1 /* ---------- To make a malloc.h, start cutting here ------------ */
4 A version of malloc/free/realloc written by Doug Lea and released to the
5 public domain. Send questions/comments/complaints/performance data
8 * VERSION 2.6.6 Sun Mar 5 19:10:03 2000 Doug Lea (dl at gee)
10 Note: There may be an updated version of this malloc obtainable at
11 ftp://g.oswego.edu/pub/misc/malloc.c
12 Check before installing!
14 * Why use this malloc?
16 This is not the fastest, most space-conserving, most portable, or
17 most tunable malloc ever written. However it is among the fastest
18 while also being among the most space-conserving, portable and tunable.
19 Consistent balance across these factors results in a good general-purpose
20 allocator. For a high-level description, see
21 http://g.oswego.edu/dl/html/malloc.html
23 * Synopsis of public routines
25 (Much fuller descriptions are contained in the program documentation below.)
28 Return a pointer to a newly allocated chunk of at least n bytes, or null
29 if no space is available.
31 Release the chunk of memory pointed to by p, or no effect if p is null.
32 realloc(Void_t* p, size_t n);
33 Return a pointer to a chunk of size n that contains the same data
34 as does chunk p up to the minimum of (n, p's size) bytes, or null
35 if no space is available. The returned pointer may or may not be
36 the same as p. If p is null, equivalent to malloc. Unless the
37 #define REALLOC_ZERO_BYTES_FREES below is set, realloc with a
38 size argument of zero (re)allocates a minimum-sized chunk.
39 memalign(size_t alignment, size_t n);
40 Return a pointer to a newly allocated chunk of n bytes, aligned
41 in accord with the alignment argument, which must be a power of
44 Equivalent to memalign(pagesize, n), where pagesize is the page
45 size of the system (or as near to this as can be figured out from
46 all the includes/defines below.)
48 Equivalent to valloc(minimum-page-that-holds(n)), that is,
49 round up n to nearest pagesize.
50 calloc(size_t unit, size_t quantity);
51 Returns a pointer to quantity * unit bytes, with all locations
54 Equivalent to free(p).
55 malloc_trim(size_t pad);
56 Release all but pad bytes of freed top-most memory back
57 to the system. Return 1 if successful, else 0.
58 malloc_usable_size(Void_t* p);
59 Report the number usable allocated bytes associated with allocated
60 chunk p. This may or may not report more bytes than were requested,
61 due to alignment and minimum size constraints.
63 Prints brief summary statistics on stderr.
65 Returns (by copy) a struct containing various summary statistics.
66 mallopt(int parameter_number, int parameter_value)
67 Changes one of the tunable parameters described below. Returns
68 1 if successful in changing the parameter, else 0.
73 8 byte alignment is currently hardwired into the design. This
74 seems to suffice for all current machines and C compilers.
76 Assumed pointer representation: 4 or 8 bytes
77 Code for 8-byte pointers is untested by me but has worked
78 reliably by Wolfram Gloger, who contributed most of the
79 changes supporting this.
81 Assumed size_t representation: 4 or 8 bytes
82 Note that size_t is allowed to be 4 bytes even if pointers are 8.
84 Minimum overhead per allocated chunk: 4 or 8 bytes
85 Each malloced chunk has a hidden overhead of 4 bytes holding size
86 and status information.
88 Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead)
89 8-byte ptrs: 24/32 bytes (including, 4/8 overhead)
91 When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte
92 ptrs but 4 byte size) or 24 (for 8/8) additional bytes are
93 needed; 4 (8) for a trailing size field
94 and 8 (16) bytes for free list pointers. Thus, the minimum
95 allocatable size is 16/24/32 bytes.
97 Even a request for zero bytes (i.e., malloc(0)) returns a
98 pointer to something of the minimum allocatable size.
100 Maximum allocated size: 4-byte size_t: 2^31 - 8 bytes
101 8-byte size_t: 2^63 - 16 bytes
103 It is assumed that (possibly signed) size_t bit values suffice to
104 represent chunk sizes. `Possibly signed' is due to the fact
105 that `size_t' may be defined on a system as either a signed or
106 an unsigned type. To be conservative, values that would appear
107 as negative numbers are avoided.
108 Requests for sizes with a negative sign bit when the request
109 size is treaded as a long will return null.
111 Maximum overhead wastage per allocated chunk: normally 15 bytes
113 Alignnment demands, plus the minimum allocatable size restriction
114 make the normal worst-case wastage 15 bytes (i.e., up to 15
115 more bytes will be allocated than were requested in malloc), with
117 1. Because requests for zero bytes allocate non-zero space,
118 the worst case wastage for a request of zero bytes is 24 bytes.
119 2. For requests >= mmap_threshold that are serviced via
120 mmap(), the worst case wastage is 8 bytes plus the remainder
121 from a system page (the minimal mmap unit); typically 4096 bytes.
125 Here are some features that are NOT currently supported
127 * No user-definable hooks for callbacks and the like.
128 * No automated mechanism for fully checking that all accesses
129 to malloced memory stay within their bounds.
130 * No support for compaction.
132 * Synopsis of compile-time options:
134 People have reported using previous versions of this malloc on all
135 versions of Unix, sometimes by tweaking some of the defines
136 below. It has been tested most extensively on Solaris and
137 Linux. It is also reported to work on WIN32 platforms.
138 People have also reported adapting this malloc for use in
139 stand-alone embedded systems.
141 The implementation is in straight, hand-tuned ANSI C. Among other
142 consequences, it uses a lot of macros. Because of this, to be at
143 all usable, this code should be compiled using an optimizing compiler
144 (for example gcc -O2) that can simplify expressions and control
147 __STD_C (default: derived from C compiler defines)
148 Nonzero if using ANSI-standard C compiler, a C++ compiler, or
149 a C compiler sufficiently close to ANSI to get away with it.
150 DEBUG (default: NOT defined)
151 Define to enable debugging. Adds fairly extensive assertion-based
152 checking to help track down memory errors, but noticeably slows down
154 REALLOC_ZERO_BYTES_FREES (default: NOT defined)
155 Define this if you think that realloc(p, 0) should be equivalent
156 to free(p). Otherwise, since malloc returns a unique pointer for
157 malloc(0), so does realloc(p, 0).
158 HAVE_MEMCPY (default: defined)
159 Define if you are not otherwise using ANSI STD C, but still
160 have memcpy and memset in your C library and want to use them.
161 Otherwise, simple internal versions are supplied.
162 USE_MEMCPY (default: 1 if HAVE_MEMCPY is defined, 0 otherwise)
163 Define as 1 if you want the C library versions of memset and
164 memcpy called in realloc and calloc (otherwise macro versions are used).
165 At least on some platforms, the simple macro versions usually
166 outperform libc versions.
167 HAVE_MMAP (default: defined as 1)
168 Define to non-zero to optionally make malloc() use mmap() to
169 allocate very large blocks.
170 HAVE_MREMAP (default: defined as 0 unless Linux libc set)
171 Define to non-zero to optionally make realloc() use mremap() to
172 reallocate very large blocks.
173 malloc_getpagesize (default: derived from system #includes)
174 Either a constant or routine call returning the system page size.
175 HAVE_USR_INCLUDE_MALLOC_H (default: NOT defined)
176 Optionally define if you are on a system with a /usr/include/malloc.h
177 that declares struct mallinfo. It is not at all necessary to
178 define this even if you do, but will ensure consistency.
179 INTERNAL_SIZE_T (default: size_t)
180 Define to a 32-bit type (probably `unsigned int') if you are on a
181 64-bit machine, yet do not want or need to allow malloc requests of
182 greater than 2^31 to be handled. This saves space, especially for
184 INTERNAL_LINUX_C_LIB (default: NOT defined)
185 Defined only when compiled as part of Linux libc.
186 Also note that there is some odd internal name-mangling via defines
187 (for example, internally, `malloc' is named `mALLOc') needed
188 when compiling in this case. These look funny but don't otherwise
190 WIN32 (default: undefined)
191 Define this on MS win (95, nt) platforms to compile in sbrk emulation.
192 LACKS_UNISTD_H (default: undefined if not WIN32)
193 Define this if your system does not have a <unistd.h>.
194 LACKS_SYS_PARAM_H (default: undefined if not WIN32)
195 Define this if your system does not have a <sys/param.h>.
196 MORECORE (default: sbrk)
197 The name of the routine to call to obtain more memory from the system.
198 MORECORE_FAILURE (default: -1)
199 The value returned upon failure of MORECORE.
200 MORECORE_CLEARS (default 1)
201 true (1) if the routine mapped to MORECORE zeroes out memory (which
203 DEFAULT_TRIM_THRESHOLD
205 DEFAULT_MMAP_THRESHOLD
207 Default values of tunable parameters (described in detail below)
208 controlling interaction with host system routines (sbrk, mmap, etc).
209 These values may also be changed dynamically via mallopt(). The
210 preset defaults are those that give best performance for typical
212 USE_DL_PREFIX (default: undefined)
213 Prefix all public routines with the string 'dl'. Useful to
214 quickly avoid procedure declaration conflicts and linker symbol
215 conflicts with existing memory allocation routines.
233 #endif /*__cplusplus*/
238 #if (__STD_C || defined(WIN32))
246 #include <stddef.h> /* for size_t */
248 #include <sys/types.h>
255 #include <stdio.h> /* needed for malloc_stats */
266 Because freed chunks may be overwritten with link fields, this
267 malloc will often die when freed memory is overwritten by user
268 programs. This can be very effective (albeit in an annoying way)
269 in helping track down dangling pointers.
271 If you compile with -DDEBUG, a number of assertion checks are
272 enabled that will catch more memory errors. You probably won't be
273 able to make much sense of the actual assertion errors, but they
274 should help you locate incorrectly overwritten memory. The
275 checking is fairly extensive, and will slow down execution
276 noticeably. Calling malloc_stats or mallinfo with DEBUG set will
277 attempt to check every non-mmapped allocated and free chunk in the
278 course of computing the summmaries. (By nature, mmapped regions
279 cannot be checked very much automatically.)
281 Setting DEBUG may also be helpful if you are trying to modify
282 this code. The assertions in the check routines spell out in more
283 detail the assumptions and invariants underlying the algorithms.
290 #define assert(x) ((void)0)
295 INTERNAL_SIZE_T is the word-size used for internal bookkeeping
296 of chunk sizes. On a 64-bit machine, you can reduce malloc
297 overhead by defining INTERNAL_SIZE_T to be a 32 bit `unsigned int'
298 at the expense of not being able to handle requests greater than
299 2^31. This limitation is hardly ever a concern; you are encouraged
300 to set this. However, the default version is the same as size_t.
303 #ifndef INTERNAL_SIZE_T
304 #define INTERNAL_SIZE_T size_t
308 REALLOC_ZERO_BYTES_FREES should be set if a call to
309 realloc with zero bytes should be the same as a call to free.
310 Some people think it should. Otherwise, since this malloc
311 returns a unique pointer for malloc(0), so does realloc(p, 0).
315 /* #define REALLOC_ZERO_BYTES_FREES */
319 WIN32 causes an emulation of sbrk to be compiled in
320 mmap-based options are not currently supported in WIN32.
325 #define MORECORE wsbrk
328 #define LACKS_UNISTD_H
329 #define LACKS_SYS_PARAM_H
332 Include 'windows.h' to get the necessary declarations for the
333 Microsoft Visual C++ data structures and routines used in the 'sbrk'
336 Define WIN32_LEAN_AND_MEAN so that only the essential Microsoft
337 Visual C++ header files are included.
339 #define WIN32_LEAN_AND_MEAN
345 HAVE_MEMCPY should be defined if you are not otherwise using
346 ANSI STD C, but still have memcpy and memset in your C library
347 and want to use them in calloc and realloc. Otherwise simple
348 macro versions are defined here.
350 USE_MEMCPY should be defined as 1 if you actually want to
351 have memset and memcpy called. People report that the macro
352 versions are often enough faster than libc versions on many
353 systems that it is better to use them.
367 #if (__STD_C || defined(HAVE_MEMCPY))
370 void* memset(void*, int, size_t);
371 void* memcpy(void*, const void*, size_t);
374 /* On Win32 platforms, 'memset()' and 'memcpy()' are already declared in */
385 /* The following macros are only invoked with (2n+1)-multiples of
386 INTERNAL_SIZE_T units, with a positive integer n. This is exploited
387 for fast inline execution when n is small. */
389 #define MALLOC_ZERO(charp, nbytes) \
391 INTERNAL_SIZE_T mzsz = (nbytes); \
392 if(mzsz <= 9*sizeof(mzsz)) { \
393 INTERNAL_SIZE_T* mz = (INTERNAL_SIZE_T*) (charp); \
394 if(mzsz >= 5*sizeof(mzsz)) { *mz++ = 0; \
396 if(mzsz >= 7*sizeof(mzsz)) { *mz++ = 0; \
398 if(mzsz >= 9*sizeof(mzsz)) { *mz++ = 0; \
403 } else memset((charp), 0, mzsz); \
406 #define MALLOC_COPY(dest,src,nbytes) \
408 INTERNAL_SIZE_T mcsz = (nbytes); \
409 if(mcsz <= 9*sizeof(mcsz)) { \
410 INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) (src); \
411 INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) (dest); \
412 if(mcsz >= 5*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \
413 *mcdst++ = *mcsrc++; \
414 if(mcsz >= 7*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \
415 *mcdst++ = *mcsrc++; \
416 if(mcsz >= 9*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \
417 *mcdst++ = *mcsrc++; }}} \
418 *mcdst++ = *mcsrc++; \
419 *mcdst++ = *mcsrc++; \
421 } else memcpy(dest, src, mcsz); \
424 #else /* !USE_MEMCPY */
426 /* Use Duff's device for good zeroing/copying performance. */
428 #define MALLOC_ZERO(charp, nbytes) \
430 INTERNAL_SIZE_T* mzp = (INTERNAL_SIZE_T*)(charp); \
431 long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn; \
432 if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \
434 case 0: for(;;) { *mzp++ = 0; \
435 case 7: *mzp++ = 0; \
436 case 6: *mzp++ = 0; \
437 case 5: *mzp++ = 0; \
438 case 4: *mzp++ = 0; \
439 case 3: *mzp++ = 0; \
440 case 2: *mzp++ = 0; \
441 case 1: *mzp++ = 0; if(mcn <= 0) break; mcn--; } \
445 #define MALLOC_COPY(dest,src,nbytes) \
447 INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) src; \
448 INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) dest; \
449 long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn; \
450 if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \
452 case 0: for(;;) { *mcdst++ = *mcsrc++; \
453 case 7: *mcdst++ = *mcsrc++; \
454 case 6: *mcdst++ = *mcsrc++; \
455 case 5: *mcdst++ = *mcsrc++; \
456 case 4: *mcdst++ = *mcsrc++; \
457 case 3: *mcdst++ = *mcsrc++; \
458 case 2: *mcdst++ = *mcsrc++; \
459 case 1: *mcdst++ = *mcsrc++; if(mcn <= 0) break; mcn--; } \
467 Define HAVE_MMAP to optionally make malloc() use mmap() to
468 allocate very large blocks. These will be returned to the
469 operating system immediately after a free().
477 Define HAVE_MREMAP to make realloc() use mremap() to re-allocate
478 large blocks. This is currently only possible on Linux with
479 kernel versions newer than 1.3.77.
483 #ifdef INTERNAL_LINUX_C_LIB
484 #define HAVE_MREMAP 1
486 #define HAVE_MREMAP 0
494 #include <sys/mman.h>
496 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
497 #define MAP_ANONYMOUS MAP_ANON
500 #endif /* HAVE_MMAP */
503 Access to system page size. To the extent possible, this malloc
504 manages memory from the system in page-size units.
506 The following mechanics for getpagesize were adapted from
507 bsd/gnu getpagesize.h
510 #ifndef LACKS_UNISTD_H
514 #ifndef malloc_getpagesize
515 # ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */
516 # ifndef _SC_PAGE_SIZE
517 # define _SC_PAGE_SIZE _SC_PAGESIZE
520 # ifdef _SC_PAGE_SIZE
521 # define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
523 # if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
524 extern size_t getpagesize();
525 # define malloc_getpagesize getpagesize()
528 # define malloc_getpagesize (4096) /* TBD: Use 'GetSystemInfo' instead */
530 # ifndef LACKS_SYS_PARAM_H
531 # include <sys/param.h>
533 # ifdef EXEC_PAGESIZE
534 # define malloc_getpagesize EXEC_PAGESIZE
538 # define malloc_getpagesize NBPG
540 # define malloc_getpagesize (NBPG * CLSIZE)
544 # define malloc_getpagesize NBPC
547 # define malloc_getpagesize PAGESIZE
549 # define malloc_getpagesize (4096) /* just guess */
562 This version of malloc supports the standard SVID/XPG mallinfo
563 routine that returns a struct containing the same kind of
564 information you can get from malloc_stats. It should work on
565 any SVID/XPG compliant system that has a /usr/include/malloc.h
566 defining struct mallinfo. (If you'd like to install such a thing
567 yourself, cut out the preliminary declarations as described above
568 and below and save them in a malloc.h file. But there's no
569 compelling reason to bother to do this.)
571 The main declaration needed is the mallinfo struct that is returned
572 (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a
573 bunch of fields, most of which are not even meaningful in this
574 version of malloc. Some of these fields are are instead filled by
575 mallinfo() with other numbers that might possibly be of interest.
577 HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
578 /usr/include/malloc.h file that includes a declaration of struct
579 mallinfo. If so, it is included; else an SVID2/XPG2 compliant
580 version is declared below. These must be precisely the same for
585 /* #define HAVE_USR_INCLUDE_MALLOC_H */
587 #if HAVE_USR_INCLUDE_MALLOC_H
588 #include "/usr/include/malloc.h"
591 /* SVID2/XPG mallinfo structure */
594 int arena; /* total space allocated from system */
595 int ordblks; /* number of non-inuse chunks */
596 int smblks; /* unused -- always zero */
597 int hblks; /* number of mmapped regions */
598 int hblkhd; /* total space in mmapped regions */
599 int usmblks; /* unused -- always zero */
600 int fsmblks; /* unused -- always zero */
601 int uordblks; /* total allocated space */
602 int fordblks; /* total non-inuse space */
603 int keepcost; /* top-most, releasable (via malloc_trim) space */
606 /* SVID2/XPG mallopt options */
608 #define M_MXFAST 1 /* UNUSED in this malloc */
609 #define M_NLBLKS 2 /* UNUSED in this malloc */
610 #define M_GRAIN 3 /* UNUSED in this malloc */
611 #define M_KEEP 4 /* UNUSED in this malloc */
615 /* mallopt options that actually do something */
617 #define M_TRIM_THRESHOLD -1
619 #define M_MMAP_THRESHOLD -3
620 #define M_MMAP_MAX -4
623 #ifndef DEFAULT_TRIM_THRESHOLD
624 #define DEFAULT_TRIM_THRESHOLD (128 * 1024)
628 M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
629 to keep before releasing via malloc_trim in free().
631 Automatic trimming is mainly useful in long-lived programs.
632 Because trimming via sbrk can be slow on some systems, and can
633 sometimes be wasteful (in cases where programs immediately
634 afterward allocate more large chunks) the value should be high
635 enough so that your overall system performance would improve by
638 The trim threshold and the mmap control parameters (see below)
639 can be traded off with one another. Trimming and mmapping are
640 two different ways of releasing unused memory back to the
641 system. Between these two, it is often possible to keep
642 system-level demands of a long-lived program down to a bare
643 minimum. For example, in one test suite of sessions measuring
644 the XF86 X server on Linux, using a trim threshold of 128K and a
645 mmap threshold of 192K led to near-minimal long term resource
648 If you are using this malloc in a long-lived program, it should
649 pay to experiment with these values. As a rough guide, you
650 might set to a value close to the average size of a process
651 (program) running on your system. Releasing this much memory
652 would allow such a process to run in memory. Generally, it's
653 worth it to tune for trimming rather tham memory mapping when a
654 program undergoes phases where several large chunks are
655 allocated and released in ways that can reuse each other's
656 storage, perhaps mixed with phases where there are no such
657 chunks at all. And in well-behaved long-lived programs,
658 controlling release of large blocks via trimming versus mapping
661 However, in most programs, these parameters serve mainly as
662 protection against the system-level effects of carrying around
663 massive amounts of unneeded memory. Since frequent calls to
664 sbrk, mmap, and munmap otherwise degrade performance, the default
665 parameters are set to relatively high values that serve only as
668 The default trim value is high enough to cause trimming only in
669 fairly extreme (by current memory consumption standards) cases.
670 It must be greater than page size to have any useful effect. To
671 disable trimming completely, you can set to (unsigned long)(-1);
677 #ifndef DEFAULT_TOP_PAD
678 #define DEFAULT_TOP_PAD (0)
682 M_TOP_PAD is the amount of extra `padding' space to allocate or
683 retain whenever sbrk is called. It is used in two ways internally:
685 * When sbrk is called to extend the top of the arena to satisfy
686 a new malloc request, this much padding is added to the sbrk
689 * When malloc_trim is called automatically from free(),
690 it is used as the `pad' argument.
692 In both cases, the actual amount of padding is rounded
693 so that the end of the arena is always a system page boundary.
695 The main reason for using padding is to avoid calling sbrk so
696 often. Having even a small pad greatly reduces the likelihood
697 that nearly every malloc request during program start-up (or
698 after trimming) will invoke sbrk, which needlessly wastes
701 Automatic rounding-up to page-size units is normally sufficient
702 to avoid measurable overhead, so the default is 0. However, in
703 systems where sbrk is relatively slow, it can pay to increase
704 this value, at the expense of carrying around more memory than
710 #ifndef DEFAULT_MMAP_THRESHOLD
711 #define DEFAULT_MMAP_THRESHOLD (128 * 1024)
716 M_MMAP_THRESHOLD is the request size threshold for using mmap()
717 to service a request. Requests of at least this size that cannot
718 be allocated using already-existing space will be serviced via mmap.
719 (If enough normal freed space already exists it is used instead.)
721 Using mmap segregates relatively large chunks of memory so that
722 they can be individually obtained and released from the host
723 system. A request serviced through mmap is never reused by any
724 other request (at least not directly; the system may just so
725 happen to remap successive requests to the same locations).
727 Segregating space in this way has the benefit that mmapped space
728 can ALWAYS be individually released back to the system, which
729 helps keep the system level memory demands of a long-lived
730 program low. Mapped memory can never become `locked' between
731 other chunks, as can happen with normally allocated chunks, which
732 menas that even trimming via malloc_trim would not release them.
734 However, it has the disadvantages that:
736 1. The space cannot be reclaimed, consolidated, and then
737 used to service later requests, as happens with normal chunks.
738 2. It can lead to more wastage because of mmap page alignment
740 3. It causes malloc performance to be more dependent on host
741 system memory management support routines which may vary in
742 implementation quality and may impose arbitrary
743 limitations. Generally, servicing a request via normal
744 malloc steps is faster than going through a system's mmap.
746 All together, these considerations should lead you to use mmap
747 only for relatively large requests.
753 #ifndef DEFAULT_MMAP_MAX
755 #define DEFAULT_MMAP_MAX (64)
757 #define DEFAULT_MMAP_MAX (0)
762 M_MMAP_MAX is the maximum number of requests to simultaneously
763 service using mmap. This parameter exists because:
765 1. Some systems have a limited number of internal tables for
767 2. In most systems, overreliance on mmap can degrade overall
769 3. If a program allocates many large regions, it is probably
770 better off using normal sbrk-based allocation routines that
771 can reclaim and reallocate normal heap memory. Using a
772 small value allows transition into this mode after the
773 first few allocations.
775 Setting to 0 disables all use of mmap. If HAVE_MMAP is not set,
776 the default value is 0, and attempts to set it to non-zero values
777 in mallopt will fail.
782 USE_DL_PREFIX will prefix all public routines with the string 'dl'.
783 Useful to quickly avoid procedure declaration conflicts and linker
784 symbol conflicts with existing memory allocation routines.
788 /* #define USE_DL_PREFIX */
793 Special defines for linux libc
795 Except when compiled using these special defines for Linux libc
796 using weak aliases, this malloc is NOT designed to work in
797 multithreaded applications. No semaphores or other concurrency
798 control are provided to ensure that multiple malloc or free calls
799 don't run at the same time, which could be disasterous. A single
800 semaphore could be used across malloc, realloc, and free (which is
801 essentially the effect of the linux weak alias approach). It would
802 be hard to obtain finer granularity.
807 #ifdef INTERNAL_LINUX_C_LIB
811 Void_t * __default_morecore_init (ptrdiff_t);
812 Void_t *(*__morecore)(ptrdiff_t) = __default_morecore_init;
816 Void_t * __default_morecore_init ();
817 Void_t *(*__morecore)() = __default_morecore_init;
821 #define MORECORE (*__morecore)
822 #define MORECORE_FAILURE 0
823 #define MORECORE_CLEARS 1
825 #else /* INTERNAL_LINUX_C_LIB */
828 extern Void_t* sbrk(ptrdiff_t);
830 extern Void_t* sbrk();
834 #define MORECORE sbrk
837 #ifndef MORECORE_FAILURE
838 #define MORECORE_FAILURE -1
841 #ifndef MORECORE_CLEARS
842 #define MORECORE_CLEARS 1
845 #endif /* INTERNAL_LINUX_C_LIB */
847 #if defined(INTERNAL_LINUX_C_LIB) && defined(__ELF__)
849 #define cALLOc __libc_calloc
850 #define fREe __libc_free
851 #define mALLOc __libc_malloc
852 #define mEMALIGn __libc_memalign
853 #define rEALLOc __libc_realloc
854 #define vALLOc __libc_valloc
855 #define pvALLOc __libc_pvalloc
856 #define mALLINFo __libc_mallinfo
857 #define mALLOPt __libc_mallopt
859 #pragma weak calloc = __libc_calloc
860 #pragma weak free = __libc_free
861 #pragma weak cfree = __libc_free
862 #pragma weak malloc = __libc_malloc
863 #pragma weak memalign = __libc_memalign
864 #pragma weak realloc = __libc_realloc
865 #pragma weak valloc = __libc_valloc
866 #pragma weak pvalloc = __libc_pvalloc
867 #pragma weak mallinfo = __libc_mallinfo
868 #pragma weak mallopt = __libc_mallopt
873 #define cALLOc dlcalloc
875 #define mALLOc dlmalloc
876 #define mEMALIGn dlmemalign
877 #define rEALLOc dlrealloc
878 #define vALLOc dlvalloc
879 #define pvALLOc dlpvalloc
880 #define mALLINFo dlmallinfo
881 #define mALLOPt dlmallopt
882 #else /* USE_DL_PREFIX */
883 #define cALLOc calloc
885 #define mALLOc malloc
886 #define mEMALIGn memalign
887 #define rEALLOc realloc
888 #define vALLOc valloc
889 #define pvALLOc pvalloc
890 #define mALLINFo mallinfo
891 #define mALLOPt mallopt
892 #endif /* USE_DL_PREFIX */
896 /* Public routines */
900 Void_t* mALLOc(size_t);
902 Void_t* rEALLOc(Void_t*, size_t);
903 Void_t* mEMALIGn(size_t, size_t);
904 Void_t* vALLOc(size_t);
905 Void_t* pvALLOc(size_t);
906 Void_t* cALLOc(size_t, size_t);
908 int malloc_trim(size_t);
909 size_t malloc_usable_size(Void_t*);
911 int mALLOPt(int, int);
912 struct mallinfo mALLINFo(void);
923 size_t malloc_usable_size();
926 struct mallinfo mALLINFo();
931 }; /* end of extern "C" */
934 /* ---------- To make a malloc.h, end cutting here ------------ */
938 Emulation of sbrk for WIN32
939 All code within the ifdef WIN32 is untested by me.
941 Thanks to Martin Fong and others for supplying this.
947 #define AlignPage(add) (((add) + (malloc_getpagesize-1)) & \
948 ~(malloc_getpagesize-1))
949 #define AlignPage64K(add) (((add) + (0x10000 - 1)) & ~(0x10000 - 1))
951 /* resrve 64MB to insure large contiguous space */
952 #define RESERVED_SIZE (1024*1024*64)
953 #define NEXT_SIZE (2048*1024)
954 #define TOP_MEMORY ((unsigned long)2*1024*1024*1024)
956 struct GmListElement;
957 typedef struct GmListElement GmListElement;
965 static GmListElement* head = 0;
966 static unsigned int gNextAddress = 0;
967 static unsigned int gAddressBase = 0;
968 static unsigned int gAllocatedSize = 0;
971 GmListElement* makeGmListElement (void* bas)
974 this = (GmListElement*)(void*)LocalAlloc (0, sizeof (GmListElement));
988 assert ( (head == NULL) || (head->base == (void*)gAddressBase));
989 if (gAddressBase && (gNextAddress - gAddressBase))
991 rval = VirtualFree ((void*)gAddressBase,
992 gNextAddress - gAddressBase,
998 GmListElement* next = head->next;
999 rval = VirtualFree (head->base, 0, MEM_RELEASE);
1007 void* findRegion (void* start_address, unsigned long size)
1009 MEMORY_BASIC_INFORMATION info;
1010 if (size >= TOP_MEMORY) return NULL;
1012 while ((unsigned long)start_address + size < TOP_MEMORY)
1014 VirtualQuery (start_address, &info, sizeof (info));
1015 if ((info.State == MEM_FREE) && (info.RegionSize >= size))
1016 return start_address;
1019 /* Requested region is not available so see if the */
1020 /* next region is available. Set 'start_address' */
1021 /* to the next region and call 'VirtualQuery()' */
1024 start_address = (char*)info.BaseAddress + info.RegionSize;
1026 /* Make sure we start looking for the next region */
1027 /* on the *next* 64K boundary. Otherwise, even if */
1028 /* the new region is free according to */
1029 /* 'VirtualQuery()', the subsequent call to */
1030 /* 'VirtualAlloc()' (which follows the call to */
1031 /* this routine in 'wsbrk()') will round *down* */
1032 /* the requested address to a 64K boundary which */
1033 /* we already know is an address in the */
1034 /* unavailable region. Thus, the subsequent call */
1035 /* to 'VirtualAlloc()' will fail and bring us back */
1036 /* here, causing us to go into an infinite loop. */
1039 (void *) AlignPage64K((unsigned long) start_address);
1047 void* wsbrk (long size)
1052 if (gAddressBase == 0)
1054 gAllocatedSize = max (RESERVED_SIZE, AlignPage (size));
1055 gNextAddress = gAddressBase =
1056 (unsigned int)VirtualAlloc (NULL, gAllocatedSize,
1057 MEM_RESERVE, PAGE_NOACCESS);
1058 } else if (AlignPage (gNextAddress + size) > (gAddressBase +
1061 long new_size = max (NEXT_SIZE, AlignPage (size));
1062 void* new_address = (void*)(gAddressBase+gAllocatedSize);
1065 new_address = findRegion (new_address, new_size);
1067 if (new_address == 0)
1070 gAddressBase = gNextAddress =
1071 (unsigned int)VirtualAlloc (new_address, new_size,
1072 MEM_RESERVE, PAGE_NOACCESS);
1073 /* repeat in case of race condition */
1074 /* The region that we found has been snagged */
1075 /* by another thread */
1077 while (gAddressBase == 0);
1079 assert (new_address == (void*)gAddressBase);
1081 gAllocatedSize = new_size;
1083 if (!makeGmListElement ((void*)gAddressBase))
1086 if ((size + gNextAddress) > AlignPage (gNextAddress))
1089 res = VirtualAlloc ((void*)AlignPage (gNextAddress),
1090 (size + gNextAddress -
1091 AlignPage (gNextAddress)),
1092 MEM_COMMIT, PAGE_READWRITE);
1096 tmp = (void*)gNextAddress;
1097 gNextAddress = (unsigned int)tmp + size;
1102 unsigned int alignedGoal = AlignPage (gNextAddress + size);
1103 /* Trim by releasing the virtual memory */
1104 if (alignedGoal >= gAddressBase)
1106 VirtualFree ((void*)alignedGoal, gNextAddress - alignedGoal,
1108 gNextAddress = gNextAddress + size;
1109 return (void*)gNextAddress;
1113 VirtualFree ((void*)gAddressBase, gNextAddress - gAddressBase,
1115 gNextAddress = gAddressBase;
1121 return (void*)gNextAddress;
1136 INTERNAL_SIZE_T prev_size; /* Size of previous chunk (if free). */
1137 INTERNAL_SIZE_T size; /* Size in bytes, including overhead. */
1138 struct malloc_chunk* fd; /* double links -- used only if free. */
1139 struct malloc_chunk* bk;
1142 typedef struct malloc_chunk* mchunkptr;
1146 malloc_chunk details:
1148 (The following includes lightly edited explanations by Colin Plumb.)
1150 Chunks of memory are maintained using a `boundary tag' method as
1151 described in e.g., Knuth or Standish. (See the paper by Paul
1152 Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
1153 survey of such techniques.) Sizes of free chunks are stored both
1154 in the front of each chunk and at the end. This makes
1155 consolidating fragmented chunks into bigger chunks very fast. The
1156 size fields also hold bits representing whether chunks are free or
1159 An allocated chunk looks like this:
1162 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1163 | Size of previous chunk, if allocated | |
1164 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1165 | Size of chunk, in bytes |P|
1166 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1167 | User data starts here... .
1169 . (malloc_usable_space() bytes) .
1171 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1173 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1176 Where "chunk" is the front of the chunk for the purpose of most of
1177 the malloc code, but "mem" is the pointer that is returned to the
1178 user. "Nextchunk" is the beginning of the next contiguous chunk.
1180 Chunks always begin on even word boundries, so the mem portion
1181 (which is returned to the user) is also on an even word boundary, and
1182 thus double-word aligned.
1184 Free chunks are stored in circular doubly-linked lists, and look like this:
1186 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1187 | Size of previous chunk |
1188 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1189 `head:' | Size of chunk, in bytes |P|
1190 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1191 | Forward pointer to next chunk in list |
1192 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1193 | Back pointer to previous chunk in list |
1194 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1195 | Unused space (may be 0 bytes long) .
1198 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1199 `foot:' | Size of chunk, in bytes |
1200 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1202 The P (PREV_INUSE) bit, stored in the unused low-order bit of the
1203 chunk size (which is always a multiple of two words), is an in-use
1204 bit for the *previous* chunk. If that bit is *clear*, then the
1205 word before the current chunk size contains the previous chunk
1206 size, and can be used to find the front of the previous chunk.
1207 (The very first chunk allocated always has this bit set,
1208 preventing access to non-existent (or non-owned) memory.)
1210 Note that the `foot' of the current chunk is actually represented
1211 as the prev_size of the NEXT chunk. (This makes it easier to
1212 deal with alignments etc).
1214 The two exceptions to all this are
1216 1. The special chunk `top', which doesn't bother using the
1217 trailing size field since there is no
1218 next contiguous chunk that would have to index off it. (After
1219 initialization, `top' is forced to always exist. If it would
1220 become less than MINSIZE bytes long, it is replenished via
1223 2. Chunks allocated via mmap, which have the second-lowest-order
1224 bit (IS_MMAPPED) set in their size fields. Because they are
1225 never merged or traversed from any other chunk, they have no
1226 foot size or inuse information.
1228 Available chunks are kept in any of several places (all declared below):
1230 * `av': An array of chunks serving as bin headers for consolidated
1231 chunks. Each bin is doubly linked. The bins are approximately
1232 proportionally (log) spaced. There are a lot of these bins
1233 (128). This may look excessive, but works very well in
1234 practice. All procedures maintain the invariant that no
1235 consolidated chunk physically borders another one. Chunks in
1236 bins are kept in size order, with ties going to the
1237 approximately least recently used chunk.
1239 The chunks in each bin are maintained in decreasing sorted order by
1240 size. This is irrelevant for the small bins, which all contain
1241 the same-sized chunks, but facilitates best-fit allocation for
1242 larger chunks. (These lists are just sequential. Keeping them in
1243 order almost never requires enough traversal to warrant using
1244 fancier ordered data structures.) Chunks of the same size are
1245 linked with the most recently freed at the front, and allocations
1246 are taken from the back. This results in LRU or FIFO allocation
1247 order, which tends to give each chunk an equal opportunity to be
1248 consolidated with adjacent freed chunks, resulting in larger free
1249 chunks and less fragmentation.
1251 * `top': The top-most available chunk (i.e., the one bordering the
1252 end of available memory) is treated specially. It is never
1253 included in any bin, is used only if no other chunk is
1254 available, and is released back to the system if it is very
1255 large (see M_TRIM_THRESHOLD).
1257 * `last_remainder': A bin holding only the remainder of the
1258 most recently split (non-top) chunk. This bin is checked
1259 before other non-fitting chunks, so as to provide better
1260 locality for runs of sequentially allocated chunks.
1262 * Implicitly, through the host system's memory mapping tables.
1263 If supported, requests greater than a threshold are usually
1264 serviced via calls to mmap, and then later released via munmap.
1272 /* sizes, alignments */
1274 #define SIZE_SZ (sizeof(INTERNAL_SIZE_T))
1275 #define MALLOC_ALIGNMENT (SIZE_SZ + SIZE_SZ)
1276 #define MALLOC_ALIGN_MASK (MALLOC_ALIGNMENT - 1)
1277 #define MINSIZE (sizeof(struct malloc_chunk))
1279 /* conversion from malloc headers to user pointers, and back */
1281 #define chunk2mem(p) ((Void_t*)((char*)(p) + 2*SIZE_SZ))
1282 #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ))
1284 /* pad request bytes into a usable size */
1286 #define request2size(req) \
1287 (((long)((req) + (SIZE_SZ + MALLOC_ALIGN_MASK)) < \
1288 (long)(MINSIZE + MALLOC_ALIGN_MASK)) ? MINSIZE : \
1289 (((req) + (SIZE_SZ + MALLOC_ALIGN_MASK)) & ~(MALLOC_ALIGN_MASK)))
1291 /* Check if m has acceptable alignment */
1293 #define aligned_OK(m) (((unsigned long)((m)) & (MALLOC_ALIGN_MASK)) == 0)
1299 Physical chunk operations
1303 /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */
1305 #define PREV_INUSE 0x1
1307 /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */
1309 #define IS_MMAPPED 0x2
1311 /* Bits to mask off when extracting size */
1313 #define SIZE_BITS (PREV_INUSE|IS_MMAPPED)
1316 /* Ptr to next physical malloc_chunk. */
1318 #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->size & ~PREV_INUSE) ))
1320 /* Ptr to previous physical malloc_chunk */
1322 #define prev_chunk(p)\
1323 ((mchunkptr)( ((char*)(p)) - ((p)->prev_size) ))
1326 /* Treat space at ptr + offset as a chunk */
1328 #define chunk_at_offset(p, s) ((mchunkptr)(((char*)(p)) + (s)))
1334 Dealing with use bits
1337 /* extract p's inuse bit */
1340 ((((mchunkptr)(((char*)(p))+((p)->size & ~PREV_INUSE)))->size) & PREV_INUSE)
1342 /* extract inuse bit of previous chunk */
1344 #define prev_inuse(p) ((p)->size & PREV_INUSE)
1346 /* check for mmap()'ed chunk */
1348 #define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED)
1350 /* set/clear chunk as in use without otherwise disturbing */
1352 #define set_inuse(p)\
1353 ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size |= PREV_INUSE
1355 #define clear_inuse(p)\
1356 ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size &= ~(PREV_INUSE)
1358 /* check/set/clear inuse bits in known places */
1360 #define inuse_bit_at_offset(p, s)\
1361 (((mchunkptr)(((char*)(p)) + (s)))->size & PREV_INUSE)
1363 #define set_inuse_bit_at_offset(p, s)\
1364 (((mchunkptr)(((char*)(p)) + (s)))->size |= PREV_INUSE)
1366 #define clear_inuse_bit_at_offset(p, s)\
1367 (((mchunkptr)(((char*)(p)) + (s)))->size &= ~(PREV_INUSE))
1373 Dealing with size fields
1376 /* Get size, ignoring use bits */
1378 #define chunksize(p) ((p)->size & ~(SIZE_BITS))
1380 /* Set size at head, without disturbing its use bit */
1382 #define set_head_size(p, s) ((p)->size = (((p)->size & PREV_INUSE) | (s)))
1384 /* Set size/use ignoring previous bits in header */
1386 #define set_head(p, s) ((p)->size = (s))
1388 /* Set size at footer (only when chunk is not in use) */
1390 #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_size = (s))
1399 The bins, `av_' are an array of pairs of pointers serving as the
1400 heads of (initially empty) doubly-linked lists of chunks, laid out
1401 in a way so that each pair can be treated as if it were in a
1402 malloc_chunk. (This way, the fd/bk offsets for linking bin heads
1403 and chunks are the same).
1405 Bins for sizes < 512 bytes contain chunks of all the same size, spaced
1406 8 bytes apart. Larger bins are approximately logarithmically
1407 spaced. (See the table below.) The `av_' array is never mentioned
1408 directly in the code, but instead via bin access macros.
1416 4 bins of size 32768
1417 2 bins of size 262144
1418 1 bin of size what's left
1420 There is actually a little bit of slop in the numbers in bin_index
1421 for the sake of speed. This makes no difference elsewhere.
1423 The special chunks `top' and `last_remainder' get their own bins,
1424 (this is implemented via yet more trickery with the av_ array),
1425 although `top' is never properly linked to its bin since it is
1426 always handled specially.
1430 #define NAV 128 /* number of bins */
1432 typedef struct malloc_chunk* mbinptr;
1436 #define bin_at(i) ((mbinptr)((char*)&(av_[2*(i) + 2]) - 2*SIZE_SZ))
1437 #define next_bin(b) ((mbinptr)((char*)(b) + 2 * sizeof(mbinptr)))
1438 #define prev_bin(b) ((mbinptr)((char*)(b) - 2 * sizeof(mbinptr)))
1441 The first 2 bins are never indexed. The corresponding av_ cells are instead
1442 used for bookkeeping. This is not to save space, but to simplify
1443 indexing, maintain locality, and avoid some initialization tests.
1446 #define top (bin_at(0)->fd) /* The topmost chunk */
1447 #define last_remainder (bin_at(1)) /* remainder from last split */
1451 Because top initially points to its own bin with initial
1452 zero size, thus forcing extension on the first malloc request,
1453 we avoid having any special code in malloc to check whether
1454 it even exists yet. But we still need to in malloc_extend_top.
1457 #define initial_top ((mchunkptr)(bin_at(0)))
1459 /* Helper macro to initialize bins */
1461 #define IAV(i) bin_at(i), bin_at(i)
1463 static mbinptr av_[NAV * 2 + 2] = {
1465 IAV(0), IAV(1), IAV(2), IAV(3), IAV(4), IAV(5), IAV(6), IAV(7),
1466 IAV(8), IAV(9), IAV(10), IAV(11), IAV(12), IAV(13), IAV(14), IAV(15),
1467 IAV(16), IAV(17), IAV(18), IAV(19), IAV(20), IAV(21), IAV(22), IAV(23),
1468 IAV(24), IAV(25), IAV(26), IAV(27), IAV(28), IAV(29), IAV(30), IAV(31),
1469 IAV(32), IAV(33), IAV(34), IAV(35), IAV(36), IAV(37), IAV(38), IAV(39),
1470 IAV(40), IAV(41), IAV(42), IAV(43), IAV(44), IAV(45), IAV(46), IAV(47),
1471 IAV(48), IAV(49), IAV(50), IAV(51), IAV(52), IAV(53), IAV(54), IAV(55),
1472 IAV(56), IAV(57), IAV(58), IAV(59), IAV(60), IAV(61), IAV(62), IAV(63),
1473 IAV(64), IAV(65), IAV(66), IAV(67), IAV(68), IAV(69), IAV(70), IAV(71),
1474 IAV(72), IAV(73), IAV(74), IAV(75), IAV(76), IAV(77), IAV(78), IAV(79),
1475 IAV(80), IAV(81), IAV(82), IAV(83), IAV(84), IAV(85), IAV(86), IAV(87),
1476 IAV(88), IAV(89), IAV(90), IAV(91), IAV(92), IAV(93), IAV(94), IAV(95),
1477 IAV(96), IAV(97), IAV(98), IAV(99), IAV(100), IAV(101), IAV(102), IAV(103),
1478 IAV(104), IAV(105), IAV(106), IAV(107), IAV(108), IAV(109), IAV(110), IAV(111),
1479 IAV(112), IAV(113), IAV(114), IAV(115), IAV(116), IAV(117), IAV(118), IAV(119),
1480 IAV(120), IAV(121), IAV(122), IAV(123), IAV(124), IAV(125), IAV(126), IAV(127)
1485 /* field-extraction macros */
1487 #define first(b) ((b)->fd)
1488 #define last(b) ((b)->bk)
1494 #define bin_index(sz) \
1495 (((((unsigned long)(sz)) >> 9) == 0) ? (((unsigned long)(sz)) >> 3): \
1496 ((((unsigned long)(sz)) >> 9) <= 4) ? 56 + (((unsigned long)(sz)) >> 6): \
1497 ((((unsigned long)(sz)) >> 9) <= 20) ? 91 + (((unsigned long)(sz)) >> 9): \
1498 ((((unsigned long)(sz)) >> 9) <= 84) ? 110 + (((unsigned long)(sz)) >> 12): \
1499 ((((unsigned long)(sz)) >> 9) <= 340) ? 119 + (((unsigned long)(sz)) >> 15): \
1500 ((((unsigned long)(sz)) >> 9) <= 1364) ? 124 + (((unsigned long)(sz)) >> 18): \
1503 bins for chunks < 512 are all spaced 8 bytes apart, and hold
1504 identically sized chunks. This is exploited in malloc.
1507 #define MAX_SMALLBIN 63
1508 #define MAX_SMALLBIN_SIZE 512
1509 #define SMALLBIN_WIDTH 8
1511 #define smallbin_index(sz) (((unsigned long)(sz)) >> 3)
1514 Requests are `small' if both the corresponding and the next bin are small
1517 #define is_small_request(nb) (nb < MAX_SMALLBIN_SIZE - SMALLBIN_WIDTH)
1522 To help compensate for the large number of bins, a one-level index
1523 structure is used for bin-by-bin searching. `binblocks' is a
1524 one-word bitvector recording whether groups of BINBLOCKWIDTH bins
1525 have any (possibly) non-empty bins, so they can be skipped over
1526 all at once during during traversals. The bits are NOT always
1527 cleared as soon as all bins in a block are empty, but instead only
1528 when all are noticed to be empty during traversal in malloc.
1531 #define BINBLOCKWIDTH 4 /* bins per block */
1533 #define binblocks (bin_at(0)->size) /* bitvector of nonempty blocks */
1535 /* bin<->block macros */
1537 #define idx2binblock(ix) ((unsigned)1 << (ix / BINBLOCKWIDTH))
1538 #define mark_binblock(ii) (binblocks |= idx2binblock(ii))
1539 #define clear_binblock(ii) (binblocks &= ~(idx2binblock(ii)))
1545 /* Other static bookkeeping data */
1547 /* variables holding tunable values */
1549 static unsigned long trim_threshold = DEFAULT_TRIM_THRESHOLD;
1550 static unsigned long top_pad = DEFAULT_TOP_PAD;
1551 static unsigned int n_mmaps_max = DEFAULT_MMAP_MAX;
1552 static unsigned long mmap_threshold = DEFAULT_MMAP_THRESHOLD;
1554 /* The first value returned from sbrk */
1555 static char* sbrk_base = (char*)(-1);
1557 /* The maximum memory obtained from system via sbrk */
1558 static unsigned long max_sbrked_mem = 0;
1560 /* The maximum via either sbrk or mmap */
1561 static unsigned long max_total_mem = 0;
1563 /* internal working copy of mallinfo */
1564 static struct mallinfo current_mallinfo = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
1566 /* The total memory obtained from system via sbrk */
1567 #define sbrked_mem (current_mallinfo.arena)
1569 /* Tracking mmaps */
1571 static unsigned int n_mmaps = 0;
1572 static unsigned int max_n_mmaps = 0;
1573 static unsigned long mmapped_mem = 0;
1574 static unsigned long max_mmapped_mem = 0;
1586 These routines make a number of assertions about the states
1587 of data structures that should be true at all times. If any
1588 are not true, it's very likely that a user program has somehow
1589 trashed memory. (It's also possible that there is a coding error
1590 in malloc. In which case, please report it!)
1594 static void do_check_chunk(mchunkptr p)
1596 static void do_check_chunk(p) mchunkptr p;
1599 INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1601 /* No checkable chunk is mmapped */
1602 assert(!chunk_is_mmapped(p));
1604 /* Check for legal address ... */
1605 assert((char*)p >= sbrk_base);
1607 assert((char*)p + sz <= (char*)top);
1609 assert((char*)p + sz <= sbrk_base + sbrked_mem);
1615 static void do_check_free_chunk(mchunkptr p)
1617 static void do_check_free_chunk(p) mchunkptr p;
1620 INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1621 mchunkptr next = chunk_at_offset(p, sz);
1625 /* Check whether it claims to be free ... */
1628 /* Unless a special marker, must have OK fields */
1629 if ((long)sz >= (long)MINSIZE)
1631 assert((sz & MALLOC_ALIGN_MASK) == 0);
1632 assert(aligned_OK(chunk2mem(p)));
1633 /* ... matching footer field */
1634 assert(next->prev_size == sz);
1635 /* ... and is fully consolidated */
1636 assert(prev_inuse(p));
1637 assert (next == top || inuse(next));
1639 /* ... and has minimally sane links */
1640 assert(p->fd->bk == p);
1641 assert(p->bk->fd == p);
1643 else /* markers are always of size SIZE_SZ */
1644 assert(sz == SIZE_SZ);
1648 static void do_check_inuse_chunk(mchunkptr p)
1650 static void do_check_inuse_chunk(p) mchunkptr p;
1653 mchunkptr next = next_chunk(p);
1656 /* Check whether it claims to be in use ... */
1659 /* ... and is surrounded by OK chunks.
1660 Since more things can be checked with free chunks than inuse ones,
1661 if an inuse chunk borders them and debug is on, it's worth doing them.
1665 mchunkptr prv = prev_chunk(p);
1666 assert(next_chunk(prv) == p);
1667 do_check_free_chunk(prv);
1671 assert(prev_inuse(next));
1672 assert(chunksize(next) >= MINSIZE);
1674 else if (!inuse(next))
1675 do_check_free_chunk(next);
1680 static void do_check_malloced_chunk(mchunkptr p, INTERNAL_SIZE_T s)
1682 static void do_check_malloced_chunk(p, s) mchunkptr p; INTERNAL_SIZE_T s;
1685 INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1688 do_check_inuse_chunk(p);
1690 /* Legal size ... */
1691 assert((long)sz >= (long)MINSIZE);
1692 assert((sz & MALLOC_ALIGN_MASK) == 0);
1694 assert(room < (long)MINSIZE);
1696 /* ... and alignment */
1697 assert(aligned_OK(chunk2mem(p)));
1700 /* ... and was allocated at front of an available chunk */
1701 assert(prev_inuse(p));
1706 #define check_free_chunk(P) do_check_free_chunk(P)
1707 #define check_inuse_chunk(P) do_check_inuse_chunk(P)
1708 #define check_chunk(P) do_check_chunk(P)
1709 #define check_malloced_chunk(P,N) do_check_malloced_chunk(P,N)
1711 #define check_free_chunk(P)
1712 #define check_inuse_chunk(P)
1713 #define check_chunk(P)
1714 #define check_malloced_chunk(P,N)
1720 Macro-based internal utilities
1725 Linking chunks in bin lists.
1726 Call these only with variables, not arbitrary expressions, as arguments.
1730 Place chunk p of size s in its bin, in size order,
1731 putting it ahead of others of same size.
1735 #define frontlink(P, S, IDX, BK, FD) \
1737 if (S < MAX_SMALLBIN_SIZE) \
1739 IDX = smallbin_index(S); \
1740 mark_binblock(IDX); \
1745 FD->bk = BK->fd = P; \
1749 IDX = bin_index(S); \
1752 if (FD == BK) mark_binblock(IDX); \
1755 while (FD != BK && S < chunksize(FD)) FD = FD->fd; \
1760 FD->bk = BK->fd = P; \
1765 /* take a chunk off a list */
1767 #define unlink(P, BK, FD) \
1775 /* Place p as the last remainder */
1777 #define link_last_remainder(P) \
1779 last_remainder->fd = last_remainder->bk = P; \
1780 P->fd = P->bk = last_remainder; \
1783 /* Clear the last_remainder bin */
1785 #define clear_last_remainder \
1786 (last_remainder->fd = last_remainder->bk = last_remainder)
1792 /* Routines dealing with mmap(). */
1797 static mchunkptr mmap_chunk(size_t size)
1799 static mchunkptr mmap_chunk(size) size_t size;
1802 size_t page_mask = malloc_getpagesize - 1;
1805 #ifndef MAP_ANONYMOUS
1809 if(n_mmaps >= n_mmaps_max) return 0; /* too many regions */
1811 /* For mmapped chunks, the overhead is one SIZE_SZ unit larger, because
1812 * there is no following chunk whose prev_size field could be used.
1814 size = (size + SIZE_SZ + page_mask) & ~page_mask;
1816 #ifdef MAP_ANONYMOUS
1817 p = (mchunkptr)mmap(0, size, PROT_READ|PROT_WRITE,
1818 MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1819 #else /* !MAP_ANONYMOUS */
1822 fd = open("/dev/zero", O_RDWR);
1823 if(fd < 0) return 0;
1825 p = (mchunkptr)mmap(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0);
1828 if(p == (mchunkptr)-1) return 0;
1831 if (n_mmaps > max_n_mmaps) max_n_mmaps = n_mmaps;
1833 /* We demand that eight bytes into a page must be 8-byte aligned. */
1834 assert(aligned_OK(chunk2mem(p)));
1836 /* The offset to the start of the mmapped region is stored
1837 * in the prev_size field of the chunk; normally it is zero,
1838 * but that can be changed in memalign().
1841 set_head(p, size|IS_MMAPPED);
1843 mmapped_mem += size;
1844 if ((unsigned long)mmapped_mem > (unsigned long)max_mmapped_mem)
1845 max_mmapped_mem = mmapped_mem;
1846 if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem)
1847 max_total_mem = mmapped_mem + sbrked_mem;
1852 static void munmap_chunk(mchunkptr p)
1854 static void munmap_chunk(p) mchunkptr p;
1857 INTERNAL_SIZE_T size = chunksize(p);
1860 assert (chunk_is_mmapped(p));
1861 assert(! ((char*)p >= sbrk_base && (char*)p < sbrk_base + sbrked_mem));
1862 assert((n_mmaps > 0));
1863 assert(((p->prev_size + size) & (malloc_getpagesize-1)) == 0);
1866 mmapped_mem -= (size + p->prev_size);
1868 ret = munmap((char *)p - p->prev_size, size + p->prev_size);
1870 /* munmap returns non-zero on failure */
1877 static mchunkptr mremap_chunk(mchunkptr p, size_t new_size)
1879 static mchunkptr mremap_chunk(p, new_size) mchunkptr p; size_t new_size;
1882 size_t page_mask = malloc_getpagesize - 1;
1883 INTERNAL_SIZE_T offset = p->prev_size;
1884 INTERNAL_SIZE_T size = chunksize(p);
1887 assert (chunk_is_mmapped(p));
1888 assert(! ((char*)p >= sbrk_base && (char*)p < sbrk_base + sbrked_mem));
1889 assert((n_mmaps > 0));
1890 assert(((size + offset) & (malloc_getpagesize-1)) == 0);
1892 /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */
1893 new_size = (new_size + offset + SIZE_SZ + page_mask) & ~page_mask;
1895 cp = (char *)mremap((char *)p - offset, size + offset, new_size, 1);
1897 if (cp == (char *)-1) return 0;
1899 p = (mchunkptr)(cp + offset);
1901 assert(aligned_OK(chunk2mem(p)));
1903 assert((p->prev_size == offset));
1904 set_head(p, (new_size - offset)|IS_MMAPPED);
1906 mmapped_mem -= size + offset;
1907 mmapped_mem += new_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;
1915 #endif /* HAVE_MREMAP */
1917 #endif /* HAVE_MMAP */
1923 Extend the top-most chunk by obtaining memory from system.
1924 Main interface to sbrk (but see also malloc_trim).
1928 static void malloc_extend_top(INTERNAL_SIZE_T nb)
1930 static void malloc_extend_top(nb) INTERNAL_SIZE_T nb;
1933 char* brk; /* return value from sbrk */
1934 INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of sbrked space */
1935 INTERNAL_SIZE_T correction; /* bytes for 2nd sbrk call */
1936 char* new_brk; /* return of 2nd sbrk call */
1937 INTERNAL_SIZE_T top_size; /* new size of top chunk */
1939 mchunkptr old_top = top; /* Record state of old top */
1940 INTERNAL_SIZE_T old_top_size = chunksize(old_top);
1941 char* old_end = (char*)(chunk_at_offset(old_top, old_top_size));
1943 /* Pad request with top_pad plus minimal overhead */
1945 INTERNAL_SIZE_T sbrk_size = nb + top_pad + MINSIZE;
1946 unsigned long pagesz = malloc_getpagesize;
1948 /* If not the first time through, round to preserve page boundary */
1949 /* Otherwise, we need to correct to a page size below anyway. */
1950 /* (We also correct below if an intervening foreign sbrk call.) */
1952 if (sbrk_base != (char*)(-1))
1953 sbrk_size = (sbrk_size + (pagesz - 1)) & ~(pagesz - 1);
1955 brk = (char*)(MORECORE (sbrk_size));
1957 /* Fail if sbrk failed or if a foreign sbrk call killed our space */
1958 if (brk == (char*)(MORECORE_FAILURE) ||
1959 (brk < old_end && old_top != initial_top))
1962 sbrked_mem += sbrk_size;
1964 if (brk == old_end) /* can just add bytes to current top */
1966 top_size = sbrk_size + old_top_size;
1967 set_head(top, top_size | PREV_INUSE);
1971 if (sbrk_base == (char*)(-1)) /* First time through. Record base */
1973 else /* Someone else called sbrk(). Count those bytes as sbrked_mem. */
1974 sbrked_mem += brk - (char*)old_end;
1976 /* Guarantee alignment of first new chunk made from this space */
1977 front_misalign = (unsigned long)chunk2mem(brk) & MALLOC_ALIGN_MASK;
1978 if (front_misalign > 0)
1980 correction = (MALLOC_ALIGNMENT) - front_misalign;
1986 /* Guarantee the next brk will be at a page boundary */
1988 correction += ((((unsigned long)(brk + sbrk_size))+(pagesz-1)) &
1989 ~(pagesz - 1)) - ((unsigned long)(brk + sbrk_size));
1991 /* Allocate correction */
1992 new_brk = (char*)(MORECORE (correction));
1993 if (new_brk == (char*)(MORECORE_FAILURE)) return;
1995 sbrked_mem += correction;
1997 top = (mchunkptr)brk;
1998 top_size = new_brk - brk + correction;
1999 set_head(top, top_size | PREV_INUSE);
2001 if (old_top != initial_top)
2004 /* There must have been an intervening foreign sbrk call. */
2005 /* A double fencepost is necessary to prevent consolidation */
2007 /* If not enough space to do this, then user did something very wrong */
2008 if (old_top_size < MINSIZE)
2010 set_head(top, PREV_INUSE); /* will force null return from malloc */
2014 /* Also keep size a multiple of MALLOC_ALIGNMENT */
2015 old_top_size = (old_top_size - 3*SIZE_SZ) & ~MALLOC_ALIGN_MASK;
2016 set_head_size(old_top, old_top_size);
2017 chunk_at_offset(old_top, old_top_size )->size =
2019 chunk_at_offset(old_top, old_top_size + SIZE_SZ)->size =
2021 /* If possible, release the rest. */
2022 if (old_top_size >= MINSIZE)
2023 fREe(chunk2mem(old_top));
2027 if ((unsigned long)sbrked_mem > (unsigned long)max_sbrked_mem)
2028 max_sbrked_mem = sbrked_mem;
2029 if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem)
2030 max_total_mem = mmapped_mem + sbrked_mem;
2032 /* We always land on a page boundary */
2033 assert(((unsigned long)((char*)top + top_size) & (pagesz - 1)) == 0);
2039 /* Main public routines */
2045 The requested size is first converted into a usable form, `nb'.
2046 This currently means to add 4 bytes overhead plus possibly more to
2047 obtain 8-byte alignment and/or to obtain a size of at least
2048 MINSIZE (currently 16 bytes), the smallest allocatable size.
2049 (All fits are considered `exact' if they are within MINSIZE bytes.)
2051 From there, the first successful of the following steps is taken:
2053 1. The bin corresponding to the request size is scanned, and if
2054 a chunk of exactly the right size is found, it is taken.
2056 2. The most recently remaindered chunk is used if it is big
2057 enough. This is a form of (roving) first fit, used only in
2058 the absence of exact fits. Runs of consecutive requests use
2059 the remainder of the chunk used for the previous such request
2060 whenever possible. This limited use of a first-fit style
2061 allocation strategy tends to give contiguous chunks
2062 coextensive lifetimes, which improves locality and can reduce
2063 fragmentation in the long run.
2065 3. Other bins are scanned in increasing size order, using a
2066 chunk big enough to fulfill the request, and splitting off
2067 any remainder. This search is strictly by best-fit; i.e.,
2068 the smallest (with ties going to approximately the least
2069 recently used) chunk that fits is selected.
2071 4. If large enough, the chunk bordering the end of memory
2072 (`top') is split off. (This use of `top' is in accord with
2073 the best-fit search rule. In effect, `top' is treated as
2074 larger (and thus less well fitting) than any other available
2075 chunk since it can be extended to be as large as necessary
2076 (up to system limitations).
2078 5. If the request size meets the mmap threshold and the
2079 system supports mmap, and there are few enough currently
2080 allocated mmapped regions, and a call to mmap succeeds,
2081 the request is allocated via direct memory mapping.
2083 6. Otherwise, the top of memory is extended by
2084 obtaining more space from the system (normally using sbrk,
2085 but definable to anything else via the MORECORE macro).
2086 Memory is gathered from the system (in system page-sized
2087 units) in a way that allows chunks obtained across different
2088 sbrk calls to be consolidated, but does not require
2089 contiguous memory. Thus, it should be safe to intersperse
2090 mallocs with other sbrk calls.
2093 All allocations are made from the the `lowest' part of any found
2094 chunk. (The implementation invariant is that prev_inuse is
2095 always true of any allocated chunk; i.e., that each allocated
2096 chunk borders either a previously allocated and still in-use chunk,
2097 or the base of its memory arena.)
2102 Void_t* mALLOc(size_t bytes)
2104 Void_t* mALLOc(bytes) size_t bytes;
2107 mchunkptr victim; /* inspected/selected chunk */
2108 INTERNAL_SIZE_T victim_size; /* its size */
2109 int idx; /* index for bin traversal */
2110 mbinptr bin; /* associated bin */
2111 mchunkptr remainder; /* remainder from a split */
2112 long remainder_size; /* its size */
2113 int remainder_index; /* its bin index */
2114 unsigned long block; /* block traverser bit */
2115 int startidx; /* first bin of a traversed block */
2116 mchunkptr fwd; /* misc temp for linking */
2117 mchunkptr bck; /* misc temp for linking */
2118 mbinptr q; /* misc temp */
2122 if ((long)bytes < 0) return 0;
2124 nb = request2size(bytes); /* padded request size; */
2126 /* Check for exact match in a bin */
2128 if (is_small_request(nb)) /* Faster version for small requests */
2130 idx = smallbin_index(nb);
2132 /* No traversal or size check necessary for small bins. */
2137 /* Also scan the next one, since it would have a remainder < MINSIZE */
2145 victim_size = chunksize(victim);
2146 unlink(victim, bck, fwd);
2147 set_inuse_bit_at_offset(victim, victim_size);
2148 check_malloced_chunk(victim, nb);
2149 return chunk2mem(victim);
2152 idx += 2; /* Set for bin scan below. We've already scanned 2 bins. */
2157 idx = bin_index(nb);
2160 for (victim = last(bin); victim != bin; victim = victim->bk)
2162 victim_size = chunksize(victim);
2163 remainder_size = victim_size - nb;
2165 if (remainder_size >= (long)MINSIZE) /* too big */
2167 --idx; /* adjust to rescan below after checking last remainder */
2171 else if (remainder_size >= 0) /* exact fit */
2173 unlink(victim, bck, fwd);
2174 set_inuse_bit_at_offset(victim, victim_size);
2175 check_malloced_chunk(victim, nb);
2176 return chunk2mem(victim);
2184 /* Try to use the last split-off remainder */
2186 if ( (victim = last_remainder->fd) != last_remainder)
2188 victim_size = chunksize(victim);
2189 remainder_size = victim_size - nb;
2191 if (remainder_size >= (long)MINSIZE) /* re-split */
2193 remainder = chunk_at_offset(victim, nb);
2194 set_head(victim, nb | PREV_INUSE);
2195 link_last_remainder(remainder);
2196 set_head(remainder, remainder_size | PREV_INUSE);
2197 set_foot(remainder, remainder_size);
2198 check_malloced_chunk(victim, nb);
2199 return chunk2mem(victim);
2202 clear_last_remainder;
2204 if (remainder_size >= 0) /* exhaust */
2206 set_inuse_bit_at_offset(victim, victim_size);
2207 check_malloced_chunk(victim, nb);
2208 return chunk2mem(victim);
2211 /* Else place in bin */
2213 frontlink(victim, victim_size, remainder_index, bck, fwd);
2217 If there are any possibly nonempty big-enough blocks,
2218 search for best fitting chunk by scanning bins in blockwidth units.
2221 if ( (block = idx2binblock(idx)) <= binblocks)
2224 /* Get to the first marked block */
2226 if ( (block & binblocks) == 0)
2228 /* force to an even block boundary */
2229 idx = (idx & ~(BINBLOCKWIDTH - 1)) + BINBLOCKWIDTH;
2231 while ((block & binblocks) == 0)
2233 idx += BINBLOCKWIDTH;
2238 /* For each possibly nonempty block ... */
2241 startidx = idx; /* (track incomplete blocks) */
2242 q = bin = bin_at(idx);
2244 /* For each bin in this block ... */
2247 /* Find and use first big enough chunk ... */
2249 for (victim = last(bin); victim != bin; victim = victim->bk)
2251 victim_size = chunksize(victim);
2252 remainder_size = victim_size - nb;
2254 if (remainder_size >= (long)MINSIZE) /* split */
2256 remainder = chunk_at_offset(victim, nb);
2257 set_head(victim, nb | PREV_INUSE);
2258 unlink(victim, bck, fwd);
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 else if (remainder_size >= 0) /* take */
2268 set_inuse_bit_at_offset(victim, victim_size);
2269 unlink(victim, bck, fwd);
2270 check_malloced_chunk(victim, nb);
2271 return chunk2mem(victim);
2276 bin = next_bin(bin);
2278 } while ((++idx & (BINBLOCKWIDTH - 1)) != 0);
2280 /* Clear out the block bit. */
2282 do /* Possibly backtrack to try to clear a partial block */
2284 if ((startidx & (BINBLOCKWIDTH - 1)) == 0)
2286 binblocks &= ~block;
2291 } while (first(q) == q);
2293 /* Get to the next possibly nonempty block */
2295 if ( (block <<= 1) <= binblocks && (block != 0) )
2297 while ((block & binblocks) == 0)
2299 idx += BINBLOCKWIDTH;
2309 /* Try to use top chunk */
2311 /* Require that there be a remainder, ensuring top always exists */
2312 if ( (remainder_size = chunksize(top) - nb) < (long)MINSIZE)
2316 /* If big and would otherwise need to extend, try to use mmap instead */
2317 if ((unsigned long)nb >= (unsigned long)mmap_threshold &&
2318 (victim = mmap_chunk(nb)) != 0)
2319 return chunk2mem(victim);
2323 malloc_extend_top(nb);
2324 if ( (remainder_size = chunksize(top) - nb) < (long)MINSIZE)
2325 return 0; /* propagate failure */
2329 set_head(victim, nb | PREV_INUSE);
2330 top = chunk_at_offset(victim, nb);
2331 set_head(top, remainder_size | PREV_INUSE);
2332 check_malloced_chunk(victim, nb);
2333 return chunk2mem(victim);
2346 1. free(0) has no effect.
2348 2. If the chunk was allocated via mmap, it is release via munmap().
2350 3. If a returned chunk borders the current high end of memory,
2351 it is consolidated into the top, and if the total unused
2352 topmost memory exceeds the trim threshold, malloc_trim is
2355 4. Other chunks are consolidated as they arrive, and
2356 placed in corresponding bins. (This includes the case of
2357 consolidating with the current `last_remainder').
2363 void fREe(Void_t* mem)
2365 void fREe(mem) Void_t* mem;
2368 mchunkptr p; /* chunk corresponding to mem */
2369 INTERNAL_SIZE_T hd; /* its head field */
2370 INTERNAL_SIZE_T sz; /* its size */
2371 int idx; /* its bin index */
2372 mchunkptr next; /* next contiguous chunk */
2373 INTERNAL_SIZE_T nextsz; /* its size */
2374 INTERNAL_SIZE_T prevsz; /* size of previous contiguous chunk */
2375 mchunkptr bck; /* misc temp for linking */
2376 mchunkptr fwd; /* misc temp for linking */
2377 int islr; /* track whether merging with last_remainder */
2379 if (mem == 0) /* free(0) has no effect */
2386 if (hd & IS_MMAPPED) /* release mmapped memory. */
2393 check_inuse_chunk(p);
2395 sz = hd & ~PREV_INUSE;
2396 next = chunk_at_offset(p, sz);
2397 nextsz = chunksize(next);
2399 if (next == top) /* merge with top */
2403 if (!(hd & PREV_INUSE)) /* consolidate backward */
2405 prevsz = p->prev_size;
2406 p = chunk_at_offset(p, -((long) prevsz));
2408 unlink(p, bck, fwd);
2411 set_head(p, sz | PREV_INUSE);
2413 if ((unsigned long)(sz) >= (unsigned long)trim_threshold)
2414 malloc_trim(top_pad);
2418 set_head(next, nextsz); /* clear inuse bit */
2422 if (!(hd & PREV_INUSE)) /* consolidate backward */
2424 prevsz = p->prev_size;
2425 p = chunk_at_offset(p, -((long) prevsz));
2428 if (p->fd == last_remainder) /* keep as last_remainder */
2431 unlink(p, bck, fwd);
2434 if (!(inuse_bit_at_offset(next, nextsz))) /* consolidate forward */
2438 if (!islr && next->fd == last_remainder) /* re-insert last_remainder */
2441 link_last_remainder(p);
2444 unlink(next, bck, fwd);
2448 set_head(p, sz | PREV_INUSE);
2451 frontlink(p, sz, idx, bck, fwd);
2462 Chunks that were obtained via mmap cannot be extended or shrunk
2463 unless HAVE_MREMAP is defined, in which case mremap is used.
2464 Otherwise, if their reallocation is for additional space, they are
2465 copied. If for less, they are just left alone.
2467 Otherwise, if the reallocation is for additional space, and the
2468 chunk can be extended, it is, else a malloc-copy-free sequence is
2469 taken. There are several different ways that a chunk could be
2470 extended. All are tried:
2472 * Extending forward into following adjacent free chunk.
2473 * Shifting backwards, joining preceding adjacent space
2474 * Both shifting backwards and extending forward.
2475 * Extending into newly sbrked space
2477 Unless the #define REALLOC_ZERO_BYTES_FREES is set, realloc with a
2478 size argument of zero (re)allocates a minimum-sized chunk.
2480 If the reallocation is for less space, and the new request is for
2481 a `small' (<512 bytes) size, then the newly unused space is lopped
2484 The old unix realloc convention of allowing the last-free'd chunk
2485 to be used as an argument to realloc is no longer supported.
2486 I don't know of any programs still relying on this feature,
2487 and allowing it would also allow too many other incorrect
2488 usages of realloc to be sensible.
2495 Void_t* rEALLOc(Void_t* oldmem, size_t bytes)
2497 Void_t* rEALLOc(oldmem, bytes) Void_t* oldmem; size_t bytes;
2500 INTERNAL_SIZE_T nb; /* padded request size */
2502 mchunkptr oldp; /* chunk corresponding to oldmem */
2503 INTERNAL_SIZE_T oldsize; /* its size */
2505 mchunkptr newp; /* chunk to return */
2506 INTERNAL_SIZE_T newsize; /* its size */
2507 Void_t* newmem; /* corresponding user mem */
2509 mchunkptr next; /* next contiguous chunk after oldp */
2510 INTERNAL_SIZE_T nextsize; /* its size */
2512 mchunkptr prev; /* previous contiguous chunk before oldp */
2513 INTERNAL_SIZE_T prevsize; /* its size */
2515 mchunkptr remainder; /* holds split off extra space from newp */
2516 INTERNAL_SIZE_T remainder_size; /* its size */
2518 mchunkptr bck; /* misc temp for linking */
2519 mchunkptr fwd; /* misc temp for linking */
2521 #ifdef REALLOC_ZERO_BYTES_FREES
2522 if (bytes == 0) { fREe(oldmem); return 0; }
2525 if ((long)bytes < 0) return 0;
2527 /* realloc of null is supposed to be same as malloc */
2528 if (oldmem == 0) return mALLOc(bytes);
2530 newp = oldp = mem2chunk(oldmem);
2531 newsize = oldsize = chunksize(oldp);
2534 nb = request2size(bytes);
2537 if (chunk_is_mmapped(oldp))
2540 newp = mremap_chunk(oldp, nb);
2541 if(newp) return chunk2mem(newp);
2543 /* Note the extra SIZE_SZ overhead. */
2544 if(oldsize - SIZE_SZ >= nb) return oldmem; /* do nothing */
2545 /* Must alloc, copy, free. */
2546 newmem = mALLOc(bytes);
2547 if (newmem == 0) return 0; /* propagate failure */
2548 MALLOC_COPY(newmem, oldmem, oldsize - 2*SIZE_SZ);
2554 check_inuse_chunk(oldp);
2556 if ((long)(oldsize) < (long)(nb))
2559 /* Try expanding forward */
2561 next = chunk_at_offset(oldp, oldsize);
2562 if (next == top || !inuse(next))
2564 nextsize = chunksize(next);
2566 /* Forward into top only if a remainder */
2569 if ((long)(nextsize + newsize) >= (long)(nb + MINSIZE))
2571 newsize += nextsize;
2572 top = chunk_at_offset(oldp, nb);
2573 set_head(top, (newsize - nb) | PREV_INUSE);
2574 set_head_size(oldp, nb);
2575 return chunk2mem(oldp);
2579 /* Forward into next chunk */
2580 else if (((long)(nextsize + newsize) >= (long)(nb)))
2582 unlink(next, bck, fwd);
2583 newsize += nextsize;
2593 /* Try shifting backwards. */
2595 if (!prev_inuse(oldp))
2597 prev = prev_chunk(oldp);
2598 prevsize = chunksize(prev);
2600 /* try forward + backward first to save a later consolidation */
2607 if ((long)(nextsize + prevsize + newsize) >= (long)(nb + MINSIZE))
2609 unlink(prev, bck, fwd);
2611 newsize += prevsize + nextsize;
2612 newmem = chunk2mem(newp);
2613 MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2614 top = chunk_at_offset(newp, nb);
2615 set_head(top, (newsize - nb) | PREV_INUSE);
2616 set_head_size(newp, nb);
2621 /* into next chunk */
2622 else if (((long)(nextsize + prevsize + newsize) >= (long)(nb)))
2624 unlink(next, bck, fwd);
2625 unlink(prev, bck, fwd);
2627 newsize += nextsize + prevsize;
2628 newmem = chunk2mem(newp);
2629 MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2635 if (prev != 0 && (long)(prevsize + newsize) >= (long)nb)
2637 unlink(prev, bck, fwd);
2639 newsize += prevsize;
2640 newmem = chunk2mem(newp);
2641 MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2648 newmem = mALLOc (bytes);
2650 if (newmem == 0) /* propagate failure */
2653 /* Avoid copy if newp is next chunk after oldp. */
2654 /* (This can only happen when new chunk is sbrk'ed.) */
2656 if ( (newp = mem2chunk(newmem)) == next_chunk(oldp))
2658 newsize += chunksize(newp);
2663 /* Otherwise copy, free, and exit */
2664 MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2670 split: /* split off extra room in old or expanded chunk */
2672 if (newsize - nb >= MINSIZE) /* split off remainder */
2674 remainder = chunk_at_offset(newp, nb);
2675 remainder_size = newsize - nb;
2676 set_head_size(newp, nb);
2677 set_head(remainder, remainder_size | PREV_INUSE);
2678 set_inuse_bit_at_offset(remainder, remainder_size);
2679 fREe(chunk2mem(remainder)); /* let free() deal with it */
2683 set_head_size(newp, newsize);
2684 set_inuse_bit_at_offset(newp, newsize);
2687 check_inuse_chunk(newp);
2688 return chunk2mem(newp);
2698 memalign requests more than enough space from malloc, finds a spot
2699 within that chunk that meets the alignment request, and then
2700 possibly frees the leading and trailing space.
2702 The alignment argument must be a power of two. This property is not
2703 checked by memalign, so misuse may result in random runtime errors.
2705 8-byte alignment is guaranteed by normal malloc calls, so don't
2706 bother calling memalign with an argument of 8 or less.
2708 Overreliance on memalign is a sure way to fragment space.
2714 Void_t* mEMALIGn(size_t alignment, size_t bytes)
2716 Void_t* mEMALIGn(alignment, bytes) size_t alignment; size_t bytes;
2719 INTERNAL_SIZE_T nb; /* padded request size */
2720 char* m; /* memory returned by malloc call */
2721 mchunkptr p; /* corresponding chunk */
2722 char* brk; /* alignment point within p */
2723 mchunkptr newp; /* chunk to return */
2724 INTERNAL_SIZE_T newsize; /* its size */
2725 INTERNAL_SIZE_T leadsize; /* leading space befor alignment point */
2726 mchunkptr remainder; /* spare room at end to split off */
2727 long remainder_size; /* its size */
2729 if ((long)bytes < 0) return 0;
2731 /* If need less alignment than we give anyway, just relay to malloc */
2733 if (alignment <= MALLOC_ALIGNMENT) return mALLOc(bytes);
2735 /* Otherwise, ensure that it is at least a minimum chunk size */
2737 if (alignment < MINSIZE) alignment = MINSIZE;
2739 /* Call malloc with worst case padding to hit alignment. */
2741 nb = request2size(bytes);
2742 m = (char*)(mALLOc(nb + alignment + MINSIZE));
2744 if (m == 0) return 0; /* propagate failure */
2748 if ((((unsigned long)(m)) % alignment) == 0) /* aligned */
2751 if(chunk_is_mmapped(p))
2752 return chunk2mem(p); /* nothing more to do */
2755 else /* misaligned */
2758 Find an aligned spot inside chunk.
2759 Since we need to give back leading space in a chunk of at
2760 least MINSIZE, if the first calculation places us at
2761 a spot with less than MINSIZE leader, we can move to the
2762 next aligned spot -- we've allocated enough total room so that
2763 this is always possible.
2766 brk = (char*)mem2chunk(((unsigned long)(m + alignment - 1)) & -((signed) alignment));
2767 if ((long)(brk - (char*)(p)) < MINSIZE) brk = brk + alignment;
2769 newp = (mchunkptr)brk;
2770 leadsize = brk - (char*)(p);
2771 newsize = chunksize(p) - leadsize;
2774 if(chunk_is_mmapped(p))
2776 newp->prev_size = p->prev_size + leadsize;
2777 set_head(newp, newsize|IS_MMAPPED);
2778 return chunk2mem(newp);
2782 /* give back leader, use the rest */
2784 set_head(newp, newsize | PREV_INUSE);
2785 set_inuse_bit_at_offset(newp, newsize);
2786 set_head_size(p, leadsize);
2790 assert (newsize >= nb && (((unsigned long)(chunk2mem(p))) % alignment) == 0);
2793 /* Also give back spare room at the end */
2795 remainder_size = chunksize(p) - nb;
2797 if (remainder_size >= (long)MINSIZE)
2799 remainder = chunk_at_offset(p, nb);
2800 set_head(remainder, remainder_size | PREV_INUSE);
2801 set_head_size(p, nb);
2802 fREe(chunk2mem(remainder));
2805 check_inuse_chunk(p);
2806 return chunk2mem(p);
2814 valloc just invokes memalign with alignment argument equal
2815 to the page size of the system (or as near to this as can
2816 be figured out from all the includes/defines above.)
2820 Void_t* vALLOc(size_t bytes)
2822 Void_t* vALLOc(bytes) size_t bytes;
2825 return mEMALIGn (malloc_getpagesize, bytes);
2829 pvalloc just invokes valloc for the nearest pagesize
2830 that will accommodate request
2835 Void_t* pvALLOc(size_t bytes)
2837 Void_t* pvALLOc(bytes) size_t bytes;
2840 size_t pagesize = malloc_getpagesize;
2841 return mEMALIGn (pagesize, (bytes + pagesize - 1) & ~(pagesize - 1));
2846 calloc calls malloc, then zeroes out the allocated chunk.
2851 Void_t* cALLOc(size_t n, size_t elem_size)
2853 Void_t* cALLOc(n, elem_size) size_t n; size_t elem_size;
2857 INTERNAL_SIZE_T csz;
2859 INTERNAL_SIZE_T sz = n * elem_size;
2862 /* check if expand_top called, in which case don't need to clear */
2864 mchunkptr oldtop = top;
2865 INTERNAL_SIZE_T oldtopsize = chunksize(top);
2867 Void_t* mem = mALLOc (sz);
2869 if ((long)n < 0) return 0;
2877 /* Two optional cases in which clearing not necessary */
2881 if (chunk_is_mmapped(p)) return mem;
2887 if (p == oldtop && csz > oldtopsize)
2889 /* clear only the bytes from non-freshly-sbrked memory */
2894 MALLOC_ZERO(mem, csz - SIZE_SZ);
2901 cfree just calls free. It is needed/defined on some systems
2902 that pair it with calloc, presumably for odd historical reasons.
2906 #if !defined(INTERNAL_LINUX_C_LIB) || !defined(__ELF__)
2908 void cfree(Void_t *mem)
2910 void cfree(mem) Void_t *mem;
2921 Malloc_trim gives memory back to the system (via negative
2922 arguments to sbrk) if there is unused memory at the `high' end of
2923 the malloc pool. You can call this after freeing large blocks of
2924 memory to potentially reduce the system-level memory requirements
2925 of a program. However, it cannot guarantee to reduce memory. Under
2926 some allocation patterns, some large free blocks of memory will be
2927 locked between two used chunks, so they cannot be given back to
2930 The `pad' argument to malloc_trim represents the amount of free
2931 trailing space to leave untrimmed. If this argument is zero,
2932 only the minimum amount of memory to maintain internal data
2933 structures will be left (one page or less). Non-zero arguments
2934 can be supplied to maintain enough trailing space to service
2935 future expected allocations without having to re-obtain memory
2938 Malloc_trim returns 1 if it actually released any memory, else 0.
2943 int malloc_trim(size_t pad)
2945 int malloc_trim(pad) size_t pad;
2948 long top_size; /* Amount of top-most memory */
2949 long extra; /* Amount to release */
2950 char* current_brk; /* address returned by pre-check sbrk call */
2951 char* new_brk; /* address returned by negative sbrk call */
2953 unsigned long pagesz = malloc_getpagesize;
2955 top_size = chunksize(top);
2956 extra = ((top_size - pad - MINSIZE + (pagesz-1)) / pagesz - 1) * pagesz;
2958 if (extra < (long)pagesz) /* Not enough memory to release */
2963 /* Test to make sure no one else called sbrk */
2964 current_brk = (char*)(MORECORE (0));
2965 if (current_brk != (char*)(top) + top_size)
2966 return 0; /* Apparently we don't own memory; must fail */
2970 new_brk = (char*)(MORECORE (-extra));
2972 if (new_brk == (char*)(MORECORE_FAILURE)) /* sbrk failed? */
2974 /* Try to figure out what we have */
2975 current_brk = (char*)(MORECORE (0));
2976 top_size = current_brk - (char*)top;
2977 if (top_size >= (long)MINSIZE) /* if not, we are very very dead! */
2979 sbrked_mem = current_brk - sbrk_base;
2980 set_head(top, top_size | PREV_INUSE);
2988 /* Success. Adjust top accordingly. */
2989 set_head(top, (top_size - extra) | PREV_INUSE);
2990 sbrked_mem -= extra;
3003 This routine tells you how many bytes you can actually use in an
3004 allocated chunk, which may be more than you requested (although
3005 often not). You can use this many bytes without worrying about
3006 overwriting other allocated objects. Not a particularly great
3007 programming practice, but still sometimes useful.
3012 size_t malloc_usable_size(Void_t* mem)
3014 size_t malloc_usable_size(mem) Void_t* mem;
3023 if(!chunk_is_mmapped(p))
3025 if (!inuse(p)) return 0;
3026 check_inuse_chunk(p);
3027 return chunksize(p) - SIZE_SZ;
3029 return chunksize(p) - 2*SIZE_SZ;
3036 /* Utility to update current_mallinfo for malloc_stats and mallinfo() */
3038 static void malloc_update_mallinfo()
3047 INTERNAL_SIZE_T avail = chunksize(top);
3048 int navail = ((long)(avail) >= (long)MINSIZE)? 1 : 0;
3050 for (i = 1; i < NAV; ++i)
3053 for (p = last(b); p != b; p = p->bk)
3056 check_free_chunk(p);
3057 for (q = next_chunk(p);
3058 q < top && inuse(q) && (long)(chunksize(q)) >= (long)MINSIZE;
3060 check_inuse_chunk(q);
3062 avail += chunksize(p);
3067 current_mallinfo.ordblks = navail;
3068 current_mallinfo.uordblks = sbrked_mem - avail;
3069 current_mallinfo.fordblks = avail;
3070 current_mallinfo.hblks = n_mmaps;
3071 current_mallinfo.hblkhd = mmapped_mem;
3072 current_mallinfo.keepcost = chunksize(top);
3082 Prints on stderr the amount of space obtain from the system (both
3083 via sbrk and mmap), the maximum amount (which may be more than
3084 current if malloc_trim and/or munmap got called), the maximum
3085 number of simultaneous mmap regions used, and the current number
3086 of bytes allocated via malloc (or realloc, etc) but not yet
3087 freed. (Note that this is the number of bytes allocated, not the
3088 number requested. It will be larger than the number requested
3089 because of alignment and bookkeeping overhead.)
3095 malloc_update_mallinfo();
3096 fprintf(stderr, "max system bytes = %10u\n",
3097 (unsigned int)(max_total_mem));
3098 fprintf(stderr, "system bytes = %10u\n",
3099 (unsigned int)(sbrked_mem + mmapped_mem));
3100 fprintf(stderr, "in use bytes = %10u\n",
3101 (unsigned int)(current_mallinfo.uordblks + mmapped_mem));
3103 fprintf(stderr, "max mmap regions = %10u\n",
3104 (unsigned int)max_n_mmaps);
3109 mallinfo returns a copy of updated current mallinfo.
3112 struct mallinfo mALLINFo()
3114 malloc_update_mallinfo();
3115 return current_mallinfo;
3124 mallopt is the general SVID/XPG interface to tunable parameters.
3125 The format is to provide a (parameter-number, parameter-value) pair.
3126 mallopt then sets the corresponding parameter to the argument
3127 value if it can (i.e., so long as the value is meaningful),
3128 and returns 1 if successful else 0.
3130 See descriptions of tunable parameters above.
3135 int mALLOPt(int param_number, int value)
3137 int mALLOPt(param_number, value) int param_number; int value;
3140 switch(param_number)
3142 case M_TRIM_THRESHOLD:
3143 trim_threshold = value; return 1;
3145 top_pad = value; return 1;
3146 case M_MMAP_THRESHOLD:
3147 mmap_threshold = value; return 1;
3150 n_mmaps_max = value; return 1;
3152 if (value != 0) return 0; else n_mmaps_max = value; return 1;
3164 V2.6.6 Sun Dec 5 07:42:19 1999 Doug Lea (dl at gee)
3165 * return null for negative arguments
3166 * Added Several WIN32 cleanups from Martin C. Fong <mcfong@yahoo.com>
3167 * Add 'LACKS_SYS_PARAM_H' for those systems without 'sys/param.h'
3168 (e.g. WIN32 platforms)
3169 * Cleanup up header file inclusion for WIN32 platforms
3170 * Cleanup code to avoid Microsoft Visual C++ compiler complaints
3171 * Add 'USE_DL_PREFIX' to quickly allow co-existence with existing
3172 memory allocation routines
3173 * Set 'malloc_getpagesize' for WIN32 platforms (needs more work)
3174 * Use 'assert' rather than 'ASSERT' in WIN32 code to conform to
3175 usage of 'assert' in non-WIN32 code
3176 * Improve WIN32 'sbrk()' emulation's 'findRegion()' routine to
3178 * Always call 'fREe()' rather than 'free()'
3180 V2.6.5 Wed Jun 17 15:57:31 1998 Doug Lea (dl at gee)
3181 * Fixed ordering problem with boundary-stamping
3183 V2.6.3 Sun May 19 08:17:58 1996 Doug Lea (dl at gee)
3184 * Added pvalloc, as recommended by H.J. Liu
3185 * Added 64bit pointer support mainly from Wolfram Gloger
3186 * Added anonymously donated WIN32 sbrk emulation
3187 * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen
3188 * malloc_extend_top: fix mask error that caused wastage after
3190 * Add linux mremap support code from HJ Liu
3192 V2.6.2 Tue Dec 5 06:52:55 1995 Doug Lea (dl at gee)
3193 * Integrated most documentation with the code.
3194 * Add support for mmap, with help from
3195 Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
3196 * Use last_remainder in more cases.
3197 * Pack bins using idea from colin@nyx10.cs.du.edu
3198 * Use ordered bins instead of best-fit threshhold
3199 * Eliminate block-local decls to simplify tracing and debugging.
3200 * Support another case of realloc via move into top
3201 * Fix error occuring when initial sbrk_base not word-aligned.
3202 * Rely on page size for units instead of SBRK_UNIT to
3203 avoid surprises about sbrk alignment conventions.
3204 * Add mallinfo, mallopt. Thanks to Raymond Nijssen
3205 (raymond@es.ele.tue.nl) for the suggestion.
3206 * Add `pad' argument to malloc_trim and top_pad mallopt parameter.
3207 * More precautions for cases where other routines call sbrk,
3208 courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
3209 * Added macros etc., allowing use in linux libc from
3210 H.J. Lu (hjl@gnu.ai.mit.edu)
3211 * Inverted this history list
3213 V2.6.1 Sat Dec 2 14:10:57 1995 Doug Lea (dl at gee)
3214 * Re-tuned and fixed to behave more nicely with V2.6.0 changes.
3215 * Removed all preallocation code since under current scheme
3216 the work required to undo bad preallocations exceeds
3217 the work saved in good cases for most test programs.
3218 * No longer use return list or unconsolidated bins since
3219 no scheme using them consistently outperforms those that don't
3220 given above changes.
3221 * Use best fit for very large chunks to prevent some worst-cases.
3222 * Added some support for debugging
3224 V2.6.0 Sat Nov 4 07:05:23 1995 Doug Lea (dl at gee)
3225 * Removed footers when chunks are in use. Thanks to
3226 Paul Wilson (wilson@cs.texas.edu) for the suggestion.
3228 V2.5.4 Wed Nov 1 07:54:51 1995 Doug Lea (dl at gee)
3229 * Added malloc_trim, with help from Wolfram Gloger
3230 (wmglo@Dent.MED.Uni-Muenchen.DE).
3232 V2.5.3 Tue Apr 26 10:16:01 1994 Doug Lea (dl at g)
3234 V2.5.2 Tue Apr 5 16:20:40 1994 Doug Lea (dl at g)
3235 * realloc: try to expand in both directions
3236 * malloc: swap order of clean-bin strategy;
3237 * realloc: only conditionally expand backwards
3238 * Try not to scavenge used bins
3239 * Use bin counts as a guide to preallocation
3240 * Occasionally bin return list chunks in first scan
3241 * Add a few optimizations from colin@nyx10.cs.du.edu
3243 V2.5.1 Sat Aug 14 15:40:43 1993 Doug Lea (dl at g)
3244 * faster bin computation & slightly different binning
3245 * merged all consolidations to one part of malloc proper
3246 (eliminating old malloc_find_space & malloc_clean_bin)
3247 * Scan 2 returns chunks (not just 1)
3248 * Propagate failure in realloc if malloc returns 0
3249 * Add stuff to allow compilation on non-ANSI compilers
3250 from kpv@research.att.com
3252 V2.5 Sat Aug 7 07:41:59 1993 Doug Lea (dl at g.oswego.edu)
3253 * removed potential for odd address access in prev_chunk
3254 * removed dependency on getpagesize.h
3255 * misc cosmetics and a bit more internal documentation
3256 * anticosmetics: mangled names in macros to evade debugger strangeness
3257 * tested on sparc, hp-700, dec-mips, rs6000
3258 with gcc & native cc (hp, dec only) allowing
3259 Detlefs & Zorn comparison study (in SIGPLAN Notices.)
3261 Trial version Fri Aug 28 13:14:29 1992 Doug Lea (dl at g.oswego.edu)
3262 * Based loosely on libg++-1.2X malloc. (It retains some of the overall
3263 structure of old version, but most details differ.)