1 /* 2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "classfile/altHashing.hpp" 27 #include "classfile/javaClasses.hpp" 28 #include "classfile/symbolTable.hpp" 29 #include "classfile/systemDictionary.hpp" 30 #include "gc_interface/collectedHeap.inline.hpp" 31 #include "memory/allocation.inline.hpp" 32 #include "memory/filemap.hpp" 33 #include "memory/gcLocker.inline.hpp" 34 #include "oops/oop.inline.hpp" 35 #include "oops/oop.inline2.hpp" 36 #include "runtime/atomic.inline.hpp" 37 #include "runtime/mutexLocker.hpp" 38 #include "utilities/hashtable.inline.hpp" 39 #if INCLUDE_ALL_GCS 40 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp" 41 #include "gc_implementation/g1/g1StringDedup.hpp" 42 #endif 43 44 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC 45 46 // -------------------------------------------------------------------------- 47 48 // the number of buckets a thread claims 49 const int ClaimChunkSize = 32; 50 51 SymbolTable* SymbolTable::_the_table = NULL; 52 // Static arena for symbols that are not deallocated 53 Arena* SymbolTable::_arena = NULL; 54 bool SymbolTable::_needs_rehashing = false; 55 56 Symbol* SymbolTable::allocate_symbol(const u1* name, int len, bool c_heap, TRAPS) { 57 assert (len <= Symbol::max_length(), "should be checked by caller"); 58 59 Symbol* sym; 60 61 if (DumpSharedSpaces) { 62 // Allocate all symbols to CLD shared metaspace 63 sym = new (len, ClassLoaderData::the_null_class_loader_data(), THREAD) Symbol(name, len, -1); 64 } else if (c_heap) { 65 // refcount starts as 1 66 sym = new (len, THREAD) Symbol(name, len, 1); 67 assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted"); 68 } else { 69 // Allocate to global arena 70 sym = new (len, arena(), THREAD) Symbol(name, len, -1); 71 } 72 return sym; 73 } 74 75 void SymbolTable::initialize_symbols(int arena_alloc_size) { 76 // Initialize the arena for global symbols, size passed in depends on CDS. 77 if (arena_alloc_size == 0) { 78 _arena = new (mtSymbol) Arena(mtSymbol); 79 } else { 80 _arena = new (mtSymbol) Arena(mtSymbol, arena_alloc_size); 81 } 82 } 83 84 // Call function for all symbols in the symbol table. 85 void SymbolTable::symbols_do(SymbolClosure *cl) { 86 const int n = the_table()->table_size(); 87 for (int i = 0; i < n; i++) { 88 for (HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i); 89 p != NULL; 90 p = p->next()) { 91 cl->do_symbol(p->literal_addr()); 92 } 93 } 94 } 95 96 int SymbolTable::_symbols_removed = 0; 97 int SymbolTable::_symbols_counted = 0; 98 volatile int SymbolTable::_parallel_claimed_idx = 0; 99 100 void SymbolTable::buckets_unlink(int start_idx, int end_idx, int* processed, int* removed, size_t* memory_total) { 101 for (int i = start_idx; i < end_idx; ++i) { 102 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i); 103 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i); 104 while (entry != NULL) { 105 // Shared entries are normally at the end of the bucket and if we run into 106 // a shared entry, then there is nothing more to remove. However, if we 107 // have rehashed the table, then the shared entries are no longer at the 108 // end of the bucket. 109 if (entry->is_shared() && !use_alternate_hashcode()) { 110 break; 111 } 112 Symbol* s = entry->literal(); 113 (*memory_total) += s->size(); 114 (*processed)++; 115 assert(s != NULL, "just checking"); 116 // If reference count is zero, remove. 117 if (s->refcount() == 0) { 118 assert(!entry->is_shared(), "shared entries should be kept live"); 119 delete s; 120 (*removed)++; 121 *p = entry->next(); 122 the_table()->free_entry(entry); 123 } else { 124 p = entry->next_addr(); 125 } 126 // get next entry 127 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p); 128 } 129 } 130 } 131 132 // Remove unreferenced symbols from the symbol table 133 // This is done late during GC. 134 void SymbolTable::unlink(int* processed, int* removed) { 135 size_t memory_total = 0; 136 buckets_unlink(0, the_table()->table_size(), processed, removed, &memory_total); 137 _symbols_removed += *removed; 138 _symbols_counted += *processed; 139 // Exclude printing for normal PrintGCDetails because people parse 140 // this output. 141 if (PrintGCDetails && Verbose && WizardMode) { 142 gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", *processed, 143 (memory_total*HeapWordSize)/1024); 144 } 145 } 146 147 void SymbolTable::possibly_parallel_unlink(int* processed, int* removed) { 148 const int limit = the_table()->table_size(); 149 150 size_t memory_total = 0; 151 152 for (;;) { 153 // Grab next set of buckets to scan 154 int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize; 155 if (start_idx >= limit) { 156 // End of table 157 break; 158 } 159 160 int end_idx = MIN2(limit, start_idx + ClaimChunkSize); 161 buckets_unlink(start_idx, end_idx, processed, removed, &memory_total); 162 } 163 Atomic::add(*processed, &_symbols_counted); 164 Atomic::add(*removed, &_symbols_removed); 165 // Exclude printing for normal PrintGCDetails because people parse 166 // this output. 167 if (PrintGCDetails && Verbose && WizardMode) { 168 gclog_or_tty->print(" [Symbols: scanned=%d removed=%d size=" SIZE_FORMAT "K] ", *processed, *removed, 169 (memory_total*HeapWordSize)/1024); 170 } 171 } 172 173 // Create a new table and using alternate hash code, populate the new table 174 // with the existing strings. Set flag to use the alternate hash code afterwards. 175 void SymbolTable::rehash_table() { 176 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint"); 177 // This should never happen with -Xshare:dump but it might in testing mode. 178 if (DumpSharedSpaces) return; 179 // Create a new symbol table 180 SymbolTable* new_table = new SymbolTable(); 181 182 the_table()->move_to(new_table); 183 184 // Delete the table and buckets (entries are reused in new table). 185 delete _the_table; 186 // Don't check if we need rehashing until the table gets unbalanced again. 187 // Then rehash with a new global seed. 188 _needs_rehashing = false; 189 _the_table = new_table; 190 } 191 192 // Lookup a symbol in a bucket. 193 194 Symbol* SymbolTable::lookup(int index, const char* name, 195 int len, unsigned int hash) { 196 int count = 0; 197 for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) { 198 count++; // count all entries in this bucket, not just ones with same hash 199 if (e->hash() == hash) { 200 Symbol* sym = e->literal(); 201 if (sym->equals(name, len)) { 202 // something is referencing this symbol now. 203 sym->increment_refcount(); 204 return sym; 205 } 206 } 207 } 208 // If the bucket size is too deep check if this hash code is insufficient. 209 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) { 210 _needs_rehashing = check_rehash_table(count); 211 } 212 return NULL; 213 } 214 215 // Pick hashing algorithm. 216 unsigned int SymbolTable::hash_symbol(const char* s, int len) { 217 return use_alternate_hashcode() ? 218 AltHashing::murmur3_32(seed(), (const jbyte*)s, len) : 219 java_lang_String::hash_code(s, len); 220 } 221 222 223 // We take care not to be blocking while holding the 224 // SymbolTable_lock. Otherwise, the system might deadlock, since the 225 // symboltable is used during compilation (VM_thread) The lock free 226 // synchronization is simplified by the fact that we do not delete 227 // entries in the symbol table during normal execution (only during 228 // safepoints). 229 230 Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) { 231 unsigned int hashValue = hash_symbol(name, len); 232 int index = the_table()->hash_to_index(hashValue); 233 234 Symbol* s = the_table()->lookup(index, name, len, hashValue); 235 236 // Found 237 if (s != NULL) return s; 238 239 // Grab SymbolTable_lock first. 240 MutexLocker ml(SymbolTable_lock, THREAD); 241 242 // Otherwise, add to symbol to table 243 return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL); 244 } 245 246 Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) { 247 char* buffer; 248 int index, len; 249 unsigned int hashValue; 250 char* name; 251 { 252 debug_only(No_Safepoint_Verifier nsv;) 253 254 name = (char*)sym->base() + begin; 255 len = end - begin; 256 hashValue = hash_symbol(name, len); 257 index = the_table()->hash_to_index(hashValue); 258 Symbol* s = the_table()->lookup(index, name, len, hashValue); 259 260 // Found 261 if (s != NULL) return s; 262 } 263 264 // Otherwise, add to symbol to table. Copy to a C string first. 265 char stack_buf[128]; 266 ResourceMark rm(THREAD); 267 if (len <= 128) { 268 buffer = stack_buf; 269 } else { 270 buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len); 271 } 272 for (int i=0; i<len; i++) { 273 buffer[i] = name[i]; 274 } 275 // Make sure there is no safepoint in the code above since name can't move. 276 // We can't include the code in No_Safepoint_Verifier because of the 277 // ResourceMark. 278 279 // Grab SymbolTable_lock first. 280 MutexLocker ml(SymbolTable_lock, THREAD); 281 282 return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL); 283 } 284 285 Symbol* SymbolTable::lookup_only(const char* name, int len, 286 unsigned int& hash) { 287 hash = hash_symbol(name, len); 288 int index = the_table()->hash_to_index(hash); 289 290 Symbol* s = the_table()->lookup(index, name, len, hash); 291 return s; 292 } 293 294 // Look up the address of the literal in the SymbolTable for this Symbol* 295 // Do not create any new symbols 296 // Do not increment the reference count to keep this alive 297 Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){ 298 unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length()); 299 int index = the_table()->hash_to_index(hash); 300 301 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) { 302 if (e->hash() == hash) { 303 Symbol* literal_sym = e->literal(); 304 if (sym == literal_sym) { 305 return e->literal_addr(); 306 } 307 } 308 } 309 return NULL; 310 } 311 312 // Suggestion: Push unicode-based lookup all the way into the hashing 313 // and probing logic, so there is no need for convert_to_utf8 until 314 // an actual new Symbol* is created. 315 Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) { 316 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length); 317 char stack_buf[128]; 318 if (utf8_length < (int) sizeof(stack_buf)) { 319 char* chars = stack_buf; 320 UNICODE::convert_to_utf8(name, utf16_length, chars); 321 return lookup(chars, utf8_length, THREAD); 322 } else { 323 ResourceMark rm(THREAD); 324 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);; 325 UNICODE::convert_to_utf8(name, utf16_length, chars); 326 return lookup(chars, utf8_length, THREAD); 327 } 328 } 329 330 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length, 331 unsigned int& hash) { 332 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length); 333 char stack_buf[128]; 334 if (utf8_length < (int) sizeof(stack_buf)) { 335 char* chars = stack_buf; 336 UNICODE::convert_to_utf8(name, utf16_length, chars); 337 return lookup_only(chars, utf8_length, hash); 338 } else { 339 ResourceMark rm; 340 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);; 341 UNICODE::convert_to_utf8(name, utf16_length, chars); 342 return lookup_only(chars, utf8_length, hash); 343 } 344 } 345 346 void SymbolTable::add(ClassLoaderData* loader_data, constantPoolHandle cp, 347 int names_count, 348 const char** names, int* lengths, int* cp_indices, 349 unsigned int* hashValues, TRAPS) { 350 // Grab SymbolTable_lock first. 351 MutexLocker ml(SymbolTable_lock, THREAD); 352 353 SymbolTable* table = the_table(); 354 bool added = table->basic_add(loader_data, cp, names_count, names, lengths, 355 cp_indices, hashValues, CHECK); 356 if (!added) { 357 // do it the hard way 358 for (int i=0; i<names_count; i++) { 359 int index = table->hash_to_index(hashValues[i]); 360 bool c_heap = !loader_data->is_the_null_class_loader_data(); 361 Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK); 362 cp->symbol_at_put(cp_indices[i], sym); 363 } 364 } 365 } 366 367 Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) { 368 unsigned int hash; 369 Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash); 370 if (result != NULL) { 371 return result; 372 } 373 // Grab SymbolTable_lock first. 374 MutexLocker ml(SymbolTable_lock, THREAD); 375 376 SymbolTable* table = the_table(); 377 int index = table->hash_to_index(hash); 378 return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD); 379 } 380 381 Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len, 382 unsigned int hashValue_arg, bool c_heap, TRAPS) { 383 assert(!Universe::heap()->is_in_reserved(name), 384 "proposed name of symbol must be stable"); 385 386 // Don't allow symbols to be created which cannot fit in a Symbol*. 387 if (len > Symbol::max_length()) { 388 THROW_MSG_0(vmSymbols::java_lang_InternalError(), 389 "name is too long to represent"); 390 } 391 392 // Cannot hit a safepoint in this function because the "this" pointer can move. 393 No_Safepoint_Verifier nsv; 394 395 // Check if the symbol table has been rehashed, if so, need to recalculate 396 // the hash value and index. 397 unsigned int hashValue; 398 int index; 399 if (use_alternate_hashcode()) { 400 hashValue = hash_symbol((const char*)name, len); 401 index = hash_to_index(hashValue); 402 } else { 403 hashValue = hashValue_arg; 404 index = index_arg; 405 } 406 407 // Since look-up was done lock-free, we need to check if another 408 // thread beat us in the race to insert the symbol. 409 Symbol* test = lookup(index, (char*)name, len, hashValue); 410 if (test != NULL) { 411 // A race occurred and another thread introduced the symbol. 412 assert(test->refcount() != 0, "lookup should have incremented the count"); 413 return test; 414 } 415 416 // Create a new symbol. 417 Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL); 418 assert(sym->equals((char*)name, len), "symbol must be properly initialized"); 419 420 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym); 421 add_entry(index, entry); 422 return sym; 423 } 424 425 // This version of basic_add adds symbols in batch from the constant pool 426 // parsing. 427 bool SymbolTable::basic_add(ClassLoaderData* loader_data, constantPoolHandle cp, 428 int names_count, 429 const char** names, int* lengths, 430 int* cp_indices, unsigned int* hashValues, 431 TRAPS) { 432 433 // Check symbol names are not too long. If any are too long, don't add any. 434 for (int i = 0; i< names_count; i++) { 435 if (lengths[i] > Symbol::max_length()) { 436 THROW_MSG_0(vmSymbols::java_lang_InternalError(), 437 "name is too long to represent"); 438 } 439 } 440 441 // Cannot hit a safepoint in this function because the "this" pointer can move. 442 No_Safepoint_Verifier nsv; 443 444 for (int i=0; i<names_count; i++) { 445 // Check if the symbol table has been rehashed, if so, need to recalculate 446 // the hash value. 447 unsigned int hashValue; 448 if (use_alternate_hashcode()) { 449 hashValue = hash_symbol(names[i], lengths[i]); 450 } else { 451 hashValue = hashValues[i]; 452 } 453 // Since look-up was done lock-free, we need to check if another 454 // thread beat us in the race to insert the symbol. 455 int index = hash_to_index(hashValue); 456 Symbol* test = lookup(index, names[i], lengths[i], hashValue); 457 if (test != NULL) { 458 // A race occurred and another thread introduced the symbol, this one 459 // will be dropped and collected. Use test instead. 460 cp->symbol_at_put(cp_indices[i], test); 461 assert(test->refcount() != 0, "lookup should have incremented the count"); 462 } else { 463 // Create a new symbol. The null class loader is never unloaded so these 464 // are allocated specially in a permanent arena. 465 bool c_heap = !loader_data->is_the_null_class_loader_data(); 466 Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false)); 467 assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized"); // why wouldn't it be??? 468 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym); 469 add_entry(index, entry); 470 cp->symbol_at_put(cp_indices[i], sym); 471 } 472 } 473 return true; 474 } 475 476 477 void SymbolTable::verify() { 478 for (int i = 0; i < the_table()->table_size(); ++i) { 479 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i); 480 for ( ; p != NULL; p = p->next()) { 481 Symbol* s = (Symbol*)(p->literal()); 482 guarantee(s != NULL, "symbol is NULL"); 483 unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length()); 484 guarantee(p->hash() == h, "broken hash in symbol table entry"); 485 guarantee(the_table()->hash_to_index(h) == i, 486 "wrong index in symbol table"); 487 } 488 } 489 } 490 491 void SymbolTable::dump(outputStream* st) { 492 the_table()->dump_table(st, "SymbolTable"); 493 } 494 495 496 //--------------------------------------------------------------------------- 497 // Non-product code 498 499 #ifndef PRODUCT 500 501 void SymbolTable::print_histogram() { 502 MutexLocker ml(SymbolTable_lock); 503 const int results_length = 100; 504 int results[results_length]; 505 int i,j; 506 507 // initialize results to zero 508 for (j = 0; j < results_length; j++) { 509 results[j] = 0; 510 } 511 512 int total = 0; 513 int max_symbols = 0; 514 int out_of_range = 0; 515 int memory_total = 0; 516 int count = 0; 517 for (i = 0; i < the_table()->table_size(); i++) { 518 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i); 519 for ( ; p != NULL; p = p->next()) { 520 memory_total += p->literal()->size(); 521 count++; 522 int counter = p->literal()->utf8_length(); 523 total += counter; 524 if (counter < results_length) { 525 results[counter]++; 526 } else { 527 out_of_range++; 528 } 529 max_symbols = MAX2(max_symbols, counter); 530 } 531 } 532 tty->print_cr("Symbol Table:"); 533 tty->print_cr("Total number of symbols %5d", count); 534 tty->print_cr("Total size in memory %5dK", 535 (memory_total*HeapWordSize)/1024); 536 tty->print_cr("Total counted %5d", _symbols_counted); 537 tty->print_cr("Total removed %5d", _symbols_removed); 538 if (_symbols_counted > 0) { 539 tty->print_cr("Percent removed %3.2f", 540 ((float)_symbols_removed/(float)_symbols_counted)* 100); 541 } 542 tty->print_cr("Reference counts %5d", Symbol::_total_count); 543 tty->print_cr("Symbol arena size %5d used %5d", 544 arena()->size_in_bytes(), arena()->used()); 545 tty->print_cr("Histogram of symbol length:"); 546 tty->print_cr("%8s %5d", "Total ", total); 547 tty->print_cr("%8s %5d", "Maximum", max_symbols); 548 tty->print_cr("%8s %3.2f", "Average", 549 ((float) total / (float) the_table()->table_size())); 550 tty->print_cr("%s", "Histogram:"); 551 tty->print_cr(" %s %29s", "Length", "Number chains that length"); 552 for (i = 0; i < results_length; i++) { 553 if (results[i] > 0) { 554 tty->print_cr("%6d %10d", i, results[i]); 555 } 556 } 557 if (Verbose) { 558 int line_length = 70; 559 tty->print_cr("%s %30s", " Length", "Number chains that length"); 560 for (i = 0; i < results_length; i++) { 561 if (results[i] > 0) { 562 tty->print("%4d", i); 563 for (j = 0; (j < results[i]) && (j < line_length); j++) { 564 tty->print("%1s", "*"); 565 } 566 if (j == line_length) { 567 tty->print("%1s", "+"); 568 } 569 tty->cr(); 570 } 571 } 572 } 573 tty->print_cr(" %s %d: %d\n", "Number chains longer than", 574 results_length, out_of_range); 575 } 576 577 void SymbolTable::print() { 578 for (int i = 0; i < the_table()->table_size(); ++i) { 579 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i); 580 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i); 581 if (entry != NULL) { 582 while (entry != NULL) { 583 tty->print(PTR_FORMAT " ", entry->literal()); 584 entry->literal()->print(); 585 tty->print(" %d", entry->literal()->refcount()); 586 p = entry->next_addr(); 587 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p); 588 } 589 tty->cr(); 590 } 591 } 592 } 593 #endif // PRODUCT 594 595 // -------------------------------------------------------------------------- 596 597 #ifdef ASSERT 598 class StableMemoryChecker : public StackObj { 599 enum { _bufsize = wordSize*4 }; 600 601 address _region; 602 jint _size; 603 u1 _save_buf[_bufsize]; 604 605 int sample(u1* save_buf) { 606 if (_size <= _bufsize) { 607 memcpy(save_buf, _region, _size); 608 return _size; 609 } else { 610 // copy head and tail 611 memcpy(&save_buf[0], _region, _bufsize/2); 612 memcpy(&save_buf[_bufsize/2], _region + _size - _bufsize/2, _bufsize/2); 613 return (_bufsize/2)*2; 614 } 615 } 616 617 public: 618 StableMemoryChecker(const void* region, jint size) { 619 _region = (address) region; 620 _size = size; 621 sample(_save_buf); 622 } 623 624 bool verify() { 625 u1 check_buf[sizeof(_save_buf)]; 626 int check_size = sample(check_buf); 627 return (0 == memcmp(_save_buf, check_buf, check_size)); 628 } 629 630 void set_region(const void* region) { _region = (address) region; } 631 }; 632 #endif 633 634 635 // -------------------------------------------------------------------------- 636 StringTable* StringTable::_the_table = NULL; 637 638 bool StringTable::_needs_rehashing = false; 639 640 volatile int StringTable::_parallel_claimed_idx = 0; 641 642 // Pick hashing algorithm 643 unsigned int StringTable::hash_string(const jchar* s, int len) { 644 return use_alternate_hashcode() ? AltHashing::murmur3_32(seed(), s, len) : 645 java_lang_String::hash_code(s, len); 646 } 647 648 oop StringTable::lookup(int index, jchar* name, 649 int len, unsigned int hash) { 650 int count = 0; 651 for (HashtableEntry<oop, mtSymbol>* l = bucket(index); l != NULL; l = l->next()) { 652 count++; 653 if (l->hash() == hash) { 654 if (java_lang_String::equals(l->literal(), name, len)) { 655 return l->literal(); 656 } 657 } 658 } 659 // If the bucket size is too deep check if this hash code is insufficient. 660 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) { 661 _needs_rehashing = check_rehash_table(count); 662 } 663 return NULL; 664 } 665 666 667 oop StringTable::basic_add(int index_arg, Handle string, jchar* name, 668 int len, unsigned int hashValue_arg, TRAPS) { 669 670 assert(java_lang_String::equals(string(), name, len), 671 "string must be properly initialized"); 672 // Cannot hit a safepoint in this function because the "this" pointer can move. 673 No_Safepoint_Verifier nsv; 674 675 // Check if the symbol table has been rehashed, if so, need to recalculate 676 // the hash value and index before second lookup. 677 unsigned int hashValue; 678 int index; 679 if (use_alternate_hashcode()) { 680 hashValue = hash_string(name, len); 681 index = hash_to_index(hashValue); 682 } else { 683 hashValue = hashValue_arg; 684 index = index_arg; 685 } 686 687 // Since look-up was done lock-free, we need to check if another 688 // thread beat us in the race to insert the symbol. 689 690 oop test = lookup(index, name, len, hashValue); // calls lookup(u1*, int) 691 if (test != NULL) { 692 // Entry already added 693 return test; 694 } 695 696 HashtableEntry<oop, mtSymbol>* entry = new_entry(hashValue, string()); 697 add_entry(index, entry); 698 return string(); 699 } 700 701 702 oop StringTable::lookup(Symbol* symbol) { 703 ResourceMark rm; 704 int length; 705 jchar* chars = symbol->as_unicode(length); 706 return lookup(chars, length); 707 } 708 709 // Tell the GC that this string was looked up in the StringTable. 710 static void ensure_string_alive(oop string) { 711 // A lookup in the StringTable could return an object that was previously 712 // considered dead. The SATB part of G1 needs to get notified about this 713 // potential resurrection, otherwise the marking might not find the object. 714 #if INCLUDE_ALL_GCS 715 if (UseG1GC && string != NULL) { 716 G1SATBCardTableModRefBS::enqueue(string); 717 } 718 #endif 719 } 720 721 oop StringTable::lookup(jchar* name, int len) { 722 unsigned int hash = hash_string(name, len); 723 int index = the_table()->hash_to_index(hash); 724 oop string = the_table()->lookup(index, name, len, hash); 725 726 ensure_string_alive(string); 727 728 return string; 729 } 730 731 732 oop StringTable::intern(Handle string_or_null, jchar* name, 733 int len, TRAPS) { 734 unsigned int hashValue = hash_string(name, len); 735 int index = the_table()->hash_to_index(hashValue); 736 oop found_string = the_table()->lookup(index, name, len, hashValue); 737 738 // Found 739 if (found_string != NULL) { 740 ensure_string_alive(found_string); 741 return found_string; 742 } 743 744 debug_only(StableMemoryChecker smc(name, len * sizeof(name[0]))); 745 assert(!Universe::heap()->is_in_reserved(name), 746 "proposed name of symbol must be stable"); 747 748 Handle string; 749 // try to reuse the string if possible 750 if (!string_or_null.is_null()) { 751 string = string_or_null; 752 } else { 753 string = java_lang_String::create_from_unicode(name, len, CHECK_NULL); 754 } 755 756 #if INCLUDE_ALL_GCS 757 if (G1StringDedup::is_enabled()) { 758 // Deduplicate the string before it is interned. Note that we should never 759 // deduplicate a string after it has been interned. Doing so will counteract 760 // compiler optimizations done on e.g. interned string literals. 761 G1StringDedup::deduplicate(string()); 762 } 763 #endif 764 765 // Grab the StringTable_lock before getting the_table() because it could 766 // change at safepoint. 767 oop added_or_found; 768 { 769 MutexLocker ml(StringTable_lock, THREAD); 770 // Otherwise, add to symbol to table 771 added_or_found = the_table()->basic_add(index, string, name, len, 772 hashValue, CHECK_NULL); 773 } 774 775 ensure_string_alive(added_or_found); 776 777 return added_or_found; 778 } 779 780 oop StringTable::intern(Symbol* symbol, TRAPS) { 781 if (symbol == NULL) return NULL; 782 ResourceMark rm(THREAD); 783 int length; 784 jchar* chars = symbol->as_unicode(length); 785 Handle string; 786 oop result = intern(string, chars, length, CHECK_NULL); 787 return result; 788 } 789 790 791 oop StringTable::intern(oop string, TRAPS) 792 { 793 if (string == NULL) return NULL; 794 ResourceMark rm(THREAD); 795 int length; 796 Handle h_string (THREAD, string); 797 jchar* chars = java_lang_String::as_unicode_string(string, length, CHECK_NULL); 798 oop result = intern(h_string, chars, length, CHECK_NULL); 799 return result; 800 } 801 802 803 oop StringTable::intern(const char* utf8_string, TRAPS) { 804 if (utf8_string == NULL) return NULL; 805 ResourceMark rm(THREAD); 806 int length = UTF8::unicode_length(utf8_string); 807 jchar* chars = NEW_RESOURCE_ARRAY(jchar, length); 808 UTF8::convert_to_unicode(utf8_string, chars, length); 809 Handle string; 810 oop result = intern(string, chars, length, CHECK_NULL); 811 return result; 812 } 813 814 void StringTable::unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int* processed, int* removed) { 815 buckets_unlink_or_oops_do(is_alive, f, 0, the_table()->table_size(), processed, removed); 816 } 817 818 void StringTable::possibly_parallel_unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int* processed, int* removed) { 819 // Readers of the table are unlocked, so we should only be removing 820 // entries at a safepoint. 821 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint"); 822 const int limit = the_table()->table_size(); 823 824 for (;;) { 825 // Grab next set of buckets to scan 826 int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize; 827 if (start_idx >= limit) { 828 // End of table 829 break; 830 } 831 832 int end_idx = MIN2(limit, start_idx + ClaimChunkSize); 833 buckets_unlink_or_oops_do(is_alive, f, start_idx, end_idx, processed, removed); 834 } 835 } 836 837 void StringTable::buckets_oops_do(OopClosure* f, int start_idx, int end_idx) { 838 const int limit = the_table()->table_size(); 839 840 assert(0 <= start_idx && start_idx <= limit, 841 err_msg("start_idx (" INT32_FORMAT ") is out of bounds", start_idx)); 842 assert(0 <= end_idx && end_idx <= limit, 843 err_msg("end_idx (" INT32_FORMAT ") is out of bounds", end_idx)); 844 assert(start_idx <= end_idx, 845 err_msg("Index ordering: start_idx=" INT32_FORMAT", end_idx=" INT32_FORMAT, 846 start_idx, end_idx)); 847 848 for (int i = start_idx; i < end_idx; i += 1) { 849 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i); 850 while (entry != NULL) { 851 assert(!entry->is_shared(), "CDS not used for the StringTable"); 852 853 f->do_oop((oop*)entry->literal_addr()); 854 855 entry = entry->next(); 856 } 857 } 858 } 859 860 void StringTable::buckets_unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int start_idx, int end_idx, int* processed, int* removed) { 861 const int limit = the_table()->table_size(); 862 863 assert(0 <= start_idx && start_idx <= limit, 864 err_msg("start_idx (" INT32_FORMAT ") is out of bounds", start_idx)); 865 assert(0 <= end_idx && end_idx <= limit, 866 err_msg("end_idx (" INT32_FORMAT ") is out of bounds", end_idx)); 867 assert(start_idx <= end_idx, 868 err_msg("Index ordering: start_idx=" INT32_FORMAT", end_idx=" INT32_FORMAT, 869 start_idx, end_idx)); 870 871 for (int i = start_idx; i < end_idx; ++i) { 872 HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i); 873 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i); 874 while (entry != NULL) { 875 assert(!entry->is_shared(), "CDS not used for the StringTable"); 876 877 if (is_alive->do_object_b(entry->literal())) { 878 if (f != NULL) { 879 f->do_oop((oop*)entry->literal_addr()); 880 } 881 p = entry->next_addr(); 882 } else { 883 *p = entry->next(); 884 the_table()->free_entry(entry); 885 (*removed)++; 886 } 887 (*processed)++; 888 entry = *p; 889 } 890 } 891 } 892 893 void StringTable::oops_do(OopClosure* f) { 894 buckets_oops_do(f, 0, the_table()->table_size()); 895 } 896 897 void StringTable::possibly_parallel_oops_do(OopClosure* f) { 898 const int limit = the_table()->table_size(); 899 900 for (;;) { 901 // Grab next set of buckets to scan 902 int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize; 903 if (start_idx >= limit) { 904 // End of table 905 break; 906 } 907 908 int end_idx = MIN2(limit, start_idx + ClaimChunkSize); 909 buckets_oops_do(f, start_idx, end_idx); 910 } 911 } 912 913 // This verification is part of Universe::verify() and needs to be quick. 914 // See StringTable::verify_and_compare() below for exhaustive verification. 915 void StringTable::verify() { 916 for (int i = 0; i < the_table()->table_size(); ++i) { 917 HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i); 918 for ( ; p != NULL; p = p->next()) { 919 oop s = p->literal(); 920 guarantee(s != NULL, "interned string is NULL"); 921 unsigned int h = java_lang_String::hash_string(s); 922 guarantee(p->hash() == h, "broken hash in string table entry"); 923 guarantee(the_table()->hash_to_index(h) == i, 924 "wrong index in string table"); 925 } 926 } 927 } 928 929 void StringTable::dump(outputStream* st) { 930 the_table()->dump_table(st, "StringTable"); 931 } 932 933 StringTable::VerifyRetTypes StringTable::compare_entries( 934 int bkt1, int e_cnt1, 935 HashtableEntry<oop, mtSymbol>* e_ptr1, 936 int bkt2, int e_cnt2, 937 HashtableEntry<oop, mtSymbol>* e_ptr2) { 938 // These entries are sanity checked by verify_and_compare_entries() 939 // before this function is called. 940 oop str1 = e_ptr1->literal(); 941 oop str2 = e_ptr2->literal(); 942 943 if (str1 == str2) { 944 tty->print_cr("ERROR: identical oop values (0x" PTR_FORMAT ") " 945 "in entry @ bucket[%d][%d] and entry @ bucket[%d][%d]", 946 (void *)str1, bkt1, e_cnt1, bkt2, e_cnt2); 947 return _verify_fail_continue; 948 } 949 950 if (java_lang_String::equals(str1, str2)) { 951 tty->print_cr("ERROR: identical String values in entry @ " 952 "bucket[%d][%d] and entry @ bucket[%d][%d]", 953 bkt1, e_cnt1, bkt2, e_cnt2); 954 return _verify_fail_continue; 955 } 956 957 return _verify_pass; 958 } 959 960 StringTable::VerifyRetTypes StringTable::verify_entry(int bkt, int e_cnt, 961 HashtableEntry<oop, mtSymbol>* e_ptr, 962 StringTable::VerifyMesgModes mesg_mode) { 963 964 VerifyRetTypes ret = _verify_pass; // be optimistic 965 966 oop str = e_ptr->literal(); 967 if (str == NULL) { 968 if (mesg_mode == _verify_with_mesgs) { 969 tty->print_cr("ERROR: NULL oop value in entry @ bucket[%d][%d]", bkt, 970 e_cnt); 971 } 972 // NULL oop means no more verifications are possible 973 return _verify_fail_done; 974 } 975 976 if (str->klass() != SystemDictionary::String_klass()) { 977 if (mesg_mode == _verify_with_mesgs) { 978 tty->print_cr("ERROR: oop is not a String in entry @ bucket[%d][%d]", 979 bkt, e_cnt); 980 } 981 // not a String means no more verifications are possible 982 return _verify_fail_done; 983 } 984 985 unsigned int h = java_lang_String::hash_string(str); 986 if (e_ptr->hash() != h) { 987 if (mesg_mode == _verify_with_mesgs) { 988 tty->print_cr("ERROR: broken hash value in entry @ bucket[%d][%d], " 989 "bkt_hash=%d, str_hash=%d", bkt, e_cnt, e_ptr->hash(), h); 990 } 991 ret = _verify_fail_continue; 992 } 993 994 if (the_table()->hash_to_index(h) != bkt) { 995 if (mesg_mode == _verify_with_mesgs) { 996 tty->print_cr("ERROR: wrong index value for entry @ bucket[%d][%d], " 997 "str_hash=%d, hash_to_index=%d", bkt, e_cnt, h, 998 the_table()->hash_to_index(h)); 999 } 1000 ret = _verify_fail_continue; 1001 } 1002 1003 return ret; 1004 } 1005 1006 // See StringTable::verify() above for the quick verification that is 1007 // part of Universe::verify(). This verification is exhaustive and 1008 // reports on every issue that is found. StringTable::verify() only 1009 // reports on the first issue that is found. 1010 // 1011 // StringTable::verify_entry() checks: 1012 // - oop value != NULL (same as verify()) 1013 // - oop value is a String 1014 // - hash(String) == hash in entry (same as verify()) 1015 // - index for hash == index of entry (same as verify()) 1016 // 1017 // StringTable::compare_entries() checks: 1018 // - oops are unique across all entries 1019 // - String values are unique across all entries 1020 // 1021 int StringTable::verify_and_compare_entries() { 1022 assert(StringTable_lock->is_locked(), "sanity check"); 1023 1024 int fail_cnt = 0; 1025 1026 // first, verify all the entries individually: 1027 for (int bkt = 0; bkt < the_table()->table_size(); bkt++) { 1028 HashtableEntry<oop, mtSymbol>* e_ptr = the_table()->bucket(bkt); 1029 for (int e_cnt = 0; e_ptr != NULL; e_ptr = e_ptr->next(), e_cnt++) { 1030 VerifyRetTypes ret = verify_entry(bkt, e_cnt, e_ptr, _verify_with_mesgs); 1031 if (ret != _verify_pass) { 1032 fail_cnt++; 1033 } 1034 } 1035 } 1036 1037 // Optimization: if the above check did not find any failures, then 1038 // the comparison loop below does not need to call verify_entry() 1039 // before calling compare_entries(). If there were failures, then we 1040 // have to call verify_entry() to see if the entry can be passed to 1041 // compare_entries() safely. When we call verify_entry() in the loop 1042 // below, we do so quietly to void duplicate messages and we don't 1043 // increment fail_cnt because the failures have already been counted. 1044 bool need_entry_verify = (fail_cnt != 0); 1045 1046 // second, verify all entries relative to each other: 1047 for (int bkt1 = 0; bkt1 < the_table()->table_size(); bkt1++) { 1048 HashtableEntry<oop, mtSymbol>* e_ptr1 = the_table()->bucket(bkt1); 1049 for (int e_cnt1 = 0; e_ptr1 != NULL; e_ptr1 = e_ptr1->next(), e_cnt1++) { 1050 if (need_entry_verify) { 1051 VerifyRetTypes ret = verify_entry(bkt1, e_cnt1, e_ptr1, 1052 _verify_quietly); 1053 if (ret == _verify_fail_done) { 1054 // cannot use the current entry to compare against other entries 1055 continue; 1056 } 1057 } 1058 1059 for (int bkt2 = bkt1; bkt2 < the_table()->table_size(); bkt2++) { 1060 HashtableEntry<oop, mtSymbol>* e_ptr2 = the_table()->bucket(bkt2); 1061 int e_cnt2; 1062 for (e_cnt2 = 0; e_ptr2 != NULL; e_ptr2 = e_ptr2->next(), e_cnt2++) { 1063 if (bkt1 == bkt2 && e_cnt2 <= e_cnt1) { 1064 // skip the entries up to and including the one that 1065 // we're comparing against 1066 continue; 1067 } 1068 1069 if (need_entry_verify) { 1070 VerifyRetTypes ret = verify_entry(bkt2, e_cnt2, e_ptr2, 1071 _verify_quietly); 1072 if (ret == _verify_fail_done) { 1073 // cannot compare against this entry 1074 continue; 1075 } 1076 } 1077 1078 // compare two entries, report and count any failures: 1079 if (compare_entries(bkt1, e_cnt1, e_ptr1, bkt2, e_cnt2, e_ptr2) 1080 != _verify_pass) { 1081 fail_cnt++; 1082 } 1083 } 1084 } 1085 } 1086 } 1087 return fail_cnt; 1088 } 1089 1090 // Create a new table and using alternate hash code, populate the new table 1091 // with the existing strings. Set flag to use the alternate hash code afterwards. 1092 void StringTable::rehash_table() { 1093 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint"); 1094 // This should never happen with -Xshare:dump but it might in testing mode. 1095 if (DumpSharedSpaces) return; 1096 StringTable* new_table = new StringTable(); 1097 1098 // Rehash the table 1099 the_table()->move_to(new_table); 1100 1101 // Delete the table and buckets (entries are reused in new table). 1102 delete _the_table; 1103 // Don't check if we need rehashing until the table gets unbalanced again. 1104 // Then rehash with a new global seed. 1105 _needs_rehashing = false; 1106 _the_table = new_table; 1107 }