Context Navigation

heap.cfa @ 5f6b2c2

ADTast-experimental

Last change on this file since 5f6b2c2 was 5a076837, checked in by Thierry Delisle <tdelisle@…>, 2 years ago
Remove unnecessary declaration in heap
Property mode set to `100644`
File size: 58.1 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 : Fri Apr 29 19:05:03 2022
13	// Update Count : 1167
14	//
15
16	#include <string.h> // memset, memcpy
17	#include <limits.h> // ULONG_MAX
18	#include <stdlib.h> // EXIT_FAILURE
19	#include <errno.h> // errno, ENOMEM, EINVAL
20	#include <unistd.h> // STDERR_FILENO, sbrk, sysconf
21	#include <malloc.h> // memalign, malloc_usable_size
22	#include <sys/mman.h> // mmap, munmap
23	#include <sys/sysinfo.h> // get_nprocs
24
25	#include "bits/align.hfa" // libAlign
26	#include "bits/defs.hfa" // likely, unlikely
27	#include "bits/locks.hfa" // __spinlock_t
28	#include "startup.hfa" // STARTUP_PRIORITY_MEMORY
29	#include "math.hfa" // min
30	#include "bitmanip.hfa" // is_pow2, ceiling2
31
32	#define FASTLOOKUP
33	#define __STATISTICS__
34
35
36	static bool traceHeap = false;
37
38	inline bool traceHeap() libcfa_public { return traceHeap; }
39
40	bool traceHeapOn() libcfa_public {
41	bool temp = traceHeap;
42	traceHeap = true;
43	return temp;
44	} // traceHeapOn
45
46	bool traceHeapOff() libcfa_public {
47	bool temp = traceHeap;
48	traceHeap = false;
49	return temp;
50	} // traceHeapOff
51
52	bool traceHeapTerm() libcfa_public { return false; }
53
54
55	static bool prtFree = false;
56
57	static bool prtFree() {
58	return prtFree;
59	} // prtFree
60
61	static bool prtFreeOn() {
62	bool temp = prtFree;
63	prtFree = true;
64	return temp;
65	} // prtFreeOn
66
67	static bool prtFreeOff() {
68	bool temp = prtFree;
69	prtFree = false;
70	return temp;
71	} // prtFreeOff
72
73
74	enum {
75	// The default extension heap amount in units of bytes. When the current heap reaches the brk address, the brk
76	// address is extended by the extension amount.
77	__CFA_DEFAULT_HEAP_EXPANSION__ = 10 * 1024 * 1024,
78
79	// The mmap crossover point during allocation. Allocations less than this amount are allocated from buckets; values
80	// greater than or equal to this value are mmap from the operating system.
81	__CFA_DEFAULT_MMAP_START__ = 512 * 1024 + 1,
82
83	// The default unfreed storage amount in units of bytes. When the uC++ program ends it subtracts this amount from
84	// the malloc/free counter to adjust for storage the program does not free.
85	__CFA_DEFAULT_HEAP_UNFREED__ = 0
86	}; // enum
87
88
89	//####################### Heap Statistics ####################
90
91
92	#ifdef __STATISTICS__
93	enum { CntTriples = 12 }; // number of counter triples
94	enum { MALLOC, AALLOC, CALLOC, MEMALIGN, AMEMALIGN, CMEMALIGN, RESIZE, REALLOC, FREE };
95
96	struct StatsOverlay { // overlay for iteration
97	unsigned int calls, calls_0;
98	unsigned long long int request, alloc;
99	};
100
101	// Heap statistics counters.
102	union HeapStatistics {
103	struct { // minimum qualification
104	unsigned int malloc_calls, malloc_0_calls;
105	unsigned long long int malloc_storage_request, malloc_storage_alloc;
106	unsigned int aalloc_calls, aalloc_0_calls;
107	unsigned long long int aalloc_storage_request, aalloc_storage_alloc;
108	unsigned int calloc_calls, calloc_0_calls;
109	unsigned long long int calloc_storage_request, calloc_storage_alloc;
110	unsigned int memalign_calls, memalign_0_calls;
111	unsigned long long int memalign_storage_request, memalign_storage_alloc;
112	unsigned int amemalign_calls, amemalign_0_calls;
113	unsigned long long int amemalign_storage_request, amemalign_storage_alloc;
114	unsigned int cmemalign_calls, cmemalign_0_calls;
115	unsigned long long int cmemalign_storage_request, cmemalign_storage_alloc;
116	unsigned int resize_calls, resize_0_calls;
117	unsigned long long int resize_storage_request, resize_storage_alloc;
118	unsigned int realloc_calls, realloc_0_calls;
119	unsigned long long int realloc_storage_request, realloc_storage_alloc;
120	unsigned int free_calls, free_null_calls;
121	unsigned long long int free_storage_request, free_storage_alloc;
122	unsigned int away_pulls, away_pushes;
123	unsigned long long int away_storage_request, away_storage_alloc;
124	unsigned int mmap_calls, mmap_0_calls; // no zero calls
125	unsigned long long int mmap_storage_request, mmap_storage_alloc;
126	unsigned int munmap_calls, munmap_0_calls; // no zero calls
127	unsigned long long int munmap_storage_request, munmap_storage_alloc;
128	};
129	struct StatsOverlay counters[CntTriples]; // overlay for iteration
130	}; // HeapStatistics
131
132	static_assert( sizeof(HeapStatistics) == CntTriples * sizeof(StatsOverlay),
133	"Heap statistics counter-triplets does not match with array size" );
134
135	static void HeapStatisticsCtor( HeapStatistics & stats ) {
136	memset( &stats, '\0', sizeof(stats) ); // very fast
137	// for ( unsigned int i = 0; i < CntTriples; i += 1 ) {
138	// stats.counters[i].calls = stats.counters[i].calls_0 = stats.counters[i].request = stats.counters[i].alloc = 0;
139	// } // for
140	} // HeapStatisticsCtor
141
142	static HeapStatistics & ?+=?( HeapStatistics & lhs, const HeapStatistics & rhs ) {
143	for ( unsigned int i = 0; i < CntTriples; i += 1 ) {
144	lhs.counters[i].calls += rhs.counters[i].calls;
145	lhs.counters[i].calls_0 += rhs.counters[i].calls_0;
146	lhs.counters[i].request += rhs.counters[i].request;
147	lhs.counters[i].alloc += rhs.counters[i].alloc;
148	} // for
149	return lhs;
150	} // ?+=?
151	#endif // __STATISTICS__
152
153
154	#define SPINLOCK 0
155	#define LOCKFREE 1
156	#define BUCKETLOCK SPINLOCK
157	#if BUCKETLOCK == SPINLOCK
158	#elif BUCKETLOCK == LOCKFREE
159	#include <stackLockFree.hfa>
160	#else
161	#error undefined lock type for bucket lock
162	#endif // LOCKFREE
163
164	// Recursive definitions: HeapManager needs size of bucket array and bucket area needs sizeof HeapManager storage.
165	// Break recursion by hardcoding number of buckets and statically checking number is correct after bucket array defined.
166	enum { NoBucketSizes = 91 }; // number of buckets sizes
167
168	struct Heap {
169	struct Storage {
170	struct Header { // header
171	union Kind {
172	struct RealHeader {
173	union {
174	struct { // 4-byte word => 8-byte header, 8-byte word => 16-byte header
175	union {
176	// 2nd low-order bit => zero filled, 3rd low-order bit => mmapped
177	// FreeHeader * home; // allocated block points back to home locations (must overlay alignment)
178	void * home; // allocated block points back to home locations (must overlay alignment)
179	size_t blockSize; // size for munmap (must overlay alignment)
180	#if BUCKETLOCK == SPINLOCK
181	Storage * next; // freed block points to next freed block of same size
182	#endif // SPINLOCK
183	};
184	size_t size; // allocation size in bytes
185	};
186	#if BUCKETLOCK == LOCKFREE
187	Link(Storage) next; // freed block points next freed block of same size (double-wide)
188	#endif // LOCKFREE
189	};
190	} real; // RealHeader
191
192	struct FakeHeader {
193	uintptr_t alignment; // 1st low-order bit => fake header & alignment
194	uintptr_t offset;
195	} fake; // FakeHeader
196	} kind; // Kind
197	} header; // Header
198
199	char pad[libAlign() - sizeof( Header )];
200	char data[0]; // storage
201	}; // Storage
202
203	static_assert( libAlign() >= sizeof( Storage ), "minimum alignment < sizeof( Storage )" );
204
205	struct FreeHeader {
206	size_t blockSize __attribute__(( aligned (8) )); // size of allocations on this list
207	#if BUCKETLOCK == SPINLOCK
208	__spinlock_t lock;
209	Storage * freeList;
210	#else
211	StackLF(Storage) freeList;
212	#endif // BUCKETLOCK
213	} __attribute__(( aligned (8) )); // FreeHeader
214
215	FreeHeader freeLists[NoBucketSizes]; // buckets for different allocation sizes
216
217	__spinlock_t extlock; // protects allocation-buffer extension
218	void * heapBegin; // start of heap
219	void * heapEnd; // logical end of heap
220	size_t heapRemaining; // amount of storage not allocated in the current chunk
221	}; // Heap
222
223	#if BUCKETLOCK == LOCKFREE
224	static inline {
225	Link(Heap.Storage) * ?`next( Heap.Storage * this ) { return &this->header.kind.real.next; }
226	void ?{}( Heap.FreeHeader & ) {}
227	void ^?{}( Heap.FreeHeader & ) {}
228	} // distribution
229	#endif // LOCKFREE
230
231	static inline size_t getKey( const Heap.FreeHeader & freeheader ) { return freeheader.blockSize; }
232
233
234	#ifdef FASTLOOKUP
235	enum { LookupSizes = 65_536 + sizeof(Heap.Storage) }; // number of fast lookup sizes
236	static unsigned char lookup[LookupSizes]; // O(1) lookup for small sizes
237	#endif // FASTLOOKUP
238
239	static const off_t mmapFd = -1; // fake or actual fd for anonymous file
240	#ifdef __CFA_DEBUG__
241	static bool heapBoot = 0; // detect recursion during boot
242	#endif // __CFA_DEBUG__
243
244
245	// Size of array must harmonize with NoBucketSizes and individual bucket sizes must be multiple of 16.
246	// Smaller multiples of 16 and powers of 2 are common allocation sizes, so make them generate the minimum required bucket size.
247	// malloc(0) returns 0p, so no bucket is necessary for 0 bytes returning an address that can be freed.
248	static const unsigned int bucketSizes[] @= { // different bucket sizes
249	16 + sizeof(Heap.Storage), 32 + sizeof(Heap.Storage), 48 + sizeof(Heap.Storage), 64 + sizeof(Heap.Storage), // 4
250	96 + sizeof(Heap.Storage), 112 + sizeof(Heap.Storage), 128 + sizeof(Heap.Storage), // 3
251	160, 192, 224, 256 + sizeof(Heap.Storage), // 4
252	320, 384, 448, 512 + sizeof(Heap.Storage), // 4
253	640, 768, 896, 1_024 + sizeof(Heap.Storage), // 4
254	1_536, 2_048 + sizeof(Heap.Storage), // 2
255	2_560, 3_072, 3_584, 4_096 + sizeof(Heap.Storage), // 4
256	6_144, 8_192 + sizeof(Heap.Storage), // 2
257	9_216, 10_240, 11_264, 12_288, 13_312, 14_336, 15_360, 16_384 + sizeof(Heap.Storage), // 8
258	18_432, 20_480, 22_528, 24_576, 26_624, 28_672, 30_720, 32_768 + sizeof(Heap.Storage), // 8
259	36_864, 40_960, 45_056, 49_152, 53_248, 57_344, 61_440, 65_536 + sizeof(Heap.Storage), // 8
260	73_728, 81_920, 90_112, 98_304, 106_496, 114_688, 122_880, 131_072 + sizeof(Heap.Storage), // 8
261	147_456, 163_840, 180_224, 196_608, 212_992, 229_376, 245_760, 262_144 + sizeof(Heap.Storage), // 8
262	294_912, 327_680, 360_448, 393_216, 425_984, 458_752, 491_520, 524_288 + sizeof(Heap.Storage), // 8
263	655_360, 786_432, 917_504, 1_048_576 + sizeof(Heap.Storage), // 4
264	1_179_648, 1_310_720, 1_441_792, 1_572_864, 1_703_936, 1_835_008, 1_966_080, 2_097_152 + sizeof(Heap.Storage), // 8
265	2_621_440, 3_145_728, 3_670_016, 4_194_304 + sizeof(Heap.Storage), // 4
266	};
267
268	static_assert( NoBucketSizes == sizeof(bucketSizes) / sizeof(bucketSizes[0] ), "size of bucket array wrong" );
269
270	// The constructor for heapManager is called explicitly in memory_startup.
271	static Heap heapManager __attribute__(( aligned (128) )) @= {}; // size of cache line to prevent false sharing
272
273
274	//####################### Memory Allocation Routines Helpers ####################
275
276
277	#ifdef __CFA_DEBUG__
278	static size_t allocUnfreed; // running total of allocations minus frees
279
280	static void prtUnfreed() {
281	if ( allocUnfreed != 0 ) {
282	// DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT.
283	char helpText[512];
284	__cfaabi_bits_print_buffer( STDERR_FILENO, helpText, sizeof(helpText),
285	"CFA warning (UNIX pid:%ld) : program terminating with %zu(0x%zx) bytes of storage allocated but not freed.\n"
286	"Possible cause is unfreed storage allocated by the program or system/library routines called from the program.\n",
287	(long int)getpid(), allocUnfreed, allocUnfreed ); // always print the UNIX pid
288	} // if
289	} // prtUnfreed
290
291	extern int cfa_main_returned; // from interpose.cfa
292	extern "C" {
293	void heapAppStart() { // called by __cfaabi_appready_startup
294	allocUnfreed = 0;
295	} // heapAppStart
296
297	void heapAppStop() { // called by __cfaabi_appready_startdown
298	fclose( stdin ); fclose( stdout );
299	if ( cfa_main_returned ) prtUnfreed(); // do not check unfreed storage if exit called
300	} // heapAppStop
301	} // extern "C"
302	#endif // __CFA_DEBUG__
303
304
305	#ifdef __STATISTICS__
306	static HeapStatistics stats; // zero filled
307	static unsigned int sbrk_calls;
308	static unsigned long long int sbrk_storage;
309	// Statistics file descriptor (changed by malloc_stats_fd).
310	static int stats_fd = STDERR_FILENO; // default stderr
311
312	#define prtFmt \
313	"\nHeap statistics: (storage request / allocation)\n" \
314	" malloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
315	" aalloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
316	" calloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
317	" memalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
318	" amemalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
319	" cmemalign >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
320	" resize >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
321	" realloc >0 calls %'u; 0 calls %'u; storage %'llu / %'llu bytes\n" \
322	" free !null calls %'u; null calls %'u; storage %'llu / %'llu bytes\n" \
323	" sbrk calls %'u; storage %'llu bytes\n" \
324	" mmap calls %'u; storage %'llu / %'llu bytes\n" \
325	" munmap calls %'u; storage %'llu / %'llu bytes\n" \
326
327	// Use "write" because streams may be shutdown when calls are made.
328	static int printStats() { // see malloc_stats
329	char helpText[sizeof(prtFmt) + 1024]; // space for message and values
330	return __cfaabi_bits_print_buffer( STDERR_FILENO, helpText, sizeof(helpText), prtFmt,
331	stats.malloc_calls, stats.malloc_0_calls, stats.malloc_storage_request, stats.malloc_storage_alloc,
332	stats.aalloc_calls, stats.aalloc_0_calls, stats.aalloc_storage_request, stats.aalloc_storage_alloc,
333	stats.calloc_calls, stats.calloc_0_calls, stats.calloc_storage_request, stats.calloc_storage_alloc,
334	stats.memalign_calls, stats.memalign_0_calls, stats.memalign_storage_request, stats.memalign_storage_alloc,
335	stats.amemalign_calls, stats.amemalign_0_calls, stats.amemalign_storage_request, stats.amemalign_storage_alloc,
336	stats.cmemalign_calls, stats.cmemalign_0_calls, stats.cmemalign_storage_request, stats.cmemalign_storage_alloc,
337	stats.resize_calls, stats.resize_0_calls, stats.resize_storage_request, stats.resize_storage_alloc,
338	stats.realloc_calls, stats.realloc_0_calls, stats.realloc_storage_request, stats.realloc_storage_alloc,
339	stats.free_calls, stats.free_null_calls, stats.free_storage_request, stats.free_storage_alloc,
340	sbrk_calls, sbrk_storage,
341	stats.mmap_calls, stats.mmap_storage_request, stats.mmap_storage_alloc,
342	stats.munmap_calls, stats.munmap_storage_request, stats.munmap_storage_alloc
343	);
344	} // printStats
345
346	#define prtFmtXML \
347	"<malloc version=\"1\">\n" \
348	"<heap nr=\"0\">\n" \
349	"<sizes>\n" \
350	"</sizes>\n" \
351	"<total type=\"malloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
352	"<total type=\"aalloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
353	"<total type=\"calloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
354	"<total type=\"memalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
355	"<total type=\"amemalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
356	"<total type=\"cmemalign\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
357	"<total type=\"resize\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
358	"<total type=\"realloc\" >0 count=\"%'u;\" 0 count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
359	"<total type=\"free\" !null=\"%'u;\" 0 null=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
360	"<total type=\"sbrk\" count=\"%'u;\" size=\"%'llu\"/> bytes\n" \
361	"<total type=\"mmap\" count=\"%'u;\" size=\"%'llu / %'llu\" / > bytes\n" \
362	"<total type=\"munmap\" count=\"%'u;\" size=\"%'llu / %'llu\"/> bytes\n" \
363	"</malloc>"
364
365	static int printStatsXML( FILE * stream ) { // see malloc_info
366	char helpText[sizeof(prtFmtXML) + 1024]; // space for message and values
367	return __cfaabi_bits_print_buffer( fileno( stream ), helpText, sizeof(helpText), prtFmtXML,
368	stats.malloc_calls, stats.malloc_0_calls, stats.malloc_storage_request, stats.malloc_storage_alloc,
369	stats.aalloc_calls, stats.aalloc_0_calls, stats.aalloc_storage_request, stats.aalloc_storage_alloc,
370	stats.calloc_calls, stats.calloc_0_calls, stats.calloc_storage_request, stats.calloc_storage_alloc,
371	stats.memalign_calls, stats.memalign_0_calls, stats.memalign_storage_request, stats.memalign_storage_alloc,
372	stats.amemalign_calls, stats.amemalign_0_calls, stats.amemalign_storage_request, stats.amemalign_storage_alloc,
373	stats.cmemalign_calls, stats.cmemalign_0_calls, stats.cmemalign_storage_request, stats.cmemalign_storage_alloc,
374	stats.resize_calls, stats.resize_0_calls, stats.resize_storage_request, stats.resize_storage_alloc,
375	stats.realloc_calls, stats.realloc_0_calls, stats.realloc_storage_request, stats.realloc_storage_alloc,
376	stats.free_calls, stats.free_null_calls, stats.free_storage_request, stats.free_storage_alloc,
377	sbrk_calls, sbrk_storage,
378	stats.mmap_calls, stats.mmap_storage_request, stats.mmap_storage_alloc,
379	stats.munmap_calls, stats.munmap_storage_request, stats.munmap_storage_alloc
380	);
381	} // printStatsXML
382	#endif // __STATISTICS__
383
384
385	// statically allocated variables => zero filled.
386	static size_t heapExpand; // sbrk advance
387	static size_t mmapStart; // cross over point for mmap
388	static unsigned int maxBucketsUsed; // maximum number of buckets in use
389	// extern visibility, used by runtime kernel
390	// would be cool to remove libcfa_public but it's needed for libcfathread
391	libcfa_public size_t __page_size; // architecture pagesize
392	libcfa_public int __map_prot; // common mmap/mprotect protection
393
394
395	// thunk problem
396	size_t Bsearchl( unsigned int key, const unsigned int * vals, size_t dim ) {
397	size_t l = 0, m, h = dim;
398	while ( l < h ) {
399	m = (l + h) / 2;
400	if ( (unsigned int &)(vals[m]) < key ) { // cast away const
401	l = m + 1;
402	} else {
403	h = m;
404	} // if
405	} // while
406	return l;
407	} // Bsearchl
408
409
410	static inline bool setMmapStart( size_t value ) { // true => mmapped, false => sbrk
411	if ( value < __page_size \|\| bucketSizes[NoBucketSizes - 1] < value ) return false;
412	mmapStart = value; // set global
413
414	// find the closest bucket size less than or equal to the mmapStart size
415	maxBucketsUsed = Bsearchl( (unsigned int)mmapStart, bucketSizes, NoBucketSizes ); // binary search
416	assert( maxBucketsUsed < NoBucketSizes ); // subscript failure ?
417	assert( mmapStart <= bucketSizes[maxBucketsUsed] ); // search failure ?
418	return true;
419	} // setMmapStart
420
421
422	// <-------+----------------------------------------------------> bsize (bucket size)
423	// \|header \|addr
424	//==================================================================================
425	// align/offset \|
426	// <-----------------<------------+-----------------------------> bsize (bucket size)
427	// \|fake-header \| addr
428	#define HeaderAddr( addr ) ((Heap.Storage.Header )( (char )addr - sizeof(Heap.Storage) ))
429	#define RealHeader( header ) ((Heap.Storage.Header )((char )header - header->kind.fake.offset))
430
431	// <-------<<--------------------- dsize ---------------------->> bsize (bucket size)
432	// \|header \|addr
433	//==================================================================================
434	// align/offset \|
435	// <------------------------------<<---------- dsize --------->>> bsize (bucket size)
436	// \|fake-header \|addr
437	#define DataStorage( bsize, addr, header ) (bsize - ( (char )addr - (char )header ))
438
439
440	static inline void checkAlign( size_t alignment ) {
441	if ( unlikely( alignment < libAlign() \|\| ! is_pow2( alignment ) ) ) {
442	abort( "** Error ** alignment %zu for memory allocation is less than %d and/or not a power of 2.", alignment, libAlign() );
443	} // if
444	} // checkAlign
445
446
447	static inline void checkHeader( bool check, const char name[], void * addr ) {
448	if ( unlikely( check ) ) { // bad address ?
449	abort( "** Error ** attempt to %s storage %p with address outside the heap.\n"
450	"Possible cause is duplicate free on same block or overwriting of memory.",
451	name, addr );
452	} // if
453	} // checkHeader
454
455
456	// Manipulate sticky bits stored in unused 3 low-order bits of an address.
457	// bit0 => alignment => fake header
458	// bit1 => zero filled (calloc)
459	// bit2 => mapped allocation versus sbrk
460	#define StickyBits( header ) (((header)->kind.real.blockSize & 0x7))
461	#define ClearStickyBits( addr ) (typeof(addr))((uintptr_t)(addr) & ~7)
462	#define MarkAlignmentBit( align ) ((align) \| 1)
463	#define AlignmentBit( header ) ((((header)->kind.fake.alignment) & 1))
464	#define ClearAlignmentBit( header ) (((header)->kind.fake.alignment) & ~1)
465	#define ZeroFillBit( header ) ((((header)->kind.real.blockSize) & 2))
466	#define ClearZeroFillBit( header ) ((((header)->kind.real.blockSize) &= ~2))
467	#define MarkZeroFilledBit( header ) ((header)->kind.real.blockSize \|= 2)
468	#define MmappedBit( header ) ((((header)->kind.real.blockSize) & 4))
469	#define MarkMmappedBit( size ) ((size) \| 4)
470
471
472	static inline void fakeHeader( Heap.Storage.Header *& header, size_t & alignment ) {
473	if ( unlikely( AlignmentBit( header ) ) ) { // fake header ?
474	alignment = ClearAlignmentBit( header ); // clear flag from value
475	#ifdef __CFA_DEBUG__
476	checkAlign( alignment ); // check alignment
477	#endif // __CFA_DEBUG__
478	header = RealHeader( header ); // backup from fake to real header
479	} else {
480	alignment = libAlign(); // => no fake header
481	} // if
482	} // fakeHeader
483
484
485	static inline bool headers( const char name[] __attribute__(( unused )), void * addr, Heap.Storage.Header *& header,
486	Heap.FreeHeader *& freeHead, size_t & size, size_t & alignment ) with( heapManager ) {
487	header = HeaderAddr( addr );
488
489	#ifdef __CFA_DEBUG__
490	checkHeader( header < (Heap.Storage.Header *)heapBegin, name, addr ); // bad low address ?
491	#endif // __CFA_DEBUG__
492
493	if ( likely( ! StickyBits( header ) ) ) { // no sticky bits ?
494	freeHead = (Heap.FreeHeader *)(header->kind.real.home);
495	alignment = libAlign();
496	} else {
497	fakeHeader( header, alignment );
498	if ( unlikely( MmappedBit( header ) ) ) { // mmapped ?
499	verify( addr < heapBegin \|\| heapEnd < addr );
500	size = ClearStickyBits( header->kind.real.blockSize ); // mmap size
501	return true;
502	} // if
503
504	freeHead = (Heap.FreeHeader *)(ClearStickyBits( header->kind.real.home ));
505	} // if
506	size = freeHead->blockSize;
507
508	#ifdef __CFA_DEBUG__
509	checkHeader( header < (Heap.Storage.Header )heapBegin \|\| (Heap.Storage.Header )heapEnd < header, name, addr ); // bad address ? (offset could be + or -)
510
511	if ( unlikely( freeHead == 0p \|\| // freed and only free-list node => null link
512	// freed and link points at another free block not to a bucket in the bucket array.
513	freeHead < &freeLists[0] \|\| &freeLists[NoBucketSizes] <= freeHead ) ) {
514	abort( "** Error ** attempt to %s storage %p with corrupted header.\n"
515	"Possible cause is duplicate free on same block or overwriting of header information.",
516	name, addr );
517	} // if
518	#endif // __CFA_DEBUG__
519
520	return false;
521	} // headers
522
523	// #ifdef __CFA_DEBUG__
524	// #if __SIZEOF_POINTER__ == 4
525	// #define MASK 0xdeadbeef
526	// #else
527	// #define MASK 0xdeadbeefdeadbeef
528	// #endif
529	// #define STRIDE size_t
530
531	// static void * Memset( void * addr, STRIDE size ) { // debug only
532	// if ( size % sizeof(STRIDE) != 0 ) abort( "Memset() : internal error, size %zd not multiple of %zd.", size, sizeof(STRIDE) );
533	// if ( (STRIDE)addr % sizeof(STRIDE) != 0 ) abort( "Memset() : internal error, addr %p not multiple of %zd.", addr, sizeof(STRIDE) );
534
535	// STRIDE * end = (STRIDE *)addr + size / sizeof(STRIDE);
536	// for ( STRIDE * p = (STRIDE )addr; p < end; p += 1 ) p = MASK;
537	// return addr;
538	// } // Memset
539	// #endif // __CFA_DEBUG__
540
541
542	#define NO_MEMORY_MSG "insufficient heap memory available for allocating %zd new bytes."
543
544	static inline void * extend( size_t size ) with( heapManager ) {
545	lock( extlock __cfaabi_dbg_ctx2 );
546
547	ptrdiff_t rem = heapRemaining - size;
548	if ( unlikely( rem < 0 ) ) {
549	// If the size requested is bigger than the current remaining storage, increase the size of the heap.
550
551	size_t increase = ceiling2( size > heapExpand ? size : heapExpand, __page_size );
552	// Do not call abort or strerror( errno ) as they may call malloc.
553	if ( sbrk( increase ) == (void *)-1 ) { // failed, no memory ?
554	unlock( extlock );
555	__cfaabi_bits_print_nolock( STDERR_FILENO, NO_MEMORY_MSG, size );
556	_exit( EXIT_FAILURE ); // give up
557	} // if
558
559	// Make storage executable for thunks.
560	if ( mprotect( (char *)heapEnd + heapRemaining, increase, __map_prot ) ) {
561	unlock( extlock );
562	__cfaabi_bits_print_nolock( STDERR_FILENO, "extend() : internal error, mprotect failure, heapEnd:%p size:%zd, errno:%d.\n", heapEnd, increase, errno );
563	_exit( EXIT_FAILURE );
564	} // if
565
566	#ifdef __STATISTICS__
567	sbrk_calls += 1;
568	sbrk_storage += increase;
569	#endif // __STATISTICS__
570
571	#ifdef __CFA_DEBUG__
572	// Set new memory to garbage so subsequent uninitialized usages might fail.
573	memset( (char *)heapEnd + heapRemaining, '\xde', increase );
574	//Memset( (char *)heapEnd + heapRemaining, increase );
575	#endif // __CFA_DEBUG__
576
577	rem = heapRemaining + increase - size;
578	} // if
579
580	Heap.Storage * block = (Heap.Storage *)heapEnd;
581	heapRemaining = rem;
582	heapEnd = (char *)heapEnd + size;
583	unlock( extlock );
584	return block;
585	} // extend
586
587
588	static inline void * doMalloc( size_t size ) with( heapManager ) {
589	Heap.Storage * block; // pointer to new block of storage
590
591	// Look up size in the size list. Make sure the user request includes space for the header that must be allocated
592	// along with the block and is a multiple of the alignment size.
593
594	size_t tsize = size + sizeof(Heap.Storage);
595
596	if ( likely( tsize < mmapStart ) ) { // small size => sbrk
597	size_t posn;
598	#ifdef FASTLOOKUP
599	if ( tsize < LookupSizes ) posn = lookup[tsize];
600	else
601	#endif // FASTLOOKUP
602	posn = Bsearchl( (unsigned int)tsize, bucketSizes, (size_t)maxBucketsUsed );
603	Heap.FreeHeader * freeElem = &freeLists[posn];
604	verify( freeElem <= &freeLists[maxBucketsUsed] ); // subscripting error ?
605	verify( tsize <= freeElem->blockSize ); // search failure ?
606	tsize = freeElem->blockSize; // total space needed for request
607
608	// Spin until the lock is acquired for this particular size of block.
609
610	#if BUCKETLOCK == SPINLOCK
611	lock( freeElem->lock __cfaabi_dbg_ctx2 );
612	block = freeElem->freeList; // remove node from stack
613	#else
614	block = pop( freeElem->freeList );
615	#endif // BUCKETLOCK
616	if ( unlikely( block == 0p ) ) { // no free block ?
617	#if BUCKETLOCK == SPINLOCK
618	unlock( freeElem->lock );
619	#endif // BUCKETLOCK
620
621	// Freelist for that size was empty, so carve it out of the heap if there's enough left, or get some more
622	// and then carve it off.
623
624	block = (Heap.Storage *)extend( tsize ); // mutual exclusion on call
625	#if BUCKETLOCK == SPINLOCK
626	} else {
627	freeElem->freeList = block->header.kind.real.next;
628	unlock( freeElem->lock );
629	#endif // BUCKETLOCK
630	} // if
631
632	block->header.kind.real.home = freeElem; // pointer back to free list of apropriate size
633	} else { // large size => mmap
634	if ( unlikely( size > ULONG_MAX - __page_size ) ) return 0p;
635	tsize = ceiling2( tsize, __page_size ); // must be multiple of page size
636	#ifdef __STATISTICS__
637	__atomic_add_fetch( &stats.mmap_calls, 1, __ATOMIC_SEQ_CST );
638	__atomic_add_fetch( &stats.mmap_storage_request, size, __ATOMIC_SEQ_CST );
639	__atomic_add_fetch( &stats.mmap_storage_alloc, tsize, __ATOMIC_SEQ_CST );
640	#endif // __STATISTICS__
641
642	block = (Heap.Storage *)mmap( 0, tsize, __map_prot, MAP_PRIVATE \| MAP_ANONYMOUS, mmapFd, 0 );
643	if ( block == (Heap.Storage *)MAP_FAILED ) { // failed ?
644	if ( errno == ENOMEM ) abort( NO_MEMORY_MSG, tsize ); // no memory
645	// Do not call strerror( errno ) as it may call malloc.
646	abort( "(Heap &)0x%p.doMalloc() : internal error, mmap failure, size:%zu errno:%d.", &heapManager, tsize, errno );
647	} //if
648	#ifdef __CFA_DEBUG__
649	// Set new memory to garbage so subsequent uninitialized usages might fail.
650	memset( block, '\xde', tsize );
651	//Memset( block, tsize );
652	#endif // __CFA_DEBUG__
653	block->header.kind.real.blockSize = MarkMmappedBit( tsize ); // storage size for munmap
654	} // if
655
656	block->header.kind.real.size = size; // store allocation size
657	void * addr = &(block->data); // adjust off header to user bytes
658	verify( ((uintptr_t)addr & (libAlign() - 1)) == 0 ); // minimum alignment ?
659
660	#ifdef __CFA_DEBUG__
661	__atomic_add_fetch( &allocUnfreed, tsize, __ATOMIC_SEQ_CST );
662	if ( traceHeap() ) {
663	char helpText[64];
664	__cfaabi_bits_print_buffer( STDERR_FILENO, helpText, sizeof(helpText),
665	"%p = Malloc( %zu ) (allocated %zu)\n", addr, size, tsize ); // print debug/nodebug
666	} // if
667	#endif // __CFA_DEBUG__
668
669	return addr;
670	} // doMalloc
671
672
673	static inline void doFree( void * addr ) with( heapManager ) {
674	#ifdef __CFA_DEBUG__
675	if ( unlikely( heapManager.heapBegin == 0p ) ) {
676	abort( "doFree( %p ) : internal error, called before heap is initialized.", addr );
677	} // if
678	#endif // __CFA_DEBUG__
679
680	Heap.Storage.Header * header;
681	Heap.FreeHeader * freeElem;
682	size_t size, alignment; // not used (see realloc)
683
684	if ( headers( "free", addr, header, freeElem, size, alignment ) ) { // mmapped ?
685	#ifdef __STATISTICS__
686	__atomic_add_fetch( &stats.munmap_calls, 1, __ATOMIC_SEQ_CST );
687	__atomic_add_fetch( &stats.munmap_storage_request, header->kind.real.size, __ATOMIC_SEQ_CST );
688	__atomic_add_fetch( &stats.munmap_storage_alloc, size, __ATOMIC_SEQ_CST );
689	#endif // __STATISTICS__
690	if ( munmap( header, size ) == -1 ) {
691	abort( "Attempt to deallocate storage %p not allocated or with corrupt header.\n"
692	"Possible cause is invalid pointer.",
693	addr );
694	} // if
695	} else {
696	#ifdef __CFA_DEBUG__
697	// Set free memory to garbage so subsequent usages might fail.
698	memset( ((Heap.Storage *)header)->data, '\xde', freeElem->blockSize - sizeof( Heap.Storage ) );
699	//Memset( ((Heap.Storage *)header)->data, freeElem->blockSize - sizeof( Heap.Storage ) );
700	#endif // __CFA_DEBUG__
701
702	#ifdef __STATISTICS__
703	__atomic_add_fetch( &stats.free_calls, 1, __ATOMIC_SEQ_CST );
704	__atomic_add_fetch( &stats.free_storage_request, header->kind.real.size, __ATOMIC_SEQ_CST );
705	__atomic_add_fetch( &stats.free_storage_alloc, size, __ATOMIC_SEQ_CST );
706	#endif // __STATISTICS__
707
708	#if BUCKETLOCK == SPINLOCK
709	lock( freeElem->lock __cfaabi_dbg_ctx2 ); // acquire spin lock
710	header->kind.real.next = freeElem->freeList; // push on stack
711	freeElem->freeList = (Heap.Storage *)header;
712	unlock( freeElem->lock ); // release spin lock
713	#else
714	push( freeElem->freeList, (Heap.Storage )header );
715	#endif // BUCKETLOCK
716	} // if
717
718	#ifdef __CFA_DEBUG__
719	__atomic_add_fetch( &allocUnfreed, -size, __ATOMIC_SEQ_CST );
720	if ( traceHeap() ) {
721	char helpText[64];
722	__cfaabi_bits_print_buffer( STDERR_FILENO, helpText, sizeof(helpText),
723	"Free( %p ) size:%zu\n", addr, size ); // print debug/nodebug
724	} // if
725	#endif // __CFA_DEBUG__
726	} // doFree
727
728
729	static size_t prtFree( Heap & manager ) with( manager ) {
730	size_t total = 0;
731	#ifdef __STATISTICS__
732	__cfaabi_bits_acquire();
733	__cfaabi_bits_print_nolock( STDERR_FILENO, "\nBin lists (bin size : free blocks on list)\n" );
734	#endif // __STATISTICS__
735	for ( unsigned int i = 0; i < maxBucketsUsed; i += 1 ) {
736	size_t size = freeLists[i].blockSize;
737	#ifdef __STATISTICS__
738	unsigned int N = 0;
739	#endif // __STATISTICS__
740
741	#if BUCKETLOCK == SPINLOCK
742	for ( Heap.Storage * p = freeLists[i].freeList; p != 0p; p = p->header.kind.real.next ) {
743	#else
744	for(;;) {
745	// for ( Heap.Storage * p = top( freeLists[i].freeList ); p != 0p; p = (p)`next->top ) {
746	// for ( Heap.Storage * p = top( freeLists[i].freeList ); p != 0p; /* p = getNext( p )->top */) {
747	// Heap.Storage * temp = p->header.kind.real.next.top; // FIX ME: direct assignent fails, initialization works`
748	// typeof(p) temp = (( p )`next)->top; // FIX ME: direct assignent fails, initialization works`
749	// p = temp;
750	#endif // BUCKETLOCK
751	total += size;
752	#ifdef __STATISTICS__
753	N += 1;
754	#endif // __STATISTICS__
755	} // for
756
757	#ifdef __STATISTICS__
758	__cfaabi_bits_print_nolock( STDERR_FILENO, "%7zu, %-7u ", size, N );
759	if ( (i + 1) % 8 == 0 ) __cfaabi_bits_print_nolock( STDERR_FILENO, "\n" );
760	#endif // __STATISTICS__
761	} // for
762	#ifdef __STATISTICS__
763	__cfaabi_bits_print_nolock( STDERR_FILENO, "\ntotal free blocks:%zu\n", total );
764	__cfaabi_bits_release();
765	#endif // __STATISTICS__
766	return (char )heapEnd - (char )heapBegin - total;
767	} // prtFree
768
769
770	static void ?{}( Heap & manager ) with( manager ) {
771	__page_size = sysconf( _SC_PAGESIZE );
772	__map_prot = PROT_READ \| PROT_WRITE \| PROT_EXEC;
773
774	for ( unsigned int i = 0; i < NoBucketSizes; i += 1 ) { // initialize the free lists
775	freeLists[i].blockSize = bucketSizes[i];
776	} // for
777
778	#ifdef FASTLOOKUP
779	unsigned int idx = 0;
780	for ( unsigned int i = 0; i < LookupSizes; i += 1 ) {
781	if ( i > bucketSizes[idx] ) idx += 1;
782	lookup[i] = idx;
783	} // for
784	#endif // FASTLOOKUP
785
786	if ( ! setMmapStart( malloc_mmap_start() ) ) {
787	abort( "Heap : internal error, mmap start initialization failure." );
788	} // if
789	heapExpand = malloc_expansion();
790
791	char * end = (char *)sbrk( 0 );
792	heapBegin = heapEnd = sbrk( (char *)ceiling2( (long unsigned int)end, __page_size ) - end ); // move start of heap to multiple of alignment
793	} // Heap
794
795
796	static void ^?{}( Heap & ) {
797	#ifdef __STATISTICS__
798	if ( traceHeapTerm() ) {
799	printStats();
800	// prtUnfreed() called in heapAppStop()
801	} // if
802	#endif // __STATISTICS__
803	} // ~Heap
804
805
806	static void memory_startup( void ) __attribute__(( constructor( STARTUP_PRIORITY_MEMORY ) ));
807	void memory_startup( void ) {
808	#ifdef __CFA_DEBUG__
809	if ( heapBoot ) { // check for recursion during system boot
810	abort( "boot() : internal error, recursively invoked during system boot." );
811	} // if
812	heapBoot = true;
813	#endif // __CFA_DEBUG__
814
815	//verify( heapManager.heapBegin != 0 );
816	//heapManager{};
817	if ( heapManager.heapBegin == 0p ) heapManager{}; // sanity check
818	} // memory_startup
819
820	static void memory_shutdown( void ) __attribute__(( destructor( STARTUP_PRIORITY_MEMORY ) ));
821	void memory_shutdown( void ) {
822	^heapManager{};
823	} // memory_shutdown
824
825
826	static inline void * mallocNoStats( size_t size ) { // necessary for malloc statistics
827	verify( heapManager.heapBegin != 0p ); // called before memory_startup ?
828	if ( unlikely( size ) == 0 ) return 0p; // 0 BYTE ALLOCATION RETURNS NULL POINTER
829
830	#if __SIZEOF_POINTER__ == 8
831	verify( size < ((typeof(size_t))1 << 48) );
832	#endif // __SIZEOF_POINTER__ == 8
833	return doMalloc( size );
834	} // mallocNoStats
835
836
837	static inline void * memalignNoStats( size_t alignment, size_t size ) {
838	if ( unlikely( size ) == 0 ) return 0p; // 0 BYTE ALLOCATION RETURNS NULL POINTER
839
840	#ifdef __CFA_DEBUG__
841	checkAlign( alignment ); // check alignment
842	#endif // __CFA_DEBUG__
843
844	// if alignment <= default alignment, do normal malloc as two headers are unnecessary
845	if ( unlikely( alignment <= libAlign() ) ) return mallocNoStats( size );
846
847	// Allocate enough storage to guarantee an address on the alignment boundary, and sufficient space before it for
848	// administrative storage. NOTE, WHILE THERE ARE 2 HEADERS, THE FIRST ONE IS IMPLICITLY CREATED BY DOMALLOC.
849	// .-------------v-----------------v----------------v----------,
850	// \| Real Header \| ... padding ... \| Fake Header \| data ... \|
851	// `-------------^-----------------^-+--------------^----------'
852	// \|<--------------------------------' offset/align \|<-- alignment boundary
853
854	// subtract libAlign() because it is already the minimum alignment
855	// add sizeof(Storage) for fake header
856	char * addr = (char *)mallocNoStats( size + alignment - libAlign() + sizeof(Heap.Storage) );
857
858	// address in the block of the "next" alignment address
859	char * user = (char *)ceiling2( (uintptr_t)(addr + sizeof(Heap.Storage)), alignment );
860
861	// address of header from malloc
862	Heap.Storage.Header * RealHeader = HeaderAddr( addr );
863	RealHeader->kind.real.size = size; // correct size to eliminate above alignment offset
864	// address of fake header * before* the alignment location
865	Heap.Storage.Header * fakeHeader = HeaderAddr( user );
866	// SKULLDUGGERY: insert the offset to the start of the actual storage block and remember alignment
867	fakeHeader->kind.fake.offset = (char )fakeHeader - (char )RealHeader;
868	// SKULLDUGGERY: odd alignment implies fake header
869	fakeHeader->kind.fake.alignment = MarkAlignmentBit( alignment );
870
871	return user;
872	} // memalignNoStats
873
874
875	//####################### Memory Allocation Routines ####################
876
877
878	extern "C" {
879	// Allocates size bytes and returns a pointer to the allocated memory. The contents are undefined. If size is 0,
880	// then malloc() returns a unique pointer value that can later be successfully passed to free().
881	void * malloc( size_t size ) libcfa_public {
882	#ifdef __STATISTICS__
883	if ( likely( size > 0 ) ) {
884	__atomic_add_fetch( &stats.malloc_calls, 1, __ATOMIC_SEQ_CST );
885	__atomic_add_fetch( &stats.malloc_storage_request, size, __ATOMIC_SEQ_CST );
886	} else {
887	__atomic_add_fetch( &stats.malloc_0_calls, 1, __ATOMIC_SEQ_CST );
888	} // if
889	#endif // __STATISTICS__
890
891	return mallocNoStats( size );
892	} // malloc
893
894
895	// Same as malloc() except size bytes is an array of dim elements each of elemSize bytes.
896	void * aalloc( size_t dim, size_t elemSize ) libcfa_public {
897	size_t size = dim * elemSize;
898	#ifdef __STATISTICS__
899	if ( likely( size > 0 ) ) {
900	__atomic_add_fetch( &stats.aalloc_calls, 1, __ATOMIC_SEQ_CST );
901	__atomic_add_fetch( &stats.aalloc_storage_request, size, __ATOMIC_SEQ_CST );
902	} else {
903	__atomic_add_fetch( &stats.aalloc_0_calls, 1, __ATOMIC_SEQ_CST );
904	} // if
905	#endif // __STATISTICS__
906
907	return mallocNoStats( size );
908	} // aalloc
909
910
911	// Same as aalloc() with memory set to zero.
912	void * calloc( size_t dim, size_t elemSize ) libcfa_public {
913	size_t size = dim * elemSize;
914	if ( unlikely( size ) == 0 ) { // 0 BYTE ALLOCATION RETURNS NULL POINTER
915	#ifdef __STATISTICS__
916	__atomic_add_fetch( &stats.calloc_0_calls, 1, __ATOMIC_SEQ_CST );
917	#endif // __STATISTICS__
918	return 0p;
919	} // if
920	#ifdef __STATISTICS__
921	__atomic_add_fetch( &stats.calloc_calls, 1, __ATOMIC_SEQ_CST );
922	__atomic_add_fetch( &stats.calloc_storage_request, dim * elemSize, __ATOMIC_SEQ_CST );
923	#endif // __STATISTICS__
924
925	char * addr = (char *)mallocNoStats( size );
926
927	Heap.Storage.Header * header;
928	Heap.FreeHeader * freeElem;
929	size_t bsize, alignment;
930
931	#ifndef __CFA_DEBUG__
932	bool mapped =
933	#endif // __CFA_DEBUG__
934	headers( "calloc", addr, header, freeElem, bsize, alignment );
935
936	#ifndef __CFA_DEBUG__
937	// Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
938	if ( ! mapped )
939	#endif // __CFA_DEBUG__
940	// <-------0000000000000000000000000000UUUUUUUUUUUUUUUUUUUUUUUUU> bsize (bucket size) U => undefined
941	// `-header`-addr `-size
942	memset( addr, '\0', size ); // set to zeros
943
944	MarkZeroFilledBit( header ); // mark as zero fill
945	return addr;
946	} // calloc
947
948
949	// Change the size of the memory block pointed to by oaddr to size bytes. The contents are undefined. If oaddr is
950	// 0p, then the call is equivalent to malloc(size), for all values of size; if size is equal to zero, and oaddr is
951	// not 0p, then the call is equivalent to free(oaddr). Unless oaddr is 0p, it must have been returned by an earlier
952	// call to malloc(), alloc(), calloc() or realloc(). If the area pointed to was moved, a free(oaddr) is done.
953	void * resize( void * oaddr, size_t size ) libcfa_public {
954	// If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
955	if ( unlikely( size == 0 ) ) { // special cases
956	#ifdef __STATISTICS__
957	__atomic_add_fetch( &stats.resize_0_calls, 1, __ATOMIC_SEQ_CST );
958	#endif // __STATISTICS__
959	free( oaddr );
960	return 0p;
961	} // if
962	#ifdef __STATISTICS__
963	__atomic_add_fetch( &stats.resize_calls, 1, __ATOMIC_SEQ_CST );
964	#endif // __STATISTICS__
965
966	if ( unlikely( oaddr == 0p ) ) {
967	#ifdef __STATISTICS__
968	__atomic_add_fetch( &stats.resize_storage_request, size, __ATOMIC_SEQ_CST );
969	#endif // __STATISTICS__
970	return mallocNoStats( size );
971	} // if
972
973	Heap.Storage.Header * header;
974	Heap.FreeHeader * freeElem;
975	size_t bsize, oalign;
976	headers( "resize", oaddr, header, freeElem, bsize, oalign );
977
978	size_t odsize = DataStorage( bsize, oaddr, header ); // data storage available in bucket
979	// same size, DO NOT preserve STICKY PROPERTIES.
980	if ( oalign == libAlign() && size <= odsize && odsize <= size * 2 ) { // allow 50% wasted storage for smaller size
981	ClearZeroFillBit( header ); // no alignment and turn off 0 fill
982	header->kind.real.size = size; // reset allocation size
983	return oaddr;
984	} // if
985
986	#ifdef __STATISTICS__
987	__atomic_add_fetch( &stats.resize_storage_request, size, __ATOMIC_SEQ_CST );
988	#endif // __STATISTICS__
989
990	// change size, DO NOT preserve STICKY PROPERTIES.
991	free( oaddr );
992	return mallocNoStats( size ); // create new area
993	} // resize
994
995
996	// Same as resize() but the contents are unchanged in the range from the start of the region up to the minimum of
997	// the old and new sizes.
998	void * realloc( void * oaddr, size_t size ) libcfa_public {
999	// If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
1000	if ( unlikely( size == 0 ) ) { // special cases
1001	#ifdef __STATISTICS__
1002	__atomic_add_fetch( &stats.realloc_0_calls, 1, __ATOMIC_SEQ_CST );
1003	#endif // __STATISTICS__
1004	free( oaddr );
1005	return 0p;
1006	} // if
1007	#ifdef __STATISTICS__
1008	__atomic_add_fetch( &stats.realloc_calls, 1, __ATOMIC_SEQ_CST );
1009	#endif // __STATISTICS__
1010
1011	if ( unlikely( oaddr == 0p ) ) {
1012	#ifdef __STATISTICS__
1013	__atomic_add_fetch( &stats.realloc_storage_request, size, __ATOMIC_SEQ_CST );
1014	#endif // __STATISTICS__
1015	return mallocNoStats( size );
1016	} // if
1017
1018	Heap.Storage.Header * header;
1019	Heap.FreeHeader * freeElem;
1020	size_t bsize, oalign;
1021	headers( "realloc", oaddr, header, freeElem, bsize, oalign );
1022
1023	size_t odsize = DataStorage( bsize, oaddr, header ); // data storage available in bucket
1024	size_t osize = header->kind.real.size; // old allocation size
1025	bool ozfill = ZeroFillBit( header ); // old allocation zero filled
1026	if ( unlikely( size <= odsize ) && odsize <= size * 2 ) { // allow up to 50% wasted storage
1027	header->kind.real.size = size; // reset allocation size
1028	if ( unlikely( ozfill ) && size > osize ) { // previous request zero fill and larger ?
1029	memset( (char *)oaddr + osize, '\0', size - osize ); // initialize added storage
1030	} // if
1031	return oaddr;
1032	} // if
1033
1034	#ifdef __STATISTICS__
1035	__atomic_add_fetch( &stats.realloc_storage_request, size, __ATOMIC_SEQ_CST );
1036	#endif // __STATISTICS__
1037
1038	// change size and copy old content to new storage
1039
1040	void * naddr;
1041	if ( likely( oalign == libAlign() ) ) { // previous request not aligned ?
1042	naddr = mallocNoStats( size ); // create new area
1043	} else {
1044	naddr = memalignNoStats( oalign, size ); // create new aligned area
1045	} // if
1046
1047	headers( "realloc", naddr, header, freeElem, bsize, oalign );
1048	memcpy( naddr, oaddr, min( osize, size ) ); // copy bytes
1049	free( oaddr );
1050
1051	if ( unlikely( ozfill ) ) { // previous request zero fill ?
1052	MarkZeroFilledBit( header ); // mark new request as zero filled
1053	if ( size > osize ) { // previous request larger ?
1054	memset( (char *)naddr + osize, '\0', size - osize ); // initialize added storage
1055	} // if
1056	} // if
1057	return naddr;
1058	} // realloc
1059
1060
1061	// Same as realloc() except the new allocation size is large enough for an array of nelem elements of size elsize.
1062	void * reallocarray( void * oaddr, size_t dim, size_t elemSize ) libcfa_public {
1063	return realloc( oaddr, dim * elemSize );
1064	} // reallocarray
1065
1066
1067	// Same as malloc() except the memory address is a multiple of alignment, which must be a power of two. (obsolete)
1068	void * memalign( size_t alignment, size_t size ) libcfa_public {
1069	#ifdef __STATISTICS__
1070	if ( likely( size > 0 ) ) {
1071	__atomic_add_fetch( &stats.memalign_calls, 1, __ATOMIC_SEQ_CST );
1072	__atomic_add_fetch( &stats.memalign_storage_request, size, __ATOMIC_SEQ_CST );
1073	} else {
1074	__atomic_add_fetch( &stats.memalign_0_calls, 1, __ATOMIC_SEQ_CST );
1075	} // if
1076	#endif // __STATISTICS__
1077
1078	return memalignNoStats( alignment, size );
1079	} // memalign
1080
1081
1082	// Same as aalloc() with memory alignment.
1083	void * amemalign( size_t alignment, size_t dim, size_t elemSize ) libcfa_public {
1084	size_t size = dim * elemSize;
1085	#ifdef __STATISTICS__
1086	if ( likely( size > 0 ) ) {
1087	__atomic_add_fetch( &stats.cmemalign_calls, 1, __ATOMIC_SEQ_CST );
1088	__atomic_add_fetch( &stats.cmemalign_storage_request, size, __ATOMIC_SEQ_CST );
1089	} else {
1090	__atomic_add_fetch( &stats.cmemalign_0_calls, 1, __ATOMIC_SEQ_CST );
1091	} // if
1092	#endif // __STATISTICS__
1093
1094	return memalignNoStats( alignment, size );
1095	} // amemalign
1096
1097
1098	// Same as calloc() with memory alignment.
1099	void * cmemalign( size_t alignment, size_t dim, size_t elemSize ) libcfa_public {
1100	size_t size = dim * elemSize;
1101	if ( unlikely( size ) == 0 ) { // 0 BYTE ALLOCATION RETURNS NULL POINTER
1102	#ifdef __STATISTICS__
1103	__atomic_add_fetch( &stats.cmemalign_0_calls, 1, __ATOMIC_SEQ_CST );
1104	#endif // __STATISTICS__
1105	return 0p;
1106	} // if
1107	#ifdef __STATISTICS__
1108	__atomic_add_fetch( &stats.cmemalign_calls, 1, __ATOMIC_SEQ_CST );
1109	__atomic_add_fetch( &stats.cmemalign_storage_request, dim * elemSize, __ATOMIC_SEQ_CST );
1110	#endif // __STATISTICS__
1111
1112	char * addr = (char *)memalignNoStats( alignment, size );
1113
1114	Heap.Storage.Header * header;
1115	Heap.FreeHeader * freeElem;
1116	size_t bsize;
1117
1118	#ifndef __CFA_DEBUG__
1119	bool mapped =
1120	#endif // __CFA_DEBUG__
1121	headers( "cmemalign", addr, header, freeElem, bsize, alignment );
1122
1123	// Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
1124	#ifndef __CFA_DEBUG__
1125	if ( ! mapped )
1126	#endif // __CFA_DEBUG__
1127	// <-------0000000000000000000000000000UUUUUUUUUUUUUUUUUUUUUUUUU> bsize (bucket size) U => undefined
1128	// `-header`-addr `-size
1129	memset( addr, '\0', size ); // set to zeros
1130
1131	MarkZeroFilledBit( header ); // mark as zero filled
1132	return addr;
1133	} // cmemalign
1134
1135
1136	// Same as memalign(), but ISO/IEC 2011 C11 Section 7.22.2 states: the value of size shall be an integral multiple
1137	// of alignment. This requirement is universally ignored.
1138	void * aligned_alloc( size_t alignment, size_t size ) libcfa_public {
1139	return memalign( alignment, size );
1140	} // aligned_alloc
1141
1142
1143	// Allocates size bytes and places the address of the allocated memory in *memptr. The address of the allocated
1144	// memory shall be a multiple of alignment, which must be a power of two and a multiple of sizeof(void *). If size
1145	// is 0, then posix_memalign() returns either 0p, or a unique pointer value that can later be successfully passed to
1146	// free(3).
1147	int posix_memalign( void ** memptr, size_t alignment, size_t size ) libcfa_public {
1148	if ( unlikely( alignment < libAlign() \|\| ! is_pow2( alignment ) ) ) return EINVAL; // check alignment
1149	*memptr = memalign( alignment, size );
1150	return 0;
1151	} // posix_memalign
1152
1153
1154	// Allocates size bytes and returns a pointer to the allocated memory. The memory address shall be a multiple of the
1155	// page size. It is equivalent to memalign(sysconf(_SC_PAGESIZE),size).
1156	void * valloc( size_t size ) libcfa_public {
1157	return memalign( __page_size, size );
1158	} // valloc
1159
1160
1161	// Same as valloc but rounds size to multiple of page size.
1162	void * pvalloc( size_t size ) libcfa_public {
1163	return memalign( __page_size, ceiling2( size, __page_size ) ); // round size to multiple of page size
1164	} // pvalloc
1165
1166
1167	// Frees the memory space pointed to by ptr, which must have been returned by a previous call to malloc(), calloc()
1168	// or realloc(). Otherwise, or if free(ptr) has already been called before, undefined behaviour occurs. If ptr is
1169	// 0p, no operation is performed.
1170	void free( void * addr ) libcfa_public {
1171	if ( unlikely( addr == 0p ) ) { // special case
1172	#ifdef __STATISTICS__
1173	__atomic_add_fetch( &stats.free_null_calls, 1, __ATOMIC_SEQ_CST );
1174	#endif // __STATISTICS__
1175
1176	// #ifdef __CFA_DEBUG__
1177	// if ( traceHeap() ) {
1178	// #define nullmsg "Free( 0x0 ) size:0\n"
1179	// // Do not debug print free( 0p ), as it can cause recursive entry from sprintf.
1180	// __cfaabi_dbg_write( nullmsg, sizeof(nullmsg) - 1 );
1181	// } // if
1182	// #endif // __CFA_DEBUG__
1183	return;
1184	} // exit
1185
1186	doFree( addr );
1187	} // free
1188
1189
1190	// Returns the alignment of an allocation.
1191	size_t malloc_alignment( void * addr ) libcfa_public {
1192	if ( unlikely( addr == 0p ) ) return libAlign(); // minimum alignment
1193	Heap.Storage.Header * header = HeaderAddr( addr );
1194	if ( unlikely( AlignmentBit( header ) ) ) { // fake header ?
1195	return ClearAlignmentBit( header ); // clear flag from value
1196	} else {
1197	return libAlign(); // minimum alignment
1198	} // if
1199	} // malloc_alignment
1200
1201
1202	// Returns true if the allocation is zero filled, e.g., allocated by calloc().
1203	bool malloc_zero_fill( void * addr ) libcfa_public {
1204	if ( unlikely( addr == 0p ) ) return false; // null allocation is not zero fill
1205	Heap.Storage.Header * header = HeaderAddr( addr );
1206	if ( unlikely( AlignmentBit( header ) ) ) { // fake header ?
1207	header = RealHeader( header ); // backup from fake to real header
1208	} // if
1209	return ZeroFillBit( header ); // zero filled ?
1210	} // malloc_zero_fill
1211
1212
1213	// Returns original total allocation size (not bucket size) => array size is dimension * sizeof(T).
1214	size_t malloc_size( void * addr ) libcfa_public {
1215	if ( unlikely( addr == 0p ) ) return 0; // null allocation has zero size
1216	Heap.Storage.Header * header = HeaderAddr( addr );
1217	if ( unlikely( AlignmentBit( header ) ) ) { // fake header ?
1218	header = RealHeader( header ); // backup from fake to real header
1219	} // if
1220	return header->kind.real.size;
1221	} // malloc_size
1222
1223
1224	// Returns the number of usable bytes in the block pointed to by ptr, a pointer to a block of memory allocated by
1225	// malloc or a related function.
1226	size_t malloc_usable_size( void * addr ) libcfa_public {
1227	if ( unlikely( addr == 0p ) ) return 0; // null allocation has 0 size
1228	Heap.Storage.Header * header;
1229	Heap.FreeHeader * freeElem;
1230	size_t bsize, alignment;
1231
1232	headers( "malloc_usable_size", addr, header, freeElem, bsize, alignment );
1233	return DataStorage( bsize, addr, header ); // data storage in bucket
1234	} // malloc_usable_size
1235
1236
1237	// Prints (on default standard error) statistics about memory allocated by malloc and related functions.
1238	void malloc_stats( void ) libcfa_public {
1239	#ifdef __STATISTICS__
1240	printStats();
1241	if ( prtFree() ) prtFree( heapManager );
1242	#endif // __STATISTICS__
1243	} // malloc_stats
1244
1245
1246	// Changes the file descriptor where malloc_stats() writes statistics.
1247	int malloc_stats_fd( int fd __attribute__(( unused )) ) libcfa_public {
1248	#ifdef __STATISTICS__
1249	int temp = stats_fd;
1250	stats_fd = fd;
1251	return temp;
1252	#else
1253	return -1; // unsupported
1254	#endif // __STATISTICS__
1255	} // malloc_stats_fd
1256
1257
1258	// Prints an XML string that describes the current state of the memory-allocation implementation in the caller.
1259	// The string is printed on the file stream stream. The exported string includes information about all arenas (see
1260	// malloc).
1261	int malloc_info( int options, FILE * stream __attribute__(( unused )) ) libcfa_public {
1262	if ( options != 0 ) { errno = EINVAL; return -1; }
1263	#ifdef __STATISTICS__
1264	return printStatsXML( stream );
1265	#else
1266	return 0; // unsupported
1267	#endif // __STATISTICS__
1268	} // malloc_info
1269
1270
1271	// Adjusts parameters that control the behaviour of the memory-allocation functions (see malloc). The param argument
1272	// specifies the parameter to be modified, and value specifies the new value for that parameter.
1273	int mallopt( int option, int value ) libcfa_public {
1274	if ( value < 0 ) return 0;
1275	choose( option ) {
1276	case M_TOP_PAD:
1277	heapExpand = ceiling2( value, __page_size );
1278	return 1;
1279	case M_MMAP_THRESHOLD:
1280	if ( setMmapStart( value ) ) return 1;
1281	} // choose
1282	return 0; // error, unsupported
1283	} // mallopt
1284
1285
1286	// Attempt to release free memory at the top of the heap (by calling sbrk with a suitable argument).
1287	int malloc_trim( size_t ) libcfa_public {
1288	return 0; // => impossible to release memory
1289	} // malloc_trim
1290
1291
1292	// Records the current state of all malloc internal bookkeeping variables (but not the actual contents of the heap
1293	// or the state of malloc_hook functions pointers). The state is recorded in a system-dependent opaque data
1294	// structure dynamically allocated via malloc, and a pointer to that data structure is returned as the function
1295	// result. (The caller must free this memory.)
1296	void * malloc_get_state( void ) libcfa_public {
1297	return 0p; // unsupported
1298	} // malloc_get_state
1299
1300
1301	// Restores the state of all malloc internal bookkeeping variables to the values recorded in the opaque data
1302	// structure pointed to by state.
1303	int malloc_set_state( void * ) libcfa_public {
1304	return 0; // unsupported
1305	} // malloc_set_state
1306
1307
1308	// Sets the amount (bytes) to extend the heap when there is insufficent free storage to service an allocation.
1309	__attribute__((weak)) size_t malloc_expansion() libcfa_public { return __CFA_DEFAULT_HEAP_EXPANSION__; }
1310
1311	// Sets the crossover point between allocations occuring in the sbrk area or separately mmapped.
1312	__attribute__((weak)) size_t malloc_mmap_start() libcfa_public { return __CFA_DEFAULT_MMAP_START__; }
1313
1314	// Amount subtracted to adjust for unfreed program storage (debug only).
1315	__attribute__((weak)) size_t malloc_unfreed() libcfa_public { return __CFA_DEFAULT_HEAP_UNFREED__; }
1316	} // extern "C"
1317
1318
1319	// Must have CFA linkage to overload with C linkage realloc.
1320	void * resize( void * oaddr, size_t nalign, size_t size ) libcfa_public {
1321	// If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
1322	if ( unlikely( size == 0 ) ) { // special cases
1323	#ifdef __STATISTICS__
1324	__atomic_add_fetch( &stats.resize_0_calls, 1, __ATOMIC_SEQ_CST );
1325	#endif // __STATISTICS__
1326	free( oaddr );
1327	return 0p;
1328	} // if
1329
1330	if ( unlikely( nalign < libAlign() ) ) nalign = libAlign(); // reset alignment to minimum
1331	#ifdef __CFA_DEBUG__
1332	else checkAlign( nalign ); // check alignment
1333	#endif // __CFA_DEBUG__
1334
1335	if ( unlikely( oaddr == 0p ) ) {
1336	#ifdef __STATISTICS__
1337	__atomic_add_fetch( &stats.resize_calls, 1, __ATOMIC_SEQ_CST );
1338	__atomic_add_fetch( &stats.resize_storage_request, size, __ATOMIC_SEQ_CST );
1339	#endif // __STATISTICS__
1340	return memalignNoStats( nalign, size );
1341	} // if
1342
1343	// Attempt to reuse existing alignment.
1344	Heap.Storage.Header * header = HeaderAddr( oaddr );
1345	bool isFakeHeader = AlignmentBit( header ); // old fake header ?
1346	size_t oalign;
1347
1348	if ( unlikely( isFakeHeader ) ) {
1349	oalign = ClearAlignmentBit( header ); // old alignment
1350	if ( unlikely( (uintptr_t)oaddr % nalign == 0 // lucky match ?
1351	&& ( oalign <= nalign // going down
1352	\|\| (oalign >= nalign && oalign <= 256) ) // little alignment storage wasted ?
1353	) ) {
1354	HeaderAddr( oaddr )->kind.fake.alignment = MarkAlignmentBit( nalign ); // update alignment (could be the same)
1355	Heap.FreeHeader * freeElem;
1356	size_t bsize, oalign;
1357	headers( "resize", oaddr, header, freeElem, bsize, oalign );
1358	size_t odsize = DataStorage( bsize, oaddr, header ); // data storage available in bucket
1359
1360	if ( size <= odsize && odsize <= size * 2 ) { // allow 50% wasted data storage
1361	HeaderAddr( oaddr )->kind.fake.alignment = MarkAlignmentBit( nalign ); // update alignment (could be the same)
1362	ClearZeroFillBit( header ); // turn off 0 fill
1363	header->kind.real.size = size; // reset allocation size
1364	return oaddr;
1365	} // if
1366	} // if
1367	} else if ( ! isFakeHeader // old real header (aligned on libAlign) ?
1368	&& nalign == libAlign() ) { // new alignment also on libAlign => no fake header needed
1369	return resize( oaddr, size ); // duplicate special case checks
1370	} // if
1371
1372	#ifdef __STATISTICS__
1373	__atomic_add_fetch( &stats.resize_storage_request, size, __ATOMIC_SEQ_CST );
1374	#endif // __STATISTICS__
1375
1376	// change size, DO NOT preserve STICKY PROPERTIES.
1377	free( oaddr );
1378	return memalignNoStats( nalign, size ); // create new aligned area
1379	} // resize
1380
1381
1382	void * realloc( void * oaddr, size_t nalign, size_t size ) libcfa_public {
1383	// If size is equal to 0, either NULL or a pointer suitable to be passed to free() is returned.
1384	if ( unlikely( size == 0 ) ) { // special cases
1385	#ifdef __STATISTICS__
1386	__atomic_add_fetch( &stats.realloc_0_calls, 1, __ATOMIC_SEQ_CST );
1387	#endif // __STATISTICS__
1388	free( oaddr );
1389	return 0p;
1390	} // if
1391
1392	if ( unlikely( nalign < libAlign() ) ) nalign = libAlign(); // reset alignment to minimum
1393	#ifdef __CFA_DEBUG__
1394	else checkAlign( nalign ); // check alignment
1395	#endif // __CFA_DEBUG__
1396
1397	if ( unlikely( oaddr == 0p ) ) {
1398	#ifdef __STATISTICS__
1399	__atomic_add_fetch( &stats.realloc_calls, 1, __ATOMIC_SEQ_CST );
1400	__atomic_add_fetch( &stats.realloc_storage_request, size, __ATOMIC_SEQ_CST );
1401	#endif // __STATISTICS__
1402	return memalignNoStats( nalign, size );
1403	} // if
1404
1405	// Attempt to reuse existing alignment.
1406	Heap.Storage.Header * header = HeaderAddr( oaddr );
1407	bool isFakeHeader = AlignmentBit( header ); // old fake header ?
1408	size_t oalign;
1409	if ( unlikely( isFakeHeader ) ) {
1410	oalign = ClearAlignmentBit( header ); // old alignment
1411	if ( unlikely( (uintptr_t)oaddr % nalign == 0 // lucky match ?
1412	&& ( oalign <= nalign // going down
1413	\|\| (oalign >= nalign && oalign <= 256) ) // little alignment storage wasted ?
1414	) ) {
1415	HeaderAddr( oaddr )->kind.fake.alignment = MarkAlignmentBit( nalign ); // update alignment (could be the same)
1416	return realloc( oaddr, size ); // duplicate special case checks
1417	} // if
1418	} else if ( ! isFakeHeader // old real header (aligned on libAlign) ?
1419	&& nalign == libAlign() ) { // new alignment also on libAlign => no fake header needed
1420	return realloc( oaddr, size ); // duplicate special case checks
1421	} // if
1422
1423	#ifdef __STATISTICS__
1424	__atomic_add_fetch( &stats.realloc_calls, 1, __ATOMIC_SEQ_CST );
1425	__atomic_add_fetch( &stats.realloc_storage_request, size, __ATOMIC_SEQ_CST );
1426	#endif // __STATISTICS__
1427
1428	Heap.FreeHeader * freeElem;
1429	size_t bsize;
1430	headers( "realloc", oaddr, header, freeElem, bsize, oalign );
1431
1432	// change size and copy old content to new storage
1433
1434	size_t osize = header->kind.real.size; // old allocation size
1435	bool ozfill = ZeroFillBit( header ); // old allocation zero filled
1436
1437	void * naddr = memalignNoStats( nalign, size ); // create new aligned area
1438
1439	headers( "realloc", naddr, header, freeElem, bsize, oalign );
1440	memcpy( naddr, oaddr, min( osize, size ) ); // copy bytes
1441	free( oaddr );
1442
1443	if ( unlikely( ozfill ) ) { // previous request zero fill ?
1444	MarkZeroFilledBit( header ); // mark new request as zero filled
1445	if ( size > osize ) { // previous request larger ?
1446	memset( (char *)naddr + osize, '\0', size - osize ); // initialize added storage
1447	} // if
1448	} // if
1449	return naddr;
1450	} // realloc
1451
1452
1453	// Local Variables: //
1454	// tab-width: 4 //
1455	// compile-command: "cfa -nodebug -O2 heap.cfa" //
1456	// End: //

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

Download in other formats:

Original Format