411 | ||
3412 | V2.5.2 Tue Apr 5 16:20:40 1994 Doug Lea (dl at g) | |
3413 | * realloc: try to expand in both directions | |
3414 | * malloc: swap order of clean-bin strategy; | |
3415 | * realloc: only conditionally expand backwards | |
3416 | * Try not to scavenge used bins | |
3417 | * Use bin counts as a guide to preallocation | |
3418 | * Occasionally bin return list chunks in first scan | |
3419 | * Add a few optimizations from colin@nyx10.cs.du.edu | |
3420 | ||
3421 | V2.5.1 Sat Aug 14 15:40:43 1993 Doug Lea (dl at g) | |
3422 | * faster bin computation & slightly different binning | |
3423 | * merged all consolidations to one part of malloc proper | |
3424 | (eliminating old malloc_find_space & malloc_clean_bin) | |
3425 | * Scan 2 returns chunks (not just 1) | |
3426 | * Propagate failure in realloc if malloc returns 0 | |
3427 | * Add stuff to allow compilation on non-ANSI compilers | |
3428 | from kpv@research.att.com | |
3429 | ||
3430 | V2.5 Sat Aug 7 07:41:59 1993 Doug Lea (dl at g.oswego.edu) | |
3431 | * removed potential for odd address access in prev_chunk | |
3432 | * removed dependency on getpagesize.h | |
3433 | * misc cosmetics and a bit more internal documentation | |
3434 | * anticosmetics: mangled names in macros to evade debugger strangeness | |
3435 | * tested on sparc, hp-700, dec-mips, rs6000 | |
3436 | with gcc & native cc (hp, dec only) allowing | |
3437 | Detlefs & Zorn comparison study (in SIGPLAN Notices.) | |
3438 | ||
3439 | Trial version Fri Aug 28 13:14:29 1992 Doug Lea (dl at g.oswego.edu) | |
3440 | * Based loosely on libg++-1.2X malloc. (It retains some of the overall | |
3441 | structure of old version, but most details differ.) | |
3442 | ||
3443 | */ |