source: libcfa/src/heap.cfa@ 8f94a63

ADT ast-experimental pthread-emulation qualifiedEnum
Last change on this file since 8f94a63 was 69ec0fb, checked in by Peter A. Buhr <pabuhr@…>, 3 years ago

fix error in second update of heap allocator

  • Property mode set to 100644
File size: 57.3 KB
RevLine 
[73abe95]1//
[c4f68dc]2// Cforall Version 1.0.0 Copyright (C) 2017 University of Waterloo
3//
4// The contents of this file are covered under the licence agreement in the
5// file "LICENCE" distributed with Cforall.
[73abe95]6//
[92aca37]7// heap.cfa --
[73abe95]8//
[c4f68dc]9// Author : Peter A. Buhr
10// Created On : Tue Dec 19 21:58:35 2017
11// Last Modified By : Peter A. Buhr
[69ec0fb]12// Last Modified On : Mon Apr 25 18:51:36 2022
13// Update Count : 1147
[73abe95]14//
[c4f68dc]15
[1e034d9]16#include <string.h> // memset, memcpy
[1076d05]17#include <limits.h> // ULONG_MAX
[31a5f418]18#include <stdlib.h> // EXIT_FAILURE
19#include <errno.h> // errno, ENOMEM, EINVAL
20#include <unistd.h> // STDERR_FILENO, sbrk, sysconf
[ada0246d]21#include <malloc.h> // memalign, malloc_usable_size
[c4f68dc]22#include <sys/mman.h> // mmap, munmap
[31a5f418]23#include <sys/sysinfo.h> // get_nprocs
[c4f68dc]24
[92aca37]25#include "bits/align.hfa" // libAlign
[bcb14b5]26#include "bits/defs.hfa" // likely, unlikely
27#include "bits/locks.hfa" // __spinlock_t
[73abe95]28#include "startup.hfa" // STARTUP_PRIORITY_MEMORY
[58c671ba]29#include "math.hfa" // min
[7cfef0d]30#include "bitmanip.hfa" // is_pow2, ceiling2
[c4f68dc]31
[31a5f418]32#define FASTLOOKUP
33#define __STATISTICS__
34
35
[93c2e0a]36static bool traceHeap = false;
[d46ed6e]37
[baf608a]38inline bool traceHeap() { return traceHeap; }
[d46ed6e]39
[93c2e0a]40bool traceHeapOn() {
41 bool temp = traceHeap;
[d46ed6e]42 traceHeap = true;
43 return temp;
44} // traceHeapOn
45
[93c2e0a]46bool traceHeapOff() {
47 bool temp = traceHeap;
[d46ed6e]48 traceHeap = false;
49 return temp;
50} // traceHeapOff
51
[baf608a]52bool traceHeapTerm() { return false; }
53
[d46ed6e]54
[95eb7cf]55static bool prtFree = false;
[d46ed6e]56
[d134b15]57bool prtFree() {
[95eb7cf]58 return prtFree;
59} // prtFree
[5d4fa18]60
[95eb7cf]61bool prtFreeOn() {
62 bool temp = prtFree;
63 prtFree = true;
[5d4fa18]64 return temp;
[95eb7cf]65} // prtFreeOn
[5d4fa18]66
[95eb7cf]67bool prtFreeOff() {
68 bool temp = prtFree;
69 prtFree = false;
[5d4fa18]70 return temp;
[95eb7cf]71} // prtFreeOff
[5d4fa18]72
73
[e723100]74enum {
[31a5f418]75 // The default extension heap amount in units of bytes. When the current heap reaches the brk address, the brk
76 // address is extended by the extension amount.
77 __CFA_DEFAULT_HEAP_EXPANSION__ = 10 * 1024 * 1024,
[1e034d9]78
[31a5f418]79 // The mmap crossover point during allocation. Allocations less than this amount are allocated from buckets; values
80 // greater than or equal to this value are mmap from the operating system.
81 __CFA_DEFAULT_MMAP_START__ = 512 * 1024 + 1,
[dd23e66]82
[31a5f418]83 // The default unfreed storage amount in units of bytes. When the uC++ program ends it subtracts this amount from
84 // the malloc/free counter to adjust for storage the program does not free.
85 __CFA_DEFAULT_HEAP_UNFREED__ = 0
86}; // enum
[e723100]87
88
[f0b3f51]89#ifdef __CFA_DEBUG__
[92aca37]90static size_t allocUnfreed; // running total of allocations minus frees
[d46ed6e]91
[95eb7cf]92static void prtUnfreed() {
[c1f38e6c]93 if ( allocUnfreed != 0 ) {
[d46ed6e]94 // DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT.
[4ea1c6d]95 char helpText[512];
[92aca37]96 int len = snprintf( helpText, sizeof(helpText), "CFA warning (UNIX pid:%ld) : program terminating with %zu(0x%zx) bytes of storage allocated but not freed.\n"
[4ea1c6d]97 "Possible cause is unfreed storage allocated by the program or system/library routines called from the program.\n",
[c1f38e6c]98 (long int)getpid(), allocUnfreed, allocUnfreed ); // always print the UNIX pid
[4ea1c6d]99 __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
[b6830d74]100 } // if
[95eb7cf]101} // prtUnfreed
[d46ed6e]102
[7dd98b6]103extern int cfa_main_returned; // from interpose.cfa
[d46ed6e]104extern "C" {
[bcb14b5]105 void heapAppStart() { // called by __cfaabi_appready_startup
[c1f38e6c]106 allocUnfreed = 0;
[bcb14b5]107 } // heapAppStart
108
109 void heapAppStop() { // called by __cfaabi_appready_startdown
110 fclose( stdin ); fclose( stdout );
[b42d0ea]111 if ( cfa_main_returned ) prtUnfreed(); // do not check unfreed storage if exit called
[bcb14b5]112 } // heapAppStop
[d46ed6e]113} // extern "C"
114#endif // __CFA_DEBUG__
115
[1e034d9]116
[e723100]117// statically allocated variables => zero filled.
118static size_t heapExpand; // sbrk advance
119static size_t mmapStart; // cross over point for mmap
120static unsigned int maxBucketsUsed; // maximum number of buckets in use
[69ec0fb]121// extern visibility, used by runtime kernel
122size_t __page_size; // architecture pagesize
123int __map_prot; // common mmap/mprotect protection
[e723100]124
125
126#define SPINLOCK 0
127#define LOCKFREE 1
128#define BUCKETLOCK SPINLOCK
[9c438546]129#if BUCKETLOCK == SPINLOCK
130#elif BUCKETLOCK == LOCKFREE
131#include <stackLockFree.hfa>
132#else
133 #error undefined lock type for bucket lock
[e723100]134#endif // LOCKFREE
135
136// Recursive definitions: HeapManager needs size of bucket array and bucket area needs sizeof HeapManager storage.
[31a5f418]137// Break recursion by hardcoding number of buckets and statically checking number is correct after bucket array defined.
[95eb7cf]138enum { NoBucketSizes = 91 }; // number of buckets sizes
[d46ed6e]139
[31a5f418]140struct Heap {
[c4f68dc]141 struct Storage {
[bcb14b5]142 struct Header { // header
[c4f68dc]143 union Kind {
144 struct RealHeader {
145 union {
[bcb14b5]146 struct { // 4-byte word => 8-byte header, 8-byte word => 16-byte header
[c4f68dc]147 union {
[31a5f418]148 // 2nd low-order bit => zero filled, 3rd low-order bit => mmapped
[9c438546]149 // FreeHeader * home; // allocated block points back to home locations (must overlay alignment)
[c4f68dc]150 void * home; // allocated block points back to home locations (must overlay alignment)
151 size_t blockSize; // size for munmap (must overlay alignment)
[9c438546]152 #if BUCKETLOCK == SPINLOCK
[31a5f418]153 Storage * next; // freed block points to next freed block of same size
[c4f68dc]154 #endif // SPINLOCK
155 };
[9c438546]156 size_t size; // allocation size in bytes
[c4f68dc]157 };
[9c438546]158 #if BUCKETLOCK == LOCKFREE
159 Link(Storage) next; // freed block points next freed block of same size (double-wide)
[c4f68dc]160 #endif // LOCKFREE
161 };
[93c2e0a]162 } real; // RealHeader
[9c438546]163
[c4f68dc]164 struct FakeHeader {
[31a5f418]165 uintptr_t alignment; // 1st low-order bit => fake header & alignment
166 uintptr_t offset;
[93c2e0a]167 } fake; // FakeHeader
168 } kind; // Kind
[bcb14b5]169 } header; // Header
[31a5f418]170
[95eb7cf]171 char pad[libAlign() - sizeof( Header )];
[bcb14b5]172 char data[0]; // storage
[c4f68dc]173 }; // Storage
174
[31a5f418]175 static_assert( libAlign() >= sizeof( Storage ), "minimum alignment < sizeof( Storage )" );
[c4f68dc]176
177 struct FreeHeader {
[9c438546]178 #if BUCKETLOCK == SPINLOCK
[bcb14b5]179 __spinlock_t lock; // must be first field for alignment
180 Storage * freeList;
[c4f68dc]181 #else
[9c438546]182 StackLF(Storage) freeList;
183 #endif // BUCKETLOCK
[bcb14b5]184 size_t blockSize; // size of allocations on this list
[c4f68dc]185 }; // FreeHeader
186
187 // must be first fields for alignment
188 __spinlock_t extlock; // protects allocation-buffer extension
189 FreeHeader freeLists[NoBucketSizes]; // buckets for different allocation sizes
190
191 void * heapBegin; // start of heap
192 void * heapEnd; // logical end of heap
193 size_t heapRemaining; // amount of storage not allocated in the current chunk
[31a5f418]194}; // Heap
[c4f68dc]195
[9c438546]196#if BUCKETLOCK == LOCKFREE
[c45d2fa]197static inline {
[31a5f418]198 Link(Heap.Storage) * ?`next( Heap.Storage * this ) { return &this->header.kind.real.next; }
199 void ?{}( Heap.FreeHeader & ) {}
200 void ^?{}( Heap.FreeHeader & ) {}
[c45d2fa]201} // distribution
[9c438546]202#endif // LOCKFREE
203
[31a5f418]204static inline size_t getKey( const Heap.FreeHeader & freeheader ) { return freeheader.blockSize; }
[5d4fa18]205
[e723100]206
[31a5f418]207#ifdef FASTLOOKUP
208enum { LookupSizes = 65_536 + sizeof(Heap.Storage) }; // number of fast lookup sizes
209static unsigned char lookup[LookupSizes]; // O(1) lookup for small sizes
210#endif // FASTLOOKUP
211
212static const off_t mmapFd = -1; // fake or actual fd for anonymous file
213#ifdef __CFA_DEBUG__
214static bool heapBoot = 0; // detect recursion during boot
215#endif // __CFA_DEBUG__
216
[5d4fa18]217
[c1f38e6c]218// Size of array must harmonize with NoBucketSizes and individual bucket sizes must be multiple of 16.
[d5d3a90]219// Smaller multiples of 16 and powers of 2 are common allocation sizes, so make them generate the minimum required bucket size.
220// malloc(0) returns 0p, so no bucket is necessary for 0 bytes returning an address that can be freed.
[e723100]221static const unsigned int bucketSizes[] @= { // different bucket sizes
[31a5f418]222 16 + sizeof(Heap.Storage), 32 + sizeof(Heap.Storage), 48 + sizeof(Heap.Storage), 64 + sizeof(Heap.Storage), // 4
223 96 + sizeof(Heap.Storage), 112 + sizeof(Heap.Storage), 128 + sizeof(Heap.Storage), // 3
224 160, 192, 224, 256 + sizeof(Heap.Storage), // 4
225 320, 384, 448, 512 + sizeof(Heap.Storage), // 4
226 640, 768, 896, 1_024 + sizeof(Heap.Storage), // 4
227 1_536, 2_048 + sizeof(Heap.Storage), // 2
228 2_560, 3_072, 3_584, 4_096 + sizeof(Heap.Storage), // 4
229 6_144, 8_192 + sizeof(Heap.Storage), // 2
230 9_216, 10_240, 11_264, 12_288, 13_312, 14_336, 15_360, 16_384 + sizeof(Heap.Storage), // 8
231 18_432, 20_480, 22_528, 24_576, 26_624, 28_672, 30_720, 32_768 + sizeof(Heap.Storage), // 8
232 36_864, 40_960, 45_056, 49_152, 53_248, 57_344, 61_440, 65_536 + sizeof(Heap.Storage), // 8
233 73_728, 81_920, 90_112, 98_304, 106_496, 114_688, 122_880, 131_072 + sizeof(Heap.Storage), // 8
234 147_456, 163_840, 180_224, 196_608, 212_992, 229_376, 245_760, 262_144 + sizeof(Heap.Storage), // 8
235 294_912, 327_680, 360_448, 393_216, 425_984, 458_752, 491_520, 524_288 + sizeof(Heap.Storage), // 8
236 655_360, 786_432, 917_504, 1_048_576 + sizeof(Heap.Storage), // 4
237 1_179_648, 1_310_720, 1_441_792, 1_572_864, 1_703_936, 1_835_008, 1_966_080, 2_097_152 + sizeof(Heap.Storage), // 8
238 2_621_440, 3_145_728, 3_670_016, 4_194_304 + sizeof(Heap.Storage), // 4
[5d4fa18]239};
[e723100]240
[c1f38e6c]241static_assert( NoBucketSizes == sizeof(bucketSizes) / sizeof(bucketSizes[0] ), "size of bucket array wrong" );
[e723100]242
[9c438546]243// The constructor for heapManager is called explicitly in memory_startup.
[31a5f418]244static Heap heapManager __attribute__(( aligned (128) )) @= {}; // size of cache line to prevent false sharing
[5d4fa18]245
[c4f68dc]246
[19e5d65d]247//####################### Memory Allocation Routines Helpers ####################
248
249
[c4f68dc]250#ifdef __STATISTICS__
[31a5f418]251enum { CntTriples = 12 }; // number of counter triples
252enum { MALLOC, AALLOC, CALLOC, MEMALIGN, AMEMALIGN, CMEMALIGN, RESIZE, REALLOC, FREE };
253
254struct StatsOverlay { // overlay for iteration
255 unsigned int calls, calls_0;
256 unsigned long long int request, alloc;
257};
258
[95eb7cf]259// Heap statistics counters.
[31a5f418]260union HeapStatistics {
261 struct { // minimum qualification
262 unsigned int malloc_calls, malloc_0_calls;
263 unsigned long long int malloc_storage_request, malloc_storage_alloc;
264 unsigned int aalloc_calls, aalloc_0_calls;
265 unsigned long long int aalloc_storage_request, aalloc_storage_alloc;
266 unsigned int calloc_calls, calloc_0_calls;
267 unsigned long long int calloc_storage_request, calloc_storage_alloc;
268 unsigned int memalign_calls, memalign_0_calls;
269 unsigned long long int memalign_storage_request, memalign_storage_alloc;
270 unsigned int amemalign_calls, amemalign_0_calls;
271 unsigned long long int amemalign_storage_request, amemalign_storage_alloc;
272 unsigned int cmemalign_calls, cmemalign_0_calls;
273 unsigned long long int cmemalign_storage_request, cmemalign_storage_alloc;
274 unsigned int resize_calls, resize_0_calls;
275 unsigned long long int resize_storage_request, resize_storage_alloc;
276 unsigned int realloc_calls, realloc_0_calls;
277 unsigned long long int realloc_storage_request, realloc_storage_alloc;
278 unsigned int free_calls, free_null_calls;
279 unsigned long long int free_storage_request, free_storage_alloc;
280 unsigned int away_pulls, away_pushes;
281 unsigned long long int away_storage_request, away_storage_alloc;
282 unsigned int mmap_calls, mmap_0_calls; // no zero calls
283 unsigned long long int mmap_storage_request, mmap_storage_alloc;
284 unsigned int munmap_calls, munmap_0_calls; // no zero calls
285 unsigned long long int munmap_storage_request, munmap_storage_alloc;
286 };
287 struct StatsOverlay counters[CntTriples]; // overlay for iteration
288}; // HeapStatistics
289
290static_assert( sizeof(HeapStatistics) == CntTriples * sizeof(StatsOverlay),
291 "Heap statistics counter-triplets does not match with array size" );
292
293static void HeapStatisticsCtor( HeapStatistics & stats ) {
294 memset( &stats, '\0', sizeof(stats) ); // very fast
295 // for ( unsigned int i = 0; i < CntTriples; i += 1 ) {
296 // stats.counters[i].calls = stats.counters[i].calls_0 = stats.counters[i].request = stats.counters[i].alloc = 0;
297 // } // for
298} // HeapStatisticsCtor
299
300static HeapStatistics & ?+=?( HeapStatistics & lhs, const HeapStatistics & rhs ) {
301 for ( unsigned int i = 0; i < CntTriples; i += 1 ) {
302 lhs.counters[i].calls += rhs.counters[i].calls;
303 lhs.counters[i].calls_0 += rhs.counters[i].calls_0;
304 lhs.counters[i].request += rhs.counters[i].request;
305 lhs.counters[i].alloc += rhs.counters[i].alloc;
306 } // for
307 return lhs;
308} // ?+=?
309
310static HeapStatistics stats; // zero filled
[c1f38e6c]311static unsigned int sbrk_calls;
312static unsigned long long int sbrk_storage;
[95eb7cf]313// Statistics file descriptor (changed by malloc_stats_fd).
[709b812]314static int stats_fd = STDERR_FILENO; // default stderr
[c4f68dc]315
[31a5f418]316#define prtFmt \
317 "\nHeap statistics: (storage request / allocation)\n" \
318 " malloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
319 " aalloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
320 " calloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
321 " memalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
322 " amemalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
323 " cmemalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
324 " resize >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
325 " realloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
326 " free !null calls %'u; null calls %'u; storage %'llu / %'llu bytes\n" \
327 " sbrk calls %'u; storage %'llu bytes\n" \
328 " mmap calls %'u; storage %'llu / %'llu bytes\n" \
329 " munmap calls %'u; storage %'llu / %'llu bytes\n" \
330
[c4f68dc]331// Use "write" because streams may be shutdown when calls are made.
[19e5d65d]332static int printStats() { // see malloc_stats
[31a5f418]333 char helpText[sizeof(prtFmt) + 1024]; // space for message and values
[19e5d65d]334 return __cfaabi_bits_print_buffer( STDERR_FILENO, helpText, sizeof(helpText), prtFmt,
[31a5f418]335 stats.malloc_calls, stats.malloc_0_calls, stats.malloc_storage_request, stats.malloc_storage_alloc,
336 stats.aalloc_calls, stats.aalloc_0_calls, stats.aalloc_storage_request, stats.aalloc_storage_alloc,
337 stats.calloc_calls, stats.calloc_0_calls, stats.calloc_storage_request, stats.calloc_storage_alloc,
338 stats.memalign_calls, stats.memalign_0_calls, stats.memalign_storage_request, stats.memalign_storage_alloc,
339 stats.amemalign_calls, stats.amemalign_0_calls, stats.amemalign_storage_request, stats.amemalign_storage_alloc,
340 stats.cmemalign_calls, stats.cmemalign_0_calls, stats.cmemalign_storage_request, stats.cmemalign_storage_alloc,
341 stats.resize_calls, stats.resize_0_calls, stats.resize_storage_request, stats.resize_storage_alloc,
342 stats.realloc_calls, stats.realloc_0_calls, stats.realloc_storage_request, stats.realloc_storage_alloc,
343 stats.free_calls, stats.free_null_calls, stats.free_storage_request, stats.free_storage_alloc,
344 sbrk_calls, sbrk_storage,
345 stats.mmap_calls, stats.mmap_storage_request, stats.mmap_storage_alloc,
346 stats.munmap_calls, stats.munmap_storage_request, stats.munmap_storage_alloc
[c4f68dc]347 );
[d46ed6e]348} // printStats
[c4f68dc]349
[31a5f418]350#define prtFmtXML \
351 "<malloc version=\"1\">\n" \
352 "<heap nr=\"0\">\n" \
353 "<sizes>\n" \
354 "</sizes>\n" \
355 "<total type=\"malloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
356 "<total type=\"aalloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
357 "<total type=\"calloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
358 "<total type=\"memalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
359 "<total type=\"amemalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
360 "<total type=\"cmemalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
361 "<total type=\"resize\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
362 "<total type=\"realloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
363 "<total type=\"free\" !null=\"%'u;\" 0 null=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
364 "<total type=\"sbrk\" count=\"%'u;\" size=\"%'llu\"/> bytes\n" \
365 "<total type=\"mmap\" count=\"%'u;\" size=\"%'llu / %'llu\" / > bytes\n" \
366 "<total type=\"munmap\" count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
367 "</malloc>"
368
[bcb14b5]369static int printStatsXML( FILE * stream ) { // see malloc_info
[31a5f418]370 char helpText[sizeof(prtFmtXML) + 1024]; // space for message and values
371 return __cfaabi_bits_print_buffer( fileno( stream ), helpText, sizeof(helpText), prtFmtXML,
372 stats.malloc_calls, stats.malloc_0_calls, stats.malloc_storage_request, stats.malloc_storage_alloc,
373 stats.aalloc_calls, stats.aalloc_0_calls, stats.aalloc_storage_request, stats.aalloc_storage_alloc,
374 stats.calloc_calls, stats.calloc_0_calls, stats.calloc_storage_request, stats.calloc_storage_alloc,
375 stats.memalign_calls, stats.memalign_0_calls, stats.memalign_storage_request, stats.memalign_storage_alloc,
376 stats.amemalign_calls, stats.amemalign_0_calls, stats.amemalign_storage_request, stats.amemalign_storage_alloc,
377 stats.cmemalign_calls, stats.cmemalign_0_calls, stats.cmemalign_storage_request, stats.cmemalign_storage_alloc,
378 stats.resize_calls, stats.resize_0_calls, stats.resize_storage_request, stats.resize_storage_alloc,
379 stats.realloc_calls, stats.realloc_0_calls, stats.realloc_storage_request, stats.realloc_storage_alloc,
380 stats.free_calls, stats.free_null_calls, stats.free_storage_request, stats.free_storage_alloc,
381 sbrk_calls, sbrk_storage,
382 stats.mmap_calls, stats.mmap_storage_request, stats.mmap_storage_alloc,
383 stats.munmap_calls, stats.munmap_storage_request, stats.munmap_storage_alloc
[c4f68dc]384 );
[d46ed6e]385} // printStatsXML
[c4f68dc]386#endif // __STATISTICS__
387
[95eb7cf]388
[1e034d9]389// thunk problem
390size_t Bsearchl( unsigned int key, const unsigned int * vals, size_t dim ) {
391 size_t l = 0, m, h = dim;
392 while ( l < h ) {
393 m = (l + h) / 2;
394 if ( (unsigned int &)(vals[m]) < key ) { // cast away const
395 l = m + 1;
396 } else {
397 h = m;
398 } // if
399 } // while
400 return l;
401} // Bsearchl
402
403
[95eb7cf]404static inline bool setMmapStart( size_t value ) { // true => mmapped, false => sbrk
[ad2dced]405 if ( value < __page_size || bucketSizes[NoBucketSizes - 1] < value ) return false;
[95eb7cf]406 mmapStart = value; // set global
407
408 // find the closest bucket size less than or equal to the mmapStart size
[1e034d9]409 maxBucketsUsed = Bsearchl( (unsigned int)mmapStart, bucketSizes, NoBucketSizes ); // binary search
[95eb7cf]410 assert( maxBucketsUsed < NoBucketSizes ); // subscript failure ?
411 assert( mmapStart <= bucketSizes[maxBucketsUsed] ); // search failure ?
[1076d05]412 return true;
[95eb7cf]413} // setMmapStart
414
415
[cfbc703d]416// <-------+----------------------------------------------------> bsize (bucket size)
417// |header |addr
418//==================================================================================
419// align/offset |
420// <-----------------<------------+-----------------------------> bsize (bucket size)
421// |fake-header | addr
[19e5d65d]422#define HeaderAddr( addr ) ((Heap.Storage.Header *)( (char *)addr - sizeof(Heap.Storage) ))
423#define RealHeader( header ) ((Heap.Storage.Header *)((char *)header - header->kind.fake.offset))
[cfbc703d]424
425// <-------<<--------------------- dsize ---------------------->> bsize (bucket size)
426// |header |addr
427//==================================================================================
428// align/offset |
429// <------------------------------<<---------- dsize --------->>> bsize (bucket size)
430// |fake-header |addr
[19e5d65d]431#define DataStorage( bsize, addr, header ) (bsize - ( (char *)addr - (char *)header ))
[cfbc703d]432
433
434static inline void checkAlign( size_t alignment ) {
[19e5d65d]435 if ( unlikely( alignment < libAlign() || ! is_pow2( alignment ) ) ) {
436 abort( "**** Error **** alignment %zu for memory allocation is less than %d and/or not a power of 2.", alignment, libAlign() );
[cfbc703d]437 } // if
438} // checkAlign
439
440
[e3fea42]441static inline void checkHeader( bool check, const char name[], void * addr ) {
[b6830d74]442 if ( unlikely( check ) ) { // bad address ?
[19e5d65d]443 abort( "**** Error **** attempt to %s storage %p with address outside the heap.\n"
[bcb14b5]444 "Possible cause is duplicate free on same block or overwriting of memory.",
445 name, addr );
[b6830d74]446 } // if
[c4f68dc]447} // checkHeader
448
[95eb7cf]449
[19e5d65d]450// Manipulate sticky bits stored in unused 3 low-order bits of an address.
451// bit0 => alignment => fake header
452// bit1 => zero filled (calloc)
453// bit2 => mapped allocation versus sbrk
454#define StickyBits( header ) (((header)->kind.real.blockSize & 0x7))
455#define ClearStickyBits( addr ) (typeof(addr))((uintptr_t)(addr) & ~7)
456#define MarkAlignmentBit( align ) ((align) | 1)
457#define AlignmentBit( header ) ((((header)->kind.fake.alignment) & 1))
458#define ClearAlignmentBit( header ) (((header)->kind.fake.alignment) & ~1)
459#define ZeroFillBit( header ) ((((header)->kind.real.blockSize) & 2))
460#define ClearZeroFillBit( header ) ((((header)->kind.real.blockSize) &= ~2))
461#define MarkZeroFilledBit( header ) ((header)->kind.real.blockSize |= 2)
462#define MmappedBit( header ) ((((header)->kind.real.blockSize) & 4))
463#define MarkMmappedBit( size ) ((size) | 4)
464
465
[31a5f418]466static inline void fakeHeader( Heap.Storage.Header *& header, size_t & alignment ) {
[19e5d65d]467 if ( unlikely( AlignmentBit( header ) ) ) { // fake header ?
468 alignment = ClearAlignmentBit( header ); // clear flag from value
[c4f68dc]469 #ifdef __CFA_DEBUG__
470 checkAlign( alignment ); // check alignment
471 #endif // __CFA_DEBUG__
[19e5d65d]472 header = RealHeader( header ); // backup from fake to real header
[d5d3a90]473 } else {
[c1f38e6c]474 alignment = libAlign(); // => no fake header
[b6830d74]475 } // if
[c4f68dc]476} // fakeHeader
477
[95eb7cf]478
[19e5d65d]479static inline bool headers( const char name[] __attribute__(( unused )), void * addr, Heap.Storage.Header *& header,
480 Heap.FreeHeader *& freeHead, size_t & size, size_t & alignment ) with( heapManager ) {
481 header = HeaderAddr( addr );
[c4f68dc]482
483 #ifdef __CFA_DEBUG__
[31a5f418]484 checkHeader( header < (Heap.Storage.Header *)heapBegin, name, addr ); // bad low address ?
[c4f68dc]485 #endif // __CFA_DEBUG__
[b6830d74]486
[19e5d65d]487 if ( likely( ! StickyBits( header ) ) ) { // no sticky bits ?
488 freeHead = (Heap.FreeHeader *)(header->kind.real.home);
489 alignment = libAlign();
490 } else {
491 fakeHeader( header, alignment );
492 if ( unlikely( MmappedBit( header ) ) ) {
493 assert( addr < heapBegin || heapEnd < addr );
494 size = ClearStickyBits( header->kind.real.blockSize ); // mmap size
495 return true;
496 } // if
497
498 freeHead = (Heap.FreeHeader *)(ClearStickyBits( header->kind.real.home ));
499 } // if
500 size = freeHead->blockSize;
501
[c4f68dc]502 #ifdef __CFA_DEBUG__
[31a5f418]503 checkHeader( header < (Heap.Storage.Header *)heapBegin || (Heap.Storage.Header *)heapEnd < header, name, addr ); // bad address ? (offset could be + or -)
[c4f68dc]504
[19e5d65d]505 if ( freeHead < &freeLists[0] || &freeLists[NoBucketSizes] <= freeHead ) {
[bcb14b5]506 abort( "Attempt to %s storage %p with corrupted header.\n"
[19e5d65d]507 "Possible cause is duplicate free on same block or overwriting of header information.",
508 name, addr );
509 } // if
[c4f68dc]510 #endif // __CFA_DEBUG__
[19e5d65d]511
[bcb14b5]512 return false;
[c4f68dc]513} // headers
514
[709b812]515// #ifdef __CFA_DEBUG__
516// #if __SIZEOF_POINTER__ == 4
517// #define MASK 0xdeadbeef
518// #else
519// #define MASK 0xdeadbeefdeadbeef
520// #endif
521// #define STRIDE size_t
[e4b6b7d3]522
[709b812]523// static void * Memset( void * addr, STRIDE size ) { // debug only
524// if ( size % sizeof(STRIDE) != 0 ) abort( "Memset() : internal error, size %zd not multiple of %zd.", size, sizeof(STRIDE) );
525// if ( (STRIDE)addr % sizeof(STRIDE) != 0 ) abort( "Memset() : internal error, addr %p not multiple of %zd.", addr, sizeof(STRIDE) );
[e4b6b7d3]526
[709b812]527// STRIDE * end = (STRIDE *)addr + size / sizeof(STRIDE);
528// for ( STRIDE * p = (STRIDE *)addr; p < end; p += 1 ) *p = MASK;
529// return addr;
530// } // Memset
531// #endif // __CFA_DEBUG__
[e4b6b7d3]532
[13fece5]533
[92aca37]534#define NO_MEMORY_MSG "insufficient heap memory available for allocating %zd new bytes."
[c4f68dc]535
[9c438546]536static inline void * extend( size_t size ) with( heapManager ) {
[b6830d74]537 lock( extlock __cfaabi_dbg_ctx2 );
[19e5d65d]538
[b6830d74]539 ptrdiff_t rem = heapRemaining - size;
[19e5d65d]540 if ( unlikely( rem < 0 ) ) {
[c4f68dc]541 // If the size requested is bigger than the current remaining storage, increase the size of the heap.
542
[69ec0fb]543 size_t increase = ceiling2( size > heapExpand ? size : heapExpand, __page_size );
[ad2dced]544 // Do not call abort or strerror( errno ) as they may call malloc.
[92aca37]545 if ( sbrk( increase ) == (void *)-1 ) { // failed, no memory ?
[c4f68dc]546 unlock( extlock );
[ad2dced]547 __cfaabi_bits_print_nolock( STDERR_FILENO, NO_MEMORY_MSG, size );
[709b812]548 _exit( EXIT_FAILURE ); // give up
[92aca37]549 } // if
[19e5d65d]550
[709b812]551 // Make storage executable for thunks.
[69ec0fb]552 if ( mprotect( (char *)heapEnd + heapRemaining, increase, __map_prot ) ) {
553 unlock( extlock );
554 __cfaabi_bits_print_nolock( STDERR_FILENO, "extend() : internal error, mprotect failure, heapEnd:%p size:%zd, errno:%d.\n", heapEnd, increase, errno );
555 _exit( EXIT_FAILURE );
556 } // if
[19e5d65d]557
[bcb14b5]558 #ifdef __STATISTICS__
[c4f68dc]559 sbrk_calls += 1;
560 sbrk_storage += increase;
[bcb14b5]561 #endif // __STATISTICS__
562 #ifdef __CFA_DEBUG__
[c4f68dc]563 // Set new memory to garbage so subsequent uninitialized usages might fail.
[13fece5]564 memset( (char *)heapEnd + heapRemaining, '\xde', increase );
[ad2dced]565 //Memset( (char *)heapEnd + heapRemaining, increase );
[bcb14b5]566 #endif // __CFA_DEBUG__
[c4f68dc]567 rem = heapRemaining + increase - size;
[b6830d74]568 } // if
[c4f68dc]569
[31a5f418]570 Heap.Storage * block = (Heap.Storage *)heapEnd;
[b6830d74]571 heapRemaining = rem;
572 heapEnd = (char *)heapEnd + size;
573 unlock( extlock );
574 return block;
[c4f68dc]575} // extend
576
577
[9c438546]578static inline void * doMalloc( size_t size ) with( heapManager ) {
[31a5f418]579 Heap.Storage * block; // pointer to new block of storage
[c4f68dc]580
[b6830d74]581 // Look up size in the size list. Make sure the user request includes space for the header that must be allocated
582 // along with the block and is a multiple of the alignment size.
[69ec0fb]583
[31a5f418]584 size_t tsize = size + sizeof(Heap.Storage);
[19e5d65d]585
[b6830d74]586 if ( likely( tsize < mmapStart ) ) { // small size => sbrk
[e723100]587 size_t posn;
588 #ifdef FASTLOOKUP
589 if ( tsize < LookupSizes ) posn = lookup[tsize];
590 else
591 #endif // FASTLOOKUP
592 posn = Bsearchl( (unsigned int)tsize, bucketSizes, (size_t)maxBucketsUsed );
[31a5f418]593 Heap.FreeHeader * freeElem = &freeLists[posn];
[c1f38e6c]594 verify( freeElem <= &freeLists[maxBucketsUsed] ); // subscripting error ?
595 verify( tsize <= freeElem->blockSize ); // search failure ?
[c4f68dc]596 tsize = freeElem->blockSize; // total space needed for request
597
598 // Spin until the lock is acquired for this particular size of block.
599
[9c438546]600 #if BUCKETLOCK == SPINLOCK
[bcb14b5]601 lock( freeElem->lock __cfaabi_dbg_ctx2 );
602 block = freeElem->freeList; // remove node from stack
[c4f68dc]603 #else
[9c438546]604 block = pop( freeElem->freeList );
605 #endif // BUCKETLOCK
[95eb7cf]606 if ( unlikely( block == 0p ) ) { // no free block ?
[9c438546]607 #if BUCKETLOCK == SPINLOCK
[c4f68dc]608 unlock( freeElem->lock );
[9c438546]609 #endif // BUCKETLOCK
[bcb14b5]610
[c4f68dc]611 // Freelist for that size was empty, so carve it out of the heap if there's enough left, or get some more
612 // and then carve it off.
613
[31a5f418]614 block = (Heap.Storage *)extend( tsize ); // mutual exclusion on call
[9c438546]615 #if BUCKETLOCK == SPINLOCK
[c4f68dc]616 } else {
617 freeElem->freeList = block->header.kind.real.next;
618 unlock( freeElem->lock );
[9c438546]619 #endif // BUCKETLOCK
[c4f68dc]620 } // if
621
622 block->header.kind.real.home = freeElem; // pointer back to free list of apropriate size
[bcb14b5]623 } else { // large size => mmap
[ad2dced]624 if ( unlikely( size > ULONG_MAX - __page_size ) ) return 0p;
625 tsize = ceiling2( tsize, __page_size ); // must be multiple of page size
[c4f68dc]626 #ifdef __STATISTICS__
[31a5f418]627 __atomic_add_fetch( &stats.mmap_calls, 1, __ATOMIC_SEQ_CST );
628 __atomic_add_fetch( &stats.mmap_storage_request, size, __ATOMIC_SEQ_CST );
629 __atomic_add_fetch( &stats.mmap_storage_alloc, tsize, __ATOMIC_SEQ_CST );
[c4f68dc]630 #endif // __STATISTICS__
[92aca37]631
[31a5f418]632 block = (Heap.Storage *)mmap( 0, tsize, __map_prot, MAP_PRIVATE | MAP_ANONYMOUS, mmapFd, 0 );
633 if ( block == (Heap.Storage *)MAP_FAILED ) { // failed ?
[92aca37]634 if ( errno == ENOMEM ) abort( NO_MEMORY_MSG, tsize ); // no memory
[c4f68dc]635 // Do not call strerror( errno ) as it may call malloc.
[31a5f418]636 abort( "(Heap &)0x%p.doMalloc() : internal error, mmap failure, size:%zu errno:%d.", &heapManager, tsize, errno );
[92aca37]637 } //if
[bcb14b5]638 #ifdef __CFA_DEBUG__
[c4f68dc]639 // Set new memory to garbage so subsequent uninitialized usages might fail.
[13fece5]640 memset( block, '\xde', tsize );
[ad2dced]641 //Memset( block, tsize );
[bcb14b5]642 #endif // __CFA_DEBUG__
[19e5d65d]643 block->header.kind.real.blockSize = MarkMmappedBit( tsize ); // storage size for munmap
[bcb14b5]644 } // if
[c4f68dc]645
[9c438546]646 block->header.kind.real.size = size; // store allocation size
[95eb7cf]647 void * addr = &(block->data); // adjust off header to user bytes
[c1f38e6c]648 verify( ((uintptr_t)addr & (libAlign() - 1)) == 0 ); // minimum alignment ?
[c4f68dc]649
650 #ifdef __CFA_DEBUG__
[c1f38e6c]651 __atomic_add_fetch( &allocUnfreed, tsize, __ATOMIC_SEQ_CST );
[bcb14b5]652 if ( traceHeap() ) {
653 enum { BufferSize = 64 };
654 char helpText[BufferSize];
[95eb7cf]655 int len = snprintf( helpText, BufferSize, "%p = Malloc( %zu ) (allocated %zu)\n", addr, size, tsize );
656 __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
[bcb14b5]657 } // if
[c4f68dc]658 #endif // __CFA_DEBUG__
659
[95eb7cf]660 return addr;
[c4f68dc]661} // doMalloc
662
663
[9c438546]664static inline void doFree( void * addr ) with( heapManager ) {
[c4f68dc]665 #ifdef __CFA_DEBUG__
[95eb7cf]666 if ( unlikely( heapManager.heapBegin == 0p ) ) {
[bcb14b5]667 abort( "doFree( %p ) : internal error, called before heap is initialized.", addr );
668 } // if
[c4f68dc]669 #endif // __CFA_DEBUG__
670
[31a5f418]671 Heap.Storage.Header * header;
672 Heap.FreeHeader * freeElem;
[b6830d74]673 size_t size, alignment; // not used (see realloc)
[c4f68dc]674
[b6830d74]675 if ( headers( "free", addr, header, freeElem, size, alignment ) ) { // mmapped ?
[c4f68dc]676 #ifdef __STATISTICS__
[31a5f418]677 __atomic_add_fetch( &stats.munmap_calls, 1, __ATOMIC_SEQ_CST );
678 __atomic_add_fetch( &stats.munmap_storage_request, header->kind.real.size, __ATOMIC_SEQ_CST );
679 __atomic_add_fetch( &stats.munmap_storage_alloc, size, __ATOMIC_SEQ_CST );
[c4f68dc]680 #endif // __STATISTICS__
681 if ( munmap( header, size ) == -1 ) {
682 abort( "Attempt to deallocate storage %p not allocated or with corrupt header.\n"
[bcb14b5]683 "Possible cause is invalid pointer.",
684 addr );
[c4f68dc]685 } // if
[bcb14b5]686 } else {
[c4f68dc]687 #ifdef __CFA_DEBUG__
[bcb14b5]688 // Set free memory to garbage so subsequent usages might fail.
[31a5f418]689 memset( ((Heap.Storage *)header)->data, '\xde', freeElem->blockSize - sizeof( Heap.Storage ) );
690 //Memset( ((Heap.Storage *)header)->data, freeElem->blockSize - sizeof( Heap.Storage ) );
[c4f68dc]691 #endif // __CFA_DEBUG__
692
693 #ifdef __STATISTICS__
[31a5f418]694 __atomic_add_fetch( &stats.free_calls, 1, __ATOMIC_SEQ_CST );
695 __atomic_add_fetch( &stats.free_storage_request, header->kind.real.size, __ATOMIC_SEQ_CST );
696 __atomic_add_fetch( &stats.free_storage_alloc, size, __ATOMIC_SEQ_CST );
[c4f68dc]697 #endif // __STATISTICS__
[b38b22f]698
[9c438546]699 #if BUCKETLOCK == SPINLOCK
[bcb14b5]700 lock( freeElem->lock __cfaabi_dbg_ctx2 ); // acquire spin lock
701 header->kind.real.next = freeElem->freeList; // push on stack
[31a5f418]702 freeElem->freeList = (Heap.Storage *)header;
[bcb14b5]703 unlock( freeElem->lock ); // release spin lock
[c4f68dc]704 #else
[31a5f418]705 push( freeElem->freeList, *(Heap.Storage *)header );
[9c438546]706 #endif // BUCKETLOCK
[bcb14b5]707 } // if
[c4f68dc]708
709 #ifdef __CFA_DEBUG__
[c1f38e6c]710 __atomic_add_fetch( &allocUnfreed, -size, __ATOMIC_SEQ_CST );
[bcb14b5]711 if ( traceHeap() ) {
[92aca37]712 char helpText[64];
[bcb14b5]713 int len = snprintf( helpText, sizeof(helpText), "Free( %p ) size:%zu\n", addr, size );
[95eb7cf]714 __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
[bcb14b5]715 } // if
[c4f68dc]716 #endif // __CFA_DEBUG__
717} // doFree
718
719
[31a5f418]720size_t prtFree( Heap & manager ) with( manager ) {
[b6830d74]721 size_t total = 0;
[c4f68dc]722 #ifdef __STATISTICS__
[95eb7cf]723 __cfaabi_bits_acquire();
724 __cfaabi_bits_print_nolock( STDERR_FILENO, "\nBin lists (bin size : free blocks on list)\n" );
[c4f68dc]725 #endif // __STATISTICS__
[b6830d74]726 for ( unsigned int i = 0; i < maxBucketsUsed; i += 1 ) {
[d46ed6e]727 size_t size = freeLists[i].blockSize;
728 #ifdef __STATISTICS__
729 unsigned int N = 0;
730 #endif // __STATISTICS__
[b6830d74]731
[9c438546]732 #if BUCKETLOCK == SPINLOCK
[31a5f418]733 for ( Heap.Storage * p = freeLists[i].freeList; p != 0p; p = p->header.kind.real.next ) {
[d46ed6e]734 #else
[b4aa1ab]735 for(;;) {
[31a5f418]736// for ( Heap.Storage * p = top( freeLists[i].freeList ); p != 0p; p = (p)`next->top ) {
737// for ( Heap.Storage * p = top( freeLists[i].freeList ); p != 0p; /* p = getNext( p )->top */) {
738// Heap.Storage * temp = p->header.kind.real.next.top; // FIX ME: direct assignent fails, initialization works`
[7cfef0d]739// typeof(p) temp = (( p )`next)->top; // FIX ME: direct assignent fails, initialization works`
740// p = temp;
[9c438546]741 #endif // BUCKETLOCK
[d46ed6e]742 total += size;
743 #ifdef __STATISTICS__
744 N += 1;
745 #endif // __STATISTICS__
[b6830d74]746 } // for
747
[d46ed6e]748 #ifdef __STATISTICS__
[95eb7cf]749 __cfaabi_bits_print_nolock( STDERR_FILENO, "%7zu, %-7u ", size, N );
750 if ( (i + 1) % 8 == 0 ) __cfaabi_bits_print_nolock( STDERR_FILENO, "\n" );
[d46ed6e]751 #endif // __STATISTICS__
752 } // for
753 #ifdef __STATISTICS__
[95eb7cf]754 __cfaabi_bits_print_nolock( STDERR_FILENO, "\ntotal free blocks:%zu\n", total );
755 __cfaabi_bits_release();
[d46ed6e]756 #endif // __STATISTICS__
757 return (char *)heapEnd - (char *)heapBegin - total;
[95eb7cf]758} // prtFree
759
760
[31a5f418]761static void ?{}( Heap & manager ) with( manager ) {
[ad2dced]762 __page_size = sysconf( _SC_PAGESIZE );
763 __map_prot = PROT_READ | PROT_WRITE | PROT_EXEC;
[95eb7cf]764
765 for ( unsigned int i = 0; i < NoBucketSizes; i += 1 ) { // initialize the free lists
766 freeLists[i].blockSize = bucketSizes[i];
767 } // for
768
769 #ifdef FASTLOOKUP
770 unsigned int idx = 0;
771 for ( unsigned int i = 0; i < LookupSizes; i += 1 ) {
772 if ( i > bucketSizes[idx] ) idx += 1;
773 lookup[i] = idx;
774 } // for
775 #endif // FASTLOOKUP
776
[31a5f418]777 if ( ! setMmapStart( malloc_mmap_start() ) ) {
778 abort( "Heap : internal error, mmap start initialization failure." );
[95eb7cf]779 } // if
[31a5f418]780 heapExpand = malloc_expansion();
[95eb7cf]781
[1e034d9]782 char * end = (char *)sbrk( 0 );
[ad2dced]783 heapBegin = heapEnd = sbrk( (char *)ceiling2( (long unsigned int)end, __page_size ) - end ); // move start of heap to multiple of alignment
[31a5f418]784} // Heap
[95eb7cf]785
786
[31a5f418]787static void ^?{}( Heap & ) {
[95eb7cf]788 #ifdef __STATISTICS__
[baf608a]789 if ( traceHeapTerm() ) {
790 printStats();
[92aca37]791 // prtUnfreed() called in heapAppStop()
[baf608a]792 } // if
[95eb7cf]793 #endif // __STATISTICS__
[31a5f418]794} // ~Heap
[95eb7cf]795
796
797static void memory_startup( void ) __attribute__(( constructor( STARTUP_PRIORITY_MEMORY ) ));
798void memory_startup( void ) {
799 #ifdef __CFA_DEBUG__
[92aca37]800 if ( heapBoot ) { // check for recursion during system boot
[95eb7cf]801 abort( "boot() : internal error, recursively invoked during system boot." );
802 } // if
803 heapBoot = true;
804 #endif // __CFA_DEBUG__
805
[c1f38e6c]806 //verify( heapManager.heapBegin != 0 );
[95eb7cf]807 //heapManager{};
[1076d05]808 if ( heapManager.heapBegin == 0p ) heapManager{}; // sanity check
[95eb7cf]809} // memory_startup
810
811static void memory_shutdown( void ) __attribute__(( destructor( STARTUP_PRIORITY_MEMORY ) ));
812void memory_shutdown( void ) {
813 ^heapManager{};
814} // memory_shutdown
[c4f68dc]815
[bcb14b5]816
817static inline void * mallocNoStats( size_t size ) { // necessary for malloc statistics
[92aca37]818 verify( heapManager.heapBegin != 0p ); // called before memory_startup ?
[dd23e66]819 if ( unlikely( size ) == 0 ) return 0p; // 0 BYTE ALLOCATION RETURNS NULL POINTER
[d5d3a90]820
[76e2113]821#if __SIZEOF_POINTER__ == 8
822 verify( size < ((typeof(size_t))1 << 48) );
823#endif // __SIZEOF_POINTER__ == 8
[d5d3a90]824 return doMalloc( size );
[bcb14b5]825} // mallocNoStats
[c4f68dc]826
827
[92aca37]828static inline void * memalignNoStats( size_t alignment, size_t size ) {
[dd23e66]829 if ( unlikely( size ) == 0 ) return 0p; // 0 BYTE ALLOCATION RETURNS NULL POINTER
[d5d3a90]830
[bcb14b5]831 #ifdef __CFA_DEBUG__
[b6830d74]832 checkAlign( alignment ); // check alignment
[bcb14b5]833 #endif // __CFA_DEBUG__
[c4f68dc]834
[b6830d74]835 // if alignment <= default alignment, do normal malloc as two headers are unnecessary
[bcb14b5]836 if ( unlikely( alignment <= libAlign() ) ) return mallocNoStats( size );
[b6830d74]837
838 // Allocate enough storage to guarantee an address on the alignment boundary, and sufficient space before it for
839 // administrative storage. NOTE, WHILE THERE ARE 2 HEADERS, THE FIRST ONE IS IMPLICITLY CREATED BY DOMALLOC.
840 // .-------------v-----------------v----------------v----------,
841 // | Real Header | ... padding ... | Fake Header | data ... |
842 // `-------------^-----------------^-+--------------^----------'
843 // |<--------------------------------' offset/align |<-- alignment boundary
844
845 // subtract libAlign() because it is already the minimum alignment
846 // add sizeof(Storage) for fake header
[31a5f418]847 char * addr = (char *)mallocNoStats( size + alignment - libAlign() + sizeof(Heap.Storage) );
[b6830d74]848
849 // address in the block of the "next" alignment address
[31a5f418]850 char * user = (char *)ceiling2( (uintptr_t)(addr + sizeof(Heap.Storage)), alignment );
[b6830d74]851
852 // address of header from malloc
[19e5d65d]853 Heap.Storage.Header * RealHeader = HeaderAddr( addr );
854 RealHeader->kind.real.size = size; // correct size to eliminate above alignment offset
[b6830d74]855 // address of fake header * before* the alignment location
[19e5d65d]856 Heap.Storage.Header * fakeHeader = HeaderAddr( user );
[b6830d74]857 // SKULLDUGGERY: insert the offset to the start of the actual storage block and remember alignment
[19e5d65d]858 fakeHeader->kind.fake.offset = (char *)fakeHeader - (char *)RealHeader;
[69ec0fb]859 // SKULLDUGGERY: odd alignment implies fake header
[19e5d65d]860 fakeHeader->kind.fake.alignment = MarkAlignmentBit( alignment );
[b6830d74]861
862 return user;
[bcb14b5]863} // memalignNoStats
[c4f68dc]864
865
[19e5d65d]866//####################### Memory Allocation Routines ####################
867
868
[c4f68dc]869extern "C" {
[61248a4]870 // Allocates size bytes and returns a pointer to the allocated memory. The contents are undefined. If size is 0,
871 // then malloc() returns a unique pointer value that can later be successfully passed to free().
[b6830d74]872 void * malloc( size_t size ) {
[c4f68dc]873 #ifdef __STATISTICS__
[709b812]874 if ( likely( size > 0 ) ) {
[31a5f418]875 __atomic_add_fetch( &stats.malloc_calls, 1, __ATOMIC_SEQ_CST );
876 __atomic_add_fetch( &stats.malloc_storage_request, size, __ATOMIC_SEQ_CST );
[709b812]877 } else {
[31a5f418]878 __atomic_add_fetch( &stats.malloc_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]879 } // if
[c4f68dc]880 #endif // __STATISTICS__
881
[bcb14b5]882 return mallocNoStats( size );
883 } // malloc
[c4f68dc]884
[76e2113]885
[61248a4]886 // Same as malloc() except size bytes is an array of dim elements each of elemSize bytes.
[76e2113]887 void * aalloc( size_t dim, size_t elemSize ) {
[92aca37]888 size_t size = dim * elemSize;
[76e2113]889 #ifdef __STATISTICS__
[709b812]890 if ( likely( size > 0 ) ) {
[31a5f418]891 __atomic_add_fetch( &stats.aalloc_calls, 1, __ATOMIC_SEQ_CST );
892 __atomic_add_fetch( &stats.aalloc_storage_request, size, __ATOMIC_SEQ_CST );
[709b812]893 } else {
[31a5f418]894 __atomic_add_fetch( &stats.aalloc_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]895 } // if
[76e2113]896 #endif // __STATISTICS__
897
[92aca37]898 return mallocNoStats( size );
[76e2113]899 } // aalloc
900
901
[61248a4]902 // Same as aalloc() with memory set to zero.
[76e2113]903 void * calloc( size_t dim, size_t elemSize ) {
[709b812]904 size_t size = dim * elemSize;
905 if ( unlikely( size ) == 0 ) { // 0 BYTE ALLOCATION RETURNS NULL POINTER
906 #ifdef __STATISTICS__
[31a5f418]907 __atomic_add_fetch( &stats.calloc_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]908 #endif // __STATISTICS__
909 return 0p;
910 } // if
[c4f68dc]911 #ifdef __STATISTICS__
[31a5f418]912 __atomic_add_fetch( &stats.calloc_calls, 1, __ATOMIC_SEQ_CST );
913 __atomic_add_fetch( &stats.calloc_storage_request, dim * elemSize, __ATOMIC_SEQ_CST );
[c4f68dc]914 #endif // __STATISTICS__
915
[709b812]916 char * addr = (char *)mallocNoStats( size );
917
[31a5f418]918 Heap.Storage.Header * header;
919 Heap.FreeHeader * freeElem;
[709b812]920 size_t bsize, alignment;
921
922 #ifndef __CFA_DEBUG__
923 bool mapped =
924 #endif // __CFA_DEBUG__
925 headers( "calloc", addr, header, freeElem, bsize, alignment );
926
927 #ifndef __CFA_DEBUG__
928 // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
929 if ( ! mapped )
930 #endif // __CFA_DEBUG__
931 // <-------0000000000000000000000000000UUUUUUUUUUUUUUUUUUUUUUUUU> bsize (bucket size) U => undefined
932 // `-header`-addr `-size
933 memset( addr, '\0', size ); // set to zeros
934
[19e5d65d]935 MarkZeroFilledBit( header ); // mark as zero fill
[709b812]936 return addr;
[bcb14b5]937 } // calloc
[c4f68dc]938
[92aca37]939
[61248a4]940 // Change the size of the memory block pointed to by oaddr to size bytes. The contents are undefined. If oaddr is
941 // 0p, then the call is equivalent to malloc(size), for all values of size; if size is equal to zero, and oaddr is
942 // not 0p, then the call is equivalent to free(oaddr). Unless oaddr is 0p, it must have been returned by an earlier
943 // call to malloc(), alloc(), calloc() or realloc(). If the area pointed to was moved, a free(oaddr) is done.
[cfbc703d]944 void * resize( void * oaddr, size_t size ) {
[709b812]945 // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
946 if ( unlikely( size == 0 ) ) { // special cases
947 #ifdef __STATISTICS__
[31a5f418]948 __atomic_add_fetch( &stats.resize_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]949 #endif // __STATISTICS__
950 free( oaddr );
951 return 0p;
952 } // if
[cfbc703d]953 #ifdef __STATISTICS__
[31a5f418]954 __atomic_add_fetch( &stats.resize_calls, 1, __ATOMIC_SEQ_CST );
[cfbc703d]955 #endif // __STATISTICS__
956
[92aca37]957 if ( unlikely( oaddr == 0p ) ) {
958 #ifdef __STATISTICS__
[31a5f418]959 __atomic_add_fetch( &stats.resize_storage_request, size, __ATOMIC_SEQ_CST );
[92aca37]960 #endif // __STATISTICS__
961 return mallocNoStats( size );
962 } // if
[cfbc703d]963
[31a5f418]964 Heap.Storage.Header * header;
965 Heap.FreeHeader * freeElem;
[92aca37]966 size_t bsize, oalign;
[cfbc703d]967 headers( "resize", oaddr, header, freeElem, bsize, oalign );
[92847f7]968
[19e5d65d]969 size_t odsize = DataStorage( bsize, oaddr, header ); // data storage available in bucket
[cfbc703d]970 // same size, DO NOT preserve STICKY PROPERTIES.
[92847f7]971 if ( oalign == libAlign() && size <= odsize && odsize <= size * 2 ) { // allow 50% wasted storage for smaller size
[19e5d65d]972 ClearZeroFillBit( header ); // no alignment and turn off 0 fill
[d5d3a90]973 header->kind.real.size = size; // reset allocation size
[cfbc703d]974 return oaddr;
975 } // if
[0f89d4f]976
[92aca37]977 #ifdef __STATISTICS__
[31a5f418]978 __atomic_add_fetch( &stats.resize_storage_request, size, __ATOMIC_SEQ_CST );
[92aca37]979 #endif // __STATISTICS__
980
[cfbc703d]981 // change size, DO NOT preserve STICKY PROPERTIES.
982 free( oaddr );
[d5d3a90]983 return mallocNoStats( size ); // create new area
[cfbc703d]984 } // resize
985
986
[61248a4]987 // Same as resize() but the contents are unchanged in the range from the start of the region up to the minimum of
[cfbc703d]988 // the old and new sizes.
[95eb7cf]989 void * realloc( void * oaddr, size_t size ) {
[709b812]990 // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
991 if ( unlikely( size == 0 ) ) { // special cases
992 #ifdef __STATISTICS__
[31a5f418]993 __atomic_add_fetch( &stats.realloc_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]994 #endif // __STATISTICS__
995 free( oaddr );
996 return 0p;
997 } // if
[c4f68dc]998 #ifdef __STATISTICS__
[31a5f418]999 __atomic_add_fetch( &stats.realloc_calls, 1, __ATOMIC_SEQ_CST );
[c4f68dc]1000 #endif // __STATISTICS__
1001
[92aca37]1002 if ( unlikely( oaddr == 0p ) ) {
1003 #ifdef __STATISTICS__
[31a5f418]1004 __atomic_add_fetch( &stats.realloc_storage_request, size, __ATOMIC_SEQ_CST );
[92aca37]1005 #endif // __STATISTICS__
1006 return mallocNoStats( size );
1007 } // if
[c4f68dc]1008
[31a5f418]1009 Heap.Storage.Header * header;
1010 Heap.FreeHeader * freeElem;
[92aca37]1011 size_t bsize, oalign;
[95eb7cf]1012 headers( "realloc", oaddr, header, freeElem, bsize, oalign );
1013
[19e5d65d]1014 size_t odsize = DataStorage( bsize, oaddr, header ); // data storage available in bucket
[d5d3a90]1015 size_t osize = header->kind.real.size; // old allocation size
[19e5d65d]1016 bool ozfill = ZeroFillBit( header ); // old allocation zero filled
[92847f7]1017 if ( unlikely( size <= odsize ) && odsize <= size * 2 ) { // allow up to 50% wasted storage
[d5d3a90]1018 header->kind.real.size = size; // reset allocation size
1019 if ( unlikely( ozfill ) && size > osize ) { // previous request zero fill and larger ?
[e4b6b7d3]1020 memset( (char *)oaddr + osize, '\0', size - osize ); // initialize added storage
[d5d3a90]1021 } // if
[95eb7cf]1022 return oaddr;
[c4f68dc]1023 } // if
1024
[92aca37]1025 #ifdef __STATISTICS__
[31a5f418]1026 __atomic_add_fetch( &stats.realloc_storage_request, size, __ATOMIC_SEQ_CST );
[92aca37]1027 #endif // __STATISTICS__
1028
[95eb7cf]1029 // change size and copy old content to new storage
1030
1031 void * naddr;
[92847f7]1032 if ( likely( oalign == libAlign() ) ) { // previous request not aligned ?
[d5d3a90]1033 naddr = mallocNoStats( size ); // create new area
[c4f68dc]1034 } else {
[d5d3a90]1035 naddr = memalignNoStats( oalign, size ); // create new aligned area
[c4f68dc]1036 } // if
[1e034d9]1037
[95eb7cf]1038 headers( "realloc", naddr, header, freeElem, bsize, oalign );
[47dd0d2]1039 memcpy( naddr, oaddr, min( osize, size ) ); // copy bytes
[95eb7cf]1040 free( oaddr );
[d5d3a90]1041
1042 if ( unlikely( ozfill ) ) { // previous request zero fill ?
[19e5d65d]1043 MarkZeroFilledBit( header ); // mark new request as zero filled
[d5d3a90]1044 if ( size > osize ) { // previous request larger ?
[e4b6b7d3]1045 memset( (char *)naddr + osize, '\0', size - osize ); // initialize added storage
[d5d3a90]1046 } // if
1047 } // if
[95eb7cf]1048 return naddr;
[b6830d74]1049 } // realloc
[c4f68dc]1050
[c1f38e6c]1051
[19e5d65d]1052 // Same as realloc() except the new allocation size is large enough for an array of nelem elements of size elsize.
1053 void * reallocarray( void * oaddr, size_t dim, size_t elemSize ) {
1054 return realloc( oaddr, dim * elemSize );
1055 } // reallocarray
1056
1057
[61248a4]1058 // Same as malloc() except the memory address is a multiple of alignment, which must be a power of two. (obsolete)
[bcb14b5]1059 void * memalign( size_t alignment, size_t size ) {
[c4f68dc]1060 #ifdef __STATISTICS__
[709b812]1061 if ( likely( size > 0 ) ) {
[31a5f418]1062 __atomic_add_fetch( &stats.memalign_calls, 1, __ATOMIC_SEQ_CST );
1063 __atomic_add_fetch( &stats.memalign_storage_request, size, __ATOMIC_SEQ_CST );
[709b812]1064 } else {
[31a5f418]1065 __atomic_add_fetch( &stats.memalign_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]1066 } // if
[c4f68dc]1067 #endif // __STATISTICS__
1068
[95eb7cf]1069 return memalignNoStats( alignment, size );
[bcb14b5]1070 } // memalign
[c4f68dc]1071
[95eb7cf]1072
[76e2113]1073 // Same as aalloc() with memory alignment.
1074 void * amemalign( size_t alignment, size_t dim, size_t elemSize ) {
[92aca37]1075 size_t size = dim * elemSize;
[76e2113]1076 #ifdef __STATISTICS__
[709b812]1077 if ( likely( size > 0 ) ) {
[31a5f418]1078 __atomic_add_fetch( &stats.cmemalign_calls, 1, __ATOMIC_SEQ_CST );
1079 __atomic_add_fetch( &stats.cmemalign_storage_request, size, __ATOMIC_SEQ_CST );
[709b812]1080 } else {
[31a5f418]1081 __atomic_add_fetch( &stats.cmemalign_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]1082 } // if
[76e2113]1083 #endif // __STATISTICS__
1084
[92aca37]1085 return memalignNoStats( alignment, size );
[76e2113]1086 } // amemalign
1087
1088
[ca7949b]1089 // Same as calloc() with memory alignment.
[76e2113]1090 void * cmemalign( size_t alignment, size_t dim, size_t elemSize ) {
[709b812]1091 size_t size = dim * elemSize;
1092 if ( unlikely( size ) == 0 ) { // 0 BYTE ALLOCATION RETURNS NULL POINTER
1093 #ifdef __STATISTICS__
[31a5f418]1094 __atomic_add_fetch( &stats.cmemalign_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]1095 #endif // __STATISTICS__
1096 return 0p;
1097 } // if
[95eb7cf]1098 #ifdef __STATISTICS__
[31a5f418]1099 __atomic_add_fetch( &stats.cmemalign_calls, 1, __ATOMIC_SEQ_CST );
1100 __atomic_add_fetch( &stats.cmemalign_storage_request, dim * elemSize, __ATOMIC_SEQ_CST );
[95eb7cf]1101 #endif // __STATISTICS__
1102
[709b812]1103 char * addr = (char *)memalignNoStats( alignment, size );
1104
[31a5f418]1105 Heap.Storage.Header * header;
1106 Heap.FreeHeader * freeElem;
[709b812]1107 size_t bsize;
1108
1109 #ifndef __CFA_DEBUG__
1110 bool mapped =
1111 #endif // __CFA_DEBUG__
1112 headers( "cmemalign", addr, header, freeElem, bsize, alignment );
1113
1114 // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
1115 #ifndef __CFA_DEBUG__
1116 if ( ! mapped )
1117 #endif // __CFA_DEBUG__
1118 // <-------0000000000000000000000000000UUUUUUUUUUUUUUUUUUUUUUUUU> bsize (bucket size) U => undefined
1119 // `-header`-addr `-size
1120 memset( addr, '\0', size ); // set to zeros
1121
[19e5d65d]1122 MarkZeroFilledBit( header ); // mark as zero filled
[709b812]1123 return addr;
[95eb7cf]1124 } // cmemalign
1125
[13fece5]1126
[ca7949b]1127 // Same as memalign(), but ISO/IEC 2011 C11 Section 7.22.2 states: the value of size shall be an integral multiple
[19e5d65d]1128 // of alignment. This requirement is universally ignored.
[b6830d74]1129 void * aligned_alloc( size_t alignment, size_t size ) {
[c4f68dc]1130 return memalign( alignment, size );
[b6830d74]1131 } // aligned_alloc
[c4f68dc]1132
1133
[ca7949b]1134 // Allocates size bytes and places the address of the allocated memory in *memptr. The address of the allocated
1135 // memory shall be a multiple of alignment, which must be a power of two and a multiple of sizeof(void *). If size
1136 // is 0, then posix_memalign() returns either 0p, or a unique pointer value that can later be successfully passed to
1137 // free(3).
[b6830d74]1138 int posix_memalign( void ** memptr, size_t alignment, size_t size ) {
[69ec0fb]1139 if ( unlikely( alignment < libAlign() || ! is_pow2( alignment ) ) ) return EINVAL; // check alignment
[19e5d65d]1140 *memptr = memalign( alignment, size );
[c4f68dc]1141 return 0;
[b6830d74]1142 } // posix_memalign
[c4f68dc]1143
[13fece5]1144
[ca7949b]1145 // Allocates size bytes and returns a pointer to the allocated memory. The memory address shall be a multiple of the
1146 // page size. It is equivalent to memalign(sysconf(_SC_PAGESIZE),size).
[b6830d74]1147 void * valloc( size_t size ) {
[ad2dced]1148 return memalign( __page_size, size );
[b6830d74]1149 } // valloc
[c4f68dc]1150
1151
[ca7949b]1152 // Same as valloc but rounds size to multiple of page size.
1153 void * pvalloc( size_t size ) {
[19e5d65d]1154 return memalign( __page_size, ceiling2( size, __page_size ) ); // round size to multiple of page size
[ca7949b]1155 } // pvalloc
1156
1157
1158 // Frees the memory space pointed to by ptr, which must have been returned by a previous call to malloc(), calloc()
[1076d05]1159 // or realloc(). Otherwise, or if free(ptr) has already been called before, undefined behaviour occurs. If ptr is
[ca7949b]1160 // 0p, no operation is performed.
[b6830d74]1161 void free( void * addr ) {
[95eb7cf]1162 if ( unlikely( addr == 0p ) ) { // special case
[709b812]1163 #ifdef __STATISTICS__
[31a5f418]1164 __atomic_add_fetch( &stats.free_null_calls, 1, __ATOMIC_SEQ_CST );
[709b812]1165 #endif // __STATISTICS__
1166
[95eb7cf]1167 // #ifdef __CFA_DEBUG__
1168 // if ( traceHeap() ) {
1169 // #define nullmsg "Free( 0x0 ) size:0\n"
[1e034d9]1170 // // Do not debug print free( 0p ), as it can cause recursive entry from sprintf.
[95eb7cf]1171 // __cfaabi_dbg_write( nullmsg, sizeof(nullmsg) - 1 );
1172 // } // if
1173 // #endif // __CFA_DEBUG__
[c4f68dc]1174 return;
1175 } // exit
1176
1177 doFree( addr );
[b6830d74]1178 } // free
[93c2e0a]1179
[c4f68dc]1180
[76e2113]1181 // Returns the alignment of an allocation.
[b6830d74]1182 size_t malloc_alignment( void * addr ) {
[95eb7cf]1183 if ( unlikely( addr == 0p ) ) return libAlign(); // minimum alignment
[19e5d65d]1184 Heap.Storage.Header * header = HeaderAddr( addr );
1185 if ( unlikely( AlignmentBit( header ) ) ) { // fake header ?
1186 return ClearAlignmentBit( header ); // clear flag from value
[c4f68dc]1187 } else {
[cfbc703d]1188 return libAlign(); // minimum alignment
[c4f68dc]1189 } // if
[bcb14b5]1190 } // malloc_alignment
[c4f68dc]1191
[92aca37]1192
[76e2113]1193 // Returns true if the allocation is zero filled, e.g., allocated by calloc().
[b6830d74]1194 bool malloc_zero_fill( void * addr ) {
[95eb7cf]1195 if ( unlikely( addr == 0p ) ) return false; // null allocation is not zero fill
[19e5d65d]1196 Heap.Storage.Header * header = HeaderAddr( addr );
1197 if ( unlikely( AlignmentBit( header ) ) ) { // fake header ?
1198 header = RealHeader( header ); // backup from fake to real header
[c4f68dc]1199 } // if
[19e5d65d]1200 return ZeroFillBit( header ); // zero filled ?
[bcb14b5]1201 } // malloc_zero_fill
[c4f68dc]1202
[19e5d65d]1203
1204 // Returns original total allocation size (not bucket size) => array size is dimension * sizeof(T).
[76e2113]1205 size_t malloc_size( void * addr ) {
[849fb370]1206 if ( unlikely( addr == 0p ) ) return 0; // null allocation has zero size
[19e5d65d]1207 Heap.Storage.Header * header = HeaderAddr( addr );
1208 if ( unlikely( AlignmentBit( header ) ) ) { // fake header ?
1209 header = RealHeader( header ); // backup from fake to real header
[cfbc703d]1210 } // if
[9c438546]1211 return header->kind.real.size;
[76e2113]1212 } // malloc_size
1213
[cfbc703d]1214
[ca7949b]1215 // Returns the number of usable bytes in the block pointed to by ptr, a pointer to a block of memory allocated by
1216 // malloc or a related function.
[95eb7cf]1217 size_t malloc_usable_size( void * addr ) {
1218 if ( unlikely( addr == 0p ) ) return 0; // null allocation has 0 size
[31a5f418]1219 Heap.Storage.Header * header;
1220 Heap.FreeHeader * freeElem;
[95eb7cf]1221 size_t bsize, alignment;
1222
1223 headers( "malloc_usable_size", addr, header, freeElem, bsize, alignment );
[19e5d65d]1224 return DataStorage( bsize, addr, header ); // data storage in bucket
[95eb7cf]1225 } // malloc_usable_size
1226
1227
[ca7949b]1228 // Prints (on default standard error) statistics about memory allocated by malloc and related functions.
[b6830d74]1229 void malloc_stats( void ) {
[c4f68dc]1230 #ifdef __STATISTICS__
[bcb14b5]1231 printStats();
[95eb7cf]1232 if ( prtFree() ) prtFree( heapManager );
[c4f68dc]1233 #endif // __STATISTICS__
[bcb14b5]1234 } // malloc_stats
[c4f68dc]1235
[92aca37]1236
[19e5d65d]1237 // Changes the file descriptor where malloc_stats() writes statistics.
[95eb7cf]1238 int malloc_stats_fd( int fd __attribute__(( unused )) ) {
[c4f68dc]1239 #ifdef __STATISTICS__
[709b812]1240 int temp = stats_fd;
1241 stats_fd = fd;
[bcb14b5]1242 return temp;
[c4f68dc]1243 #else
[19e5d65d]1244 return -1; // unsupported
[c4f68dc]1245 #endif // __STATISTICS__
[bcb14b5]1246 } // malloc_stats_fd
[c4f68dc]1247
[95eb7cf]1248
[19e5d65d]1249 // Prints an XML string that describes the current state of the memory-allocation implementation in the caller.
1250 // The string is printed on the file stream stream. The exported string includes information about all arenas (see
1251 // malloc).
1252 int malloc_info( int options, FILE * stream __attribute__(( unused )) ) {
1253 if ( options != 0 ) { errno = EINVAL; return -1; }
1254 #ifdef __STATISTICS__
1255 return printStatsXML( stream );
1256 #else
1257 return 0; // unsupported
1258 #endif // __STATISTICS__
1259 } // malloc_info
1260
1261
[1076d05]1262 // Adjusts parameters that control the behaviour of the memory-allocation functions (see malloc). The param argument
[ca7949b]1263 // specifies the parameter to be modified, and value specifies the new value for that parameter.
[95eb7cf]1264 int mallopt( int option, int value ) {
[19e5d65d]1265 if ( value < 0 ) return 0;
[95eb7cf]1266 choose( option ) {
1267 case M_TOP_PAD:
[19e5d65d]1268 heapExpand = ceiling2( value, __page_size );
1269 return 1;
[95eb7cf]1270 case M_MMAP_THRESHOLD:
1271 if ( setMmapStart( value ) ) return 1;
[19e5d65d]1272 } // choose
[95eb7cf]1273 return 0; // error, unsupported
1274 } // mallopt
1275
[c1f38e6c]1276
[ca7949b]1277 // Attempt to release free memory at the top of the heap (by calling sbrk with a suitable argument).
[95eb7cf]1278 int malloc_trim( size_t ) {
1279 return 0; // => impossible to release memory
1280 } // malloc_trim
1281
1282
[ca7949b]1283 // Records the current state of all malloc internal bookkeeping variables (but not the actual contents of the heap
1284 // or the state of malloc_hook functions pointers). The state is recorded in a system-dependent opaque data
1285 // structure dynamically allocated via malloc, and a pointer to that data structure is returned as the function
1286 // result. (The caller must free this memory.)
[c4f68dc]1287 void * malloc_get_state( void ) {
[95eb7cf]1288 return 0p; // unsupported
[c4f68dc]1289 } // malloc_get_state
1290
[bcb14b5]1291
[ca7949b]1292 // Restores the state of all malloc internal bookkeeping variables to the values recorded in the opaque data
1293 // structure pointed to by state.
[92aca37]1294 int malloc_set_state( void * ) {
[bcb14b5]1295 return 0; // unsupported
[c4f68dc]1296 } // malloc_set_state
[31a5f418]1297
[19e5d65d]1298
[31a5f418]1299 // Sets the amount (bytes) to extend the heap when there is insufficent free storage to service an allocation.
1300 __attribute__((weak)) size_t malloc_expansion() { return __CFA_DEFAULT_HEAP_EXPANSION__; }
1301
1302 // Sets the crossover point between allocations occuring in the sbrk area or separately mmapped.
1303 __attribute__((weak)) size_t malloc_mmap_start() { return __CFA_DEFAULT_MMAP_START__; }
1304
1305 // Amount subtracted to adjust for unfreed program storage (debug only).
1306 __attribute__((weak)) size_t malloc_unfreed() { return __CFA_DEFAULT_HEAP_UNFREED__; }
[c4f68dc]1307} // extern "C"
1308
1309
[95eb7cf]1310// Must have CFA linkage to overload with C linkage realloc.
[cfbc703d]1311void * resize( void * oaddr, size_t nalign, size_t size ) {
[709b812]1312 // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
1313 if ( unlikely( size == 0 ) ) { // special cases
1314 #ifdef __STATISTICS__
[31a5f418]1315 __atomic_add_fetch( &stats.resize_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]1316 #endif // __STATISTICS__
1317 free( oaddr );
1318 return 0p;
1319 } // if
[95eb7cf]1320
[c86f587]1321 if ( unlikely( nalign < libAlign() ) ) nalign = libAlign(); // reset alignment to minimum
1322 #ifdef __CFA_DEBUG__
[709b812]1323 else checkAlign( nalign ); // check alignment
[c86f587]1324 #endif // __CFA_DEBUG__
1325
[92aca37]1326 if ( unlikely( oaddr == 0p ) ) {
1327 #ifdef __STATISTICS__
[31a5f418]1328 __atomic_add_fetch( &stats.resize_calls, 1, __ATOMIC_SEQ_CST );
1329 __atomic_add_fetch( &stats.resize_storage_request, size, __ATOMIC_SEQ_CST );
[92aca37]1330 #endif // __STATISTICS__
1331 return memalignNoStats( nalign, size );
1332 } // if
[cfbc703d]1333
[92847f7]1334 // Attempt to reuse existing alignment.
[19e5d65d]1335 Heap.Storage.Header * header = HeaderAddr( oaddr );
1336 bool isFakeHeader = AlignmentBit( header ); // old fake header ?
[92847f7]1337 size_t oalign;
[19e5d65d]1338
1339 if ( unlikely( isFakeHeader ) ) {
1340 oalign = ClearAlignmentBit( header ); // old alignment
1341 if ( unlikely( (uintptr_t)oaddr % nalign == 0 // lucky match ?
[92847f7]1342 && ( oalign <= nalign // going down
1343 || (oalign >= nalign && oalign <= 256) ) // little alignment storage wasted ?
[19e5d65d]1344 ) ) {
1345 HeaderAddr( oaddr )->kind.fake.alignment = MarkAlignmentBit( nalign ); // update alignment (could be the same)
[31a5f418]1346 Heap.FreeHeader * freeElem;
[92847f7]1347 size_t bsize, oalign;
1348 headers( "resize", oaddr, header, freeElem, bsize, oalign );
[19e5d65d]1349 size_t odsize = DataStorage( bsize, oaddr, header ); // data storage available in bucket
[a3ade94]1350
[92847f7]1351 if ( size <= odsize && odsize <= size * 2 ) { // allow 50% wasted data storage
[19e5d65d]1352 HeaderAddr( oaddr )->kind.fake.alignment = MarkAlignmentBit( nalign ); // update alignment (could be the same)
1353 ClearZeroFillBit( header ); // turn off 0 fill
[92847f7]1354 header->kind.real.size = size; // reset allocation size
1355 return oaddr;
1356 } // if
[cfbc703d]1357 } // if
[92847f7]1358 } else if ( ! isFakeHeader // old real header (aligned on libAlign) ?
1359 && nalign == libAlign() ) { // new alignment also on libAlign => no fake header needed
[113d785]1360 return resize( oaddr, size ); // duplicate special case checks
[cfbc703d]1361 } // if
1362
[92aca37]1363 #ifdef __STATISTICS__
[31a5f418]1364 __atomic_add_fetch( &stats.resize_storage_request, size, __ATOMIC_SEQ_CST );
[92aca37]1365 #endif // __STATISTICS__
1366
[dd23e66]1367 // change size, DO NOT preserve STICKY PROPERTIES.
[cfbc703d]1368 free( oaddr );
[dd23e66]1369 return memalignNoStats( nalign, size ); // create new aligned area
[cfbc703d]1370} // resize
1371
1372
1373void * realloc( void * oaddr, size_t nalign, size_t size ) {
[709b812]1374 // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
1375 if ( unlikely( size == 0 ) ) { // special cases
1376 #ifdef __STATISTICS__
[31a5f418]1377 __atomic_add_fetch( &stats.realloc_0_calls, 1, __ATOMIC_SEQ_CST );
[709b812]1378 #endif // __STATISTICS__
1379 free( oaddr );
1380 return 0p;
1381 } // if
1382
[c1f38e6c]1383 if ( unlikely( nalign < libAlign() ) ) nalign = libAlign(); // reset alignment to minimum
[cfbc703d]1384 #ifdef __CFA_DEBUG__
[709b812]1385 else checkAlign( nalign ); // check alignment
[cfbc703d]1386 #endif // __CFA_DEBUG__
1387
[c86f587]1388 if ( unlikely( oaddr == 0p ) ) {
1389 #ifdef __STATISTICS__
[31a5f418]1390 __atomic_add_fetch( &stats.realloc_calls, 1, __ATOMIC_SEQ_CST );
1391 __atomic_add_fetch( &stats.realloc_storage_request, size, __ATOMIC_SEQ_CST );
[c86f587]1392 #endif // __STATISTICS__
1393 return memalignNoStats( nalign, size );
1394 } // if
1395
[92847f7]1396 // Attempt to reuse existing alignment.
[19e5d65d]1397 Heap.Storage.Header * header = HeaderAddr( oaddr );
1398 bool isFakeHeader = AlignmentBit( header ); // old fake header ?
[92847f7]1399 size_t oalign;
[19e5d65d]1400 if ( unlikely( isFakeHeader ) ) {
1401 oalign = ClearAlignmentBit( header ); // old alignment
1402 if ( unlikely( (uintptr_t)oaddr % nalign == 0 // lucky match ?
[92847f7]1403 && ( oalign <= nalign // going down
1404 || (oalign >= nalign && oalign <= 256) ) // little alignment storage wasted ?
[19e5d65d]1405 ) ) {
1406 HeaderAddr( oaddr )->kind.fake.alignment = MarkAlignmentBit( nalign ); // update alignment (could be the same)
1407 return realloc( oaddr, size ); // duplicate special case checks
[92847f7]1408 } // if
1409 } else if ( ! isFakeHeader // old real header (aligned on libAlign) ?
[19e5d65d]1410 && nalign == libAlign() ) { // new alignment also on libAlign => no fake header needed
1411 return realloc( oaddr, size ); // duplicate special case checks
1412 } // if
[cfbc703d]1413
[1e034d9]1414 #ifdef __STATISTICS__
[31a5f418]1415 __atomic_add_fetch( &stats.realloc_calls, 1, __ATOMIC_SEQ_CST );
1416 __atomic_add_fetch( &stats.realloc_storage_request, size, __ATOMIC_SEQ_CST );
[1e034d9]1417 #endif // __STATISTICS__
1418
[31a5f418]1419 Heap.FreeHeader * freeElem;
[92847f7]1420 size_t bsize;
1421 headers( "realloc", oaddr, header, freeElem, bsize, oalign );
1422
1423 // change size and copy old content to new storage
1424
[dd23e66]1425 size_t osize = header->kind.real.size; // old allocation size
[19e5d65d]1426 bool ozfill = ZeroFillBit( header ); // old allocation zero filled
[dd23e66]1427
1428 void * naddr = memalignNoStats( nalign, size ); // create new aligned area
[95eb7cf]1429
[1e034d9]1430 headers( "realloc", naddr, header, freeElem, bsize, oalign );
[47dd0d2]1431 memcpy( naddr, oaddr, min( osize, size ) ); // copy bytes
[1e034d9]1432 free( oaddr );
[d5d3a90]1433
1434 if ( unlikely( ozfill ) ) { // previous request zero fill ?
[19e5d65d]1435 MarkZeroFilledBit( header ); // mark new request as zero filled
[d5d3a90]1436 if ( size > osize ) { // previous request larger ?
[e4b6b7d3]1437 memset( (char *)naddr + osize, '\0', size - osize ); // initialize added storage
[d5d3a90]1438 } // if
1439 } // if
[1e034d9]1440 return naddr;
[95eb7cf]1441} // realloc
1442
1443
[c4f68dc]1444// Local Variables: //
1445// tab-width: 4 //
[f8cd310]1446// compile-command: "cfa -nodebug -O2 heap.cfa" //
[c4f68dc]1447// End: //
Note: See TracBrowser for help on using the repository browser.