[b6830d74] | 1 | // #comment TD : this file uses both spaces and tabs for indentation |
---|
| 2 | |
---|
[73abe95] | 3 | // |
---|
[c4f68dc] | 4 | // Cforall Version 1.0.0 Copyright (C) 2017 University of Waterloo |
---|
| 5 | // |
---|
| 6 | // The contents of this file are covered under the licence agreement in the |
---|
| 7 | // file "LICENCE" distributed with Cforall. |
---|
[73abe95] | 8 | // |
---|
| 9 | // heap.c -- |
---|
| 10 | // |
---|
[c4f68dc] | 11 | // Author : Peter A. Buhr |
---|
| 12 | // Created On : Tue Dec 19 21:58:35 2017 |
---|
| 13 | // Last Modified By : Peter A. Buhr |
---|
[93c2e0a] | 14 | // Last Modified On : Sat Aug 11 08:22:16 2018 |
---|
| 15 | // Update Count : 495 |
---|
[73abe95] | 16 | // |
---|
[c4f68dc] | 17 | |
---|
| 18 | #include <unistd.h> // sbrk, sysconf |
---|
| 19 | #include <stdbool.h> // true, false |
---|
| 20 | #include <stdio.h> // snprintf, fileno |
---|
| 21 | #include <errno.h> // errno |
---|
| 22 | extern "C" { |
---|
| 23 | #include <sys/mman.h> // mmap, munmap |
---|
| 24 | } // extern "C" |
---|
| 25 | |
---|
[b6830d74] | 26 | // #comment TD : Many of these should be merged into math I believe |
---|
[73abe95] | 27 | #include "bits/align.hfa" // libPow2 |
---|
| 28 | #include "bits/defs.hfa" // likely, unlikely |
---|
| 29 | #include "bits/locks.hfa" // __spinlock_t |
---|
| 30 | #include "startup.hfa" // STARTUP_PRIORITY_MEMORY |
---|
[58b6d1b] | 31 | #include "stdlib.hfa" // bsearchl |
---|
[c4f68dc] | 32 | #include "malloc.h" |
---|
| 33 | |
---|
| 34 | |
---|
| 35 | enum { |
---|
| 36 | __CFA_DEFAULT_MMAP_START__ = (512 * 1024 + 1), |
---|
| 37 | __CFA_DEFAULT_HEAP_EXPANSION__ = (1 * 1024 * 1024), |
---|
| 38 | }; |
---|
| 39 | |
---|
| 40 | size_t default_mmap_start() __attribute__(( weak )) { |
---|
[b6830d74] | 41 | return __CFA_DEFAULT_MMAP_START__; |
---|
[c4f68dc] | 42 | } // default_mmap_start |
---|
| 43 | |
---|
| 44 | size_t default_heap_expansion() __attribute__(( weak )) { |
---|
[b6830d74] | 45 | return __CFA_DEFAULT_HEAP_EXPANSION__; |
---|
[c4f68dc] | 46 | } // default_heap_expansion |
---|
| 47 | |
---|
| 48 | |
---|
| 49 | // supported mallopt options |
---|
| 50 | #ifndef M_MMAP_THRESHOLD |
---|
| 51 | #define M_MMAP_THRESHOLD (-1) |
---|
| 52 | #endif // M_TOP_PAD |
---|
| 53 | #ifndef M_TOP_PAD |
---|
| 54 | #define M_TOP_PAD (-2) |
---|
| 55 | #endif // M_TOP_PAD |
---|
| 56 | |
---|
| 57 | #define FASTLOOKUP |
---|
| 58 | #define __STATISTICS__ |
---|
| 59 | |
---|
| 60 | #define SPINLOCK 0 |
---|
| 61 | #define LOCKFREE 1 |
---|
| 62 | #define BUCKETLOCK SPINLOCK |
---|
| 63 | #if BUCKETLOCK == LOCKFREE |
---|
| 64 | #include <uStackLF.h> |
---|
| 65 | #endif // LOCKFREE |
---|
| 66 | |
---|
[b6830d74] | 67 | // #comment TD : This defined is significantly different from the __ALIGN__ define from locks.hfa |
---|
[c4f68dc] | 68 | #define ALIGN 16 |
---|
| 69 | |
---|
| 70 | // enum { NoBucketSizes = 93, // number of buckets sizes |
---|
| 71 | // #ifdef FASTLOOKUP |
---|
| 72 | // LookupSizes = 65536, // number of fast lookup sizes |
---|
| 73 | // #endif // FASTLOOKUP |
---|
| 74 | // }; |
---|
| 75 | #define NoBucketSizes 93 // number of buckets sizes |
---|
| 76 | #ifdef FASTLOOKUP |
---|
| 77 | #define LookupSizes 65536 // number of fast lookup sizes |
---|
| 78 | #endif // FASTLOOKUP |
---|
| 79 | |
---|
| 80 | |
---|
[93c2e0a] | 81 | static bool traceHeap = false; |
---|
[d46ed6e] | 82 | |
---|
[93c2e0a] | 83 | inline bool traceHeap() { |
---|
[d46ed6e] | 84 | return traceHeap; |
---|
| 85 | } // traceHeap |
---|
| 86 | |
---|
[93c2e0a] | 87 | bool traceHeapOn() { |
---|
| 88 | bool temp = traceHeap; |
---|
[d46ed6e] | 89 | traceHeap = true; |
---|
| 90 | return temp; |
---|
| 91 | } // traceHeapOn |
---|
| 92 | |
---|
[93c2e0a] | 93 | bool traceHeapOff() { |
---|
| 94 | bool temp = traceHeap; |
---|
[d46ed6e] | 95 | traceHeap = false; |
---|
| 96 | return temp; |
---|
| 97 | } // traceHeapOff |
---|
| 98 | |
---|
| 99 | |
---|
[93c2e0a] | 100 | static bool checkFree = false; |
---|
[d46ed6e] | 101 | |
---|
[93c2e0a] | 102 | inline bool checkFree() { |
---|
[5d4fa18] | 103 | return checkFree; |
---|
| 104 | } // checkFree |
---|
| 105 | |
---|
[93c2e0a] | 106 | bool checkFreeOn() { |
---|
| 107 | bool temp = checkFree; |
---|
[5d4fa18] | 108 | checkFree = true; |
---|
| 109 | return temp; |
---|
| 110 | } // checkFreeOn |
---|
| 111 | |
---|
[93c2e0a] | 112 | bool checkFreeOff() { |
---|
| 113 | bool temp = checkFree; |
---|
[5d4fa18] | 114 | checkFree = false; |
---|
| 115 | return temp; |
---|
| 116 | } // checkFreeOff |
---|
| 117 | |
---|
| 118 | |
---|
[93c2e0a] | 119 | // static bool traceHeapTerm = false; |
---|
[5d4fa18] | 120 | |
---|
[93c2e0a] | 121 | // inline bool traceHeapTerm() { |
---|
[5d4fa18] | 122 | // return traceHeapTerm; |
---|
| 123 | // } // traceHeapTerm |
---|
[d46ed6e] | 124 | |
---|
[93c2e0a] | 125 | // bool traceHeapTermOn() { |
---|
| 126 | // bool temp = traceHeapTerm; |
---|
[5d4fa18] | 127 | // traceHeapTerm = true; |
---|
[d46ed6e] | 128 | // return temp; |
---|
[5d4fa18] | 129 | // } // traceHeapTermOn |
---|
[d46ed6e] | 130 | |
---|
[93c2e0a] | 131 | // bool traceHeapTermOff() { |
---|
| 132 | // bool temp = traceHeapTerm; |
---|
[5d4fa18] | 133 | // traceHeapTerm = false; |
---|
[d46ed6e] | 134 | // return temp; |
---|
[5d4fa18] | 135 | // } // traceHeapTermOff |
---|
[d46ed6e] | 136 | |
---|
| 137 | |
---|
[f0b3f51] | 138 | #ifdef __CFA_DEBUG__ |
---|
[93c2e0a] | 139 | static unsigned int allocFree; // running total of allocations minus frees |
---|
[d46ed6e] | 140 | |
---|
| 141 | static void checkUnfreed() { |
---|
[b6830d74] | 142 | if ( allocFree != 0 ) { |
---|
[d46ed6e] | 143 | // DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT. |
---|
| 144 | // char helpText[512]; |
---|
[93c2e0a] | 145 | // int len = snprintf( helpText, sizeof(helpText), "CFA warning (UNIX pid:%ld) : program terminating with %u(0x%x) bytes of storage allocated but not freed.\n" |
---|
[d46ed6e] | 146 | // "Possible cause is unfreed storage allocated by the program or system/library routines called from the program.\n", |
---|
[93c2e0a] | 147 | // (long int)getpid(), allocFree, allocFree ); // always print the UNIX pid |
---|
[d46ed6e] | 148 | // __cfaabi_dbg_bits_write( helpText, len ); |
---|
[b6830d74] | 149 | } // if |
---|
[d46ed6e] | 150 | } // checkUnfreed |
---|
| 151 | |
---|
| 152 | extern "C" { |
---|
| 153 | void heapAppStart() { // called by __cfaabi_appready_startup |
---|
[93c2e0a] | 154 | allocFree = 0; |
---|
[d46ed6e] | 155 | } // heapAppStart |
---|
| 156 | |
---|
| 157 | void heapAppStop() { // called by __cfaabi_appready_startdown |
---|
[93c2e0a] | 158 | fclose( stdin ); fclose( stdout ); |
---|
[d46ed6e] | 159 | checkUnfreed(); |
---|
| 160 | } // heapAppStop |
---|
| 161 | } // extern "C" |
---|
| 162 | #endif // __CFA_DEBUG__ |
---|
| 163 | |
---|
| 164 | |
---|
[c4f68dc] | 165 | struct HeapManager { |
---|
| 166 | // struct FreeHeader; // forward declaration |
---|
| 167 | |
---|
| 168 | struct Storage { |
---|
| 169 | struct Header { // header |
---|
| 170 | union Kind { |
---|
| 171 | struct RealHeader { |
---|
| 172 | union { |
---|
[b6830d74] | 173 | // #comment TD : this code use byte size but the comment uses bit size |
---|
| 174 | |
---|
[c4f68dc] | 175 | struct { // 32-bit word => 64-bit header, 64-bit word => 128-bit header |
---|
[f0b3f51] | 176 | #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4 |
---|
[c4f68dc] | 177 | uint32_t padding; // unused, force home/blocksize to overlay alignment in fake header |
---|
[f0b3f51] | 178 | #endif // __ORDER_BIG_ENDIAN__ && __U_WORDSIZE__ == 32 |
---|
[c4f68dc] | 179 | |
---|
| 180 | union { |
---|
| 181 | // FreeHeader * home; // allocated block points back to home locations (must overlay alignment) |
---|
| 182 | void * home; // allocated block points back to home locations (must overlay alignment) |
---|
| 183 | size_t blockSize; // size for munmap (must overlay alignment) |
---|
| 184 | #if BUCKLOCK == SPINLOCK |
---|
| 185 | Storage * next; // freed block points next freed block of same size |
---|
| 186 | #endif // SPINLOCK |
---|
| 187 | }; |
---|
| 188 | |
---|
[f0b3f51] | 189 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4 |
---|
[c4f68dc] | 190 | uint32_t padding; // unused, force home/blocksize to overlay alignment in fake header |
---|
[f0b3f51] | 191 | #endif // __ORDER_LITTLE_ENDIAN__ && __U_WORDSIZE__ == 32 |
---|
[c4f68dc] | 192 | |
---|
| 193 | }; |
---|
[b6830d74] | 194 | |
---|
| 195 | // #comment TD : C++ code |
---|
[c4f68dc] | 196 | #if BUCKLOCK == LOCKFREE |
---|
| 197 | Stack<Storage>::Link next; // freed block points next freed block of same size (double-wide) |
---|
| 198 | #endif // LOCKFREE |
---|
| 199 | }; |
---|
[93c2e0a] | 200 | } real; // RealHeader |
---|
[c4f68dc] | 201 | struct FakeHeader { |
---|
| 202 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
---|
| 203 | uint32_t alignment; // low-order bits of home/blockSize used for tricks |
---|
[f0b3f51] | 204 | #endif // __ORDER_LITTLE_ENDIAN__ |
---|
[c4f68dc] | 205 | |
---|
| 206 | uint32_t offset; |
---|
| 207 | |
---|
| 208 | #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
---|
| 209 | uint32_t alignment; // low-order bits of home/blockSize used for tricks |
---|
[f0b3f51] | 210 | #endif // __ORDER_BIG_ENDIAN__ |
---|
[93c2e0a] | 211 | } fake; // FakeHeader |
---|
| 212 | } kind; // Kind |
---|
[c4f68dc] | 213 | } header; // Header |
---|
| 214 | char pad[ALIGN - sizeof( Header )]; |
---|
| 215 | char data[0]; // storage |
---|
| 216 | }; // Storage |
---|
| 217 | |
---|
| 218 | static_assert( ALIGN >= sizeof( Storage ), "ALIGN < sizeof( Storage )" ); |
---|
| 219 | |
---|
| 220 | struct FreeHeader { |
---|
| 221 | #if BUCKLOCK == SPINLOCK |
---|
| 222 | __spinlock_t lock; // must be first field for alignment |
---|
| 223 | Storage * freeList; |
---|
| 224 | #elif BUCKLOCK == LOCKFREE |
---|
[b6830d74] | 225 | // #comment TD : C++ code |
---|
[c4f68dc] | 226 | StackLF<Storage> freeList; |
---|
| 227 | #else |
---|
| 228 | #error undefined lock type for bucket lock |
---|
| 229 | #endif // SPINLOCK |
---|
| 230 | size_t blockSize; // size of allocations on this list |
---|
| 231 | }; // FreeHeader |
---|
| 232 | |
---|
| 233 | // must be first fields for alignment |
---|
| 234 | __spinlock_t extlock; // protects allocation-buffer extension |
---|
| 235 | FreeHeader freeLists[NoBucketSizes]; // buckets for different allocation sizes |
---|
| 236 | |
---|
| 237 | void * heapBegin; // start of heap |
---|
| 238 | void * heapEnd; // logical end of heap |
---|
| 239 | size_t heapRemaining; // amount of storage not allocated in the current chunk |
---|
| 240 | }; // HeapManager |
---|
| 241 | |
---|
[5d4fa18] | 242 | static inline size_t getKey( const HeapManager.FreeHeader & freeheader ) { return freeheader.blockSize; } |
---|
| 243 | // statically allocated variables => zero filled. |
---|
| 244 | |
---|
| 245 | |
---|
| 246 | static size_t pageSize; // architecture pagesize |
---|
| 247 | static size_t heapExpand; // sbrk advance |
---|
| 248 | static size_t mmapStart; // cross over point for mmap |
---|
| 249 | static unsigned int maxBucketsUsed; // maximum number of buckets in use |
---|
| 250 | |
---|
[b6830d74] | 251 | // #comment TD : This array is not const but it feels like it should be |
---|
[5d4fa18] | 252 | // Powers of 2 are common allocation sizes, so make powers of 2 generate the minimum required size. |
---|
| 253 | static unsigned int bucketSizes[NoBucketSizes] @= { // different bucket sizes |
---|
[b6830d74] | 254 | 16, 32, 48, 64, |
---|
| 255 | 64 + sizeof(HeapManager.Storage), 96, 112, 128, 128 + sizeof(HeapManager.Storage), 160, 192, 224, |
---|
| 256 | 256 + sizeof(HeapManager.Storage), 320, 384, 448, 512 + sizeof(HeapManager.Storage), 640, 768, 896, |
---|
| 257 | 1_024 + sizeof(HeapManager.Storage), 1_536, 2_048 + sizeof(HeapManager.Storage), 2_560, 3_072, 3_584, 4_096 + sizeof(HeapManager.Storage), 6_144, |
---|
| 258 | 8_192 + sizeof(HeapManager.Storage), 9_216, 10_240, 11_264, 12_288, 13_312, 14_336, 15_360, |
---|
| 259 | 16_384 + sizeof(HeapManager.Storage), 18_432, 20_480, 22_528, 24_576, 26_624, 28_672, 30_720, |
---|
| 260 | 32_768 + sizeof(HeapManager.Storage), 36_864, 40_960, 45_056, 49_152, 53_248, 57_344, 61_440, |
---|
| 261 | 65_536 + sizeof(HeapManager.Storage), 73_728, 81_920, 90_112, 98_304, 106_496, 114_688, 122_880, |
---|
| 262 | 131_072 + sizeof(HeapManager.Storage), 147_456, 163_840, 180_224, 196_608, 212_992, 229_376, 245_760, |
---|
| 263 | 262_144 + sizeof(HeapManager.Storage), 294_912, 327_680, 360_448, 393_216, 425_984, 458_752, 491_520, |
---|
| 264 | 524_288 + sizeof(HeapManager.Storage), 655_360, 786_432, 917_504, 1_048_576 + sizeof(HeapManager.Storage), 1_179_648, 1_310_720, 1_441_792, |
---|
| 265 | 1_572_864, 1_703_936, 1_835_008, 1_966_080, 2_097_152 + sizeof(HeapManager.Storage), 2_621_440, 3_145_728, 3_670_016, |
---|
| 266 | 4_194_304 + sizeof(HeapManager.Storage) |
---|
[5d4fa18] | 267 | }; |
---|
| 268 | #ifdef FASTLOOKUP |
---|
| 269 | static unsigned char lookup[LookupSizes]; // O(1) lookup for small sizes |
---|
| 270 | #endif // FASTLOOKUP |
---|
| 271 | static int mmapFd = -1; // fake or actual fd for anonymous file |
---|
| 272 | |
---|
| 273 | |
---|
| 274 | #ifdef __CFA_DEBUG__ |
---|
[93c2e0a] | 275 | static bool heapBoot = 0; // detect recursion during boot |
---|
[5d4fa18] | 276 | #endif // __CFA_DEBUG__ |
---|
| 277 | static HeapManager heapManager __attribute__(( aligned (128) )) @= {}; // size of cache line to prevent false sharing |
---|
| 278 | |
---|
[b6830d74] | 279 | // #comment TD : The return type of this function should be commented |
---|
[93c2e0a] | 280 | static inline bool setMmapStart( size_t value ) { |
---|
[b6830d74] | 281 | if ( value < pageSize || bucketSizes[NoBucketSizes - 1] < value ) return true; |
---|
| 282 | mmapStart = value; // set global |
---|
| 283 | |
---|
| 284 | // find the closest bucket size less than or equal to the mmapStart size |
---|
| 285 | maxBucketsUsed = bsearchl( (unsigned int)mmapStart, bucketSizes, NoBucketSizes ); // binary search |
---|
| 286 | assert( maxBucketsUsed < NoBucketSizes ); // subscript failure ? |
---|
| 287 | assert( mmapStart <= bucketSizes[maxBucketsUsed] ); // search failure ? |
---|
| 288 | return false; |
---|
[d46ed6e] | 289 | } // setMmapStart |
---|
| 290 | |
---|
| 291 | |
---|
| 292 | static void ?{}( HeapManager & manager ) with ( manager ) { |
---|
[b6830d74] | 293 | pageSize = sysconf( _SC_PAGESIZE ); |
---|
[73abe95] | 294 | |
---|
[b6830d74] | 295 | for ( unsigned int i = 0; i < NoBucketSizes; i += 1 ) { // initialize the free lists |
---|
[d46ed6e] | 296 | freeLists[i].blockSize = bucketSizes[i]; |
---|
[b6830d74] | 297 | } // for |
---|
[d46ed6e] | 298 | |
---|
| 299 | #ifdef FASTLOOKUP |
---|
[b6830d74] | 300 | unsigned int idx = 0; |
---|
| 301 | for ( unsigned int i = 0; i < LookupSizes; i += 1 ) { |
---|
[d46ed6e] | 302 | if ( i > bucketSizes[idx] ) idx += 1; |
---|
| 303 | lookup[i] = idx; |
---|
[b6830d74] | 304 | } // for |
---|
[d46ed6e] | 305 | #endif // FASTLOOKUP |
---|
| 306 | |
---|
[b6830d74] | 307 | if ( setMmapStart( default_mmap_start() ) ) { |
---|
[d46ed6e] | 308 | abort( "HeapManager : internal error, mmap start initialization failure." ); |
---|
[b6830d74] | 309 | } // if |
---|
| 310 | heapExpand = default_heap_expansion(); |
---|
[d46ed6e] | 311 | |
---|
[b6830d74] | 312 | char * End = (char *)sbrk( 0 ); |
---|
| 313 | sbrk( (char *)libCeiling( (long unsigned int)End, libAlign() ) - End ); // move start of heap to multiple of alignment |
---|
| 314 | heapBegin = heapEnd = sbrk( 0 ); // get new start point |
---|
[d46ed6e] | 315 | } // HeapManager |
---|
| 316 | |
---|
| 317 | |
---|
| 318 | static void ^?{}( HeapManager & ) { |
---|
| 319 | #ifdef __STATISTICS__ |
---|
[5d4fa18] | 320 | // if ( traceHeapTerm() ) { |
---|
[d46ed6e] | 321 | // printStats(); |
---|
[5d4fa18] | 322 | // if ( checkfree() ) checkFree( heapManager, true ); |
---|
[d46ed6e] | 323 | // } // if |
---|
| 324 | #endif // __STATISTICS__ |
---|
| 325 | } // ~HeapManager |
---|
| 326 | |
---|
| 327 | |
---|
| 328 | static void memory_startup( void ) __attribute__(( constructor( STARTUP_PRIORITY_MEMORY ) )); |
---|
| 329 | void memory_startup( void ) { |
---|
| 330 | #ifdef __CFA_DEBUG__ |
---|
| 331 | if ( unlikely( heapBoot ) ) { // check for recursion during system boot |
---|
| 332 | // DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT. |
---|
| 333 | abort( "boot() : internal error, recursively invoked during system boot." ); |
---|
| 334 | } // if |
---|
| 335 | heapBoot = true; |
---|
| 336 | #endif // __CFA_DEBUG__ |
---|
[c4f68dc] | 337 | |
---|
[b6830d74] | 338 | // #comment TD : This assertion seems redundent with the above code |
---|
[f0b3f51] | 339 | assert( heapManager.heapBegin == 0 ); |
---|
[d46ed6e] | 340 | heapManager{}; |
---|
| 341 | } // memory_startup |
---|
[c4f68dc] | 342 | |
---|
[d46ed6e] | 343 | static void memory_shutdown( void ) __attribute__(( destructor( STARTUP_PRIORITY_MEMORY ) )); |
---|
| 344 | void memory_shutdown( void ) { |
---|
| 345 | ^heapManager{}; |
---|
| 346 | } // memory_shutdown |
---|
| 347 | |
---|
[c4f68dc] | 348 | |
---|
| 349 | #ifdef __STATISTICS__ |
---|
[d46ed6e] | 350 | static unsigned long long int mmap_storage; // heap statistics counters |
---|
[c4f68dc] | 351 | static unsigned int mmap_calls; |
---|
| 352 | static unsigned long long int munmap_storage; |
---|
| 353 | static unsigned int munmap_calls; |
---|
| 354 | static unsigned long long int sbrk_storage; |
---|
| 355 | static unsigned int sbrk_calls; |
---|
| 356 | static unsigned long long int malloc_storage; |
---|
| 357 | static unsigned int malloc_calls; |
---|
| 358 | static unsigned long long int free_storage; |
---|
| 359 | static unsigned int free_calls; |
---|
| 360 | static unsigned long long int calloc_storage; |
---|
| 361 | static unsigned int calloc_calls; |
---|
| 362 | static unsigned long long int memalign_storage; |
---|
| 363 | static unsigned int memalign_calls; |
---|
| 364 | static unsigned long long int cmemalign_storage; |
---|
| 365 | static unsigned int cmemalign_calls; |
---|
| 366 | static unsigned long long int realloc_storage; |
---|
| 367 | static unsigned int realloc_calls; |
---|
[d46ed6e] | 368 | |
---|
| 369 | static int statfd; // statistics file descriptor (changed by malloc_stats_fd) |
---|
[c4f68dc] | 370 | |
---|
| 371 | |
---|
| 372 | // Use "write" because streams may be shutdown when calls are made. |
---|
[d46ed6e] | 373 | static void printStats() { |
---|
[b6830d74] | 374 | char helpText[512]; |
---|
[93c2e0a] | 375 | __cfaabi_dbg_bits_print_buffer( helpText, sizeof(helpText), |
---|
[5d4fa18] | 376 | "\nHeap statistics:\n" |
---|
| 377 | " malloc: calls %u / storage %llu\n" |
---|
| 378 | " calloc: calls %u / storage %llu\n" |
---|
| 379 | " memalign: calls %u / storage %llu\n" |
---|
| 380 | " cmemalign: calls %u / storage %llu\n" |
---|
| 381 | " realloc: calls %u / storage %llu\n" |
---|
| 382 | " free: calls %u / storage %llu\n" |
---|
| 383 | " mmap: calls %u / storage %llu\n" |
---|
| 384 | " munmap: calls %u / storage %llu\n" |
---|
| 385 | " sbrk: calls %u / storage %llu\n", |
---|
| 386 | malloc_calls, malloc_storage, |
---|
| 387 | calloc_calls, calloc_storage, |
---|
| 388 | memalign_calls, memalign_storage, |
---|
| 389 | cmemalign_calls, cmemalign_storage, |
---|
| 390 | realloc_calls, realloc_storage, |
---|
| 391 | free_calls, free_storage, |
---|
| 392 | mmap_calls, mmap_storage, |
---|
| 393 | munmap_calls, munmap_storage, |
---|
| 394 | sbrk_calls, sbrk_storage |
---|
[c4f68dc] | 395 | ); |
---|
[d46ed6e] | 396 | } // printStats |
---|
[c4f68dc] | 397 | |
---|
[b6830d74] | 398 | // #comment TD : Why do we have this? |
---|
[d46ed6e] | 399 | static int printStatsXML( FILE * stream ) { |
---|
[b6830d74] | 400 | char helpText[512]; |
---|
| 401 | int len = snprintf( helpText, sizeof(helpText), |
---|
[c4f68dc] | 402 | "<malloc version=\"1\">\n" |
---|
| 403 | "<heap nr=\"0\">\n" |
---|
| 404 | "<sizes>\n" |
---|
| 405 | "</sizes>\n" |
---|
| 406 | "<total type=\"malloc\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 407 | "<total type=\"calloc\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 408 | "<total type=\"memalign\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 409 | "<total type=\"cmemalign\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 410 | "<total type=\"realloc\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 411 | "<total type=\"free\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 412 | "<total type=\"mmap\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 413 | "<total type=\"munmap\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 414 | "<total type=\"sbrk\" count=\"%u\" size=\"%llu\"/>\n" |
---|
| 415 | "</malloc>", |
---|
| 416 | malloc_calls, malloc_storage, |
---|
| 417 | calloc_calls, calloc_storage, |
---|
| 418 | memalign_calls, memalign_storage, |
---|
| 419 | cmemalign_calls, cmemalign_storage, |
---|
| 420 | realloc_calls, realloc_storage, |
---|
| 421 | free_calls, free_storage, |
---|
| 422 | mmap_calls, mmap_storage, |
---|
| 423 | munmap_calls, munmap_storage, |
---|
| 424 | sbrk_calls, sbrk_storage |
---|
| 425 | ); |
---|
[b6830d74] | 426 | return write( fileno( stream ), helpText, len ); // -1 => error |
---|
[d46ed6e] | 427 | } // printStatsXML |
---|
[c4f68dc] | 428 | #endif // __STATISTICS__ |
---|
| 429 | |
---|
[b6830d74] | 430 | // #comment TD : Is this the samething as Out-of-Memory? |
---|
[c4f68dc] | 431 | static inline void noMemory() { |
---|
[b6830d74] | 432 | abort( "Heap memory exhausted at %zu bytes.\n" |
---|
[c4f68dc] | 433 | "Possible cause is very large memory allocation and/or large amount of unfreed storage allocated by the program or system/library routines.", |
---|
| 434 | ((char *)(sbrk( 0 )) - (char *)(heapManager.heapBegin)) ); |
---|
| 435 | } // noMemory |
---|
| 436 | |
---|
| 437 | |
---|
| 438 | static inline void checkAlign( size_t alignment ) { |
---|
[b6830d74] | 439 | if ( alignment < sizeof(void *) || ! libPow2( alignment ) ) { |
---|
[c4f68dc] | 440 | abort( "Alignment %zu for memory allocation is less than sizeof(void *) and/or not a power of 2.", alignment ); |
---|
[b6830d74] | 441 | } // if |
---|
[c4f68dc] | 442 | } // checkAlign |
---|
| 443 | |
---|
| 444 | |
---|
[93c2e0a] | 445 | static inline bool setHeapExpand( size_t value ) { |
---|
[b6830d74] | 446 | if ( heapExpand < pageSize ) return true; |
---|
| 447 | heapExpand = value; |
---|
| 448 | return false; |
---|
[c4f68dc] | 449 | } // setHeapExpand |
---|
| 450 | |
---|
| 451 | |
---|
[93c2e0a] | 452 | static inline void checkHeader( bool check, const char * name, void * addr ) { |
---|
[b6830d74] | 453 | if ( unlikely( check ) ) { // bad address ? |
---|
[c4f68dc] | 454 | abort( "Attempt to %s storage %p with address outside the heap.\n" |
---|
| 455 | "Possible cause is duplicate free on same block or overwriting of memory.", |
---|
| 456 | name, addr ); |
---|
[b6830d74] | 457 | } // if |
---|
[c4f68dc] | 458 | } // checkHeader |
---|
| 459 | |
---|
[b6830d74] | 460 | // #comment TD : function should be commented and/or have a more evocative name |
---|
| 461 | // this isn't either a check or a constructor which is what I would expect this function to be |
---|
[c4f68dc] | 462 | static inline void fakeHeader( HeapManager.Storage.Header *& header, size_t & size, size_t & alignment ) { |
---|
[b6830d74] | 463 | if ( unlikely( (header->kind.fake.alignment & 1) == 1 ) ) { // fake header ? |
---|
[c4f68dc] | 464 | size_t offset = header->kind.fake.offset; |
---|
| 465 | alignment = header->kind.fake.alignment & -2; // remove flag from value |
---|
| 466 | #ifdef __CFA_DEBUG__ |
---|
| 467 | checkAlign( alignment ); // check alignment |
---|
| 468 | #endif // __CFA_DEBUG__ |
---|
| 469 | header = (HeapManager.Storage.Header *)((char *)header - offset); |
---|
[b6830d74] | 470 | } // if |
---|
[c4f68dc] | 471 | } // fakeHeader |
---|
| 472 | |
---|
[b6830d74] | 473 | // #comment TD : Why is this a define |
---|
[c4f68dc] | 474 | #define headerAddr( addr ) ((HeapManager.Storage.Header *)( (char *)addr - sizeof(HeapManager.Storage) )) |
---|
| 475 | |
---|
[93c2e0a] | 476 | static inline bool headers( const char * name, void * addr, HeapManager.Storage.Header *& header, HeapManager.FreeHeader *& freeElem, size_t & size, size_t & alignment ) with ( heapManager ) { |
---|
[b6830d74] | 477 | header = headerAddr( addr ); |
---|
[c4f68dc] | 478 | |
---|
[b6830d74] | 479 | if ( unlikely( heapEnd < addr ) ) { // mmapped ? |
---|
[c4f68dc] | 480 | fakeHeader( header, size, alignment ); |
---|
| 481 | size = header->kind.real.blockSize & -3; // mmap size |
---|
| 482 | return true; |
---|
[b6830d74] | 483 | } // if |
---|
[c4f68dc] | 484 | |
---|
| 485 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 486 | checkHeader( addr < heapBegin || header < (HeapManager.Storage.Header *)heapBegin, name, addr ); // bad low address ? |
---|
[c4f68dc] | 487 | #endif // __CFA_DEBUG__ |
---|
[b6830d74] | 488 | |
---|
| 489 | // #comment TD : This code looks weird... |
---|
| 490 | // It's called as the first statement of both branches of the last if, with the same parameters in all cases |
---|
| 491 | |
---|
| 492 | // header may be safe to dereference |
---|
| 493 | fakeHeader( header, size, alignment ); |
---|
[c4f68dc] | 494 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 495 | checkHeader( header < (HeapManager.Storage.Header *)heapBegin || (HeapManager.Storage.Header *)heapEnd < header, name, addr ); // bad address ? (offset could be + or -) |
---|
[c4f68dc] | 496 | #endif // __CFA_DEBUG__ |
---|
| 497 | |
---|
[b6830d74] | 498 | freeElem = (HeapManager.FreeHeader *)((size_t)header->kind.real.home & -3); |
---|
[c4f68dc] | 499 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 500 | if ( freeElem < &freeLists[0] || &freeLists[NoBucketSizes] <= freeElem ) { |
---|
| 501 | abort( "Attempt to %s storage %p with corrupted header.\n" |
---|
| 502 | "Possible cause is duplicate free on same block or overwriting of header information.", |
---|
| 503 | name, addr ); |
---|
| 504 | } // if |
---|
[c4f68dc] | 505 | #endif // __CFA_DEBUG__ |
---|
[b6830d74] | 506 | size = freeElem->blockSize; |
---|
| 507 | return false; |
---|
[c4f68dc] | 508 | } // headers |
---|
| 509 | |
---|
| 510 | |
---|
| 511 | static inline void * extend( size_t size ) with ( heapManager ) { |
---|
[b6830d74] | 512 | lock( extlock __cfaabi_dbg_ctx2 ); |
---|
| 513 | ptrdiff_t rem = heapRemaining - size; |
---|
| 514 | if ( rem < 0 ) { |
---|
[c4f68dc] | 515 | // If the size requested is bigger than the current remaining storage, increase the size of the heap. |
---|
| 516 | |
---|
| 517 | size_t increase = libCeiling( size > heapExpand ? size : heapExpand, libAlign() ); |
---|
| 518 | if ( sbrk( increase ) == (void *)-1 ) { |
---|
| 519 | unlock( extlock ); |
---|
| 520 | errno = ENOMEM; |
---|
| 521 | return 0; |
---|
| 522 | } // if |
---|
| 523 | #ifdef __STATISTICS__ |
---|
| 524 | sbrk_calls += 1; |
---|
| 525 | sbrk_storage += increase; |
---|
| 526 | #endif // __STATISTICS__ |
---|
| 527 | #ifdef __CFA_DEBUG__ |
---|
| 528 | // Set new memory to garbage so subsequent uninitialized usages might fail. |
---|
| 529 | memset( (char *)heapEnd + heapRemaining, '\377', increase ); |
---|
| 530 | #endif // __CFA_DEBUG__ |
---|
| 531 | rem = heapRemaining + increase - size; |
---|
[b6830d74] | 532 | } // if |
---|
[c4f68dc] | 533 | |
---|
[b6830d74] | 534 | HeapManager.Storage * block = (HeapManager.Storage *)heapEnd; |
---|
| 535 | heapRemaining = rem; |
---|
| 536 | heapEnd = (char *)heapEnd + size; |
---|
| 537 | unlock( extlock ); |
---|
| 538 | return block; |
---|
[c4f68dc] | 539 | } // extend |
---|
| 540 | |
---|
| 541 | |
---|
| 542 | static inline void * doMalloc( size_t size ) with ( heapManager ) { |
---|
[b6830d74] | 543 | HeapManager.Storage * block; |
---|
[c4f68dc] | 544 | |
---|
[b6830d74] | 545 | // Look up size in the size list. Make sure the user request includes space for the header that must be allocated |
---|
| 546 | // along with the block and is a multiple of the alignment size. |
---|
[c4f68dc] | 547 | |
---|
[b6830d74] | 548 | size_t tsize = size + sizeof(HeapManager.Storage); |
---|
| 549 | if ( likely( tsize < mmapStart ) ) { // small size => sbrk |
---|
[c4f68dc] | 550 | HeapManager.FreeHeader * freeElem = |
---|
| 551 | #ifdef FASTLOOKUP |
---|
| 552 | tsize < LookupSizes ? &freeLists[lookup[tsize]] : |
---|
| 553 | #endif // FASTLOOKUP |
---|
| 554 | bsearchl( tsize, freeLists, (size_t)maxBucketsUsed ); // binary search |
---|
| 555 | assert( freeElem <= &freeLists[maxBucketsUsed] ); // subscripting error ? |
---|
| 556 | assert( tsize <= freeElem->blockSize ); // search failure ? |
---|
| 557 | tsize = freeElem->blockSize; // total space needed for request |
---|
| 558 | |
---|
| 559 | // Spin until the lock is acquired for this particular size of block. |
---|
| 560 | |
---|
| 561 | #if defined( SPINLOCK ) |
---|
[b6830d74] | 562 | lock( freeElem->lock __cfaabi_dbg_ctx2 ); |
---|
| 563 | block = freeElem->freeList; // remove node from stack |
---|
[c4f68dc] | 564 | #else |
---|
[b6830d74] | 565 | block = freeElem->freeList.pop(); |
---|
[c4f68dc] | 566 | #endif // SPINLOCK |
---|
| 567 | if ( unlikely( block == 0 ) ) { // no free block ? |
---|
| 568 | #if defined( SPINLOCK ) |
---|
| 569 | unlock( freeElem->lock ); |
---|
| 570 | #endif // SPINLOCK |
---|
| 571 | // Freelist for that size was empty, so carve it out of the heap if there's enough left, or get some more |
---|
| 572 | // and then carve it off. |
---|
| 573 | |
---|
| 574 | block = (HeapManager.Storage *)extend( tsize ); // mutual exclusion on call |
---|
| 575 | if ( unlikely( block == 0 ) ) return 0; |
---|
| 576 | #if defined( SPINLOCK ) |
---|
| 577 | } else { |
---|
| 578 | freeElem->freeList = block->header.kind.real.next; |
---|
| 579 | unlock( freeElem->lock ); |
---|
| 580 | #endif // SPINLOCK |
---|
| 581 | } // if |
---|
| 582 | |
---|
| 583 | block->header.kind.real.home = freeElem; // pointer back to free list of apropriate size |
---|
[b6830d74] | 584 | } else { // large size => mmap |
---|
[c4f68dc] | 585 | tsize = libCeiling( tsize, pageSize ); // must be multiple of page size |
---|
| 586 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 587 | __atomic_add_fetch( &mmap_calls, 1, __ATOMIC_SEQ_CST ); |
---|
| 588 | __atomic_add_fetch( &mmap_storage, tsize, __ATOMIC_SEQ_CST ); |
---|
[c4f68dc] | 589 | #endif // __STATISTICS__ |
---|
| 590 | block = (HeapManager.Storage *)mmap( 0, tsize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, mmapFd, 0 ); |
---|
| 591 | if ( block == (HeapManager.Storage *)MAP_FAILED ) { |
---|
| 592 | // Do not call strerror( errno ) as it may call malloc. |
---|
| 593 | abort( "(HeapManager &)0x%p.doMalloc() : internal error, mmap failure, size:%zu error:%d.", &heapManager, tsize, errno ); |
---|
| 594 | } // if |
---|
| 595 | #ifdef __CFA_DEBUG__ |
---|
| 596 | // Set new memory to garbage so subsequent uninitialized usages might fail. |
---|
| 597 | memset( block, '\377', tsize ); |
---|
| 598 | #endif // __CFA_DEBUG__ |
---|
| 599 | block->header.kind.real.blockSize = tsize; // storage size for munmap |
---|
[b6830d74] | 600 | } // if |
---|
[c4f68dc] | 601 | |
---|
[b6830d74] | 602 | void * area = &(block->data); // adjust off header to user bytes |
---|
[c4f68dc] | 603 | |
---|
| 604 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 605 | assert( ((uintptr_t)area & (libAlign() - 1)) == 0 ); // minimum alignment ? |
---|
| 606 | __atomic_add_fetch( &allocFree, tsize, __ATOMIC_SEQ_CST ); |
---|
| 607 | if ( traceHeap() ) { |
---|
| 608 | enum { BufferSize = 64 }; |
---|
| 609 | char helpText[BufferSize]; |
---|
| 610 | int len = snprintf( helpText, BufferSize, "%p = Malloc( %zu ) (allocated %zu)\n", area, size, tsize ); |
---|
| 611 | // int len = snprintf( helpText, BufferSize, "Malloc %p %zu\n", area, size ); |
---|
| 612 | __cfaabi_dbg_bits_write( helpText, len ); |
---|
| 613 | } // if |
---|
[c4f68dc] | 614 | #endif // __CFA_DEBUG__ |
---|
| 615 | |
---|
[b6830d74] | 616 | return area; |
---|
[c4f68dc] | 617 | } // doMalloc |
---|
| 618 | |
---|
| 619 | |
---|
| 620 | static inline void doFree( void * addr ) with ( heapManager ) { |
---|
| 621 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 622 | if ( unlikely( heapManager.heapBegin == 0 ) ) { |
---|
| 623 | abort( "doFree( %p ) : internal error, called before heap is initialized.", addr ); |
---|
| 624 | } // if |
---|
[c4f68dc] | 625 | #endif // __CFA_DEBUG__ |
---|
| 626 | |
---|
[b6830d74] | 627 | HeapManager.Storage.Header * header; |
---|
| 628 | HeapManager.FreeHeader * freeElem; |
---|
| 629 | size_t size, alignment; // not used (see realloc) |
---|
[c4f68dc] | 630 | |
---|
[b6830d74] | 631 | if ( headers( "free", addr, header, freeElem, size, alignment ) ) { // mmapped ? |
---|
[c4f68dc] | 632 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 633 | __atomic_add_fetch( &munmap_calls, 1, __ATOMIC_SEQ_CST ); |
---|
| 634 | __atomic_add_fetch( &munmap_storage, size, __ATOMIC_SEQ_CST ); |
---|
[c4f68dc] | 635 | #endif // __STATISTICS__ |
---|
| 636 | if ( munmap( header, size ) == -1 ) { |
---|
| 637 | #ifdef __CFA_DEBUG__ |
---|
| 638 | abort( "Attempt to deallocate storage %p not allocated or with corrupt header.\n" |
---|
| 639 | "Possible cause is invalid pointer.", |
---|
| 640 | addr ); |
---|
| 641 | #endif // __CFA_DEBUG__ |
---|
| 642 | } // if |
---|
[b6830d74] | 643 | } else { |
---|
[c4f68dc] | 644 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 645 | // Set free memory to garbage so subsequent usages might fail. |
---|
| 646 | memset( ((HeapManager.Storage *)header)->data, '\377', freeElem->blockSize - sizeof( HeapManager.Storage ) ); |
---|
[c4f68dc] | 647 | #endif // __CFA_DEBUG__ |
---|
| 648 | |
---|
| 649 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 650 | free_storage += size; |
---|
[c4f68dc] | 651 | #endif // __STATISTICS__ |
---|
| 652 | #if defined( SPINLOCK ) |
---|
[b6830d74] | 653 | lock( freeElem->lock __cfaabi_dbg_ctx2 ); // acquire spin lock |
---|
| 654 | header->kind.real.next = freeElem->freeList; // push on stack |
---|
| 655 | freeElem->freeList = (HeapManager.Storage *)header; |
---|
| 656 | unlock( freeElem->lock ); // release spin lock |
---|
[c4f68dc] | 657 | #else |
---|
[b6830d74] | 658 | freeElem->freeList.push( *(HeapManager.Storage *)header ); |
---|
[c4f68dc] | 659 | #endif // SPINLOCK |
---|
[b6830d74] | 660 | } // if |
---|
[c4f68dc] | 661 | |
---|
| 662 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 663 | __atomic_add_fetch( &allocFree, -size, __ATOMIC_SEQ_CST ); |
---|
| 664 | if ( traceHeap() ) { |
---|
| 665 | char helpText[64]; |
---|
| 666 | int len = snprintf( helpText, sizeof(helpText), "Free( %p ) size:%zu\n", addr, size ); |
---|
| 667 | __cfaabi_dbg_bits_write( helpText, len ); |
---|
| 668 | } // if |
---|
[c4f68dc] | 669 | #endif // __CFA_DEBUG__ |
---|
| 670 | } // doFree |
---|
| 671 | |
---|
| 672 | |
---|
[5d4fa18] | 673 | size_t checkFree( HeapManager & manager ) with ( manager ) { |
---|
[b6830d74] | 674 | size_t total = 0; |
---|
[c4f68dc] | 675 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 676 | __cfaabi_dbg_bits_acquire(); |
---|
| 677 | __cfaabi_dbg_bits_print_nolock( "\nBin lists (bin size : free blocks on list)\n" ); |
---|
[c4f68dc] | 678 | #endif // __STATISTICS__ |
---|
[b6830d74] | 679 | for ( unsigned int i = 0; i < maxBucketsUsed; i += 1 ) { |
---|
[d46ed6e] | 680 | size_t size = freeLists[i].blockSize; |
---|
| 681 | #ifdef __STATISTICS__ |
---|
| 682 | unsigned int N = 0; |
---|
| 683 | #endif // __STATISTICS__ |
---|
[b6830d74] | 684 | |
---|
[d46ed6e] | 685 | #if defined( SPINLOCK ) |
---|
| 686 | for ( HeapManager.Storage * p = freeLists[i].freeList; p != 0; p = p->header.kind.real.next ) { |
---|
| 687 | #else |
---|
| 688 | for ( HeapManager.Storage * p = freeLists[i].freeList.top(); p != 0; p = p->header.kind.real.next.top ) { |
---|
| 689 | #endif // SPINLOCK |
---|
| 690 | total += size; |
---|
| 691 | #ifdef __STATISTICS__ |
---|
| 692 | N += 1; |
---|
| 693 | #endif // __STATISTICS__ |
---|
[b6830d74] | 694 | } // for |
---|
| 695 | |
---|
[d46ed6e] | 696 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 697 | __cfaabi_dbg_bits_print_nolock( "%7zu, %-7u ", size, N ); |
---|
| 698 | if ( (i + 1) % 8 == 0 ) __cfaabi_dbg_bits_print_nolock( "\n" ); |
---|
[d46ed6e] | 699 | #endif // __STATISTICS__ |
---|
| 700 | } // for |
---|
| 701 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 702 | __cfaabi_dbg_bits_print_nolock( "\ntotal free blocks:%zu\n", total ); |
---|
| 703 | __cfaabi_dbg_bits_release(); |
---|
[d46ed6e] | 704 | #endif // __STATISTICS__ |
---|
| 705 | return (char *)heapEnd - (char *)heapBegin - total; |
---|
| 706 | } // checkFree |
---|
[c4f68dc] | 707 | |
---|
[b6830d74] | 708 | // #comment TD : This is not a good name, plus this feels like it could easily be folded into doMalloc |
---|
[c4f68dc] | 709 | static inline void * malloc2( size_t size ) { // necessary for malloc statistics |
---|
[f0b3f51] | 710 | assert( heapManager.heapBegin != 0 ); |
---|
[b6830d74] | 711 | void * area = doMalloc( size ); |
---|
| 712 | if ( unlikely( area == 0 ) ) errno = ENOMEM; // POSIX |
---|
| 713 | return area; |
---|
[c4f68dc] | 714 | } // malloc2 |
---|
| 715 | |
---|
| 716 | |
---|
| 717 | static inline void * memalign2( size_t alignment, size_t size ) { // necessary for malloc statistics |
---|
| 718 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 719 | checkAlign( alignment ); // check alignment |
---|
[c4f68dc] | 720 | #endif // __CFA_DEBUG__ |
---|
| 721 | |
---|
[b6830d74] | 722 | // if alignment <= default alignment, do normal malloc as two headers are unnecessary |
---|
| 723 | if ( unlikely( alignment <= libAlign() ) ) return malloc2( size ); |
---|
| 724 | |
---|
| 725 | // Allocate enough storage to guarantee an address on the alignment boundary, and sufficient space before it for |
---|
| 726 | // administrative storage. NOTE, WHILE THERE ARE 2 HEADERS, THE FIRST ONE IS IMPLICITLY CREATED BY DOMALLOC. |
---|
| 727 | // .-------------v-----------------v----------------v----------, |
---|
| 728 | // | Real Header | ... padding ... | Fake Header | data ... | |
---|
| 729 | // `-------------^-----------------^-+--------------^----------' |
---|
| 730 | // |<--------------------------------' offset/align |<-- alignment boundary |
---|
| 731 | |
---|
| 732 | // subtract libAlign() because it is already the minimum alignment |
---|
| 733 | // add sizeof(Storage) for fake header |
---|
| 734 | // #comment TD : this is the only place that calls doMalloc without calling malloc2, why ? |
---|
| 735 | char * area = (char *)doMalloc( size + alignment - libAlign() + sizeof(HeapManager.Storage) ); |
---|
| 736 | if ( unlikely( area == 0 ) ) return area; |
---|
| 737 | |
---|
| 738 | // address in the block of the "next" alignment address |
---|
| 739 | char * user = (char *)libCeiling( (uintptr_t)(area + sizeof(HeapManager.Storage)), alignment ); |
---|
| 740 | |
---|
| 741 | // address of header from malloc |
---|
| 742 | HeapManager.Storage.Header * realHeader = headerAddr( area ); |
---|
| 743 | // address of fake header * before* the alignment location |
---|
| 744 | HeapManager.Storage.Header * fakeHeader = headerAddr( user ); |
---|
| 745 | // SKULLDUGGERY: insert the offset to the start of the actual storage block and remember alignment |
---|
| 746 | fakeHeader->kind.fake.offset = (char *)fakeHeader - (char *)realHeader; |
---|
| 747 | // SKULLDUGGERY: odd alignment imples fake header |
---|
| 748 | fakeHeader->kind.fake.alignment = alignment | 1; |
---|
| 749 | |
---|
| 750 | return user; |
---|
[c4f68dc] | 751 | } // memalign2 |
---|
| 752 | |
---|
| 753 | |
---|
| 754 | extern "C" { |
---|
[b6830d74] | 755 | // The malloc() function allocates size bytes and returns a pointer to the |
---|
| 756 | // allocated memory. The memory is not initialized. If size is 0, then malloc() |
---|
| 757 | // returns either NULL, or a unique pointer value that can later be successfully |
---|
| 758 | // passed to free(). |
---|
| 759 | void * malloc( size_t size ) { |
---|
[c4f68dc] | 760 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 761 | __atomic_add_fetch( &malloc_calls, 1, __ATOMIC_SEQ_CST ); |
---|
| 762 | __atomic_add_fetch( &malloc_storage, size, __ATOMIC_SEQ_CST ); |
---|
[c4f68dc] | 763 | #endif // __STATISTICS__ |
---|
| 764 | |
---|
| 765 | return malloc2( size ); |
---|
[b6830d74] | 766 | } // malloc |
---|
[c4f68dc] | 767 | |
---|
[b6830d74] | 768 | // The calloc() function allocates memory for an array of nmemb elements of |
---|
| 769 | // size bytes each and returns a pointer to the allocated memory. The memory |
---|
| 770 | // is set to zero. If nmemb or size is 0, then calloc() returns either NULL, |
---|
| 771 | // or a unique pointer value that can later be successfully passed to free(). |
---|
| 772 | void * calloc( size_t noOfElems, size_t elemSize ) { |
---|
[c4f68dc] | 773 | size_t size = noOfElems * elemSize; |
---|
| 774 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 775 | __atomic_add_fetch( &calloc_calls, 1, __ATOMIC_SEQ_CST ); |
---|
| 776 | __atomic_add_fetch( &calloc_storage, size, __ATOMIC_SEQ_CST ); |
---|
[c4f68dc] | 777 | #endif // __STATISTICS__ |
---|
| 778 | |
---|
| 779 | char * area = (char *)malloc2( size ); |
---|
| 780 | if ( unlikely( area == 0 ) ) return 0; |
---|
[b6830d74] | 781 | |
---|
[c4f68dc] | 782 | HeapManager.Storage.Header * header; |
---|
| 783 | HeapManager.FreeHeader * freeElem; |
---|
| 784 | size_t asize, alignment; |
---|
[93c2e0a] | 785 | bool mapped __attribute__(( unused )) = headers( "calloc", area, header, freeElem, asize, alignment ); |
---|
[c4f68dc] | 786 | #ifndef __CFA_DEBUG__ |
---|
[73abe95] | 787 | // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero. |
---|
[c4f68dc] | 788 | if ( ! mapped ) |
---|
| 789 | #endif // __CFA_DEBUG__ |
---|
| 790 | memset( area, '\0', asize - sizeof(HeapManager.Storage) ); // set to zeros |
---|
[b6830d74] | 791 | |
---|
[c4f68dc] | 792 | header->kind.real.blockSize |= 2; // mark as zero filled |
---|
| 793 | return area; |
---|
[b6830d74] | 794 | } // calloc |
---|
[c4f68dc] | 795 | |
---|
[b6830d74] | 796 | // #comment TD : Document this function |
---|
| 797 | void * cmemalign( size_t alignment, size_t noOfElems, size_t elemSize ) { |
---|
[c4f68dc] | 798 | size_t size = noOfElems * elemSize; |
---|
| 799 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 800 | __atomic_add_fetch( &cmemalign_calls, 1, __ATOMIC_SEQ_CST ); |
---|
| 801 | __atomic_add_fetch( &cmemalign_storage, size, __ATOMIC_SEQ_CST ); |
---|
[c4f68dc] | 802 | #endif // __STATISTICS__ |
---|
| 803 | |
---|
| 804 | char * area = (char *)memalign2( alignment, size ); |
---|
| 805 | if ( unlikely( area == 0 ) ) return 0; |
---|
| 806 | HeapManager.Storage.Header * header; |
---|
| 807 | HeapManager.FreeHeader * freeElem; |
---|
| 808 | size_t asize; |
---|
[93c2e0a] | 809 | bool mapped __attribute__(( unused )) = headers( "cmemalign", area, header, freeElem, asize, alignment ); |
---|
[c4f68dc] | 810 | #ifndef __CFA_DEBUG__ |
---|
[73abe95] | 811 | // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero. |
---|
[c4f68dc] | 812 | if ( ! mapped ) |
---|
| 813 | #endif // __CFA_DEBUG__ |
---|
| 814 | memset( area, '\0', asize - ( (char *)area - (char *)header ) ); // set to zeros |
---|
| 815 | header->kind.real.blockSize |= 2; // mark as zero filled |
---|
| 816 | |
---|
| 817 | return area; |
---|
[b6830d74] | 818 | } // cmemalign |
---|
| 819 | |
---|
| 820 | // The realloc() function changes the size of the memory block pointed to by |
---|
| 821 | // ptr to size bytes. The contents will be unchanged in the range from the |
---|
| 822 | // start of the region up to the minimum of the old and new sizes. If the new |
---|
| 823 | // size is larger than the old size, the added memory will not be initialized. |
---|
| 824 | // If ptr is NULL, then the call is equivalent to malloc(size), for all values |
---|
| 825 | // of size; if size is equal to zero, and ptr is not NULL, then the call is |
---|
| 826 | // equivalent to free(ptr). Unless ptr is NULL, it must have been returned by |
---|
| 827 | // an earlier call to malloc(), calloc() or realloc(). If the area pointed to |
---|
| 828 | // was moved, a free(ptr) is done. |
---|
| 829 | void * realloc( void * addr, size_t size ) { |
---|
[c4f68dc] | 830 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 831 | __atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST ); |
---|
[c4f68dc] | 832 | #endif // __STATISTICS__ |
---|
| 833 | |
---|
| 834 | if ( unlikely( addr == 0 ) ) return malloc2( size ); // special cases |
---|
| 835 | if ( unlikely( size == 0 ) ) { free( addr ); return 0; } |
---|
| 836 | |
---|
| 837 | HeapManager.Storage.Header * header; |
---|
| 838 | HeapManager.FreeHeader * freeElem; |
---|
| 839 | size_t asize, alignment = 0; |
---|
| 840 | headers( "realloc", addr, header, freeElem, asize, alignment ); |
---|
| 841 | |
---|
| 842 | size_t usize = asize - ( (char *)addr - (char *)header ); // compute the amount of user storage in the block |
---|
| 843 | if ( usize >= size ) { // already sufficient storage |
---|
| 844 | // This case does not result in a new profiler entry because the previous one still exists and it must match with |
---|
| 845 | // the free for this memory. Hence, this realloc does not appear in the profiler output. |
---|
| 846 | return addr; |
---|
| 847 | } // if |
---|
| 848 | |
---|
| 849 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 850 | __atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST ); |
---|
[c4f68dc] | 851 | #endif // __STATISTICS__ |
---|
| 852 | |
---|
| 853 | void * area; |
---|
| 854 | if ( unlikely( alignment != 0 ) ) { // previous request memalign? |
---|
| 855 | area = memalign( alignment, size ); // create new area |
---|
| 856 | } else { |
---|
| 857 | area = malloc2( size ); // create new area |
---|
| 858 | } // if |
---|
| 859 | if ( unlikely( area == 0 ) ) return 0; |
---|
| 860 | if ( unlikely( header->kind.real.blockSize & 2 ) ) { // previous request zero fill (calloc/cmemalign) ? |
---|
| 861 | assert( (header->kind.real.blockSize & 1) == 0 ); |
---|
[93c2e0a] | 862 | bool mapped __attribute__(( unused )) = headers( "realloc", area, header, freeElem, asize, alignment ); |
---|
[c4f68dc] | 863 | #ifndef __CFA_DEBUG__ |
---|
[73abe95] | 864 | // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero. |
---|
[c4f68dc] | 865 | if ( ! mapped ) |
---|
| 866 | #endif // __CFA_DEBUG__ |
---|
| 867 | memset( (char *)area + usize, '\0', asize - ( (char *)area - (char *)header ) - usize ); // zero-fill back part |
---|
| 868 | header->kind.real.blockSize |= 2; // mark new request as zero fill |
---|
| 869 | } // if |
---|
| 870 | memcpy( area, addr, usize ); // copy bytes |
---|
| 871 | free( addr ); |
---|
| 872 | return area; |
---|
[b6830d74] | 873 | } // realloc |
---|
[c4f68dc] | 874 | |
---|
| 875 | |
---|
[b6830d74] | 876 | // The obsolete function memalign() allocates size bytes and returns |
---|
| 877 | // a pointer to the allocated memory. The memory address will be a |
---|
| 878 | // multiple of alignment, which must be a power of two. |
---|
| 879 | void * memalign( size_t alignment, size_t size ) __attribute__ ((deprecated)); |
---|
| 880 | void * memalign( size_t alignment, size_t size ) { |
---|
[c4f68dc] | 881 | #ifdef __STATISTICS__ |
---|
| 882 | __atomic_add_fetch( &memalign_calls, 1, __ATOMIC_SEQ_CST ); |
---|
| 883 | __atomic_add_fetch( &memalign_storage, size, __ATOMIC_SEQ_CST ); |
---|
| 884 | #endif // __STATISTICS__ |
---|
| 885 | |
---|
| 886 | void * area = memalign2( alignment, size ); |
---|
| 887 | |
---|
| 888 | return area; |
---|
[b6830d74] | 889 | } // memalign |
---|
[c4f68dc] | 890 | |
---|
[b6830d74] | 891 | // The function aligned_alloc() is the same as memalign(), except for |
---|
| 892 | // the added restriction that size should be a multiple of alignment. |
---|
| 893 | void * aligned_alloc( size_t alignment, size_t size ) { |
---|
[c4f68dc] | 894 | return memalign( alignment, size ); |
---|
[b6830d74] | 895 | } // aligned_alloc |
---|
[c4f68dc] | 896 | |
---|
| 897 | |
---|
[b6830d74] | 898 | // The function posix_memalign() allocates size bytes and places the address |
---|
| 899 | // of the allocated memory in *memptr. The address of the allocated memory |
---|
| 900 | // will be a multiple of alignment, which must be a power of two and a multiple |
---|
| 901 | // of sizeof(void *). If size is 0, then posix_memalign() returns either NULL, |
---|
| 902 | // or a unique pointer value that can later be successfully passed to free(3). |
---|
| 903 | int posix_memalign( void ** memptr, size_t alignment, size_t size ) { |
---|
[c4f68dc] | 904 | if ( alignment < sizeof(void *) || ! libPow2( alignment ) ) return EINVAL; // check alignment |
---|
| 905 | * memptr = memalign( alignment, size ); |
---|
| 906 | if ( unlikely( * memptr == 0 ) ) return ENOMEM; |
---|
| 907 | return 0; |
---|
[b6830d74] | 908 | } // posix_memalign |
---|
[c4f68dc] | 909 | |
---|
[b6830d74] | 910 | // The obsolete function valloc() allocates size bytes and returns a pointer |
---|
| 911 | // to the allocated memory. The memory address will be a multiple of the page size. |
---|
| 912 | // It is equivalent to memalign(sysconf(_SC_PAGESIZE),size). |
---|
| 913 | void * valloc( size_t size ) __attribute__ ((deprecated)); |
---|
| 914 | void * valloc( size_t size ) { |
---|
[c4f68dc] | 915 | return memalign( pageSize, size ); |
---|
[b6830d74] | 916 | } // valloc |
---|
[c4f68dc] | 917 | |
---|
| 918 | |
---|
[b6830d74] | 919 | // The free() function frees the memory space pointed to by ptr, which must |
---|
| 920 | // have been returned by a previous call to malloc(), calloc() or realloc(). |
---|
| 921 | // Otherwise, or if free(ptr) has already been called before, undefined |
---|
| 922 | // behavior occurs. If ptr is NULL, no operation is performed. |
---|
| 923 | void free( void * addr ) { |
---|
[c4f68dc] | 924 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 925 | __atomic_add_fetch( &free_calls, 1, __ATOMIC_SEQ_CST ); |
---|
[c4f68dc] | 926 | #endif // __STATISTICS__ |
---|
| 927 | |
---|
[b6830d74] | 928 | // #comment TD : To decrease nesting I would but the special case in the |
---|
| 929 | // else instead, plus it reads more naturally to have the |
---|
| 930 | // short / normal case instead |
---|
[c4f68dc] | 931 | if ( unlikely( addr == 0 ) ) { // special case |
---|
[d46ed6e] | 932 | #ifdef __CFA_DEBUG__ |
---|
[b6830d74] | 933 | if ( traceHeap() ) { |
---|
| 934 | #define nullmsg "Free( 0x0 ) size:0\n" |
---|
| 935 | // Do not debug print free( 0 ), as it can cause recursive entry from sprintf. |
---|
| 936 | __cfaabi_dbg_bits_write( nullmsg, sizeof(nullmsg) - 1 ); |
---|
| 937 | } // if |
---|
[d46ed6e] | 938 | #endif // __CFA_DEBUG__ |
---|
[c4f68dc] | 939 | return; |
---|
| 940 | } // exit |
---|
| 941 | |
---|
| 942 | doFree( addr ); |
---|
[b6830d74] | 943 | } // free |
---|
[93c2e0a] | 944 | |
---|
[b6830d74] | 945 | // The mallopt() function adjusts parameters that control the behavior of the |
---|
| 946 | // memory-allocation functions (see malloc(3)). The param argument specifies |
---|
| 947 | // the parameter to be modified, and value specifies the new value for that |
---|
| 948 | // parameter. |
---|
| 949 | int mallopt( int option, int value ) { |
---|
[c4f68dc] | 950 | choose( option ) { |
---|
[b6830d74] | 951 | case M_TOP_PAD: |
---|
| 952 | if ( setHeapExpand( value ) ) fallthru default; |
---|
| 953 | case M_MMAP_THRESHOLD: |
---|
| 954 | if ( setMmapStart( value ) ) fallthru default; |
---|
| 955 | default: |
---|
| 956 | // #comment TD : 1 for unsopported feels wrong |
---|
| 957 | return 1; // success, or unsupported |
---|
[c4f68dc] | 958 | } // switch |
---|
| 959 | return 0; // error |
---|
[b6830d74] | 960 | } // mallopt |
---|
[c4f68dc] | 961 | |
---|
[b6830d74] | 962 | // The malloc_trim() function attempts to release free memory at the top |
---|
| 963 | // of the heap (by calling sbrk(2) with a suitable argument). |
---|
[c4f68dc] | 964 | int malloc_trim( size_t ) { |
---|
| 965 | return 0; // => impossible to release memory |
---|
| 966 | } // malloc_trim |
---|
| 967 | |
---|
[b6830d74] | 968 | // The malloc_usable_size() function returns the number of usable bytes in the |
---|
| 969 | // block pointed to by ptr, a pointer to a block of memory allocated by |
---|
| 970 | // malloc(3) or a related function. |
---|
| 971 | size_t malloc_usable_size( void * addr ) { |
---|
[c4f68dc] | 972 | if ( unlikely( addr == 0 ) ) return 0; // null allocation has 0 size |
---|
[b6830d74] | 973 | |
---|
[c4f68dc] | 974 | HeapManager.Storage.Header * header; |
---|
| 975 | HeapManager.FreeHeader * freeElem; |
---|
| 976 | size_t size, alignment; |
---|
| 977 | |
---|
| 978 | headers( "malloc_usable_size", addr, header, freeElem, size, alignment ); |
---|
| 979 | size_t usize = size - ( (char *)addr - (char *)header ); // compute the amount of user storage in the block |
---|
| 980 | return usize; |
---|
[b6830d74] | 981 | } // malloc_usable_size |
---|
[c4f68dc] | 982 | |
---|
| 983 | |
---|
[b6830d74] | 984 | // #comment TD : Document this function |
---|
| 985 | size_t malloc_alignment( void * addr ) { |
---|
[c4f68dc] | 986 | if ( unlikely( addr == 0 ) ) return libAlign(); // minimum alignment |
---|
| 987 | HeapManager.Storage.Header * header = (HeapManager.Storage.Header *)( (char *)addr - sizeof(HeapManager.Storage) ); |
---|
| 988 | if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ? |
---|
| 989 | return header->kind.fake.alignment & -2; // remove flag from value |
---|
| 990 | } else { |
---|
| 991 | return libAlign (); // minimum alignment |
---|
| 992 | } // if |
---|
[b6830d74] | 993 | } // malloc_alignment |
---|
[c4f68dc] | 994 | |
---|
| 995 | |
---|
[b6830d74] | 996 | // #comment TD : Document this function |
---|
| 997 | bool malloc_zero_fill( void * addr ) { |
---|
[c4f68dc] | 998 | if ( unlikely( addr == 0 ) ) return false; // null allocation is not zero fill |
---|
[b6830d74] | 999 | |
---|
[c4f68dc] | 1000 | HeapManager.Storage.Header * header = (HeapManager.Storage.Header *)( (char *)addr - sizeof(HeapManager.Storage) ); |
---|
| 1001 | if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ? |
---|
| 1002 | header = (HeapManager.Storage.Header *)((char *)header - header->kind.fake.offset); |
---|
| 1003 | } // if |
---|
| 1004 | return (header->kind.real.blockSize & 2) != 0; // zero filled (calloc/cmemalign) ? |
---|
[b6830d74] | 1005 | } // malloc_zero_fill |
---|
[c4f68dc] | 1006 | |
---|
| 1007 | |
---|
[b6830d74] | 1008 | // #comment TD : Document this function |
---|
| 1009 | void malloc_stats( void ) { |
---|
[c4f68dc] | 1010 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 1011 | printStats(); |
---|
| 1012 | if ( checkFree() ) checkFree( heapManager ); |
---|
[c4f68dc] | 1013 | #endif // __STATISTICS__ |
---|
[b6830d74] | 1014 | } // malloc_stats |
---|
[c4f68dc] | 1015 | |
---|
[b6830d74] | 1016 | // #comment TD : Document this function |
---|
| 1017 | int malloc_stats_fd( int fd ) { |
---|
[c4f68dc] | 1018 | #ifdef __STATISTICS__ |
---|
[b6830d74] | 1019 | int temp = statfd; |
---|
| 1020 | statfd = fd; |
---|
| 1021 | return temp; |
---|
[c4f68dc] | 1022 | #else |
---|
[b6830d74] | 1023 | return -1; |
---|
[c4f68dc] | 1024 | #endif // __STATISTICS__ |
---|
[b6830d74] | 1025 | } // malloc_stats_fd |
---|
[c4f68dc] | 1026 | |
---|
| 1027 | |
---|
[b6830d74] | 1028 | // #comment TD : Document this function |
---|
[c4f68dc] | 1029 | int malloc_info( int options, FILE * stream ) { |
---|
[d46ed6e] | 1030 | return printStatsXML( stream ); |
---|
[c4f68dc] | 1031 | } // malloc_info |
---|
| 1032 | |
---|
| 1033 | |
---|
[b6830d74] | 1034 | // #comment TD : What are these two functions for? |
---|
[c4f68dc] | 1035 | void * malloc_get_state( void ) { |
---|
| 1036 | return 0; |
---|
| 1037 | } // malloc_get_state |
---|
| 1038 | |
---|
| 1039 | int malloc_set_state( void * ptr ) { |
---|
| 1040 | return 0; |
---|
| 1041 | } // malloc_set_state |
---|
| 1042 | } // extern "C" |
---|
| 1043 | |
---|
| 1044 | |
---|
| 1045 | // Local Variables: // |
---|
| 1046 | // tab-width: 4 // |
---|
| 1047 | // compile-command: "cfa -nodebug -O2 heap.c" // |
---|
| 1048 | // End: // |
---|