xref: /openbmc/u-boot/common/dlmalloc.c (revision b46694df)
1 #include <common.h>
2 
3 #if 0	/* Moved to malloc.h */
4 /* ---------- To make a malloc.h, start cutting here ------------ */
5 
6 /*
7   A version of malloc/free/realloc written by Doug Lea and released to the
8   public domain.  Send questions/comments/complaints/performance data
9   to dl@cs.oswego.edu
10 
11 * VERSION 2.6.6  Sun Mar  5 19:10:03 2000  Doug Lea  (dl at gee)
12 
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!
16 
17 * Why use this malloc?
18 
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
25 
26 * Synopsis of public routines
27 
28   (Much fuller descriptions are contained in the program documentation below.)
29 
30   malloc(size_t n);
31      Return a pointer to a newly allocated chunk of at least n bytes, or null
32      if no space is available.
33   free(Void_t* p);
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
45      two.
46   valloc(size_t n);
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.)
50   pvalloc(size_t n);
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
55      set to zero.
56   cfree(Void_t* p);
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.
65   malloc_stats();
66      Prints brief summary statistics.
67   mallinfo()
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.
72 
73 * Vital statistics:
74 
75   Alignment:                            8-byte
76        8 byte alignment is currently hardwired into the design.  This
77        seems to suffice for all current machines and C compilers.
78 
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.
83 
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.
86 
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.
90 
91   Minimum allocated size: 4-byte ptrs:  16 bytes    (including 4 overhead)
92 			  8-byte ptrs:  24/32 bytes (including, 4/8 overhead)
93 
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.
99 
100        Even a request for zero bytes (i.e., malloc(0)) returns a
101        pointer to something of the minimum allocatable size.
102 
103   Maximum allocated size: 4-byte size_t: 2^31 -  8 bytes
104 			  8-byte size_t: 2^63 - 16 bytes
105 
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.
113 
114   Maximum overhead wastage per allocated chunk: normally 15 bytes
115 
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
119        two exceptions:
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.
125 
126 * Limitations
127 
128     Here are some features that are NOT currently supported
129 
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.
134 
135 * Synopsis of compile-time options:
136 
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.
143 
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
148     paths.
149 
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
156      execution.
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
186      very small chunks.
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
192      affect anything.
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
205      holds for sbrk).
206   DEFAULT_TRIM_THRESHOLD
207   DEFAULT_TOP_PAD
208   DEFAULT_MMAP_THRESHOLD
209   DEFAULT_MMAP_MAX
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
214      programs/systems.
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.
219 
220 
221 */
222 
223 
224 
225 /* Preliminaries */
226 
227 #ifndef __STD_C
228 #ifdef __STDC__
229 #define __STD_C     1
230 #else
231 #if __cplusplus
232 #define __STD_C     1
233 #else
234 #define __STD_C     0
235 #endif /*__cplusplus*/
236 #endif /*__STDC__*/
237 #endif /*__STD_C*/
238 
239 #ifndef Void_t
240 #if (__STD_C || defined(WIN32))
241 #define Void_t      void
242 #else
243 #define Void_t      char
244 #endif
245 #endif /*Void_t*/
246 
247 #if __STD_C
248 #include <stddef.h>   /* for size_t */
249 #else
250 #include <sys/types.h>
251 #endif
252 
253 #ifdef __cplusplus
254 extern "C" {
255 #endif
256 
257 #include <stdio.h>    /* needed for malloc_stats */
258 
259 
260 /*
261   Compile-time options
262 */
263 
264 
265 /*
266     Debugging:
267 
268     Because freed chunks may be overwritten with link fields, this
269     malloc will often die when freed memory is overwritten by user
270     programs.  This can be very effective (albeit in an annoying way)
271     in helping track down dangling pointers.
272 
273     If you compile with -DDEBUG, a number of assertion checks are
274     enabled that will catch more memory errors. You probably won't be
275     able to make much sense of the actual assertion errors, but they
276     should help you locate incorrectly overwritten memory.  The
277     checking is fairly extensive, and will slow down execution
278     noticeably. Calling malloc_stats or mallinfo with DEBUG set will
279     attempt to check every non-mmapped allocated and free chunk in the
280     course of computing the summmaries. (By nature, mmapped regions
281     cannot be checked very much automatically.)
282 
283     Setting DEBUG may also be helpful if you are trying to modify
284     this code. The assertions in the check routines spell out in more
285     detail the assumptions and invariants underlying the algorithms.
286 
287 */
288 
289 /*
290   INTERNAL_SIZE_T is the word-size used for internal bookkeeping
291   of chunk sizes. On a 64-bit machine, you can reduce malloc
292   overhead by defining INTERNAL_SIZE_T to be a 32 bit `unsigned int'
293   at the expense of not being able to handle requests greater than
294   2^31. This limitation is hardly ever a concern; you are encouraged
295   to set this. However, the default version is the same as size_t.
296 */
297 
298 #ifndef INTERNAL_SIZE_T
299 #define INTERNAL_SIZE_T size_t
300 #endif
301 
302 /*
303   REALLOC_ZERO_BYTES_FREES should be set if a call to
304   realloc with zero bytes should be the same as a call to free.
305   Some people think it should. Otherwise, since this malloc
306   returns a unique pointer for malloc(0), so does realloc(p, 0).
307 */
308 
309 
310 /*   #define REALLOC_ZERO_BYTES_FREES */
311 
312 
313 /*
314   WIN32 causes an emulation of sbrk to be compiled in
315   mmap-based options are not currently supported in WIN32.
316 */
317 
318 /* #define WIN32 */
319 #ifdef WIN32
320 #define MORECORE wsbrk
321 #define HAVE_MMAP 0
322 
323 #define LACKS_UNISTD_H
324 #define LACKS_SYS_PARAM_H
325 
326 /*
327   Include 'windows.h' to get the necessary declarations for the
328   Microsoft Visual C++ data structures and routines used in the 'sbrk'
329   emulation.
330 
331   Define WIN32_LEAN_AND_MEAN so that only the essential Microsoft
332   Visual C++ header files are included.
333 */
334 #define WIN32_LEAN_AND_MEAN
335 #include <windows.h>
336 #endif
337 
338 
339 /*
340   HAVE_MEMCPY should be defined if you are not otherwise using
341   ANSI STD C, but still have memcpy and memset in your C library
342   and want to use them in calloc and realloc. Otherwise simple
343   macro versions are defined here.
344 
345   USE_MEMCPY should be defined as 1 if you actually want to
346   have memset and memcpy called. People report that the macro
347   versions are often enough faster than libc versions on many
348   systems that it is better to use them.
349 
350 */
351 
352 #define HAVE_MEMCPY
353 
354 #ifndef USE_MEMCPY
355 #ifdef HAVE_MEMCPY
356 #define USE_MEMCPY 1
357 #else
358 #define USE_MEMCPY 0
359 #endif
360 #endif
361 
362 #if (__STD_C || defined(HAVE_MEMCPY))
363 
364 #if __STD_C
365 void* memset(void*, int, size_t);
366 void* memcpy(void*, const void*, size_t);
367 #else
368 #ifdef WIN32
369 /* On Win32 platforms, 'memset()' and 'memcpy()' are already declared in */
370 /* 'windows.h' */
371 #else
372 Void_t* memset();
373 Void_t* memcpy();
374 #endif
375 #endif
376 #endif
377 
378 #if USE_MEMCPY
379 
380 /* The following macros are only invoked with (2n+1)-multiples of
381    INTERNAL_SIZE_T units, with a positive integer n. This is exploited
382    for fast inline execution when n is small. */
383 
384 #define MALLOC_ZERO(charp, nbytes)                                            \
385 do {                                                                          \
386   INTERNAL_SIZE_T mzsz = (nbytes);                                            \
387   if(mzsz <= 9*sizeof(mzsz)) {                                                \
388     INTERNAL_SIZE_T* mz = (INTERNAL_SIZE_T*) (charp);                         \
389     if(mzsz >= 5*sizeof(mzsz)) {     *mz++ = 0;                               \
390 				     *mz++ = 0;                               \
391       if(mzsz >= 7*sizeof(mzsz)) {   *mz++ = 0;                               \
392 				     *mz++ = 0;                               \
393 	if(mzsz >= 9*sizeof(mzsz)) { *mz++ = 0;                               \
394 				     *mz++ = 0; }}}                           \
395 				     *mz++ = 0;                               \
396 				     *mz++ = 0;                               \
397 				     *mz   = 0;                               \
398   } else memset((charp), 0, mzsz);                                            \
399 } while(0)
400 
401 #define MALLOC_COPY(dest,src,nbytes)                                          \
402 do {                                                                          \
403   INTERNAL_SIZE_T mcsz = (nbytes);                                            \
404   if(mcsz <= 9*sizeof(mcsz)) {                                                \
405     INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) (src);                        \
406     INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) (dest);                       \
407     if(mcsz >= 5*sizeof(mcsz)) {     *mcdst++ = *mcsrc++;                     \
408 				     *mcdst++ = *mcsrc++;                     \
409       if(mcsz >= 7*sizeof(mcsz)) {   *mcdst++ = *mcsrc++;                     \
410 				     *mcdst++ = *mcsrc++;                     \
411 	if(mcsz >= 9*sizeof(mcsz)) { *mcdst++ = *mcsrc++;                     \
412 				     *mcdst++ = *mcsrc++; }}}                 \
413 				     *mcdst++ = *mcsrc++;                     \
414 				     *mcdst++ = *mcsrc++;                     \
415 				     *mcdst   = *mcsrc  ;                     \
416   } else memcpy(dest, src, mcsz);                                             \
417 } while(0)
418 
419 #else /* !USE_MEMCPY */
420 
421 /* Use Duff's device for good zeroing/copying performance. */
422 
423 #define MALLOC_ZERO(charp, nbytes)                                            \
424 do {                                                                          \
425   INTERNAL_SIZE_T* mzp = (INTERNAL_SIZE_T*)(charp);                           \
426   long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn;                         \
427   if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; }             \
428   switch (mctmp) {                                                            \
429     case 0: for(;;) { *mzp++ = 0;                                             \
430     case 7:           *mzp++ = 0;                                             \
431     case 6:           *mzp++ = 0;                                             \
432     case 5:           *mzp++ = 0;                                             \
433     case 4:           *mzp++ = 0;                                             \
434     case 3:           *mzp++ = 0;                                             \
435     case 2:           *mzp++ = 0;                                             \
436     case 1:           *mzp++ = 0; if(mcn <= 0) break; mcn--; }                \
437   }                                                                           \
438 } while(0)
439 
440 #define MALLOC_COPY(dest,src,nbytes)                                          \
441 do {                                                                          \
442   INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) src;                            \
443   INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) dest;                           \
444   long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn;                         \
445   if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; }             \
446   switch (mctmp) {                                                            \
447     case 0: for(;;) { *mcdst++ = *mcsrc++;                                    \
448     case 7:           *mcdst++ = *mcsrc++;                                    \
449     case 6:           *mcdst++ = *mcsrc++;                                    \
450     case 5:           *mcdst++ = *mcsrc++;                                    \
451     case 4:           *mcdst++ = *mcsrc++;                                    \
452     case 3:           *mcdst++ = *mcsrc++;                                    \
453     case 2:           *mcdst++ = *mcsrc++;                                    \
454     case 1:           *mcdst++ = *mcsrc++; if(mcn <= 0) break; mcn--; }       \
455   }                                                                           \
456 } while(0)
457 
458 #endif
459 
460 
461 /*
462   Define HAVE_MMAP to optionally make malloc() use mmap() to
463   allocate very large blocks.  These will be returned to the
464   operating system immediately after a free().
465 */
466 
467 #ifndef HAVE_MMAP
468 #define HAVE_MMAP 1
469 #endif
470 
471 /*
472   Define HAVE_MREMAP to make realloc() use mremap() to re-allocate
473   large blocks.  This is currently only possible on Linux with
474   kernel versions newer than 1.3.77.
475 */
476 
477 #ifndef HAVE_MREMAP
478 #ifdef INTERNAL_LINUX_C_LIB
479 #define HAVE_MREMAP 1
480 #else
481 #define HAVE_MREMAP 0
482 #endif
483 #endif
484 
485 #if HAVE_MMAP
486 
487 #include <unistd.h>
488 #include <fcntl.h>
489 #include <sys/mman.h>
490 
491 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
492 #define MAP_ANONYMOUS MAP_ANON
493 #endif
494 
495 #endif /* HAVE_MMAP */
496 
497 /*
498   Access to system page size. To the extent possible, this malloc
499   manages memory from the system in page-size units.
500 
501   The following mechanics for getpagesize were adapted from
502   bsd/gnu getpagesize.h
503 */
504 
505 #ifndef LACKS_UNISTD_H
506 #  include <unistd.h>
507 #endif
508 
509 #ifndef malloc_getpagesize
510 #  ifdef _SC_PAGESIZE         /* some SVR4 systems omit an underscore */
511 #    ifndef _SC_PAGE_SIZE
512 #      define _SC_PAGE_SIZE _SC_PAGESIZE
513 #    endif
514 #  endif
515 #  ifdef _SC_PAGE_SIZE
516 #    define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
517 #  else
518 #    if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
519        extern size_t getpagesize();
520 #      define malloc_getpagesize getpagesize()
521 #    else
522 #      ifdef WIN32
523 #        define malloc_getpagesize (4096) /* TBD: Use 'GetSystemInfo' instead */
524 #      else
525 #        ifndef LACKS_SYS_PARAM_H
526 #          include <sys/param.h>
527 #        endif
528 #        ifdef EXEC_PAGESIZE
529 #          define malloc_getpagesize EXEC_PAGESIZE
530 #        else
531 #          ifdef NBPG
532 #            ifndef CLSIZE
533 #              define malloc_getpagesize NBPG
534 #            else
535 #              define malloc_getpagesize (NBPG * CLSIZE)
536 #            endif
537 #          else
538 #            ifdef NBPC
539 #              define malloc_getpagesize NBPC
540 #            else
541 #              ifdef PAGESIZE
542 #                define malloc_getpagesize PAGESIZE
543 #              else
544 #                define malloc_getpagesize (4096) /* just guess */
545 #              endif
546 #            endif
547 #          endif
548 #        endif
549 #      endif
550 #    endif
551 #  endif
552 #endif
553 
554 
555 /*
556 
557   This version of malloc supports the standard SVID/XPG mallinfo
558   routine that returns a struct containing the same kind of
559   information you can get from malloc_stats. It should work on
560   any SVID/XPG compliant system that has a /usr/include/malloc.h
561   defining struct mallinfo. (If you'd like to install such a thing
562   yourself, cut out the preliminary declarations as described above
563   and below and save them in a malloc.h file. But there's no
564   compelling reason to bother to do this.)
565 
566   The main declaration needed is the mallinfo struct that is returned
567   (by-copy) by mallinfo().  The SVID/XPG malloinfo struct contains a
568   bunch of fields, most of which are not even meaningful in this
569   version of malloc. Some of these fields are are instead filled by
570   mallinfo() with other numbers that might possibly be of interest.
571 
572   HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
573   /usr/include/malloc.h file that includes a declaration of struct
574   mallinfo.  If so, it is included; else an SVID2/XPG2 compliant
575   version is declared below.  These must be precisely the same for
576   mallinfo() to work.
577 
578 */
579 
580 /* #define HAVE_USR_INCLUDE_MALLOC_H */
581 
582 #if HAVE_USR_INCLUDE_MALLOC_H
583 #include "/usr/include/malloc.h"
584 #else
585 
586 /* SVID2/XPG mallinfo structure */
587 
588 struct mallinfo {
589   int arena;    /* total space allocated from system */
590   int ordblks;  /* number of non-inuse chunks */
591   int smblks;   /* unused -- always zero */
592   int hblks;    /* number of mmapped regions */
593   int hblkhd;   /* total space in mmapped regions */
594   int usmblks;  /* unused -- always zero */
595   int fsmblks;  /* unused -- always zero */
596   int uordblks; /* total allocated space */
597   int fordblks; /* total non-inuse space */
598   int keepcost; /* top-most, releasable (via malloc_trim) space */
599 };
600 
601 /* SVID2/XPG mallopt options */
602 
603 #define M_MXFAST  1    /* UNUSED in this malloc */
604 #define M_NLBLKS  2    /* UNUSED in this malloc */
605 #define M_GRAIN   3    /* UNUSED in this malloc */
606 #define M_KEEP    4    /* UNUSED in this malloc */
607 
608 #endif
609 
610 /* mallopt options that actually do something */
611 
612 #define M_TRIM_THRESHOLD    -1
613 #define M_TOP_PAD           -2
614 #define M_MMAP_THRESHOLD    -3
615 #define M_MMAP_MAX          -4
616 
617 
618 #ifndef DEFAULT_TRIM_THRESHOLD
619 #define DEFAULT_TRIM_THRESHOLD (128 * 1024)
620 #endif
621 
622 /*
623     M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
624       to keep before releasing via malloc_trim in free().
625 
626       Automatic trimming is mainly useful in long-lived programs.
627       Because trimming via sbrk can be slow on some systems, and can
628       sometimes be wasteful (in cases where programs immediately
629       afterward allocate more large chunks) the value should be high
630       enough so that your overall system performance would improve by
631       releasing.
632 
633       The trim threshold and the mmap control parameters (see below)
634       can be traded off with one another. Trimming and mmapping are
635       two different ways of releasing unused memory back to the
636       system. Between these two, it is often possible to keep
637       system-level demands of a long-lived program down to a bare
638       minimum. For example, in one test suite of sessions measuring
639       the XF86 X server on Linux, using a trim threshold of 128K and a
640       mmap threshold of 192K led to near-minimal long term resource
641       consumption.
642 
643       If you are using this malloc in a long-lived program, it should
644       pay to experiment with these values.  As a rough guide, you
645       might set to a value close to the average size of a process
646       (program) running on your system.  Releasing this much memory
647       would allow such a process to run in memory.  Generally, it's
648       worth it to tune for trimming rather tham memory mapping when a
649       program undergoes phases where several large chunks are
650       allocated and released in ways that can reuse each other's
651       storage, perhaps mixed with phases where there are no such
652       chunks at all.  And in well-behaved long-lived programs,
653       controlling release of large blocks via trimming versus mapping
654       is usually faster.
655 
656       However, in most programs, these parameters serve mainly as
657       protection against the system-level effects of carrying around
658       massive amounts of unneeded memory. Since frequent calls to
659       sbrk, mmap, and munmap otherwise degrade performance, the default
660       parameters are set to relatively high values that serve only as
661       safeguards.
662 
663       The default trim value is high enough to cause trimming only in
664       fairly extreme (by current memory consumption standards) cases.
665       It must be greater than page size to have any useful effect.  To
666       disable trimming completely, you can set to (unsigned long)(-1);
667 
668 
669 */
670 
671 
672 #ifndef DEFAULT_TOP_PAD
673 #define DEFAULT_TOP_PAD        (0)
674 #endif
675 
676 /*
677     M_TOP_PAD is the amount of extra `padding' space to allocate or
678       retain whenever sbrk is called. It is used in two ways internally:
679 
680       * When sbrk is called to extend the top of the arena to satisfy
681 	a new malloc request, this much padding is added to the sbrk
682 	request.
683 
684       * When malloc_trim is called automatically from free(),
685 	it is used as the `pad' argument.
686 
687       In both cases, the actual amount of padding is rounded
688       so that the end of the arena is always a system page boundary.
689 
690       The main reason for using padding is to avoid calling sbrk so
691       often. Having even a small pad greatly reduces the likelihood
692       that nearly every malloc request during program start-up (or
693       after trimming) will invoke sbrk, which needlessly wastes
694       time.
695 
696       Automatic rounding-up to page-size units is normally sufficient
697       to avoid measurable overhead, so the default is 0.  However, in
698       systems where sbrk is relatively slow, it can pay to increase
699       this value, at the expense of carrying around more memory than
700       the program needs.
701 
702 */
703 
704 
705 #ifndef DEFAULT_MMAP_THRESHOLD
706 #define DEFAULT_MMAP_THRESHOLD (128 * 1024)
707 #endif
708 
709 /*
710 
711     M_MMAP_THRESHOLD is the request size threshold for using mmap()
712       to service a request. Requests of at least this size that cannot
713       be allocated using already-existing space will be serviced via mmap.
714       (If enough normal freed space already exists it is used instead.)
715 
716       Using mmap segregates relatively large chunks of memory so that
717       they can be individually obtained and released from the host
718       system. A request serviced through mmap is never reused by any
719       other request (at least not directly; the system may just so
720       happen to remap successive requests to the same locations).
721 
722       Segregating space in this way has the benefit that mmapped space
723       can ALWAYS be individually released back to the system, which
724       helps keep the system level memory demands of a long-lived
725       program low. Mapped memory can never become `locked' between
726       other chunks, as can happen with normally allocated chunks, which
727       menas that even trimming via malloc_trim would not release them.
728 
729       However, it has the disadvantages that:
730 
731 	 1. The space cannot be reclaimed, consolidated, and then
732 	    used to service later requests, as happens with normal chunks.
733 	 2. It can lead to more wastage because of mmap page alignment
734 	    requirements
735 	 3. It causes malloc performance to be more dependent on host
736 	    system memory management support routines which may vary in
737 	    implementation quality and may impose arbitrary
738 	    limitations. Generally, servicing a request via normal
739 	    malloc steps is faster than going through a system's mmap.
740 
741       All together, these considerations should lead you to use mmap
742       only for relatively large requests.
743 
744 
745 */
746 
747 
748 #ifndef DEFAULT_MMAP_MAX
749 #if HAVE_MMAP
750 #define DEFAULT_MMAP_MAX       (64)
751 #else
752 #define DEFAULT_MMAP_MAX       (0)
753 #endif
754 #endif
755 
756 /*
757     M_MMAP_MAX is the maximum number of requests to simultaneously
758       service using mmap. This parameter exists because:
759 
760 	 1. Some systems have a limited number of internal tables for
761 	    use by mmap.
762 	 2. In most systems, overreliance on mmap can degrade overall
763 	    performance.
764 	 3. If a program allocates many large regions, it is probably
765 	    better off using normal sbrk-based allocation routines that
766 	    can reclaim and reallocate normal heap memory. Using a
767 	    small value allows transition into this mode after the
768 	    first few allocations.
769 
770       Setting to 0 disables all use of mmap.  If HAVE_MMAP is not set,
771       the default value is 0, and attempts to set it to non-zero values
772       in mallopt will fail.
773 */
774 
775 
776 /*
777     USE_DL_PREFIX will prefix all public routines with the string 'dl'.
778       Useful to quickly avoid procedure declaration conflicts and linker
779       symbol conflicts with existing memory allocation routines.
780 
781 */
782 
783 /* #define USE_DL_PREFIX */
784 
785 
786 /*
787 
788   Special defines for linux libc
789 
790   Except when compiled using these special defines for Linux libc
791   using weak aliases, this malloc is NOT designed to work in
792   multithreaded applications.  No semaphores or other concurrency
793   control are provided to ensure that multiple malloc or free calls
794   don't run at the same time, which could be disasterous. A single
795   semaphore could be used across malloc, realloc, and free (which is
796   essentially the effect of the linux weak alias approach). It would
797   be hard to obtain finer granularity.
798 
799 */
800 
801 
802 #ifdef INTERNAL_LINUX_C_LIB
803 
804 #if __STD_C
805 
806 Void_t * __default_morecore_init (ptrdiff_t);
807 Void_t *(*__morecore)(ptrdiff_t) = __default_morecore_init;
808 
809 #else
810 
811 Void_t * __default_morecore_init ();
812 Void_t *(*__morecore)() = __default_morecore_init;
813 
814 #endif
815 
816 #define MORECORE (*__morecore)
817 #define MORECORE_FAILURE 0
818 #define MORECORE_CLEARS 1
819 
820 #else /* INTERNAL_LINUX_C_LIB */
821 
822 #if __STD_C
823 extern Void_t*     sbrk(ptrdiff_t);
824 #else
825 extern Void_t*     sbrk();
826 #endif
827 
828 #ifndef MORECORE
829 #define MORECORE sbrk
830 #endif
831 
832 #ifndef MORECORE_FAILURE
833 #define MORECORE_FAILURE -1
834 #endif
835 
836 #ifndef MORECORE_CLEARS
837 #define MORECORE_CLEARS 1
838 #endif
839 
840 #endif /* INTERNAL_LINUX_C_LIB */
841 
842 #if defined(INTERNAL_LINUX_C_LIB) && defined(__ELF__)
843 
844 #define cALLOc		__libc_calloc
845 #define fREe		__libc_free
846 #define mALLOc		__libc_malloc
847 #define mEMALIGn	__libc_memalign
848 #define rEALLOc		__libc_realloc
849 #define vALLOc		__libc_valloc
850 #define pvALLOc		__libc_pvalloc
851 #define mALLINFo	__libc_mallinfo
852 #define mALLOPt		__libc_mallopt
853 
854 #pragma weak calloc = __libc_calloc
855 #pragma weak free = __libc_free
856 #pragma weak cfree = __libc_free
857 #pragma weak malloc = __libc_malloc
858 #pragma weak memalign = __libc_memalign
859 #pragma weak realloc = __libc_realloc
860 #pragma weak valloc = __libc_valloc
861 #pragma weak pvalloc = __libc_pvalloc
862 #pragma weak mallinfo = __libc_mallinfo
863 #pragma weak mallopt = __libc_mallopt
864 
865 #else
866 
867 #ifdef USE_DL_PREFIX
868 #define cALLOc		dlcalloc
869 #define fREe		dlfree
870 #define mALLOc		dlmalloc
871 #define mEMALIGn	dlmemalign
872 #define rEALLOc		dlrealloc
873 #define vALLOc		dlvalloc
874 #define pvALLOc		dlpvalloc
875 #define mALLINFo	dlmallinfo
876 #define mALLOPt		dlmallopt
877 #else /* USE_DL_PREFIX */
878 #define cALLOc		calloc
879 #define fREe		free
880 #define mALLOc		malloc
881 #define mEMALIGn	memalign
882 #define rEALLOc		realloc
883 #define vALLOc		valloc
884 #define pvALLOc		pvalloc
885 #define mALLINFo	mallinfo
886 #define mALLOPt		mallopt
887 #endif /* USE_DL_PREFIX */
888 
889 #endif
890 
891 /* Public routines */
892 
893 #if __STD_C
894 
895 Void_t* mALLOc(size_t);
896 void    fREe(Void_t*);
897 Void_t* rEALLOc(Void_t*, size_t);
898 Void_t* mEMALIGn(size_t, size_t);
899 Void_t* vALLOc(size_t);
900 Void_t* pvALLOc(size_t);
901 Void_t* cALLOc(size_t, size_t);
902 void    cfree(Void_t*);
903 int     malloc_trim(size_t);
904 size_t  malloc_usable_size(Void_t*);
905 void    malloc_stats();
906 int     mALLOPt(int, int);
907 struct mallinfo mALLINFo(void);
908 #else
909 Void_t* mALLOc();
910 void    fREe();
911 Void_t* rEALLOc();
912 Void_t* mEMALIGn();
913 Void_t* vALLOc();
914 Void_t* pvALLOc();
915 Void_t* cALLOc();
916 void    cfree();
917 int     malloc_trim();
918 size_t  malloc_usable_size();
919 void    malloc_stats();
920 int     mALLOPt();
921 struct mallinfo mALLINFo();
922 #endif
923 
924 
925 #ifdef __cplusplus
926 };  /* end of extern "C" */
927 #endif
928 
929 /* ---------- To make a malloc.h, end cutting here ------------ */
930 #endif	/* 0 */			/* Moved to malloc.h */
931 
932 #include <malloc.h>
933 #ifdef DEBUG
934 #if __STD_C
935 static void malloc_update_mallinfo (void);
936 void malloc_stats (void);
937 #else
938 static void malloc_update_mallinfo ();
939 void malloc_stats();
940 #endif
941 #endif	/* DEBUG */
942 
943 DECLARE_GLOBAL_DATA_PTR;
944 
945 /*
946   Emulation of sbrk for WIN32
947   All code within the ifdef WIN32 is untested by me.
948 
949   Thanks to Martin Fong and others for supplying this.
950 */
951 
952 
953 #ifdef WIN32
954 
955 #define AlignPage(add) (((add) + (malloc_getpagesize-1)) & \
956 ~(malloc_getpagesize-1))
957 #define AlignPage64K(add) (((add) + (0x10000 - 1)) & ~(0x10000 - 1))
958 
959 /* resrve 64MB to insure large contiguous space */
960 #define RESERVED_SIZE (1024*1024*64)
961 #define NEXT_SIZE (2048*1024)
962 #define TOP_MEMORY ((unsigned long)2*1024*1024*1024)
963 
964 struct GmListElement;
965 typedef struct GmListElement GmListElement;
966 
967 struct GmListElement
968 {
969 	GmListElement* next;
970 	void* base;
971 };
972 
973 static GmListElement* head = 0;
974 static unsigned int gNextAddress = 0;
975 static unsigned int gAddressBase = 0;
976 static unsigned int gAllocatedSize = 0;
977 
978 static
979 GmListElement* makeGmListElement (void* bas)
980 {
981 	GmListElement* this;
982 	this = (GmListElement*)(void*)LocalAlloc (0, sizeof (GmListElement));
983 	assert (this);
984 	if (this)
985 	{
986 		this->base = bas;
987 		this->next = head;
988 		head = this;
989 	}
990 	return this;
991 }
992 
993 void gcleanup ()
994 {
995 	BOOL rval;
996 	assert ( (head == NULL) || (head->base == (void*)gAddressBase));
997 	if (gAddressBase && (gNextAddress - gAddressBase))
998 	{
999 		rval = VirtualFree ((void*)gAddressBase,
1000 							gNextAddress - gAddressBase,
1001 							MEM_DECOMMIT);
1002 	assert (rval);
1003 	}
1004 	while (head)
1005 	{
1006 		GmListElement* next = head->next;
1007 		rval = VirtualFree (head->base, 0, MEM_RELEASE);
1008 		assert (rval);
1009 		LocalFree (head);
1010 		head = next;
1011 	}
1012 }
1013 
1014 static
1015 void* findRegion (void* start_address, unsigned long size)
1016 {
1017 	MEMORY_BASIC_INFORMATION info;
1018 	if (size >= TOP_MEMORY) return NULL;
1019 
1020 	while ((unsigned long)start_address + size < TOP_MEMORY)
1021 	{
1022 		VirtualQuery (start_address, &info, sizeof (info));
1023 		if ((info.State == MEM_FREE) && (info.RegionSize >= size))
1024 			return start_address;
1025 		else
1026 		{
1027 			/* Requested region is not available so see if the */
1028 			/* next region is available.  Set 'start_address' */
1029 			/* to the next region and call 'VirtualQuery()' */
1030 			/* again. */
1031 
1032 			start_address = (char*)info.BaseAddress + info.RegionSize;
1033 
1034 			/* Make sure we start looking for the next region */
1035 			/* on the *next* 64K boundary.  Otherwise, even if */
1036 			/* the new region is free according to */
1037 			/* 'VirtualQuery()', the subsequent call to */
1038 			/* 'VirtualAlloc()' (which follows the call to */
1039 			/* this routine in 'wsbrk()') will round *down* */
1040 			/* the requested address to a 64K boundary which */
1041 			/* we already know is an address in the */
1042 			/* unavailable region.  Thus, the subsequent call */
1043 			/* to 'VirtualAlloc()' will fail and bring us back */
1044 			/* here, causing us to go into an infinite loop. */
1045 
1046 			start_address =
1047 				(void *) AlignPage64K((unsigned long) start_address);
1048 		}
1049 	}
1050 	return NULL;
1051 
1052 }
1053 
1054 
1055 void* wsbrk (long size)
1056 {
1057 	void* tmp;
1058 	if (size > 0)
1059 	{
1060 		if (gAddressBase == 0)
1061 		{
1062 			gAllocatedSize = max (RESERVED_SIZE, AlignPage (size));
1063 			gNextAddress = gAddressBase =
1064 				(unsigned int)VirtualAlloc (NULL, gAllocatedSize,
1065 											MEM_RESERVE, PAGE_NOACCESS);
1066 		} else if (AlignPage (gNextAddress + size) > (gAddressBase +
1067 gAllocatedSize))
1068 		{
1069 			long new_size = max (NEXT_SIZE, AlignPage (size));
1070 			void* new_address = (void*)(gAddressBase+gAllocatedSize);
1071 			do
1072 			{
1073 				new_address = findRegion (new_address, new_size);
1074 
1075 				if (new_address == 0)
1076 					return (void*)-1;
1077 
1078 				gAddressBase = gNextAddress =
1079 					(unsigned int)VirtualAlloc (new_address, new_size,
1080 												MEM_RESERVE, PAGE_NOACCESS);
1081 				/* repeat in case of race condition */
1082 				/* The region that we found has been snagged */
1083 				/* by another thread */
1084 			}
1085 			while (gAddressBase == 0);
1086 
1087 			assert (new_address == (void*)gAddressBase);
1088 
1089 			gAllocatedSize = new_size;
1090 
1091 			if (!makeGmListElement ((void*)gAddressBase))
1092 				return (void*)-1;
1093 		}
1094 		if ((size + gNextAddress) > AlignPage (gNextAddress))
1095 		{
1096 			void* res;
1097 			res = VirtualAlloc ((void*)AlignPage (gNextAddress),
1098 								(size + gNextAddress -
1099 								 AlignPage (gNextAddress)),
1100 								MEM_COMMIT, PAGE_READWRITE);
1101 			if (res == 0)
1102 				return (void*)-1;
1103 		}
1104 		tmp = (void*)gNextAddress;
1105 		gNextAddress = (unsigned int)tmp + size;
1106 		return tmp;
1107 	}
1108 	else if (size < 0)
1109 	{
1110 		unsigned int alignedGoal = AlignPage (gNextAddress + size);
1111 		/* Trim by releasing the virtual memory */
1112 		if (alignedGoal >= gAddressBase)
1113 		{
1114 			VirtualFree ((void*)alignedGoal, gNextAddress - alignedGoal,
1115 						 MEM_DECOMMIT);
1116 			gNextAddress = gNextAddress + size;
1117 			return (void*)gNextAddress;
1118 		}
1119 		else
1120 		{
1121 			VirtualFree ((void*)gAddressBase, gNextAddress - gAddressBase,
1122 						 MEM_DECOMMIT);
1123 			gNextAddress = gAddressBase;
1124 			return (void*)-1;
1125 		}
1126 	}
1127 	else
1128 	{
1129 		return (void*)gNextAddress;
1130 	}
1131 }
1132 
1133 #endif
1134 
1135 
1136 
1137 /*
1138   Type declarations
1139 */
1140 
1141 
1142 struct malloc_chunk
1143 {
1144   INTERNAL_SIZE_T prev_size; /* Size of previous chunk (if free). */
1145   INTERNAL_SIZE_T size;      /* Size in bytes, including overhead. */
1146   struct malloc_chunk* fd;   /* double links -- used only if free. */
1147   struct malloc_chunk* bk;
1148 } __attribute__((__may_alias__)) ;
1149 
1150 typedef struct malloc_chunk* mchunkptr;
1151 
1152 /*
1153 
1154    malloc_chunk details:
1155 
1156     (The following includes lightly edited explanations by Colin Plumb.)
1157 
1158     Chunks of memory are maintained using a `boundary tag' method as
1159     described in e.g., Knuth or Standish.  (See the paper by Paul
1160     Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
1161     survey of such techniques.)  Sizes of free chunks are stored both
1162     in the front of each chunk and at the end.  This makes
1163     consolidating fragmented chunks into bigger chunks very fast.  The
1164     size fields also hold bits representing whether chunks are free or
1165     in use.
1166 
1167     An allocated chunk looks like this:
1168 
1169 
1170     chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1171 	    |             Size of previous chunk, if allocated            | |
1172 	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1173 	    |             Size of chunk, in bytes                         |P|
1174       mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1175 	    |             User data starts here...                          .
1176 	    .                                                               .
1177 	    .             (malloc_usable_space() bytes)                     .
1178 	    .                                                               |
1179 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1180 	    |             Size of chunk                                     |
1181 	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1182 
1183 
1184     Where "chunk" is the front of the chunk for the purpose of most of
1185     the malloc code, but "mem" is the pointer that is returned to the
1186     user.  "Nextchunk" is the beginning of the next contiguous chunk.
1187 
1188     Chunks always begin on even word boundries, so the mem portion
1189     (which is returned to the user) is also on an even word boundary, and
1190     thus double-word aligned.
1191 
1192     Free chunks are stored in circular doubly-linked lists, and look like this:
1193 
1194     chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1195 	    |             Size of previous chunk                            |
1196 	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1197     `head:' |             Size of chunk, in bytes                         |P|
1198       mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1199 	    |             Forward pointer to next chunk in list             |
1200 	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1201 	    |             Back pointer to previous chunk in list            |
1202 	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1203 	    |             Unused space (may be 0 bytes long)                .
1204 	    .                                                               .
1205 	    .                                                               |
1206 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1207     `foot:' |             Size of chunk, in bytes                           |
1208 	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1209 
1210     The P (PREV_INUSE) bit, stored in the unused low-order bit of the
1211     chunk size (which is always a multiple of two words), is an in-use
1212     bit for the *previous* chunk.  If that bit is *clear*, then the
1213     word before the current chunk size contains the previous chunk
1214     size, and can be used to find the front of the previous chunk.
1215     (The very first chunk allocated always has this bit set,
1216     preventing access to non-existent (or non-owned) memory.)
1217 
1218     Note that the `foot' of the current chunk is actually represented
1219     as the prev_size of the NEXT chunk. (This makes it easier to
1220     deal with alignments etc).
1221 
1222     The two exceptions to all this are
1223 
1224      1. The special chunk `top', which doesn't bother using the
1225 	trailing size field since there is no
1226 	next contiguous chunk that would have to index off it. (After
1227 	initialization, `top' is forced to always exist.  If it would
1228 	become less than MINSIZE bytes long, it is replenished via
1229 	malloc_extend_top.)
1230 
1231      2. Chunks allocated via mmap, which have the second-lowest-order
1232 	bit (IS_MMAPPED) set in their size fields.  Because they are
1233 	never merged or traversed from any other chunk, they have no
1234 	foot size or inuse information.
1235 
1236     Available chunks are kept in any of several places (all declared below):
1237 
1238     * `av': An array of chunks serving as bin headers for consolidated
1239        chunks. Each bin is doubly linked.  The bins are approximately
1240        proportionally (log) spaced.  There are a lot of these bins
1241        (128). This may look excessive, but works very well in
1242        practice.  All procedures maintain the invariant that no
1243        consolidated chunk physically borders another one. Chunks in
1244        bins are kept in size order, with ties going to the
1245        approximately least recently used chunk.
1246 
1247        The chunks in each bin are maintained in decreasing sorted order by
1248        size.  This is irrelevant for the small bins, which all contain
1249        the same-sized chunks, but facilitates best-fit allocation for
1250        larger chunks. (These lists are just sequential. Keeping them in
1251        order almost never requires enough traversal to warrant using
1252        fancier ordered data structures.)  Chunks of the same size are
1253        linked with the most recently freed at the front, and allocations
1254        are taken from the back.  This results in LRU or FIFO allocation
1255        order, which tends to give each chunk an equal opportunity to be
1256        consolidated with adjacent freed chunks, resulting in larger free
1257        chunks and less fragmentation.
1258 
1259     * `top': The top-most available chunk (i.e., the one bordering the
1260        end of available memory) is treated specially. It is never
1261        included in any bin, is used only if no other chunk is
1262        available, and is released back to the system if it is very
1263        large (see M_TRIM_THRESHOLD).
1264 
1265     * `last_remainder': A bin holding only the remainder of the
1266        most recently split (non-top) chunk. This bin is checked
1267        before other non-fitting chunks, so as to provide better
1268        locality for runs of sequentially allocated chunks.
1269 
1270     *  Implicitly, through the host system's memory mapping tables.
1271        If supported, requests greater than a threshold are usually
1272        serviced via calls to mmap, and then later released via munmap.
1273 
1274 */
1275 
1276 /*  sizes, alignments */
1277 
1278 #define SIZE_SZ                (sizeof(INTERNAL_SIZE_T))
1279 #define MALLOC_ALIGNMENT       (SIZE_SZ + SIZE_SZ)
1280 #define MALLOC_ALIGN_MASK      (MALLOC_ALIGNMENT - 1)
1281 #define MINSIZE                (sizeof(struct malloc_chunk))
1282 
1283 /* conversion from malloc headers to user pointers, and back */
1284 
1285 #define chunk2mem(p)   ((Void_t*)((char*)(p) + 2*SIZE_SZ))
1286 #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ))
1287 
1288 /* pad request bytes into a usable size */
1289 
1290 #define request2size(req) \
1291  (((long)((req) + (SIZE_SZ + MALLOC_ALIGN_MASK)) < \
1292   (long)(MINSIZE + MALLOC_ALIGN_MASK)) ? MINSIZE : \
1293    (((req) + (SIZE_SZ + MALLOC_ALIGN_MASK)) & ~(MALLOC_ALIGN_MASK)))
1294 
1295 /* Check if m has acceptable alignment */
1296 
1297 #define aligned_OK(m)    (((unsigned long)((m)) & (MALLOC_ALIGN_MASK)) == 0)
1298 
1299 
1300 
1301 
1302 /*
1303   Physical chunk operations
1304 */
1305 
1306 
1307 /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */
1308 
1309 #define PREV_INUSE 0x1
1310 
1311 /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */
1312 
1313 #define IS_MMAPPED 0x2
1314 
1315 /* Bits to mask off when extracting size */
1316 
1317 #define SIZE_BITS (PREV_INUSE|IS_MMAPPED)
1318 
1319 
1320 /* Ptr to next physical malloc_chunk. */
1321 
1322 #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->size & ~PREV_INUSE) ))
1323 
1324 /* Ptr to previous physical malloc_chunk */
1325 
1326 #define prev_chunk(p)\
1327    ((mchunkptr)( ((char*)(p)) - ((p)->prev_size) ))
1328 
1329 
1330 /* Treat space at ptr + offset as a chunk */
1331 
1332 #define chunk_at_offset(p, s)  ((mchunkptr)(((char*)(p)) + (s)))
1333 
1334 
1335 
1336 
1337 /*
1338   Dealing with use bits
1339 */
1340 
1341 /* extract p's inuse bit */
1342 
1343 #define inuse(p)\
1344 ((((mchunkptr)(((char*)(p))+((p)->size & ~PREV_INUSE)))->size) & PREV_INUSE)
1345 
1346 /* extract inuse bit of previous chunk */
1347 
1348 #define prev_inuse(p)  ((p)->size & PREV_INUSE)
1349 
1350 /* check for mmap()'ed chunk */
1351 
1352 #define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED)
1353 
1354 /* set/clear chunk as in use without otherwise disturbing */
1355 
1356 #define set_inuse(p)\
1357 ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size |= PREV_INUSE
1358 
1359 #define clear_inuse(p)\
1360 ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size &= ~(PREV_INUSE)
1361 
1362 /* check/set/clear inuse bits in known places */
1363 
1364 #define inuse_bit_at_offset(p, s)\
1365  (((mchunkptr)(((char*)(p)) + (s)))->size & PREV_INUSE)
1366 
1367 #define set_inuse_bit_at_offset(p, s)\
1368  (((mchunkptr)(((char*)(p)) + (s)))->size |= PREV_INUSE)
1369 
1370 #define clear_inuse_bit_at_offset(p, s)\
1371  (((mchunkptr)(((char*)(p)) + (s)))->size &= ~(PREV_INUSE))
1372 
1373 
1374 
1375 
1376 /*
1377   Dealing with size fields
1378 */
1379 
1380 /* Get size, ignoring use bits */
1381 
1382 #define chunksize(p)          ((p)->size & ~(SIZE_BITS))
1383 
1384 /* Set size at head, without disturbing its use bit */
1385 
1386 #define set_head_size(p, s)   ((p)->size = (((p)->size & PREV_INUSE) | (s)))
1387 
1388 /* Set size/use ignoring previous bits in header */
1389 
1390 #define set_head(p, s)        ((p)->size = (s))
1391 
1392 /* Set size at footer (only when chunk is not in use) */
1393 
1394 #define set_foot(p, s)   (((mchunkptr)((char*)(p) + (s)))->prev_size = (s))
1395 
1396 
1397 
1398 
1399 
1400 /*
1401    Bins
1402 
1403     The bins, `av_' are an array of pairs of pointers serving as the
1404     heads of (initially empty) doubly-linked lists of chunks, laid out
1405     in a way so that each pair can be treated as if it were in a
1406     malloc_chunk. (This way, the fd/bk offsets for linking bin heads
1407     and chunks are the same).
1408 
1409     Bins for sizes < 512 bytes contain chunks of all the same size, spaced
1410     8 bytes apart. Larger bins are approximately logarithmically
1411     spaced. (See the table below.) The `av_' array is never mentioned
1412     directly in the code, but instead via bin access macros.
1413 
1414     Bin layout:
1415 
1416     64 bins of size       8
1417     32 bins of size      64
1418     16 bins of size     512
1419      8 bins of size    4096
1420      4 bins of size   32768
1421      2 bins of size  262144
1422      1 bin  of size what's left
1423 
1424     There is actually a little bit of slop in the numbers in bin_index
1425     for the sake of speed. This makes no difference elsewhere.
1426 
1427     The special chunks `top' and `last_remainder' get their own bins,
1428     (this is implemented via yet more trickery with the av_ array),
1429     although `top' is never properly linked to its bin since it is
1430     always handled specially.
1431 
1432 */
1433 
1434 #define NAV             128   /* number of bins */
1435 
1436 typedef struct malloc_chunk* mbinptr;
1437 
1438 /* access macros */
1439 
1440 #define bin_at(i)      ((mbinptr)((char*)&(av_[2*(i) + 2]) - 2*SIZE_SZ))
1441 #define next_bin(b)    ((mbinptr)((char*)(b) + 2 * sizeof(mbinptr)))
1442 #define prev_bin(b)    ((mbinptr)((char*)(b) - 2 * sizeof(mbinptr)))
1443 
1444 /*
1445    The first 2 bins are never indexed. The corresponding av_ cells are instead
1446    used for bookkeeping. This is not to save space, but to simplify
1447    indexing, maintain locality, and avoid some initialization tests.
1448 */
1449 
1450 #define top            (av_[2])          /* The topmost chunk */
1451 #define last_remainder (bin_at(1))       /* remainder from last split */
1452 
1453 
1454 /*
1455    Because top initially points to its own bin with initial
1456    zero size, thus forcing extension on the first malloc request,
1457    we avoid having any special code in malloc to check whether
1458    it even exists yet. But we still need to in malloc_extend_top.
1459 */
1460 
1461 #define initial_top    ((mchunkptr)(bin_at(0)))
1462 
1463 /* Helper macro to initialize bins */
1464 
1465 #define IAV(i)  bin_at(i), bin_at(i)
1466 
1467 static mbinptr av_[NAV * 2 + 2] = {
1468  NULL, NULL,
1469  IAV(0),   IAV(1),   IAV(2),   IAV(3),   IAV(4),   IAV(5),   IAV(6),   IAV(7),
1470  IAV(8),   IAV(9),   IAV(10),  IAV(11),  IAV(12),  IAV(13),  IAV(14),  IAV(15),
1471  IAV(16),  IAV(17),  IAV(18),  IAV(19),  IAV(20),  IAV(21),  IAV(22),  IAV(23),
1472  IAV(24),  IAV(25),  IAV(26),  IAV(27),  IAV(28),  IAV(29),  IAV(30),  IAV(31),
1473  IAV(32),  IAV(33),  IAV(34),  IAV(35),  IAV(36),  IAV(37),  IAV(38),  IAV(39),
1474  IAV(40),  IAV(41),  IAV(42),  IAV(43),  IAV(44),  IAV(45),  IAV(46),  IAV(47),
1475  IAV(48),  IAV(49),  IAV(50),  IAV(51),  IAV(52),  IAV(53),  IAV(54),  IAV(55),
1476  IAV(56),  IAV(57),  IAV(58),  IAV(59),  IAV(60),  IAV(61),  IAV(62),  IAV(63),
1477  IAV(64),  IAV(65),  IAV(66),  IAV(67),  IAV(68),  IAV(69),  IAV(70),  IAV(71),
1478  IAV(72),  IAV(73),  IAV(74),  IAV(75),  IAV(76),  IAV(77),  IAV(78),  IAV(79),
1479  IAV(80),  IAV(81),  IAV(82),  IAV(83),  IAV(84),  IAV(85),  IAV(86),  IAV(87),
1480  IAV(88),  IAV(89),  IAV(90),  IAV(91),  IAV(92),  IAV(93),  IAV(94),  IAV(95),
1481  IAV(96),  IAV(97),  IAV(98),  IAV(99),  IAV(100), IAV(101), IAV(102), IAV(103),
1482  IAV(104), IAV(105), IAV(106), IAV(107), IAV(108), IAV(109), IAV(110), IAV(111),
1483  IAV(112), IAV(113), IAV(114), IAV(115), IAV(116), IAV(117), IAV(118), IAV(119),
1484  IAV(120), IAV(121), IAV(122), IAV(123), IAV(124), IAV(125), IAV(126), IAV(127)
1485 };
1486 
1487 #ifdef CONFIG_NEEDS_MANUAL_RELOC
1488 static void malloc_bin_reloc(void)
1489 {
1490 	mbinptr *p = &av_[2];
1491 	size_t i;
1492 
1493 	for (i = 2; i < ARRAY_SIZE(av_); ++i, ++p)
1494 		*p = (mbinptr)((ulong)*p + gd->reloc_off);
1495 }
1496 #else
1497 static inline void malloc_bin_reloc(void) {}
1498 #endif
1499 
1500 ulong mem_malloc_start = 0;
1501 ulong mem_malloc_end = 0;
1502 ulong mem_malloc_brk = 0;
1503 
1504 void *sbrk(ptrdiff_t increment)
1505 {
1506 	ulong old = mem_malloc_brk;
1507 	ulong new = old + increment;
1508 
1509 	/*
1510 	 * if we are giving memory back make sure we clear it out since
1511 	 * we set MORECORE_CLEARS to 1
1512 	 */
1513 	if (increment < 0)
1514 		memset((void *)new, 0, -increment);
1515 
1516 	if ((new < mem_malloc_start) || (new > mem_malloc_end))
1517 		return (void *)MORECORE_FAILURE;
1518 
1519 	mem_malloc_brk = new;
1520 
1521 	return (void *)old;
1522 }
1523 
1524 void mem_malloc_init(ulong start, ulong size)
1525 {
1526 	mem_malloc_start = start;
1527 	mem_malloc_end = start + size;
1528 	mem_malloc_brk = start;
1529 
1530 	memset((void *)mem_malloc_start, 0, size);
1531 
1532 	malloc_bin_reloc();
1533 }
1534 
1535 /* field-extraction macros */
1536 
1537 #define first(b) ((b)->fd)
1538 #define last(b)  ((b)->bk)
1539 
1540 /*
1541   Indexing into bins
1542 */
1543 
1544 #define bin_index(sz)                                                          \
1545 (((((unsigned long)(sz)) >> 9) ==    0) ?       (((unsigned long)(sz)) >>  3): \
1546  ((((unsigned long)(sz)) >> 9) <=    4) ?  56 + (((unsigned long)(sz)) >>  6): \
1547  ((((unsigned long)(sz)) >> 9) <=   20) ?  91 + (((unsigned long)(sz)) >>  9): \
1548  ((((unsigned long)(sz)) >> 9) <=   84) ? 110 + (((unsigned long)(sz)) >> 12): \
1549  ((((unsigned long)(sz)) >> 9) <=  340) ? 119 + (((unsigned long)(sz)) >> 15): \
1550  ((((unsigned long)(sz)) >> 9) <= 1364) ? 124 + (((unsigned long)(sz)) >> 18): \
1551 					  126)
1552 /*
1553   bins for chunks < 512 are all spaced 8 bytes apart, and hold
1554   identically sized chunks. This is exploited in malloc.
1555 */
1556 
1557 #define MAX_SMALLBIN         63
1558 #define MAX_SMALLBIN_SIZE   512
1559 #define SMALLBIN_WIDTH        8
1560 
1561 #define smallbin_index(sz)  (((unsigned long)(sz)) >> 3)
1562 
1563 /*
1564    Requests are `small' if both the corresponding and the next bin are small
1565 */
1566 
1567 #define is_small_request(nb) (nb < MAX_SMALLBIN_SIZE - SMALLBIN_WIDTH)
1568 
1569 
1570 
1571 /*
1572     To help compensate for the large number of bins, a one-level index
1573     structure is used for bin-by-bin searching.  `binblocks' is a
1574     one-word bitvector recording whether groups of BINBLOCKWIDTH bins
1575     have any (possibly) non-empty bins, so they can be skipped over
1576     all at once during during traversals. The bits are NOT always
1577     cleared as soon as all bins in a block are empty, but instead only
1578     when all are noticed to be empty during traversal in malloc.
1579 */
1580 
1581 #define BINBLOCKWIDTH     4   /* bins per block */
1582 
1583 #define binblocks_r     ((INTERNAL_SIZE_T)av_[1]) /* bitvector of nonempty blocks */
1584 #define binblocks_w     (av_[1])
1585 
1586 /* bin<->block macros */
1587 
1588 #define idx2binblock(ix)    ((unsigned)1 << (ix / BINBLOCKWIDTH))
1589 #define mark_binblock(ii)   (binblocks_w = (mbinptr)(binblocks_r | idx2binblock(ii)))
1590 #define clear_binblock(ii)  (binblocks_w = (mbinptr)(binblocks_r & ~(idx2binblock(ii))))
1591 
1592 
1593 
1594 
1595 
1596 /*  Other static bookkeeping data */
1597 
1598 /* variables holding tunable values */
1599 
1600 static unsigned long trim_threshold   = DEFAULT_TRIM_THRESHOLD;
1601 static unsigned long top_pad          = DEFAULT_TOP_PAD;
1602 static unsigned int  n_mmaps_max      = DEFAULT_MMAP_MAX;
1603 static unsigned long mmap_threshold   = DEFAULT_MMAP_THRESHOLD;
1604 
1605 /* The first value returned from sbrk */
1606 static char* sbrk_base = (char*)(-1);
1607 
1608 /* The maximum memory obtained from system via sbrk */
1609 static unsigned long max_sbrked_mem = 0;
1610 
1611 /* The maximum via either sbrk or mmap */
1612 static unsigned long max_total_mem = 0;
1613 
1614 /* internal working copy of mallinfo */
1615 static struct mallinfo current_mallinfo = {  0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
1616 
1617 /* The total memory obtained from system via sbrk */
1618 #define sbrked_mem  (current_mallinfo.arena)
1619 
1620 /* Tracking mmaps */
1621 
1622 #ifdef DEBUG
1623 static unsigned int n_mmaps = 0;
1624 #endif	/* DEBUG */
1625 static unsigned long mmapped_mem = 0;
1626 #if HAVE_MMAP
1627 static unsigned int max_n_mmaps = 0;
1628 static unsigned long max_mmapped_mem = 0;
1629 #endif
1630 
1631 
1632 
1633 /*
1634   Debugging support
1635 */
1636 
1637 #ifdef DEBUG
1638 
1639 
1640 /*
1641   These routines make a number of assertions about the states
1642   of data structures that should be true at all times. If any
1643   are not true, it's very likely that a user program has somehow
1644   trashed memory. (It's also possible that there is a coding error
1645   in malloc. In which case, please report it!)
1646 */
1647 
1648 #if __STD_C
1649 static void do_check_chunk(mchunkptr p)
1650 #else
1651 static void do_check_chunk(p) mchunkptr p;
1652 #endif
1653 {
1654   INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1655 
1656   /* No checkable chunk is mmapped */
1657   assert(!chunk_is_mmapped(p));
1658 
1659   /* Check for legal address ... */
1660   assert((char*)p >= sbrk_base);
1661   if (p != top)
1662     assert((char*)p + sz <= (char*)top);
1663   else
1664     assert((char*)p + sz <= sbrk_base + sbrked_mem);
1665 
1666 }
1667 
1668 
1669 #if __STD_C
1670 static void do_check_free_chunk(mchunkptr p)
1671 #else
1672 static void do_check_free_chunk(p) mchunkptr p;
1673 #endif
1674 {
1675   INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1676   mchunkptr next = chunk_at_offset(p, sz);
1677 
1678   do_check_chunk(p);
1679 
1680   /* Check whether it claims to be free ... */
1681   assert(!inuse(p));
1682 
1683   /* Unless a special marker, must have OK fields */
1684   if ((long)sz >= (long)MINSIZE)
1685   {
1686     assert((sz & MALLOC_ALIGN_MASK) == 0);
1687     assert(aligned_OK(chunk2mem(p)));
1688     /* ... matching footer field */
1689     assert(next->prev_size == sz);
1690     /* ... and is fully consolidated */
1691     assert(prev_inuse(p));
1692     assert (next == top || inuse(next));
1693 
1694     /* ... and has minimally sane links */
1695     assert(p->fd->bk == p);
1696     assert(p->bk->fd == p);
1697   }
1698   else /* markers are always of size SIZE_SZ */
1699     assert(sz == SIZE_SZ);
1700 }
1701 
1702 #if __STD_C
1703 static void do_check_inuse_chunk(mchunkptr p)
1704 #else
1705 static void do_check_inuse_chunk(p) mchunkptr p;
1706 #endif
1707 {
1708   mchunkptr next = next_chunk(p);
1709   do_check_chunk(p);
1710 
1711   /* Check whether it claims to be in use ... */
1712   assert(inuse(p));
1713 
1714   /* ... and is surrounded by OK chunks.
1715     Since more things can be checked with free chunks than inuse ones,
1716     if an inuse chunk borders them and debug is on, it's worth doing them.
1717   */
1718   if (!prev_inuse(p))
1719   {
1720     mchunkptr prv = prev_chunk(p);
1721     assert(next_chunk(prv) == p);
1722     do_check_free_chunk(prv);
1723   }
1724   if (next == top)
1725   {
1726     assert(prev_inuse(next));
1727     assert(chunksize(next) >= MINSIZE);
1728   }
1729   else if (!inuse(next))
1730     do_check_free_chunk(next);
1731 
1732 }
1733 
1734 #if __STD_C
1735 static void do_check_malloced_chunk(mchunkptr p, INTERNAL_SIZE_T s)
1736 #else
1737 static void do_check_malloced_chunk(p, s) mchunkptr p; INTERNAL_SIZE_T s;
1738 #endif
1739 {
1740   INTERNAL_SIZE_T sz = p->size & ~PREV_INUSE;
1741   long room = sz - s;
1742 
1743   do_check_inuse_chunk(p);
1744 
1745   /* Legal size ... */
1746   assert((long)sz >= (long)MINSIZE);
1747   assert((sz & MALLOC_ALIGN_MASK) == 0);
1748   assert(room >= 0);
1749   assert(room < (long)MINSIZE);
1750 
1751   /* ... and alignment */
1752   assert(aligned_OK(chunk2mem(p)));
1753 
1754 
1755   /* ... and was allocated at front of an available chunk */
1756   assert(prev_inuse(p));
1757 
1758 }
1759 
1760 
1761 #define check_free_chunk(P)  do_check_free_chunk(P)
1762 #define check_inuse_chunk(P) do_check_inuse_chunk(P)
1763 #define check_chunk(P) do_check_chunk(P)
1764 #define check_malloced_chunk(P,N) do_check_malloced_chunk(P,N)
1765 #else
1766 #define check_free_chunk(P)
1767 #define check_inuse_chunk(P)
1768 #define check_chunk(P)
1769 #define check_malloced_chunk(P,N)
1770 #endif
1771 
1772 
1773 
1774 /*
1775   Macro-based internal utilities
1776 */
1777 
1778 
1779 /*
1780   Linking chunks in bin lists.
1781   Call these only with variables, not arbitrary expressions, as arguments.
1782 */
1783 
1784 /*
1785   Place chunk p of size s in its bin, in size order,
1786   putting it ahead of others of same size.
1787 */
1788 
1789 
1790 #define frontlink(P, S, IDX, BK, FD)                                          \
1791 {                                                                             \
1792   if (S < MAX_SMALLBIN_SIZE)                                                  \
1793   {                                                                           \
1794     IDX = smallbin_index(S);                                                  \
1795     mark_binblock(IDX);                                                       \
1796     BK = bin_at(IDX);                                                         \
1797     FD = BK->fd;                                                              \
1798     P->bk = BK;                                                               \
1799     P->fd = FD;                                                               \
1800     FD->bk = BK->fd = P;                                                      \
1801   }                                                                           \
1802   else                                                                        \
1803   {                                                                           \
1804     IDX = bin_index(S);                                                       \
1805     BK = bin_at(IDX);                                                         \
1806     FD = BK->fd;                                                              \
1807     if (FD == BK) mark_binblock(IDX);                                         \
1808     else                                                                      \
1809     {                                                                         \
1810       while (FD != BK && S < chunksize(FD)) FD = FD->fd;                      \
1811       BK = FD->bk;                                                            \
1812     }                                                                         \
1813     P->bk = BK;                                                               \
1814     P->fd = FD;                                                               \
1815     FD->bk = BK->fd = P;                                                      \
1816   }                                                                           \
1817 }
1818 
1819 
1820 /* take a chunk off a list */
1821 
1822 #define unlink(P, BK, FD)                                                     \
1823 {                                                                             \
1824   BK = P->bk;                                                                 \
1825   FD = P->fd;                                                                 \
1826   FD->bk = BK;                                                                \
1827   BK->fd = FD;                                                                \
1828 }                                                                             \
1829 
1830 /* Place p as the last remainder */
1831 
1832 #define link_last_remainder(P)                                                \
1833 {                                                                             \
1834   last_remainder->fd = last_remainder->bk =  P;                               \
1835   P->fd = P->bk = last_remainder;                                             \
1836 }
1837 
1838 /* Clear the last_remainder bin */
1839 
1840 #define clear_last_remainder \
1841   (last_remainder->fd = last_remainder->bk = last_remainder)
1842 
1843 
1844 
1845 
1846 
1847 /* Routines dealing with mmap(). */
1848 
1849 #if HAVE_MMAP
1850 
1851 #if __STD_C
1852 static mchunkptr mmap_chunk(size_t size)
1853 #else
1854 static mchunkptr mmap_chunk(size) size_t size;
1855 #endif
1856 {
1857   size_t page_mask = malloc_getpagesize - 1;
1858   mchunkptr p;
1859 
1860 #ifndef MAP_ANONYMOUS
1861   static int fd = -1;
1862 #endif
1863 
1864   if(n_mmaps >= n_mmaps_max) return 0; /* too many regions */
1865 
1866   /* For mmapped chunks, the overhead is one SIZE_SZ unit larger, because
1867    * there is no following chunk whose prev_size field could be used.
1868    */
1869   size = (size + SIZE_SZ + page_mask) & ~page_mask;
1870 
1871 #ifdef MAP_ANONYMOUS
1872   p = (mchunkptr)mmap(0, size, PROT_READ|PROT_WRITE,
1873 		      MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
1874 #else /* !MAP_ANONYMOUS */
1875   if (fd < 0)
1876   {
1877     fd = open("/dev/zero", O_RDWR);
1878     if(fd < 0) return 0;
1879   }
1880   p = (mchunkptr)mmap(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0);
1881 #endif
1882 
1883   if(p == (mchunkptr)-1) return 0;
1884 
1885   n_mmaps++;
1886   if (n_mmaps > max_n_mmaps) max_n_mmaps = n_mmaps;
1887 
1888   /* We demand that eight bytes into a page must be 8-byte aligned. */
1889   assert(aligned_OK(chunk2mem(p)));
1890 
1891   /* The offset to the start of the mmapped region is stored
1892    * in the prev_size field of the chunk; normally it is zero,
1893    * but that can be changed in memalign().
1894    */
1895   p->prev_size = 0;
1896   set_head(p, size|IS_MMAPPED);
1897 
1898   mmapped_mem += size;
1899   if ((unsigned long)mmapped_mem > (unsigned long)max_mmapped_mem)
1900     max_mmapped_mem = mmapped_mem;
1901   if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem)
1902     max_total_mem = mmapped_mem + sbrked_mem;
1903   return p;
1904 }
1905 
1906 #if __STD_C
1907 static void munmap_chunk(mchunkptr p)
1908 #else
1909 static void munmap_chunk(p) mchunkptr p;
1910 #endif
1911 {
1912   INTERNAL_SIZE_T size = chunksize(p);
1913   int ret;
1914 
1915   assert (chunk_is_mmapped(p));
1916   assert(! ((char*)p >= sbrk_base && (char*)p < sbrk_base + sbrked_mem));
1917   assert((n_mmaps > 0));
1918   assert(((p->prev_size + size) & (malloc_getpagesize-1)) == 0);
1919 
1920   n_mmaps--;
1921   mmapped_mem -= (size + p->prev_size);
1922 
1923   ret = munmap((char *)p - p->prev_size, size + p->prev_size);
1924 
1925   /* munmap returns non-zero on failure */
1926   assert(ret == 0);
1927 }
1928 
1929 #if HAVE_MREMAP
1930 
1931 #if __STD_C
1932 static mchunkptr mremap_chunk(mchunkptr p, size_t new_size)
1933 #else
1934 static mchunkptr mremap_chunk(p, new_size) mchunkptr p; size_t new_size;
1935 #endif
1936 {
1937   size_t page_mask = malloc_getpagesize - 1;
1938   INTERNAL_SIZE_T offset = p->prev_size;
1939   INTERNAL_SIZE_T size = chunksize(p);
1940   char *cp;
1941 
1942   assert (chunk_is_mmapped(p));
1943   assert(! ((char*)p >= sbrk_base && (char*)p < sbrk_base + sbrked_mem));
1944   assert((n_mmaps > 0));
1945   assert(((size + offset) & (malloc_getpagesize-1)) == 0);
1946 
1947   /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */
1948   new_size = (new_size + offset + SIZE_SZ + page_mask) & ~page_mask;
1949 
1950   cp = (char *)mremap((char *)p - offset, size + offset, new_size, 1);
1951 
1952   if (cp == (char *)-1) return 0;
1953 
1954   p = (mchunkptr)(cp + offset);
1955 
1956   assert(aligned_OK(chunk2mem(p)));
1957 
1958   assert((p->prev_size == offset));
1959   set_head(p, (new_size - offset)|IS_MMAPPED);
1960 
1961   mmapped_mem -= size + offset;
1962   mmapped_mem += new_size;
1963   if ((unsigned long)mmapped_mem > (unsigned long)max_mmapped_mem)
1964     max_mmapped_mem = mmapped_mem;
1965   if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem)
1966     max_total_mem = mmapped_mem + sbrked_mem;
1967   return p;
1968 }
1969 
1970 #endif /* HAVE_MREMAP */
1971 
1972 #endif /* HAVE_MMAP */
1973 
1974 
1975 
1976 
1977 /*
1978   Extend the top-most chunk by obtaining memory from system.
1979   Main interface to sbrk (but see also malloc_trim).
1980 */
1981 
1982 #if __STD_C
1983 static void malloc_extend_top(INTERNAL_SIZE_T nb)
1984 #else
1985 static void malloc_extend_top(nb) INTERNAL_SIZE_T nb;
1986 #endif
1987 {
1988   char*     brk;                  /* return value from sbrk */
1989   INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of sbrked space */
1990   INTERNAL_SIZE_T correction;     /* bytes for 2nd sbrk call */
1991   char*     new_brk;              /* return of 2nd sbrk call */
1992   INTERNAL_SIZE_T top_size;       /* new size of top chunk */
1993 
1994   mchunkptr old_top     = top;  /* Record state of old top */
1995   INTERNAL_SIZE_T old_top_size = chunksize(old_top);
1996   char*     old_end      = (char*)(chunk_at_offset(old_top, old_top_size));
1997 
1998   /* Pad request with top_pad plus minimal overhead */
1999 
2000   INTERNAL_SIZE_T    sbrk_size     = nb + top_pad + MINSIZE;
2001   unsigned long pagesz    = malloc_getpagesize;
2002 
2003   /* If not the first time through, round to preserve page boundary */
2004   /* Otherwise, we need to correct to a page size below anyway. */
2005   /* (We also correct below if an intervening foreign sbrk call.) */
2006 
2007   if (sbrk_base != (char*)(-1))
2008     sbrk_size = (sbrk_size + (pagesz - 1)) & ~(pagesz - 1);
2009 
2010   brk = (char*)(MORECORE (sbrk_size));
2011 
2012   /* Fail if sbrk failed or if a foreign sbrk call killed our space */
2013   if (brk == (char*)(MORECORE_FAILURE) ||
2014       (brk < old_end && old_top != initial_top))
2015     return;
2016 
2017   sbrked_mem += sbrk_size;
2018 
2019   if (brk == old_end) /* can just add bytes to current top */
2020   {
2021     top_size = sbrk_size + old_top_size;
2022     set_head(top, top_size | PREV_INUSE);
2023   }
2024   else
2025   {
2026     if (sbrk_base == (char*)(-1))  /* First time through. Record base */
2027       sbrk_base = brk;
2028     else  /* Someone else called sbrk().  Count those bytes as sbrked_mem. */
2029       sbrked_mem += brk - (char*)old_end;
2030 
2031     /* Guarantee alignment of first new chunk made from this space */
2032     front_misalign = (unsigned long)chunk2mem(brk) & MALLOC_ALIGN_MASK;
2033     if (front_misalign > 0)
2034     {
2035       correction = (MALLOC_ALIGNMENT) - front_misalign;
2036       brk += correction;
2037     }
2038     else
2039       correction = 0;
2040 
2041     /* Guarantee the next brk will be at a page boundary */
2042 
2043     correction += ((((unsigned long)(brk + sbrk_size))+(pagesz-1)) &
2044 		   ~(pagesz - 1)) - ((unsigned long)(brk + sbrk_size));
2045 
2046     /* Allocate correction */
2047     new_brk = (char*)(MORECORE (correction));
2048     if (new_brk == (char*)(MORECORE_FAILURE)) return;
2049 
2050     sbrked_mem += correction;
2051 
2052     top = (mchunkptr)brk;
2053     top_size = new_brk - brk + correction;
2054     set_head(top, top_size | PREV_INUSE);
2055 
2056     if (old_top != initial_top)
2057     {
2058 
2059       /* There must have been an intervening foreign sbrk call. */
2060       /* A double fencepost is necessary to prevent consolidation */
2061 
2062       /* If not enough space to do this, then user did something very wrong */
2063       if (old_top_size < MINSIZE)
2064       {
2065 	set_head(top, PREV_INUSE); /* will force null return from malloc */
2066 	return;
2067       }
2068 
2069       /* Also keep size a multiple of MALLOC_ALIGNMENT */
2070       old_top_size = (old_top_size - 3*SIZE_SZ) & ~MALLOC_ALIGN_MASK;
2071       set_head_size(old_top, old_top_size);
2072       chunk_at_offset(old_top, old_top_size          )->size =
2073 	SIZE_SZ|PREV_INUSE;
2074       chunk_at_offset(old_top, old_top_size + SIZE_SZ)->size =
2075 	SIZE_SZ|PREV_INUSE;
2076       /* If possible, release the rest. */
2077       if (old_top_size >= MINSIZE)
2078 	fREe(chunk2mem(old_top));
2079     }
2080   }
2081 
2082   if ((unsigned long)sbrked_mem > (unsigned long)max_sbrked_mem)
2083     max_sbrked_mem = sbrked_mem;
2084   if ((unsigned long)(mmapped_mem + sbrked_mem) > (unsigned long)max_total_mem)
2085     max_total_mem = mmapped_mem + sbrked_mem;
2086 
2087   /* We always land on a page boundary */
2088   assert(((unsigned long)((char*)top + top_size) & (pagesz - 1)) == 0);
2089 }
2090 
2091 
2092 
2093 
2094 /* Main public routines */
2095 
2096 
2097 /*
2098   Malloc Algorthim:
2099 
2100     The requested size is first converted into a usable form, `nb'.
2101     This currently means to add 4 bytes overhead plus possibly more to
2102     obtain 8-byte alignment and/or to obtain a size of at least
2103     MINSIZE (currently 16 bytes), the smallest allocatable size.
2104     (All fits are considered `exact' if they are within MINSIZE bytes.)
2105 
2106     From there, the first successful of the following steps is taken:
2107 
2108       1. The bin corresponding to the request size is scanned, and if
2109 	 a chunk of exactly the right size is found, it is taken.
2110 
2111       2. The most recently remaindered chunk is used if it is big
2112 	 enough.  This is a form of (roving) first fit, used only in
2113 	 the absence of exact fits. Runs of consecutive requests use
2114 	 the remainder of the chunk used for the previous such request
2115 	 whenever possible. This limited use of a first-fit style
2116 	 allocation strategy tends to give contiguous chunks
2117 	 coextensive lifetimes, which improves locality and can reduce
2118 	 fragmentation in the long run.
2119 
2120       3. Other bins are scanned in increasing size order, using a
2121 	 chunk big enough to fulfill the request, and splitting off
2122 	 any remainder.  This search is strictly by best-fit; i.e.,
2123 	 the smallest (with ties going to approximately the least
2124 	 recently used) chunk that fits is selected.
2125 
2126       4. If large enough, the chunk bordering the end of memory
2127 	 (`top') is split off. (This use of `top' is in accord with
2128 	 the best-fit search rule.  In effect, `top' is treated as
2129 	 larger (and thus less well fitting) than any other available
2130 	 chunk since it can be extended to be as large as necessary
2131 	 (up to system limitations).
2132 
2133       5. If the request size meets the mmap threshold and the
2134 	 system supports mmap, and there are few enough currently
2135 	 allocated mmapped regions, and a call to mmap succeeds,
2136 	 the request is allocated via direct memory mapping.
2137 
2138       6. Otherwise, the top of memory is extended by
2139 	 obtaining more space from the system (normally using sbrk,
2140 	 but definable to anything else via the MORECORE macro).
2141 	 Memory is gathered from the system (in system page-sized
2142 	 units) in a way that allows chunks obtained across different
2143 	 sbrk calls to be consolidated, but does not require
2144 	 contiguous memory. Thus, it should be safe to intersperse
2145 	 mallocs with other sbrk calls.
2146 
2147 
2148       All allocations are made from the the `lowest' part of any found
2149       chunk. (The implementation invariant is that prev_inuse is
2150       always true of any allocated chunk; i.e., that each allocated
2151       chunk borders either a previously allocated and still in-use chunk,
2152       or the base of its memory arena.)
2153 
2154 */
2155 
2156 #if __STD_C
2157 Void_t* mALLOc(size_t bytes)
2158 #else
2159 Void_t* mALLOc(bytes) size_t bytes;
2160 #endif
2161 {
2162   mchunkptr victim;                  /* inspected/selected chunk */
2163   INTERNAL_SIZE_T victim_size;       /* its size */
2164   int       idx;                     /* index for bin traversal */
2165   mbinptr   bin;                     /* associated bin */
2166   mchunkptr remainder;               /* remainder from a split */
2167   long      remainder_size;          /* its size */
2168   int       remainder_index;         /* its bin index */
2169   unsigned long block;               /* block traverser bit */
2170   int       startidx;                /* first bin of a traversed block */
2171   mchunkptr fwd;                     /* misc temp for linking */
2172   mchunkptr bck;                     /* misc temp for linking */
2173   mbinptr q;                         /* misc temp */
2174 
2175   INTERNAL_SIZE_T nb;
2176 
2177   /* check if mem_malloc_init() was run */
2178   if ((mem_malloc_start == 0) && (mem_malloc_end == 0)) {
2179     /* not initialized yet */
2180     return NULL;
2181   }
2182 
2183   if ((long)bytes < 0) return NULL;
2184 
2185   nb = request2size(bytes);  /* padded request size; */
2186 
2187   /* Check for exact match in a bin */
2188 
2189   if (is_small_request(nb))  /* Faster version for small requests */
2190   {
2191     idx = smallbin_index(nb);
2192 
2193     /* No traversal or size check necessary for small bins.  */
2194 
2195     q = bin_at(idx);
2196     victim = last(q);
2197 
2198     /* Also scan the next one, since it would have a remainder < MINSIZE */
2199     if (victim == q)
2200     {
2201       q = next_bin(q);
2202       victim = last(q);
2203     }
2204     if (victim != q)
2205     {
2206       victim_size = chunksize(victim);
2207       unlink(victim, bck, fwd);
2208       set_inuse_bit_at_offset(victim, victim_size);
2209       check_malloced_chunk(victim, nb);
2210       return chunk2mem(victim);
2211     }
2212 
2213     idx += 2; /* Set for bin scan below. We've already scanned 2 bins. */
2214 
2215   }
2216   else
2217   {
2218     idx = bin_index(nb);
2219     bin = bin_at(idx);
2220 
2221     for (victim = last(bin); victim != bin; victim = victim->bk)
2222     {
2223       victim_size = chunksize(victim);
2224       remainder_size = victim_size - nb;
2225 
2226       if (remainder_size >= (long)MINSIZE) /* too big */
2227       {
2228 	--idx; /* adjust to rescan below after checking last remainder */
2229 	break;
2230       }
2231 
2232       else if (remainder_size >= 0) /* exact fit */
2233       {
2234 	unlink(victim, bck, fwd);
2235 	set_inuse_bit_at_offset(victim, victim_size);
2236 	check_malloced_chunk(victim, nb);
2237 	return chunk2mem(victim);
2238       }
2239     }
2240 
2241     ++idx;
2242 
2243   }
2244 
2245   /* Try to use the last split-off remainder */
2246 
2247   if ( (victim = last_remainder->fd) != last_remainder)
2248   {
2249     victim_size = chunksize(victim);
2250     remainder_size = victim_size - nb;
2251 
2252     if (remainder_size >= (long)MINSIZE) /* re-split */
2253     {
2254       remainder = chunk_at_offset(victim, nb);
2255       set_head(victim, nb | PREV_INUSE);
2256       link_last_remainder(remainder);
2257       set_head(remainder, remainder_size | PREV_INUSE);
2258       set_foot(remainder, remainder_size);
2259       check_malloced_chunk(victim, nb);
2260       return chunk2mem(victim);
2261     }
2262 
2263     clear_last_remainder;
2264 
2265     if (remainder_size >= 0)  /* exhaust */
2266     {
2267       set_inuse_bit_at_offset(victim, victim_size);
2268       check_malloced_chunk(victim, nb);
2269       return chunk2mem(victim);
2270     }
2271 
2272     /* Else place in bin */
2273 
2274     frontlink(victim, victim_size, remainder_index, bck, fwd);
2275   }
2276 
2277   /*
2278      If there are any possibly nonempty big-enough blocks,
2279      search for best fitting chunk by scanning bins in blockwidth units.
2280   */
2281 
2282   if ( (block = idx2binblock(idx)) <= binblocks_r)
2283   {
2284 
2285     /* Get to the first marked block */
2286 
2287     if ( (block & binblocks_r) == 0)
2288     {
2289       /* force to an even block boundary */
2290       idx = (idx & ~(BINBLOCKWIDTH - 1)) + BINBLOCKWIDTH;
2291       block <<= 1;
2292       while ((block & binblocks_r) == 0)
2293       {
2294 	idx += BINBLOCKWIDTH;
2295 	block <<= 1;
2296       }
2297     }
2298 
2299     /* For each possibly nonempty block ... */
2300     for (;;)
2301     {
2302       startidx = idx;          /* (track incomplete blocks) */
2303       q = bin = bin_at(idx);
2304 
2305       /* For each bin in this block ... */
2306       do
2307       {
2308 	/* Find and use first big enough chunk ... */
2309 
2310 	for (victim = last(bin); victim != bin; victim = victim->bk)
2311 	{
2312 	  victim_size = chunksize(victim);
2313 	  remainder_size = victim_size - nb;
2314 
2315 	  if (remainder_size >= (long)MINSIZE) /* split */
2316 	  {
2317 	    remainder = chunk_at_offset(victim, nb);
2318 	    set_head(victim, nb | PREV_INUSE);
2319 	    unlink(victim, bck, fwd);
2320 	    link_last_remainder(remainder);
2321 	    set_head(remainder, remainder_size | PREV_INUSE);
2322 	    set_foot(remainder, remainder_size);
2323 	    check_malloced_chunk(victim, nb);
2324 	    return chunk2mem(victim);
2325 	  }
2326 
2327 	  else if (remainder_size >= 0)  /* take */
2328 	  {
2329 	    set_inuse_bit_at_offset(victim, victim_size);
2330 	    unlink(victim, bck, fwd);
2331 	    check_malloced_chunk(victim, nb);
2332 	    return chunk2mem(victim);
2333 	  }
2334 
2335 	}
2336 
2337        bin = next_bin(bin);
2338 
2339       } while ((++idx & (BINBLOCKWIDTH - 1)) != 0);
2340 
2341       /* Clear out the block bit. */
2342 
2343       do   /* Possibly backtrack to try to clear a partial block */
2344       {
2345 	if ((startidx & (BINBLOCKWIDTH - 1)) == 0)
2346 	{
2347 	  av_[1] = (mbinptr)(binblocks_r & ~block);
2348 	  break;
2349 	}
2350 	--startidx;
2351        q = prev_bin(q);
2352       } while (first(q) == q);
2353 
2354       /* Get to the next possibly nonempty block */
2355 
2356       if ( (block <<= 1) <= binblocks_r && (block != 0) )
2357       {
2358 	while ((block & binblocks_r) == 0)
2359 	{
2360 	  idx += BINBLOCKWIDTH;
2361 	  block <<= 1;
2362 	}
2363       }
2364       else
2365 	break;
2366     }
2367   }
2368 
2369 
2370   /* Try to use top chunk */
2371 
2372   /* Require that there be a remainder, ensuring top always exists  */
2373   if ( (remainder_size = chunksize(top) - nb) < (long)MINSIZE)
2374   {
2375 
2376 #if HAVE_MMAP
2377     /* If big and would otherwise need to extend, try to use mmap instead */
2378     if ((unsigned long)nb >= (unsigned long)mmap_threshold &&
2379 	(victim = mmap_chunk(nb)) != 0)
2380       return chunk2mem(victim);
2381 #endif
2382 
2383     /* Try to extend */
2384     malloc_extend_top(nb);
2385     if ( (remainder_size = chunksize(top) - nb) < (long)MINSIZE)
2386       return NULL; /* propagate failure */
2387   }
2388 
2389   victim = top;
2390   set_head(victim, nb | PREV_INUSE);
2391   top = chunk_at_offset(victim, nb);
2392   set_head(top, remainder_size | PREV_INUSE);
2393   check_malloced_chunk(victim, nb);
2394   return chunk2mem(victim);
2395 
2396 }
2397 
2398 
2399 
2400 
2401 /*
2402 
2403   free() algorithm :
2404 
2405     cases:
2406 
2407        1. free(0) has no effect.
2408 
2409        2. If the chunk was allocated via mmap, it is release via munmap().
2410 
2411        3. If a returned chunk borders the current high end of memory,
2412 	  it is consolidated into the top, and if the total unused
2413 	  topmost memory exceeds the trim threshold, malloc_trim is
2414 	  called.
2415 
2416        4. Other chunks are consolidated as they arrive, and
2417 	  placed in corresponding bins. (This includes the case of
2418 	  consolidating with the current `last_remainder').
2419 
2420 */
2421 
2422 
2423 #if __STD_C
2424 void fREe(Void_t* mem)
2425 #else
2426 void fREe(mem) Void_t* mem;
2427 #endif
2428 {
2429   mchunkptr p;         /* chunk corresponding to mem */
2430   INTERNAL_SIZE_T hd;  /* its head field */
2431   INTERNAL_SIZE_T sz;  /* its size */
2432   int       idx;       /* its bin index */
2433   mchunkptr next;      /* next contiguous chunk */
2434   INTERNAL_SIZE_T nextsz; /* its size */
2435   INTERNAL_SIZE_T prevsz; /* size of previous contiguous chunk */
2436   mchunkptr bck;       /* misc temp for linking */
2437   mchunkptr fwd;       /* misc temp for linking */
2438   int       islr;      /* track whether merging with last_remainder */
2439 
2440   if (mem == NULL)                              /* free(0) has no effect */
2441     return;
2442 
2443   p = mem2chunk(mem);
2444   hd = p->size;
2445 
2446 #if HAVE_MMAP
2447   if (hd & IS_MMAPPED)                       /* release mmapped memory. */
2448   {
2449     munmap_chunk(p);
2450     return;
2451   }
2452 #endif
2453 
2454   check_inuse_chunk(p);
2455 
2456   sz = hd & ~PREV_INUSE;
2457   next = chunk_at_offset(p, sz);
2458   nextsz = chunksize(next);
2459 
2460   if (next == top)                            /* merge with top */
2461   {
2462     sz += nextsz;
2463 
2464     if (!(hd & PREV_INUSE))                    /* consolidate backward */
2465     {
2466       prevsz = p->prev_size;
2467       p = chunk_at_offset(p, -((long) prevsz));
2468       sz += prevsz;
2469       unlink(p, bck, fwd);
2470     }
2471 
2472     set_head(p, sz | PREV_INUSE);
2473     top = p;
2474     if ((unsigned long)(sz) >= (unsigned long)trim_threshold)
2475       malloc_trim(top_pad);
2476     return;
2477   }
2478 
2479   set_head(next, nextsz);                    /* clear inuse bit */
2480 
2481   islr = 0;
2482 
2483   if (!(hd & PREV_INUSE))                    /* consolidate backward */
2484   {
2485     prevsz = p->prev_size;
2486     p = chunk_at_offset(p, -((long) prevsz));
2487     sz += prevsz;
2488 
2489     if (p->fd == last_remainder)             /* keep as last_remainder */
2490       islr = 1;
2491     else
2492       unlink(p, bck, fwd);
2493   }
2494 
2495   if (!(inuse_bit_at_offset(next, nextsz)))   /* consolidate forward */
2496   {
2497     sz += nextsz;
2498 
2499     if (!islr && next->fd == last_remainder)  /* re-insert last_remainder */
2500     {
2501       islr = 1;
2502       link_last_remainder(p);
2503     }
2504     else
2505       unlink(next, bck, fwd);
2506   }
2507 
2508 
2509   set_head(p, sz | PREV_INUSE);
2510   set_foot(p, sz);
2511   if (!islr)
2512     frontlink(p, sz, idx, bck, fwd);
2513 }
2514 
2515 
2516 
2517 
2518 
2519 /*
2520 
2521   Realloc algorithm:
2522 
2523     Chunks that were obtained via mmap cannot be extended or shrunk
2524     unless HAVE_MREMAP is defined, in which case mremap is used.
2525     Otherwise, if their reallocation is for additional space, they are
2526     copied.  If for less, they are just left alone.
2527 
2528     Otherwise, if the reallocation is for additional space, and the
2529     chunk can be extended, it is, else a malloc-copy-free sequence is
2530     taken.  There are several different ways that a chunk could be
2531     extended. All are tried:
2532 
2533        * Extending forward into following adjacent free chunk.
2534        * Shifting backwards, joining preceding adjacent space
2535        * Both shifting backwards and extending forward.
2536        * Extending into newly sbrked space
2537 
2538     Unless the #define REALLOC_ZERO_BYTES_FREES is set, realloc with a
2539     size argument of zero (re)allocates a minimum-sized chunk.
2540 
2541     If the reallocation is for less space, and the new request is for
2542     a `small' (<512 bytes) size, then the newly unused space is lopped
2543     off and freed.
2544 
2545     The old unix realloc convention of allowing the last-free'd chunk
2546     to be used as an argument to realloc is no longer supported.
2547     I don't know of any programs still relying on this feature,
2548     and allowing it would also allow too many other incorrect
2549     usages of realloc to be sensible.
2550 
2551 
2552 */
2553 
2554 
2555 #if __STD_C
2556 Void_t* rEALLOc(Void_t* oldmem, size_t bytes)
2557 #else
2558 Void_t* rEALLOc(oldmem, bytes) Void_t* oldmem; size_t bytes;
2559 #endif
2560 {
2561   INTERNAL_SIZE_T    nb;      /* padded request size */
2562 
2563   mchunkptr oldp;             /* chunk corresponding to oldmem */
2564   INTERNAL_SIZE_T    oldsize; /* its size */
2565 
2566   mchunkptr newp;             /* chunk to return */
2567   INTERNAL_SIZE_T    newsize; /* its size */
2568   Void_t*   newmem;           /* corresponding user mem */
2569 
2570   mchunkptr next;             /* next contiguous chunk after oldp */
2571   INTERNAL_SIZE_T  nextsize;  /* its size */
2572 
2573   mchunkptr prev;             /* previous contiguous chunk before oldp */
2574   INTERNAL_SIZE_T  prevsize;  /* its size */
2575 
2576   mchunkptr remainder;        /* holds split off extra space from newp */
2577   INTERNAL_SIZE_T  remainder_size;   /* its size */
2578 
2579   mchunkptr bck;              /* misc temp for linking */
2580   mchunkptr fwd;              /* misc temp for linking */
2581 
2582 #ifdef REALLOC_ZERO_BYTES_FREES
2583   if (bytes == 0) { fREe(oldmem); return 0; }
2584 #endif
2585 
2586   if ((long)bytes < 0) return NULL;
2587 
2588   /* realloc of null is supposed to be same as malloc */
2589   if (oldmem == NULL) return mALLOc(bytes);
2590 
2591   newp    = oldp    = mem2chunk(oldmem);
2592   newsize = oldsize = chunksize(oldp);
2593 
2594 
2595   nb = request2size(bytes);
2596 
2597 #if HAVE_MMAP
2598   if (chunk_is_mmapped(oldp))
2599   {
2600 #if HAVE_MREMAP
2601     newp = mremap_chunk(oldp, nb);
2602     if(newp) return chunk2mem(newp);
2603 #endif
2604     /* Note the extra SIZE_SZ overhead. */
2605     if(oldsize - SIZE_SZ >= nb) return oldmem; /* do nothing */
2606     /* Must alloc, copy, free. */
2607     newmem = mALLOc(bytes);
2608     if (newmem == 0) return 0; /* propagate failure */
2609     MALLOC_COPY(newmem, oldmem, oldsize - 2*SIZE_SZ);
2610     munmap_chunk(oldp);
2611     return newmem;
2612   }
2613 #endif
2614 
2615   check_inuse_chunk(oldp);
2616 
2617   if ((long)(oldsize) < (long)(nb))
2618   {
2619 
2620     /* Try expanding forward */
2621 
2622     next = chunk_at_offset(oldp, oldsize);
2623     if (next == top || !inuse(next))
2624     {
2625       nextsize = chunksize(next);
2626 
2627       /* Forward into top only if a remainder */
2628       if (next == top)
2629       {
2630 	if ((long)(nextsize + newsize) >= (long)(nb + MINSIZE))
2631 	{
2632 	  newsize += nextsize;
2633 	  top = chunk_at_offset(oldp, nb);
2634 	  set_head(top, (newsize - nb) | PREV_INUSE);
2635 	  set_head_size(oldp, nb);
2636 	  return chunk2mem(oldp);
2637 	}
2638       }
2639 
2640       /* Forward into next chunk */
2641       else if (((long)(nextsize + newsize) >= (long)(nb)))
2642       {
2643 	unlink(next, bck, fwd);
2644 	newsize  += nextsize;
2645 	goto split;
2646       }
2647     }
2648     else
2649     {
2650       next = NULL;
2651       nextsize = 0;
2652     }
2653 
2654     /* Try shifting backwards. */
2655 
2656     if (!prev_inuse(oldp))
2657     {
2658       prev = prev_chunk(oldp);
2659       prevsize = chunksize(prev);
2660 
2661       /* try forward + backward first to save a later consolidation */
2662 
2663       if (next != NULL)
2664       {
2665 	/* into top */
2666 	if (next == top)
2667 	{
2668 	  if ((long)(nextsize + prevsize + newsize) >= (long)(nb + MINSIZE))
2669 	  {
2670 	    unlink(prev, bck, fwd);
2671 	    newp = prev;
2672 	    newsize += prevsize + nextsize;
2673 	    newmem = chunk2mem(newp);
2674 	    MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2675 	    top = chunk_at_offset(newp, nb);
2676 	    set_head(top, (newsize - nb) | PREV_INUSE);
2677 	    set_head_size(newp, nb);
2678 	    return newmem;
2679 	  }
2680 	}
2681 
2682 	/* into next chunk */
2683 	else if (((long)(nextsize + prevsize + newsize) >= (long)(nb)))
2684 	{
2685 	  unlink(next, bck, fwd);
2686 	  unlink(prev, bck, fwd);
2687 	  newp = prev;
2688 	  newsize += nextsize + prevsize;
2689 	  newmem = chunk2mem(newp);
2690 	  MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2691 	  goto split;
2692 	}
2693       }
2694 
2695       /* backward only */
2696       if (prev != NULL && (long)(prevsize + newsize) >= (long)nb)
2697       {
2698 	unlink(prev, bck, fwd);
2699 	newp = prev;
2700 	newsize += prevsize;
2701 	newmem = chunk2mem(newp);
2702 	MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2703 	goto split;
2704       }
2705     }
2706 
2707     /* Must allocate */
2708 
2709     newmem = mALLOc (bytes);
2710 
2711     if (newmem == NULL)  /* propagate failure */
2712       return NULL;
2713 
2714     /* Avoid copy if newp is next chunk after oldp. */
2715     /* (This can only happen when new chunk is sbrk'ed.) */
2716 
2717     if ( (newp = mem2chunk(newmem)) == next_chunk(oldp))
2718     {
2719       newsize += chunksize(newp);
2720       newp = oldp;
2721       goto split;
2722     }
2723 
2724     /* Otherwise copy, free, and exit */
2725     MALLOC_COPY(newmem, oldmem, oldsize - SIZE_SZ);
2726     fREe(oldmem);
2727     return newmem;
2728   }
2729 
2730 
2731  split:  /* split off extra room in old or expanded chunk */
2732 
2733   if (newsize - nb >= MINSIZE) /* split off remainder */
2734   {
2735     remainder = chunk_at_offset(newp, nb);
2736     remainder_size = newsize - nb;
2737     set_head_size(newp, nb);
2738     set_head(remainder, remainder_size | PREV_INUSE);
2739     set_inuse_bit_at_offset(remainder, remainder_size);
2740     fREe(chunk2mem(remainder)); /* let free() deal with it */
2741   }
2742   else
2743   {
2744     set_head_size(newp, newsize);
2745     set_inuse_bit_at_offset(newp, newsize);
2746   }
2747 
2748   check_inuse_chunk(newp);
2749   return chunk2mem(newp);
2750 }
2751 
2752 
2753 
2754 
2755 /*
2756 
2757   memalign algorithm:
2758 
2759     memalign requests more than enough space from malloc, finds a spot
2760     within that chunk that meets the alignment request, and then
2761     possibly frees the leading and trailing space.
2762 
2763     The alignment argument must be a power of two. This property is not
2764     checked by memalign, so misuse may result in random runtime errors.
2765 
2766     8-byte alignment is guaranteed by normal malloc calls, so don't
2767     bother calling memalign with an argument of 8 or less.
2768 
2769     Overreliance on memalign is a sure way to fragment space.
2770 
2771 */
2772 
2773 
2774 #if __STD_C
2775 Void_t* mEMALIGn(size_t alignment, size_t bytes)
2776 #else
2777 Void_t* mEMALIGn(alignment, bytes) size_t alignment; size_t bytes;
2778 #endif
2779 {
2780   INTERNAL_SIZE_T    nb;      /* padded  request size */
2781   char*     m;                /* memory returned by malloc call */
2782   mchunkptr p;                /* corresponding chunk */
2783   char*     brk;              /* alignment point within p */
2784   mchunkptr newp;             /* chunk to return */
2785   INTERNAL_SIZE_T  newsize;   /* its size */
2786   INTERNAL_SIZE_T  leadsize;  /* leading space befor alignment point */
2787   mchunkptr remainder;        /* spare room at end to split off */
2788   long      remainder_size;   /* its size */
2789 
2790   if ((long)bytes < 0) return NULL;
2791 
2792   /* If need less alignment than we give anyway, just relay to malloc */
2793 
2794   if (alignment <= MALLOC_ALIGNMENT) return mALLOc(bytes);
2795 
2796   /* Otherwise, ensure that it is at least a minimum chunk size */
2797 
2798   if (alignment <  MINSIZE) alignment = MINSIZE;
2799 
2800   /* Call malloc with worst case padding to hit alignment. */
2801 
2802   nb = request2size(bytes);
2803   m  = (char*)(mALLOc(nb + alignment + MINSIZE));
2804 
2805   if (m == NULL) return NULL; /* propagate failure */
2806 
2807   p = mem2chunk(m);
2808 
2809   if ((((unsigned long)(m)) % alignment) == 0) /* aligned */
2810   {
2811 #if HAVE_MMAP
2812     if(chunk_is_mmapped(p))
2813       return chunk2mem(p); /* nothing more to do */
2814 #endif
2815   }
2816   else /* misaligned */
2817   {
2818     /*
2819       Find an aligned spot inside chunk.
2820       Since we need to give back leading space in a chunk of at
2821       least MINSIZE, if the first calculation places us at
2822       a spot with less than MINSIZE leader, we can move to the
2823       next aligned spot -- we've allocated enough total room so that
2824       this is always possible.
2825     */
2826 
2827     brk = (char*)mem2chunk(((unsigned long)(m + alignment - 1)) & -((signed) alignment));
2828     if ((long)(brk - (char*)(p)) < MINSIZE) brk = brk + alignment;
2829 
2830     newp = (mchunkptr)brk;
2831     leadsize = brk - (char*)(p);
2832     newsize = chunksize(p) - leadsize;
2833 
2834 #if HAVE_MMAP
2835     if(chunk_is_mmapped(p))
2836     {
2837       newp->prev_size = p->prev_size + leadsize;
2838       set_head(newp, newsize|IS_MMAPPED);
2839       return chunk2mem(newp);
2840     }
2841 #endif
2842 
2843     /* give back leader, use the rest */
2844 
2845     set_head(newp, newsize | PREV_INUSE);
2846     set_inuse_bit_at_offset(newp, newsize);
2847     set_head_size(p, leadsize);
2848     fREe(chunk2mem(p));
2849     p = newp;
2850 
2851     assert (newsize >= nb && (((unsigned long)(chunk2mem(p))) % alignment) == 0);
2852   }
2853 
2854   /* Also give back spare room at the end */
2855 
2856   remainder_size = chunksize(p) - nb;
2857 
2858   if (remainder_size >= (long)MINSIZE)
2859   {
2860     remainder = chunk_at_offset(p, nb);
2861     set_head(remainder, remainder_size | PREV_INUSE);
2862     set_head_size(p, nb);
2863     fREe(chunk2mem(remainder));
2864   }
2865 
2866   check_inuse_chunk(p);
2867   return chunk2mem(p);
2868 
2869 }
2870 
2871 
2872 
2873 
2874 /*
2875     valloc just invokes memalign with alignment argument equal
2876     to the page size of the system (or as near to this as can
2877     be figured out from all the includes/defines above.)
2878 */
2879 
2880 #if __STD_C
2881 Void_t* vALLOc(size_t bytes)
2882 #else
2883 Void_t* vALLOc(bytes) size_t bytes;
2884 #endif
2885 {
2886   return mEMALIGn (malloc_getpagesize, bytes);
2887 }
2888 
2889 /*
2890   pvalloc just invokes valloc for the nearest pagesize
2891   that will accommodate request
2892 */
2893 
2894 
2895 #if __STD_C
2896 Void_t* pvALLOc(size_t bytes)
2897 #else
2898 Void_t* pvALLOc(bytes) size_t bytes;
2899 #endif
2900 {
2901   size_t pagesize = malloc_getpagesize;
2902   return mEMALIGn (pagesize, (bytes + pagesize - 1) & ~(pagesize - 1));
2903 }
2904 
2905 /*
2906 
2907   calloc calls malloc, then zeroes out the allocated chunk.
2908 
2909 */
2910 
2911 #if __STD_C
2912 Void_t* cALLOc(size_t n, size_t elem_size)
2913 #else
2914 Void_t* cALLOc(n, elem_size) size_t n; size_t elem_size;
2915 #endif
2916 {
2917   mchunkptr p;
2918   INTERNAL_SIZE_T csz;
2919 
2920   INTERNAL_SIZE_T sz = n * elem_size;
2921 
2922 
2923   /* check if expand_top called, in which case don't need to clear */
2924 #if MORECORE_CLEARS
2925   mchunkptr oldtop = top;
2926   INTERNAL_SIZE_T oldtopsize = chunksize(top);
2927 #endif
2928   Void_t* mem = mALLOc (sz);
2929 
2930   if ((long)n < 0) return NULL;
2931 
2932   if (mem == NULL)
2933     return NULL;
2934   else
2935   {
2936     p = mem2chunk(mem);
2937 
2938     /* Two optional cases in which clearing not necessary */
2939 
2940 
2941 #if HAVE_MMAP
2942     if (chunk_is_mmapped(p)) return mem;
2943 #endif
2944 
2945     csz = chunksize(p);
2946 
2947 #if MORECORE_CLEARS
2948     if (p == oldtop && csz > oldtopsize)
2949     {
2950       /* clear only the bytes from non-freshly-sbrked memory */
2951       csz = oldtopsize;
2952     }
2953 #endif
2954 
2955     MALLOC_ZERO(mem, csz - SIZE_SZ);
2956     return mem;
2957   }
2958 }
2959 
2960 /*
2961 
2962   cfree just calls free. It is needed/defined on some systems
2963   that pair it with calloc, presumably for odd historical reasons.
2964 
2965 */
2966 
2967 #if !defined(INTERNAL_LINUX_C_LIB) || !defined(__ELF__)
2968 #if __STD_C
2969 void cfree(Void_t *mem)
2970 #else
2971 void cfree(mem) Void_t *mem;
2972 #endif
2973 {
2974   fREe(mem);
2975 }
2976 #endif
2977 
2978 
2979 
2980 /*
2981 
2982     Malloc_trim gives memory back to the system (via negative
2983     arguments to sbrk) if there is unused memory at the `high' end of
2984     the malloc pool. You can call this after freeing large blocks of
2985     memory to potentially reduce the system-level memory requirements
2986     of a program. However, it cannot guarantee to reduce memory. Under
2987     some allocation patterns, some large free blocks of memory will be
2988     locked between two used chunks, so they cannot be given back to
2989     the system.
2990 
2991     The `pad' argument to malloc_trim represents the amount of free
2992     trailing space to leave untrimmed. If this argument is zero,
2993     only the minimum amount of memory to maintain internal data
2994     structures will be left (one page or less). Non-zero arguments
2995     can be supplied to maintain enough trailing space to service
2996     future expected allocations without having to re-obtain memory
2997     from the system.
2998 
2999     Malloc_trim returns 1 if it actually released any memory, else 0.
3000 
3001 */
3002 
3003 #if __STD_C
3004 int malloc_trim(size_t pad)
3005 #else
3006 int malloc_trim(pad) size_t pad;
3007 #endif
3008 {
3009   long  top_size;        /* Amount of top-most memory */
3010   long  extra;           /* Amount to release */
3011   char* current_brk;     /* address returned by pre-check sbrk call */
3012   char* new_brk;         /* address returned by negative sbrk call */
3013 
3014   unsigned long pagesz = malloc_getpagesize;
3015 
3016   top_size = chunksize(top);
3017   extra = ((top_size - pad - MINSIZE + (pagesz-1)) / pagesz - 1) * pagesz;
3018 
3019   if (extra < (long)pagesz)  /* Not enough memory to release */
3020     return 0;
3021 
3022   else
3023   {
3024     /* Test to make sure no one else called sbrk */
3025     current_brk = (char*)(MORECORE (0));
3026     if (current_brk != (char*)(top) + top_size)
3027       return 0;     /* Apparently we don't own memory; must fail */
3028 
3029     else
3030     {
3031       new_brk = (char*)(MORECORE (-extra));
3032 
3033       if (new_brk == (char*)(MORECORE_FAILURE)) /* sbrk failed? */
3034       {
3035 	/* Try to figure out what we have */
3036 	current_brk = (char*)(MORECORE (0));
3037 	top_size = current_brk - (char*)top;
3038 	if (top_size >= (long)MINSIZE) /* if not, we are very very dead! */
3039 	{
3040 	  sbrked_mem = current_brk - sbrk_base;
3041 	  set_head(top, top_size | PREV_INUSE);
3042 	}
3043 	check_chunk(top);
3044 	return 0;
3045       }
3046 
3047       else
3048       {
3049 	/* Success. Adjust top accordingly. */
3050 	set_head(top, (top_size - extra) | PREV_INUSE);
3051 	sbrked_mem -= extra;
3052 	check_chunk(top);
3053 	return 1;
3054       }
3055     }
3056   }
3057 }
3058 
3059 
3060 
3061 /*
3062   malloc_usable_size:
3063 
3064     This routine tells you how many bytes you can actually use in an
3065     allocated chunk, which may be more than you requested (although
3066     often not). You can use this many bytes without worrying about
3067     overwriting other allocated objects. Not a particularly great
3068     programming practice, but still sometimes useful.
3069 
3070 */
3071 
3072 #if __STD_C
3073 size_t malloc_usable_size(Void_t* mem)
3074 #else
3075 size_t malloc_usable_size(mem) Void_t* mem;
3076 #endif
3077 {
3078   mchunkptr p;
3079   if (mem == NULL)
3080     return 0;
3081   else
3082   {
3083     p = mem2chunk(mem);
3084     if(!chunk_is_mmapped(p))
3085     {
3086       if (!inuse(p)) return 0;
3087       check_inuse_chunk(p);
3088       return chunksize(p) - SIZE_SZ;
3089     }
3090     return chunksize(p) - 2*SIZE_SZ;
3091   }
3092 }
3093 
3094 
3095 
3096 
3097 /* Utility to update current_mallinfo for malloc_stats and mallinfo() */
3098 
3099 #ifdef DEBUG
3100 static void malloc_update_mallinfo()
3101 {
3102   int i;
3103   mbinptr b;
3104   mchunkptr p;
3105 #ifdef DEBUG
3106   mchunkptr q;
3107 #endif
3108 
3109   INTERNAL_SIZE_T avail = chunksize(top);
3110   int   navail = ((long)(avail) >= (long)MINSIZE)? 1 : 0;
3111 
3112   for (i = 1; i < NAV; ++i)
3113   {
3114     b = bin_at(i);
3115     for (p = last(b); p != b; p = p->bk)
3116     {
3117 #ifdef DEBUG
3118       check_free_chunk(p);
3119       for (q = next_chunk(p);
3120 	   q < top && inuse(q) && (long)(chunksize(q)) >= (long)MINSIZE;
3121 	   q = next_chunk(q))
3122 	check_inuse_chunk(q);
3123 #endif
3124       avail += chunksize(p);
3125       navail++;
3126     }
3127   }
3128 
3129   current_mallinfo.ordblks = navail;
3130   current_mallinfo.uordblks = sbrked_mem - avail;
3131   current_mallinfo.fordblks = avail;
3132   current_mallinfo.hblks = n_mmaps;
3133   current_mallinfo.hblkhd = mmapped_mem;
3134   current_mallinfo.keepcost = chunksize(top);
3135 
3136 }
3137 #endif	/* DEBUG */
3138 
3139 
3140 
3141 /*
3142 
3143   malloc_stats:
3144 
3145     Prints on the amount of space obtain from the system (both
3146     via sbrk and mmap), the maximum amount (which may be more than
3147     current if malloc_trim and/or munmap got called), the maximum
3148     number of simultaneous mmap regions used, and the current number
3149     of bytes allocated via malloc (or realloc, etc) but not yet
3150     freed. (Note that this is the number of bytes allocated, not the
3151     number requested. It will be larger than the number requested
3152     because of alignment and bookkeeping overhead.)
3153 
3154 */
3155 
3156 #ifdef DEBUG
3157 void malloc_stats()
3158 {
3159   malloc_update_mallinfo();
3160   printf("max system bytes = %10u\n",
3161 	  (unsigned int)(max_total_mem));
3162   printf("system bytes     = %10u\n",
3163 	  (unsigned int)(sbrked_mem + mmapped_mem));
3164   printf("in use bytes     = %10u\n",
3165 	  (unsigned int)(current_mallinfo.uordblks + mmapped_mem));
3166 #if HAVE_MMAP
3167   printf("max mmap regions = %10u\n",
3168 	  (unsigned int)max_n_mmaps);
3169 #endif
3170 }
3171 #endif	/* DEBUG */
3172 
3173 /*
3174   mallinfo returns a copy of updated current mallinfo.
3175 */
3176 
3177 #ifdef DEBUG
3178 struct mallinfo mALLINFo()
3179 {
3180   malloc_update_mallinfo();
3181   return current_mallinfo;
3182 }
3183 #endif	/* DEBUG */
3184 
3185 
3186 
3187 
3188 /*
3189   mallopt:
3190 
3191     mallopt is the general SVID/XPG interface to tunable parameters.
3192     The format is to provide a (parameter-number, parameter-value) pair.
3193     mallopt then sets the corresponding parameter to the argument
3194     value if it can (i.e., so long as the value is meaningful),
3195     and returns 1 if successful else 0.
3196 
3197     See descriptions of tunable parameters above.
3198 
3199 */
3200 
3201 #if __STD_C
3202 int mALLOPt(int param_number, int value)
3203 #else
3204 int mALLOPt(param_number, value) int param_number; int value;
3205 #endif
3206 {
3207   switch(param_number)
3208   {
3209     case M_TRIM_THRESHOLD:
3210       trim_threshold = value; return 1;
3211     case M_TOP_PAD:
3212       top_pad = value; return 1;
3213     case M_MMAP_THRESHOLD:
3214       mmap_threshold = value; return 1;
3215     case M_MMAP_MAX:
3216 #if HAVE_MMAP
3217       n_mmaps_max = value; return 1;
3218 #else
3219       if (value != 0) return 0; else  n_mmaps_max = value; return 1;
3220 #endif
3221 
3222     default:
3223       return 0;
3224   }
3225 }
3226 
3227 /*
3228 
3229 History:
3230 
3231     V2.6.6 Sun Dec  5 07:42:19 1999  Doug Lea  (dl at gee)
3232       * return null for negative arguments
3233       * Added Several WIN32 cleanups from Martin C. Fong <mcfong@yahoo.com>
3234 	 * Add 'LACKS_SYS_PARAM_H' for those systems without 'sys/param.h'
3235 	  (e.g. WIN32 platforms)
3236 	 * Cleanup up header file inclusion for WIN32 platforms
3237 	 * Cleanup code to avoid Microsoft Visual C++ compiler complaints
3238 	 * Add 'USE_DL_PREFIX' to quickly allow co-existence with existing
3239 	   memory allocation routines
3240 	 * Set 'malloc_getpagesize' for WIN32 platforms (needs more work)
3241 	 * Use 'assert' rather than 'ASSERT' in WIN32 code to conform to
3242 	   usage of 'assert' in non-WIN32 code
3243 	 * Improve WIN32 'sbrk()' emulation's 'findRegion()' routine to
3244 	   avoid infinite loop
3245       * Always call 'fREe()' rather than 'free()'
3246 
3247     V2.6.5 Wed Jun 17 15:57:31 1998  Doug Lea  (dl at gee)
3248       * Fixed ordering problem with boundary-stamping
3249 
3250     V2.6.3 Sun May 19 08:17:58 1996  Doug Lea  (dl at gee)
3251       * Added pvalloc, as recommended by H.J. Liu
3252       * Added 64bit pointer support mainly from Wolfram Gloger
3253       * Added anonymously donated WIN32 sbrk emulation
3254       * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen
3255       * malloc_extend_top: fix mask error that caused wastage after
3256 	foreign sbrks
3257       * Add linux mremap support code from HJ Liu
3258 
3259     V2.6.2 Tue Dec  5 06:52:55 1995  Doug Lea  (dl at gee)
3260       * Integrated most documentation with the code.
3261       * Add support for mmap, with help from
3262 	Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
3263       * Use last_remainder in more cases.
3264       * Pack bins using idea from  colin@nyx10.cs.du.edu
3265       * Use ordered bins instead of best-fit threshhold
3266       * Eliminate block-local decls to simplify tracing and debugging.
3267       * Support another case of realloc via move into top
3268       * Fix error occuring when initial sbrk_base not word-aligned.
3269       * Rely on page size for units instead of SBRK_UNIT to
3270 	avoid surprises about sbrk alignment conventions.
3271       * Add mallinfo, mallopt. Thanks to Raymond Nijssen
3272 	(raymond@es.ele.tue.nl) for the suggestion.
3273       * Add `pad' argument to malloc_trim and top_pad mallopt parameter.
3274       * More precautions for cases where other routines call sbrk,
3275 	courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
3276       * Added macros etc., allowing use in linux libc from
3277 	H.J. Lu (hjl@gnu.ai.mit.edu)
3278       * Inverted this history list
3279 
3280     V2.6.1 Sat Dec  2 14:10:57 1995  Doug Lea  (dl at gee)
3281       * Re-tuned and fixed to behave more nicely with V2.6.0 changes.
3282       * Removed all preallocation code since under current scheme
3283 	the work required to undo bad preallocations exceeds
3284 	the work saved in good cases for most test programs.
3285       * No longer use return list or unconsolidated bins since
3286 	no scheme using them consistently outperforms those that don't
3287 	given above changes.
3288       * Use best fit for very large chunks to prevent some worst-cases.
3289       * Added some support for debugging
3290 
3291     V2.6.0 Sat Nov  4 07:05:23 1995  Doug Lea  (dl at gee)
3292       * Removed footers when chunks are in use. Thanks to
3293 	Paul Wilson (wilson@cs.texas.edu) for the suggestion.
3294 
3295     V2.5.4 Wed Nov  1 07:54:51 1995  Doug Lea  (dl at gee)
3296       * Added malloc_trim, with help from Wolfram Gloger
3297 	(wmglo@Dent.MED.Uni-Muenchen.DE).
3298 
3299     V2.5.3 Tue Apr 26 10:16:01 1994  Doug Lea  (dl at g)
3300 
3301     V2.5.2 Tue Apr  5 16:20:40 1994  Doug Lea  (dl at g)
3302       * realloc: try to expand in both directions
3303       * malloc: swap order of clean-bin strategy;
3304       * realloc: only conditionally expand backwards
3305       * Try not to scavenge used bins
3306       * Use bin counts as a guide to preallocation
3307       * Occasionally bin return list chunks in first scan
3308       * Add a few optimizations from colin@nyx10.cs.du.edu
3309 
3310     V2.5.1 Sat Aug 14 15:40:43 1993  Doug Lea  (dl at g)
3311       * faster bin computation & slightly different binning
3312       * merged all consolidations to one part of malloc proper
3313 	 (eliminating old malloc_find_space & malloc_clean_bin)
3314       * Scan 2 returns chunks (not just 1)
3315       * Propagate failure in realloc if malloc returns 0
3316       * Add stuff to allow compilation on non-ANSI compilers
3317 	  from kpv@research.att.com
3318 
3319     V2.5 Sat Aug  7 07:41:59 1993  Doug Lea  (dl at g.oswego.edu)
3320       * removed potential for odd address access in prev_chunk
3321       * removed dependency on getpagesize.h
3322       * misc cosmetics and a bit more internal documentation
3323       * anticosmetics: mangled names in macros to evade debugger strangeness
3324       * tested on sparc, hp-700, dec-mips, rs6000
3325 	  with gcc & native cc (hp, dec only) allowing
3326 	  Detlefs & Zorn comparison study (in SIGPLAN Notices.)
3327 
3328     Trial version Fri Aug 28 13:14:29 1992  Doug Lea  (dl at g.oswego.edu)
3329       * Based loosely on libg++-1.2X malloc. (It retains some of the overall
3330 	 structure of old version,  but most details differ.)
3331 
3332 */
3333