Context Navigation

heap.cfa @ 7a15b7e

ADTast-experimentalenumforall-pointer-decaypthread-emulationqualifiedEnum

Last change on this file since 7a15b7e was 578ec01c, checked in by Peter A. Buhr <pabuhr@…>, 3 years ago
update heap statistics to new format (incomplete)
Property mode set to `100644`
File size: 56.3 KB

Line
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.cfa --
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 : Sun Jan 2 23:29:41 2022
13	// Update Count : 1058
14	//
15
16	#include <unistd.h> // sbrk, sysconf
17	#include <stdlib.h> // EXIT_FAILURE
18	#include <stdbool.h> // true, false
19	#include <stdio.h> // snprintf, fileno
20	#include <errno.h> // errno
21	#include <string.h> // memset, memcpy
22	#include <limits.h> // ULONG_MAX
23	#include <malloc.h> // memalign, malloc_usable_size
24	#include <sys/mman.h> // mmap, munmap
25
26	#include "bits/align.hfa" // libAlign
27	#include "bits/defs.hfa" // likely, unlikely
28	#include "bits/locks.hfa" // __spinlock_t
29	#include "startup.hfa" // STARTUP_PRIORITY_MEMORY
30	#include "math.hfa" // min
31	#include "bitmanip.hfa" // is_pow2, ceiling2
32
33	static bool traceHeap = false;
34
35	inline bool traceHeap() { return traceHeap; }
36
37	bool traceHeapOn() {
38	bool temp = traceHeap;
39	traceHeap = true;
40	return temp;
41	} // traceHeapOn
42
43	bool traceHeapOff() {
44	bool temp = traceHeap;
45	traceHeap = false;
46	return temp;
47	} // traceHeapOff
48
49	bool traceHeapTerm() { return false; }
50
51
52	static bool prtFree = false;
53
54	bool prtFree() {
55	return prtFree;
56	} // prtFree
57
58	bool prtFreeOn() {
59	bool temp = prtFree;
60	prtFree = true;
61	return temp;
62	} // prtFreeOn
63
64	bool prtFreeOff() {
65	bool temp = prtFree;
66	prtFree = false;
67	return temp;
68	} // prtFreeOff
69
70
71	enum {
72	// Define the default extension heap amount in units of bytes. When the uC++ supplied heap reaches the brk address,
73	// the brk address is extended by the extension amount.
74	__CFA_DEFAULT_HEAP_EXPANSION__ = (10 * 1024 * 1024),
75
76	// Define the mmap crossover point during allocation. Allocations less than this amount are allocated from buckets;
77	// values greater than or equal to this value are mmap from the operating system.
78	__CFA_DEFAULT_MMAP_START__ = (512 * 1024 + 1),
79	};
80
81	size_t default_mmap_start() __attribute__(( weak )) {
82	return __CFA_DEFAULT_MMAP_START__;
83	} // default_mmap_start
84
85	size_t default_heap_expansion() __attribute__(( weak )) {
86	return __CFA_DEFAULT_HEAP_EXPANSION__;
87	} // default_heap_expansion
88
89
90	#ifdef __CFA_DEBUG__
91	static size_t allocUnfreed; // running total of allocations minus frees
92
93	static void prtUnfreed() {
94	if ( allocUnfreed != 0 ) {
95	// DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT.
96	char helpText[512];
97	int len = snprintf( helpText, sizeof(helpText), "CFA warning (UNIX pid:%ld) : program terminating with %zu(0x%zx) bytes of storage allocated but not freed.\n"
98	"Possible cause is unfreed storage allocated by the program or system/library routines called from the program.\n",
99	(long int)getpid(), allocUnfreed, allocUnfreed ); // always print the UNIX pid
100	__cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
101	} // if
102	} // prtUnfreed
103
104	extern int cfa_main_returned; // from interpose.cfa
105	extern "C" {
106	void heapAppStart() { // called by __cfaabi_appready_startup
107	allocUnfreed = 0;
108	} // heapAppStart
109
110	void heapAppStop() { // called by __cfaabi_appready_startdown
111	fclose( stdin ); fclose( stdout );
112	if ( cfa_main_returned ) prtUnfreed(); // do not check unfreed storage if exit called
113	} // heapAppStop
114	} // extern "C"
115	#endif // __CFA_DEBUG__
116
117
118	// statically allocated variables => zero filled.
119	size_t __page_size; // architecture pagesize
120	int __map_prot; // common mmap/mprotect protection
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 == SPINLOCK
130	#elif BUCKETLOCK == LOCKFREE
131	#include <stackLockFree.hfa>
132	#else
133	#error undefined lock type for bucket lock
134	#endif // LOCKFREE
135
136	// Recursive definitions: HeapManager needs size of bucket array and bucket area needs sizeof HeapManager storage.
137	// Break recusion by hardcoding number of buckets and statically checking number is correct after bucket array defined.
138	enum { NoBucketSizes = 91 }; // number of buckets sizes
139
140	struct HeapManager {
141	struct Storage {
142	struct Header { // header
143	union Kind {
144	struct RealHeader {
145	union {
146	struct { // 4-byte word => 8-byte header, 8-byte word => 16-byte header
147	#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4
148	uint64_t padding; // unused, force home/blocksize to overlay alignment in fake header
149	#endif // __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4
150
151	union {
152	// FreeHeader * home; // allocated block points back to home locations (must overlay alignment)
153	// 2nd low-order bit => zero filled
154	void * home; // allocated block points back to home locations (must overlay alignment)
155	size_t blockSize; // size for munmap (must overlay alignment)
156	#if BUCKETLOCK == SPINLOCK
157	Storage * next; // freed block points next freed block of same size
158	#endif // SPINLOCK
159	};
160	size_t size; // allocation size in bytes
161
162	#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4
163	uint64_t padding; // unused, force home/blocksize to overlay alignment in fake header
164	#endif // __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4
165	};
166	#if BUCKETLOCK == LOCKFREE
167	Link(Storage) next; // freed block points next freed block of same size (double-wide)
168	#endif // LOCKFREE
169	};
170	} real; // RealHeader
171
172	struct FakeHeader {
173	#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
174	uint32_t alignment; // 1st low-order bit => fake header & alignment
175	#endif // __ORDER_LITTLE_ENDIAN__
176
177	uint32_t offset;
178
179	#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
180	uint32_t alignment; // low-order bits of home/blockSize used for tricks
181	#endif // __ORDER_BIG_ENDIAN__
182	} fake; // FakeHeader
183	} kind; // Kind
184	} header; // Header
185	char pad[libAlign() - sizeof( Header )];
186	char data[0]; // storage
187	}; // Storage
188
189	static_assert( libAlign() >= sizeof( Storage ), "libAlign() < sizeof( Storage )" );
190
191	struct FreeHeader {
192	#if BUCKETLOCK == SPINLOCK
193	__spinlock_t lock; // must be first field for alignment
194	Storage * freeList;
195	#else
196	StackLF(Storage) freeList;
197	#endif // BUCKETLOCK
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	#if BUCKETLOCK == LOCKFREE
211	static inline {
212	Link(HeapManager.Storage) * ?`next( HeapManager.Storage * this ) { return &this->header.kind.real.next; }
213	void ?{}( HeapManager.FreeHeader & ) {}
214	void ^?{}( HeapManager.FreeHeader & ) {}
215	} // distribution
216	#endif // LOCKFREE
217
218	static inline size_t getKey( const HeapManager.FreeHeader & freeheader ) { return freeheader.blockSize; }
219
220
221	#define FASTLOOKUP
222	#define __STATISTICS__
223
224	// Size of array must harmonize with NoBucketSizes and individual bucket sizes must be multiple of 16.
225	// Smaller multiples of 16 and powers of 2 are common allocation sizes, so make them generate the minimum required bucket size.
226	// malloc(0) returns 0p, so no bucket is necessary for 0 bytes returning an address that can be freed.
227	static const unsigned int bucketSizes[] @= { // different bucket sizes
228	16 + sizeof(HeapManager.Storage), 32 + sizeof(HeapManager.Storage), 48 + sizeof(HeapManager.Storage), 64 + sizeof(HeapManager.Storage), // 4
229	96 + sizeof(HeapManager.Storage), 112 + sizeof(HeapManager.Storage), 128 + sizeof(HeapManager.Storage), // 3
230	160, 192, 224, 256 + sizeof(HeapManager.Storage), // 4
231	320, 384, 448, 512 + sizeof(HeapManager.Storage), // 4
232	640, 768, 896, 1_024 + sizeof(HeapManager.Storage), // 4
233	1_536, 2_048 + sizeof(HeapManager.Storage), // 2
234	2_560, 3_072, 3_584, 4_096 + sizeof(HeapManager.Storage), // 4
235	6_144, 8_192 + sizeof(HeapManager.Storage), // 2
236	9_216, 10_240, 11_264, 12_288, 13_312, 14_336, 15_360, 16_384 + sizeof(HeapManager.Storage), // 8
237	18_432, 20_480, 22_528, 24_576, 26_624, 28_672, 30_720, 32_768 + sizeof(HeapManager.Storage), // 8
238	36_864, 40_960, 45_056, 49_152, 53_248, 57_344, 61_440, 65_536 + sizeof(HeapManager.Storage), // 8
239	73_728, 81_920, 90_112, 98_304, 106_496, 114_688, 122_880, 131_072 + sizeof(HeapManager.Storage), // 8
240	147_456, 163_840, 180_224, 196_608, 212_992, 229_376, 245_760, 262_144 + sizeof(HeapManager.Storage), // 8
241	294_912, 327_680, 360_448, 393_216, 425_984, 458_752, 491_520, 524_288 + sizeof(HeapManager.Storage), // 8
242	655_360, 786_432, 917_504, 1_048_576 + sizeof(HeapManager.Storage), // 4
243	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
244	2_621_440, 3_145_728, 3_670_016, 4_194_304 + sizeof(HeapManager.Storage), // 4
245	};
246
247	static_assert( NoBucketSizes == sizeof(bucketSizes) / sizeof(bucketSizes[0] ), "size of bucket array wrong" );
248
249	#ifdef FASTLOOKUP
250	enum { LookupSizes = 65_536 + sizeof(HeapManager.Storage) }; // number of fast lookup sizes
251	static unsigned char lookup[LookupSizes]; // O(1) lookup for small sizes
252	#endif // FASTLOOKUP
253
254	static const off_t mmapFd = -1; // fake or actual fd for anonymous file
255	#ifdef __CFA_DEBUG__
256	static bool heapBoot = 0; // detect recursion during boot
257	#endif // __CFA_DEBUG__
258
259	// The constructor for heapManager is called explicitly in memory_startup.
260	static HeapManager heapManager __attribute__(( aligned (128) )) @= {}; // size of cache line to prevent false sharing
261
262
263	#ifdef __STATISTICS__
264	// Heap statistics counters.
265	static unsigned int malloc_calls, malloc_0_calls;
266	static unsigned long long int malloc_storage_request, malloc_storage_alloc;
267	static unsigned int aalloc_calls, aalloc_0_calls;
268	static unsigned long long int aalloc_storage_request, aalloc_storage_alloc;
269	static unsigned int calloc_calls, calloc_0_calls;
270	static unsigned long long int calloc_storage_request, calloc_storage_alloc;
271	static unsigned int memalign_calls, memalign_0_calls;
272	static unsigned long long int memalign_storage_request, memalign_storage_alloc;
273	static unsigned int amemalign_calls, amemalign_0_calls;
274	static unsigned long long int amemalign_storage_request, amemalign_storage_alloc;
275	static unsigned int cmemalign_calls, cmemalign_0_calls;
276	static unsigned long long int cmemalign_storage_request, cmemalign_storage_alloc;
277	static unsigned int resize_calls, resize_0_calls;
278	static unsigned long long int resize_storage_request, resize_storage_alloc;
279	static unsigned int realloc_calls, realloc_0_calls;
280	static unsigned long long int realloc_storage_request, realloc_storage_alloc;
281	static unsigned int free_calls, free_null_calls;
282	static unsigned long long int free_storage_request, free_storage_alloc;
283	static unsigned int mmap_calls;
284	static unsigned long long int mmap_storage_request, mmap_storage_alloc;
285	static unsigned int munmap_calls;
286	static unsigned long long int munmap_storage_request, munmap_storage_alloc;
287	static unsigned int sbrk_calls;
288	static unsigned long long int sbrk_storage;
289	// Statistics file descriptor (changed by malloc_stats_fd).
290	static int stats_fd = STDERR_FILENO; // default stderr
291
292	// Use "write" because streams may be shutdown when calls are made.
293	static void printStats() {
294	char helpText[1024];
295	__cfaabi_bits_print_buffer( STDERR_FILENO, helpText, sizeof(helpText),
296	"\nHeap statistics: (storage request / allocation + header)\n"
297	" malloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n"
298	" aalloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n"
299	" calloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n"
300	" memalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n"
301	" amemalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n"
302	" cmemalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n"
303	" resize >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n"
304	" realloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n"
305	" free !null calls %'u; null calls %'u; storage %'llu / %'llu bytes\n"
306	" sbrk calls %'u; storage %'llu bytes\n"
307	" mmap calls %'u; storage %'llu / %'llu bytes\n"
308	" munmap calls %'u; storage %'llu / %'llu bytes\n",
309	malloc_calls, malloc_0_calls, malloc_storage_request, malloc_storage_alloc,
310	aalloc_calls, aalloc_0_calls, aalloc_storage_request, aalloc_storage_alloc,
311	calloc_calls, calloc_0_calls, calloc_storage_request, calloc_storage_alloc,
312	memalign_calls, memalign_0_calls, memalign_storage_request, memalign_storage_alloc,
313	amemalign_calls, amemalign_0_calls, amemalign_storage_request, amemalign_storage_alloc,
314	cmemalign_calls, cmemalign_0_calls, cmemalign_storage_request, cmemalign_storage_alloc,
315	resize_calls, resize_0_calls, resize_storage_request, resize_storage_alloc,
316	realloc_calls, realloc_0_calls, realloc_storage_request, realloc_storage_alloc,
317	free_calls, free_null_calls, free_storage_request, free_storage_alloc,
318	sbrk_calls, sbrk_storage,
319	mmap_calls, mmap_storage_request, mmap_storage_alloc,
320	munmap_calls, munmap_storage_request, munmap_storage_alloc
321	);
322	} // printStats
323
324	static int printStatsXML( FILE * stream ) { // see malloc_info
325	char helpText[1024];
326	int len = snprintf( helpText, sizeof(helpText),
327	"<malloc version=\"1\">\n"
328	"<heap nr=\"0\">\n"
329	"<sizes>\n"
330	"</sizes>\n"
331	"<total type=\"malloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
332	"<total type=\"aalloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
333	"<total type=\"calloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
334	"<total type=\"memalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
335	"<total type=\"amemalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
336	"<total type=\"cmemalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
337	"<total type=\"resize\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
338	"<total type=\"realloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
339	"<total type=\"free\" !null=\"%'u;\" 0 null=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
340	"<total type=\"sbrk\" count=\"%'u;\" size=\"%'llu\"/> bytes\n"
341	"<total type=\"mmap\" count=\"%'u;\" size=\"%'llu / %'llu\" / > bytes\n"
342	"<total type=\"munmap\" count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n"
343	"</malloc>",
344	malloc_calls, malloc_0_calls, malloc_storage_request, malloc_storage_alloc,
345	aalloc_calls, aalloc_0_calls, aalloc_storage_request, aalloc_storage_alloc,
346	calloc_calls, calloc_0_calls, calloc_storage_request, calloc_storage_alloc,
347	memalign_calls, memalign_0_calls, memalign_storage_request, memalign_storage_alloc,
348	amemalign_calls, amemalign_0_calls, amemalign_storage_request, amemalign_storage_alloc,
349	cmemalign_calls, cmemalign_0_calls, cmemalign_storage_request, cmemalign_storage_alloc,
350	resize_calls, resize_0_calls, resize_storage_request, resize_storage_alloc,
351	realloc_calls, realloc_0_calls, realloc_storage_request, realloc_storage_alloc,
352	free_calls, free_null_calls, free_storage_request, free_storage_alloc,
353	sbrk_calls, sbrk_storage,
354	mmap_calls, mmap_storage_request, mmap_storage_alloc,
355	munmap_calls, munmap_storage_request, munmap_storage_alloc
356	);
357	__cfaabi_bits_write( fileno( stream ), helpText, len ); // ensures all bytes written or exit
358	return len;
359	} // printStatsXML
360	#endif // __STATISTICS__
361
362
363	// thunk problem
364	size_t Bsearchl( unsigned int key, const unsigned int * vals, size_t dim ) {
365	size_t l = 0, m, h = dim;
366	while ( l < h ) {
367	m = (l + h) / 2;
368	if ( (unsigned int &)(vals[m]) < key ) { // cast away const
369	l = m + 1;
370	} else {
371	h = m;
372	} // if
373	} // while
374	return l;
375	} // Bsearchl
376
377
378	static inline bool setMmapStart( size_t value ) { // true => mmapped, false => sbrk
379	if ( value < __page_size \|\| bucketSizes[NoBucketSizes - 1] < value ) return false;
380	mmapStart = value; // set global
381
382	// find the closest bucket size less than or equal to the mmapStart size
383	maxBucketsUsed = Bsearchl( (unsigned int)mmapStart, bucketSizes, NoBucketSizes ); // binary search
384	assert( maxBucketsUsed < NoBucketSizes ); // subscript failure ?
385	assert( mmapStart <= bucketSizes[maxBucketsUsed] ); // search failure ?
386	return true;
387	} // setMmapStart
388
389
390	// <-------+----------------------------------------------------> bsize (bucket size)
391	// \|header \|addr
392	//==================================================================================
393	// align/offset \|
394	// <-----------------<------------+-----------------------------> bsize (bucket size)
395	// \|fake-header \| addr
396	#define headerAddr( addr ) ((HeapManager.Storage.Header )( (char )addr - sizeof(HeapManager.Storage) ))
397	#define realHeader( header ) ((HeapManager.Storage.Header )((char )header - header->kind.fake.offset))
398
399	// <-------<<--------------------- dsize ---------------------->> bsize (bucket size)
400	// \|header \|addr
401	//==================================================================================
402	// align/offset \|
403	// <------------------------------<<---------- dsize --------->>> bsize (bucket size)
404	// \|fake-header \|addr
405	#define dataStorage( bsize, addr, header ) (bsize - ( (char )addr - (char )header ))
406
407
408	static inline void checkAlign( size_t alignment ) {
409	if ( alignment < libAlign() \|\| ! is_pow2( alignment ) ) {
410	abort( "Alignment %zu for memory allocation is less than %d and/or not a power of 2.", alignment, libAlign() );
411	} // if
412	} // checkAlign
413
414
415	static inline void checkHeader( bool check, const char name[], void * addr ) {
416	if ( unlikely( check ) ) { // bad address ?
417	abort( "Attempt to %s storage %p with address outside the heap.\n"
418	"Possible cause is duplicate free on same block or overwriting of memory.",
419	name, addr );
420	} // if
421	} // checkHeader
422
423
424	static inline void fakeHeader( HeapManager.Storage.Header *& header, size_t & alignment ) {
425	if ( unlikely( (header->kind.fake.alignment & 1) == 1 ) ) { // fake header ?
426	alignment = header->kind.fake.alignment & -2; // remove flag from value
427	#ifdef __CFA_DEBUG__
428	checkAlign( alignment ); // check alignment
429	#endif // __CFA_DEBUG__
430	header = realHeader( header ); // backup from fake to real header
431	} else {
432	alignment = libAlign(); // => no fake header
433	} // if
434	} // fakeHeader
435
436
437	static inline bool headers( const char name[] __attribute__(( unused )), void * addr, HeapManager.Storage.Header & header, HeapManager.FreeHeader & freeElem,
438	size_t & size, size_t & alignment ) with( heapManager ) {
439	header = headerAddr( addr );
440
441	if ( unlikely( addr < heapBegin \|\| heapEnd < addr ) ) { // mmapped ?
442	fakeHeader( header, alignment );
443	size = header->kind.real.blockSize & -3; // mmap size
444	return true;
445	} // if
446
447	#ifdef __CFA_DEBUG__
448	checkHeader( header < (HeapManager.Storage.Header *)heapBegin, name, addr ); // bad low address ?
449	#endif // __CFA_DEBUG__
450
451	// header may be safe to dereference
452	fakeHeader( header, alignment );
453	#ifdef __CFA_DEBUG__
454	checkHeader( header < (HeapManager.Storage.Header )heapBegin \|\| (HeapManager.Storage.Header )heapEnd < header, name, addr ); // bad address ? (offset could be + or -)
455	#endif // __CFA_DEBUG__
456
457	freeElem = (HeapManager.FreeHeader *)((size_t)header->kind.real.home & -3);
458	#ifdef __CFA_DEBUG__
459	if ( freeElem < &freeLists[0] \|\| &freeLists[NoBucketSizes] <= freeElem ) {
460	abort( "Attempt to %s storage %p with corrupted header.\n"
461	"Possible cause is duplicate free on same block or overwriting of header information.",
462	name, addr );
463	} // if
464	#endif // __CFA_DEBUG__
465	size = freeElem->blockSize;
466	return false;
467	} // headers
468
469	// #ifdef __CFA_DEBUG__
470	// #if __SIZEOF_POINTER__ == 4
471	// #define MASK 0xdeadbeef
472	// #else
473	// #define MASK 0xdeadbeefdeadbeef
474	// #endif
475	// #define STRIDE size_t
476
477	// static void * Memset( void * addr, STRIDE size ) { // debug only
478	// if ( size % sizeof(STRIDE) != 0 ) abort( "Memset() : internal error, size %zd not multiple of %zd.", size, sizeof(STRIDE) );
479	// if ( (STRIDE)addr % sizeof(STRIDE) != 0 ) abort( "Memset() : internal error, addr %p not multiple of %zd.", addr, sizeof(STRIDE) );
480
481	// STRIDE * end = (STRIDE *)addr + size / sizeof(STRIDE);
482	// for ( STRIDE * p = (STRIDE )addr; p < end; p += 1 ) p = MASK;
483	// return addr;
484	// } // Memset
485	// #endif // __CFA_DEBUG__
486
487
488	#define NO_MEMORY_MSG "insufficient heap memory available for allocating %zd new bytes."
489
490	static inline void * extend( size_t size ) with( heapManager ) {
491	lock( extlock __cfaabi_dbg_ctx2 );
492	ptrdiff_t rem = heapRemaining - size;
493	if ( rem < 0 ) {
494	// If the size requested is bigger than the current remaining storage, increase the size of the heap.
495
496	size_t increase = ceiling2( size > heapExpand ? size : heapExpand, __page_size );
497	// Do not call abort or strerror( errno ) as they may call malloc.
498	if ( sbrk( increase ) == (void *)-1 ) { // failed, no memory ?
499	unlock( extlock );
500	__cfaabi_bits_print_nolock( STDERR_FILENO, NO_MEMORY_MSG, size );
501	_exit( EXIT_FAILURE ); // give up
502	} // if
503	// Make storage executable for thunks.
504	if ( mprotect( (char *)heapEnd + heapRemaining, increase, __map_prot ) ) {
505	unlock( extlock );
506	__cfaabi_bits_print_nolock( STDERR_FILENO, "extend() : internal error, mprotect failure, heapEnd:%p size:%zd, errno:%d.\n", heapEnd, increase, errno );
507	_exit( EXIT_FAILURE );
508	} // if
509	#ifdef __STATISTICS__
510	sbrk_calls += 1;
511	sbrk_storage += increase;
512	#endif // __STATISTICS__
513	#ifdef __CFA_DEBUG__
514	// Set new memory to garbage so subsequent uninitialized usages might fail.
515	memset( (char *)heapEnd + heapRemaining, '\xde', increase );
516	//Memset( (char *)heapEnd + heapRemaining, increase );
517	#endif // __CFA_DEBUG__
518	rem = heapRemaining + increase - size;
519	} // if
520
521	HeapManager.Storage * block = (HeapManager.Storage *)heapEnd;
522	heapRemaining = rem;
523	heapEnd = (char *)heapEnd + size;
524	unlock( extlock );
525	return block;
526	} // extend
527
528
529	static inline void * doMalloc( size_t size ) with( heapManager ) {
530	HeapManager.Storage * block; // pointer to new block of storage
531
532	// Look up size in the size list. Make sure the user request includes space for the header that must be allocated
533	// along with the block and is a multiple of the alignment size.
534
535	if ( unlikely( size > ULONG_MAX - sizeof(HeapManager.Storage) ) ) return 0p;
536	size_t tsize = size + sizeof(HeapManager.Storage);
537	if ( likely( tsize < mmapStart ) ) { // small size => sbrk
538	size_t posn;
539	#ifdef FASTLOOKUP
540	if ( tsize < LookupSizes ) posn = lookup[tsize];
541	else
542	#endif // FASTLOOKUP
543	posn = Bsearchl( (unsigned int)tsize, bucketSizes, (size_t)maxBucketsUsed );
544	HeapManager.FreeHeader * freeElem = &freeLists[posn];
545	verify( freeElem <= &freeLists[maxBucketsUsed] ); // subscripting error ?
546	verify( tsize <= freeElem->blockSize ); // search failure ?
547	tsize = freeElem->blockSize; // total space needed for request
548
549	// Spin until the lock is acquired for this particular size of block.
550
551	#if BUCKETLOCK == SPINLOCK
552	lock( freeElem->lock __cfaabi_dbg_ctx2 );
553	block = freeElem->freeList; // remove node from stack
554	#else
555	block = pop( freeElem->freeList );
556	#endif // BUCKETLOCK
557	if ( unlikely( block == 0p ) ) { // no free block ?
558	#if BUCKETLOCK == SPINLOCK
559	unlock( freeElem->lock );
560	#endif // BUCKETLOCK
561
562	// Freelist for that size was empty, so carve it out of the heap if there's enough left, or get some more
563	// and then carve it off.
564
565	block = (HeapManager.Storage *)extend( tsize ); // mutual exclusion on call
566	#if BUCKETLOCK == SPINLOCK
567	} else {
568	freeElem->freeList = block->header.kind.real.next;
569	unlock( freeElem->lock );
570	#endif // BUCKETLOCK
571	} // if
572
573	block->header.kind.real.home = freeElem; // pointer back to free list of apropriate size
574	} else { // large size => mmap
575	if ( unlikely( size > ULONG_MAX - __page_size ) ) return 0p;
576	tsize = ceiling2( tsize, __page_size ); // must be multiple of page size
577	#ifdef __STATISTICS__
578	__atomic_add_fetch( &mmap_calls, 1, __ATOMIC_SEQ_CST );
579	__atomic_add_fetch( &mmap_storage_request, size, __ATOMIC_SEQ_CST );
580	__atomic_add_fetch( &mmap_storage_alloc, tsize, __ATOMIC_SEQ_CST );
581	#endif // __STATISTICS__
582
583	block = (HeapManager.Storage *)mmap( 0, tsize, __map_prot, MAP_PRIVATE \| MAP_ANONYMOUS, mmapFd, 0 );
584	if ( block == (HeapManager.Storage *)MAP_FAILED ) { // failed ?
585	if ( errno == ENOMEM ) abort( NO_MEMORY_MSG, tsize ); // no memory
586	// Do not call strerror( errno ) as it may call malloc.
587	abort( "(HeapManager &)0x%p.doMalloc() : internal error, mmap failure, size:%zu errno:%d.", &heapManager, tsize, errno );
588	} //if
589	#ifdef __CFA_DEBUG__
590	// Set new memory to garbage so subsequent uninitialized usages might fail.
591	memset( block, '\xde', tsize );
592	//Memset( block, tsize );
593	#endif // __CFA_DEBUG__
594	block->header.kind.real.blockSize = tsize; // storage size for munmap
595	} // if
596
597	block->header.kind.real.size = size; // store allocation size
598	void * addr = &(block->data); // adjust off header to user bytes
599	verify( ((uintptr_t)addr & (libAlign() - 1)) == 0 ); // minimum alignment ?
600
601	#ifdef __CFA_DEBUG__
602	__atomic_add_fetch( &allocUnfreed, tsize, __ATOMIC_SEQ_CST );
603	if ( traceHeap() ) {
604	enum { BufferSize = 64 };
605	char helpText[BufferSize];
606	int len = snprintf( helpText, BufferSize, "%p = Malloc( %zu ) (allocated %zu)\n", addr, size, tsize );
607	__cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
608	} // if
609	#endif // __CFA_DEBUG__
610
611	return addr;
612	} // doMalloc
613
614
615	static inline void doFree( void * addr ) with( heapManager ) {
616	#ifdef __CFA_DEBUG__
617	if ( unlikely( heapManager.heapBegin == 0p ) ) {
618	abort( "doFree( %p ) : internal error, called before heap is initialized.", addr );
619	} // if
620	#endif // __CFA_DEBUG__
621
622	HeapManager.Storage.Header * header;
623	HeapManager.FreeHeader * freeElem;
624	size_t size, alignment; // not used (see realloc)
625
626	if ( headers( "free", addr, header, freeElem, size, alignment ) ) { // mmapped ?
627	#ifdef __STATISTICS__
628	__atomic_add_fetch( &munmap_calls, 1, __ATOMIC_SEQ_CST );
629	__atomic_add_fetch( &munmap_storage_request, header->kind.real.size, __ATOMIC_SEQ_CST );
630	__atomic_add_fetch( &munmap_storage_alloc, size, __ATOMIC_SEQ_CST );
631	#endif // __STATISTICS__
632	if ( munmap( header, size ) == -1 ) {
633	abort( "Attempt to deallocate storage %p not allocated or with corrupt header.\n"
634	"Possible cause is invalid pointer.",
635	addr );
636	} // if
637	} else {
638	#ifdef __CFA_DEBUG__
639	// Set free memory to garbage so subsequent usages might fail.
640	memset( ((HeapManager.Storage *)header)->data, '\xde', freeElem->blockSize - sizeof( HeapManager.Storage ) );
641	//Memset( ((HeapManager.Storage *)header)->data, freeElem->blockSize - sizeof( HeapManager.Storage ) );
642	#endif // __CFA_DEBUG__
643
644	#ifdef __STATISTICS__
645	__atomic_add_fetch( &free_calls, 1, __ATOMIC_SEQ_CST );
646	__atomic_add_fetch( &free_storage_request, header->kind.real.size, __ATOMIC_SEQ_CST );
647	__atomic_add_fetch( &free_storage_alloc, size, __ATOMIC_SEQ_CST );
648	#endif // __STATISTICS__
649
650	#if BUCKETLOCK == SPINLOCK
651	lock( freeElem->lock __cfaabi_dbg_ctx2 ); // acquire spin lock
652	header->kind.real.next = freeElem->freeList; // push on stack
653	freeElem->freeList = (HeapManager.Storage *)header;
654	unlock( freeElem->lock ); // release spin lock
655	#else
656	push( freeElem->freeList, (HeapManager.Storage )header );
657	#endif // BUCKETLOCK
658	} // if
659
660	#ifdef __CFA_DEBUG__
661	__atomic_add_fetch( &allocUnfreed, -size, __ATOMIC_SEQ_CST );
662	if ( traceHeap() ) {
663	char helpText[64];
664	int len = snprintf( helpText, sizeof(helpText), "Free( %p ) size:%zu\n", addr, size );
665	__cfaabi_bits_write( STDERR_FILENO, helpText, len ); // print debug/nodebug
666	} // if
667	#endif // __CFA_DEBUG__
668	} // doFree
669
670
671	size_t prtFree( HeapManager & manager ) with( manager ) {
672	size_t total = 0;
673	#ifdef __STATISTICS__
674	__cfaabi_bits_acquire();
675	__cfaabi_bits_print_nolock( STDERR_FILENO, "\nBin lists (bin size : free blocks on list)\n" );
676	#endif // __STATISTICS__
677	for ( unsigned int i = 0; i < maxBucketsUsed; i += 1 ) {
678	size_t size = freeLists[i].blockSize;
679	#ifdef __STATISTICS__
680	unsigned int N = 0;
681	#endif // __STATISTICS__
682
683	#if BUCKETLOCK == SPINLOCK
684	for ( HeapManager.Storage * p = freeLists[i].freeList; p != 0p; p = p->header.kind.real.next ) {
685	#else
686	for(;;) {
687	// for ( HeapManager.Storage * p = top( freeLists[i].freeList ); p != 0p; p = (p)`next->top ) {
688	// for ( HeapManager.Storage * p = top( freeLists[i].freeList ); p != 0p; /* p = getNext( p )->top */) {
689	// HeapManager.Storage * temp = p->header.kind.real.next.top; // FIX ME: direct assignent fails, initialization works`
690	// typeof(p) temp = (( p )`next)->top; // FIX ME: direct assignent fails, initialization works`
691	// p = temp;
692	#endif // BUCKETLOCK
693	total += size;
694	#ifdef __STATISTICS__
695	N += 1;
696	#endif // __STATISTICS__
697	} // for
698
699	#ifdef __STATISTICS__
700	__cfaabi_bits_print_nolock( STDERR_FILENO, "%7zu, %-7u ", size, N );
701	if ( (i + 1) % 8 == 0 ) __cfaabi_bits_print_nolock( STDERR_FILENO, "\n" );
702	#endif // __STATISTICS__
703	} // for
704	#ifdef __STATISTICS__
705	__cfaabi_bits_print_nolock( STDERR_FILENO, "\ntotal free blocks:%zu\n", total );
706	__cfaabi_bits_release();
707	#endif // __STATISTICS__
708	return (char )heapEnd - (char )heapBegin - total;
709	} // prtFree
710
711
712	static void ?{}( HeapManager & manager ) with( manager ) {
713	__page_size = sysconf( _SC_PAGESIZE );
714	__map_prot = PROT_READ \| PROT_WRITE \| PROT_EXEC;
715
716	for ( unsigned int i = 0; i < NoBucketSizes; i += 1 ) { // initialize the free lists
717	freeLists[i].blockSize = bucketSizes[i];
718	} // for
719
720	#ifdef FASTLOOKUP
721	unsigned int idx = 0;
722	for ( unsigned int i = 0; i < LookupSizes; i += 1 ) {
723	if ( i > bucketSizes[idx] ) idx += 1;
724	lookup[i] = idx;
725	} // for
726	#endif // FASTLOOKUP
727
728	if ( ! setMmapStart( default_mmap_start() ) ) {
729	abort( "HeapManager : internal error, mmap start initialization failure." );
730	} // if
731	heapExpand = default_heap_expansion();
732
733	char * end = (char *)sbrk( 0 );
734	heapBegin = heapEnd = sbrk( (char *)ceiling2( (long unsigned int)end, __page_size ) - end ); // move start of heap to multiple of alignment
735	} // HeapManager
736
737
738	static void ^?{}( HeapManager & ) {
739	#ifdef __STATISTICS__
740	if ( traceHeapTerm() ) {
741	printStats();
742	// prtUnfreed() called in heapAppStop()
743	} // if
744	#endif // __STATISTICS__
745	} // ~HeapManager
746
747
748	static void memory_startup( void ) __attribute__(( constructor( STARTUP_PRIORITY_MEMORY ) ));
749	void memory_startup( void ) {
750	#ifdef __CFA_DEBUG__
751	if ( heapBoot ) { // check for recursion during system boot
752	abort( "boot() : internal error, recursively invoked during system boot." );
753	} // if
754	heapBoot = true;
755	#endif // __CFA_DEBUG__
756
757	//verify( heapManager.heapBegin != 0 );
758	//heapManager{};
759	if ( heapManager.heapBegin == 0p ) heapManager{}; // sanity check
760	} // memory_startup
761
762	static void memory_shutdown( void ) __attribute__(( destructor( STARTUP_PRIORITY_MEMORY ) ));
763	void memory_shutdown( void ) {
764	^heapManager{};
765	} // memory_shutdown
766
767
768	static inline void * mallocNoStats( size_t size ) { // necessary for malloc statistics
769	verify( heapManager.heapBegin != 0p ); // called before memory_startup ?
770	if ( unlikely( size ) == 0 ) return 0p; // 0 BYTE ALLOCATION RETURNS NULL POINTER
771
772	#if __SIZEOF_POINTER__ == 8
773	verify( size < ((typeof(size_t))1 << 48) );
774	#endif // __SIZEOF_POINTER__ == 8
775	return doMalloc( size );
776	} // mallocNoStats
777
778
779	static inline void * memalignNoStats( size_t alignment, size_t size ) {
780	if ( unlikely( size ) == 0 ) return 0p; // 0 BYTE ALLOCATION RETURNS NULL POINTER
781
782	#ifdef __CFA_DEBUG__
783	checkAlign( alignment ); // check alignment
784	#endif // __CFA_DEBUG__
785
786	// if alignment <= default alignment, do normal malloc as two headers are unnecessary
787	if ( unlikely( alignment <= libAlign() ) ) return mallocNoStats( size );
788
789	// Allocate enough storage to guarantee an address on the alignment boundary, and sufficient space before it for
790	// administrative storage. NOTE, WHILE THERE ARE 2 HEADERS, THE FIRST ONE IS IMPLICITLY CREATED BY DOMALLOC.
791	// .-------------v-----------------v----------------v----------,
792	// \| Real Header \| ... padding ... \| Fake Header \| data ... \|
793	// `-------------^-----------------^-+--------------^----------'
794	// \|<--------------------------------' offset/align \|<-- alignment boundary
795
796	// subtract libAlign() because it is already the minimum alignment
797	// add sizeof(Storage) for fake header
798	char * addr = (char *)mallocNoStats( size + alignment - libAlign() + sizeof(HeapManager.Storage) );
799
800	// address in the block of the "next" alignment address
801	char * user = (char *)ceiling2( (uintptr_t)(addr + sizeof(HeapManager.Storage)), alignment );
802
803	// address of header from malloc
804	HeapManager.Storage.Header * realHeader = headerAddr( addr );
805	realHeader->kind.real.size = size; // correct size to eliminate above alignment offset
806	// address of fake header * before* the alignment location
807	HeapManager.Storage.Header * fakeHeader = headerAddr( user );
808	// SKULLDUGGERY: insert the offset to the start of the actual storage block and remember alignment
809	fakeHeader->kind.fake.offset = (char )fakeHeader - (char )realHeader;
810	// SKULLDUGGERY: odd alignment imples fake header
811	fakeHeader->kind.fake.alignment = alignment \| 1;
812
813	return user;
814	} // memalignNoStats
815
816
817	extern "C" {
818	// Allocates size bytes and returns a pointer to the allocated memory. The contents are undefined. If size is 0,
819	// then malloc() returns a unique pointer value that can later be successfully passed to free().
820	void * malloc( size_t size ) {
821	#ifdef __STATISTICS__
822	if ( likely( size > 0 ) ) {
823	__atomic_add_fetch( &malloc_calls, 1, __ATOMIC_SEQ_CST );
824	__atomic_add_fetch( &malloc_storage_request, size, __ATOMIC_SEQ_CST );
825	} else {
826	__atomic_add_fetch( &malloc_0_calls, 1, __ATOMIC_SEQ_CST );
827	} // if
828	#endif // __STATISTICS__
829
830	return mallocNoStats( size );
831	} // malloc
832
833
834	// Same as malloc() except size bytes is an array of dim elements each of elemSize bytes.
835	void * aalloc( size_t dim, size_t elemSize ) {
836	size_t size = dim * elemSize;
837	#ifdef __STATISTICS__
838	if ( likely( size > 0 ) ) {
839	__atomic_add_fetch( &aalloc_calls, 1, __ATOMIC_SEQ_CST );
840	__atomic_add_fetch( &aalloc_storage_request, size, __ATOMIC_SEQ_CST );
841	} else {
842	__atomic_add_fetch( &aalloc_0_calls, 1, __ATOMIC_SEQ_CST );
843	} // if
844	#endif // __STATISTICS__
845
846	return mallocNoStats( size );
847	} // aalloc
848
849
850	// Same as aalloc() with memory set to zero.
851	void * calloc( size_t dim, size_t elemSize ) {
852	size_t size = dim * elemSize;
853	if ( unlikely( size ) == 0 ) { // 0 BYTE ALLOCATION RETURNS NULL POINTER
854	#ifdef __STATISTICS__
855	__atomic_add_fetch( &calloc_0_calls, 1, __ATOMIC_SEQ_CST );
856	#endif // __STATISTICS__
857	return 0p;
858	} // if
859	#ifdef __STATISTICS__
860	__atomic_add_fetch( &calloc_calls, 1, __ATOMIC_SEQ_CST );
861	__atomic_add_fetch( &calloc_storage_request, dim * elemSize, __ATOMIC_SEQ_CST );
862	#endif // __STATISTICS__
863
864	char * addr = (char *)mallocNoStats( size );
865
866	HeapManager.Storage.Header * header;
867	HeapManager.FreeHeader * freeElem;
868	size_t bsize, alignment;
869
870	#ifndef __CFA_DEBUG__
871	bool mapped =
872	#endif // __CFA_DEBUG__
873	headers( "calloc", addr, header, freeElem, bsize, alignment );
874
875	#ifndef __CFA_DEBUG__
876	// Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
877	if ( ! mapped )
878	#endif // __CFA_DEBUG__
879	// <-------0000000000000000000000000000UUUUUUUUUUUUUUUUUUUUUUUUU> bsize (bucket size) U => undefined
880	// `-header`-addr `-size
881	memset( addr, '\0', size ); // set to zeros
882
883	header->kind.real.blockSize \|= 2; // mark as zero filled
884	return addr;
885	} // calloc
886
887
888	// Change the size of the memory block pointed to by oaddr to size bytes. The contents are undefined. If oaddr is
889	// 0p, then the call is equivalent to malloc(size), for all values of size; if size is equal to zero, and oaddr is
890	// not 0p, then the call is equivalent to free(oaddr). Unless oaddr is 0p, it must have been returned by an earlier
891	// call to malloc(), alloc(), calloc() or realloc(). If the area pointed to was moved, a free(oaddr) is done.
892	void * resize( void * oaddr, size_t size ) {
893	// If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
894	if ( unlikely( size == 0 ) ) { // special cases
895	#ifdef __STATISTICS__
896	__atomic_add_fetch( &resize_0_calls, 1, __ATOMIC_SEQ_CST );
897	#endif // __STATISTICS__
898	free( oaddr );
899	return 0p;
900	} // if
901	#ifdef __STATISTICS__
902	__atomic_add_fetch( &resize_calls, 1, __ATOMIC_SEQ_CST );
903	#endif // __STATISTICS__
904
905	if ( unlikely( oaddr == 0p ) ) {
906	#ifdef __STATISTICS__
907	__atomic_add_fetch( &resize_storage_request, size, __ATOMIC_SEQ_CST );
908	#endif // __STATISTICS__
909	return mallocNoStats( size );
910	} // if
911
912	HeapManager.Storage.Header * header;
913	HeapManager.FreeHeader * freeElem;
914	size_t bsize, oalign;
915	headers( "resize", oaddr, header, freeElem, bsize, oalign );
916
917	size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket
918	// same size, DO NOT preserve STICKY PROPERTIES.
919	if ( oalign == libAlign() && size <= odsize && odsize <= size * 2 ) { // allow 50% wasted storage for smaller size
920	header->kind.real.blockSize &= -2; // no alignment and turn off 0 fill
921	header->kind.real.size = size; // reset allocation size
922	return oaddr;
923	} // if
924
925	#ifdef __STATISTICS__
926	__atomic_add_fetch( &resize_storage_request, size, __ATOMIC_SEQ_CST );
927	#endif // __STATISTICS__
928
929	// change size, DO NOT preserve STICKY PROPERTIES.
930	free( oaddr );
931	return mallocNoStats( size ); // create new area
932	} // resize
933
934
935	// Same as resize() but the contents are unchanged in the range from the start of the region up to the minimum of
936	// the old and new sizes.
937	void * realloc( void * oaddr, size_t size ) {
938	// If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
939	if ( unlikely( size == 0 ) ) { // special cases
940	#ifdef __STATISTICS__
941	__atomic_add_fetch( &realloc_0_calls, 1, __ATOMIC_SEQ_CST );
942	#endif // __STATISTICS__
943	free( oaddr );
944	return 0p;
945	} // if
946	#ifdef __STATISTICS__
947	__atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
948	#endif // __STATISTICS__
949
950	if ( unlikely( oaddr == 0p ) ) {
951	#ifdef __STATISTICS__
952	__atomic_add_fetch( &realloc_storage_request, size, __ATOMIC_SEQ_CST );
953	#endif // __STATISTICS__
954	return mallocNoStats( size );
955	} // if
956
957	HeapManager.Storage.Header * header;
958	HeapManager.FreeHeader * freeElem;
959	size_t bsize, oalign;
960	headers( "realloc", oaddr, header, freeElem, bsize, oalign );
961
962	size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket
963	size_t osize = header->kind.real.size; // old allocation size
964	bool ozfill = (header->kind.real.blockSize & 2); // old allocation zero filled
965	if ( unlikely( size <= odsize ) && odsize <= size * 2 ) { // allow up to 50% wasted storage
966	header->kind.real.size = size; // reset allocation size
967	if ( unlikely( ozfill ) && size > osize ) { // previous request zero fill and larger ?
968	memset( (char *)oaddr + osize, '\0', size - osize ); // initialize added storage
969	} // if
970	return oaddr;
971	} // if
972
973	#ifdef __STATISTICS__
974	__atomic_add_fetch( &realloc_storage_request, size, __ATOMIC_SEQ_CST );
975	#endif // __STATISTICS__
976
977	// change size and copy old content to new storage
978
979	void * naddr;
980	if ( likely( oalign == libAlign() ) ) { // previous request not aligned ?
981	naddr = mallocNoStats( size ); // create new area
982	} else {
983	naddr = memalignNoStats( oalign, size ); // create new aligned area
984	} // if
985
986	headers( "realloc", naddr, header, freeElem, bsize, oalign );
987	memcpy( naddr, oaddr, min( osize, size ) ); // copy bytes
988	free( oaddr );
989
990	if ( unlikely( ozfill ) ) { // previous request zero fill ?
991	header->kind.real.blockSize \|= 2; // mark new request as zero filled
992	if ( size > osize ) { // previous request larger ?
993	memset( (char *)naddr + osize, '\0', size - osize ); // initialize added storage
994	} // if
995	} // if
996	return naddr;
997	} // realloc
998
999
1000	// Same as malloc() except the memory address is a multiple of alignment, which must be a power of two. (obsolete)
1001	void * memalign( size_t alignment, size_t size ) {
1002	#ifdef __STATISTICS__
1003	if ( likely( size > 0 ) ) {
1004	__atomic_add_fetch( &memalign_calls, 1, __ATOMIC_SEQ_CST );
1005	__atomic_add_fetch( &memalign_storage_request, size, __ATOMIC_SEQ_CST );
1006	} else {
1007	__atomic_add_fetch( &memalign_0_calls, 1, __ATOMIC_SEQ_CST );
1008	} // if
1009	#endif // __STATISTICS__
1010
1011	return memalignNoStats( alignment, size );
1012	} // memalign
1013
1014
1015	// Same as aalloc() with memory alignment.
1016	void * amemalign( size_t alignment, size_t dim, size_t elemSize ) {
1017	size_t size = dim * elemSize;
1018	#ifdef __STATISTICS__
1019	if ( likely( size > 0 ) ) {
1020	__atomic_add_fetch( &cmemalign_calls, 1, __ATOMIC_SEQ_CST );
1021	__atomic_add_fetch( &cmemalign_storage_request, size, __ATOMIC_SEQ_CST );
1022	} else {
1023	__atomic_add_fetch( &cmemalign_0_calls, 1, __ATOMIC_SEQ_CST );
1024	} // if
1025	#endif // __STATISTICS__
1026
1027	return memalignNoStats( alignment, size );
1028	} // amemalign
1029
1030
1031	// Same as calloc() with memory alignment.
1032	void * cmemalign( size_t alignment, size_t dim, size_t elemSize ) {
1033	size_t size = dim * elemSize;
1034	if ( unlikely( size ) == 0 ) { // 0 BYTE ALLOCATION RETURNS NULL POINTER
1035	#ifdef __STATISTICS__
1036	__atomic_add_fetch( &cmemalign_0_calls, 1, __ATOMIC_SEQ_CST );
1037	#endif // __STATISTICS__
1038	return 0p;
1039	} // if
1040	#ifdef __STATISTICS__
1041	__atomic_add_fetch( &cmemalign_calls, 1, __ATOMIC_SEQ_CST );
1042	__atomic_add_fetch( &cmemalign_storage_request, dim * elemSize, __ATOMIC_SEQ_CST );
1043	#endif // __STATISTICS__
1044
1045	char * addr = (char *)memalignNoStats( alignment, size );
1046
1047	HeapManager.Storage.Header * header;
1048	HeapManager.FreeHeader * freeElem;
1049	size_t bsize;
1050
1051	#ifndef __CFA_DEBUG__
1052	bool mapped =
1053	#endif // __CFA_DEBUG__
1054	headers( "cmemalign", addr, header, freeElem, bsize, alignment );
1055
1056	// Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
1057	#ifndef __CFA_DEBUG__
1058	if ( ! mapped )
1059	#endif // __CFA_DEBUG__
1060	// <-------0000000000000000000000000000UUUUUUUUUUUUUUUUUUUUUUUUU> bsize (bucket size) U => undefined
1061	// `-header`-addr `-size
1062	memset( addr, '\0', size ); // set to zeros
1063
1064	header->kind.real.blockSize \|= 2; // mark as zero filled
1065	return addr;
1066	} // cmemalign
1067
1068
1069	// Same as memalign(), but ISO/IEC 2011 C11 Section 7.22.2 states: the value of size shall be an integral multiple
1070	// of alignment. This requirement is universally ignored.
1071	void * aligned_alloc( size_t alignment, size_t size ) {
1072	return memalign( alignment, size );
1073	} // aligned_alloc
1074
1075
1076	// Allocates size bytes and places the address of the allocated memory in *memptr. The address of the allocated
1077	// memory shall be a multiple of alignment, which must be a power of two and a multiple of sizeof(void *). If size
1078	// is 0, then posix_memalign() returns either 0p, or a unique pointer value that can later be successfully passed to
1079	// free(3).
1080	int posix_memalign( void ** memptr, size_t alignment, size_t size ) {
1081	if ( alignment < libAlign() \|\| ! is_pow2( alignment ) ) return EINVAL; // check alignment
1082	* memptr = memalign( alignment, size );
1083	return 0;
1084	} // posix_memalign
1085
1086
1087	// Allocates size bytes and returns a pointer to the allocated memory. The memory address shall be a multiple of the
1088	// page size. It is equivalent to memalign(sysconf(_SC_PAGESIZE),size).
1089	void * valloc( size_t size ) {
1090	return memalign( __page_size, size );
1091	} // valloc
1092
1093
1094	// Same as valloc but rounds size to multiple of page size.
1095	void * pvalloc( size_t size ) {
1096	return memalign( __page_size, ceiling2( size, __page_size ) );
1097	} // pvalloc
1098
1099
1100	// Frees the memory space pointed to by ptr, which must have been returned by a previous call to malloc(), calloc()
1101	// or realloc(). Otherwise, or if free(ptr) has already been called before, undefined behaviour occurs. If ptr is
1102	// 0p, no operation is performed.
1103	void free( void * addr ) {
1104	if ( unlikely( addr == 0p ) ) { // special case
1105	#ifdef __STATISTICS__
1106	__atomic_add_fetch( &free_null_calls, 1, __ATOMIC_SEQ_CST );
1107	#endif // __STATISTICS__
1108
1109	// #ifdef __CFA_DEBUG__
1110	// if ( traceHeap() ) {
1111	// #define nullmsg "Free( 0x0 ) size:0\n"
1112	// // Do not debug print free( 0p ), as it can cause recursive entry from sprintf.
1113	// __cfaabi_dbg_write( nullmsg, sizeof(nullmsg) - 1 );
1114	// } // if
1115	// #endif // __CFA_DEBUG__
1116	return;
1117	} // exit
1118
1119	doFree( addr );
1120	} // free
1121
1122
1123	// Returns the alignment of an allocation.
1124	size_t malloc_alignment( void * addr ) {
1125	if ( unlikely( addr == 0p ) ) return libAlign(); // minimum alignment
1126	HeapManager.Storage.Header * header = headerAddr( addr );
1127	if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
1128	return header->kind.fake.alignment & -2; // remove flag from value
1129	} else {
1130	return libAlign(); // minimum alignment
1131	} // if
1132	} // malloc_alignment
1133
1134
1135	// Set the alignment for an the allocation and return previous alignment or 0 if no alignment.
1136	size_t malloc_alignment_set$( void * addr, size_t alignment ) {
1137	if ( unlikely( addr == 0p ) ) return libAlign(); // minimum alignment
1138	size_t ret;
1139	HeapManager.Storage.Header * header = headerAddr( addr );
1140	if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
1141	ret = header->kind.fake.alignment & -2; // remove flag from old value
1142	header->kind.fake.alignment = alignment \| 1; // add flag to new value
1143	} else {
1144	ret = 0; // => no alignment to change
1145	} // if
1146	return ret;
1147	} // malloc_alignment_set$
1148
1149
1150	// Returns true if the allocation is zero filled, e.g., allocated by calloc().
1151	bool malloc_zero_fill( void * addr ) {
1152	if ( unlikely( addr == 0p ) ) return false; // null allocation is not zero fill
1153	HeapManager.Storage.Header * header = headerAddr( addr );
1154	if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
1155	header = realHeader( header ); // backup from fake to real header
1156	} // if
1157	return (header->kind.real.blockSize & 2) != 0; // zero filled ?
1158	} // malloc_zero_fill
1159
1160	// Set allocation is zero filled and return previous zero filled.
1161	bool malloc_zero_fill_set$( void * addr ) {
1162	if ( unlikely( addr == 0p ) ) return false; // null allocation is not zero fill
1163	HeapManager.Storage.Header * header = headerAddr( addr );
1164	if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
1165	header = realHeader( header ); // backup from fake to real header
1166	} // if
1167	bool ret = (header->kind.real.blockSize & 2) != 0; // zero filled ?
1168	header->kind.real.blockSize \|= 2; // mark as zero filled
1169	return ret;
1170	} // malloc_zero_fill_set$
1171
1172
1173	// Returns original total allocation size (not bucket size) => array size is dimension * sizeif(T).
1174	size_t malloc_size( void * addr ) {
1175	if ( unlikely( addr == 0p ) ) return 0; // null allocation has zero size
1176	HeapManager.Storage.Header * header = headerAddr( addr );
1177	if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
1178	header = realHeader( header ); // backup from fake to real header
1179	} // if
1180	return header->kind.real.size;
1181	} // malloc_size
1182
1183	// Set allocation size and return previous size.
1184	size_t malloc_size_set$( void * addr, size_t size ) {
1185	if ( unlikely( addr == 0p ) ) return 0; // null allocation has 0 size
1186	HeapManager.Storage.Header * header = headerAddr( addr );
1187	if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
1188	header = realHeader( header ); // backup from fake to real header
1189	} // if
1190	size_t ret = header->kind.real.size;
1191	header->kind.real.size = size;
1192	return ret;
1193	} // malloc_size_set$
1194
1195
1196	// Returns the number of usable bytes in the block pointed to by ptr, a pointer to a block of memory allocated by
1197	// malloc or a related function.
1198	size_t malloc_usable_size( void * addr ) {
1199	if ( unlikely( addr == 0p ) ) return 0; // null allocation has 0 size
1200	HeapManager.Storage.Header * header;
1201	HeapManager.FreeHeader * freeElem;
1202	size_t bsize, alignment;
1203
1204	headers( "malloc_usable_size", addr, header, freeElem, bsize, alignment );
1205	return dataStorage( bsize, addr, header ); // data storage in bucket
1206	} // malloc_usable_size
1207
1208
1209	// Prints (on default standard error) statistics about memory allocated by malloc and related functions.
1210	void malloc_stats( void ) {
1211	#ifdef __STATISTICS__
1212	printStats();
1213	if ( prtFree() ) prtFree( heapManager );
1214	#endif // __STATISTICS__
1215	} // malloc_stats
1216
1217
1218	// Changes the file descripter where malloc_stats() writes statistics.
1219	int malloc_stats_fd( int fd __attribute__(( unused )) ) {
1220	#ifdef __STATISTICS__
1221	int temp = stats_fd;
1222	stats_fd = fd;
1223	return temp;
1224	#else
1225	return -1;
1226	#endif // __STATISTICS__
1227	} // malloc_stats_fd
1228
1229
1230	// Adjusts parameters that control the behaviour of the memory-allocation functions (see malloc). The param argument
1231	// specifies the parameter to be modified, and value specifies the new value for that parameter.
1232	int mallopt( int option, int value ) {
1233	choose( option ) {
1234	case M_TOP_PAD:
1235	heapExpand = ceiling2( value, __page_size ); return 1;
1236	case M_MMAP_THRESHOLD:
1237	if ( setMmapStart( value ) ) return 1;
1238	break;
1239	} // switch
1240	return 0; // error, unsupported
1241	} // mallopt
1242
1243
1244	// Attempt to release free memory at the top of the heap (by calling sbrk with a suitable argument).
1245	int malloc_trim( size_t ) {
1246	return 0; // => impossible to release memory
1247	} // malloc_trim
1248
1249
1250	// Exports an XML string that describes the current state of the memory-allocation implementation in the caller.
1251	// The string is printed on the file stream stream. The exported string includes information about all arenas (see
1252	// malloc).
1253	int malloc_info( int options, FILE * stream __attribute__(( unused )) ) {
1254	if ( options != 0 ) { errno = EINVAL; return -1; }
1255	#ifdef __STATISTICS__
1256	return printStatsXML( stream );
1257	#else
1258	return 0; // unsupported
1259	#endif // __STATISTICS__
1260	} // malloc_info
1261
1262
1263	// Records the current state of all malloc internal bookkeeping variables (but not the actual contents of the heap
1264	// or the state of malloc_hook functions pointers). The state is recorded in a system-dependent opaque data
1265	// structure dynamically allocated via malloc, and a pointer to that data structure is returned as the function
1266	// result. (The caller must free this memory.)
1267	void * malloc_get_state( void ) {
1268	return 0p; // unsupported
1269	} // malloc_get_state
1270
1271
1272	// Restores the state of all malloc internal bookkeeping variables to the values recorded in the opaque data
1273	// structure pointed to by state.
1274	int malloc_set_state( void * ) {
1275	return 0; // unsupported
1276	} // malloc_set_state
1277	} // extern "C"
1278
1279
1280	// Must have CFA linkage to overload with C linkage realloc.
1281	void * resize( void * oaddr, size_t nalign, size_t size ) {
1282	// If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
1283	if ( unlikely( size == 0 ) ) { // special cases
1284	#ifdef __STATISTICS__
1285	__atomic_add_fetch( &resize_0_calls, 1, __ATOMIC_SEQ_CST );
1286	#endif // __STATISTICS__
1287	free( oaddr );
1288	return 0p;
1289	} // if
1290
1291	if ( unlikely( nalign < libAlign() ) ) nalign = libAlign(); // reset alignment to minimum
1292	#ifdef __CFA_DEBUG__
1293	else checkAlign( nalign ); // check alignment
1294	#endif // __CFA_DEBUG__
1295
1296	if ( unlikely( oaddr == 0p ) ) {
1297	#ifdef __STATISTICS__
1298	__atomic_add_fetch( &resize_calls, 1, __ATOMIC_SEQ_CST );
1299	__atomic_add_fetch( &resize_storage_request, size, __ATOMIC_SEQ_CST );
1300	#endif // __STATISTICS__
1301	return memalignNoStats( nalign, size );
1302	} // if
1303
1304	// Attempt to reuse existing alignment.
1305	HeapManager.Storage.Header * header = headerAddr( oaddr );
1306	bool isFakeHeader = header->kind.fake.alignment & 1; // old fake header ?
1307	size_t oalign;
1308	if ( isFakeHeader ) {
1309	oalign = header->kind.fake.alignment & -2; // old alignment
1310	if ( (uintptr_t)oaddr % nalign == 0 // lucky match ?
1311	&& ( oalign <= nalign // going down
1312	\|\| (oalign >= nalign && oalign <= 256) ) // little alignment storage wasted ?
1313	) {
1314	headerAddr( oaddr )->kind.fake.alignment = nalign \| 1; // update alignment (could be the same)
1315	HeapManager.FreeHeader * freeElem;
1316	size_t bsize, oalign;
1317	headers( "resize", oaddr, header, freeElem, bsize, oalign );
1318	size_t odsize = dataStorage( bsize, oaddr, header ); // data storage available in bucket
1319
1320	if ( size <= odsize && odsize <= size * 2 ) { // allow 50% wasted data storage
1321	headerAddr( oaddr )->kind.fake.alignment = nalign \| 1; // update alignment (could be the same)
1322
1323	header->kind.real.blockSize &= -2; // turn off 0 fill
1324	header->kind.real.size = size; // reset allocation size
1325	return oaddr;
1326	} // if
1327	} // if
1328	} else if ( ! isFakeHeader // old real header (aligned on libAlign) ?
1329	&& nalign == libAlign() ) { // new alignment also on libAlign => no fake header needed
1330	return resize( oaddr, size ); // duplicate special case checks
1331	} // if
1332
1333	#ifdef __STATISTICS__
1334	__atomic_add_fetch( &resize_storage_request, size, __ATOMIC_SEQ_CST );
1335	#endif // __STATISTICS__
1336
1337	// change size, DO NOT preserve STICKY PROPERTIES.
1338	free( oaddr );
1339	return memalignNoStats( nalign, size ); // create new aligned area
1340	} // resize
1341
1342
1343	void * realloc( void * oaddr, size_t nalign, size_t size ) {
1344	// If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
1345	if ( unlikely( size == 0 ) ) { // special cases
1346	#ifdef __STATISTICS__
1347	__atomic_add_fetch( &realloc_0_calls, 1, __ATOMIC_SEQ_CST );
1348	#endif // __STATISTICS__
1349	free( oaddr );
1350	return 0p;
1351	} // if
1352
1353	if ( unlikely( nalign < libAlign() ) ) nalign = libAlign(); // reset alignment to minimum
1354	#ifdef __CFA_DEBUG__
1355	else checkAlign( nalign ); // check alignment
1356	#endif // __CFA_DEBUG__
1357
1358	if ( unlikely( oaddr == 0p ) ) {
1359	#ifdef __STATISTICS__
1360	__atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
1361	__atomic_add_fetch( &realloc_storage_request, size, __ATOMIC_SEQ_CST );
1362	#endif // __STATISTICS__
1363	return memalignNoStats( nalign, size );
1364	} // if
1365
1366	// Attempt to reuse existing alignment.
1367	HeapManager.Storage.Header * header = headerAddr( oaddr );
1368	bool isFakeHeader = header->kind.fake.alignment & 1; // old fake header ?
1369	size_t oalign;
1370	if ( isFakeHeader ) {
1371	oalign = header->kind.fake.alignment & -2; // old alignment
1372	if ( (uintptr_t)oaddr % nalign == 0 // lucky match ?
1373	&& ( oalign <= nalign // going down
1374	\|\| (oalign >= nalign && oalign <= 256) ) // little alignment storage wasted ?
1375	) {
1376	headerAddr( oaddr )->kind.fake.alignment = nalign \| 1; // update alignment (could be the same)
1377	return realloc( oaddr, size ); // duplicate alignment and special case checks
1378	} // if
1379	} else if ( ! isFakeHeader // old real header (aligned on libAlign) ?
1380	&& nalign == libAlign() ) // new alignment also on libAlign => no fake header needed
1381	return realloc( oaddr, size ); // duplicate alignment and special case checks
1382
1383	#ifdef __STATISTICS__
1384	__atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
1385	__atomic_add_fetch( &realloc_storage_request, size, __ATOMIC_SEQ_CST );
1386	#endif // __STATISTICS__
1387
1388	HeapManager.FreeHeader * freeElem;
1389	size_t bsize;
1390	headers( "realloc", oaddr, header, freeElem, bsize, oalign );
1391
1392	// change size and copy old content to new storage
1393
1394	size_t osize = header->kind.real.size; // old allocation size
1395	bool ozfill = (header->kind.real.blockSize & 2); // old allocation zero filled
1396
1397	void * naddr = memalignNoStats( nalign, size ); // create new aligned area
1398
1399	headers( "realloc", naddr, header, freeElem, bsize, oalign );
1400	memcpy( naddr, oaddr, min( osize, size ) ); // copy bytes
1401	free( oaddr );
1402
1403	if ( unlikely( ozfill ) ) { // previous request zero fill ?
1404	header->kind.real.blockSize \|= 2; // mark new request as zero filled
1405	if ( size > osize ) { // previous request larger ?
1406	memset( (char *)naddr + osize, '\0', size - osize ); // initialize added storage
1407	} // if
1408	} // if
1409	return naddr;
1410	} // realloc
1411
1412
1413	// Local Variables: //
1414	// tab-width: 4 //
1415	// compile-command: "cfa -nodebug -O2 heap.cfa" //
1416	// End: //

Note: See TracBrowser for help on using the repository browser.

Download in other formats:

Original Format