Context Navigation

heap.cfa @ f1ec88a

ADTarm-ehast-experimentalcleanup-dtorsenumforall-pointer-decayjacob/cs343-translationjenkins-sandboxnew-astnew-ast-unique-exprpthread-emulationqualifiedEnum

Last change on this file since f1ec88a was 7b149bc, checked in by Peter A. Buhr <pabuhr@…>, 5 years ago
formatting
Property mode set to `100644`
File size: 39.5 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.c --
8	//
9	// Author : Peter A. Buhr
10	// Created On : Tue Dec 19 21:58:35 2017
11	// Last Modified By : Peter A. Buhr
12	// Last Modified On : Thu May 9 16:29:12 2019
13	// Update Count : 516
14	//
15
16	#include <unistd.h> // sbrk, sysconf
17	#include <stdbool.h> // true, false
18	#include <stdio.h> // snprintf, fileno
19	#include <errno.h> // errno
20	extern "C" {
21	#include <sys/mman.h> // mmap, munmap
22	} // extern "C"
23
24	// #comment TD : Many of these should be merged into math I believe
25	#include "bits/align.hfa" // libPow2
26	#include "bits/defs.hfa" // likely, unlikely
27	#include "bits/locks.hfa" // __spinlock_t
28	#include "startup.hfa" // STARTUP_PRIORITY_MEMORY
29	#include "stdlib.hfa" // bsearchl
30	#include "malloc.h"
31
32
33	enum {
34	__CFA_DEFAULT_MMAP_START__ = (512 * 1024 + 1),
35	__CFA_DEFAULT_HEAP_EXPANSION__ = (1 * 1024 * 1024),
36	};
37
38	size_t default_mmap_start() __attribute__(( weak )) {
39	return __CFA_DEFAULT_MMAP_START__;
40	} // default_mmap_start
41
42	size_t default_heap_expansion() __attribute__(( weak )) {
43	return __CFA_DEFAULT_HEAP_EXPANSION__;
44	} // default_heap_expansion
45
46
47	// supported mallopt options
48	#ifndef M_MMAP_THRESHOLD
49	#define M_MMAP_THRESHOLD (-1)
50	#endif // M_TOP_PAD
51	#ifndef M_TOP_PAD
52	#define M_TOP_PAD (-2)
53	#endif // M_TOP_PAD
54
55	#define FASTLOOKUP
56	#define __STATISTICS__
57
58	#define SPINLOCK 0
59	#define LOCKFREE 1
60	#define BUCKETLOCK SPINLOCK
61	#if BUCKETLOCK == LOCKFREE
62	#include <uStackLF.h>
63	#endif // LOCKFREE
64
65	// #comment TD : This defined is significantly different from the __ALIGN__ define from locks.hfa
66	#define ALIGN 16
67
68	// enum { NoBucketSizes = 93, // number of buckets sizes
69	// #ifdef FASTLOOKUP
70	// LookupSizes = 65536, // number of fast lookup sizes
71	// #endif // FASTLOOKUP
72	// };
73	#define NoBucketSizes 93 // number of buckets sizes
74	#ifdef FASTLOOKUP
75	#define LookupSizes 65536 // number of fast lookup sizes
76	#endif // FASTLOOKUP
77
78
79	static bool traceHeap = false;
80
81	inline bool traceHeap() {
82	return traceHeap;
83	} // traceHeap
84
85	bool traceHeapOn() {
86	bool temp = traceHeap;
87	traceHeap = true;
88	return temp;
89	} // traceHeapOn
90
91	bool traceHeapOff() {
92	bool temp = traceHeap;
93	traceHeap = false;
94	return temp;
95	} // traceHeapOff
96
97
98	static bool checkFree = false;
99
100	inline bool checkFree() {
101	return checkFree;
102	} // checkFree
103
104	bool checkFreeOn() {
105	bool temp = checkFree;
106	checkFree = true;
107	return temp;
108	} // checkFreeOn
109
110	bool checkFreeOff() {
111	bool temp = checkFree;
112	checkFree = false;
113	return temp;
114	} // checkFreeOff
115
116
117	// static bool traceHeapTerm = false;
118
119	// inline bool traceHeapTerm() {
120	// return traceHeapTerm;
121	// } // traceHeapTerm
122
123	// bool traceHeapTermOn() {
124	// bool temp = traceHeapTerm;
125	// traceHeapTerm = true;
126	// return temp;
127	// } // traceHeapTermOn
128
129	// bool traceHeapTermOff() {
130	// bool temp = traceHeapTerm;
131	// traceHeapTerm = false;
132	// return temp;
133	// } // traceHeapTermOff
134
135
136	#ifdef __CFA_DEBUG__
137	static unsigned int allocFree; // running total of allocations minus frees
138
139	static void checkUnfreed() {
140	if ( allocFree != 0 ) {
141	// DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT.
142	// char helpText[512];
143	// int len = snprintf( helpText, sizeof(helpText), "CFA warning (UNIX pid:%ld) : program terminating with %u(0x%x) bytes of storage allocated but not freed.\n"
144	// "Possible cause is unfreed storage allocated by the program or system/library routines called from the program.\n",
145	// (long int)getpid(), allocFree, allocFree ); // always print the UNIX pid
146	// __cfaabi_dbg_bits_write( helpText, len );
147	} // if
148	} // checkUnfreed
149
150	extern "C" {
151	void heapAppStart() { // called by __cfaabi_appready_startup
152	allocFree = 0;
153	} // heapAppStart
154
155	void heapAppStop() { // called by __cfaabi_appready_startdown
156	fclose( stdin ); fclose( stdout );
157	checkUnfreed();
158	} // heapAppStop
159	} // extern "C"
160	#endif // __CFA_DEBUG__
161
162
163	struct HeapManager {
164	// struct FreeHeader; // forward declaration
165
166	struct Storage {
167	struct Header { // header
168	union Kind {
169	struct RealHeader {
170	union {
171	struct { // 4-byte word => 8-byte header, 8-byte word => 16-byte header
172	#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4
173	uint32_t padding; // unused, force home/blocksize to overlay alignment in fake header
174	#endif // __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ && __SIZEOF_POINTER__ == 4
175
176	union {
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 BUCKLOCK == SPINLOCK
181	Storage * next; // freed block points next freed block of same size
182	#endif // SPINLOCK
183	};
184
185	#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4
186	uint32_t padding; // unused, force home/blocksize to overlay alignment in fake header
187	#endif // __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ && __SIZEOF_POINTER__ == 4
188	};
189	// future code
190	#if BUCKLOCK == LOCKFREE
191	Stack<Storage>::Link next; // freed block points next freed block of same size (double-wide)
192	#endif // LOCKFREE
193	};
194	} real; // RealHeader
195	struct FakeHeader {
196	#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
197	uint32_t alignment; // low-order bits of home/blockSize used for tricks
198	#endif // __ORDER_LITTLE_ENDIAN__
199
200	uint32_t offset;
201
202	#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
203	uint32_t alignment; // low-order bits of home/blockSize used for tricks
204	#endif // __ORDER_BIG_ENDIAN__
205	} fake; // FakeHeader
206	} kind; // Kind
207	} header; // Header
208	char pad[ALIGN - sizeof( Header )];
209	char data[0]; // storage
210	}; // Storage
211
212	static_assert( ALIGN >= sizeof( Storage ), "ALIGN < sizeof( Storage )" );
213
214	struct FreeHeader {
215	#if BUCKLOCK == SPINLOCK
216	__spinlock_t lock; // must be first field for alignment
217	Storage * freeList;
218	#elif BUCKLOCK == LOCKFREE
219	// future code
220	StackLF<Storage> freeList;
221	#else
222	#error undefined lock type for bucket lock
223	#endif // SPINLOCK
224	size_t blockSize; // size of allocations on this list
225	}; // FreeHeader
226
227	// must be first fields for alignment
228	__spinlock_t extlock; // protects allocation-buffer extension
229	FreeHeader freeLists[NoBucketSizes]; // buckets for different allocation sizes
230
231	void * heapBegin; // start of heap
232	void * heapEnd; // logical end of heap
233	size_t heapRemaining; // amount of storage not allocated in the current chunk
234	}; // HeapManager
235
236
237	static inline size_t getKey( const HeapManager.FreeHeader & freeheader ) { return freeheader.blockSize; }
238
239	// statically allocated variables => zero filled.
240	static size_t pageSize; // architecture pagesize
241	static size_t heapExpand; // sbrk advance
242	static size_t mmapStart; // cross over point for mmap
243	static unsigned int maxBucketsUsed; // maximum number of buckets in use
244
245	// Powers of 2 are common allocation sizes, so make powers of 2 generate the minimum required size.
246	static const unsigned int bucketSizes[NoBucketSizes] @= { // different bucket sizes
247	16, 32, 48, 64,
248	64 + sizeof(HeapManager.Storage), 96, 112, 128, 128 + sizeof(HeapManager.Storage), 160, 192, 224,
249	256 + sizeof(HeapManager.Storage), 320, 384, 448, 512 + sizeof(HeapManager.Storage), 640, 768, 896,
250	1_024 + sizeof(HeapManager.Storage), 1_536, 2_048 + sizeof(HeapManager.Storage), 2_560, 3_072, 3_584, 4_096 + sizeof(HeapManager.Storage), 6_144,
251	8_192 + sizeof(HeapManager.Storage), 9_216, 10_240, 11_264, 12_288, 13_312, 14_336, 15_360,
252	16_384 + sizeof(HeapManager.Storage), 18_432, 20_480, 22_528, 24_576, 26_624, 28_672, 30_720,
253	32_768 + sizeof(HeapManager.Storage), 36_864, 40_960, 45_056, 49_152, 53_248, 57_344, 61_440,
254	65_536 + sizeof(HeapManager.Storage), 73_728, 81_920, 90_112, 98_304, 106_496, 114_688, 122_880,
255	131_072 + sizeof(HeapManager.Storage), 147_456, 163_840, 180_224, 196_608, 212_992, 229_376, 245_760,
256	262_144 + sizeof(HeapManager.Storage), 294_912, 327_680, 360_448, 393_216, 425_984, 458_752, 491_520,
257	524_288 + sizeof(HeapManager.Storage), 655_360, 786_432, 917_504, 1_048_576 + sizeof(HeapManager.Storage), 1_179_648, 1_310_720, 1_441_792,
258	1_572_864, 1_703_936, 1_835_008, 1_966_080, 2_097_152 + sizeof(HeapManager.Storage), 2_621_440, 3_145_728, 3_670_016,
259	4_194_304 + sizeof(HeapManager.Storage)
260	};
261	#ifdef FASTLOOKUP
262	static unsigned char lookup[LookupSizes]; // O(1) lookup for small sizes
263	#endif // FASTLOOKUP
264	static int mmapFd = -1; // fake or actual fd for anonymous file
265
266
267	#ifdef __CFA_DEBUG__
268	static bool heapBoot = 0; // detect recursion during boot
269	#endif // __CFA_DEBUG__
270	static HeapManager heapManager __attribute__(( aligned (128) )) @= {}; // size of cache line to prevent false sharing
271
272	// #comment TD : The return type of this function should be commented
273	static inline bool setMmapStart( size_t value ) {
274	if ( value < pageSize \|\| bucketSizes[NoBucketSizes - 1] < value ) return true;
275	mmapStart = value; // set global
276
277	// find the closest bucket size less than or equal to the mmapStart size
278	maxBucketsUsed = bsearchl( (unsigned int)mmapStart, bucketSizes, NoBucketSizes ); // binary search
279	assert( maxBucketsUsed < NoBucketSizes ); // subscript failure ?
280	assert( mmapStart <= bucketSizes[maxBucketsUsed] ); // search failure ?
281	return false;
282	} // setMmapStart
283
284
285	static void ?{}( HeapManager & manager ) with ( manager ) {
286	pageSize = sysconf( _SC_PAGESIZE );
287
288	for ( unsigned int i = 0; i < NoBucketSizes; i += 1 ) { // initialize the free lists
289	freeLists[i].blockSize = bucketSizes[i];
290	} // for
291
292	#ifdef FASTLOOKUP
293	unsigned int idx = 0;
294	for ( unsigned int i = 0; i < LookupSizes; i += 1 ) {
295	if ( i > bucketSizes[idx] ) idx += 1;
296	lookup[i] = idx;
297	} // for
298	#endif // FASTLOOKUP
299
300	if ( setMmapStart( default_mmap_start() ) ) {
301	abort( "HeapManager : internal error, mmap start initialization failure." );
302	} // if
303	heapExpand = default_heap_expansion();
304
305	char * End = (char *)sbrk( 0 );
306	sbrk( (char *)libCeiling( (long unsigned int)End, libAlign() ) - End ); // move start of heap to multiple of alignment
307	heapBegin = heapEnd = sbrk( 0 ); // get new start point
308	} // HeapManager
309
310
311	static void ^?{}( HeapManager & ) {
312	#ifdef __STATISTICS__
313	// if ( traceHeapTerm() ) {
314	// printStats();
315	// if ( checkfree() ) checkFree( heapManager, true );
316	// } // if
317	#endif // __STATISTICS__
318	} // ~HeapManager
319
320
321	static void memory_startup( void ) __attribute__(( constructor( STARTUP_PRIORITY_MEMORY ) ));
322	void memory_startup( void ) {
323	#ifdef __CFA_DEBUG__
324	if ( unlikely( heapBoot ) ) { // check for recursion during system boot
325	// DO NOT USE STREAMS AS THEY MAY BE UNAVAILABLE AT THIS POINT.
326	abort( "boot() : internal error, recursively invoked during system boot." );
327	} // if
328	heapBoot = true;
329	#endif // __CFA_DEBUG__
330
331	//assert( heapManager.heapBegin != 0 );
332	//heapManager{};
333	if ( heapManager.heapBegin == 0 ) heapManager{};
334	} // memory_startup
335
336	static void memory_shutdown( void ) __attribute__(( destructor( STARTUP_PRIORITY_MEMORY ) ));
337	void memory_shutdown( void ) {
338	^heapManager{};
339	} // memory_shutdown
340
341
342	#ifdef __STATISTICS__
343	static unsigned long long int mmap_storage; // heap statistics counters
344	static unsigned int mmap_calls;
345	static unsigned long long int munmap_storage;
346	static unsigned int munmap_calls;
347	static unsigned long long int sbrk_storage;
348	static unsigned int sbrk_calls;
349	static unsigned long long int malloc_storage;
350	static unsigned int malloc_calls;
351	static unsigned long long int free_storage;
352	static unsigned int free_calls;
353	static unsigned long long int calloc_storage;
354	static unsigned int calloc_calls;
355	static unsigned long long int memalign_storage;
356	static unsigned int memalign_calls;
357	static unsigned long long int cmemalign_storage;
358	static unsigned int cmemalign_calls;
359	static unsigned long long int realloc_storage;
360	static unsigned int realloc_calls;
361
362	static int statfd; // statistics file descriptor (changed by malloc_stats_fd)
363
364
365	// Use "write" because streams may be shutdown when calls are made.
366	static void printStats() {
367	char helpText[512];
368	__cfaabi_dbg_bits_print_buffer( helpText, sizeof(helpText),
369	"\nHeap statistics:\n"
370	" malloc: calls %u / storage %llu\n"
371	" calloc: calls %u / storage %llu\n"
372	" memalign: calls %u / storage %llu\n"
373	" cmemalign: calls %u / storage %llu\n"
374	" realloc: calls %u / storage %llu\n"
375	" free: calls %u / storage %llu\n"
376	" mmap: calls %u / storage %llu\n"
377	" munmap: calls %u / storage %llu\n"
378	" sbrk: calls %u / storage %llu\n",
379	malloc_calls, malloc_storage,
380	calloc_calls, calloc_storage,
381	memalign_calls, memalign_storage,
382	cmemalign_calls, cmemalign_storage,
383	realloc_calls, realloc_storage,
384	free_calls, free_storage,
385	mmap_calls, mmap_storage,
386	munmap_calls, munmap_storage,
387	sbrk_calls, sbrk_storage
388	);
389	} // printStats
390
391	static int printStatsXML( FILE * stream ) { // see malloc_info
392	char helpText[512];
393	int len = snprintf( helpText, sizeof(helpText),
394	"<malloc version=\"1\">\n"
395	"<heap nr=\"0\">\n"
396	"<sizes>\n"
397	"</sizes>\n"
398	"<total type=\"malloc\" count=\"%u\" size=\"%llu\"/>\n"
399	"<total type=\"calloc\" count=\"%u\" size=\"%llu\"/>\n"
400	"<total type=\"memalign\" count=\"%u\" size=\"%llu\"/>\n"
401	"<total type=\"cmemalign\" count=\"%u\" size=\"%llu\"/>\n"
402	"<total type=\"realloc\" count=\"%u\" size=\"%llu\"/>\n"
403	"<total type=\"free\" count=\"%u\" size=\"%llu\"/>\n"
404	"<total type=\"mmap\" count=\"%u\" size=\"%llu\"/>\n"
405	"<total type=\"munmap\" count=\"%u\" size=\"%llu\"/>\n"
406	"<total type=\"sbrk\" count=\"%u\" size=\"%llu\"/>\n"
407	"</malloc>",
408	malloc_calls, malloc_storage,
409	calloc_calls, calloc_storage,
410	memalign_calls, memalign_storage,
411	cmemalign_calls, cmemalign_storage,
412	realloc_calls, realloc_storage,
413	free_calls, free_storage,
414	mmap_calls, mmap_storage,
415	munmap_calls, munmap_storage,
416	sbrk_calls, sbrk_storage
417	);
418	return write( fileno( stream ), helpText, len ); // -1 => error
419	} // printStatsXML
420	#endif // __STATISTICS__
421
422	// #comment TD : Is this the samething as Out-of-Memory?
423	static inline void noMemory() {
424	abort( "Heap memory exhausted at %zu bytes.\n"
425	"Possible cause is very large memory allocation and/or large amount of unfreed storage allocated by the program or system/library routines.",
426	((char )(sbrk( 0 )) - (char )(heapManager.heapBegin)) );
427	} // noMemory
428
429
430	static inline void checkAlign( size_t alignment ) {
431	if ( alignment < sizeof(void *) \|\| ! libPow2( alignment ) ) {
432	abort( "Alignment %zu for memory allocation is less than sizeof(void *) and/or not a power of 2.", alignment );
433	} // if
434	} // checkAlign
435
436
437	static inline bool setHeapExpand( size_t value ) {
438	if ( heapExpand < pageSize ) return true;
439	heapExpand = value;
440	return false;
441	} // setHeapExpand
442
443
444	static inline void checkHeader( bool check, const char * name, void * addr ) {
445	if ( unlikely( check ) ) { // bad address ?
446	abort( "Attempt to %s storage %p with address outside the heap.\n"
447	"Possible cause is duplicate free on same block or overwriting of memory.",
448	name, addr );
449	} // if
450	} // checkHeader
451
452	// #comment TD : function should be commented and/or have a more evocative name
453	// this isn't either a check or a constructor which is what I would expect this function to be
454	static inline void fakeHeader( HeapManager.Storage.Header *& header, size_t & size, size_t & alignment ) {
455	if ( unlikely( (header->kind.fake.alignment & 1) == 1 ) ) { // fake header ?
456	size_t offset = header->kind.fake.offset;
457	alignment = header->kind.fake.alignment & -2; // remove flag from value
458	#ifdef __CFA_DEBUG__
459	checkAlign( alignment ); // check alignment
460	#endif // __CFA_DEBUG__
461	header = (HeapManager.Storage.Header )((char )header - offset);
462	} // if
463	} // fakeHeader
464
465	// #comment TD : Why is this a define
466	#define headerAddr( addr ) ((HeapManager.Storage.Header )( (char )addr - sizeof(HeapManager.Storage) ))
467
468	static inline bool headers( const char * name, void * addr, HeapManager.Storage.Header & header, HeapManager.FreeHeader & freeElem, size_t & size, size_t & alignment ) with ( heapManager ) {
469	header = headerAddr( addr );
470
471	if ( unlikely( heapEnd < addr ) ) { // mmapped ?
472	fakeHeader( header, size, alignment );
473	size = header->kind.real.blockSize & -3; // mmap size
474	return true;
475	} // if
476
477	#ifdef __CFA_DEBUG__
478	checkHeader( addr < heapBegin \|\| header < (HeapManager.Storage.Header *)heapBegin, name, addr ); // bad low address ?
479	#endif // __CFA_DEBUG__
480
481	// #comment TD : This code looks weird...
482	// It's called as the first statement of both branches of the last if, with the same parameters in all cases
483
484	// header may be safe to dereference
485	fakeHeader( header, size, alignment );
486	#ifdef __CFA_DEBUG__
487	checkHeader( header < (HeapManager.Storage.Header )heapBegin \|\| (HeapManager.Storage.Header )heapEnd < header, name, addr ); // bad address ? (offset could be + or -)
488	#endif // __CFA_DEBUG__
489
490	freeElem = (HeapManager.FreeHeader *)((size_t)header->kind.real.home & -3);
491	#ifdef __CFA_DEBUG__
492	if ( freeElem < &freeLists[0] \|\| &freeLists[NoBucketSizes] <= freeElem ) {
493	abort( "Attempt to %s storage %p with corrupted header.\n"
494	"Possible cause is duplicate free on same block or overwriting of header information.",
495	name, addr );
496	} // if
497	#endif // __CFA_DEBUG__
498	size = freeElem->blockSize;
499	return false;
500	} // headers
501
502
503	static inline void * extend( size_t size ) with ( heapManager ) {
504	lock( extlock __cfaabi_dbg_ctx2 );
505	ptrdiff_t rem = heapRemaining - size;
506	if ( rem < 0 ) {
507	// If the size requested is bigger than the current remaining storage, increase the size of the heap.
508
509	size_t increase = libCeiling( size > heapExpand ? size : heapExpand, libAlign() );
510	if ( sbrk( increase ) == (void *)-1 ) {
511	unlock( extlock );
512	errno = ENOMEM;
513	return 0;
514	} // if
515	#ifdef __STATISTICS__
516	sbrk_calls += 1;
517	sbrk_storage += increase;
518	#endif // __STATISTICS__
519	#ifdef __CFA_DEBUG__
520	// Set new memory to garbage so subsequent uninitialized usages might fail.
521	memset( (char *)heapEnd + heapRemaining, '\377', increase );
522	#endif // __CFA_DEBUG__
523	rem = heapRemaining + increase - size;
524	} // if
525
526	HeapManager.Storage * block = (HeapManager.Storage *)heapEnd;
527	heapRemaining = rem;
528	heapEnd = (char *)heapEnd + size;
529	unlock( extlock );
530	return block;
531	} // extend
532
533
534	static inline void * doMalloc( size_t size ) with ( heapManager ) {
535	HeapManager.Storage * block; // pointer to new block of storage
536
537	// Look up size in the size list. Make sure the user request includes space for the header that must be allocated
538	// along with the block and is a multiple of the alignment size.
539
540	size_t tsize = size + sizeof(HeapManager.Storage);
541	if ( likely( tsize < mmapStart ) ) { // small size => sbrk
542	HeapManager.FreeHeader * freeElem =
543	#ifdef FASTLOOKUP
544	tsize < LookupSizes ? &freeLists[lookup[tsize]] :
545	#endif // FASTLOOKUP
546	bsearchl( tsize, freeLists, (size_t)maxBucketsUsed ); // binary search
547	assert( freeElem <= &freeLists[maxBucketsUsed] ); // subscripting error ?
548	assert( tsize <= freeElem->blockSize ); // search failure ?
549	tsize = freeElem->blockSize; // total space needed for request
550
551	// Spin until the lock is acquired for this particular size of block.
552
553	#if defined( SPINLOCK )
554	lock( freeElem->lock __cfaabi_dbg_ctx2 );
555	block = freeElem->freeList; // remove node from stack
556	#else
557	block = freeElem->freeList.pop();
558	#endif // SPINLOCK
559	if ( unlikely( block == 0 ) ) { // no free block ?
560	#if defined( SPINLOCK )
561	unlock( freeElem->lock );
562	#endif // SPINLOCK
563
564	// Freelist for that size was empty, so carve it out of the heap if there's enough left, or get some more
565	// and then carve it off.
566
567	block = (HeapManager.Storage *)extend( tsize ); // mutual exclusion on call
568	if ( unlikely( block == 0 ) ) return 0;
569	#if defined( SPINLOCK )
570	} else {
571	freeElem->freeList = block->header.kind.real.next;
572	unlock( freeElem->lock );
573	#endif // SPINLOCK
574	} // if
575
576	block->header.kind.real.home = freeElem; // pointer back to free list of apropriate size
577	} else { // large size => mmap
578	tsize = libCeiling( tsize, pageSize ); // must be multiple of page size
579	#ifdef __STATISTICS__
580	__atomic_add_fetch( &mmap_calls, 1, __ATOMIC_SEQ_CST );
581	__atomic_add_fetch( &mmap_storage, tsize, __ATOMIC_SEQ_CST );
582	#endif // __STATISTICS__
583	block = (HeapManager.Storage *)mmap( 0, tsize, PROT_READ \| PROT_WRITE, MAP_PRIVATE \| MAP_ANONYMOUS, mmapFd, 0 );
584	if ( block == (HeapManager.Storage *)MAP_FAILED ) {
585	// Do not call strerror( errno ) as it may call malloc.
586	abort( "(HeapManager &)0x%p.doMalloc() : internal error, mmap failure, size:%zu error:%d.", &heapManager, tsize, errno );
587	} // if
588	#ifdef __CFA_DEBUG__
589	// Set new memory to garbage so subsequent uninitialized usages might fail.
590	memset( block, '\377', tsize );
591	#endif // __CFA_DEBUG__
592	block->header.kind.real.blockSize = tsize; // storage size for munmap
593	} // if
594
595	void * area = &(block->data); // adjust off header to user bytes
596
597	#ifdef __CFA_DEBUG__
598	assert( ((uintptr_t)area & (libAlign() - 1)) == 0 ); // minimum alignment ?
599	__atomic_add_fetch( &allocFree, tsize, __ATOMIC_SEQ_CST );
600	if ( traceHeap() ) {
601	enum { BufferSize = 64 };
602	char helpText[BufferSize];
603	int len = snprintf( helpText, BufferSize, "%p = Malloc( %zu ) (allocated %zu)\n", area, size, tsize );
604	// int len = snprintf( helpText, BufferSize, "Malloc %p %zu\n", area, size );
605	__cfaabi_dbg_bits_write( helpText, len );
606	} // if
607	#endif // __CFA_DEBUG__
608
609	return area;
610	} // doMalloc
611
612
613	static inline void doFree( void * addr ) with ( heapManager ) {
614	#ifdef __CFA_DEBUG__
615	if ( unlikely( heapManager.heapBegin == 0 ) ) {
616	abort( "doFree( %p ) : internal error, called before heap is initialized.", addr );
617	} // if
618	#endif // __CFA_DEBUG__
619
620	HeapManager.Storage.Header * header;
621	HeapManager.FreeHeader * freeElem;
622	size_t size, alignment; // not used (see realloc)
623
624	if ( headers( "free", addr, header, freeElem, size, alignment ) ) { // mmapped ?
625	#ifdef __STATISTICS__
626	__atomic_add_fetch( &munmap_calls, 1, __ATOMIC_SEQ_CST );
627	__atomic_add_fetch( &munmap_storage, size, __ATOMIC_SEQ_CST );
628	#endif // __STATISTICS__
629	if ( munmap( header, size ) == -1 ) {
630	#ifdef __CFA_DEBUG__
631	abort( "Attempt to deallocate storage %p not allocated or with corrupt header.\n"
632	"Possible cause is invalid pointer.",
633	addr );
634	#endif // __CFA_DEBUG__
635	} // if
636	} else {
637	#ifdef __CFA_DEBUG__
638	// Set free memory to garbage so subsequent usages might fail.
639	memset( ((HeapManager.Storage *)header)->data, '\377', freeElem->blockSize - sizeof( HeapManager.Storage ) );
640	#endif // __CFA_DEBUG__
641
642	#ifdef __STATISTICS__
643	free_storage += size;
644	#endif // __STATISTICS__
645	#if defined( SPINLOCK )
646	lock( freeElem->lock __cfaabi_dbg_ctx2 ); // acquire spin lock
647	header->kind.real.next = freeElem->freeList; // push on stack
648	freeElem->freeList = (HeapManager.Storage *)header;
649	unlock( freeElem->lock ); // release spin lock
650	#else
651	freeElem->freeList.push( (HeapManager.Storage )header );
652	#endif // SPINLOCK
653	} // if
654
655	#ifdef __CFA_DEBUG__
656	__atomic_add_fetch( &allocFree, -size, __ATOMIC_SEQ_CST );
657	if ( traceHeap() ) {
658	enum { BufferSize = 64 };
659	char helpText[BufferSize];
660	int len = snprintf( helpText, sizeof(helpText), "Free( %p ) size:%zu\n", addr, size );
661	__cfaabi_dbg_bits_write( helpText, len );
662	} // if
663	#endif // __CFA_DEBUG__
664	} // doFree
665
666
667	size_t checkFree( HeapManager & manager ) with ( manager ) {
668	size_t total = 0;
669	#ifdef __STATISTICS__
670	__cfaabi_dbg_bits_acquire();
671	__cfaabi_dbg_bits_print_nolock( "\nBin lists (bin size : free blocks on list)\n" );
672	#endif // __STATISTICS__
673	for ( unsigned int i = 0; i < maxBucketsUsed; i += 1 ) {
674	size_t size = freeLists[i].blockSize;
675	#ifdef __STATISTICS__
676	unsigned int N = 0;
677	#endif // __STATISTICS__
678
679	#if defined( SPINLOCK )
680	for ( HeapManager.Storage * p = freeLists[i].freeList; p != 0; p = p->header.kind.real.next ) {
681	#else
682	for ( HeapManager.Storage * p = freeLists[i].freeList.top(); p != 0; p = p->header.kind.real.next.top ) {
683	#endif // SPINLOCK
684	total += size;
685	#ifdef __STATISTICS__
686	N += 1;
687	#endif // __STATISTICS__
688	} // for
689
690	#ifdef __STATISTICS__
691	__cfaabi_dbg_bits_print_nolock( "%7zu, %-7u ", size, N );
692	if ( (i + 1) % 8 == 0 ) __cfaabi_dbg_bits_print_nolock( "\n" );
693	#endif // __STATISTICS__
694	} // for
695	#ifdef __STATISTICS__
696	__cfaabi_dbg_bits_print_nolock( "\ntotal free blocks:%zu\n", total );
697	__cfaabi_dbg_bits_release();
698	#endif // __STATISTICS__
699	return (char )heapEnd - (char )heapBegin - total;
700	} // checkFree
701
702
703	static inline void * mallocNoStats( size_t size ) { // necessary for malloc statistics
704	//assert( heapManager.heapBegin != 0 );
705	if ( unlikely( heapManager.heapBegin == 0 ) ) heapManager{}; // called before memory_startup ?
706	void * area = doMalloc( size );
707	if ( unlikely( area == 0 ) ) errno = ENOMEM; // POSIX
708	return area;
709	} // mallocNoStats
710
711
712	static inline void * memalignNoStats( size_t alignment, size_t size ) { // necessary for malloc statistics
713	#ifdef __CFA_DEBUG__
714	checkAlign( alignment ); // check alignment
715	#endif // __CFA_DEBUG__
716
717	// if alignment <= default alignment, do normal malloc as two headers are unnecessary
718	if ( unlikely( alignment <= libAlign() ) ) return mallocNoStats( size );
719
720	// Allocate enough storage to guarantee an address on the alignment boundary, and sufficient space before it for
721	// administrative storage. NOTE, WHILE THERE ARE 2 HEADERS, THE FIRST ONE IS IMPLICITLY CREATED BY DOMALLOC.
722	// .-------------v-----------------v----------------v----------,
723	// \| Real Header \| ... padding ... \| Fake Header \| data ... \|
724	// `-------------^-----------------^-+--------------^----------'
725	// \|<--------------------------------' offset/align \|<-- alignment boundary
726
727	// subtract libAlign() because it is already the minimum alignment
728	// add sizeof(Storage) for fake header
729	// #comment TD : this is the only place that calls doMalloc without calling mallocNoStats, why ?
730	char * area = (char *)doMalloc( size + alignment - libAlign() + sizeof(HeapManager.Storage) );
731	if ( unlikely( area == 0 ) ) return area;
732
733	// address in the block of the "next" alignment address
734	char * user = (char *)libCeiling( (uintptr_t)(area + sizeof(HeapManager.Storage)), alignment );
735
736	// address of header from malloc
737	HeapManager.Storage.Header * realHeader = headerAddr( area );
738	// address of fake header * before* the alignment location
739	HeapManager.Storage.Header * fakeHeader = headerAddr( user );
740	// SKULLDUGGERY: insert the offset to the start of the actual storage block and remember alignment
741	fakeHeader->kind.fake.offset = (char )fakeHeader - (char )realHeader;
742	// SKULLDUGGERY: odd alignment imples fake header
743	fakeHeader->kind.fake.alignment = alignment \| 1;
744
745	return user;
746	} // memalignNoStats
747
748
749	extern "C" {
750	// The malloc() function allocates size bytes and returns a pointer to the allocated memory. The memory is not
751	// initialized. If size is 0, then malloc() returns either NULL, or a unique pointer value that can later be
752	// successfully passed to free().
753	void * malloc( size_t size ) {
754	#ifdef __STATISTICS__
755	__atomic_add_fetch( &malloc_calls, 1, __ATOMIC_SEQ_CST );
756	__atomic_add_fetch( &malloc_storage, size, __ATOMIC_SEQ_CST );
757	#endif // __STATISTICS__
758
759	return mallocNoStats( size );
760	} // malloc
761
762	// The calloc() function allocates memory for an array of nmemb elements of size bytes each and returns a pointer to
763	// the allocated memory. The memory is set to zero. If nmemb or size is 0, then calloc() returns either NULL, or a
764	// unique pointer value that can later be successfully passed to free().
765	void * calloc( size_t noOfElems, size_t elemSize ) {
766	size_t size = noOfElems * elemSize;
767	#ifdef __STATISTICS__
768	__atomic_add_fetch( &calloc_calls, 1, __ATOMIC_SEQ_CST );
769	__atomic_add_fetch( &calloc_storage, size, __ATOMIC_SEQ_CST );
770	#endif // __STATISTICS__
771
772	char * area = (char *)mallocNoStats( size );
773	if ( unlikely( area == 0 ) ) return 0;
774
775	HeapManager.Storage.Header * header;
776	HeapManager.FreeHeader * freeElem;
777	size_t asize, alignment;
778	bool mapped __attribute__(( unused )) = headers( "calloc", area, header, freeElem, asize, alignment );
779	#ifndef __CFA_DEBUG__
780	// Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
781	if ( ! mapped )
782	#endif // __CFA_DEBUG__
783	memset( area, '\0', asize - sizeof(HeapManager.Storage) ); // set to zeros
784
785	header->kind.real.blockSize \|= 2; // mark as zero filled
786	return area;
787	} // calloc
788
789	// #comment TD : Document this function
790	void * cmemalign( size_t alignment, size_t noOfElems, size_t elemSize ) {
791	size_t size = noOfElems * elemSize;
792	#ifdef __STATISTICS__
793	__atomic_add_fetch( &cmemalign_calls, 1, __ATOMIC_SEQ_CST );
794	__atomic_add_fetch( &cmemalign_storage, size, __ATOMIC_SEQ_CST );
795	#endif // __STATISTICS__
796
797	char * area = (char *)memalignNoStats( alignment, size );
798	if ( unlikely( area == 0 ) ) return 0;
799	HeapManager.Storage.Header * header;
800	HeapManager.FreeHeader * freeElem;
801	size_t asize;
802	bool mapped __attribute__(( unused )) = headers( "cmemalign", area, header, freeElem, asize, alignment );
803	#ifndef __CFA_DEBUG__
804	// Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
805	if ( ! mapped )
806	#endif // __CFA_DEBUG__
807	memset( area, '\0', asize - ( (char )area - (char )header ) ); // set to zeros
808	header->kind.real.blockSize \|= 2; // mark as zero filled
809
810	return area;
811	} // cmemalign
812
813	// The realloc() function changes the size of the memory block pointed to by ptr to size bytes. The contents will be
814	// unchanged in the range from the start of the region up to the minimum of the old and new sizes. If the new size
815	// is larger than the old size, the added memory will not be initialized. If ptr is NULL, then the call is
816	// equivalent to malloc(size), for all values of size; if size is equal to zero, and ptr is not NULL, then the call
817	// is equivalent to free(ptr). Unless ptr is NULL, it must have been returned by an earlier call to malloc(),
818	// calloc() or realloc(). If the area pointed to was moved, a free(ptr) is done.
819	void * realloc( void * addr, size_t size ) {
820	#ifdef __STATISTICS__
821	__atomic_add_fetch( &realloc_calls, 1, __ATOMIC_SEQ_CST );
822	#endif // __STATISTICS__
823
824	if ( unlikely( addr == 0 ) ) return mallocNoStats( size ); // special cases
825	if ( unlikely( size == 0 ) ) { free( addr ); return 0; }
826
827	HeapManager.Storage.Header * header;
828	HeapManager.FreeHeader * freeElem;
829	size_t asize, alignment = 0;
830	headers( "realloc", addr, header, freeElem, asize, alignment );
831
832	size_t usize = asize - ( (char )addr - (char )header ); // compute the amount of user storage in the block
833	if ( usize >= size ) { // already sufficient storage
834	// This case does not result in a new profiler entry because the previous one still exists and it must match with
835	// the free for this memory. Hence, this realloc does not appear in the profiler output.
836	return addr;
837	} // if
838
839	#ifdef __STATISTICS__
840	__atomic_add_fetch( &realloc_storage, size, __ATOMIC_SEQ_CST );
841	#endif // __STATISTICS__
842
843	void * area;
844	if ( unlikely( alignment != 0 ) ) { // previous request memalign?
845	area = memalign( alignment, size ); // create new area
846	} else {
847	area = mallocNoStats( size ); // create new area
848	} // if
849	if ( unlikely( area == 0 ) ) return 0;
850	if ( unlikely( header->kind.real.blockSize & 2 ) ) { // previous request zero fill (calloc/cmemalign) ?
851	assert( (header->kind.real.blockSize & 1) == 0 );
852	bool mapped __attribute__(( unused )) = headers( "realloc", area, header, freeElem, asize, alignment );
853	#ifndef __CFA_DEBUG__
854	// Mapped storage is zero filled, but in debug mode mapped memory is scrubbed in doMalloc, so it has to be reset to zero.
855	if ( ! mapped )
856	#endif // __CFA_DEBUG__
857	memset( (char )area + usize, '\0', asize - ( (char )area - (char *)header ) - usize ); // zero-fill back part
858	header->kind.real.blockSize \|= 2; // mark new request as zero fill
859	} // if
860	memcpy( area, addr, usize ); // copy bytes
861	free( addr );
862	return area;
863	} // realloc
864
865
866	// The obsolete function memalign() allocates size bytes and returns a pointer to the allocated memory. The memory
867	// address will be a multiple of alignment, which must be a power of two.
868	void * memalign( size_t alignment, size_t size ) {
869	#ifdef __STATISTICS__
870	__atomic_add_fetch( &memalign_calls, 1, __ATOMIC_SEQ_CST );
871	__atomic_add_fetch( &memalign_storage, size, __ATOMIC_SEQ_CST );
872	#endif // __STATISTICS__
873
874	void * area = memalignNoStats( alignment, size );
875
876	return area;
877	} // memalign
878
879	// The function aligned_alloc() is the same as memalign(), except for the added restriction that size should be a
880	// multiple of alignment.
881	void * aligned_alloc( size_t alignment, size_t size ) {
882	return memalign( alignment, size );
883	} // aligned_alloc
884
885
886	// The function posix_memalign() allocates size bytes and places the address of the allocated memory in *memptr. The
887	// address of the allocated memory will be a multiple of alignment, which must be a power of two and a multiple of
888	// sizeof(void *). If size is 0, then posix_memalign() returns either NULL, or a unique pointer value that can later
889	// be successfully passed to free(3).
890	int posix_memalign( void ** memptr, size_t alignment, size_t size ) {
891	if ( alignment < sizeof(void *) \|\| ! libPow2( alignment ) ) return EINVAL; // check alignment
892	* memptr = memalign( alignment, size );
893	if ( unlikely( * memptr == 0 ) ) return ENOMEM;
894	return 0;
895	} // posix_memalign
896
897	// The obsolete function valloc() allocates size bytes and returns a pointer to the allocated memory. The memory
898	// address will be a multiple of the page size. It is equivalent to memalign(sysconf(_SC_PAGESIZE),size).
899	void * valloc( size_t size ) {
900	return memalign( pageSize, size );
901	} // valloc
902
903
904	// The free() function frees the memory space pointed to by ptr, which must have been returned by a previous call to
905	// malloc(), calloc() or realloc(). Otherwise, or if free(ptr) has already been called before, undefined behavior
906	// occurs. If ptr is NULL, no operation is performed.
907	void free( void * addr ) {
908	#ifdef __STATISTICS__
909	__atomic_add_fetch( &free_calls, 1, __ATOMIC_SEQ_CST );
910	#endif // __STATISTICS__
911
912	// #comment TD : To decrease nesting I would but the special case in the
913	// else instead, plus it reads more naturally to have the
914	// short / normal case instead
915	if ( unlikely( addr == 0 ) ) { // special case
916	#ifdef __CFA_DEBUG__
917	if ( traceHeap() ) {
918	#define nullmsg "Free( 0x0 ) size:0\n"
919	// Do not debug print free( 0 ), as it can cause recursive entry from sprintf.
920	__cfaabi_dbg_bits_write( nullmsg, sizeof(nullmsg) - 1 );
921	} // if
922	#endif // __CFA_DEBUG__
923	return;
924	} // exit
925
926	doFree( addr );
927	} // free
928
929	// The mallopt() function adjusts parameters that control the behavior of the memory-allocation functions (see
930	// malloc(3)). The param argument specifies the parameter to be modified, and value specifies the new value for that
931	// parameter.
932	int mallopt( int option, int value ) {
933	choose( option ) {
934	case M_TOP_PAD:
935	if ( setHeapExpand( value ) ) fallthru default;
936	case M_MMAP_THRESHOLD:
937	if ( setMmapStart( value ) ) fallthru default;
938	default:
939	// #comment TD : 1 for unsopported feels wrong
940	return 1; // success, or unsupported
941	} // switch
942	return 0; // error
943	} // mallopt
944
945	// The malloc_trim() function attempts to release free memory at the top of the heap (by calling sbrk(2) with a
946	// suitable argument).
947	int malloc_trim( size_t ) {
948	return 0; // => impossible to release memory
949	} // malloc_trim
950
951	// The malloc_usable_size() function returns the number of usable bytes in the block pointed to by ptr, a pointer to
952	// a block of memory allocated by malloc(3) or a related function.
953	size_t malloc_usable_size( void * addr ) {
954	if ( unlikely( addr == 0 ) ) return 0; // null allocation has 0 size
955
956	HeapManager.Storage.Header * header;
957	HeapManager.FreeHeader * freeElem;
958	size_t size, alignment;
959
960	headers( "malloc_usable_size", addr, header, freeElem, size, alignment );
961	size_t usize = size - ( (char )addr - (char )header ); // compute the amount of user storage in the block
962	return usize;
963	} // malloc_usable_size
964
965
966	// The malloc_alignment() function returns the alignment of the allocation.
967	size_t malloc_alignment( void * addr ) {
968	if ( unlikely( addr == 0 ) ) return libAlign(); // minimum alignment
969	HeapManager.Storage.Header * header = (HeapManager.Storage.Header )( (char )addr - sizeof(HeapManager.Storage) );
970	if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
971	return header->kind.fake.alignment & -2; // remove flag from value
972	} else {
973	return libAlign (); // minimum alignment
974	} // if
975	} // malloc_alignment
976
977
978	// The malloc_zero_fill() function returns true if the allocation is zero filled, i.e., initially allocated by calloc().
979	bool malloc_zero_fill( void * addr ) {
980	if ( unlikely( addr == 0 ) ) return false; // null allocation is not zero fill
981
982	HeapManager.Storage.Header * header = (HeapManager.Storage.Header )( (char )addr - sizeof(HeapManager.Storage) );
983	if ( (header->kind.fake.alignment & 1) == 1 ) { // fake header ?
984	header = (HeapManager.Storage.Header )((char )header - header->kind.fake.offset);
985	} // if
986	return (header->kind.real.blockSize & 2) != 0; // zero filled (calloc/cmemalign) ?
987	} // malloc_zero_fill
988
989
990	// The malloc_stats() function prints (on default standard error) statistics about memory allocated by malloc(3) and
991	// related functions.
992	void malloc_stats( void ) {
993	#ifdef __STATISTICS__
994	printStats();
995	if ( checkFree() ) checkFree( heapManager );
996	#endif // __STATISTICS__
997	} // malloc_stats
998
999	// The malloc_stats_fd() function changes the file descripter where malloc_stats() writes the statistics.
1000	int malloc_stats_fd( int fd ) {
1001	#ifdef __STATISTICS__
1002	int temp = statfd;
1003	statfd = fd;
1004	return temp;
1005	#else
1006	return -1;
1007	#endif // __STATISTICS__
1008	} // malloc_stats_fd
1009
1010	// The malloc_info() function exports an XML string that describes the current state of the memory-allocation
1011	// implementation in the caller. The string is printed on the file stream stream. The exported string includes
1012	// information about all arenas (see malloc(3)).
1013	int malloc_info( int options, FILE * stream ) {
1014	return printStatsXML( stream );
1015	} // malloc_info
1016
1017
1018	// The malloc_get_state() function records the current state of all malloc(3) internal bookkeeping variables (but
1019	// not the actual contents of the heap or the state of malloc_hook(3) functions pointers). The state is recorded in
1020	// a system-dependent opaque data structure dynamically allocated via malloc(3), and a pointer to that data
1021	// structure is returned as the function result. (It is the caller's responsibility to free(3) this memory.)
1022	void * malloc_get_state( void ) {
1023	return 0; // unsupported
1024	} // malloc_get_state
1025
1026
1027	// The malloc_set_state() function restores the state of all malloc(3) internal bookkeeping variables to the values
1028	// recorded in the opaque data structure pointed to by state.
1029	int malloc_set_state( void * ptr ) {
1030	return 0; // unsupported
1031	} // malloc_set_state
1032	} // extern "C"
1033
1034
1035	// Local Variables: //
1036	// tab-width: 4 //
1037	// compile-command: "cfa -nodebug -O2 heap.cfa" //
1038	// End: //

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

Download in other formats:

Original Format