build/stlport/src/allocators.cpp

Sat, 03 Jan 2015 20:18:00 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Sat, 03 Jan 2015 20:18:00 +0100
branch
TOR_BUG_3246
changeset 7
129ffea94266
permissions
-rw-r--r--

Conditionally enable double key logic according to:
private browsing mode or privacy.thirdparty.isolate preference and
implement in GetCookieStringCommon and FindCookie where it counts...
With some reservations of how to convince FindCookie users to test
condition and pass a nullptr when disabling double key logic.

michael@0 1 /*
michael@0 2 *
michael@0 3 * Copyright (c) 1996,1997
michael@0 4 * Silicon Graphics Computer Systems, Inc.
michael@0 5 *
michael@0 6 * Copyright (c) 1997
michael@0 7 * Moscow Center for SPARC Technology
michael@0 8 *
michael@0 9 * Copyright (c) 1999
michael@0 10 * Boris Fomitchev
michael@0 11 *
michael@0 12 * This material is provided "as is", with absolutely no warranty expressed
michael@0 13 * or implied. Any use is at your own risk.
michael@0 14 *
michael@0 15 * Permission to use or copy this software for any purpose is hereby granted
michael@0 16 * without fee, provided the above notices are retained on all copies.
michael@0 17 * Permission to modify the code and to distribute modified code is granted,
michael@0 18 * provided the above notices are retained, and a notice that the code was
michael@0 19 * modified is included with the above copyright notice.
michael@0 20 *
michael@0 21 */
michael@0 22
michael@0 23 #include "stlport_prefix.h"
michael@0 24
michael@0 25 #include <memory>
michael@0 26
michael@0 27 #if defined (__GNUC__) && (defined (__CYGWIN__) || defined (__MINGW32__))
michael@0 28 # include <malloc.h>
michael@0 29 #endif
michael@0 30
michael@0 31 #if defined (_STLP_PTHREADS) && !defined (_STLP_NO_THREADS)
michael@0 32 # include <pthread_alloc>
michael@0 33 # include <cerrno>
michael@0 34 #endif
michael@0 35
michael@0 36 #include <stl/_threads.h>
michael@0 37
michael@0 38 #include "lock_free_slist.h"
michael@0 39
michael@0 40 #if defined (__WATCOMC__)
michael@0 41 # pragma warning 13 9
michael@0 42 # pragma warning 367 9
michael@0 43 # pragma warning 368 9
michael@0 44 #endif
michael@0 45
michael@0 46 #if defined (_STLP_SGI_THREADS)
michael@0 47 // We test whether threads are in use before locking.
michael@0 48 // Perhaps this should be moved into stl_threads.h, but that
michael@0 49 // probably makes it harder to avoid the procedure call when
michael@0 50 // it isn't needed.
michael@0 51 extern "C" {
michael@0 52 extern int __us_rsthread_malloc;
michael@0 53 }
michael@0 54 #endif
michael@0 55
michael@0 56 // Specialised debug form of new operator which does not provide "false"
michael@0 57 // memory leaks when run with debug CRT libraries.
michael@0 58 #if defined (_STLP_MSVC) && (_STLP_MSVC >= 1020 && defined (_STLP_DEBUG_ALLOC)) && !defined (_STLP_WCE)
michael@0 59 # include <crtdbg.h>
michael@0 60 inline char* __stlp_new_chunk(size_t __bytes) {
michael@0 61 void *__chunk = _STLP_CHECK_NULL_ALLOC(::operator new(__bytes, __FILE__, __LINE__));
michael@0 62 return __STATIC_CAST(char*, __chunk);
michael@0 63 }
michael@0 64 inline void __stlp_delete_chunck(void* __p) { ::operator delete(__p, __FILE__, __LINE__); }
michael@0 65 #else
michael@0 66 # ifdef _STLP_NODE_ALLOC_USE_MALLOC
michael@0 67 # include <cstdlib>
michael@0 68 inline char* __stlp_new_chunk(size_t __bytes) {
michael@0 69 // do not use _STLP_CHECK_NULL_ALLOC, this macro is dedicated to new operator.
michael@0 70 void *__chunk = _STLP_VENDOR_CSTD::malloc(__bytes);
michael@0 71 if (__chunk == 0) {
michael@0 72 _STLP_THROW_BAD_ALLOC;
michael@0 73 }
michael@0 74 return __STATIC_CAST(char*, __chunk);
michael@0 75 }
michael@0 76 inline void __stlp_delete_chunck(void* __p) { _STLP_VENDOR_CSTD::free(__p); }
michael@0 77 # else
michael@0 78 inline char* __stlp_new_chunk(size_t __bytes)
michael@0 79 { return __STATIC_CAST(char*, _STLP_STD::__stl_new(__bytes)); }
michael@0 80 inline void __stlp_delete_chunck(void* __p) { _STLP_STD::__stl_delete(__p); }
michael@0 81 # endif
michael@0 82 #endif
michael@0 83
michael@0 84 /* This is an additional atomic operations to the ones already defined in
michael@0 85 * stl/_threads.h, platform should try to support it to improve performance.
michael@0 86 * __add_atomic_t _STLP_ATOMIC_ADD(volatile __add_atomic_t* __target, __add_atomic_t __val) :
michael@0 87 * does *__target = *__target + __val and returns the old *__target value */
michael@0 88 typedef long __add_atomic_t;
michael@0 89 typedef unsigned long __uadd_atomic_t;
michael@0 90
michael@0 91 #if defined (__GNUC__) && defined (__i386__)
michael@0 92 inline long _STLP_atomic_add_gcc_x86(long volatile* p, long addend) {
michael@0 93 long result;
michael@0 94 __asm__ __volatile__
michael@0 95 ("lock; xaddl %1, %0;"
michael@0 96 :"=m" (*p), "=r" (result)
michael@0 97 :"m" (*p), "1" (addend)
michael@0 98 :"cc");
michael@0 99 return result + addend;
michael@0 100 }
michael@0 101 # define _STLP_ATOMIC_ADD(__dst, __val) _STLP_atomic_add_gcc_x86(__dst, __val)
michael@0 102 #elif defined (_STLP_WIN32THREADS)
michael@0 103 // The Win32 API function InterlockedExchangeAdd is not available on Windows 95.
michael@0 104 # if !defined (_STLP_WIN95_LIKE)
michael@0 105 # if defined (_STLP_NEW_PLATFORM_SDK)
michael@0 106 # define _STLP_ATOMIC_ADD(__dst, __val) InterlockedExchangeAdd(__dst, __val)
michael@0 107 # else
michael@0 108 # define _STLP_ATOMIC_ADD(__dst, __val) InterlockedExchangeAdd(__CONST_CAST(__add_atomic_t*, __dst), __val)
michael@0 109 # endif
michael@0 110 # endif
michael@0 111 #endif
michael@0 112
michael@0 113 #if defined (__OS400__)
michael@0 114 // dums 02/05/2007: is it really necessary ?
michael@0 115 enum { _ALIGN = 16, _ALIGN_SHIFT = 4 };
michael@0 116 #else
michael@0 117 enum { _ALIGN = 2 * sizeof(void*), _ALIGN_SHIFT = 2 + sizeof(void*) / 4 };
michael@0 118 #endif
michael@0 119
michael@0 120 #define _S_FREELIST_INDEX(__bytes) ((__bytes - size_t(1)) >> (int)_ALIGN_SHIFT)
michael@0 121
michael@0 122 _STLP_BEGIN_NAMESPACE
michael@0 123
michael@0 124 // malloc_alloc out-of-memory handling
michael@0 125 static __oom_handler_type __oom_handler = __STATIC_CAST(__oom_handler_type, 0);
michael@0 126
michael@0 127 #ifdef _STLP_THREADS
michael@0 128 _STLP_mutex __oom_handler_lock;
michael@0 129 #endif
michael@0 130
michael@0 131 void* _STLP_CALL __malloc_alloc::allocate(size_t __n)
michael@0 132 {
michael@0 133 void *__result = malloc(__n);
michael@0 134 if ( 0 == __result ) {
michael@0 135 __oom_handler_type __my_malloc_handler;
michael@0 136
michael@0 137 for (;;) {
michael@0 138 {
michael@0 139 #ifdef _STLP_THREADS
michael@0 140 _STLP_auto_lock _l( __oom_handler_lock );
michael@0 141 #endif
michael@0 142 __my_malloc_handler = __oom_handler;
michael@0 143 }
michael@0 144 if ( 0 == __my_malloc_handler) {
michael@0 145 _STLP_THROW_BAD_ALLOC;
michael@0 146 }
michael@0 147 (*__my_malloc_handler)();
michael@0 148 __result = malloc(__n);
michael@0 149 if ( __result )
michael@0 150 return __result;
michael@0 151 }
michael@0 152 }
michael@0 153 return __result;
michael@0 154 }
michael@0 155
michael@0 156 __oom_handler_type _STLP_CALL __malloc_alloc::set_malloc_handler(__oom_handler_type __f)
michael@0 157 {
michael@0 158 #ifdef _STLP_THREADS
michael@0 159 _STLP_auto_lock _l( __oom_handler_lock );
michael@0 160 #endif
michael@0 161 __oom_handler_type __old = __oom_handler;
michael@0 162 __oom_handler = __f;
michael@0 163 return __old;
michael@0 164 }
michael@0 165
michael@0 166 // *******************************************************
michael@0 167 // Default node allocator.
michael@0 168 // With a reasonable compiler, this should be roughly as fast as the
michael@0 169 // original STL class-specific allocators, but with less fragmentation.
michael@0 170 //
michael@0 171 // Important implementation properties:
michael@0 172 // 1. If the client request an object of size > _MAX_BYTES, the resulting
michael@0 173 // object will be obtained directly from malloc.
michael@0 174 // 2. In all other cases, we allocate an object of size exactly
michael@0 175 // _S_round_up(requested_size). Thus the client has enough size
michael@0 176 // information that we can return the object to the proper free list
michael@0 177 // without permanently losing part of the object.
michael@0 178 //
michael@0 179
michael@0 180 #define _STLP_NFREELISTS 16
michael@0 181
michael@0 182 #if defined (_STLP_LEAKS_PEDANTIC) && defined (_STLP_USE_DYNAMIC_LIB)
michael@0 183 /*
michael@0 184 * We can only do cleanup of the node allocator memory pool if we are
michael@0 185 * sure that the STLport library is used as a shared one as it guaranties
michael@0 186 * the unicity of the node allocator instance. Without that guaranty node
michael@0 187 * allocator instances might exchange memory blocks making the implementation
michael@0 188 * of a cleaning process much more complicated.
michael@0 189 */
michael@0 190 # define _STLP_DO_CLEAN_NODE_ALLOC
michael@0 191 #endif
michael@0 192
michael@0 193 /* When STLport is used without multi threaded safety we use the node allocator
michael@0 194 * implementation with locks as locks becomes no-op. The lock free implementation
michael@0 195 * always use system specific atomic operations which are slower than 'normal'
michael@0 196 * ones.
michael@0 197 */
michael@0 198 #if defined (_STLP_THREADS) && \
michael@0 199 defined (_STLP_HAS_ATOMIC_FREELIST) && defined (_STLP_ATOMIC_ADD)
michael@0 200 /*
michael@0 201 * We have an implementation of the atomic freelist (_STLP_atomic_freelist)
michael@0 202 * for this architecture and compiler. That means we can use the non-blocking
michael@0 203 * implementation of the node-allocation engine.*/
michael@0 204 # define _STLP_USE_LOCK_FREE_IMPLEMENTATION
michael@0 205 #endif
michael@0 206
michael@0 207 #if !defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 208 # if defined (_STLP_THREADS)
michael@0 209
michael@0 210 class _Node_Alloc_Lock {
michael@0 211 static _STLP_STATIC_MUTEX& _S_Mutex() {
michael@0 212 static _STLP_STATIC_MUTEX mutex _STLP_MUTEX_INITIALIZER;
michael@0 213 return mutex;
michael@0 214 }
michael@0 215 public:
michael@0 216 _Node_Alloc_Lock() {
michael@0 217 # if defined (_STLP_SGI_THREADS)
michael@0 218 if (__us_rsthread_malloc)
michael@0 219 # endif
michael@0 220 _S_Mutex()._M_acquire_lock();
michael@0 221 }
michael@0 222
michael@0 223 ~_Node_Alloc_Lock() {
michael@0 224 # if defined (_STLP_SGI_THREADS)
michael@0 225 if (__us_rsthread_malloc)
michael@0 226 # endif
michael@0 227 _S_Mutex()._M_release_lock();
michael@0 228 }
michael@0 229 };
michael@0 230
michael@0 231 # else
michael@0 232
michael@0 233 class _Node_Alloc_Lock {
michael@0 234 public:
michael@0 235 _Node_Alloc_Lock() { }
michael@0 236 ~_Node_Alloc_Lock() { }
michael@0 237 };
michael@0 238
michael@0 239 # endif
michael@0 240
michael@0 241 struct _Node_alloc_obj {
michael@0 242 _Node_alloc_obj * _M_next;
michael@0 243 };
michael@0 244 #endif
michael@0 245
michael@0 246 class __node_alloc_impl {
michael@0 247 static inline size_t _STLP_CALL _S_round_up(size_t __bytes)
michael@0 248 { return (((__bytes) + (size_t)_ALIGN-1) & ~((size_t)_ALIGN - 1)); }
michael@0 249
michael@0 250 #if defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 251 typedef _STLP_atomic_freelist::item _Obj;
michael@0 252 typedef _STLP_atomic_freelist _Freelist;
michael@0 253 typedef _STLP_atomic_freelist _ChunkList;
michael@0 254
michael@0 255 // Header of blocks of memory that have been allocated as part of
michael@0 256 // a larger chunk but have not yet been chopped up into nodes.
michael@0 257 struct _FreeBlockHeader : public _STLP_atomic_freelist::item {
michael@0 258 char* _M_end; // pointer to end of free memory
michael@0 259 };
michael@0 260 #else
michael@0 261 typedef _Node_alloc_obj _Obj;
michael@0 262 typedef _Obj* _STLP_VOLATILE _Freelist;
michael@0 263 typedef _Obj* _ChunkList;
michael@0 264 #endif
michael@0 265
michael@0 266 private:
michael@0 267 // Returns an object of size __n, and optionally adds to size __n free list.
michael@0 268 static _Obj* _S_refill(size_t __n);
michael@0 269 // Allocates a chunk for nobjs of size __p_size. nobjs may be reduced
michael@0 270 // if it is inconvenient to allocate the requested number.
michael@0 271 static char* _S_chunk_alloc(size_t __p_size, int& __nobjs);
michael@0 272 // Chunk allocation state.
michael@0 273 static _Freelist _S_free_list[_STLP_NFREELISTS];
michael@0 274 // Amount of total allocated memory
michael@0 275 #if defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 276 static _STLP_VOLATILE __add_atomic_t _S_heap_size;
michael@0 277 #else
michael@0 278 static size_t _S_heap_size;
michael@0 279 #endif
michael@0 280
michael@0 281 #if defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 282 // List of blocks of free memory
michael@0 283 static _STLP_atomic_freelist _S_free_mem_blocks;
michael@0 284 #else
michael@0 285 // Start of the current free memory buffer
michael@0 286 static char* _S_start_free;
michael@0 287 // End of the current free memory buffer
michael@0 288 static char* _S_end_free;
michael@0 289 #endif
michael@0 290
michael@0 291 #if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 292 public:
michael@0 293 // Methods to report alloc/dealloc calls to the counter system.
michael@0 294 # if defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 295 typedef _STLP_VOLATILE __stl_atomic_t _AllocCounter;
michael@0 296 # else
michael@0 297 typedef __stl_atomic_t _AllocCounter;
michael@0 298 # endif
michael@0 299 static _AllocCounter& _STLP_CALL _S_alloc_counter();
michael@0 300 static void _S_alloc_call();
michael@0 301 static void _S_dealloc_call();
michael@0 302
michael@0 303 private:
michael@0 304 // Free all the allocated chuncks of memory
michael@0 305 static void _S_chunk_dealloc();
michael@0 306 // Beginning of the linked list of allocated chunks of memory
michael@0 307 static _ChunkList _S_chunks;
michael@0 308 #endif /* _STLP_DO_CLEAN_NODE_ALLOC */
michael@0 309
michael@0 310 public:
michael@0 311 /* __n must be > 0 */
michael@0 312 static void* _M_allocate(size_t& __n);
michael@0 313 /* __p may not be 0 */
michael@0 314 static void _M_deallocate(void *__p, size_t __n);
michael@0 315 };
michael@0 316
michael@0 317 #if !defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 318 void* __node_alloc_impl::_M_allocate(size_t& __n) {
michael@0 319 __n = _S_round_up(__n);
michael@0 320 _Obj * _STLP_VOLATILE * __my_free_list = _S_free_list + _S_FREELIST_INDEX(__n);
michael@0 321 _Obj *__r;
michael@0 322
michael@0 323 // Acquire the lock here with a constructor call.
michael@0 324 // This ensures that it is released in exit or during stack
michael@0 325 // unwinding.
michael@0 326 _Node_Alloc_Lock __lock_instance;
michael@0 327
michael@0 328 if ( (__r = *__my_free_list) != 0 ) {
michael@0 329 *__my_free_list = __r->_M_next;
michael@0 330 } else {
michael@0 331 __r = _S_refill(__n);
michael@0 332 }
michael@0 333 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 334 _S_alloc_call();
michael@0 335 # endif
michael@0 336 // lock is released here
michael@0 337 return __r;
michael@0 338 }
michael@0 339
michael@0 340 void __node_alloc_impl::_M_deallocate(void *__p, size_t __n) {
michael@0 341 _Obj * _STLP_VOLATILE * __my_free_list = _S_free_list + _S_FREELIST_INDEX(__n);
michael@0 342 _Obj * __pobj = __STATIC_CAST(_Obj*, __p);
michael@0 343
michael@0 344 // acquire lock
michael@0 345 _Node_Alloc_Lock __lock_instance;
michael@0 346 __pobj->_M_next = *__my_free_list;
michael@0 347 *__my_free_list = __pobj;
michael@0 348
michael@0 349 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 350 _S_dealloc_call();
michael@0 351 # endif
michael@0 352 // lock is released here
michael@0 353 }
michael@0 354
michael@0 355 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 356 # define _STLP_OFFSET sizeof(_Obj)
michael@0 357 # else
michael@0 358 # define _STLP_OFFSET 0
michael@0 359 # endif
michael@0 360
michael@0 361 /* We allocate memory in large chunks in order to avoid fragmenting */
michael@0 362 /* the malloc heap too much. */
michael@0 363 /* We assume that size is properly aligned. */
michael@0 364 /* We hold the allocation lock. */
michael@0 365 char* __node_alloc_impl::_S_chunk_alloc(size_t _p_size, int& __nobjs) {
michael@0 366 char* __result;
michael@0 367 size_t __total_bytes = _p_size * __nobjs;
michael@0 368 size_t __bytes_left = _S_end_free - _S_start_free;
michael@0 369
michael@0 370 if (__bytes_left > 0) {
michael@0 371 if (__bytes_left >= __total_bytes) {
michael@0 372 __result = _S_start_free;
michael@0 373 _S_start_free += __total_bytes;
michael@0 374 return __result;
michael@0 375 }
michael@0 376
michael@0 377 if (__bytes_left >= _p_size) {
michael@0 378 __nobjs = (int)(__bytes_left / _p_size);
michael@0 379 __total_bytes = _p_size * __nobjs;
michael@0 380 __result = _S_start_free;
michael@0 381 _S_start_free += __total_bytes;
michael@0 382 return __result;
michael@0 383 }
michael@0 384
michael@0 385 // Try to make use of the left-over piece.
michael@0 386 _Obj* _STLP_VOLATILE* __my_free_list = _S_free_list + _S_FREELIST_INDEX(__bytes_left);
michael@0 387 __REINTERPRET_CAST(_Obj*, _S_start_free)->_M_next = *__my_free_list;
michael@0 388 *__my_free_list = __REINTERPRET_CAST(_Obj*, _S_start_free);
michael@0 389 _S_start_free = _S_end_free = 0;
michael@0 390 }
michael@0 391
michael@0 392 size_t __bytes_to_get = 2 * __total_bytes + _S_round_up(_S_heap_size) + _STLP_OFFSET;
michael@0 393
michael@0 394 _STLP_TRY {
michael@0 395 _S_start_free = __stlp_new_chunk(__bytes_to_get);
michael@0 396 }
michael@0 397 #if defined (_STLP_USE_EXCEPTIONS)
michael@0 398 catch (const _STLP_STD::bad_alloc&) {
michael@0 399 _Obj* _STLP_VOLATILE* __my_free_list;
michael@0 400 _Obj* __p;
michael@0 401 // Try to do with what we have. That can't hurt.
michael@0 402 // We do not try smaller requests, since that tends
michael@0 403 // to result in disaster on multi-process machines.
michael@0 404 for (size_t __i = _p_size; __i <= (size_t)_MAX_BYTES; __i += (size_t)_ALIGN) {
michael@0 405 __my_free_list = _S_free_list + _S_FREELIST_INDEX(__i);
michael@0 406 __p = *__my_free_list;
michael@0 407 if (0 != __p) {
michael@0 408 *__my_free_list = __p -> _M_next;
michael@0 409 _S_start_free = __REINTERPRET_CAST(char*, __p);
michael@0 410 _S_end_free = _S_start_free + __i;
michael@0 411 return _S_chunk_alloc(_p_size, __nobjs);
michael@0 412 // Any leftover piece will eventually make it to the
michael@0 413 // right free list.
michael@0 414 }
michael@0 415 }
michael@0 416 __bytes_to_get = __total_bytes + _STLP_OFFSET;
michael@0 417 _S_start_free = __stlp_new_chunk(__bytes_to_get);
michael@0 418 }
michael@0 419 #endif
michael@0 420
michael@0 421 _S_heap_size += __bytes_to_get >> 4;
michael@0 422 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 423 __REINTERPRET_CAST(_Obj*, _S_start_free)->_M_next = _S_chunks;
michael@0 424 _S_chunks = __REINTERPRET_CAST(_Obj*, _S_start_free);
michael@0 425 # endif
michael@0 426 _S_end_free = _S_start_free + __bytes_to_get;
michael@0 427 _S_start_free += _STLP_OFFSET;
michael@0 428 return _S_chunk_alloc(_p_size, __nobjs);
michael@0 429 }
michael@0 430
michael@0 431 /* Returns an object of size __n, and optionally adds to size __n free list.*/
michael@0 432 /* We assume that __n is properly aligned. */
michael@0 433 /* We hold the allocation lock. */
michael@0 434 _Node_alloc_obj* __node_alloc_impl::_S_refill(size_t __n) {
michael@0 435 int __nobjs = 20;
michael@0 436 char* __chunk = _S_chunk_alloc(__n, __nobjs);
michael@0 437
michael@0 438 if (1 == __nobjs) return __REINTERPRET_CAST(_Obj*, __chunk);
michael@0 439
michael@0 440 _Obj* _STLP_VOLATILE* __my_free_list = _S_free_list + _S_FREELIST_INDEX(__n);
michael@0 441 _Obj* __result;
michael@0 442 _Obj* __current_obj;
michael@0 443 _Obj* __next_obj;
michael@0 444
michael@0 445 /* Build free list in chunk */
michael@0 446 __result = __REINTERPRET_CAST(_Obj*, __chunk);
michael@0 447 *__my_free_list = __next_obj = __REINTERPRET_CAST(_Obj*, __chunk + __n);
michael@0 448 for (--__nobjs; --__nobjs; ) {
michael@0 449 __current_obj = __next_obj;
michael@0 450 __next_obj = __REINTERPRET_CAST(_Obj*, __REINTERPRET_CAST(char*, __next_obj) + __n);
michael@0 451 __current_obj->_M_next = __next_obj;
michael@0 452 }
michael@0 453 __next_obj->_M_next = 0;
michael@0 454 return __result;
michael@0 455 }
michael@0 456
michael@0 457 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 458 void __node_alloc_impl::_S_alloc_call()
michael@0 459 { ++_S_alloc_counter(); }
michael@0 460
michael@0 461 void __node_alloc_impl::_S_dealloc_call() {
michael@0 462 __stl_atomic_t &counter = _S_alloc_counter();
michael@0 463 if (--counter == 0)
michael@0 464 { _S_chunk_dealloc(); }
michael@0 465 }
michael@0 466
michael@0 467 /* We deallocate all the memory chunks */
michael@0 468 void __node_alloc_impl::_S_chunk_dealloc() {
michael@0 469 _Obj *__pcur = _S_chunks, *__pnext;
michael@0 470 while (__pcur != 0) {
michael@0 471 __pnext = __pcur->_M_next;
michael@0 472 __stlp_delete_chunck(__pcur);
michael@0 473 __pcur = __pnext;
michael@0 474 }
michael@0 475 _S_chunks = 0;
michael@0 476 _S_start_free = _S_end_free = 0;
michael@0 477 _S_heap_size = 0;
michael@0 478 memset(__REINTERPRET_CAST(char*, __CONST_CAST(_Obj**, &_S_free_list[0])), 0, _STLP_NFREELISTS * sizeof(_Obj*));
michael@0 479 }
michael@0 480 # endif
michael@0 481
michael@0 482 #else
michael@0 483
michael@0 484 void* __node_alloc_impl::_M_allocate(size_t& __n) {
michael@0 485 __n = _S_round_up(__n);
michael@0 486 _Obj* __r = _S_free_list[_S_FREELIST_INDEX(__n)].pop();
michael@0 487 if (__r == 0)
michael@0 488 { __r = _S_refill(__n); }
michael@0 489
michael@0 490 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 491 _S_alloc_call();
michael@0 492 # endif
michael@0 493 return __r;
michael@0 494 }
michael@0 495
michael@0 496 void __node_alloc_impl::_M_deallocate(void *__p, size_t __n) {
michael@0 497 _S_free_list[_S_FREELIST_INDEX(__n)].push(__STATIC_CAST(_Obj*, __p));
michael@0 498
michael@0 499 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 500 _S_dealloc_call();
michael@0 501 # endif
michael@0 502 }
michael@0 503
michael@0 504 /* Returns an object of size __n, and optionally adds additional ones to */
michael@0 505 /* freelist of objects of size __n. */
michael@0 506 /* We assume that __n is properly aligned. */
michael@0 507 __node_alloc_impl::_Obj* __node_alloc_impl::_S_refill(size_t __n) {
michael@0 508 int __nobjs = 20;
michael@0 509 char* __chunk = _S_chunk_alloc(__n, __nobjs);
michael@0 510
michael@0 511 if (__nobjs <= 1)
michael@0 512 return __REINTERPRET_CAST(_Obj*, __chunk);
michael@0 513
michael@0 514 // Push all new nodes (minus first one) onto freelist
michael@0 515 _Obj* __result = __REINTERPRET_CAST(_Obj*, __chunk);
michael@0 516 _Obj* __cur_item = __result;
michael@0 517 _Freelist* __my_freelist = _S_free_list + _S_FREELIST_INDEX(__n);
michael@0 518 for (--__nobjs; __nobjs != 0; --__nobjs) {
michael@0 519 __cur_item = __REINTERPRET_CAST(_Obj*, __REINTERPRET_CAST(char*, __cur_item) + __n);
michael@0 520 __my_freelist->push(__cur_item);
michael@0 521 }
michael@0 522 return __result;
michael@0 523 }
michael@0 524
michael@0 525 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 526 # define _STLP_OFFSET _ALIGN
michael@0 527 # else
michael@0 528 # define _STLP_OFFSET 0
michael@0 529 # endif
michael@0 530
michael@0 531 /* We allocate memory in large chunks in order to avoid fragmenting */
michael@0 532 /* the malloc heap too much. */
michael@0 533 /* We assume that size is properly aligned. */
michael@0 534 char* __node_alloc_impl::_S_chunk_alloc(size_t _p_size, int& __nobjs) {
michael@0 535 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 536 //We are going to add a small memory block to keep all the allocated blocks
michael@0 537 //address, we need to do so respecting the memory alignment. The following
michael@0 538 //static assert checks that the reserved block is big enough to store a pointer.
michael@0 539 _STLP_STATIC_ASSERT(sizeof(_Obj) <= _ALIGN)
michael@0 540 # endif
michael@0 541 char* __result = 0;
michael@0 542 __add_atomic_t __total_bytes = __STATIC_CAST(__add_atomic_t, _p_size) * __nobjs;
michael@0 543
michael@0 544 _FreeBlockHeader* __block = __STATIC_CAST(_FreeBlockHeader*, _S_free_mem_blocks.pop());
michael@0 545 if (__block != 0) {
michael@0 546 // We checked a block out and can now mess with it with impugnity.
michael@0 547 // We'll put the remainder back into the list if we're done with it below.
michael@0 548 char* __buf_start = __REINTERPRET_CAST(char*, __block);
michael@0 549 __add_atomic_t __bytes_left = __block->_M_end - __buf_start;
michael@0 550
michael@0 551 if ((__bytes_left < __total_bytes) && (__bytes_left >= __STATIC_CAST(__add_atomic_t, _p_size))) {
michael@0 552 // There's enough left for at least one object, but not as much as we wanted
michael@0 553 __result = __buf_start;
michael@0 554 __nobjs = (int)(__bytes_left/_p_size);
michael@0 555 __total_bytes = __STATIC_CAST(__add_atomic_t, _p_size) * __nobjs;
michael@0 556 __bytes_left -= __total_bytes;
michael@0 557 __buf_start += __total_bytes;
michael@0 558 }
michael@0 559 else if (__bytes_left >= __total_bytes) {
michael@0 560 // The block has enough left to satisfy all that was asked for
michael@0 561 __result = __buf_start;
michael@0 562 __bytes_left -= __total_bytes;
michael@0 563 __buf_start += __total_bytes;
michael@0 564 }
michael@0 565
michael@0 566 if (__bytes_left != 0) {
michael@0 567 // There is still some memory left over in block after we satisfied our request.
michael@0 568 if ((__result != 0) && (__bytes_left >= (__add_atomic_t)sizeof(_FreeBlockHeader))) {
michael@0 569 // We were able to allocate at least one object and there is still enough
michael@0 570 // left to put remainder back into list.
michael@0 571 _FreeBlockHeader* __newblock = __REINTERPRET_CAST(_FreeBlockHeader*, __buf_start);
michael@0 572 __newblock->_M_end = __block->_M_end;
michael@0 573 _S_free_mem_blocks.push(__newblock);
michael@0 574 }
michael@0 575 else {
michael@0 576 // We were not able to allocate enough for at least one object.
michael@0 577 // Shove into freelist of nearest (rounded-down!) size.
michael@0 578 size_t __rounded_down = _S_round_up(__bytes_left + 1) - (size_t)_ALIGN;
michael@0 579 if (__rounded_down > 0)
michael@0 580 _S_free_list[_S_FREELIST_INDEX(__rounded_down)].push((_Obj*)__buf_start);
michael@0 581 }
michael@0 582 }
michael@0 583 if (__result != 0)
michael@0 584 return __result;
michael@0 585 }
michael@0 586
michael@0 587 // We couldn't satisfy it from the list of free blocks, get new memory.
michael@0 588 __add_atomic_t __bytes_to_get = 2 * __total_bytes +
michael@0 589 __STATIC_CAST(__add_atomic_t,
michael@0 590 _S_round_up(__STATIC_CAST(__uadd_atomic_t, _STLP_ATOMIC_ADD(&_S_heap_size, 0)))) +
michael@0 591 _STLP_OFFSET;
michael@0 592 _STLP_TRY {
michael@0 593 __result = __stlp_new_chunk(__bytes_to_get);
michael@0 594 }
michael@0 595 #if defined (_STLP_USE_EXCEPTIONS)
michael@0 596 catch (const bad_alloc&) {
michael@0 597 // Allocation failed; try to canibalize from freelist of a larger object size.
michael@0 598 for (size_t __i = _p_size; __i <= (size_t)_MAX_BYTES; __i += (size_t)_ALIGN) {
michael@0 599 _Obj* __p = _S_free_list[_S_FREELIST_INDEX(__i)].pop();
michael@0 600 if (0 != __p) {
michael@0 601 if (__i < sizeof(_FreeBlockHeader)) {
michael@0 602 // Not enough to put into list of free blocks, divvy it up here.
michael@0 603 // Use as much as possible for this request and shove remainder into freelist.
michael@0 604 __nobjs = (int)(__i/_p_size);
michael@0 605 __total_bytes = __nobjs * __STATIC_CAST(__add_atomic_t, _p_size);
michael@0 606 size_t __bytes_left = __i - __total_bytes;
michael@0 607 size_t __rounded_down = _S_round_up(__bytes_left+1) - (size_t)_ALIGN;
michael@0 608 if (__rounded_down > 0) {
michael@0 609 _S_free_list[_S_FREELIST_INDEX(__rounded_down)].push(__REINTERPRET_CAST(_Obj*, __REINTERPRET_CAST(char*, __p) + __total_bytes));
michael@0 610 }
michael@0 611 return __REINTERPRET_CAST(char*, __p);
michael@0 612 }
michael@0 613 else {
michael@0 614 // Add node to list of available blocks and recursively allocate from it.
michael@0 615 _FreeBlockHeader* __newblock = (_FreeBlockHeader*)__p;
michael@0 616 __newblock->_M_end = __REINTERPRET_CAST(char*, __p) + __i;
michael@0 617 _S_free_mem_blocks.push(__newblock);
michael@0 618 return _S_chunk_alloc(_p_size, __nobjs);
michael@0 619 }
michael@0 620 }
michael@0 621 }
michael@0 622
michael@0 623 // We were not able to find something in a freelist, try to allocate a smaller amount.
michael@0 624 __bytes_to_get = __total_bytes + _STLP_OFFSET;
michael@0 625 __result = __stlp_new_chunk(__bytes_to_get);
michael@0 626
michael@0 627 // This should either throw an exception or remedy the situation.
michael@0 628 // Thus we assume it succeeded.
michael@0 629 }
michael@0 630 #endif
michael@0 631 // Alignment check
michael@0 632 _STLP_VERBOSE_ASSERT(((__REINTERPRET_CAST(size_t, __result) & __STATIC_CAST(size_t, _ALIGN - 1)) == 0),
michael@0 633 _StlMsg_DBA_DELETED_TWICE)
michael@0 634 _STLP_ATOMIC_ADD(&_S_heap_size, __bytes_to_get >> 4);
michael@0 635
michael@0 636 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 637 // We have to track the allocated memory chunks for release on exit.
michael@0 638 _S_chunks.push(__REINTERPRET_CAST(_Obj*, __result));
michael@0 639 __result += _ALIGN;
michael@0 640 __bytes_to_get -= _ALIGN;
michael@0 641 # endif
michael@0 642
michael@0 643 if (__bytes_to_get > __total_bytes) {
michael@0 644 // Push excess memory allocated in this chunk into list of free memory blocks
michael@0 645 _FreeBlockHeader* __freeblock = __REINTERPRET_CAST(_FreeBlockHeader*, __result + __total_bytes);
michael@0 646 __freeblock->_M_end = __result + __bytes_to_get;
michael@0 647 _S_free_mem_blocks.push(__freeblock);
michael@0 648 }
michael@0 649 return __result;
michael@0 650 }
michael@0 651
michael@0 652 # if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 653 void __node_alloc_impl::_S_alloc_call()
michael@0 654 { _STLP_ATOMIC_INCREMENT(&_S_alloc_counter()); }
michael@0 655
michael@0 656 void __node_alloc_impl::_S_dealloc_call() {
michael@0 657 _STLP_VOLATILE __stl_atomic_t *pcounter = &_S_alloc_counter();
michael@0 658 if (_STLP_ATOMIC_DECREMENT(pcounter) == 0)
michael@0 659 _S_chunk_dealloc();
michael@0 660 }
michael@0 661
michael@0 662 /* We deallocate all the memory chunks */
michael@0 663 void __node_alloc_impl::_S_chunk_dealloc() {
michael@0 664 // Note: The _Node_alloc_helper class ensures that this function
michael@0 665 // will only be called when the (shared) library is unloaded or the
michael@0 666 // process is shutdown. It's thus not possible that another thread
michael@0 667 // is currently trying to allocate a node (we're not thread-safe here).
michael@0 668 //
michael@0 669
michael@0 670 // Clear the free blocks and all freelistst. This makes sure that if
michael@0 671 // for some reason more memory is allocated again during shutdown
michael@0 672 // (it'd also be really nasty to leave references to deallocated memory).
michael@0 673 _S_free_mem_blocks.clear();
michael@0 674 _S_heap_size = 0;
michael@0 675
michael@0 676 for (size_t __i = 0; __i < _STLP_NFREELISTS; ++__i) {
michael@0 677 _S_free_list[__i].clear();
michael@0 678 }
michael@0 679
michael@0 680 // Detach list of chunks and free them all
michael@0 681 _Obj* __chunk = _S_chunks.clear();
michael@0 682 while (__chunk != 0) {
michael@0 683 _Obj* __next = __chunk->_M_next;
michael@0 684 __stlp_delete_chunck(__chunk);
michael@0 685 __chunk = __next;
michael@0 686 }
michael@0 687 }
michael@0 688 # endif
michael@0 689
michael@0 690 #endif
michael@0 691
michael@0 692 #if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 693 struct __node_alloc_cleaner {
michael@0 694 ~__node_alloc_cleaner()
michael@0 695 { __node_alloc_impl::_S_dealloc_call(); }
michael@0 696 };
michael@0 697
michael@0 698 # if defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 699 _STLP_VOLATILE __stl_atomic_t& _STLP_CALL
michael@0 700 # else
michael@0 701 __stl_atomic_t& _STLP_CALL
michael@0 702 # endif
michael@0 703 __node_alloc_impl::_S_alloc_counter() {
michael@0 704 static _AllocCounter _S_counter = 1;
michael@0 705 static __node_alloc_cleaner _S_node_alloc_cleaner;
michael@0 706 return _S_counter;
michael@0 707 }
michael@0 708 #endif
michael@0 709
michael@0 710 #if !defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 711 _Node_alloc_obj * _STLP_VOLATILE
michael@0 712 __node_alloc_impl::_S_free_list[_STLP_NFREELISTS]
michael@0 713 = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
michael@0 714 // The 16 zeros are necessary to make version 4.1 of the SunPro
michael@0 715 // compiler happy. Otherwise it appears to allocate too little
michael@0 716 // space for the array.
michael@0 717 #else
michael@0 718 _STLP_atomic_freelist __node_alloc_impl::_S_free_list[_STLP_NFREELISTS];
michael@0 719 _STLP_atomic_freelist __node_alloc_impl::_S_free_mem_blocks;
michael@0 720 #endif
michael@0 721
michael@0 722 #if !defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 723 char *__node_alloc_impl::_S_start_free = 0;
michael@0 724 char *__node_alloc_impl::_S_end_free = 0;
michael@0 725 #endif
michael@0 726
michael@0 727 #if defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 728 _STLP_VOLATILE __add_atomic_t
michael@0 729 #else
michael@0 730 size_t
michael@0 731 #endif
michael@0 732 __node_alloc_impl::_S_heap_size = 0;
michael@0 733
michael@0 734 #if defined (_STLP_DO_CLEAN_NODE_ALLOC)
michael@0 735 # if defined (_STLP_USE_LOCK_FREE_IMPLEMENTATION)
michael@0 736 _STLP_atomic_freelist __node_alloc_impl::_S_chunks;
michael@0 737 # else
michael@0 738 _Node_alloc_obj* __node_alloc_impl::_S_chunks = 0;
michael@0 739 # endif
michael@0 740 #endif
michael@0 741
michael@0 742 void * _STLP_CALL __node_alloc::_M_allocate(size_t& __n)
michael@0 743 { return __node_alloc_impl::_M_allocate(__n); }
michael@0 744
michael@0 745 void _STLP_CALL __node_alloc::_M_deallocate(void *__p, size_t __n)
michael@0 746 { __node_alloc_impl::_M_deallocate(__p, __n); }
michael@0 747
michael@0 748 #if defined (_STLP_PTHREADS) && !defined (_STLP_NO_THREADS)
michael@0 749
michael@0 750 # define _STLP_DATA_ALIGNMENT 8
michael@0 751
michael@0 752 _STLP_MOVE_TO_PRIV_NAMESPACE
michael@0 753
michael@0 754 // *******************************************************
michael@0 755 // __perthread_alloc implementation
michael@0 756 union _Pthread_alloc_obj {
michael@0 757 union _Pthread_alloc_obj * __free_list_link;
michael@0 758 char __client_data[_STLP_DATA_ALIGNMENT]; /* The client sees this. */
michael@0 759 };
michael@0 760
michael@0 761 // Pthread allocators don't appear to the client to have meaningful
michael@0 762 // instances. We do in fact need to associate some state with each
michael@0 763 // thread. That state is represented by _Pthread_alloc_per_thread_state.
michael@0 764
michael@0 765 struct _Pthread_alloc_per_thread_state {
michael@0 766 typedef _Pthread_alloc_obj __obj;
michael@0 767 enum { _S_NFREELISTS = _MAX_BYTES / _STLP_DATA_ALIGNMENT };
michael@0 768
michael@0 769 // Free list link for list of available per thread structures.
michael@0 770 // When one of these becomes available for reuse due to thread
michael@0 771 // termination, any objects in its free list remain associated
michael@0 772 // with it. The whole structure may then be used by a newly
michael@0 773 // created thread.
michael@0 774 _Pthread_alloc_per_thread_state() : __next(0)
michael@0 775 { memset((void *)__CONST_CAST(_Pthread_alloc_obj**, __free_list), 0, (size_t)_S_NFREELISTS * sizeof(__obj *)); }
michael@0 776 // Returns an object of size __n, and possibly adds to size n free list.
michael@0 777 void *_M_refill(size_t __n);
michael@0 778
michael@0 779 _Pthread_alloc_obj* volatile __free_list[_S_NFREELISTS];
michael@0 780 _Pthread_alloc_per_thread_state *__next;
michael@0 781 // this data member is only to be used by per_thread_allocator, which returns memory to the originating thread.
michael@0 782 _STLP_mutex _M_lock;
michael@0 783 };
michael@0 784
michael@0 785 // Pthread-specific allocator.
michael@0 786 class _Pthread_alloc_impl {
michael@0 787 public: // but only for internal use:
michael@0 788 typedef _Pthread_alloc_per_thread_state __state_type;
michael@0 789 typedef char value_type;
michael@0 790
michael@0 791 // Allocates a chunk for nobjs of size size. nobjs may be reduced
michael@0 792 // if it is inconvenient to allocate the requested number.
michael@0 793 static char *_S_chunk_alloc(size_t __size, size_t &__nobjs, __state_type*);
michael@0 794
michael@0 795 enum {_S_ALIGN = _STLP_DATA_ALIGNMENT};
michael@0 796
michael@0 797 static size_t _S_round_up(size_t __bytes)
michael@0 798 { return (((__bytes) + (int)_S_ALIGN - 1) & ~((int)_S_ALIGN - 1)); }
michael@0 799 static size_t _S_freelist_index(size_t __bytes)
michael@0 800 { return (((__bytes) + (int)_S_ALIGN - 1) / (int)_S_ALIGN - 1); }
michael@0 801
michael@0 802 private:
michael@0 803 // Chunk allocation state. And other shared state.
michael@0 804 // Protected by _S_chunk_allocator_lock.
michael@0 805 static _STLP_STATIC_MUTEX _S_chunk_allocator_lock;
michael@0 806 static char *_S_start_free;
michael@0 807 static char *_S_end_free;
michael@0 808 static size_t _S_heap_size;
michael@0 809 static __state_type *_S_free_per_thread_states;
michael@0 810 static pthread_key_t _S_key;
michael@0 811 static bool _S_key_initialized;
michael@0 812 // Pthread key under which per thread state is stored.
michael@0 813 // Allocator instances that are currently unclaimed by any thread.
michael@0 814 static void _S_destructor(void *instance);
michael@0 815 // Function to be called on thread exit to reclaim per thread
michael@0 816 // state.
michael@0 817 static __state_type *_S_new_per_thread_state();
michael@0 818 public:
michael@0 819 // Return a recycled or new per thread state.
michael@0 820 static __state_type *_S_get_per_thread_state();
michael@0 821 private:
michael@0 822 // ensure that the current thread has an associated
michael@0 823 // per thread state.
michael@0 824 class _M_lock;
michael@0 825 friend class _M_lock;
michael@0 826 class _M_lock {
michael@0 827 public:
michael@0 828 _M_lock () { _S_chunk_allocator_lock._M_acquire_lock(); }
michael@0 829 ~_M_lock () { _S_chunk_allocator_lock._M_release_lock(); }
michael@0 830 };
michael@0 831
michael@0 832 public:
michael@0 833
michael@0 834 /* n must be > 0 */
michael@0 835 static void * allocate(size_t& __n);
michael@0 836
michael@0 837 /* p may not be 0 */
michael@0 838 static void deallocate(void *__p, size_t __n);
michael@0 839
michael@0 840 // boris : versions for per_thread_allocator
michael@0 841 /* n must be > 0 */
michael@0 842 static void * allocate(size_t& __n, __state_type* __a);
michael@0 843
michael@0 844 /* p may not be 0 */
michael@0 845 static void deallocate(void *__p, size_t __n, __state_type* __a);
michael@0 846
michael@0 847 static void * reallocate(void *__p, size_t __old_sz, size_t& __new_sz);
michael@0 848 };
michael@0 849
michael@0 850 /* Returns an object of size n, and optionally adds to size n free list.*/
michael@0 851 /* We assume that n is properly aligned. */
michael@0 852 /* We hold the allocation lock. */
michael@0 853 void *_Pthread_alloc_per_thread_state::_M_refill(size_t __n) {
michael@0 854 typedef _Pthread_alloc_obj __obj;
michael@0 855 size_t __nobjs = 128;
michael@0 856 char * __chunk = _Pthread_alloc_impl::_S_chunk_alloc(__n, __nobjs, this);
michael@0 857 __obj * volatile * __my_free_list;
michael@0 858 __obj * __result;
michael@0 859 __obj * __current_obj, * __next_obj;
michael@0 860 size_t __i;
michael@0 861
michael@0 862 if (1 == __nobjs) {
michael@0 863 return __chunk;
michael@0 864 }
michael@0 865
michael@0 866 __my_free_list = __free_list + _Pthread_alloc_impl::_S_freelist_index(__n);
michael@0 867
michael@0 868 /* Build free list in chunk */
michael@0 869 __result = (__obj *)__chunk;
michael@0 870 *__my_free_list = __next_obj = (__obj *)(__chunk + __n);
michael@0 871 for (__i = 1; ; ++__i) {
michael@0 872 __current_obj = __next_obj;
michael@0 873 __next_obj = (__obj *)((char *)__next_obj + __n);
michael@0 874 if (__nobjs - 1 == __i) {
michael@0 875 __current_obj -> __free_list_link = 0;
michael@0 876 break;
michael@0 877 } else {
michael@0 878 __current_obj -> __free_list_link = __next_obj;
michael@0 879 }
michael@0 880 }
michael@0 881 return __result;
michael@0 882 }
michael@0 883
michael@0 884 void _Pthread_alloc_impl::_S_destructor(void *__instance) {
michael@0 885 _M_lock __lock_instance; // Need to acquire lock here.
michael@0 886 _Pthread_alloc_per_thread_state* __s = (_Pthread_alloc_per_thread_state*)__instance;
michael@0 887 __s -> __next = _S_free_per_thread_states;
michael@0 888 _S_free_per_thread_states = __s;
michael@0 889 }
michael@0 890
michael@0 891 _Pthread_alloc_per_thread_state* _Pthread_alloc_impl::_S_new_per_thread_state() {
michael@0 892 /* lock already held here. */
michael@0 893 if (0 != _S_free_per_thread_states) {
michael@0 894 _Pthread_alloc_per_thread_state *__result = _S_free_per_thread_states;
michael@0 895 _S_free_per_thread_states = _S_free_per_thread_states -> __next;
michael@0 896 return __result;
michael@0 897 }
michael@0 898 else {
michael@0 899 return new _Pthread_alloc_per_thread_state;
michael@0 900 }
michael@0 901 }
michael@0 902
michael@0 903 _Pthread_alloc_per_thread_state* _Pthread_alloc_impl::_S_get_per_thread_state() {
michael@0 904 int __ret_code;
michael@0 905 __state_type* __result;
michael@0 906
michael@0 907 if (_S_key_initialized && (__result = (__state_type*) pthread_getspecific(_S_key)))
michael@0 908 return __result;
michael@0 909
michael@0 910 /*REFERENCED*/
michael@0 911 _M_lock __lock_instance; // Need to acquire lock here.
michael@0 912 if (!_S_key_initialized) {
michael@0 913 if (pthread_key_create(&_S_key, _S_destructor)) {
michael@0 914 _STLP_THROW_BAD_ALLOC; // failed
michael@0 915 }
michael@0 916 _S_key_initialized = true;
michael@0 917 }
michael@0 918
michael@0 919 __result = _S_new_per_thread_state();
michael@0 920 __ret_code = pthread_setspecific(_S_key, __result);
michael@0 921 if (__ret_code) {
michael@0 922 if (__ret_code == ENOMEM) {
michael@0 923 _STLP_THROW_BAD_ALLOC;
michael@0 924 } else {
michael@0 925 // EINVAL
michael@0 926 _STLP_ABORT();
michael@0 927 }
michael@0 928 }
michael@0 929 return __result;
michael@0 930 }
michael@0 931
michael@0 932 /* We allocate memory in large chunks in order to avoid fragmenting */
michael@0 933 /* the malloc heap too much. */
michael@0 934 /* We assume that size is properly aligned. */
michael@0 935 char *_Pthread_alloc_impl::_S_chunk_alloc(size_t __p_size, size_t &__nobjs, _Pthread_alloc_per_thread_state *__a) {
michael@0 936 typedef _Pthread_alloc_obj __obj;
michael@0 937 {
michael@0 938 char * __result;
michael@0 939 size_t __total_bytes;
michael@0 940 size_t __bytes_left;
michael@0 941 /*REFERENCED*/
michael@0 942 _M_lock __lock_instance; // Acquire lock for this routine
michael@0 943
michael@0 944 __total_bytes = __p_size * __nobjs;
michael@0 945 __bytes_left = _S_end_free - _S_start_free;
michael@0 946 if (__bytes_left >= __total_bytes) {
michael@0 947 __result = _S_start_free;
michael@0 948 _S_start_free += __total_bytes;
michael@0 949 return __result;
michael@0 950 } else if (__bytes_left >= __p_size) {
michael@0 951 __nobjs = __bytes_left/__p_size;
michael@0 952 __total_bytes = __p_size * __nobjs;
michael@0 953 __result = _S_start_free;
michael@0 954 _S_start_free += __total_bytes;
michael@0 955 return __result;
michael@0 956 } else {
michael@0 957 size_t __bytes_to_get = 2 * __total_bytes + _S_round_up(_S_heap_size);
michael@0 958 // Try to make use of the left-over piece.
michael@0 959 if (__bytes_left > 0) {
michael@0 960 __obj * volatile * __my_free_list = __a->__free_list + _S_freelist_index(__bytes_left);
michael@0 961 ((__obj *)_S_start_free) -> __free_list_link = *__my_free_list;
michael@0 962 *__my_free_list = (__obj *)_S_start_free;
michael@0 963 }
michael@0 964 # ifdef _SGI_SOURCE
michael@0 965 // Try to get memory that's aligned on something like a
michael@0 966 // cache line boundary, so as to avoid parceling out
michael@0 967 // parts of the same line to different threads and thus
michael@0 968 // possibly different processors.
michael@0 969 {
michael@0 970 const int __cache_line_size = 128; // probable upper bound
michael@0 971 __bytes_to_get &= ~(__cache_line_size-1);
michael@0 972 _S_start_free = (char *)memalign(__cache_line_size, __bytes_to_get);
michael@0 973 if (0 == _S_start_free) {
michael@0 974 _S_start_free = (char *)__malloc_alloc::allocate(__bytes_to_get);
michael@0 975 }
michael@0 976 }
michael@0 977 # else /* !SGI_SOURCE */
michael@0 978 _S_start_free = (char *)__malloc_alloc::allocate(__bytes_to_get);
michael@0 979 # endif
michael@0 980 _S_heap_size += __bytes_to_get >> 4;
michael@0 981 _S_end_free = _S_start_free + __bytes_to_get;
michael@0 982 }
michael@0 983 }
michael@0 984 // lock is released here
michael@0 985 return _S_chunk_alloc(__p_size, __nobjs, __a);
michael@0 986 }
michael@0 987
michael@0 988
michael@0 989 /* n must be > 0 */
michael@0 990 void *_Pthread_alloc_impl::allocate(size_t& __n) {
michael@0 991 typedef _Pthread_alloc_obj __obj;
michael@0 992 __obj * volatile * __my_free_list;
michael@0 993 __obj * __result;
michael@0 994 __state_type* __a;
michael@0 995
michael@0 996 if (__n > _MAX_BYTES) {
michael@0 997 return __malloc_alloc::allocate(__n);
michael@0 998 }
michael@0 999
michael@0 1000 __n = _S_round_up(__n);
michael@0 1001 __a = _S_get_per_thread_state();
michael@0 1002
michael@0 1003 __my_free_list = __a->__free_list + _S_freelist_index(__n);
michael@0 1004 __result = *__my_free_list;
michael@0 1005 if (__result == 0) {
michael@0 1006 void *__r = __a->_M_refill(__n);
michael@0 1007 return __r;
michael@0 1008 }
michael@0 1009 *__my_free_list = __result->__free_list_link;
michael@0 1010 return __result;
michael@0 1011 };
michael@0 1012
michael@0 1013 /* p may not be 0 */
michael@0 1014 void _Pthread_alloc_impl::deallocate(void *__p, size_t __n) {
michael@0 1015 typedef _Pthread_alloc_obj __obj;
michael@0 1016 __obj *__q = (__obj *)__p;
michael@0 1017 __obj * volatile * __my_free_list;
michael@0 1018 __state_type* __a;
michael@0 1019
michael@0 1020 if (__n > _MAX_BYTES) {
michael@0 1021 __malloc_alloc::deallocate(__p, __n);
michael@0 1022 return;
michael@0 1023 }
michael@0 1024
michael@0 1025 __a = _S_get_per_thread_state();
michael@0 1026
michael@0 1027 __my_free_list = __a->__free_list + _S_freelist_index(__n);
michael@0 1028 __q -> __free_list_link = *__my_free_list;
michael@0 1029 *__my_free_list = __q;
michael@0 1030 }
michael@0 1031
michael@0 1032 // boris : versions for per_thread_allocator
michael@0 1033 /* n must be > 0 */
michael@0 1034 void *_Pthread_alloc_impl::allocate(size_t& __n, __state_type* __a) {
michael@0 1035 typedef _Pthread_alloc_obj __obj;
michael@0 1036 __obj * volatile * __my_free_list;
michael@0 1037 __obj * __result;
michael@0 1038
michael@0 1039 if (__n > _MAX_BYTES) {
michael@0 1040 return __malloc_alloc::allocate(__n);
michael@0 1041 }
michael@0 1042 __n = _S_round_up(__n);
michael@0 1043
michael@0 1044 // boris : here, we have to lock per thread state, as we may be getting memory from
michael@0 1045 // different thread pool.
michael@0 1046 _STLP_auto_lock __lock(__a->_M_lock);
michael@0 1047
michael@0 1048 __my_free_list = __a->__free_list + _S_freelist_index(__n);
michael@0 1049 __result = *__my_free_list;
michael@0 1050 if (__result == 0) {
michael@0 1051 void *__r = __a->_M_refill(__n);
michael@0 1052 return __r;
michael@0 1053 }
michael@0 1054 *__my_free_list = __result->__free_list_link;
michael@0 1055 return __result;
michael@0 1056 };
michael@0 1057
michael@0 1058 /* p may not be 0 */
michael@0 1059 void _Pthread_alloc_impl::deallocate(void *__p, size_t __n, __state_type* __a) {
michael@0 1060 typedef _Pthread_alloc_obj __obj;
michael@0 1061 __obj *__q = (__obj *)__p;
michael@0 1062 __obj * volatile * __my_free_list;
michael@0 1063
michael@0 1064 if (__n > _MAX_BYTES) {
michael@0 1065 __malloc_alloc::deallocate(__p, __n);
michael@0 1066 return;
michael@0 1067 }
michael@0 1068
michael@0 1069 // boris : here, we have to lock per thread state, as we may be returning memory from
michael@0 1070 // different thread.
michael@0 1071 _STLP_auto_lock __lock(__a->_M_lock);
michael@0 1072
michael@0 1073 __my_free_list = __a->__free_list + _S_freelist_index(__n);
michael@0 1074 __q -> __free_list_link = *__my_free_list;
michael@0 1075 *__my_free_list = __q;
michael@0 1076 }
michael@0 1077
michael@0 1078 void *_Pthread_alloc_impl::reallocate(void *__p, size_t __old_sz, size_t& __new_sz) {
michael@0 1079 void * __result;
michael@0 1080 size_t __copy_sz;
michael@0 1081
michael@0 1082 if (__old_sz > _MAX_BYTES && __new_sz > _MAX_BYTES) {
michael@0 1083 return realloc(__p, __new_sz);
michael@0 1084 }
michael@0 1085
michael@0 1086 if (_S_round_up(__old_sz) == _S_round_up(__new_sz)) return __p;
michael@0 1087 __result = allocate(__new_sz);
michael@0 1088 __copy_sz = __new_sz > __old_sz? __old_sz : __new_sz;
michael@0 1089 memcpy(__result, __p, __copy_sz);
michael@0 1090 deallocate(__p, __old_sz);
michael@0 1091 return __result;
michael@0 1092 }
michael@0 1093
michael@0 1094 _Pthread_alloc_per_thread_state* _Pthread_alloc_impl::_S_free_per_thread_states = 0;
michael@0 1095 pthread_key_t _Pthread_alloc_impl::_S_key = 0;
michael@0 1096 _STLP_STATIC_MUTEX _Pthread_alloc_impl::_S_chunk_allocator_lock _STLP_MUTEX_INITIALIZER;
michael@0 1097 bool _Pthread_alloc_impl::_S_key_initialized = false;
michael@0 1098 char *_Pthread_alloc_impl::_S_start_free = 0;
michael@0 1099 char *_Pthread_alloc_impl::_S_end_free = 0;
michael@0 1100 size_t _Pthread_alloc_impl::_S_heap_size = 0;
michael@0 1101
michael@0 1102 void * _STLP_CALL _Pthread_alloc::allocate(size_t& __n)
michael@0 1103 { return _Pthread_alloc_impl::allocate(__n); }
michael@0 1104 void _STLP_CALL _Pthread_alloc::deallocate(void *__p, size_t __n)
michael@0 1105 { _Pthread_alloc_impl::deallocate(__p, __n); }
michael@0 1106 void * _STLP_CALL _Pthread_alloc::allocate(size_t& __n, __state_type* __a)
michael@0 1107 { return _Pthread_alloc_impl::allocate(__n, __a); }
michael@0 1108 void _STLP_CALL _Pthread_alloc::deallocate(void *__p, size_t __n, __state_type* __a)
michael@0 1109 { _Pthread_alloc_impl::deallocate(__p, __n, __a); }
michael@0 1110 void * _STLP_CALL _Pthread_alloc::reallocate(void *__p, size_t __old_sz, size_t& __new_sz)
michael@0 1111 { return _Pthread_alloc_impl::reallocate(__p, __old_sz, __new_sz); }
michael@0 1112 _Pthread_alloc_per_thread_state* _STLP_CALL _Pthread_alloc::_S_get_per_thread_state()
michael@0 1113 { return _Pthread_alloc_impl::_S_get_per_thread_state(); }
michael@0 1114
michael@0 1115 _STLP_MOVE_TO_STD_NAMESPACE
michael@0 1116
michael@0 1117 #endif
michael@0 1118
michael@0 1119 _STLP_END_NAMESPACE
michael@0 1120
michael@0 1121 #undef _S_FREELIST_INDEX

mercurial