1/* 2 A version of malloc/free/realloc written by Doug Lea and released to the 3 public domain. Send questions/comments/complaints/performance data 4 to dl@cs.oswego.edu 5 6* VERSION 2.6.6 Sun Mar 5 19:10:03 2000 Doug Lea (dl at gee) 7 8 Note: There may be an updated version of this malloc obtainable at 9 ftp://g.oswego.edu/pub/misc/malloc.c 10 Check before installing! 11 12* Why use this malloc? 13 14 This is not the fastest, most space-conserving, most portable, or 15 most tunable malloc ever written. However it is among the fastest 16 while also being among the most space-conserving, portable and tunable. 17 Consistent balance across these factors results in a good general-purpose 18 allocator. For a high-level description, see 19 http://g.oswego.edu/dl/html/malloc.html 20 21* Synopsis of public routines 22 23 (Much fuller descriptions are contained in the program documentation below.) 24 25 malloc(size_t n); 26 Return a pointer to a newly allocated chunk of at least n bytes, or null 27 if no space is available. 28 free(Void_t* p); 29 Release the chunk of memory pointed to by p, or no effect if p is null. 30 realloc(Void_t* p, size_t n); 31 Return a pointer to a chunk of size n that contains the same data 32 as does chunk p up to the minimum of (n, p's size) bytes, or null 33 if no space is available. The returned pointer may or may not be 34 the same as p. If p is null, equivalent to malloc. Unless the 35 #define REALLOC_ZERO_BYTES_FREES below is set, realloc with a 36 size argument of zero (re)allocates a minimum-sized chunk. 37 memalign(size_t alignment, size_t n); 38 Return a pointer to a newly allocated chunk of n bytes, aligned 39 in accord with the alignment argument, which must be a power of 40 two. 41 valloc(size_t n); 42 Equivalent to memalign(pagesize, n), where pagesize is the page 43 size of the system (or as near to this as can be figured out from 44 all the includes/defines below.) 45 pvalloc(size_t n); 46 Equivalent to valloc(minimum-page-that-holds(n)), that is, 47 round up n to nearest pagesize. 48 calloc(size_t unit, size_t quantity); 49 Returns a pointer to quantity * unit bytes, with all locations 50 set to zero. 51 cfree(Void_t* p); 52 Equivalent to free(p). 53 malloc_trim(size_t pad); 54 Release all but pad bytes of freed top-most memory back 55 to the system. Return 1 if successful, else 0. 56 malloc_usable_size(Void_t* p); 57 Report the number usable allocated bytes associated with allocated 58 chunk p. This may or may not report more bytes than were requested, 59 due to alignment and minimum size constraints. 60 malloc_stats(); 61 Prints brief summary statistics on stderr. 62 mallinfo() 63 Returns (by copy) a struct containing various summary statistics. 64 mallopt(int parameter_number, int parameter_value) 65 Changes one of the tunable parameters described below. Returns 66 1 if successful in changing the parameter, else 0. 67 68* Vital statistics: 69 70 Alignment: 8-byte 71 8 byte alignment is currently hardwired into the design. This 72 seems to suffice for all current machines and C compilers. 73 74 Assumed pointer representation: 4 or 8 bytes 75 Code for 8-byte pointers is untested by me but has worked 76 reliably by Wolfram Gloger, who contributed most of the 77 changes supporting this. 78 79 Assumed size_t representation: 4 or 8 bytes 80 Note that size_t is allowed to be 4 bytes even if pointers are 8. 81 82 Minimum overhead per allocated chunk: 4 or 8 bytes 83 Each malloced chunk has a hidden overhead of 4 bytes holding size 84 and status information. 85 86 Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead) 87 8-byte ptrs: 24/32 bytes (including, 4/8 overhead) 88 89 When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte 90 ptrs but 4 byte size) or 24 (for 8/8) additional bytes are 91 needed; 4 (8) for a trailing size field 92 and 8 (16) bytes for free list pointers. Thus, the minimum 93 allocatable size is 16/24/32 bytes. 94 95 Even a request for zero bytes (i.e., malloc(0)) returns a 96 pointer to something of the minimum allocatable size. 97 98 Maximum allocated size: 4-byte size_t: 2^31 - 8 bytes 99 8-byte size_t: 2^63 - 16 bytes 100 101 It is assumed that (possibly signed) size_t bit values suffice to 102 represent chunk sizes. `Possibly signed' is due to the fact 103 that `size_t' may be defined on a system as either a signed or 104 an unsigned type. To be conservative, values that would appear 105 as negative numbers are avoided. 106 Requests for sizes with a negative sign bit when the request 107 size is treaded as a long will return null. 108 109 Maximum overhead wastage per allocated chunk: normally 15 bytes 110 111 Alignnment demands, plus the minimum allocatable size restriction 112 make the normal worst-case wastage 15 bytes (i.e., up to 15 113 more bytes will be allocated than were requested in malloc), with 114 two exceptions: 115 1. Because requests for zero bytes allocate non-zero space, 116 the worst case wastage for a request of zero bytes is 24 bytes. 117 2. For requests >= mmap_threshold that are serviced via 118 mmap(), the worst case wastage is 8 bytes plus the remainder 119 from a system page (the minimal mmap unit); typically 4096 bytes. 120 121* Limitations 122 123 Here are some features that are NOT currently supported 124 125 * No user-definable hooks for callbacks and the like. 126 * No automated mechanism for fully checking that all accesses 127 to malloced memory stay within their bounds. 128 * No support for compaction. 129 130* Synopsis of compile-time options: 131 132 People have reported using previous versions of this malloc on all 133 versions of Unix, sometimes by tweaking some of the defines 134 below. It has been tested most extensively on Solaris and 135 Linux. It is also reported to work on WIN32 platforms. 136 People have also reported adapting this malloc for use in 137 stand-alone embedded systems. 138 139 The implementation is in straight, hand-tuned ANSI C. Among other 140 consequences, it uses a lot of macros. Because of this, to be at 141 all usable, this code should be compiled using an optimizing compiler 142 (for example gcc -O2) that can simplify expressions and control 143 paths. 144 145 __STD_C (default: derived from C compiler defines) 146 Nonzero if using ANSI-standard C compiler, a C++ compiler, or 147 a C compiler sufficiently close to ANSI to get away with it. 148 DEBUG (default: NOT defined) 149 Define to enable debugging. Adds fairly extensive assertion-based 150 checking to help track down memory errors, but noticeably slows down 151 execution. 152 REALLOC_ZERO_BYTES_FREES (default: NOT defined) 153 Define this if you think that realloc(p, 0) should be equivalent 154 to free(p). Otherwise, since malloc returns a unique pointer for 155 malloc(0), so does realloc(p, 0). 156 HAVE_MEMCPY (default: defined) 157 Define if you are not otherwise using ANSI STD C, but still 158 have memcpy and memset in your C library and want to use them. 159 Otherwise, simple internal versions are supplied. 160 USE_MEMCPY (default: 1 if HAVE_MEMCPY is defined, 0 otherwise) 161 Define as 1 if you want the C library versions of memset and 162 memcpy called in realloc and calloc (otherwise macro versions are used). 163 At least on some platforms, the simple macro versions usually 164 outperform libc versions. 165 HAVE_MMAP (default: defined as 1) 166 Define to non-zero to optionally make malloc() use mmap() to 167 allocate very large blocks. 168 HAVE_MREMAP (default: defined as 0 unless Linux libc set) 169 Define to non-zero to optionally make realloc() use mremap() to 170 reallocate very large blocks. 171 malloc_getpagesize (default: derived from system #includes) 172 Either a constant or routine call returning the system page size. 173 HAVE_USR_INCLUDE_MALLOC_H (default: NOT defined) 174 Optionally define if you are on a system with a /usr/include/malloc.h 175 that declares struct mallinfo. It is not at all necessary to 176 define this even if you do, but will ensure consistency. 177 INTERNAL_SIZE_T (default: size_t) 178 Define to a 32-bit type (probably `unsigned int') if you are on a 179 64-bit machine, yet do not want or need to allow malloc requests of 180 greater than 2^31 to be handled. This saves space, especially for 181 very small chunks. 182 INTERNAL_LINUX_C_LIB (default: NOT defined) 183 Defined only when compiled as part of Linux libc. 184 Also note that there is some odd internal name-mangling via defines 185 (for example, internally, `malloc' is named `mALLOc') needed 186 when compiling in this case. These look funny but don't otherwise 187 affect anything. 188 WIN32 (default: undefined) 189 Define this on MS win (95, nt) platforms to compile in sbrk emulation. 190 LACKS_UNISTD_H (default: undefined if not WIN32) 191 Define this if your system does not have a <unistd.h>. 192 LACKS_SYS_PARAM_H (default: undefined if not WIN32) 193 Define this if your system does not have a <sys/param.h>. 194 MORECORE (default: sbrk) 195 The name of the routine to call to obtain more memory from the system. 196 MORECORE_FAILURE (default: -1) 197 The value returned upon failure of MORECORE. 198 MORECORE_CLEARS (default 1) 199 true (1) if the routine mapped to MORECORE zeroes out memory (which 200 holds for sbrk). 201 DEFAULT_TRIM_THRESHOLD 202 DEFAULT_TOP_PAD 203 DEFAULT_MMAP_THRESHOLD 204 DEFAULT_MMAP_MAX 205 Default values of tunable parameters (described in detail below) 206 controlling interaction with host system routines (sbrk, mmap, etc). 207 These values may also be changed dynamically via mallopt(). The 208 preset defaults are those that give best performance for typical 209 programs/systems. 210 USE_DL_PREFIX (default: undefined) 211 Prefix all public routines with the string 'dl'. Useful to 212 quickly avoid procedure declaration conflicts and linker symbol 213 conflicts with existing memory allocation routines. 214 215 216*/ 217 218 219#ifndef __MALLOC_H__ 220#define __MALLOC_H__ 221 222/* Preliminaries */ 223 224#ifndef __STD_C 225#ifdef __STDC__ 226#define __STD_C 1 227#else 228#if __cplusplus 229#define __STD_C 1 230#else 231#define __STD_C 0 232#endif /*__cplusplus*/ 233#endif /*__STDC__*/ 234#endif /*__STD_C*/ 235 236#ifndef Void_t 237#if (__STD_C || defined(WIN32)) 238#define Void_t void 239#else 240#define Void_t char 241#endif 242#endif /*Void_t*/ 243 244#if __STD_C 245#include <linux/stddef.h> /* for size_t */ 246#else 247#include <sys/types.h> 248#endif /* __STD_C */ 249 250#ifdef __cplusplus 251extern "C" { 252#endif 253 254#if 0 /* not for U-Boot */ 255#include <stdio.h> /* needed for malloc_stats */ 256#endif 257 258 259/* 260 Compile-time options 261*/ 262 263 264/* 265 Debugging: 266 267 Because freed chunks may be overwritten with link fields, this 268 malloc will often die when freed memory is overwritten by user 269 programs. This can be very effective (albeit in an annoying way) 270 in helping track down dangling pointers. 271 272 If you compile with -DDEBUG, a number of assertion checks are 273 enabled that will catch more memory errors. You probably won't be 274 able to make much sense of the actual assertion errors, but they 275 should help you locate incorrectly overwritten memory. The 276 checking is fairly extensive, and will slow down execution 277 noticeably. Calling malloc_stats or mallinfo with DEBUG set will 278 attempt to check every non-mmapped allocated and free chunk in the 279 course of computing the summmaries. (By nature, mmapped regions 280 cannot be checked very much automatically.) 281 282 Setting DEBUG may also be helpful if you are trying to modify 283 this code. The assertions in the check routines spell out in more 284 detail the assumptions and invariants underlying the algorithms. 285 286*/ 287 288/* 289 INTERNAL_SIZE_T is the word-size used for internal bookkeeping 290 of chunk sizes. On a 64-bit machine, you can reduce malloc 291 overhead by defining INTERNAL_SIZE_T to be a 32 bit `unsigned int' 292 at the expense of not being able to handle requests greater than 293 2^31. This limitation is hardly ever a concern; you are encouraged 294 to set this. However, the default version is the same as size_t. 295*/ 296 297#ifndef INTERNAL_SIZE_T 298#define INTERNAL_SIZE_T size_t 299#endif 300 301/* 302 REALLOC_ZERO_BYTES_FREES should be set if a call to 303 realloc with zero bytes should be the same as a call to free. 304 Some people think it should. Otherwise, since this malloc 305 returns a unique pointer for malloc(0), so does realloc(p, 0). 306*/ 307 308 309/* #define REALLOC_ZERO_BYTES_FREES */ 310 311 312/* 313 WIN32 causes an emulation of sbrk to be compiled in 314 mmap-based options are not currently supported in WIN32. 315*/ 316 317/* #define WIN32 */ 318#ifdef WIN32 319#define MORECORE wsbrk 320#define HAVE_MMAP 0 321 322#define LACKS_UNISTD_H 323#define LACKS_SYS_PARAM_H 324 325/* 326 Include 'windows.h' to get the necessary declarations for the 327 Microsoft Visual C++ data structures and routines used in the 'sbrk' 328 emulation. 329 330 Define WIN32_LEAN_AND_MEAN so that only the essential Microsoft 331 Visual C++ header files are included. 332*/ 333#define WIN32_LEAN_AND_MEAN 334#include <windows.h> 335#endif 336 337 338/* 339 HAVE_MEMCPY should be defined if you are not otherwise using 340 ANSI STD C, but still have memcpy and memset in your C library 341 and want to use them in calloc and realloc. Otherwise simple 342 macro versions are defined here. 343 344 USE_MEMCPY should be defined as 1 if you actually want to 345 have memset and memcpy called. People report that the macro 346 versions are often enough faster than libc versions on many 347 systems that it is better to use them. 348 349*/ 350 351#define HAVE_MEMCPY 352 353#ifndef USE_MEMCPY 354#ifdef HAVE_MEMCPY 355#define USE_MEMCPY 1 356#else 357#define USE_MEMCPY 0 358#endif 359#endif 360 361#if (__STD_C || defined(HAVE_MEMCPY)) 362 363#if __STD_C 364void* memset(void*, int, size_t); 365void* memcpy(void*, const void*, size_t); 366#else 367#ifdef WIN32 368/* On Win32 platforms, 'memset()' and 'memcpy()' are already declared in */ 369/* 'windows.h' */ 370#else 371Void_t* memset(); 372Void_t* memcpy(); 373#endif 374#endif 375#endif 376 377#if USE_MEMCPY 378 379/* The following macros are only invoked with (2n+1)-multiples of 380 INTERNAL_SIZE_T units, with a positive integer n. This is exploited 381 for fast inline execution when n is small. */ 382 383#define MALLOC_ZERO(charp, nbytes) \ 384do { \ 385 INTERNAL_SIZE_T mzsz = (nbytes); \ 386 if(mzsz <= 9*sizeof(mzsz)) { \ 387 INTERNAL_SIZE_T* mz = (INTERNAL_SIZE_T*) (charp); \ 388 if(mzsz >= 5*sizeof(mzsz)) { *mz++ = 0; \ 389 *mz++ = 0; \ 390 if(mzsz >= 7*sizeof(mzsz)) { *mz++ = 0; \ 391 *mz++ = 0; \ 392 if(mzsz >= 9*sizeof(mzsz)) { *mz++ = 0; \ 393 *mz++ = 0; }}} \ 394 *mz++ = 0; \ 395 *mz++ = 0; \ 396 *mz = 0; \ 397 } else memset((charp), 0, mzsz); \ 398} while(0) 399 400#define MALLOC_COPY(dest,src,nbytes) \ 401do { \ 402 INTERNAL_SIZE_T mcsz = (nbytes); \ 403 if(mcsz <= 9*sizeof(mcsz)) { \ 404 INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) (src); \ 405 INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) (dest); \ 406 if(mcsz >= 5*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \ 407 *mcdst++ = *mcsrc++; \ 408 if(mcsz >= 7*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \ 409 *mcdst++ = *mcsrc++; \ 410 if(mcsz >= 9*sizeof(mcsz)) { *mcdst++ = *mcsrc++; \ 411 *mcdst++ = *mcsrc++; }}} \ 412 *mcdst++ = *mcsrc++; \ 413 *mcdst++ = *mcsrc++; \ 414 *mcdst = *mcsrc ; \ 415 } else memcpy(dest, src, mcsz); \ 416} while(0) 417 418#else /* !USE_MEMCPY */ 419 420/* Use Duff's device for good zeroing/copying performance. */ 421 422#define MALLOC_ZERO(charp, nbytes) \ 423do { \ 424 INTERNAL_SIZE_T* mzp = (INTERNAL_SIZE_T*)(charp); \ 425 long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn; \ 426 if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \ 427 switch (mctmp) { \ 428 case 0: for(;;) { *mzp++ = 0; \ 429 case 7: *mzp++ = 0; \ 430 case 6: *mzp++ = 0; \ 431 case 5: *mzp++ = 0; \ 432 case 4: *mzp++ = 0; \ 433 case 3: *mzp++ = 0; \ 434 case 2: *mzp++ = 0; \ 435 case 1: *mzp++ = 0; if(mcn <= 0) break; mcn--; } \ 436 } \ 437} while(0) 438 439#define MALLOC_COPY(dest,src,nbytes) \ 440do { \ 441 INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) src; \ 442 INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) dest; \ 443 long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T), mcn; \ 444 if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \ 445 switch (mctmp) { \ 446 case 0: for(;;) { *mcdst++ = *mcsrc++; \ 447 case 7: *mcdst++ = *mcsrc++; \ 448 case 6: *mcdst++ = *mcsrc++; \ 449 case 5: *mcdst++ = *mcsrc++; \ 450 case 4: *mcdst++ = *mcsrc++; \ 451 case 3: *mcdst++ = *mcsrc++; \ 452 case 2: *mcdst++ = *mcsrc++; \ 453 case 1: *mcdst++ = *mcsrc++; if(mcn <= 0) break; mcn--; } \ 454 } \ 455} while(0) 456 457#endif 458 459 460/* 461 Define HAVE_MMAP to optionally make malloc() use mmap() to 462 allocate very large blocks. These will be returned to the 463 operating system immediately after a free(). 464*/ 465 466/*** 467#ifndef HAVE_MMAP 468#define HAVE_MMAP 1 469#endif 470***/ 471#undef HAVE_MMAP /* Not available for U-Boot */ 472 473/* 474 Define HAVE_MREMAP to make realloc() use mremap() to re-allocate 475 large blocks. This is currently only possible on Linux with 476 kernel versions newer than 1.3.77. 477*/ 478 479/*** 480#ifndef HAVE_MREMAP 481#ifdef INTERNAL_LINUX_C_LIB 482#define HAVE_MREMAP 1 483#else 484#define HAVE_MREMAP 0 485#endif 486#endif 487***/ 488#undef HAVE_MREMAP /* Not available for U-Boot */ 489 490#ifdef HAVE_MMAP 491 492#include <unistd.h> 493#include <fcntl.h> 494#include <sys/mman.h> 495 496#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON) 497#define MAP_ANONYMOUS MAP_ANON 498#endif 499 500#endif /* HAVE_MMAP */ 501 502/* 503 Access to system page size. To the extent possible, this malloc 504 manages memory from the system in page-size units. 505 506 The following mechanics for getpagesize were adapted from 507 bsd/gnu getpagesize.h 508*/ 509 510#define LACKS_UNISTD_H /* Shortcut for U-Boot */ 511#define malloc_getpagesize 4096 512 513#ifndef LACKS_UNISTD_H 514# include <unistd.h> 515#endif 516 517#ifndef malloc_getpagesize 518# ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */ 519# ifndef _SC_PAGE_SIZE 520# define _SC_PAGE_SIZE _SC_PAGESIZE 521# endif 522# endif 523# ifdef _SC_PAGE_SIZE 524# define malloc_getpagesize sysconf(_SC_PAGE_SIZE) 525# else 526# if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE) 527 extern size_t getpagesize(); 528# define malloc_getpagesize getpagesize() 529# else 530# ifdef WIN32 531# define malloc_getpagesize (4096) /* TBD: Use 'GetSystemInfo' instead */ 532# else 533# ifndef LACKS_SYS_PARAM_H 534# include <sys/param.h> 535# endif 536# ifdef EXEC_PAGESIZE 537# define malloc_getpagesize EXEC_PAGESIZE 538# else 539# ifdef NBPG 540# ifndef CLSIZE 541# define malloc_getpagesize NBPG 542# else 543# define malloc_getpagesize (NBPG * CLSIZE) 544# endif 545# else 546# ifdef NBPC 547# define malloc_getpagesize NBPC 548# else 549# ifdef PAGESIZE 550# define malloc_getpagesize PAGESIZE 551# else 552# define malloc_getpagesize (4096) /* just guess */ 553# endif 554# endif 555# endif 556# endif 557# endif 558# endif 559# endif 560#endif 561 562 563/* 564 565 This version of malloc supports the standard SVID/XPG mallinfo 566 routine that returns a struct containing the same kind of 567 information you can get from malloc_stats. It should work on 568 any SVID/XPG compliant system that has a /usr/include/malloc.h 569 defining struct mallinfo. (If you'd like to install such a thing 570 yourself, cut out the preliminary declarations as described above 571 and below and save them in a malloc.h file. But there's no 572 compelling reason to bother to do this.) 573 574 The main declaration needed is the mallinfo struct that is returned 575 (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a 576 bunch of fields, most of which are not even meaningful in this 577 version of malloc. Some of these fields are are instead filled by 578 mallinfo() with other numbers that might possibly be of interest. 579 580 HAVE_USR_INCLUDE_MALLOC_H should be set if you have a 581 /usr/include/malloc.h file that includes a declaration of struct 582 mallinfo. If so, it is included; else an SVID2/XPG2 compliant 583 version is declared below. These must be precisely the same for 584 mallinfo() to work. 585 586*/ 587 588/* #define HAVE_USR_INCLUDE_MALLOC_H */ 589 590#ifdef HAVE_USR_INCLUDE_MALLOC_H 591#include "/usr/include/malloc.h" 592#else 593 594/* SVID2/XPG mallinfo structure */ 595 596struct mallinfo { 597 int arena; /* total space allocated from system */ 598 int ordblks; /* number of non-inuse chunks */ 599 int smblks; /* unused -- always zero */ 600 int hblks; /* number of mmapped regions */ 601 int hblkhd; /* total space in mmapped regions */ 602 int usmblks; /* unused -- always zero */ 603 int fsmblks; /* unused -- always zero */ 604 int uordblks; /* total allocated space */ 605 int fordblks; /* total non-inuse space */ 606 int keepcost; /* top-most, releasable (via malloc_trim) space */ 607}; 608 609/* SVID2/XPG mallopt options */ 610 611#define M_MXFAST 1 /* UNUSED in this malloc */ 612#define M_NLBLKS 2 /* UNUSED in this malloc */ 613#define M_GRAIN 3 /* UNUSED in this malloc */ 614#define M_KEEP 4 /* UNUSED in this malloc */ 615 616#endif 617 618/* mallopt options that actually do something */ 619 620#define M_TRIM_THRESHOLD -1 621#define M_TOP_PAD -2 622#define M_MMAP_THRESHOLD -3 623#define M_MMAP_MAX -4 624 625 626#ifndef DEFAULT_TRIM_THRESHOLD 627#define DEFAULT_TRIM_THRESHOLD (128 * 1024) 628#endif 629 630/* 631 M_TRIM_THRESHOLD is the maximum amount of unused top-most memory 632 to keep before releasing via malloc_trim in free(). 633 634 Automatic trimming is mainly useful in long-lived programs. 635 Because trimming via sbrk can be slow on some systems, and can 636 sometimes be wasteful (in cases where programs immediately 637 afterward allocate more large chunks) the value should be high 638 enough so that your overall system performance would improve by 639 releasing. 640 641 The trim threshold and the mmap control parameters (see below) 642 can be traded off with one another. Trimming and mmapping are 643 two different ways of releasing unused memory back to the 644 system. Between these two, it is often possible to keep 645 system-level demands of a long-lived program down to a bare 646 minimum. For example, in one test suite of sessions measuring 647 the XF86 X server on Linux, using a trim threshold of 128K and a 648 mmap threshold of 192K led to near-minimal long term resource 649 consumption. 650 651 If you are using this malloc in a long-lived program, it should 652 pay to experiment with these values. As a rough guide, you 653 might set to a value close to the average size of a process 654 (program) running on your system. Releasing this much memory 655 would allow such a process to run in memory. Generally, it's 656 worth it to tune for trimming rather tham memory mapping when a 657 program undergoes phases where several large chunks are 658 allocated and released in ways that can reuse each other's 659 storage, perhaps mixed with phases where there are no such 660 chunks at all. And in well-behaved long-lived programs, 661 controlling release of large blocks via trimming versus mapping 662 is usually faster. 663 664 However, in most programs, these parameters serve mainly as 665 protection against the system-level effects of carrying around 666 massive amounts of unneeded memory. Since frequent calls to 667 sbrk, mmap, and munmap otherwise degrade performance, the default 668 parameters are set to relatively high values that serve only as 669 safeguards. 670 671 The default trim value is high enough to cause trimming only in 672 fairly extreme (by current memory consumption standards) cases. 673 It must be greater than page size to have any useful effect. To 674 disable trimming completely, you can set to (unsigned long)(-1); 675 676 677*/ 678 679 680#ifndef DEFAULT_TOP_PAD 681#define DEFAULT_TOP_PAD (0) 682#endif 683 684/* 685 M_TOP_PAD is the amount of extra `padding' space to allocate or 686 retain whenever sbrk is called. It is used in two ways internally: 687 688 * When sbrk is called to extend the top of the arena to satisfy 689 a new malloc request, this much padding is added to the sbrk 690 request. 691 692 * When malloc_trim is called automatically from free(), 693 it is used as the `pad' argument. 694 695 In both cases, the actual amount of padding is rounded 696 so that the end of the arena is always a system page boundary. 697 698 The main reason for using padding is to avoid calling sbrk so 699 often. Having even a small pad greatly reduces the likelihood 700 that nearly every malloc request during program start-up (or 701 after trimming) will invoke sbrk, which needlessly wastes 702 time. 703 704 Automatic rounding-up to page-size units is normally sufficient 705 to avoid measurable overhead, so the default is 0. However, in 706 systems where sbrk is relatively slow, it can pay to increase 707 this value, at the expense of carrying around more memory than 708 the program needs. 709 710*/ 711 712 713#ifndef DEFAULT_MMAP_THRESHOLD 714#define DEFAULT_MMAP_THRESHOLD (128 * 1024) 715#endif 716 717/* 718 719 M_MMAP_THRESHOLD is the request size threshold for using mmap() 720 to service a request. Requests of at least this size that cannot 721 be allocated using already-existing space will be serviced via mmap. 722 (If enough normal freed space already exists it is used instead.) 723 724 Using mmap segregates relatively large chunks of memory so that 725 they can be individually obtained and released from the host 726 system. A request serviced through mmap is never reused by any 727 other request (at least not directly; the system may just so 728 happen to remap successive requests to the same locations). 729 730 Segregating space in this way has the benefit that mmapped space 731 can ALWAYS be individually released back to the system, which 732 helps keep the system level memory demands of a long-lived 733 program low. Mapped memory can never become `locked' between 734 other chunks, as can happen with normally allocated chunks, which 735 menas that even trimming via malloc_trim would not release them. 736 737 However, it has the disadvantages that: 738 739 1. The space cannot be reclaimed, consolidated, and then 740 used to service later requests, as happens with normal chunks. 741 2. It can lead to more wastage because of mmap page alignment 742 requirements 743 3. It causes malloc performance to be more dependent on host 744 system memory management support routines which may vary in 745 implementation quality and may impose arbitrary 746 limitations. Generally, servicing a request via normal 747 malloc steps is faster than going through a system's mmap. 748 749 All together, these considerations should lead you to use mmap 750 only for relatively large requests. 751 752 753*/ 754 755 756#ifndef DEFAULT_MMAP_MAX 757#ifdef HAVE_MMAP 758#define DEFAULT_MMAP_MAX (64) 759#else 760#define DEFAULT_MMAP_MAX (0) 761#endif 762#endif 763 764/* 765 M_MMAP_MAX is the maximum number of requests to simultaneously 766 service using mmap. This parameter exists because: 767 768 1. Some systems have a limited number of internal tables for 769 use by mmap. 770 2. In most systems, overreliance on mmap can degrade overall 771 performance. 772 3. If a program allocates many large regions, it is probably 773 better off using normal sbrk-based allocation routines that 774 can reclaim and reallocate normal heap memory. Using a 775 small value allows transition into this mode after the 776 first few allocations. 777 778 Setting to 0 disables all use of mmap. If HAVE_MMAP is not set, 779 the default value is 0, and attempts to set it to non-zero values 780 in mallopt will fail. 781*/ 782 783 784/* 785 USE_DL_PREFIX will prefix all public routines with the string 'dl'. 786 Useful to quickly avoid procedure declaration conflicts and linker 787 symbol conflicts with existing memory allocation routines. 788 789*/ 790 791/* #define USE_DL_PREFIX */ 792 793 794/* 795 796 Special defines for linux libc 797 798 Except when compiled using these special defines for Linux libc 799 using weak aliases, this malloc is NOT designed to work in 800 multithreaded applications. No semaphores or other concurrency 801 control are provided to ensure that multiple malloc or free calls 802 don't run at the same time, which could be disasterous. A single 803 semaphore could be used across malloc, realloc, and free (which is 804 essentially the effect of the linux weak alias approach). It would 805 be hard to obtain finer granularity. 806 807*/ 808 809 810#ifdef INTERNAL_LINUX_C_LIB 811 812#if __STD_C 813 814Void_t * __default_morecore_init (ptrdiff_t); 815Void_t *(*__morecore)(ptrdiff_t) = __default_morecore_init; 816 817#else 818 819Void_t * __default_morecore_init (); 820Void_t *(*__morecore)() = __default_morecore_init; 821 822#endif 823 824#define MORECORE (*__morecore) 825#define MORECORE_FAILURE 0 826#define MORECORE_CLEARS 1 827 828#else /* INTERNAL_LINUX_C_LIB */ 829 830#if __STD_C 831extern Void_t* sbrk(ptrdiff_t); 832#else 833extern Void_t* sbrk(); 834#endif 835 836#ifndef MORECORE 837#define MORECORE sbrk 838#endif 839 840#ifndef MORECORE_FAILURE 841#define MORECORE_FAILURE -1 842#endif 843 844#ifndef MORECORE_CLEARS 845#define MORECORE_CLEARS 1 846#endif 847 848#endif /* INTERNAL_LINUX_C_LIB */ 849 850#if defined(INTERNAL_LINUX_C_LIB) && defined(__ELF__) 851 852#define cALLOc __libc_calloc 853#define fREe __libc_free 854#define mALLOc __libc_malloc 855#define mEMALIGn __libc_memalign 856#define rEALLOc __libc_realloc 857#define vALLOc __libc_valloc 858#define pvALLOc __libc_pvalloc 859#define mALLINFo __libc_mallinfo 860#define mALLOPt __libc_mallopt 861 862#pragma weak calloc = __libc_calloc 863#pragma weak free = __libc_free 864#pragma weak cfree = __libc_free 865#pragma weak malloc = __libc_malloc 866#pragma weak memalign = __libc_memalign 867#pragma weak realloc = __libc_realloc 868#pragma weak valloc = __libc_valloc 869#pragma weak pvalloc = __libc_pvalloc 870#pragma weak mallinfo = __libc_mallinfo 871#pragma weak mallopt = __libc_mallopt 872 873#else 874 875#if CONFIG_IS_ENABLED(SYS_MALLOC_SIMPLE) 876#define malloc malloc_simple 877#define realloc realloc_simple 878#define memalign memalign_simple 879static inline void free(void *ptr) {} 880void *calloc(size_t nmemb, size_t size); 881void *realloc_simple(void *ptr, size_t size); 882void malloc_simple_info(void); 883#else 884 885# ifdef USE_DL_PREFIX 886# define cALLOc dlcalloc 887# define fREe dlfree 888# define mALLOc dlmalloc 889# define mEMALIGn dlmemalign 890# define rEALLOc dlrealloc 891# define vALLOc dlvalloc 892# define pvALLOc dlpvalloc 893# define mALLINFo dlmallinfo 894# define mALLOPt dlmallopt 895# else /* USE_DL_PREFIX */ 896# define cALLOc calloc 897# define fREe free 898# define mALLOc malloc 899# define mEMALIGn memalign 900# define rEALLOc realloc 901# define vALLOc valloc 902# define pvALLOc pvalloc 903# define mALLINFo mallinfo 904# define mALLOPt mallopt 905# endif /* USE_DL_PREFIX */ 906 907#endif 908 909/* Set up pre-relocation malloc() ready for use */ 910int initf_malloc(void); 911 912/* Public routines */ 913 914/* Simple versions which can be used when space is tight */ 915void *malloc_simple(size_t size); 916void *memalign_simple(size_t alignment, size_t bytes); 917 918#pragma GCC visibility push(hidden) 919# if __STD_C 920 921Void_t* mALLOc(size_t); 922void fREe(Void_t*); 923Void_t* rEALLOc(Void_t*, size_t); 924Void_t* mEMALIGn(size_t, size_t); 925Void_t* vALLOc(size_t); 926Void_t* pvALLOc(size_t); 927Void_t* cALLOc(size_t, size_t); 928void cfree(Void_t*); 929int malloc_trim(size_t); 930size_t malloc_usable_size(Void_t*); 931void malloc_stats(void); 932int mALLOPt(int, int); 933struct mallinfo mALLINFo(void); 934# else 935Void_t* mALLOc(); 936void fREe(); 937Void_t* rEALLOc(); 938Void_t* mEMALIGn(); 939Void_t* vALLOc(); 940Void_t* pvALLOc(); 941Void_t* cALLOc(); 942void cfree(); 943int malloc_trim(); 944size_t malloc_usable_size(); 945void malloc_stats(); 946int mALLOPt(); 947struct mallinfo mALLINFo(); 948# endif 949#endif 950#pragma GCC visibility pop 951 952/* 953 * Begin and End of memory area for malloc(), and current "brk" 954 */ 955extern ulong mem_malloc_start; 956extern ulong mem_malloc_end; 957extern ulong mem_malloc_brk; 958 959void mem_malloc_init(ulong start, ulong size); 960 961#ifdef __cplusplus 962}; /* end of extern "C" */ 963#endif 964 965#endif /* __MALLOC_H__ */ 966