3 #if 0 /* Moved to malloc.h */
4 /* ---------- To make a malloc.h, start cutting here ------------ */
7 A version of malloc/free/realloc written by Doug Lea and released to the
8 public domain. Send questions/comments/complaints/performance data
11 * VERSION 2.6.6 Sun Mar 5 19:10:03 2000 Doug Lea (dl at gee)
13 Note: There may be an updated version of this malloc obtainable at
14 ftp://g.oswego.edu/pub/misc/malloc.c
15 Check before installing!
17 * Why use this malloc?
19 This is not the fastest, most space-conserving, most portable, or
20 most tunable malloc ever written. However it is among the fastest
21 while also being among the most space-conserving, portable and tunable.
22 Consistent balance across these factors results in a good general-purpose
23 allocator. For a high-level description, see
24 http://g.oswego.edu/dl/html/malloc.html
26 * Synopsis of public routines
28 (Much fuller descriptions are contained in the program documentation below.)
31 Return a pointer to a newly allocated chunk of at least n bytes, or null
32 if no space is available.
34 Release the chunk of memory pointed to by p, or no effect if p is null.
35 realloc(Void_t* p, size_t n);
36 Return a pointer to a chunk of size n that contains the same data
37 as does chunk p up to the minimum of (n, p's size) bytes, or null
38 if no space is available. The returned pointer may or may not be
39 the same as p. If p is null, equivalent to malloc. Unless the
40 #define REALLOC_ZERO_BYTES_FREES below is set, realloc with a
41 size argument of zero (re)allocates a minimum-sized chunk.
42 memalign(size_t alignment, size_t n);
43 Return a pointer to a newly allocated chunk of n bytes, aligned
44 in accord with the alignment argument, which must be a power of
47 Equivalent to memalign(pagesize, n), where pagesize is the page
48 size of the system (or as near to this as can be figured out from
49 all the includes/defines below.)
51 Equivalent to valloc(minimum-page-that-holds(n)), that is,
52 round up n to nearest pagesize.
53 calloc(size_t unit, size_t quantity);
54 Returns a pointer to quantity * unit bytes, with all locations
57 Equivalent to free(p).
58 malloc_trim(size_t pad);
59 Release all but pad bytes of freed top-most memory back
60 to the system. Return 1 if successful, else 0.
61 malloc_usable_size(Void_t* p);
62 Report the number usable allocated bytes associated with allocated
63 chunk p. This may or may not report more bytes than were requested,
64 due to alignment and minimum size constraints.
66 Prints brief summary statistics.
68 Returns (by copy) a struct containing various summary statistics.
69 mallopt(int parameter_number, int parameter_value)
70 Changes one of the tunable parameters described below. Returns
71 1 if successful in changing the parameter, else 0.
76 8 byte alignment is currently hardwired into the design. This
77 seems to suffice for all current machines and C compilers.
79 Assumed pointer representation: 4 or 8 bytes
80 Code for 8-byte pointers is untested by me but has worked
81 reliably by Wolfram Gloger, who contributed most of the
82 changes supporting this.
84 Assumed size_t representation: 4 or 8 bytes
85 Note that size_t is allowed to be 4 bytes even if pointers are 8.
87 Minimum overhead per allocated chunk: 4 or 8 bytes
88 Each malloced chunk has a hidden overhead of 4 bytes holding size
89 and status information.
91 Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead)
92 8-byte ptrs: 24/32 bytes (including, 4/8 overhead)
94 When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte
95 ptrs but 4 byte size) or 24 (for 8/8) additional bytes are
96 needed; 4 (8) for a trailing size field
97 and 8 (16) bytes for free list pointers. Thus, the minimum
98 allocatable size is 16/24/32 bytes.
100 Even a request for zero bytes (i.e., malloc(0)) returns a
101 pointer to something of the minimum allocatable size.
103 Maximum allocated size: 4-byte size_t: 2^31 - 8 bytes
104 8-byte size_t: 2^63 - 16 bytes
106 It is assumed that (possibly signed) size_t bit values suffice to
107 represent chunk sizes. `Possibly signed' is due to the fact
108 that `size_t' may be defined on a system as either a signed or
109 an unsigned type. To be conservative, values that would appear
110 as negative numbers are avoided.
111 Requests for sizes with a negative sign bit when the request
112 size is treaded as a long will return null.
114 Maximum overhead wastage per allocated chunk: normally 15 bytes
116 Alignnment demands, plus the minimum allocatable size restriction
117 make the normal worst-case wastage 15 bytes (i.e., up to 15
118 more bytes will be allocated than were requested in malloc), with
120 1. Because requests for zero bytes allocate non-zero space,
121 the worst case wastage for a request of zero bytes is 24 bytes.
122 2. For requests >= mmap_threshold that are serviced via
123 mmap(), the worst case wastage is 8 bytes plus the remainder
124 from a system page (the minimal mmap unit); typically 4096 bytes.
128 Here are some features that are NOT currently supported
130 * No user-definable hooks for callbacks and the like.
131 * No automated mechanism for fully checking that all accesses
132 to malloced memory stay within their bounds.
133 * No support for compaction.
135 * Synopsis of compile-time options:
137 People have reported using previous versions of this malloc on all
138 versions of Unix, sometimes by tweaking some of the defines
139 below. It has been tested most extensively on Solaris and
140 Linux. It is also reported to work on WIN32 platforms.
141 People have also reported adapting this malloc for use in
142 stand-alone embedded systems.
144 The implementation is in straight, hand-tuned ANSI C. Among other
145 consequences, it uses a lot of macros. Because of this, to be at
146 all usable, this code should be compiled using an optimizing compiler
147 (for example gcc -O2) that can simplify expressions and control
150 __STD_C (default: derived from C compiler defines)
151 Nonzero if using ANSI-standard C compiler, a C++ compiler, or
152 a C compiler sufficiently close to ANSI to get away with it.
153 DEBUG (default: NOT defined)
154 Define to enable debugging. Adds fairly extensive assertion-based
155 checking to help track down memory errors, but noticeably slows down
157 REALLOC_ZERO_BYTES_FREES (default: NOT defined)
158 Define this if you think that realloc(p, 0) should be equivalent
159 to free(p). Otherwise, since malloc returns a unique pointer for
160 malloc(0), so does realloc(p, 0).
161 HAVE_MEMCPY (default: defined)
162 Define if you are not otherwise using ANSI STD C, but still
163 have memcpy and memset in your C library and want to use them.
164 Otherwise, simple internal versions are supplied.
165 USE_MEMCPY (default: 1 if HAVE_MEMCPY is defined, 0 otherwise)
166 Define as 1 if you want the C library versions of memset and
167 memcpy called in realloc and calloc (otherwise macro versions are used).
168 At least on some platforms, the simple macro versions usually
169 outperform libc versions.
170 HAVE_MMAP (default: defined as 1)
171 Define to non-zero to optionally make malloc() use mmap() to
172 allocate very large blocks.
173 HAVE_MREMAP (default: defined as 0 unless Linux libc set)
174 Define to non-zero to optionally make realloc() use mremap() to
175 reallocate very large blocks.
176 malloc_getpagesize (default: derived from system #includes)
177 Either a constant or routine call returning the system page size.
178 HAVE_USR_INCLUDE_MALLOC_H (default: NOT defined)
179 Optionally define if you are on a system with a /usr/include/malloc.h
180 that declares struct mallinfo. It is not at all necessary to
181 define this even if you do, but will ensure consistency.
182 INTERNAL_SIZE_T (default: size_t)
183 Define to a 32-bit type (probably `unsigned int') if you are on a
184 64-bit machine, yet do not want or need to allow malloc requests of
185 greater than 2^31 to be handled. This saves space, especially for
187 INTERNAL_LINUX_C_LIB (default: NOT defined)
188 Defined only when compiled as part of Linux libc.
189 Also note that there is some odd internal name-mangling via defines
190 (for example, internally, `malloc' is named `mALLOc') needed
191 when compiling in this case. These look funny but don't otherwise
193 WIN32 (default: undefined)
194 Define this on MS win (95, nt) platforms to compile in sbrk emulation.
195 LACKS_UNISTD_H (default: undefined if not WIN32)
196 Define this if your system does not have a <unistd.h>.
197 LACKS_SYS_PARAM_H (default: undefined if not WIN32)
198 Define this if your system does not have a <sys/param.h>.
199 MORECORE (default: sbrk)
200 The name of the routine to call to obtain more memory from the system.
201 MORECORE_FAILURE (default: -1)
202 The value returned upon failure of MORECORE.
203 MORECORE_CLEARS (default 1)
204 True (1) if the routine mapped to MORECORE zeroes out memory (which
206 DEFAULT_TRIM_THRESHOLD
208 DEFAULT_MMAP_THRESHOLD
210 Default values of tunable parameters (described in detail below)
211 controlling interaction with host system routines (sbrk, mmap, etc).
212 These values may also be changed dynamically via mallopt(). The
213 preset defaults are those that give best performance for typical
215 USE_DL_PREFIX (default: undefined)
216 Prefix all public routines with the string 'dl'. Useful to
217 quickly avoid procedure declaration conflicts and linker symbol
218 conflicts with existing memory allocation routines.
236 #endif /*__cplusplus*/
241 #if (__STD_C || defined(WIN32))
249 #include <stddef.h> /* for size_t */
251 #include <sys/types.h>
258 #include <stdio.h> /* needed for malloc_stats */
269 Because freed chunks may be overwritten with link fields, this
270 malloc will often die when freed memory is overwritten by user
271 programs. This can be very effective (albeit in an annoying way)
272 in helping track down dangling pointers.
274 If you compile with -DDEBUG, a number of assertion checks are
275 enabled that will catch more memory errors. You probably won't be
276 able to make much sense of the actual assertion errors, but they
277 should help you locate incorrectly overwritten memory. The
278 checking is fairly extensive, and will slow down execution
279 noticeably. Calling malloc_stats or mallinfo with DEBUG set will
280 attempt to check every non-mmapped allocated and free chunk in the
281 course of computing the summmaries. (By nature, mmapped regions
282 cannot be checked very much automatically.)
284 Setting DEBUG may also be helpful if you are trying to modify
285 this code. The assertions in the check routines spell out in more
286 detail the assumptions and invariants underlying the algorithms.
293 #define assert(x) ((void)0)
298 INTERNAL_SIZE_T is the word-size used for internal bookkeeping
299 of chunk sizes. On a 64-bit machine, you can reduce malloc
300 overhead by defining INTERNAL_SIZE_T to be a 32 bit `unsigned int'
301 at the expense of not being able to handle requests greater than
302 2^31. This limitation is hardly ever a concern; you are encouraged
303 to set this. However, the default version is the same as size_t.
306 #ifndef INTERNAL_SIZE_T
307 #define INTERNAL_SIZE_T size_t
311 REALLOC_ZERO_BYTES_FREES should be set if a call to
312 realloc with zero bytes should be the same as a call to free.
313 Some people think it should. Otherwise, since this malloc
314 returns a unique pointer for malloc(0), so does realloc(p, 0).
318 /* #define REALLOC_ZERO_BYTES_FREES */
322 WIN32 causes an emulation of sbrk to be compiled in
323 mmap-based options are not currently supported in WIN32.
328 #define MORECORE wsbrk
331 #define LACKS_UNISTD_H
332 #define LACKS_SYS_PARAM_H
335 Include 'windows.h' to get the necessary declarations for the
336 Microsoft Visual C++ data structures and routines used in the 'sbrk'
339 Define WIN32_LEAN_AND_MEAN so that only the essential Microsoft
340 Visual C++ header files are included.
342 #define WIN32_LEAN_AND_MEAN
348 HAVE_MEMCPY should be defined if you are not otherwise using
349 ANSI STD C, but still have memcpy and memset in your C library
350 and want to use them in calloc and realloc. Otherwise simple
351 macro versions are defined here.
353 USE_MEMCPY should be defined as 1 if you actually want to
354 have memset and memcpy called. People report that the macro
355 versions are often enough faster than libc versions on many
356 systems that it is better to use them.
370 #if (__STD_C || defined(HAVE_MEMCPY))
373 void* memset(void*, int, size_t);
374 void* memcpy(void*, const void*, size_t);
377 /* On Win32 platforms, 'memset()' and 'memcpy()' are already declared in */
388 /* The following macros are only invoked with (2n+1)-multiples of
389 INTERNAL_SIZE_T units, with a positive integer n. This is exploited
390 for fast inline execution when n is small. */
392 #define MALLOC_ZERO(charp, nbytes) \
394 INTERNAL_SIZE_T mzsz = (nbytes); \
395 if(mzsz <= 9*sizeof(mzsz)) { \
396 INTERNAL_SIZE_T* mz = (INTERNAL_SIZE_T*) (charp); \
397 if(mzsz >= 5*sizeof(mzsz)) { *mz++ = 0; \
399 if(mzsz >= 7*sizeof(mzsz)) { *mz++ = 0; \
401 if(mzsz >= 9*sizeof(mzsz)) { *mz++ = 0; \
406 } else memset((charp), 0, mzsz); \
409 #define MALLOC_COPY(dest,src,nbytes) \
411 INTERNAL_SIZE_T mcsz = (nbytes); \
412 if(mcsz <= 9*sizeof(mcsz)) { \
413 INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) (src); \
414 INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) (dest); \
415 if(mcsz >= 5*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \
416 *mcdst++ = *mcsrc++; \
417 if(mcsz >= 7*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \
418 *mcdst++ = *mcsrc++; \
419 if(mcsz >= 9*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \
420 *mcdst++ = *mcsrc++; }}} \
421 *mcdst++ = *mcsrc++; \
422 *mcdst++ = *mcsrc++; \
424 } else memcpy(dest, src, mcsz); \
427 #else /* !USE_MEMCPY */
429 /* Use Duff's device for good zeroing/copying performance. */
431 #define MALLOC_ZERO(charp, nbytes) \
433 INTERNAL_SIZE_T* mzp = (INTERNAL_SIZE_T*)(charp); \
434 long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn; \
435 if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \
437 case 0: for(;;) { *mzp++ = 0; \
438 case 7: *mzp++ = 0; \
439 case 6: *mzp++ = 0; \
440 case 5: *mzp++ = 0; \
441 case 4: *mzp++ = 0; \
442 case 3: *mzp++ = 0; \
443 case 2: *mzp++ = 0; \
444 case 1: *mzp++ = 0; if(mcn <= 0) break; mcn--; } \
448 #define MALLOC_COPY(dest,src,nbytes) \
450 INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) src; \
451 INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) dest; \
452 long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn; \
453 if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \
455 case 0: for(;;) { *mcdst++ = *mcsrc++; \
456 case 7: *mcdst++ = *mcsrc++; \
457 case 6: *mcdst++ = *mcsrc++; \
458 case 5: *mcdst++ = *mcsrc++; \
459 case 4: *mcdst++ = *mcsrc++; \
460 case 3: *mcdst++ = *mcsrc++; \
461 case 2: *mcdst++ = *mcsrc++; \
462 case 1: *mcdst++ = *mcsrc++; if(mcn <= 0) break; mcn--; } \
470 Define HAVE_MMAP to optionally make malloc() use mmap() to
471 allocate very large blocks. These will be returned to the
472 operating system immediately after a free().
480 Define HAVE_MREMAP to make realloc() use mremap() to re-allocate
481 large blocks. This is currently only possible on Linux with
482 kernel versions newer than 1.3.77.
486 #ifdef INTERNAL_LINUX_C_LIB
487 #define HAVE_MREMAP 1
489 #define HAVE_MREMAP 0
497 #include <sys/mman.h>
499 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
500 #define MAP_ANONYMOUS MAP_ANON
503 #endif /* HAVE_MMAP */
506 Access to system page size. To the extent possible, this malloc
507 manages memory from the system in page-size units.
509 The following mechanics for getpagesize were adapted from
510 bsd/gnu getpagesize.h
513 #ifndef LACKS_UNISTD_H
517 #ifndef malloc_getpagesize
518 # ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */
519 # ifndef _SC_PAGE_SIZE
520 # define _SC_PAGE_SIZE _SC_PAGESIZE
523 # ifdef _SC_PAGE_SIZE
524 # define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
526 # if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
527 extern size_t getpagesize();
528 # define malloc_getpagesize getpagesize()
531 # define malloc_getpagesize (4096) /* TBD: Use 'GetSystemInfo' instead */
533 # ifndef LACKS_SYS_PARAM_H
534 # include <sys/param.h>
536 # ifdef EXEC_PAGESIZE
537 # define malloc_getpagesize EXEC_PAGESIZE
541 # define malloc_getpagesize NBPG
543 # define malloc_getpagesize (NBPG * CLSIZE)
547 # define malloc_getpagesize NBPC
550 # define malloc_getpagesize PAGESIZE
552 # define malloc_getpagesize (4096) /* just guess */
565 This version of malloc supports the standard SVID/XPG mallinfo
566 routine that returns a struct containing the same kind of
567 information you can get from malloc_stats. It should work on
568 any SVID/XPG compliant system that has a /usr/include/malloc.h
569 defining struct mallinfo. (If you'd like to install such a thing
570 yourself, cut out the preliminary declarations as described above
571 and below and save them in a malloc.h file. But there's no
572 compelling reason to bother to do this.)
574 The main declaration needed is the mallinfo struct that is returned
575 (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a
576 bunch of fields, most of which are not even meaningful in this
577 version of malloc. Some of these fields are are instead filled by
578 mallinfo() with other numbers that might possibly be of interest.
580 HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
581 /usr/include/malloc.h file that includes a declaration of struct
582 mallinfo. If so, it is included; else an SVID2/XPG2 compliant
583 version is declared below. These must be precisely the same for
588 /* #define HAVE_USR_INCLUDE_MALLOC_H */
590 #if HAVE_USR_INCLUDE_MALLOC_H
591 #include "/usr/include/malloc.h"
594 /* SVID2/XPG mallinfo structure */
597 int arena; /* total space allocated from system */
598 int ordblks; /* number of non-inuse chunks */
599 int smblks; /* unused -- always zero */
600 int hblks; /* number of mmapped regions */
601 int hblkhd; /* total space in mmapped regions */
602 int usmblks; /* unused -- always zero */
603 int fsmblks; /* unused -- always zero */
604 int uordblks; /* total allocated space */
605 int fordblks; /* total non-inuse space */
606 int keepcost; /* top-most, releasable (via malloc_trim) space */
609 /* SVID2/XPG mallopt options */
611 #define M_MXFAST 1 /* UNUSED in this malloc */
612 #define M_NLBLKS 2 /* UNUSED in this malloc */
613 #define M_GRAIN 3 /* UNUSED in this malloc */
614 #define M_KEEP 4 /* UNUSED in this malloc */
618 /* mallopt options that actually do something */
620 #define M_TRIM_THRESHOLD -1
622 #define M_MMAP_THRESHOLD -3
623 #define M_MMAP_MAX -4
626 #ifndef DEFAULT_TRIM_THRESHOLD
627 #define DEFAULT_TRIM_THRESHOLD (128 * 1024)
631 M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
632 to keep before releasing via malloc_trim in free().
634 Automatic trimming is mainly useful in long-lived programs.
635 Because trimming via sbrk can be slow on some systems, and can
636 sometimes be wasteful (in cases where programs immediately
637 afterward allocate more large chunks) the value should be high
638 enough so that your overall system performance would improve by
641 The trim threshold and the mmap control parameters (see below)
642 can be traded off with one another. Trimming and mmapping are
643 two different ways of releasing unused memory back to the
644 system. Between these two, it is often possible to keep
645 system-level demands of a long-lived program down to a bare
646 minimum. For example, in one test suite of sessions measuring
647 the XF86 X server on Linux, using a trim threshold of 128K and a
648 mmap threshold of 192K led to near-minimal long term resource
651 If you are using this malloc in a long-lived program, it should
652 pay to experiment with these values. As a rough guide, you
653 might set to a value close to the average size of a process
654 (program) running on your system. Releasing this much memory
655 would allow such a process to run in memory. Generally, it's
656 worth it to tune for trimming rather tham memory mapping when a
657 program undergoes phases where several large chunks are
658 allocated and released in ways that can reuse each other's
659 storage, perhaps mixed with phases where there are no such
660 chunks at all. And in well-behaved long-lived programs,
661 controlling release of large blocks via trimming versus mapping
664 However, in most programs, these parameters serve mainly as
665 protection against the system-level effects of carrying around
666 massive amounts of unneeded memory. Since frequent calls to
667 sbrk, mmap, and munmap otherwise degrade performance, the default
668 parameters are set to relatively high values that serve only as
671 The default trim value is high enough to cause trimming only in
672 fairly extreme (by current memory consumption standards) cases.
673 It must be greater than page size to have any useful effect. To
674 disable trimming completely, you can set to (unsigned long)(-1);
680 #ifndef DEFAULT_TOP_PAD
681 #define DEFAULT_TOP_PAD (0)
685 M_TOP_PAD is the amount of extra `padding' space to allocate or
686 retain whenever sbrk is called. It is used in two ways internally:
688 * When sbrk is called to extend the top of the arena to satisfy
689 a new malloc request, this much padding is added to the sbrk
692 * When malloc_trim is called automatically from free(),
693 it is used as the `pad' argument.
695 In both cases, the actual amount of padding is rounded
696 so that the end of the arena is always a system page boundary.
698 The main reason for using padding is to avoid calling sbrk so
699 often. Having even a small pad greatly reduces the likelihood
700 that nearly every malloc request during program start-up (or
701 after trimming) will invoke sbrk, which needlessly wastes
704 Automatic rounding-up to page-size units is normally sufficient
705 to avoid measurable overhead, so the default is 0. However, in
706 systems where sbrk is relatively slow, it can pay to increase
707 this value, at the expense of carrying around more memory than
713 #ifndef DEFAULT_MMAP_THRESHOLD
714 #define DEFAULT_MMAP_THRESHOLD (128 * 1024)
719 M_MMAP_THRESHOLD is the request size threshold for using mmap()
720 to service a request. Requests of at least this size that cannot
721 be allocated using already-existing space will be serviced via mmap.
722 (If enough normal freed space already exists it is used instead.)
724 Using mmap segregates relatively large chunks of memory so that
725 they can be individually obtained and released from the host
726 system. A request serviced through mmap is never reused by any
727 other request (at least not directly; the system may just so
728 happen to remap successive requests to the same locations).
730 Segregating space in this way has the benefit that mmapped space
731 can ALWAYS be individually released back to the system, which
732 helps keep the system level memory demands of a long-lived
733 program low. Mapped memory can never become `locked' between
734 other chunks, as can happen with normally allocated chunks, which
735 menas that even trimming via malloc_trim would not release them.
737 However, it has the disadvantages that:
739 1. The space cannot be reclaimed, consolidated, and then
740 used to service later requests, as happens with normal chunks.
741 2. It can lead to more wastage because of mmap page alignment
743 3. It causes malloc performance to be more dependent on host
744 system memory management support routines which may vary in
745 implementation quality and may impose arbitrary
746 limitations. Generally, servicing a request via normal
747 malloc steps is faster than going through a system's mmap.
749 All together, these considerations should lead you to use mmap
750 only for relatively large requests.
756 #ifndef DEFAULT_MMAP_MAX
758 #define DEFAULT_MMAP_MAX (64)
760 #define DEFAULT_MMAP_MAX (0)
765 M_MMAP_MAX is the maximum number of requests to simultaneously
766 service using mmap. This parameter exists because:
768 1. Some systems have a limited number of internal tables for
770 2. In most systems, overreliance on mmap can degrade overall
772 3. If a program allocates many large regions, it is probably
773 better off using normal sbrk-based allocation routines that
774 can reclaim and reallocate normal heap memory. Using a
775 small value allows transition into this mode after the
776 first few allocations.
778 Setting to 0 disables all use of mmap. If HAVE_MMAP is not set,
779 the default value is 0, and attempts to set it to non-zero values
780 in mallopt will fail.
785 USE_DL_PREFIX will prefix all public routines with the string 'dl'.
786 Useful to quickly avoid procedure declaration conflicts and linker
787 symbol conflicts with existing memory allocation routines.
791 /* #define USE_DL_PREFIX */
796 Special defines for linux libc
798 Except when compiled using these special defines for Linux libc
799 using weak aliases, this malloc is NOT designed to work in
800 multithreaded applications. No semaphores or other concurrency
801 control are provided to ensure that multiple malloc or free calls
802 don't run at the same time, which could be disasterous. A single
803 semaphore could be used across malloc, realloc, and free (which is
804 essentially the effect of the linux weak alias approach). It would
805 be hard to obtain finer granularity.
810 #ifdef INTERNAL_LINUX_C_LIB
814 Void_t * __default_morecore_init (ptrdiff_t);
815 Void_t *(*__morecore)(ptrdiff_t) = __default_morecore_init;
819 Void_t * __default_morecore_init ();
820 Void_t *(*__morecore)() = __default_morecore_init;
824 #define MORECORE (*__morecore)
825 #define MORECORE_FAILURE 0
826 #define MORECORE_CLEARS 1
828 #else /* INTERNAL_LINUX_C_LIB */
831 extern Void_t* sbrk(ptrdiff_t);
833 extern Void_t* sbrk();
837 #define MORECORE sbrk
840 #ifndef MORECORE_FAILURE
841 #define MORECORE_FAILURE -1
844 #ifndef MORECORE_CLEARS
845 #define MORECORE_CLEARS 1
848 #endif /* INTERNAL_LINUX_C_LIB */
850 #if defined(INTERNAL_LINUX_C_LIB) && defined(__ELF__)
852 #define cALLOc __libc_calloc
853 #define fREe __libc_free
854 #define mALLOc __libc_malloc
855 #define mEMALIGn __libc_memalign
856 #define rEALLOc __libc_realloc
857 #define vALLOc __libc_valloc
858 #define pvALLOc __libc_pvalloc
859 #define mALLINFo __libc_mallinfo
860 #define mALLOPt __libc_mallopt
862 #pragma weak calloc = __libc_calloc
863 #pragma weak free = __libc_free
864 #pragma weak cfree = __libc_free
865 #pragma weak malloc = __libc_malloc
866 #pragma weak memalign = __libc_memalign
867 #pragma weak realloc = __libc_realloc
868 #pragma weak valloc = __libc_valloc
869 #pragma weak pvalloc = __libc_pvalloc
870 #pragma weak mallinfo = __libc_mallinfo
871 #pragma weak mallopt = __libc_mallopt
876 #define cALLOc dlcalloc
878 #define mALLOc dlmalloc
879 #define mEMALIGn dlmemalign
880 #define rEALLOc dlrealloc
881 #define vALLOc dlvalloc
882 #define pvALLOc dlpvalloc
883 #define mALLINFo dlmallinfo
884 #define mALLOPt dlmallopt
885 #else /* USE_DL_PREFIX */
886 #define cALLOc calloc
888 #define mALLOc malloc
889 #define mEMALIGn memalign
890 #define rEALLOc realloc
891 #define vALLOc valloc
892 #define pvALLOc pvalloc
893 #define mALLINFo mallinfo
894 #define mALLOPt mallopt
895 #endif /* USE_DL_PREFIX */
899 /* Public routines */
903 Void_t* mALLOc(size_t);
905 Void_t* rEALLOc(Void_t*, size_t);
906 Void_t* mEMALIGn(size_t, size_t);
907 Void_t* vALLOc(size_t);
908 Void_t* pvALLOc(size_t);
909 Void_t* cALLOc(size_t, size_t);
911 int malloc_trim(size_t);
912 size_t malloc_usable_size(Void_t*);
914 int mALLOPt(int, int);
915 struct mallinfo mALLINFo(void);
926 size_t malloc_usable_size();
929 struct mallinfo mALLINFo();
934 }; /* end of extern "C" */
937 /* ---------- To make a malloc.h, end cutting here ------------ */
938 #else /* Moved to malloc.h */
943 static void malloc_update_mallinfo (void);
944 void malloc_stats (void);
946 static void malloc_update_mallinfo ();
951 #endif /* 0 */ /* Moved to malloc.h */
953 DECLARE_GLOBAL_DATA_PTR;
956 Emulation of sbrk for WIN32
957 All code within the ifdef WIN32 is untested by me.
959 Thanks to Martin Fong and others for supplying this.
965 #define AlignPage(add) (((add) + (malloc_getpagesize-1)) & \
966 ~(malloc_getpagesize-1))
967 #define AlignPage64K(add) (((add) + (0x10000 - 1)) & ~(0x10000 - 1))
969 /* resrve 64MB to insure large contiguous space */
970 #define RESERVED_SIZE (1024*1024*64)
971 #define NEXT_SIZE (2048*1024)
972 #define TOP_MEMORY ((unsigned long)2*1024*1024*1024)
974 struct GmListElement;
975 typedef struct GmListElement GmListElement;
983 static GmListElement* head = 0;
984 static unsigned int gNextAddress = 0;
985 static unsigned int gAddressBase = 0;
986 static unsigned int gAllocatedSize = 0;
989 GmListElement* makeGmListElement (void* bas)
992 this = (GmListElement*)(void*)LocalAlloc (0, sizeof (GmListElement));
1006 assert ( (head == NULL) || (head->base == (void*)gAddressBase));
1007 if (gAddressBase && (gNextAddress - gAddressBase))
1009 rval = VirtualFree ((void*)gAddressBase,
1010 gNextAddress - gAddressBase,
1016 GmListElement* next = head->next;
1017 rval = VirtualFree (head->base, 0, MEM_RELEASE);
1025 void* findRegion (void* start_address, unsigned long size)
1027 MEMORY_BASIC_INFORMATION info;
1028 if (size >= TOP_MEMORY) return NULL;
1030 while ((unsigned long)start_address + size < TOP_MEMORY)
1032 VirtualQuery (start_address, &info, sizeof (info));
1033 if ((info.State == MEM_FREE) && (info.RegionSize >= size))
1034 return start_address;
1037 /* Requested region is not available so see if the */
1038 /* next region is available. Set 'start_address' */
1039 /* to the next region and call 'VirtualQuery()' */
1042 start_address = (char*)info.BaseAddress + info.RegionSize;
1044 /* Make sure we start looking for the next region */
1045 /* on the *next* 64K boundary. Otherwise, even if */
1046 /* the new region is free according to */
1047 /* 'VirtualQuery()', the subsequent call to */
1048 /* 'VirtualAlloc()' (which follows the call to */
1049 /* this routine in 'wsbrk()') will round *down* */
1050 /* the requested address to a 64K boundary which */
1051 /* we already know is an address in the */
1052 /* unavailable region. Thus, the subsequent call */
1053 /* to 'VirtualAlloc()' will fail and bring us back */
1054 /* here, causing us to go into an infinite loop. */
1057 (void *) AlignPage64K((unsigned long) start_address);
1065 void* wsbrk (long size)
1070 if (gAddressBase == 0)
1072 gAllocatedSize = max (RESERVED_SIZE, AlignPage (size));
1073 gNextAddress = gAddressBase =
1074 (unsigned int)VirtualAlloc (NULL, gAllocatedSize,
1075 MEM_RESERVE, PAGE_NOACCESS);
1076 } else if (AlignPage (gNextAddress + size) > (gAddressBase +
1079 long new_size = max (NEXT_SIZE, AlignPage (size));
1080 void* new_address = (void*)(gAddressBase+gAllocatedSize);
1083 new_address = findRegion (new_address, new_size);
1085 if (new_address == 0)
1088 gAddressBase = gNextAddress =
1089 (unsigned int)VirtualAlloc (new_address, new_size,
1090 MEM_RESERVE, PAGE_NOACCESS);
1091 /* repeat in case of race condition */
1092 /* The region that we found has been snagged */
1093 /* by another thread */
1095 while (gAddressBase == 0);
1097 assert (new_address == (void*)gAddressBase);
1099 gAllocatedSize = new_size;
1101 if (!makeGmListElement ((void*)gAddressBase))
1104 if ((size + gNextAddress) > AlignPage (gNextAddress))
1107 res = VirtualAlloc ((void*)AlignPage (gNextAddress),
1108 (size + gNextAddress -
1109 AlignPage (gNextAddress)),
1110 MEM_COMMIT, PAGE_READWRITE);
1114 tmp = (void*)gNextAddress;
1115 gNextAddress = (unsigned int)tmp + size;
1120 unsigned int alignedGoal = AlignPage (gNextAddress + size);
1121 /* Trim by releasing the virtual memory */
1122 if (alignedGoal >= gAddressBase)
1124 VirtualFree ((void*)alignedGoal, gNextAddress - alignedGoal,
1126 gNextAddress = gNextAddress + size;
1127 return (void*)gNextAddress;
1131 VirtualFree ((void*)gAddressBase, gNextAddress - gAddressBase,
1133 gNextAddress = gAddressBase;
1139 return (void*)gNextAddress;
1154 INTERNAL_SIZE_T prev_size; /* Size of previous chunk (if free). */
1155 INTERNAL_SIZE_T size; /* Size in bytes, including overhead. */
1156 struct malloc_chunk* fd; /* double links -- used only if free. */
1157 struct malloc_chunk* bk;
1160 typedef struct malloc_chunk* mchunkptr;
1164 malloc_chunk details:
1166 (The following includes lightly edited explanations by Colin Plumb.)
1168 Chunks of memory are maintained using a `boundary tag' method as
1169 described in e.g., Knuth or Standish. (See the paper by Paul
1170 Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
1171 survey of such techniques.) Sizes of free chunks are stored both
1172 in the front of each chunk and at the end. This makes
1173 consolidating fragmented chunks into bigger chunks very fast. The
1174 size fields also hold bits representing whether chunks are free or
1177 An allocated chunk looks like this:
1180 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1181 | Size of previous chunk, if allocated | |
1182 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1183 | Size of chunk, in bytes |P|
1184 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1185 | User data starts here... .
1187 . (malloc_usable_space() bytes) .
1189 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1191 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1194 Where "chunk" is the front of the chunk for the purpose of most of
1195 the malloc code, but "mem" is the pointer that is returned to the
1196 user. "Nextchunk" is the beginning of the next contiguous chunk.
1198 Chunks always begin on even word boundries, so the mem portion
1199 (which is returned to the user) is also on an even word boundary, and
1200 thus double-word aligned.
1202 Free chunks are stored in circular doubly-linked lists, and look like this:
1204 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1205 | Size of previous chunk |
1206 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1207 `head:' | Size of chunk, in bytes |P|
1208 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1209 | Forward pointer to next chunk in list |
1210 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1211 | Back pointer to previous chunk in list |
1212 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1213 | Unused space (may be 0 bytes long) .
1216 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1217 `foot:' | Size of chunk, in bytes |
1218 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1220 The P (PREV_INUSE) bit, stored in the unused low-order bit of the
1221 chunk size (which is always a multiple of two words), is an in-use
1222 bit for the *previous* chunk. If that bit is *clear*, then the
1223 word before the current chunk size contains the previous chunk
1224 size, and can be used to find the front of the previous chunk.
1225 (The very first chunk allocated always has this bit set,
1226 preventing access to non-existent (or non-owned) memory.)
1228 Note that the `foot' of the current chunk is actually represented
1229 as the prev_size of the NEXT chunk. (This makes it easier to
1230 deal with alignments etc).
1232 The two exceptions to all this are
1234 1. The special chunk `top', which doesn't bother using the
1235 trailing size field since there is no
1236 next contiguous chunk that would have to index off it. (After
1237 initialization, `top' is forced to always exist. If it would
1238 become less than MINSIZE bytes long, it is replenished via
1241 2. Chunks allocated via mmap, which have the second-lowest-order
1242 bit (IS_MMAPPED) set in their size fields. Because they are
1243 never merged or traversed from any other chunk, they have no
1244 foot size or inuse information.
1246 Available chunks are kept in any of several places (all declared below):
1248 * `av': An array of chunks serving as bin headers for consolidated
1249 chunks. Each bin is doubly linked. The bins are approximately
1250 proportionally (log) spaced. There are a lot of these bins
1251 (128). This may look excessive, but works very well in
1252 practice. All procedures maintain the invariant that no
1253 consolidated chunk physically borders another one. Chunks in
1254 bins are kept in size order, with ties going to the
1255 approximately least recently used chunk.
1257 The chunks in each bin are maintained in decreasing sorted order by
1258 size. This is irrelevant for the small bins, which all contain
1259 the same-sized chunks, but facilitates best-fit allocation for
1260 larger chunks. (These lists are just sequential. Keeping them in
1261 order almost never requires enough traversal to warrant using
1262 fancier ordered data structures.) Chunks of the same size are
1263 linked with the most recently freed at the front, and allocations
1264 are taken from the back. This results in LRU or FIFO allocation
1265 order, which tends to give each chunk an equal opportunity to be
1266 consolidated with adjacent freed chunks, resulting in larger free
1267 chunks and less fragmentation.
1269 * `top': The top-most available chunk (i.e., the one bordering the
1270 end of available memory) is treated specially. It is never
1271 included in any bin, is used only if no other chunk is
1272 available, and is released back to the system if it is very
1273 large (see M_TRIM_THRESHOLD).
1275 * `last_remainder': A bin holding only the remainder of the
1276 most recently split (non-top) chunk. This bin is checked
1277 before other non-fitting chunks, so as to provide better
1278 locality for runs of sequentially allocated chunks.
1280 * Implicitly, through the host system's memory mapping tables.
1281 If supported, requests greater than a threshold are usually
1282 serviced via calls to mmap, and then later released via munmap.
1286 /* sizes, alignments */
1288 #define SIZE_SZ (sizeof(INTERNAL_SIZE_T))
1289 #define MALLOC_ALIGNMENT (SIZE_SZ + SIZE_SZ)
1290 #define MALLOC_ALIGN_MASK (MALLOC_ALIGNMENT - 1)
1291 #define MINSIZE (sizeof(struct malloc_chunk))
1293 /* conversion from malloc headers to user pointers, and back */
1295 #define chunk2mem(p) ((Void_t*)((char*)(p) + 2*SIZE_SZ))
1296 #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ))
1298 /* pad request bytes into a usable size */
1300 #define request2size(req) \
1301 (((long)((req) + (SIZE_SZ + MALLOC_ALIGN_MASK)) < \
1302 (long)(MINSIZE + MALLOC_ALIGN_MASK)) ? MINSIZE : \
1303 (((req) + (SIZE_SZ + MALLOC_ALIGN_MASK)) & ~(MALLOC_ALIGN_MASK)))
1305 /* Check if m has acceptable alignment */
1307 #define aligned_OK(m) (((unsigned long)((m)) & (MALLOC_ALIGN_MASK)) == 0)
1313 Physical chunk operations
1317 /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */
1319 #define PREV_INUSE 0x1
1321 /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */
1323 #define IS_MMAPPED 0x2
1325 /* Bits to mask off when extracting size */
1327 #define SIZE_BITS (PREV_INUSE|IS_MMAPPED)
1330 /* Ptr to next physical malloc_chunk. */
1332 #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->size & ~PREV_INUSE) ))
1334 /* Ptr to previous physical malloc_chunk */
1336 #define prev_chunk(p)\
1337 ((mchunkptr)( ((char*)(p)) - ((p)->prev_size) ))
1340 /* Treat space at ptr + offset as a chunk */
1342 #define chunk_at_offset(p, s) ((mchunkptr)(((char*)(p)) + (s)))
1348 Dealing with use bits
1351 /* extract p's inuse bit */
1354 ((((mchunkptr)(((char*)(p))+((p)->size & ~PREV_INUSE)))->size) & PREV_INUSE)
1356 /* extract inuse bit of previous chunk */
1358 #define prev_inuse(p) ((p)->size & PREV_INUSE)
1360 /* check for mmap()'ed chunk */
1362 #define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED)
1364 /* set/clear chunk as in use without otherwise disturbing */
1366 #define set_inuse(p)\
1367 ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size |= PREV_INUSE
1369 #define clear_inuse(p)\
1370 ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size &= ~(PREV_INUSE)
1372 /* check/set/clear inuse bits in known places */
1374 #define inuse_bit_at_offset(p, s)\
1375 (((mchunkptr)(((char*)(p)) + (s)))->size & PREV_INUSE)
1377 #define set_inuse_bit_at_offset(p, s)\
1378 (((mchunkptr)(((char*)(p)) + (s)))->size |= PREV_INUSE)
1380 #define clear_inuse_bit_at_offset(p, s)\
1381 (((mchunkptr)(((char*)(p)) + (s)))->size &= ~(PREV_INUSE))
1387 Dealing with size fields
1390 /* Get size, ignoring use bits */
1392 #define chunksize(p) ((p)->size & ~(SIZE_BITS))
1394 /* Set size at head, without disturbing its use bit */
1396 #define set_head_size(p, s) ((p)->size = (((p)->size & PREV_INUSE) | (s)))
1398 /* Set size/use ignoring previous bits in header */
1400 #define set_head(p, s) ((p)->size = (s))
1402 /* Set size at footer (only when chunk is not in use) */
1404 #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_size = (s))
1413 The bins, `av_' are an array of pairs of pointers serving as the
1414 heads of (initially empty) doubly-linked lists of chunks, laid out
1415 in a way so that each pair can be treated as if it were in a
1416 malloc_chunk. (This way, the fd/bk offsets for linking bin heads
1417 and chunks are the same).
1419 Bins for sizes < 512 bytes contain chunks of all the same size, spaced
1420 8 bytes apart. Larger bins are approximately logarithmically
1421 spaced. (See the table below.) The `av_' array is never mentioned
1422 directly in the code, but instead via bin access macros.
1430 4 bins of size 32768
1431 2 bins of size 262144
1432 1 bin of size what's left
1434 There is actually a little bit of slop in the numbers in bin_index
1435 for the sake of speed. This makes no difference elsewhere.
1437 The special chunks `top' and `last_remainder' get their own bins,
1438 (this is implemented via yet more trickery with the av_ array),
1439 although `top' is never properly linked to its bin since it is
1440 always handled specially.
1444 #define NAV 128 /* number of bins */
1446 typedef struct malloc_chunk* mbinptr;
1450 #define bin_at(i) ((mbinptr)((char*)&(av_[2*(i) + 2]) - 2*SIZE_SZ))
1451 #define next_bin(b) ((mbinptr)((char*)(b) + 2 * sizeof(mbinptr)))
1452 #define prev_bin(b) ((mbinptr)((char*)(b) - 2 * sizeof(mbinptr)))
1455 The first 2 bins are never indexed. The corresponding av_ cells are instead
1456 used for bookkeeping. This is not to save space, but to simplify
1457 indexing, maintain locality, and avoid some initialization tests.
1460 #define top (av_[2]) /* The topmost chunk */
1461 #define last_remainder (bin_at(1)) /* remainder from last split */
1465 Because top initially points to its own bin with initial
1466 zero size, thus forcing extension on the first malloc request,
1467 we avoid having any special code in malloc to check whether
1468 it even exists yet. But we still need to in malloc_extend_top.
1471 #define initial_top ((mchunkptr)(bin_at(0)))
1473 /* Helper macro to initialize bins */
1475 #define IAV(i) bin_at(i), bin_at(i)
1477 static mbinptr av_[NAV * 2 + 2] = {
1479 IAV(0), IAV(1), IAV(2), IAV(3), IAV(4), IAV(5), IAV(6), IAV(7),
1480 IAV(8), IAV(9), IAV(10), IAV(11), IAV(12), IAV(13), IAV(14), IAV(15),
1481 IAV(16), IAV(17), IAV(18), IAV(19), IAV(20), IAV(21), IAV(22), IAV(23),
1482 IAV(24), IAV(25), IAV(26), IAV(27), IAV(28), IAV(29), IAV(30), IAV(31),
1483 IAV(32), IAV(33), IAV(34), IAV(35), IAV(36), IAV(37), IAV(38), IAV(39),
1484 IAV(40), IAV(41), IAV(42), IAV(43), IAV(44), IAV(45), IAV(46), IAV(47),
1485 IAV(48), IAV(49), IAV(50), IAV(51), IAV(52), IAV(53), IAV(54), IAV(55),
1486 IAV(56), IAV(57), IAV(58), IAV(59), IAV(60), IAV(61), IAV(62), IAV(63),
1487 IAV(64), IAV(65), IAV(66), IAV(67), IAV(68), IAV(69), IAV(70), IAV(71),
1488 IAV(72), IAV(73), IAV(74), IAV(75), IAV(76), IAV(77), IAV(78), IAV(79),
1489 IAV(80), IAV(81), IAV(82), IAV(83), IAV(84), IAV(85), IAV(86), IAV(87),
1490 IAV(88), IAV(89), IAV(90), IAV(91), IAV(92), IAV(93), IAV(94), IAV(95),
1491 IAV(96), IAV(97), IAV(98), IAV(99), IAV(100), IAV(101), IAV(102), IAV(103),
1492 IAV(104), IAV(105), IAV(106), IAV(107), IAV(108), IAV(109), IAV(110), IAV(111),
1493 IAV(112), IAV(113), IAV(114), IAV(115), IAV(116), IAV(117), IAV(118), IAV(119),
1494 IAV(120), IAV(121), IAV(122), IAV(123), IAV(124), IAV(125), IAV(126), IAV(127)
1497 void malloc_bin_reloc (void)
1499 unsigned long *p = (unsigned long *)(&av_[2]);
1501 for (i=2; i<(sizeof(av_)/sizeof(mbinptr)); ++i) {
1502 *p++ += gd->reloc_off;
1506 ulong mem_malloc_start = 0;
1507 ulong mem_malloc_end = 0;
1508 ulong mem_malloc_brk = 0;
1510 void *sbrk(ptrdiff_t increment)
1512 ulong old = mem_malloc_brk;
1513 ulong new = old + increment;
1515 if ((new < mem_malloc_start) || (new > mem_malloc_end))
1518 mem_malloc_brk = new;
1525 * x86 boards use a slightly different init sequence thus they implement
1526 * their own version of mem_malloc_init()
1528 void mem_malloc_init(ulong start, ulong size)
1530 mem_malloc_start = start;
1531 mem_malloc_end = start + size;
1532 mem_malloc_brk = start;
1534 memset((void *)mem_malloc_start, 0, size);
1538 /* field-extraction macros */
1540 #define first(b) ((b)->fd)
1541 #define last(b) ((b)->bk)
1547 #define bin_index(sz) \
1548 (((((unsigned long)(sz)) >> 9) == 0) ? (((unsigned long)(sz)) >> 3): \
1549 ((((unsigned long)(sz)) >> 9) <= 4) ? 56 + (((unsigned long)(sz)) >> 6): \
1550 ((((unsigned long)(sz)) >> 9) <= 20) ? 91 + (((unsigned long)(sz)) >> 9): \
1551 ((((unsigned long)(sz)) >> 9) <= 84) ? 110 + (((unsigned long)(sz)) >> 12): \
1552 ((((unsigned long)(sz)) >> 9) <= 340) ? 119 + (((unsigned long)(sz)) >> 15): \
1553 ((((unsigned long)(sz)) >> 9) <= 1364) ? 124 + (((unsigned long)(sz)) >> 18): \
1556 bins for chunks < 512 are all spaced 8 bytes apart, and hold
1557 identically sized chunks. This is exploited in malloc.
1560 #define MAX_SMALLBIN 63
1561 #define MAX_SMALLBIN_SIZE 512
1562 #define SMALLBIN_WIDTH 8
1564 #define smallbin_index(sz) (((unsigned long)(sz)) >> 3)
1567 Requests are `small' if both the corresponding and the next bin are small
1570 #define is_small_request(nb) (nb < MAX_SMALLBIN_SIZE - SMALLBIN_WIDTH)
1575 To help compensate for the large number of bins, a one-level index
1576 structure is used for bin-by-bin searching. `binblocks' is a
1577 one-word bitvector recording whether groups of BINBLOCKWIDTH bins
1578 have any (possibly) non-empty bins, so they can be skipped over
1579 all at once during during traversals. The bits are NOT always
1580 cleared as soon as all bins in a block are empty, but instead only
1581 when all are noticed to be empty during traversal in malloc.
1584 #define BINBLOCKWIDTH 4 /* bins per block */
1586 #define binblocks_r ((INTERNAL_SIZE_T)av_[1]) /* bitvector of nonempty blocks */
1587 #define binblocks_w (av_[1])
1589 /* bin<->block macros */
1591 #define idx2binblock(ix) ((unsigned)1 << (ix / BINBLOCKWIDTH))
1592 #define mark_binblock(ii) (binblocks_w = (mbinptr)(binblocks_r | idx2binblock(ii)))
1593 #define clear_binblock(ii) (binblocks_w = (mbinptr)(binblocks_r & ~(idx2binblock(ii))))
1599 /* Other static bookkeeping data */
1601 /* variables holding tunable values */
1603 static unsigned long trim_threshold = DEFAULT_TRIM_THRESHOLD;
1604 static unsigned long top_pad = DEFAULT_TOP_PAD;
1605 static unsigned int n_mmaps_max = DEFAULT_MMAP_MAX;
1606 static unsigned long mmap_threshold = DEFAULT_MMAP_THRESHOLD;
1608 /* The first value returned from sbrk */
1609 static char* sbrk_base = (char*)(-1);
1611 /* The maximum memory obtained from system via sbrk */
1612 static unsigned long max_sbrked_mem = 0;
1614 /* The maximum via either sbrk or mmap */
1615 static unsigned long max_total_mem = 0;
1617 /* internal working copy of mallinfo */
1618 static struct mallinfo current_mallinfo = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
1620 /* The total memory obtained from system via sbrk */
1621 #define sbrked_mem (current_mallinfo.arena)
1623 /* Tracking mmaps */
1626 static unsigned int n_mmaps = 0;
1628 static unsigned long mmapped_mem = 0;
1630 static unsigned int max_n_mmaps = 0;
1631 static unsigned long max_mmapped_mem = 0;
1644 These routines make a number of assertions about the states
1645 of data structures that should be true at all times. If any
1646 are not true, it's very likely that a user program has somehow
1647 trashed memory. (It's also possible that there is a coding error
1648 in malloc. In which case, please report it!)
1652 static void do_check_chunk(mchunkptr p)
1654 static void do_check_chunk(p) mchunkptr p;
1657 #if 0 /* causes warnings because assert() is off */
1658 INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1661 /* No checkable chunk is mmapped */
1662 assert(!chunk_is_mmapped(p));
1664 /* Check for legal address ... */
1665 assert((char*)p >= sbrk_base);
1667 assert((char*)p + sz <= (char*)top);
1669 assert((char*)p + sz <= sbrk_base + sbrked_mem);
1675 static void do_check_free_chunk(mchunkptr p)
1677 static void do_check_free_chunk(p) mchunkptr p;
1680 INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1681 #if 0 /* causes warnings because assert() is off */
1682 mchunkptr next = chunk_at_offset(p, sz);
1687 /* Check whether it claims to be free ... */
1690 /* Unless a special marker, must have OK fields */
1691 if ((long)sz >= (long)MINSIZE)
1693 assert((sz & MALLOC_ALIGN_MASK) == 0);
1694 assert(aligned_OK(chunk2mem(p)));
1695 /* ... matching footer field */
1696 assert(next->prev_size == sz);
1697 /* ... and is fully consolidated */
1698 assert(prev_inuse(p));
1699 assert (next == top || inuse(next));
1701 /* ... and has minimally sane links */
1702 assert(p->fd->bk == p);
1703 assert(p->bk->fd == p);
1705 else /* markers are always of size SIZE_SZ */
1706 assert(sz == SIZE_SZ);
1710 static void do_check_inuse_chunk(mchunkptr p)
1712 static void do_check_inuse_chunk(p) mchunkptr p;
1715 mchunkptr next = next_chunk(p);
1718 /* Check whether it claims to be in use ... */
1721 /* ... and is surrounded by OK chunks.
1722 Since more things can be checked with free chunks than inuse ones,
1723 if an inuse chunk borders them and debug is on, it's worth doing them.
1727 mchunkptr prv = prev_chunk(p);
1728 assert(next_chunk(prv) == p);
1729 do_check_free_chunk(prv);
1733 assert(prev_inuse(next));
1734 assert(chunksize(next) >= MINSIZE);
1736 else if (!inuse(next))
1737 do_check_free_chunk(next);
1742 static void do_check_malloced_chunk(mchunkptr p, INTERNAL_SIZE_T s)
1744 static void do_check_malloced_chunk(p, s) mchunkptr p; INTERNAL_SIZE_T s;
1747 #if 0 /* causes warnings because assert() is off */
1748 INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1752 do_check_inuse_chunk(p);
1754 /* Legal size ... */
1755 assert((long)sz >= (long)MINSIZE);
1756 assert((sz & MALLOC_ALIGN_MASK) == 0);
1758 assert(room < (long)MINSIZE);
1760 /* ... and alignment */
1761 assert(aligned_OK(chunk2mem(p)));
1764 /* ... and was allocated at front of an available chunk */
1765 assert(prev_inuse(p));
1770 #define check_free_chunk(P) do_check_free_chunk(P)
1771 #define check_inuse_chunk(P) do_check_inuse_chunk(P)
1772 #define check_chunk(P) do_check_chunk(P)
1773 #define check_malloced_chunk(P,N) do_check_malloced_chunk(P,N)
1775 #define check_free_chunk(P)
1776 #define check_inuse_chunk(P)
1777 #define check_chunk(P)
1778 #define check_malloced_chunk(P,N)
1784 Macro-based internal utilities
1789 Linking chunks in bin lists.
1790 Call these only with variables, not arbitrary expressions, as arguments.
1794 Place chunk p of size s in its bin, in size order,
1795 putting it ahead of others of same size.
1799 #define frontlink(P, S, IDX, BK, FD) \
1801 if (S < MAX_SMALLBIN_SIZE) \
1803 IDX = smallbin_index(S); \
1804 mark_binblock(IDX); \
1809 FD->bk = BK->fd = P; \
1813 IDX = bin_index(S); \
1816 if (FD == BK) mark_binblock(IDX); \
1819 while (FD != BK && S < chunksize(FD)) FD = FD->fd; \
1824 FD->bk = BK->fd = P; \
1829 /* take a chunk off a list */
1831 #define unlink(P, BK, FD) \
1839 /* Place p as the last remainder */
1841 #define link_last_remainder(P) \
1843 last_remainder->fd = last_remainder->bk = P; \
1844 P->fd = P->bk = last_remainder; \
1847 /* Clear the last_remainder bin */
1849 #define clear_last_remainder \
1850 (last_remainder->fd = last_remainder->bk = last_remainder)
1856 /* Routines dealing with mmap(). */
1861 static mchunkptr mmap_chunk(size_t size)
1863 static mchunkptr mmap_chunk(size) size_t size;
1866 size_t page_mask = malloc_getpagesize - 1;
1869 #ifndef MAP_ANONYMOUS
1873 if(n_mmaps >= n_mmaps_max) return 0; /* too many regions */
1875 /* For mmapped chunks, the overhead is one SIZE_SZ unit larger, because
1876 * there is no following chunk whose prev_size field could be used.
1878 size = (size + SIZE_SZ + page_mask) & ~page_mask;
1880 #ifdef MAP_ANONYMOUS
1881 p = (mchunkptr)mmap(0, size, PROT_READ|PROT_WRITE,
1882 MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1883 #else /* !MAP_ANONYMOUS */
1886 fd = open("/dev/zero", O_RDWR);
1887 if(fd < 0) return 0;
1889 p = (mchunkptr)mmap(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0);
1892 if(p == (mchunkptr)-1) return 0;
1895 if (n_mmaps > max_n_mmaps) max_n_mmaps = n_mmaps;
1897 /* We demand that eight bytes into a page must be 8-byte aligned. */
1898 assert(aligned_OK(chunk2mem(p)));
1900 /* The offset to the start of the mmapped region is stored
1901 * in the prev_size field of the chunk; normally it is zero,
1902 * but that can be changed in memalign().
1905 set_head(p, size|IS_MMAPPED);
1907 mmapped_mem += size;
1908 if ((unsigned long)mmapped_mem > (unsigned long)max_mmapped_mem)
1909 max_mmapped_mem = mmapped_mem;
1910 if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem)
1911 max_total_mem = mmapped_mem + sbrked_mem;
1916 static void munmap_chunk(mchunkptr p)
1918 static void munmap_chunk(p) mchunkptr p;
1921 INTERNAL_SIZE_T size = chunksize(p);
1924 assert (chunk_is_mmapped(p));
1925 assert(! ((char*)p >= sbrk_base && (char*)p < sbrk_base + sbrked_mem));
1926 assert((n_mmaps > 0));
1927 assert(((p->prev_size + size) & (malloc_getpagesize-1)) == 0);
1930 mmapped_mem -= (size + p->prev_size);
1932 ret = munmap((char *)p - p->prev_size, size + p->prev_size);
1934 /* munmap returns non-zero on failure */
1941 static mchunkptr mremap_chunk(mchunkptr p, size_t new_size)
1943 static mchunkptr mremap_chunk(p, new_size) mchunkptr p; size_t new_size;
1946 size_t page_mask = malloc_getpagesize - 1;
1947 INTERNAL_SIZE_T offset = p->prev_size;
1948 INTERNAL_SIZE_T size = chunksize(p);
1951 assert (chunk_is_mmapped(p));
1952 assert(! ((char*)p >= sbrk_base && (char*)p < sbrk_base + sbrked_mem));
1953 assert((n_mmaps > 0));
1954 assert(((size + offset) & (malloc_getpagesize-1)) == 0);
1956 /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */
1957 new_size = (new_size + offset + SIZE_SZ + page_mask) & ~page_mask;
1959 cp = (char *)mremap((char *)p - offset, size + offset, new_size, 1);
1961 if (cp == (char *)-1) return 0;
1963 p = (mchunkptr)(cp + offset);
1965 assert(aligned_OK(chunk2mem(p)));
1967 assert((p->prev_size == offset));
1968 set_head(p, (new_size - offset)|IS_MMAPPED);
1970 mmapped_mem -= size + offset;
1971 mmapped_mem += new_size;
1972 if ((unsigned long)mmapped_mem > (unsigned long)max_mmapped_mem)
1973 max_mmapped_mem = mmapped_mem;
1974 if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem)
1975 max_total_mem = mmapped_mem + sbrked_mem;
1979 #endif /* HAVE_MREMAP */
1981 #endif /* HAVE_MMAP */
1987 Extend the top-most chunk by obtaining memory from system.
1988 Main interface to sbrk (but see also malloc_trim).
1992 static void malloc_extend_top(INTERNAL_SIZE_T nb)
1994 static void malloc_extend_top(nb) INTERNAL_SIZE_T nb;
1997 char* brk; /* return value from sbrk */
1998 INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of sbrked space */
1999 INTERNAL_SIZE_T correction; /* bytes for 2nd sbrk call */
2000 char* new_brk; /* return of 2nd sbrk call */
2001 INTERNAL_SIZE_T top_size; /* new size of top chunk */
2003 mchunkptr old_top = top; /* Record state of old top */
2004 INTERNAL_SIZE_T old_top_size = chunksize(old_top);
2005 char* old_end = (char*)(chunk_at_offset(old_top, old_top_size));
2007 /* Pad request with top_pad plus minimal overhead */
2009 INTERNAL_SIZE_T sbrk_size = nb + top_pad + MINSIZE;
2010 unsigned long pagesz = malloc_getpagesize;
2012 /* If not the first time through, round to preserve page boundary */
2013 /* Otherwise, we need to correct to a page size below anyway. */
2014 /* (We also correct below if an intervening foreign sbrk call.) */
2016 if (sbrk_base != (char*)(-1))
2017 sbrk_size = (sbrk_size + (pagesz - 1)) & ~(pagesz - 1);
2019 brk = (char*)(MORECORE (sbrk_size));
2021 /* Fail if sbrk failed or if a foreign sbrk call killed our space */
2022 if (brk == (char*)(MORECORE_FAILURE) ||
2023 (brk < old_end && old_top != initial_top))
2026 sbrked_mem += sbrk_size;
2028 if (brk == old_end) /* can just add bytes to current top */
2030 top_size = sbrk_size + old_top_size;
2031 set_head(top, top_size | PREV_INUSE);
2035 if (sbrk_base == (char*)(-1)) /* First time through. Record base */
2037 else /* Someone else called sbrk(). Count those bytes as sbrked_mem. */
2038 sbrked_mem += brk - (char*)old_end;
2040 /* Guarantee alignment of first new chunk made from this space */
2041 front_misalign = (unsigned long)chunk2mem(brk) & MALLOC_ALIGN_MASK;
2042 if (front_misalign > 0)
2044 correction = (MALLOC_ALIGNMENT) - front_misalign;
2050 /* Guarantee the next brk will be at a page boundary */
2052 correction += ((((unsigned long)(brk + sbrk_size))+(pagesz-1)) &
2053 ~(pagesz - 1)) - ((unsigned long)(brk + sbrk_size));
2055 /* Allocate correction */
2056 new_brk = (char*)(MORECORE (correction));
2057 if (new_brk == (char*)(MORECORE_FAILURE)) return;
2059 sbrked_mem += correction;
2061 top = (mchunkptr)brk;
2062 top_size = new_brk - brk + correction;
2063 set_head(top, top_size | PREV_INUSE);
2065 if (old_top != initial_top)
2068 /* There must have been an intervening foreign sbrk call. */
2069 /* A double fencepost is necessary to prevent consolidation */
2071 /* If not enough space to do this, then user did something very wrong */
2072 if (old_top_size < MINSIZE)
2074 set_head(top, PREV_INUSE); /* will force null return from malloc */
2078 /* Also keep size a multiple of MALLOC_ALIGNMENT */
2079 old_top_size = (old_top_size - 3*SIZE_SZ) & ~MALLOC_ALIGN_MASK;
2080 set_head_size(old_top, old_top_size);
2081 chunk_at_offset(old_top, old_top_size )->size =
2083 chunk_at_offset(old_top, old_top_size + SIZE_SZ)->size =
2085 /* If possible, release the rest. */
2086 if (old_top_size >= MINSIZE)
2087 fREe(chunk2mem(old_top));
2091 if ((unsigned long)sbrked_mem > (unsigned long)max_sbrked_mem)
2092 max_sbrked_mem = sbrked_mem;
2093 if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem)
2094 max_total_mem = mmapped_mem + sbrked_mem;
2096 /* We always land on a page boundary */
2097 assert(((unsigned long)((char*)top + top_size) & (pagesz - 1)) == 0);
2103 /* Main public routines */
2109 The requested size is first converted into a usable form, `nb'.
2110 This currently means to add 4 bytes overhead plus possibly more to
2111 obtain 8-byte alignment and/or to obtain a size of at least
2112 MINSIZE (currently 16 bytes), the smallest allocatable size.
2113 (All fits are considered `exact' if they are within MINSIZE bytes.)
2115 From there, the first successful of the following steps is taken:
2117 1. The bin corresponding to the request size is scanned, and if
2118 a chunk of exactly the right size is found, it is taken.
2120 2. The most recently remaindered chunk is used if it is big
2121 enough. This is a form of (roving) first fit, used only in
2122 the absence of exact fits. Runs of consecutive requests use
2123 the remainder of the chunk used for the previous such request
2124 whenever possible. This limited use of a first-fit style
2125 allocation strategy tends to give contiguous chunks
2126 coextensive lifetimes, which improves locality and can reduce
2127 fragmentation in the long run.
2129 3. Other bins are scanned in increasing size order, using a
2130 chunk big enough to fulfill the request, and splitting off
2131 any remainder. This search is strictly by best-fit; i.e.,
2132 the smallest (with ties going to approximately the least
2133 recently used) chunk that fits is selected.
2135 4. If large enough, the chunk bordering the end of memory
2136 (`top') is split off. (This use of `top' is in accord with
2137 the best-fit search rule. In effect, `top' is treated as
2138 larger (and thus less well fitting) than any other available
2139 chunk since it can be extended to be as large as necessary
2140 (up to system limitations).
2142 5. If the request size meets the mmap threshold and the
2143 system supports mmap, and there are few enough currently
2144 allocated mmapped regions, and a call to mmap succeeds,
2145 the request is allocated via direct memory mapping.
2147 6. Otherwise, the top of memory is extended by
2148 obtaining more space from the system (normally using sbrk,
2149 but definable to anything else via the MORECORE macro).
2150 Memory is gathered from the system (in system page-sized
2151 units) in a way that allows chunks obtained across different
2152 sbrk calls to be consolidated, but does not require
2153 contiguous memory. Thus, it should be safe to intersperse
2154 mallocs with other sbrk calls.
2157 All allocations are made from the the `lowest' part of any found
2158 chunk. (The implementation invariant is that prev_inuse is
2159 always true of any allocated chunk; i.e., that each allocated
2160 chunk borders either a previously allocated and still in-use chunk,
2161 or the base of its memory arena.)
2166 Void_t* mALLOc(size_t bytes)
2168 Void_t* mALLOc(bytes) size_t bytes;
2171 mchunkptr victim; /* inspected/selected chunk */
2172 INTERNAL_SIZE_T victim_size; /* its size */
2173 int idx; /* index for bin traversal */
2174 mbinptr bin; /* associated bin */
2175 mchunkptr remainder; /* remainder from a split */
2176 long remainder_size; /* its size */
2177 int remainder_index; /* its bin index */
2178 unsigned long block; /* block traverser bit */
2179 int startidx; /* first bin of a traversed block */
2180 mchunkptr fwd; /* misc temp for linking */
2181 mchunkptr bck; /* misc temp for linking */
2182 mbinptr q; /* misc temp */
2186 if ((long)bytes < 0) return 0;
2188 nb = request2size(bytes); /* padded request size; */
2190 /* Check for exact match in a bin */
2192 if (is_small_request(nb)) /* Faster version for small requests */
2194 idx = smallbin_index(nb);
2196 /* No traversal or size check necessary for small bins. */
2201 /* Also scan the next one, since it would have a remainder < MINSIZE */
2209 victim_size = chunksize(victim);
2210 unlink(victim, bck, fwd);
2211 set_inuse_bit_at_offset(victim, victim_size);
2212 check_malloced_chunk(victim, nb);
2213 return chunk2mem(victim);
2216 idx += 2; /* Set for bin scan below. We've already scanned 2 bins. */
2221 idx = bin_index(nb);
2224 for (victim = last(bin); victim != bin; victim = victim->bk)
2226 victim_size = chunksize(victim);
2227 remainder_size = victim_size - nb;
2229 if (remainder_size >= (long)MINSIZE) /* too big */
2231 --idx; /* adjust to rescan below after checking last remainder */
2235 else if (remainder_size >= 0) /* exact fit */
2237 unlink(victim, bck, fwd);
2238 set_inuse_bit_at_offset(victim, victim_size);
2239 check_malloced_chunk(victim, nb);
2240 return chunk2mem(victim);
2248 /* Try to use the last split-off remainder */
2250 if ( (victim = last_remainder->fd) != last_remainder)
2252 victim_size = chunksize(victim);
2253 remainder_size = victim_size - nb;
2255 if (remainder_size >= (long)MINSIZE) /* re-split */
2257 remainder = chunk_at_offset(victim, nb);
2258 set_head(victim, nb | PREV_INUSE);
2259 link_last_remainder(remainder);
2260 set_head(remainder, remainder_size | PREV_INUSE);
2261 set_foot(remainder, remainder_size);
2262 check_malloced_chunk(victim, nb);
2263 return chunk2mem(victim);
2266 clear_last_remainder;
2268 if (remainder_size >= 0) /* exhaust */
2270 set_inuse_bit_at_offset(victim, victim_size);
2271 check_malloced_chunk(victim, nb);
2272 return chunk2mem(victim);
2275 /* Else place in bin */
2277 frontlink(victim, victim_size, remainder_index, bck, fwd);
2281 If there are any possibly nonempty big-enough blocks,
2282 search for best fitting chunk by scanning bins in blockwidth units.
2285 if ( (block = idx2binblock(idx)) <= binblocks_r)
2288 /* Get to the first marked block */
2290 if ( (block & binblocks_r) == 0)
2292 /* force to an even block boundary */
2293 idx = (idx & ~(BINBLOCKWIDTH - 1)) + BINBLOCKWIDTH;
2295 while ((block & binblocks_r) == 0)
2297 idx += BINBLOCKWIDTH;
2302 /* For each possibly nonempty block ... */
2305 startidx = idx; /* (track incomplete blocks) */
2306 q = bin = bin_at(idx);
2308 /* For each bin in this block ... */
2311 /* Find and use first big enough chunk ... */
2313 for (victim = last(bin); victim != bin; victim = victim->bk)
2315 victim_size = chunksize(victim);
2316 remainder_size = victim_size - nb;
2318 if (remainder_size >= (long)MINSIZE) /* split */
2320 remainder = chunk_at_offset(victim, nb);
2321 set_head(victim, nb | PREV_INUSE);
2322 unlink(victim, bck, fwd);
2323 link_last_remainder(remainder);
2324 set_head(remainder, remainder_size | PREV_INUSE);
2325 set_foot(remainder, remainder_size);
2326 check_malloced_chunk(victim, nb);
2327 return chunk2mem(victim);
2330 else if (remainder_size >= 0) /* take */
2332 set_inuse_bit_at_offset(victim, victim_size);
2333 unlink(victim, bck, fwd);
2334 check_malloced_chunk(victim, nb);
2335 return chunk2mem(victim);
2340 bin = next_bin(bin);
2342 } while ((++idx & (BINBLOCKWIDTH - 1)) != 0);
2344 /* Clear out the block bit. */
2346 do /* Possibly backtrack to try to clear a partial block */
2348 if ((startidx & (BINBLOCKWIDTH - 1)) == 0)
2350 av_[1] = (mbinptr)(binblocks_r & ~block);
2355 } while (first(q) == q);
2357 /* Get to the next possibly nonempty block */
2359 if ( (block <<= 1) <= binblocks_r && (block != 0) )
2361 while ((block & binblocks_r) == 0)
2363 idx += BINBLOCKWIDTH;
2373 /* Try to use top chunk */
2375 /* Require that there be a remainder, ensuring top always exists */
2376 if ( (remainder_size = chunksize(top) - nb) < (long)MINSIZE)
2380 /* If big and would otherwise need to extend, try to use mmap instead */
2381 if ((unsigned long)nb >= (unsigned long)mmap_threshold &&
2382 (victim = mmap_chunk(nb)) != 0)
2383 return chunk2mem(victim);
2387 malloc_extend_top(nb);
2388 if ( (remainder_size = chunksize(top) - nb) < (long)MINSIZE)
2389 return 0; /* propagate failure */
2393 set_head(victim, nb | PREV_INUSE);
2394 top = chunk_at_offset(victim, nb);
2395 set_head(top, remainder_size | PREV_INUSE);
2396 check_malloced_chunk(victim, nb);
2397 return chunk2mem(victim);
2410 1. free(0) has no effect.
2412 2. If the chunk was allocated via mmap, it is release via munmap().
2414 3. If a returned chunk borders the current high end of memory,
2415 it is consolidated into the top, and if the total unused
2416 topmost memory exceeds the trim threshold, malloc_trim is
2419 4. Other chunks are consolidated as they arrive, and
2420 placed in corresponding bins. (This includes the case of
2421 consolidating with the current `last_remainder').
2427 void fREe(Void_t* mem)
2429 void fREe(mem) Void_t* mem;
2432 mchunkptr p; /* chunk corresponding to mem */
2433 INTERNAL_SIZE_T hd; /* its head field */
2434 INTERNAL_SIZE_T sz; /* its size */
2435 int idx; /* its bin index */
2436 mchunkptr next; /* next contiguous chunk */
2437 INTERNAL_SIZE_T nextsz; /* its size */
2438 INTERNAL_SIZE_T prevsz; /* size of previous contiguous chunk */
2439 mchunkptr bck; /* misc temp for linking */
2440 mchunkptr fwd; /* misc temp for linking */
2441 int islr; /* track whether merging with last_remainder */
2443 if (mem == 0) /* free(0) has no effect */
2450 if (hd & IS_MMAPPED) /* release mmapped memory. */
2457 check_inuse_chunk(p);
2459 sz = hd & ~PREV_INUSE;
2460 next = chunk_at_offset(p, sz);
2461 nextsz = chunksize(next);
2463 if (next == top) /* merge with top */
2467 if (!(hd & PREV_INUSE)) /* consolidate backward */
2469 prevsz = p->prev_size;
2470 p = chunk_at_offset(p, -((long) prevsz));
2472 unlink(p, bck, fwd);
2475 set_head(p, sz | PREV_INUSE);
2477 if ((unsigned long)(sz) >= (unsigned long)trim_threshold)
2478 malloc_trim(top_pad);
2482 set_head(next, nextsz); /* clear inuse bit */
2486 if (!(hd & PREV_INUSE)) /* consolidate backward */
2488 prevsz = p->prev_size;
2489 p = chunk_at_offset(p, -((long) prevsz));
2492 if (p->fd == last_remainder) /* keep as last_remainder */
2495 unlink(p, bck, fwd);
2498 if (!(inuse_bit_at_offset(next, nextsz))) /* consolidate forward */
2502 if (!islr && next->fd == last_remainder) /* re-insert last_remainder */
2505 link_last_remainder(p);
2508 unlink(next, bck, fwd);
2512 set_head(p, sz | PREV_INUSE);
2515 frontlink(p, sz, idx, bck, fwd);
2526 Chunks that were obtained via mmap cannot be extended or shrunk
2527 unless HAVE_MREMAP is defined, in which case mremap is used.
2528 Otherwise, if their reallocation is for additional space, they are
2529 copied. If for less, they are just left alone.
2531 Otherwise, if the reallocation is for additional space, and the
2532 chunk can be extended, it is, else a malloc-copy-free sequence is
2533 taken. There are several different ways that a chunk could be
2534 extended. All are tried:
2536 * Extending forward into following adjacent free chunk.
2537 * Shifting backwards, joining preceding adjacent space
2538 * Both shifting backwards and extending forward.
2539 * Extending into newly sbrked space
2541 Unless the #define REALLOC_ZERO_BYTES_FREES is set, realloc with a
2542 size argument of zero (re)allocates a minimum-sized chunk.
2544 If the reallocation is for less space, and the new request is for
2545 a `small' (<512 bytes) size, then the newly unused space is lopped
2548 The old unix realloc convention of allowing the last-free'd chunk
2549 to be used as an argument to realloc is no longer supported.
2550 I don't know of any programs still relying on this feature,
2551 and allowing it would also allow too many other incorrect
2552 usages of realloc to be sensible.
2559 Void_t* rEALLOc(Void_t* oldmem, size_t bytes)
2561 Void_t* rEALLOc(oldmem, bytes) Void_t* oldmem; size_t bytes;
2564 INTERNAL_SIZE_T nb; /* padded request size */
2566 mchunkptr oldp; /* chunk corresponding to oldmem */
2567 INTERNAL_SIZE_T oldsize; /* its size */
2569 mchunkptr newp; /* chunk to return */
2570 INTERNAL_SIZE_T newsize; /* its size */
2571 Void_t* newmem; /* corresponding user mem */
2573 mchunkptr next; /* next contiguous chunk after oldp */
2574 INTERNAL_SIZE_T nextsize; /* its size */
2576 mchunkptr prev; /* previous contiguous chunk before oldp */
2577 INTERNAL_SIZE_T prevsize; /* its size */
2579 mchunkptr remainder; /* holds split off extra space from newp */
2580 INTERNAL_SIZE_T remainder_size; /* its size */
2582 mchunkptr bck; /* misc temp for linking */
2583 mchunkptr fwd; /* misc temp for linking */
2585 #ifdef REALLOC_ZERO_BYTES_FREES
2586 if (bytes == 0) { fREe(oldmem); return 0; }
2589 if ((long)bytes < 0) return 0;
2591 /* realloc of null is supposed to be same as malloc */
2592 if (oldmem == 0) return mALLOc(bytes);
2594 newp = oldp = mem2chunk(oldmem);
2595 newsize = oldsize = chunksize(oldp);
2598 nb = request2size(bytes);
2601 if (chunk_is_mmapped(oldp))
2604 newp = mremap_chunk(oldp, nb);
2605 if(newp) return chunk2mem(newp);
2607 /* Note the extra SIZE_SZ overhead. */
2608 if(oldsize - SIZE_SZ >= nb) return oldmem; /* do nothing */
2609 /* Must alloc, copy, free. */
2610 newmem = mALLOc(bytes);
2611 if (newmem == 0) return 0; /* propagate failure */
2612 MALLOC_COPY(newmem, oldmem, oldsize - 2*SIZE_SZ);
2618 check_inuse_chunk(oldp);
2620 if ((long)(oldsize) < (long)(nb))
2623 /* Try expanding forward */
2625 next = chunk_at_offset(oldp, oldsize);
2626 if (next == top || !inuse(next))
2628 nextsize = chunksize(next);
2630 /* Forward into top only if a remainder */
2633 if ((long)(nextsize + newsize) >= (long)(nb + MINSIZE))
2635 newsize += nextsize;
2636 top = chunk_at_offset(oldp, nb);
2637 set_head(top, (newsize - nb) | PREV_INUSE);
2638 set_head_size(oldp, nb);
2639 return chunk2mem(oldp);
2643 /* Forward into next chunk */
2644 else if (((long)(nextsize + newsize) >= (long)(nb)))
2646 unlink(next, bck, fwd);
2647 newsize += nextsize;
2657 /* Try shifting backwards. */
2659 if (!prev_inuse(oldp))
2661 prev = prev_chunk(oldp);
2662 prevsize = chunksize(prev);
2664 /* try forward + backward first to save a later consolidation */
2671 if ((long)(nextsize + prevsize + newsize) >= (long)(nb + MINSIZE))
2673 unlink(prev, bck, fwd);
2675 newsize += prevsize + nextsize;
2676 newmem = chunk2mem(newp);
2677 MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2678 top = chunk_at_offset(newp, nb);
2679 set_head(top, (newsize - nb) | PREV_INUSE);
2680 set_head_size(newp, nb);
2685 /* into next chunk */
2686 else if (((long)(nextsize + prevsize + newsize) >= (long)(nb)))
2688 unlink(next, bck, fwd);
2689 unlink(prev, bck, fwd);
2691 newsize += nextsize + prevsize;
2692 newmem = chunk2mem(newp);
2693 MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2699 if (prev != 0 && (long)(prevsize + newsize) >= (long)nb)
2701 unlink(prev, bck, fwd);
2703 newsize += prevsize;
2704 newmem = chunk2mem(newp);
2705 MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2712 newmem = mALLOc (bytes);
2714 if (newmem == 0) /* propagate failure */
2717 /* Avoid copy if newp is next chunk after oldp. */
2718 /* (This can only happen when new chunk is sbrk'ed.) */
2720 if ( (newp = mem2chunk(newmem)) == next_chunk(oldp))
2722 newsize += chunksize(newp);
2727 /* Otherwise copy, free, and exit */
2728 MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2734 split: /* split off extra room in old or expanded chunk */
2736 if (newsize - nb >= MINSIZE) /* split off remainder */
2738 remainder = chunk_at_offset(newp, nb);
2739 remainder_size = newsize - nb;
2740 set_head_size(newp, nb);
2741 set_head(remainder, remainder_size | PREV_INUSE);
2742 set_inuse_bit_at_offset(remainder, remainder_size);
2743 fREe(chunk2mem(remainder)); /* let free() deal with it */
2747 set_head_size(newp, newsize);
2748 set_inuse_bit_at_offset(newp, newsize);
2751 check_inuse_chunk(newp);
2752 return chunk2mem(newp);
2762 memalign requests more than enough space from malloc, finds a spot
2763 within that chunk that meets the alignment request, and then
2764 possibly frees the leading and trailing space.
2766 The alignment argument must be a power of two. This property is not
2767 checked by memalign, so misuse may result in random runtime errors.
2769 8-byte alignment is guaranteed by normal malloc calls, so don't
2770 bother calling memalign with an argument of 8 or less.
2772 Overreliance on memalign is a sure way to fragment space.
2778 Void_t* mEMALIGn(size_t alignment, size_t bytes)
2780 Void_t* mEMALIGn(alignment, bytes) size_t alignment; size_t bytes;
2783 INTERNAL_SIZE_T nb; /* padded request size */
2784 char* m; /* memory returned by malloc call */
2785 mchunkptr p; /* corresponding chunk */
2786 char* brk; /* alignment point within p */
2787 mchunkptr newp; /* chunk to return */
2788 INTERNAL_SIZE_T newsize; /* its size */
2789 INTERNAL_SIZE_T leadsize; /* leading space befor alignment point */
2790 mchunkptr remainder; /* spare room at end to split off */
2791 long remainder_size; /* its size */
2793 if ((long)bytes < 0) return 0;
2795 /* If need less alignment than we give anyway, just relay to malloc */
2797 if (alignment <= MALLOC_ALIGNMENT) return mALLOc(bytes);
2799 /* Otherwise, ensure that it is at least a minimum chunk size */
2801 if (alignment < MINSIZE) alignment = MINSIZE;
2803 /* Call malloc with worst case padding to hit alignment. */
2805 nb = request2size(bytes);
2806 m = (char*)(mALLOc(nb + alignment + MINSIZE));
2808 if (m == 0) return 0; /* propagate failure */
2812 if ((((unsigned long)(m)) % alignment) == 0) /* aligned */
2815 if(chunk_is_mmapped(p))
2816 return chunk2mem(p); /* nothing more to do */
2819 else /* misaligned */
2822 Find an aligned spot inside chunk.
2823 Since we need to give back leading space in a chunk of at
2824 least MINSIZE, if the first calculation places us at
2825 a spot with less than MINSIZE leader, we can move to the
2826 next aligned spot -- we've allocated enough total room so that
2827 this is always possible.
2830 brk = (char*)mem2chunk(((unsigned long)(m + alignment - 1)) & -((signed) alignment));
2831 if ((long)(brk - (char*)(p)) < MINSIZE) brk = brk + alignment;
2833 newp = (mchunkptr)brk;
2834 leadsize = brk - (char*)(p);
2835 newsize = chunksize(p) - leadsize;
2838 if(chunk_is_mmapped(p))
2840 newp->prev_size = p->prev_size + leadsize;
2841 set_head(newp, newsize|IS_MMAPPED);
2842 return chunk2mem(newp);
2846 /* give back leader, use the rest */
2848 set_head(newp, newsize | PREV_INUSE);
2849 set_inuse_bit_at_offset(newp, newsize);
2850 set_head_size(p, leadsize);
2854 assert (newsize >= nb && (((unsigned long)(chunk2mem(p))) % alignment) == 0);
2857 /* Also give back spare room at the end */
2859 remainder_size = chunksize(p) - nb;
2861 if (remainder_size >= (long)MINSIZE)
2863 remainder = chunk_at_offset(p, nb);
2864 set_head(remainder, remainder_size | PREV_INUSE);
2865 set_head_size(p, nb);
2866 fREe(chunk2mem(remainder));
2869 check_inuse_chunk(p);
2870 return chunk2mem(p);
2878 valloc just invokes memalign with alignment argument equal
2879 to the page size of the system (or as near to this as can
2880 be figured out from all the includes/defines above.)
2884 Void_t* vALLOc(size_t bytes)
2886 Void_t* vALLOc(bytes) size_t bytes;
2889 return mEMALIGn (malloc_getpagesize, bytes);
2893 pvalloc just invokes valloc for the nearest pagesize
2894 that will accommodate request
2899 Void_t* pvALLOc(size_t bytes)
2901 Void_t* pvALLOc(bytes) size_t bytes;
2904 size_t pagesize = malloc_getpagesize;
2905 return mEMALIGn (pagesize, (bytes + pagesize - 1) & ~(pagesize - 1));
2910 calloc calls malloc, then zeroes out the allocated chunk.
2915 Void_t* cALLOc(size_t n, size_t elem_size)
2917 Void_t* cALLOc(n, elem_size) size_t n; size_t elem_size;
2921 INTERNAL_SIZE_T csz;
2923 INTERNAL_SIZE_T sz = n * elem_size;
2926 /* check if expand_top called, in which case don't need to clear */
2928 mchunkptr oldtop = top;
2929 INTERNAL_SIZE_T oldtopsize = chunksize(top);
2931 Void_t* mem = mALLOc (sz);
2933 if ((long)n < 0) return 0;
2941 /* Two optional cases in which clearing not necessary */
2945 if (chunk_is_mmapped(p)) return mem;
2951 if (p == oldtop && csz > oldtopsize)
2953 /* clear only the bytes from non-freshly-sbrked memory */
2958 MALLOC_ZERO(mem, csz - SIZE_SZ);
2965 cfree just calls free. It is needed/defined on some systems
2966 that pair it with calloc, presumably for odd historical reasons.
2970 #if !defined(INTERNAL_LINUX_C_LIB) || !defined(__ELF__)
2972 void cfree(Void_t *mem)
2974 void cfree(mem) Void_t *mem;
2985 Malloc_trim gives memory back to the system (via negative
2986 arguments to sbrk) if there is unused memory at the `high' end of
2987 the malloc pool. You can call this after freeing large blocks of
2988 memory to potentially reduce the system-level memory requirements
2989 of a program. However, it cannot guarantee to reduce memory. Under
2990 some allocation patterns, some large free blocks of memory will be
2991 locked between two used chunks, so they cannot be given back to
2994 The `pad' argument to malloc_trim represents the amount of free
2995 trailing space to leave untrimmed. If this argument is zero,
2996 only the minimum amount of memory to maintain internal data
2997 structures will be left (one page or less). Non-zero arguments
2998 can be supplied to maintain enough trailing space to service
2999 future expected allocations without having to re-obtain memory
3002 Malloc_trim returns 1 if it actually released any memory, else 0.
3007 int malloc_trim(size_t pad)
3009 int malloc_trim(pad) size_t pad;
3012 long top_size; /* Amount of top-most memory */
3013 long extra; /* Amount to release */
3014 char* current_brk; /* address returned by pre-check sbrk call */
3015 char* new_brk; /* address returned by negative sbrk call */
3017 unsigned long pagesz = malloc_getpagesize;
3019 top_size = chunksize(top);
3020 extra = ((top_size - pad - MINSIZE + (pagesz-1)) / pagesz - 1) * pagesz;
3022 if (extra < (long)pagesz) /* Not enough memory to release */
3027 /* Test to make sure no one else called sbrk */
3028 current_brk = (char*)(MORECORE (0));
3029 if (current_brk != (char*)(top) + top_size)
3030 return 0; /* Apparently we don't own memory; must fail */
3034 new_brk = (char*)(MORECORE (-extra));
3036 if (new_brk == (char*)(MORECORE_FAILURE)) /* sbrk failed? */
3038 /* Try to figure out what we have */
3039 current_brk = (char*)(MORECORE (0));
3040 top_size = current_brk - (char*)top;
3041 if (top_size >= (long)MINSIZE) /* if not, we are very very dead! */
3043 sbrked_mem = current_brk - sbrk_base;
3044 set_head(top, top_size | PREV_INUSE);
3052 /* Success. Adjust top accordingly. */
3053 set_head(top, (top_size - extra) | PREV_INUSE);
3054 sbrked_mem -= extra;
3067 This routine tells you how many bytes you can actually use in an
3068 allocated chunk, which may be more than you requested (although
3069 often not). You can use this many bytes without worrying about
3070 overwriting other allocated objects. Not a particularly great
3071 programming practice, but still sometimes useful.
3076 size_t malloc_usable_size(Void_t* mem)
3078 size_t malloc_usable_size(mem) Void_t* mem;
3087 if(!chunk_is_mmapped(p))
3089 if (!inuse(p)) return 0;
3090 check_inuse_chunk(p);
3091 return chunksize(p) - SIZE_SZ;
3093 return chunksize(p) - 2*SIZE_SZ;
3100 /* Utility to update current_mallinfo for malloc_stats and mallinfo() */
3103 static void malloc_update_mallinfo()
3112 INTERNAL_SIZE_T avail = chunksize(top);
3113 int navail = ((long)(avail) >= (long)MINSIZE)? 1 : 0;
3115 for (i = 1; i < NAV; ++i)
3118 for (p = last(b); p != b; p = p->bk)
3121 check_free_chunk(p);
3122 for (q = next_chunk(p);
3123 q < top && inuse(q) && (long)(chunksize(q)) >= (long)MINSIZE;
3125 check_inuse_chunk(q);
3127 avail += chunksize(p);
3132 current_mallinfo.ordblks = navail;
3133 current_mallinfo.uordblks = sbrked_mem - avail;
3134 current_mallinfo.fordblks = avail;
3135 current_mallinfo.hblks = n_mmaps;
3136 current_mallinfo.hblkhd = mmapped_mem;
3137 current_mallinfo.keepcost = chunksize(top);
3148 Prints on the amount of space obtain from the system (both
3149 via sbrk and mmap), the maximum amount (which may be more than
3150 current if malloc_trim and/or munmap got called), the maximum
3151 number of simultaneous mmap regions used, and the current number
3152 of bytes allocated via malloc (or realloc, etc) but not yet
3153 freed. (Note that this is the number of bytes allocated, not the
3154 number requested. It will be larger than the number requested
3155 because of alignment and bookkeeping overhead.)
3162 malloc_update_mallinfo();
3163 printf("max system bytes = %10u\n",
3164 (unsigned int)(max_total_mem));
3165 printf("system bytes = %10u\n",
3166 (unsigned int)(sbrked_mem + mmapped_mem));
3167 printf("in use bytes = %10u\n",
3168 (unsigned int)(current_mallinfo.uordblks + mmapped_mem));
3170 printf("max mmap regions = %10u\n",
3171 (unsigned int)max_n_mmaps);
3177 mallinfo returns a copy of updated current mallinfo.
3181 struct mallinfo mALLINFo()
3183 malloc_update_mallinfo();
3184 return current_mallinfo;
3194 mallopt is the general SVID/XPG interface to tunable parameters.
3195 The format is to provide a (parameter-number, parameter-value) pair.
3196 mallopt then sets the corresponding parameter to the argument
3197 value if it can (i.e., so long as the value is meaningful),
3198 and returns 1 if successful else 0.
3200 See descriptions of tunable parameters above.
3205 int mALLOPt(int param_number, int value)
3207 int mALLOPt(param_number, value) int param_number; int value;
3210 switch(param_number)
3212 case M_TRIM_THRESHOLD:
3213 trim_threshold = value; return 1;
3215 top_pad = value; return 1;
3216 case M_MMAP_THRESHOLD:
3217 mmap_threshold = value; return 1;
3220 n_mmaps_max = value; return 1;
3222 if (value != 0) return 0; else n_mmaps_max = value; return 1;
3234 V2.6.6 Sun Dec 5 07:42:19 1999 Doug Lea (dl at gee)
3235 * return null for negative arguments
3236 * Added Several WIN32 cleanups from Martin C. Fong <mcfong@yahoo.com>
3237 * Add 'LACKS_SYS_PARAM_H' for those systems without 'sys/param.h'
3238 (e.g. WIN32 platforms)
3239 * Cleanup up header file inclusion for WIN32 platforms
3240 * Cleanup code to avoid Microsoft Visual C++ compiler complaints
3241 * Add 'USE_DL_PREFIX' to quickly allow co-existence with existing
3242 memory allocation routines
3243 * Set 'malloc_getpagesize' for WIN32 platforms (needs more work)
3244 * Use 'assert' rather than 'ASSERT' in WIN32 code to conform to
3245 usage of 'assert' in non-WIN32 code
3246 * Improve WIN32 'sbrk()' emulation's 'findRegion()' routine to
3248 * Always call 'fREe()' rather than 'free()'
3250 V2.6.5 Wed Jun 17 15:57:31 1998 Doug Lea (dl at gee)
3251 * Fixed ordering problem with boundary-stamping
3253 V2.6.3 Sun May 19 08:17:58 1996 Doug Lea (dl at gee)
3254 * Added pvalloc, as recommended by H.J. Liu
3255 * Added 64bit pointer support mainly from Wolfram Gloger
3256 * Added anonymously donated WIN32 sbrk emulation
3257 * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen
3258 * malloc_extend_top: fix mask error that caused wastage after
3260 * Add linux mremap support code from HJ Liu
3262 V2.6.2 Tue Dec 5 06:52:55 1995 Doug Lea (dl at gee)
3263 * Integrated most documentation with the code.
3264 * Add support for mmap, with help from
3265 Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
3266 * Use last_remainder in more cases.
3267 * Pack bins using idea from colin@nyx10.cs.du.edu
3268 * Use ordered bins instead of best-fit threshhold
3269 * Eliminate block-local decls to simplify tracing and debugging.
3270 * Support another case of realloc via move into top
3271 * Fix error occuring when initial sbrk_base not word-aligned.
3272 * Rely on page size for units instead of SBRK_UNIT to
3273 avoid surprises about sbrk alignment conventions.
3274 * Add mallinfo, mallopt. Thanks to Raymond Nijssen
3275 (raymond@es.ele.tue.nl) for the suggestion.
3276 * Add `pad' argument to malloc_trim and top_pad mallopt parameter.
3277 * More precautions for cases where other routines call sbrk,
3278 courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
3279 * Added macros etc., allowing use in linux libc from
3280 H.J. Lu (hjl@gnu.ai.mit.edu)
3281 * Inverted this history list
3283 V2.6.1 Sat Dec 2 14:10:57 1995 Doug Lea (dl at gee)
3284 * Re-tuned and fixed to behave more nicely with V2.6.0 changes.
3285 * Removed all preallocation code since under current scheme
3286 the work required to undo bad preallocations exceeds
3287 the work saved in good cases for most test programs.
3288 * No longer use return list or unconsolidated bins since
3289 no scheme using them consistently outperforms those that don't
3290 given above changes.
3291 * Use best fit for very large chunks to prevent some worst-cases.
3292 * Added some support for debugging
3294 V2.6.0 Sat Nov 4 07:05:23 1995 Doug Lea (dl at gee)
3295 * Removed footers when chunks are in use. Thanks to
3296 Paul Wilson (wilson@cs.texas.edu) for the suggestion.
3298 V2.5.4 Wed Nov 1 07:54:51 1995 Doug Lea (dl at gee)
3299 * Added malloc_trim, with help from Wolfram Gloger
3300 (wmglo@Dent.MED.Uni-Muenchen.DE).
3302 V2.5.3 Tue Apr 26 10:16:01 1994 Doug Lea (dl at g)
3304 V2.5.2 Tue Apr 5 16:20:40 1994 Doug Lea (dl at g)
3305 * realloc: try to expand in both directions
3306 * malloc: swap order of clean-bin strategy;
3307 * realloc: only conditionally expand backwards
3308 * Try not to scavenge used bins
3309 * Use bin counts as a guide to preallocation
3310 * Occasionally bin return list chunks in first scan
3311 * Add a few optimizations from colin@nyx10.cs.du.edu
3313 V2.5.1 Sat Aug 14 15:40:43 1993 Doug Lea (dl at g)
3314 * faster bin computation & slightly different binning
3315 * merged all consolidations to one part of malloc proper
3316 (eliminating old malloc_find_space & malloc_clean_bin)
3317 * Scan 2 returns chunks (not just 1)
3318 * Propagate failure in realloc if malloc returns 0
3319 * Add stuff to allow compilation on non-ANSI compilers
3320 from kpv@research.att.com
3322 V2.5 Sat Aug 7 07:41:59 1993 Doug Lea (dl at g.oswego.edu)
3323 * removed potential for odd address access in prev_chunk
3324 * removed dependency on getpagesize.h
3325 * misc cosmetics and a bit more internal documentation
3326 * anticosmetics: mangled names in macros to evade debugger strangeness
3327 * tested on sparc, hp-700, dec-mips, rs6000
3328 with gcc & native cc (hp, dec only) allowing
3329 Detlefs & Zorn comparison study (in SIGPLAN Notices.)
3331 Trial version Fri Aug 28 13:14:29 1992 Doug Lea (dl at g.oswego.edu)
3332 * Based loosely on libg++-1.2X malloc. (It retains some of the overall
3333 structure of old version, but most details differ.)