1 | // |
---|
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. |
---|
6 | // |
---|
7 | // heap.c -- |
---|
8 | // |
---|
9 | // Author : Peter A. Buhr |
---|
10 | // Created On : Tue Dec 19 21:58:35 2017 |
---|
11 | // Last Modified By : Peter A. Buhr |
---|
12 | // Last Modified On : Tue Feb 4 10:04:51 2020 |
---|
13 | // Update Count : 648 |
---|
14 | // |
---|
15 | |
---|
16 | #include <unistd.h> // sbrk, sysconf |
---|
17 | #include <stdbool.h> // true, false |
---|
18 | #include <stdio.h> // snprintf, fileno |
---|
19 | #include <errno.h> // errno |
---|
20 | #include <string.h> // memset, memcpy |
---|
21 | extern "C" { |
---|
22 | #include <sys/mman.h> // mmap, munmap |
---|
23 | } // extern "C" |
---|
24 | |
---|
25 | // #comment TD : Many of these should be merged into math I believe |
---|
26 | #include "bits/align.hfa" // libPow2 |
---|
27 | #include "bits/defs.hfa" // likely, unlikely |
---|
28 | #include "bits/locks.hfa" // __spinlock_t |
---|
29 | #include "startup.hfa" // STARTUP_PRIORITY_MEMORY |
---|
30 | //#include "stdlib.hfa" // bsearchl |
---|
31 | #include "malloc.h" |
---|
32 | |
---|
33 | #define MIN(x, y) (y > x ? x : y) |
---|
34 | |
---|
35 | static bool traceHeap = false; |
---|
36 | |
---|
37 | inline bool traceHeap() { return traceHeap; } |
---|
38 | |
---|
39 | bool traceHeapOn() { |
---|
40 | bool temp = traceHeap; |
---|
41 | traceHeap = true; |
---|
42 | return temp; |
---|
43 | } // traceHeapOn |
---|
44 | |
---|
45 | bool traceHeapOff() { |
---|
46 | bool temp = traceHeap; |
---|
47 | traceHeap = false; |
---|
48 | return temp; |
---|
49 | } // traceHeapOff |
---|
50 | |
---|
51 | bool traceHeapTerm() { return false; } |
---|
52 | |
---|
53 | |
---|
54 | static bool prtFree = false; |
---|
55 | |
---|
56 | inline bool prtFree() { |
---|
57 | return prtFree; |
---|
58 | } // prtFree |
---|
59 | |
---|
60 | bool prtFreeOn() { |
---|
61 | bool temp = prtFree; |
---|
62 | prtFree = true; |
---|
63 | return temp; |
---|
64 | } // prtFreeOn |
---|
65 | |
---|
66 | bool prtFreeOff() { |
---|
67 | bool temp = prtFree; |
---|
68 | prtFree = false; |
---|
69 | return temp; |
---|
70 | } // prtFreeOff |
---|
71 | |
---|
72 | |
---|
73 | enum { |
---|
74 | // Define the default extension heap amount in units of bytes. When the uC++ supplied heap reaches the brk address, |
---|
75 | // the brk address is extended by the extension amount. |
---|
76 | __CFA_DEFAULT_HEAP_EXPANSION__ = (1 * 1024 * 1024), |
---|
77 | |
---|
78 | // Define the mmap crossover point during allocation. Allocations less than this amount are allocated from buckets; |
---|
79 | // values greater than or equal to this value are mmap from the operating system. |
---|
80 | __CFA_DEFAULT_MMAP_START__ = (512 * 1024 + 1), |
---|
81 | }; |
---|
82 | |
---|
83 | size_t default_mmap_start() __attribute__(( weak )) { |
---|
84 | return __CFA_DEFAULT_MMAP_START__; |
---|
85 | } // default_mmap_start |
---|
86 | |
---|
87 | size_t default_heap_expansion() __attribute__(( weak )) { |
---|
88 | return __CFA_DEFAULT_HEAP_EXPANSION__; |
---|
89 | } // default_heap_expansion |
---|
90 | |
---|
91 | |
---|
92 | #ifdef __CFA_DEBUG__ |
---|
93 | static unsigned int allocFree; // running total of allocations minus frees |
---|
94 | |
---|
95 | static void prtUnfreed() { |
---|
96 | if ( allocFree != 0 ) { |
---|
97 | // DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT. |
---|
98 | char helpText[512]; |
---|
99 | 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" |
---|
100 | "Possible cause is unfreed storage allocated by the program or system/library routines called from the program.\n", |
---|
101 | (long int)getpid(), allocFree, allocFree ); // always print the UNIX pid |
---|
102 | __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug |
---|
103 | } // if |
---|
104 | } // prtUnfreed |
---|
105 | |
---|
106 | extern "C" { |
---|
107 | void heapAppStart() { // called by __cfaabi_appready_startup |
---|
108 | allocFree = 0; |
---|
109 | } // heapAppStart |
---|
110 | |
---|
111 | void heapAppStop() { // called by __cfaabi_appready_startdown |
---|
112 | fclose( stdin ); fclose( stdout ); |
---|
113 | prtUnfreed(); |
---|
114 | } // heapAppStop |
---|
115 | } // extern "C" |
---|
116 | #endif // __CFA_DEBUG__ |
---|
117 | |
---|
118 | |
---|
119 | // statically allocated variables => zero filled. |
---|
120 | static size_t pageSize; // architecture pagesize |
---|
121 | static size_t heapExpand; // sbrk advance |
---|
122 | static size_t mmapStart; // cross over point for mmap |
---|
123 | static unsigned int maxBucketsUsed; // maximum number of buckets in use |
---|
124 | |
---|
125 | |
---|
126 | #define SPINLOCK 0 |
---|
127 | #define LOCKFREE 1 |
---|
128 | #define BUCKETLOCK SPINLOCK |
---|
129 | #if BUCKETLOCK == LOCKFREE |
---|
130 | #include <uStackLF.h> |
---|
131 | #endif // LOCKFREE |
---|
132 | |
---|
133 | // Recursive definitions: HeapManager needs size of bucket array and bucket area needs sizeof HeapManager storage. |
---|
134 | // Break recusion by hardcoding number of buckets and statically checking number is correct after bucket array defined. |
---|
135 | enum { NoBucketSizes = 91 }; // number of buckets sizes |
---|
136 | |
---|
137 | struct HeapManager { |
---|
138 | // struct FreeHeader; // forward declaration |
---|
139 | |
---|
140 | struct Storage { |
---|
141 | struct Header { // header |
---|
142 | union Kind { |
---|
143 | struct RealHeader { |
---|
144 | union { |
---|
145 | struct { // 4-byte word => 8-byte header, 8-byte word => 16-byte header |
---|
146 | #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4 |
---|
147 | uint32_t padding; // unused, force home/blocksize to overlay alignment in fake header |
---|
148 | #endif // __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4 |
---|
149 | |
---|
150 | union { |
---|
151 | // FreeHeader * home; // allocated block points back to home locations (must overlay alignment) |
---|
152 | void * home; // allocated block points back to home locations (must overlay alignment) |
---|
153 | size_t blockSize; // size for munmap (must overlay alignment) |
---|
154 | #if BUCKLOCK == SPINLOCK |
---|
155 | Storage * next; // freed block points next freed block of same size |
---|
156 | #endif // SPINLOCK |
---|
157 | }; |
---|
158 | |
---|
159 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4 |
---|
160 | uint32_t padding; // unused, force home/blocksize to overlay alignment in fake header |
---|
161 | #endif // __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4 |
---|
162 | }; |
---|
163 | // future code |
---|
164 | #if BUCKLOCK == LOCKFREE |
---|
165 | Stack<Storage>::Link next; // freed block points next freed block of same size (double-wide) |
---|
166 | #endif // LOCKFREE |
---|
167 | }; |
---|
168 | } real; // RealHeader |
---|
169 | struct FakeHeader { |
---|
170 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
---|
171 | uint32_t alignment; // low-order bits of home/blockSize used for tricks |
---|
172 | #endif // __ORDER_LITTLE_ENDIAN__ |
---|
173 | |
---|
174 | uint32_t offset; |
---|
175 | |
---|
176 | #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
---|
177 | uint32_t alignment; // low-order bits of home/blockSize used for tricks |
---|
178 | #endif // __ORDER_BIG_ENDIAN__ |
---|
179 | } fake; // FakeHeader |
---|
180 | } kind; // Kind |
---|
181 | } header; // Header |
---|
182 | char pad[libAlign() - sizeof( Header )]; |
---|
183 | char data[0]; // storage |
---|
184 | }; // Storage |
---|
185 | |
---|
186 | static_assert( libAlign() >= sizeof( Storage ), "libAlign() < sizeof( Storage )" ); |
---|
187 | |
---|
188 | struct FreeHeader { |
---|
189 | #if BUCKLOCK == SPINLOCK |
---|
190 | __spinlock_t lock; // must be first field for alignment |
---|
191 | Storage * freeList; |
---|
192 | #elif BUCKLOCK == LOCKFREE |
---|
193 | // future code |
---|
194 | StackLF<Storage> freeList; |
---|
195 | #else |
---|
196 | #error undefined lock type for bucket lock |
---|
197 | #endif // SPINLOCK |
---|
198 | size_t blockSize; // size of allocations on this list |
---|
199 | }; // FreeHeader |
---|
200 | |
---|
201 | // must be first fields for alignment |
---|
202 | __spinlock_t extlock; // protects allocation-buffer extension |
---|
203 | FreeHeader freeLists[NoBucketSizes]; // buckets for different allocation sizes |
---|
204 | |
---|
205 | void * heapBegin; // start of heap |
---|
206 | void * heapEnd; // logical end of heap |
---|
207 | size_t heapRemaining; // amount of storage not allocated in the current chunk |
---|
208 | }; // HeapManager |
---|
209 | |
---|
210 | static inline size_t getKey( const HeapManager.FreeHeader & freeheader ) { return freeheader.blockSize; } |
---|
211 | |
---|
212 | |
---|
213 | #define FASTLOOKUP |
---|
214 | #define __STATISTICS__ |
---|
215 | |
---|
216 | // Bucket size must be multiple of 16. |
---|
217 | // Powers of 2 are common allocation sizes, so make powers of 2 generate the minimum required size. |
---|
218 | static const unsigned int bucketSizes[] @= { // different bucket sizes |
---|
219 | 16, 32, 48, 64 + sizeof(HeapManager.Storage), // 4 |
---|
220 | 96, 112, 128 + sizeof(HeapManager.Storage), // 3 |
---|
221 | 160, 192, 224, 256 + sizeof(HeapManager.Storage), // 4 |
---|
222 | 320, 384, 448, 512 + sizeof(HeapManager.Storage), // 4 |
---|
223 | 640, 768, 896, 1_024 + sizeof(HeapManager.Storage), // 4 |
---|
224 | 1_536, 2_048 + sizeof(HeapManager.Storage), // 2 |
---|
225 | 2_560, 3_072, 3_584, 4_096 + sizeof(HeapManager.Storage), // 4 |
---|
226 | 6_144, 8_192 + sizeof(HeapManager.Storage), // 2 |
---|
227 | 9_216, 10_240, 11_264, 12_288, 13_312, 14_336, 15_360, 16_384 + sizeof(HeapManager.Storage), // 8 |
---|
228 | 18_432, 20_480, 22_528, 24_576, 26_624, 28_672, 30_720, 32_768 + sizeof(HeapManager.Storage), // 8 |
---|
229 | 36_864, 40_960, 45_056, 49_152, 53_248, 57_344, 61_440, 65_536 + sizeof(HeapManager.Storage), // 8 |
---|
230 | 73_728, 81_920, 90_112, 98_304, 106_496, 114_688, 122_880, 131_072 + sizeof(HeapManager.Storage), // 8 |
---|
231 | 147_456, 163_840, 180_224, 196_608, 212_992, 229_376, 245_760, 262_144 + sizeof(HeapManager.Storage), // 8 |
---|
232 | 294_912, 327_680, 360_448, 393_216, 425_984, 458_752, 491_520, 524_288 + sizeof(HeapManager.Storage), // 8 |
---|
233 | 655_360, 786_432, 917_504, 1_048_576 + sizeof(HeapManager.Storage), // 4 |
---|
234 | 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(HeapManager.Storage), // 8 |
---|
235 | 2_621_440, 3_145_728, 3_670_016, 4_194_304 + sizeof(HeapManager.Storage), // 4 |
---|
236 | }; |
---|
237 | |
---|
238 | static_assert( NoBucketSizes == sizeof(bucketSizes) / sizeof(bucketSizes[0]), "size of bucket array wrong" ); |
---|
239 | |
---|
240 | #ifdef FASTLOOKUP |
---|
241 | enum { LookupSizes = 65_536 + sizeof(HeapManager.Storage) }; // number of fast lookup sizes |
---|
242 | static unsigned char lookup[LookupSizes]; // O(1) lookup for small sizes |
---|
243 | #endif // FASTLOOKUP |
---|
244 | |
---|
245 | static int mmapFd = -1; // fake or actual fd for anonymous file |
---|
246 | #ifdef __CFA_DEBUG__ |
---|
247 | static bool heapBoot = 0; // detect recursion during boot |
---|
248 | #endif // __CFA_DEBUG__ |
---|
249 | static HeapManager heapManager __attribute__(( aligned (128) )) @= {}; // size of cache line to prevent false sharing |
---|
250 | |
---|
251 | |
---|
252 | #ifdef __STATISTICS__ |
---|
253 | // Heap statistics counters. |
---|
254 | static unsigned long long int mmap_storage; |
---|
255 | static unsigned int mmap_calls; |
---|
256 | static unsigned long long int munmap_storage; |
---|
257 | static unsigned int munmap_calls; |
---|
258 | static unsigned long long int sbrk_storage; |
---|
259 | static unsigned int sbrk_calls; |
---|
260 | static unsigned long long int malloc_storage; |
---|
261 | static unsigned int malloc_calls; |
---|
262 | static unsigned long long int free_storage; |
---|
263 | static unsigned int free_calls; |
---|
264 | static unsigned long long int calloc_storage; |
---|
265 | static unsigned int calloc_calls; |
---|
266 | static unsigned long long int memalign_storage; |
---|
267 | static unsigned int memalign_calls; |
---|
268 | static unsigned long long int cmemalign_storage; |
---|
269 | static unsigned int cmemalign_calls; |
---|
270 | static unsigned long long int realloc_storage; |
---|
271 | static unsigned int realloc_calls; |
---|
272 | // Statistics file descriptor (changed by malloc_stats_fd). |
---|
273 | static int statfd = STDERR_FILENO; // default stderr |
---|
274 | |
---|
275 | // Use "write" because streams may be shutdown when calls are made. |
---|
276 | static void printStats() { |
---|
277 | char helpText[512]; |
---|
278 | __cfaabi_bits_print_buffer( STDERR_FILENO, helpText, sizeof(helpText), |
---|
279 | "\nHeap statistics:\n" |
---|
280 | " malloc: calls %u / storage %llu\n" |
---|
281 | " calloc: calls %u / storage %llu\n" |
---|
282 | " memalign: calls %u / storage %llu\n" |
---|
283 | " cmemalign: calls %u / storage %llu\n" |
---|
284 | " realloc: calls %u / storage %llu\n" |
---|
285 | " free: calls %u / storage %llu\n" |
---|
286 | " mmap: calls %u / storage %llu\n" |
---|
287 | " munmap: calls %u / storage %llu\n" |
---|
288 | " sbrk: calls %u / storage %llu\n", |
---|
289 | malloc_calls, malloc_storage, |
---|
290 | calloc_calls, calloc_storage, |
---|
291 | memalign_calls, memalign_storage, |
---|
292 | cmemalign_calls, cmemalign_storage, |
---|
293 | realloc_calls, realloc_storage, |
---|
294 | free_calls, free_storage, |
---|
295 | mmap_calls, mmap_storage, |
---|
296 | munmap_calls, munmap_storage, |
---|
297 | sbrk_calls, sbrk_storage |
---|
298 | ); |
---|
299 | } // printStats |
---|
300 | |
---|
301 | static int printStatsXML( FILE * stream ) { // see malloc_info |
---|
302 | char helpText[512]; |
---|
303 | int len = snprintf( helpText, sizeof(helpText), |
---|
304 | "<malloc version=\"1\">\n" |
---|
305 | "<heap nr=\"0\">\n" |
---|
306 | "<sizes>\n" |
---|
307 | "</sizes>\n" |
---|
308 | "<total type=\"malloc\" count=\"%u\" size=\"%llu\"/>\n" |
---|
309 | "<total type=\"calloc\" count=\"%u\" size=\"%llu\"/>\n" |
---|
310 | "<total type=\"memalign\" count=\"%u\" size=\"%llu\"/>\n" |
---|
311 | "<total type=\"cmemalign\" count=\"%u\" size=\"%llu\"/>\n" |
---|
312 | "<total type=\"realloc\" count=\"%u\" size=\"%llu\"/>\n" |
---|
313 | "<total type=\"free\" count=\"%u\" size=\"%llu\"/>\n" |
---|
314 | "<total type=\"mmap\" count=\"%u\" size=\"%llu\"/>\n" |
---|
315 | "<total type=\"munmap\" count=\"%u\" size=\"%llu\"/>\n" |
---|
316 | "<total type=\"sbrk\" count=\"%u\" size=\"%llu\"/>\n" |
---|
317 | "</malloc>", |
---|
318 | malloc_calls, malloc_storage, |
---|
319 | calloc_calls, calloc_storage, |
---|
320 | memalign_calls, memalign_storage, |
---|
321 | cmemalign_calls, cmemalign_storage, |
---|
322 | realloc_calls, realloc_storage, |
---|
323 | free_calls, free_storage, |
---|
324 | mmap_calls, mmap_storage, |
---|
325 | munmap_calls, munmap_storage, |
---|
326 | sbrk_calls, sbrk_storage |
---|
327 | ); |
---|
328 | __cfaabi_bits_write( fileno( stream ), helpText, len ); // ensures all bytes written or exit |
---|
329 | return len; |
---|
330 | } // printStatsXML |
---|
331 | #endif // __STATISTICS__ |
---|
332 | |
---|
333 | |
---|
334 | // static inline void noMemory() { |
---|
335 | // abort( "Heap memory exhausted at %zu bytes.\n" |
---|
336 | // "Possible cause is very large memory allocation and/or large amount of unfreed storage allocated by the program or system/library routines.", |
---|
337 | // ((char *)(sbrk( 0 )) - (char *)(heapManager.heapBegin)) ); |
---|
338 | // } // noMemory |
---|
339 | |
---|
340 | |
---|
341 | static inline void checkAlign( size_t alignment ) { |
---|
342 | if ( alignment < libAlign() || ! libPow2( alignment ) ) { |
---|
343 | abort( "Alignment %zu for memory allocation is less than %d and/or not a power of 2.", alignment, libAlign() ); |
---|
344 | } // if |
---|
345 | } // checkAlign |
---|
346 | |
---|
347 | |
---|
348 | static inline bool setHeapExpand( size_t value ) { |
---|
349 | if ( heapExpand < pageSize ) return true; |
---|
350 | heapExpand = value; |
---|
351 | return false; |
---|
352 | } // setHeapExpand |
---|
353 | |
---|
354 | |
---|
355 | // thunk problem |
---|
356 | size_t Bsearchl( unsigned int key, const unsigned int * vals, size_t dim ) { |
---|
357 | size_t l = 0, m, h = dim; |
---|
358 | while ( l < h ) { |
---|
359 | m = (l + h) / 2; |
---|
360 | if ( (unsigned int &)(vals[m]) < key ) { // cast away const |
---|
361 | l = m + 1; |
---|
362 | } else { |
---|
363 | h = m; |
---|
364 | } // if |
---|
365 | } // while |
---|
366 | return l; |
---|
367 | } // Bsearchl |
---|
368 | |
---|
369 | |
---|
370 | static inline bool setMmapStart( size_t value ) { // true => mmapped, false => sbrk |
---|
371 | if ( value < pageSize || bucketSizes[NoBucketSizes - 1] < value ) return true; |
---|
372 | mmapStart = value; // set global |
---|
373 | |
---|
374 | // find the closest bucket size less than or equal to the mmapStart size |
---|
375 | maxBucketsUsed = Bsearchl( (unsigned int)mmapStart, bucketSizes, NoBucketSizes ); // binary search |
---|
376 | assert( maxBucketsUsed < NoBucketSizes ); // subscript failure ? |
---|
377 | assert( mmapStart <= bucketSizes[maxBucketsUsed] ); // search failure ? |
---|
378 | return false; |
---|
379 | } // setMmapStart |
---|
380 | |
---|
381 | |
---|
382 | static inline void checkHeader( bool check, const char name[], void * addr ) { |
---|
383 | if ( unlikely( check ) ) { // bad address ? |
---|
384 | abort( "Attempt to %s storage %p with address outside the heap.\n" |
---|
385 | "Possible cause is duplicate free on same block or overwriting of memory.", |
---|
386 | name, addr ); |
---|
387 | } // if |
---|
388 | } // checkHeader |
---|
389 | |
---|
390 | |
---|
391 | static inline void fakeHeader( HeapManager.Storage.Header *& header, size_t & alignment ) { |
---|
392 | if ( unlikely( (header->kind.fake.alignment & 1) == 1 ) ) { // fake header ? |
---|
393 | size_t offset = header->kind.fake.offset; |
---|
394 | alignment = header->kind.fake.alignment & -2; // remove flag from value |
---|
395 | #ifdef __CFA_DEBUG__ |
---|
396 | checkAlign( alignment ); // check alignment |
---|
397 | #endif // __CFA_DEBUG__ |
---|
398 | header = (HeapManager.Storage.Header *)((char *)header - offset); |
---|
399 | } // if |
---|
400 | } // fakeHeader |
---|
401 | |
---|
402 | |
---|
403 | // <-------+----------------------------------------------------> bsize (bucket size) |
---|
404 | // |header |addr |
---|
405 | //================================================================================== |
---|
406 | // | alignment |
---|
407 | // <-----------------<------------+-----------------------------> bsize (bucket size) |
---|
408 | // |fake-header | addr |
---|
409 | #define headerAddr( addr ) ((HeapManager.Storage.Header *)( (char *)addr - sizeof(HeapManager.Storage) )) |
---|
410 | |
---|
411 | // <-------<<--------------------- dsize ---------------------->> bsize (bucket size) |
---|
412 | // |header |addr |
---|
413 | //================================================================================== |
---|
414 | // | alignment |
---|
415 | // <------------------------------<<---------- dsize --------->>> bsize (bucket size) |
---|
416 | // |fake-header |addr |
---|
417 | #define dataStorage( bsize, addr, header ) (bsize - ( (char *)addr - (char *)header )) |
---|
418 | |
---|
419 | |
---|
420 | static inline bool headers( const char name[] __attribute__(( unused )), void * addr, HeapManager.Storage.Header *& header, HeapManager.FreeHeader *& freeElem, size_t & size, size_t & alignment ) with ( heapManager ) { |
---|
421 | header = headerAddr( addr ); |
---|
422 | |
---|
423 | if ( unlikely( heapEnd < addr ) ) { // mmapped ? |
---|
424 | fakeHeader( header, alignment ); |
---|
425 | size = header->kind.real.blockSize & -3; // mmap size |
---|
426 | return true; |
---|
427 | } // if |
---|
428 | |
---|
429 | #ifdef __CFA_DEBUG__ |
---|
430 | checkHeader( addr < heapBegin || header < (HeapManager.Storage.Header *)heapBegin, name, addr ); // bad low address ? |
---|
431 | #endif // __CFA_DEBUG__ |
---|
432 | |
---|
433 | // header may be safe to dereference |
---|
434 | fakeHeader( header, alignment ); |
---|
435 | #ifdef __CFA_DEBUG__ |
---|
436 | checkHeader( header < (HeapManager.Storage.Header *)heapBegin || (HeapManager.Storage.Header *)heapEnd < header, name, addr ); // bad address ? (offset could be + or -) |
---|
437 | #endif // __CFA_DEBUG__ |
---|
438 | |
---|
439 | freeElem = (HeapManager.FreeHeader *)((size_t)header->kind.real.home & -3); |
---|
440 | #ifdef __CFA_DEBUG__ |
---|
441 | if ( freeElem < &freeLists[0] || &freeLists[NoBucketSizes] <= freeElem ) { |
---|
442 | abort( "Attempt to %s storage %p with corrupted header.\n" |
---|
443 | "Possible cause is duplicate free on same block or overwriting of header information.", |
---|
444 | name, addr ); |
---|
445 | } // if |
---|
446 | #endif // __CFA_DEBUG__ |
---|
447 | size = freeElem->blockSize; |
---|
448 | return false; |
---|
449 | } // headers |
---|
450 | |
---|
451 | |
---|
452 | static inline void * extend( size_t size ) with ( heapManager ) { |
---|
453 | lock( extlock __cfaabi_dbg_ctx2 ); |
---|
454 | ptrdiff_t rem = heapRemaining - size; |
---|
455 | if ( rem < 0 ) { |
---|
456 | // If the size requested is bigger than the current remaining storage, increase the size of the heap. |
---|
457 | |
---|
458 | size_t increase = libCeiling( size > heapExpand ? size : heapExpand, libAlign() ); |
---|
459 | if ( sbrk( increase ) == (void *)-1 ) { |
---|
460 | unlock( extlock ); |
---|
461 | errno = ENOMEM; |
---|
462 | return 0p; |
---|
463 | } // if |
---|
464 | #ifdef __STATISTICS__ |
---|
465 | sbrk_calls += 1; |
---|
466 | sbrk_storage += increase; |
---|
467 | #endif // __STATISTICS__ |
---|
468 | #ifdef __CFA_DEBUG__ |
---|
469 | // Set new memory to garbage so subsequent uninitialized usages might fail. |
---|
470 | memset( (char *)heapEnd + heapRemaining, '\377', increase ); |
---|
471 | #endif // __CFA_DEBUG__ |
---|
472 | rem = heapRemaining + increase - size; |
---|
473 | } // if |
---|
474 | |
---|
475 | HeapManager.Storage * block = (HeapManager.Storage *)heapEnd; |
---|
476 | heapRemaining = rem; |
---|
477 | heapEnd = (char *)heapEnd + size; |
---|
478 | unlock( extlock ); |
---|
479 | return block; |
---|
480 | } // extend |
---|
481 | |
---|
482 | |
---|
483 | static inline void * doMalloc( size_t size ) with ( heapManager ) { |
---|
484 | HeapManager.Storage * block; // pointer to new block of storage |
---|
485 | |
---|
486 | // Look up size in the size list. Make sure the user request includes space for the header that must be allocated |
---|
487 | // along with the block and is a multiple of the alignment size. |
---|
488 | |
---|
489 | if ( unlikely( size > ~0ul - sizeof(HeapManager.Storage) ) ) return 0p; |
---|
490 | size_t tsize = size + sizeof(HeapManager.Storage); |
---|
491 | if ( likely( tsize < mmapStart ) ) { // small size => sbrk |
---|
492 | size_t posn; |
---|
493 | #ifdef FASTLOOKUP |
---|
494 | if ( tsize < LookupSizes ) posn = lookup[tsize]; |
---|
495 | else |
---|
496 | #endif // FASTLOOKUP |
---|
497 | posn = Bsearchl( (unsigned int)tsize, bucketSizes, (size_t)maxBucketsUsed ); |
---|
498 | HeapManager.FreeHeader * freeElem = &freeLists[posn]; |
---|
499 | // #ifdef FASTLOOKUP |
---|
500 | // if ( tsize < LookupSizes ) |
---|
501 | // freeElem = &freeLists[lookup[tsize]]; |
---|
502 | // else |
---|
503 | // #endif // FASTLOOKUP |
---|
504 | // freeElem = bsearchl( tsize, freeLists, (size_t)maxBucketsUsed ); // binary search |
---|
505 | // HeapManager.FreeHeader * freeElem = |
---|
506 | // #ifdef FASTLOOKUP |
---|
507 | // tsize < LookupSizes ? &freeLists[lookup[tsize]] : |
---|
508 | // #endif // FASTLOOKUP |
---|
509 | // bsearchl( tsize, freeLists, (size_t)maxBucketsUsed ); // binary search |
---|
510 | assert( freeElem <= &freeLists[maxBucketsUsed] ); // subscripting error ? |
---|
511 | assert( tsize <= freeElem->blockSize ); // search failure ? |
---|
512 | tsize = freeElem->blockSize; // total space needed for request |
---|
513 | |
---|
514 | // Spin until the lock is acquired for this particular size of block. |
---|
515 | |
---|
516 | #if defined( SPINLOCK ) |
---|
517 | lock( freeElem->lock __cfaabi_dbg_ctx2 ); |
---|
518 | block = freeElem->freeList; // remove node from stack |
---|
519 | #else |
---|
520 | block = freeElem->freeList.pop(); |
---|
521 | #endif // SPINLOCK |
---|
522 | if ( unlikely( block == 0p ) ) { // no free block ? |
---|
523 | #if defined( SPINLOCK ) |
---|
524 | unlock( freeElem->lock ); |
---|
525 | #endif // SPINLOCK |
---|
526 | |
---|
527 | // Freelist for that size was empty, so carve it out of the heap if there's enough left, or get some more |
---|
528 | // and then carve it off. |
---|
529 | |
---|
530 | block = (HeapManager.Storage *)extend( tsize ); // mutual exclusion on call |
---|
531 | if ( unlikely( block == 0p ) ) return 0p; |
---|
532 | #if defined( SPINLOCK ) |
---|
533 | } else { |
---|
534 | freeElem->freeList = block->header.kind.real.next; |
---|
535 | unlock( freeElem->lock ); |
---|
536 | #endif // SPINLOCK |
---|
537 | } // if |
---|
538 | |
---|
539 | block->header.kind.real.home = freeElem; // pointer back to free list of apropriate size |
---|
540 | } else { // large size => mmap |
---|
541 | if ( unlikely( size > ~0ul - pageSize ) ) return 0p; |
---|
542 | tsize = libCeiling( tsize, pageSize ); // must be multiple of page size |
---|
543 | #ifdef __STATISTICS__ |
---|
544 | __atomic_add_fetch( &mmap_calls, 1, __ATOMIC_SEQ_CST ); |
---|
545 | __atomic_add_fetch( &mmap_storage, tsize, __ATOMIC_SEQ_CST ); |
---|
546 | #endif // __STATISTICS__ |
---|
547 | block = (HeapManager.Storage *)mmap( 0, tsize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, mmapFd, 0 ); |
---|
548 | if ( block == (HeapManager.Storage *)MAP_FAILED ) { |
---|
549 | // Do not call strerror( errno ) as it may call malloc. |
---|
550 | abort( "(HeapManager &)0x%p.doMalloc() : internal error, mmap failure, size:%zu error:%d.", &heapManager, tsize, errno ); |
---|
551 | } // if |
---|
552 | #ifdef __CFA_DEBUG__ |
---|
553 | // Set new memory to garbage so subsequent uninitialized usages might fail. |
---|
554 | memset( block, '\377', tsize ); |
---|
555 | #endif // __CFA_DEBUG__ |
---|
556 | block->header.kind.real.blockSize = tsize; // storage size for munmap |
---|
557 | } // if |
---|
558 | |
---|
559 | void * addr = &(block->data); // adjust off header to user bytes |
---|
560 | |
---|
561 | #ifdef __CFA_DEBUG__ |
---|
562 | assert( ((uintptr_t)addr & (libAlign() - 1)) == 0 ); // minimum alignment ? |
---|
563 | __atomic_add_fetch( &allocFree, tsize, __ATOMIC_SEQ_CST ); |
---|
564 | if ( traceHeap() ) { |
---|
565 | enum { BufferSize = 64 }; |
---|
566 | char helpText[BufferSize]; |
---|
567 | int len = snprintf( helpText, BufferSize, "%p = Malloc( %zu ) (allocated %zu)\n", addr, size, tsize ); |
---|
568 | // int len = snprintf( helpText, BufferSize, "Malloc %p %zu\n", addr, size ); |
---|
569 | __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug |
---|
570 | } // if |
---|
571 | #endif // __CFA_DEBUG__ |
---|
572 | |
---|
573 | return addr; |
---|
574 | } // doMalloc |
---|
575 | |
---|
576 | |
---|
577 | static inline void doFree( void * addr ) with ( heapManager ) { |
---|
578 | #ifdef __CFA_DEBUG__ |
---|
579 | if ( unlikely( heapManager.heapBegin == 0p ) ) { |
---|
580 | abort( "doFree( %p ) : internal error, called before heap is initialized.", addr ); |
---|
581 | } // if |
---|
582 | #endif // __CFA_DEBUG__ |
---|
583 | |
---|
584 | HeapManager.Storage.Header * header; |
---|
585 | HeapManager.FreeHeader * freeElem; |
---|
586 | size_t size, alignment; // not used (see realloc) |
---|
587 | |
---|
588 | if ( headers( "free", addr, header, freeElem, size, alignment ) ) { // mmapped ? |
---|
589 | #ifdef __STATISTICS__ |
---|
590 | __atomic_add_fetch( &munmap_calls, 1, __ATOMIC_SEQ_CST ); |
---|
591 | __atomic_add_fetch( &munmap_storage, size, __ATOMIC_SEQ_CST ); |
---|
592 | #endif // __STATISTICS__ |
---|
593 | if ( munmap( header, size ) == -1 ) { |
---|
594 | #ifdef __CFA_DEBUG__ |
---|
595 | abort( "Attempt to deallocate storage %p not allocated or with corrupt header.\n" |
---|
596 | "Possible cause is invalid pointer.", |
---|
597 | addr ); |
---|
598 | #endif // __CFA_DEBUG__ |
---|
599 | } // if |
---|
600 | } else { |
---|
601 | #ifdef __CFA_DEBUG__ |
---|
602 | // Set free memory to garbage so subsequent usages might fail. |
---|
603 | memset( ((HeapManager.Storage *)header)->data, '\377', freeElem->blockSize - sizeof( HeapManager.Storage ) ); |
---|
604 | #endif // __CFA_DEBUG__ |
---|
605 | |
---|
606 | #ifdef __STATISTICS__ |
---|
607 | free_storage += size; |
---|
608 | #endif // __STATISTICS__ |
---|
609 | #if defined( SPINLOCK ) |
---|
610 | lock( freeElem->lock __cfaabi_dbg_ctx2 ); // acquire spin lock |
---|
611 | header->kind.real.next = freeElem->freeList; // push on stack |
---|
612 | freeElem->freeList = (HeapManager.Storage *)header; |
---|
613 | unlock( freeElem->lock ); // release spin lock |
---|
614 | #else |
---|
615 | freeElem->freeList.push( *(HeapManager.Storage *)header ); |
---|
616 | #endif // SPINLOCK |
---|
617 | } // if |
---|
618 | |
---|
619 | #ifdef __CFA_DEBUG__ |
---|
620 | __atomic_add_fetch( &allocFree, -size, __ATOMIC_SEQ_CST ); |
---|
621 | if ( traceHeap() ) { |
---|
622 | enum { BufferSize = 64 }; |
---|
623 | char helpText[BufferSize]; |
---|
624 | int len = snprintf( helpText, sizeof(helpText), "Free( %p ) size:%zu\n", addr, size ); |
---|
625 | __cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug |
---|
626 | } // if |
---|
627 | #endif // __CFA_DEBUG__ |
---|
628 | } // doFree |
---|
629 | |
---|
630 | |
---|
631 | size_t prtFree( HeapManager & manager ) with ( manager ) { |
---|
632 | size_t total = 0; |
---|
633 | #ifdef __STATISTICS__ |
---|
634 | __cfaabi_bits_acquire(); |
---|
635 | __cfaabi_bits_print_nolock( STDERR_FILENO, "\nBin lists (bin size : free blocks on list)\n" ); |
---|
636 | #endif // __STATISTICS__ |
---|
637 | for ( unsigned int i = 0; i < maxBucketsUsed; i += 1 ) { |
---|
638 | size_t size = freeLists[i].blockSize; |
---|
639 | #ifdef __STATISTICS__ |
---|
640 | unsigned int N = 0; |
---|
641 | #endif // __STATISTICS__ |
---|
642 | |
---|
643 | #if defined( SPINLOCK ) |
---|
644 | for ( HeapManager.Storage * p = freeLists[i].freeList; p != 0p; p = p->header.kind.real.next ) { |
---|
645 | #else |
---|
646 | for ( HeapManager.Storage * p = freeLists[i].freeList.top(); p != 0p; p = p->header.kind.real.next.top ) { |
---|
647 | #endif // SPINLOCK |
---|
648 | total += size; |
---|
649 | #ifdef __STATISTICS__ |
---|
650 | N += 1; |
---|
651 | #endif // __STATISTICS__ |
---|
652 | } // for |
---|
653 | |
---|
654 | #ifdef __STATISTICS__ |
---|
655 | __cfaabi_bits_print_nolock( STDERR_FILENO, "%7zu, %-7u ", size, N ); |
---|
656 | if ( (i + 1) % 8 == 0 ) __cfaabi_bits_print_nolock( STDERR_FILENO, "\n" ); |
---|
657 | #endif // __STATISTICS__ |
---|
658 | } // for |
---|
659 | #ifdef __STATISTICS__ |
---|
660 | __cfaabi_bits_print_nolock( STDERR_FILENO, "\ntotal free blocks:%zu\n", total ); |
---|
661 | __cfaabi_bits_release(); |
---|
662 | #endif // __STATISTICS__ |
---|
663 | return (char *)heapEnd - (char *)heapBegin - total; |
---|
664 | } // prtFree |
---|
665 | |
---|
666 | |
---|
667 | static void ?{}( HeapManager & manager ) with ( manager ) { |
---|
668 | pageSize = sysconf( _SC_PAGESIZE ); |
---|
669 | |
---|
670 | for ( unsigned int i = 0; i < NoBucketSizes; i += 1 ) { // initialize the free lists |
---|
671 | freeLists[i].blockSize = bucketSizes[i]; |
---|
672 | } // for |
---|
673 | |
---|
674 | #ifdef FASTLOOKUP |
---|
675 | unsigned int idx = 0; |
---|
676 | for ( unsigned int i = 0; i < LookupSizes; i += 1 ) { |
---|
677 | if ( i > bucketSizes[idx] ) idx += 1; |
---|
678 | lookup[i] = idx; |
---|
679 | } // for |
---|
680 | #endif // FASTLOOKUP |
---|
681 | |
---|
682 | if ( setMmapStart( default_mmap_start() ) ) { |
---|
683 | abort( "HeapManager : internal error, mmap start initialization failure." ); |
---|
684 | } // if |
---|
685 | heapExpand = default_heap_expansion(); |
---|
686 | |
---|
687 | char * end = (char *)sbrk( 0 ); |
---|
688 | sbrk( (char *)libCeiling( (long unsigned int)end, libAlign() ) - end ); // move start of heap to multiple of alignment |
---|
689 | heapBegin = heapEnd = sbrk( 0 ); // get new start point |
---|
690 | } // HeapManager |
---|
691 | |
---|
692 | |
---|
693 | static void ^?{}( HeapManager & ) { |
---|
694 | #ifdef __STATISTICS__ |
---|
695 | if ( traceHeapTerm() ) { |
---|
696 | printStats(); |
---|
697 | // if ( prtfree() ) prtFree( heapManager, true ); |
---|
698 | } // if |
---|
699 | #endif // __STATISTICS__ |
---|
700 | } // ~HeapManager |
---|
701 | |
---|
702 | |
---|
703 | static void memory_startup( void ) __attribute__(( constructor( STARTUP_PRIORITY_MEMORY ) )); |
---|
704 | void memory_startup( void ) { |
---|
705 | #ifdef __CFA_DEBUG__ |
---|
706 | if ( unlikely( heapBoot ) ) { // check for recursion during system boot |
---|
707 | // DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT. |
---|
708 | abort( "boot() : internal error, recursively invoked during system boot." ); |
---|
709 | } // if |
---|
710 | heapBoot = true; |
---|
711 | #endif // __CFA_DEBUG__ |
---|
712 | |
---|
713 | //assert( heapManager.heapBegin != 0 ); |
---|
714 | //heapManager{}; |
---|
715 | if ( heapManager.heapBegin == 0p ) heapManager{}; |
---|
716 | } // memory_startup |
---|
717 | |
---|
718 | static void memory_shutdown( void ) __attribute__(( destructor( STARTUP_PRIORITY_MEMORY ) )); |
---|
719 | void memory_shutdown( void ) { |
---|
720 | ^heapManager{}; |
---|
721 | } // memory_shutdown |
---|
722 | |
---|
723 | |
---|
724 | static inline void * mallocNoStats( size_t size ) { // necessary for malloc statistics |
---|
725 | //assert( heapManager.heapBegin != 0 ); |
---|
726 | if ( unlikely( heapManager.heapBegin == 0p ) ) heapManager{}; // called before memory_startup ? |
---|
727 | void * addr = doMalloc( size ); |
---|
728 | if ( unlikely( addr == 0p ) ) errno = ENOMEM; // POSIX |
---|
729 | return addr; |
---|
730 | } // mallocNoStats |
---|
731 | |
---|
732 | |
---|
733 | static inline void * callocNoStats( size_t noOfElems, size_t elemSize ) { |
---|
734 | size_t size = noOfElems * elemSize; |
---|
735 | char * addr = (char *)mallocNoStats( size ); |
---|
736 | if ( unlikely( addr == 0p ) ) return 0p; |
---|
737 | |
---|
738 | HeapManager.Storage.Header * header; |
---|
739 | HeapManager.FreeHeader * freeElem; |
---|
740 | size_t bsize, alignment; |
---|
741 | bool mapped __attribute__(( unused )) = headers( "calloc", addr, header, freeElem, bsize, alignment ); |
---|
742 | #ifndef __CFA_DEBUG__ |
---|
743 | // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero. |
---|
744 | if ( ! mapped ) |
---|
745 | #endif // __CFA_DEBUG__ |
---|
746 | // Zero entire data space even when > than size => realloc without a new allocation and zero fill works. |
---|
747 | // <-------00000000000000000000000000000000000000000000000000000> bsize (bucket size) |
---|
748 | // `-header`-addr `-size |
---|
749 | memset( addr, '\0', bsize - sizeof(HeapManager.Storage) ); // set to zeros |
---|
750 | |
---|
751 | header->kind.real.blockSize |= 2; // mark as zero filled |
---|
752 | return addr; |
---|
753 | } // callocNoStats |
---|
754 | |
---|
755 | |
---|
756 | static inline void * memalignNoStats( size_t alignment, size_t size ) { // necessary for malloc statistics |
---|
757 | #ifdef __CFA_DEBUG__ |
---|
758 | checkAlign( alignment ); // check alignment |
---|
759 | #endif // __CFA_DEBUG__ |
---|
760 | |
---|
761 | // if alignment <= default alignment, do normal malloc as two headers are unnecessary |
---|
762 | if ( unlikely( alignment <= libAlign() ) ) return mallocNoStats( size ); |
---|
763 | |
---|
764 | // Allocate enough storage to guarantee an address on the alignment boundary, and sufficient space before it for |
---|
765 | // administrative storage. NOTE, WHILE THERE ARE 2 HEADERS, THE FIRST ONE IS IMPLICITLY CREATED BY DOMALLOC. |
---|
766 | // .-------------v-----------------v----------------v----------, |
---|
767 | // | Real Header | ... padding ... | Fake Header | data ... | |
---|
768 | // `-------------^-----------------^-+--------------^----------' |
---|
769 | // |<--------------------------------' offset/align |<-- alignment boundary |
---|
770 | |
---|
771 | // subtract libAlign() because it is already the minimum alignment |
---|
772 | // add sizeof(Storage) for fake header |
---|
773 | char * addr = (char *)mallocNoStats( size + alignment - libAlign() + sizeof(HeapManager.Storage) ); |
---|
774 | if ( unlikely( addr == 0p ) ) return addr; |
---|
775 | |
---|
776 | // address in the block of the "next" alignment address |
---|
777 | char * user = (char *)libCeiling( (uintptr_t)(addr + sizeof(HeapManager.Storage)), alignment ); |
---|
778 | |
---|
779 | // address of header from malloc |
---|
780 | HeapManager.Storage.Header * realHeader = headerAddr( addr ); |
---|
781 | // address of fake header * before* the alignment location |
---|
782 | HeapManager.Storage.Header * fakeHeader = headerAddr( user ); |
---|
783 | // SKULLDUGGERY: insert the offset to the start of the actual storage block and remember alignment |
---|
784 | fakeHeader->kind.fake.offset = (char *)fakeHeader - (char *)realHeader; |
---|
785 | // SKULLDUGGERY: odd alignment imples fake header |
---|
786 | fakeHeader->kind.fake.alignment = alignment | 1; |
---|
787 | |
---|
788 | return user; |
---|
789 | } // memalignNoStats |
---|
790 | |
---|
791 | |
---|
792 | static inline void * cmemalignNoStats( size_t alignment, size_t noOfElems, size_t elemSize ) { |
---|
793 | size_t size = noOfElems * elemSize; |
---|
794 | char * addr = (char *)memalignNoStats( alignment, size ); |
---|
795 | if ( unlikely( addr == 0p ) ) return 0p; |
---|
796 | HeapManager.Storage.Header * header; |
---|
797 | HeapManager.FreeHeader * freeElem; |
---|
798 | size_t bsize; |
---|
799 | bool mapped __attribute__(( unused )) = headers( "cmemalign", addr, header, freeElem, bsize, alignment ); |
---|
800 | #ifndef __CFA_DEBUG__ |
---|
801 | // Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero. |
---|
802 | if ( ! mapped ) |
---|
803 | #endif // __CFA_DEBUG__ |
---|
804 | memset( addr, '\0', dataStorage( bsize, addr, header ) ); // set to zeros |
---|
805 | header->kind.real.blockSize |= 2; // mark as zero filled |
---|
806 | |
---|
807 | return addr; |
---|
808 | } // cmemalignNoStats |
---|
809 | |
---|
810 | |
---|
811 | // supported mallopt options |
---|
812 | #ifndef M_MMAP_THRESHOLD |
---|
813 | #define M_MMAP_THRESHOLD (-1) |
---|
814 | #endif // M_TOP_PAD |
---|
815 | #ifndef M_TOP_PAD |
---|
816 | #define M_TOP_PAD (-2) |
---|
817 | #endif // M_TOP_PAD |
---|
818 | |
---|
819 | |
---|
820 | extern "C" { |
---|
821 | // The malloc() function allocates size bytes and returns a pointer to the allocated memory. The memory is not |
---|
822 | // initialized. If size is 0, then malloc() returns either 0p, or a unique pointer value that can later be |
---|
823 | // successfully passed to free(). |
---|
824 | void * malloc( size_t size ) { |
---|
825 | #ifdef __STATISTICS__ |
---|
826 | __atomic_add_fetch( &malloc_calls, 1, __ATOMIC_SEQ_CST ); |
---|
827 | __atomic_add_fetch( &malloc_storage, size, __ATOMIC_SEQ_CST ); |
---|
828 | #endif // __STATISTICS__ |
---|
829 | |
---|
830 | return mallocNoStats( size ); |
---|
831 | } // malloc |
---|
832 | |
---|
833 | // The calloc() function allocates memory for an array of nmemb elements of size bytes each and returns a pointer to |
---|
834 | // the allocated memory. The memory is set to zero. If nmemb or size is 0, then calloc() returns either 0p, or a |
---|
835 | // unique pointer value that can later be successfully passed to free(). |
---|
836 | void * calloc( size_t noOfElems, size_t elemSize ) { |
---|
837 | #ifdef __STATISTICS__ |
---|
838 | __atomic_add_fetch( &calloc_calls, 1, __ATOMIC_SEQ_CST ); |
---|
839 | __atomic_add_fetch( &calloc_storage, noOfElems * elemSize, __ATOMIC_SEQ_CST ); |
---|
840 | #endif // __STATISTICS__ |
---|
841 | |
---|
842 | return callocNoStats( noOfElems, elemSize ); |
---|
843 | } // calloc |
---|
844 | |
---|
845 | // The realloc() function changes the size of the memory block pointed to by ptr to size bytes. The contents will be |
---|
846 | // unchanged in the range from the start of the region up to the minimum of the old and new sizes. If the new size |
---|
847 | // is larger than the old size, the added memory will not be initialized. If ptr is 0p, then the call is |
---|
848 | // equivalent to malloc(size), for all values of size; if size is equal to zero, and ptr is not 0p, then the call |
---|
849 | // is equivalent to free(ptr). Unless ptr is 0p, it must have been returned by an earlier call to malloc(), |
---|
850 | // calloc() or realloc(). If the area pointed to was moved, a free(ptr) is done. |
---|
851 | void * realloc( void * oaddr, size_t size ) { |
---|
852 | #ifdef __STATISTICS__ |
---|
853 | __atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST ); |
---|
854 | #endif // __STATISTICS__ |
---|
855 | |
---|
856 | // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned. |
---|
857 | if ( unlikely( size == 0 ) ) { free( oaddr ); return mallocNoStats( size ); } // special cases |
---|
858 | if ( unlikely( oaddr == 0p ) ) return mallocNoStats( size ); |
---|
859 | |
---|
860 | HeapManager.Storage.Header * header; |
---|
861 | HeapManager.FreeHeader * freeElem; |
---|
862 | size_t bsize, oalign = 0; |
---|
863 | headers( "realloc", oaddr, header, freeElem, bsize, oalign ); |
---|
864 | |
---|
865 | size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket |
---|
866 | if ( size <= odsize && odsize <= size * 2 ) { // allow up to 50% wasted storage in smaller size |
---|
867 | // Do not know size of original allocation => cannot do 0 fill for any additional space because do not know |
---|
868 | // where to start filling, i.e., do not overwrite existing values in space. |
---|
869 | // |
---|
870 | // This case does not result in a new profiler entry because the previous one still exists and it must match with |
---|
871 | // the free for this memory. Hence, this realloc does not appear in the profiler output. |
---|
872 | return oaddr; |
---|
873 | } // if |
---|
874 | |
---|
875 | #ifdef __STATISTICS__ |
---|
876 | __atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST ); |
---|
877 | #endif // __STATISTICS__ |
---|
878 | |
---|
879 | // change size and copy old content to new storage |
---|
880 | |
---|
881 | void * naddr; |
---|
882 | if ( unlikely( oalign != 0 ) ) { // previous request memalign? |
---|
883 | if ( unlikely( header->kind.real.blockSize & 2 ) ) { // previous request zero fill |
---|
884 | naddr = cmemalignNoStats( oalign, 1, size ); // create new aligned area |
---|
885 | } else { |
---|
886 | naddr = memalignNoStats( oalign, size ); // create new aligned area |
---|
887 | } // if |
---|
888 | } else { |
---|
889 | if ( unlikely( header->kind.real.blockSize & 2 ) ) { // previous request zero fill |
---|
890 | naddr = callocNoStats( 1, size ); // create new area |
---|
891 | } else { |
---|
892 | naddr = mallocNoStats( size ); // create new area |
---|
893 | } // if |
---|
894 | } // if |
---|
895 | if ( unlikely( naddr == 0p ) ) return 0p; |
---|
896 | |
---|
897 | headers( "realloc", naddr, header, freeElem, bsize, oalign ); |
---|
898 | size_t ndsize = dataStorage( bsize, naddr, header ); // data storage avilable in bucket |
---|
899 | // To preserve prior fill, the entire bucket must be copied versus the size. |
---|
900 | memcpy( naddr, oaddr, MIN( odsize, ndsize ) ); // copy bytes |
---|
901 | free( oaddr ); |
---|
902 | return naddr; |
---|
903 | } // realloc |
---|
904 | |
---|
905 | // The obsolete function memalign() allocates size bytes and returns a pointer to the allocated memory. The memory |
---|
906 | // address will be a multiple of alignment, which must be a power of two. |
---|
907 | void * memalign( size_t alignment, size_t size ) { |
---|
908 | #ifdef __STATISTICS__ |
---|
909 | __atomic_add_fetch( &memalign_calls, 1, __ATOMIC_SEQ_CST ); |
---|
910 | __atomic_add_fetch( &memalign_storage, size, __ATOMIC_SEQ_CST ); |
---|
911 | #endif // __STATISTICS__ |
---|
912 | |
---|
913 | return memalignNoStats( alignment, size ); |
---|
914 | } // memalign |
---|
915 | |
---|
916 | |
---|
917 | // The cmemalign() function is the same as calloc() with memory alignment. |
---|
918 | void * cmemalign( size_t alignment, size_t noOfElems, size_t elemSize ) { |
---|
919 | #ifdef __STATISTICS__ |
---|
920 | __atomic_add_fetch( &cmemalign_calls, 1, __ATOMIC_SEQ_CST ); |
---|
921 | __atomic_add_fetch( &cmemalign_storage, noOfElems * elemSize, __ATOMIC_SEQ_CST ); |
---|
922 | #endif // __STATISTICS__ |
---|
923 | |
---|
924 | return cmemalignNoStats( alignment, noOfElems, elemSize ); |
---|
925 | } // cmemalign |
---|
926 | |
---|
927 | // The function aligned_alloc() is the same as memalign(), except for the added restriction that size should be a |
---|
928 | // multiple of alignment. |
---|
929 | void * aligned_alloc( size_t alignment, size_t size ) { |
---|
930 | return memalign( alignment, size ); |
---|
931 | } // aligned_alloc |
---|
932 | |
---|
933 | |
---|
934 | // The function posix_memalign() allocates size bytes and places the address of the allocated memory in *memptr. The |
---|
935 | // address of the allocated memory will be a multiple of alignment, which must be a power of two and a multiple of |
---|
936 | // sizeof(void *). If size is 0, then posix_memalign() returns either 0p, or a unique pointer value that can later |
---|
937 | // be successfully passed to free(3). |
---|
938 | int posix_memalign( void ** memptr, size_t alignment, size_t size ) { |
---|
939 | if ( alignment < sizeof(void *) || ! libPow2( alignment ) ) return EINVAL; // check alignment |
---|
940 | * memptr = memalign( alignment, size ); |
---|
941 | if ( unlikely( * memptr == 0p ) ) return ENOMEM; |
---|
942 | return 0; |
---|
943 | } // posix_memalign |
---|
944 | |
---|
945 | // The obsolete function valloc() allocates size bytes and returns a pointer to the allocated memory. The memory |
---|
946 | // address will be a multiple of the page size. It is equivalent to memalign(sysconf(_SC_PAGESIZE),size). |
---|
947 | void * valloc( size_t size ) { |
---|
948 | return memalign( pageSize, size ); |
---|
949 | } // valloc |
---|
950 | |
---|
951 | |
---|
952 | // The free() function frees the memory space pointed to by ptr, which must have been returned by a previous call to |
---|
953 | // malloc(), calloc() or realloc(). Otherwise, or if free(ptr) has already been called before, undefined behavior |
---|
954 | // occurs. If ptr is 0p, no operation is performed. |
---|
955 | void free( void * addr ) { |
---|
956 | #ifdef __STATISTICS__ |
---|
957 | __atomic_add_fetch( &free_calls, 1, __ATOMIC_SEQ_CST ); |
---|
958 | #endif // __STATISTICS__ |
---|
959 | |
---|
960 | if ( unlikely( addr == 0p ) ) { // special case |
---|
961 | // #ifdef __CFA_DEBUG__ |
---|
962 | // if ( traceHeap() ) { |
---|
963 | // #define nullmsg "Free( 0x0 ) size:0\n" |
---|
964 | // // Do not debug print free( 0p ), as it can cause recursive entry from sprintf. |
---|
965 | // __cfaabi_dbg_write( nullmsg, sizeof(nullmsg) - 1 ); |
---|
966 | // } // if |
---|
967 | // #endif // __CFA_DEBUG__ |
---|
968 | return; |
---|
969 | } // exit |
---|
970 | |
---|
971 | doFree( addr ); |
---|
972 | } // free |
---|
973 | |
---|
974 | |
---|
975 | // The malloc_alignment() function returns the alignment of the allocation. |
---|
976 | size_t malloc_alignment( void * addr ) { |
---|
977 | if ( unlikely( addr == 0p ) ) return libAlign(); // minimum alignment |
---|
978 | HeapManager.Storage.Header * header = headerAddr( addr ); |
---|
979 | if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ? |
---|
980 | return header->kind.fake.alignment & -2; // remove flag from value |
---|
981 | } else { |
---|
982 | return libAlign (); // minimum alignment |
---|
983 | } // if |
---|
984 | } // malloc_alignment |
---|
985 | |
---|
986 | |
---|
987 | // The malloc_zero_fill() function returns true if the allocation is zero filled, i.e., initially allocated by calloc(). |
---|
988 | bool malloc_zero_fill( void * addr ) { |
---|
989 | if ( unlikely( addr == 0p ) ) return false; // null allocation is not zero fill |
---|
990 | HeapManager.Storage.Header * header = headerAddr( addr ); |
---|
991 | if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ? |
---|
992 | header = (HeapManager.Storage.Header *)((char *)header - header->kind.fake.offset); |
---|
993 | } // if |
---|
994 | return (header->kind.real.blockSize & 2) != 0; // zero filled (calloc/cmemalign) ? |
---|
995 | } // malloc_zero_fill |
---|
996 | |
---|
997 | |
---|
998 | // The malloc_usable_size() function returns the number of usable bytes in the block pointed to by ptr, a pointer to |
---|
999 | // a block of memory allocated by malloc(3) or a related function. |
---|
1000 | size_t malloc_usable_size( void * addr ) { |
---|
1001 | if ( unlikely( addr == 0p ) ) return 0; // null allocation has 0 size |
---|
1002 | HeapManager.Storage.Header * header; |
---|
1003 | HeapManager.FreeHeader * freeElem; |
---|
1004 | size_t bsize, alignment; |
---|
1005 | |
---|
1006 | headers( "malloc_usable_size", addr, header, freeElem, bsize, alignment ); |
---|
1007 | return dataStorage( bsize, addr, header ); // data storage in bucket |
---|
1008 | } // malloc_usable_size |
---|
1009 | |
---|
1010 | |
---|
1011 | // The malloc_stats() function prints (on default standard error) statistics about memory allocated by malloc(3) and |
---|
1012 | // related functions. |
---|
1013 | void malloc_stats( void ) { |
---|
1014 | #ifdef __STATISTICS__ |
---|
1015 | printStats(); |
---|
1016 | if ( prtFree() ) prtFree( heapManager ); |
---|
1017 | #endif // __STATISTICS__ |
---|
1018 | } // malloc_stats |
---|
1019 | |
---|
1020 | // The malloc_stats_fd() function changes the file descripter where malloc_stats() writes the statistics. |
---|
1021 | int malloc_stats_fd( int fd __attribute__(( unused )) ) { |
---|
1022 | #ifdef __STATISTICS__ |
---|
1023 | int temp = statfd; |
---|
1024 | statfd = fd; |
---|
1025 | return temp; |
---|
1026 | #else |
---|
1027 | return -1; |
---|
1028 | #endif // __STATISTICS__ |
---|
1029 | } // malloc_stats_fd |
---|
1030 | |
---|
1031 | |
---|
1032 | // The mallopt() function adjusts parameters that control the behavior of the memory-allocation functions (see |
---|
1033 | // malloc(3)). The param argument specifies the parameter to be modified, and value specifies the new value for that |
---|
1034 | // parameter. |
---|
1035 | int mallopt( int option, int value ) { |
---|
1036 | choose( option ) { |
---|
1037 | case M_TOP_PAD: |
---|
1038 | if ( setHeapExpand( value ) ) return 1; |
---|
1039 | case M_MMAP_THRESHOLD: |
---|
1040 | if ( setMmapStart( value ) ) return 1; |
---|
1041 | } // switch |
---|
1042 | return 0; // error, unsupported |
---|
1043 | } // mallopt |
---|
1044 | |
---|
1045 | // The malloc_trim() function attempts to release free memory at the top of the heap (by calling sbrk(2) with a |
---|
1046 | // suitable argument). |
---|
1047 | int malloc_trim( size_t ) { |
---|
1048 | return 0; // => impossible to release memory |
---|
1049 | } // malloc_trim |
---|
1050 | |
---|
1051 | |
---|
1052 | // The malloc_info() function exports an XML string that describes the current state of the memory-allocation |
---|
1053 | // implementation in the caller. The string is printed on the file stream stream. The exported string includes |
---|
1054 | // information about all arenas (see malloc(3)). |
---|
1055 | int malloc_info( int options, FILE * stream ) { |
---|
1056 | if ( options != 0 ) { errno = EINVAL; return -1; } |
---|
1057 | return printStatsXML( stream ); |
---|
1058 | } // malloc_info |
---|
1059 | |
---|
1060 | |
---|
1061 | // The malloc_get_state() function records the current state of all malloc(3) internal bookkeeping variables (but |
---|
1062 | // not the actual contents of the heap or the state of malloc_hook(3) functions pointers). The state is recorded in |
---|
1063 | // a system-dependent opaque data structure dynamically allocated via malloc(3), and a pointer to that data |
---|
1064 | // structure is returned as the function result. (It is the caller's responsibility to free(3) this memory.) |
---|
1065 | void * malloc_get_state( void ) { |
---|
1066 | return 0p; // unsupported |
---|
1067 | } // malloc_get_state |
---|
1068 | |
---|
1069 | |
---|
1070 | // The malloc_set_state() function restores the state of all malloc(3) internal bookkeeping variables to the values |
---|
1071 | // recorded in the opaque data structure pointed to by state. |
---|
1072 | int malloc_set_state( void * ptr ) { |
---|
1073 | return 0; // unsupported |
---|
1074 | } // malloc_set_state |
---|
1075 | } // extern "C" |
---|
1076 | |
---|
1077 | |
---|
1078 | // Must have CFA linkage to overload with C linkage realloc. |
---|
1079 | void * realloc( void * oaddr, size_t nalign, size_t size ) { |
---|
1080 | #ifdef __STATISTICS__ |
---|
1081 | __atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST ); |
---|
1082 | #endif // __STATISTICS__ |
---|
1083 | |
---|
1084 | // If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned. |
---|
1085 | if ( unlikely( size == 0 ) ) { free( oaddr ); return mallocNoStats( size ); } // special cases |
---|
1086 | if ( unlikely( oaddr == 0p ) ) return mallocNoStats( size ); |
---|
1087 | |
---|
1088 | if ( unlikely( nalign == 0 ) ) nalign = libAlign(); // reset alignment to minimum |
---|
1089 | #ifdef __CFA_DEBUG__ |
---|
1090 | else |
---|
1091 | checkAlign( nalign ); // check alignment |
---|
1092 | #endif // __CFA_DEBUG__ |
---|
1093 | |
---|
1094 | HeapManager.Storage.Header * header; |
---|
1095 | HeapManager.FreeHeader * freeElem; |
---|
1096 | size_t bsize, oalign = 0; |
---|
1097 | headers( "realloc", oaddr, header, freeElem, bsize, oalign ); |
---|
1098 | size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket |
---|
1099 | |
---|
1100 | if ( oalign != 0 && (uintptr_t)oaddr % nalign == 0 ) { // has alignment and just happens to work out |
---|
1101 | headerAddr( oaddr )->kind.fake.alignment = nalign | 1; // update alignment (could be the same) |
---|
1102 | return realloc( oaddr, size ); |
---|
1103 | } // if |
---|
1104 | |
---|
1105 | #ifdef __STATISTICS__ |
---|
1106 | __atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST ); |
---|
1107 | #endif // __STATISTICS__ |
---|
1108 | |
---|
1109 | // change size and copy old content to new storage |
---|
1110 | |
---|
1111 | void * naddr; |
---|
1112 | if ( unlikely( header->kind.real.blockSize & 2 ) ) { // previous request zero fill |
---|
1113 | naddr = cmemalignNoStats( nalign, 1, size ); // create new aligned area |
---|
1114 | } else { |
---|
1115 | naddr = memalignNoStats( nalign, size ); // create new aligned area |
---|
1116 | } // if |
---|
1117 | |
---|
1118 | headers( "realloc", naddr, header, freeElem, bsize, oalign ); |
---|
1119 | size_t ndsize = dataStorage( bsize, naddr, header ); // data storage avilable in bucket |
---|
1120 | // To preserve prior fill, the entire bucket must be copied versus the size. |
---|
1121 | memcpy( naddr, oaddr, MIN( odsize, ndsize ) ); // copy bytes |
---|
1122 | free( oaddr ); |
---|
1123 | return naddr; |
---|
1124 | } // realloc |
---|
1125 | |
---|
1126 | |
---|
1127 | // Local Variables: // |
---|
1128 | // tab-width: 4 // |
---|
1129 | // compile-command: "cfa -nodebug -O2 heap.cfa" // |
---|
1130 | // End: // |
---|