src/share/vm/classfile/symbolTable.cpp

Wed, 04 Jul 2012 15:55:45 -0400

author
coleenp
date
Wed, 04 Jul 2012 15:55:45 -0400
changeset 3904
ace99a6ffc83
parent 3900
d2a62e0f25eb
child 4037
da91efe96a93
permissions
-rw-r--r--

7181200: JVM new hashing code breaks SA in product mode
Summary: Made new_hash() overloaded rather than a virtual function so SA code doesn't need to be changed.
Reviewed-by: kvn, acorn, dholmes, fparain

duke@435 1 /*
coleenp@3682 2 * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
duke@435 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435 4 *
duke@435 5 * This code is free software; you can redistribute it and/or modify it
duke@435 6 * under the terms of the GNU General Public License version 2 only, as
duke@435 7 * published by the Free Software Foundation.
duke@435 8 *
duke@435 9 * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@435 12 * version 2 for more details (a copy is included in the LICENSE file that
duke@435 13 * accompanied this code).
duke@435 14 *
duke@435 15 * You should have received a copy of the GNU General Public License version
duke@435 16 * 2 along with this work; if not, write to the Free Software Foundation,
duke@435 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435 18 *
trims@1907 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907 20 * or visit www.oracle.com if you need additional information or have any
trims@1907 21 * questions.
duke@435 22 *
duke@435 23 */
duke@435 24
stefank@2314 25 #include "precompiled.hpp"
coleenp@3865 26 #include "classfile/altHashing.hpp"
stefank@2314 27 #include "classfile/javaClasses.hpp"
stefank@2314 28 #include "classfile/symbolTable.hpp"
stefank@2314 29 #include "classfile/systemDictionary.hpp"
stefank@2314 30 #include "gc_interface/collectedHeap.inline.hpp"
coleenp@3682 31 #include "memory/allocation.inline.hpp"
stefank@2314 32 #include "memory/filemap.hpp"
stefank@2314 33 #include "memory/gcLocker.inline.hpp"
stefank@2314 34 #include "oops/oop.inline.hpp"
stefank@2314 35 #include "oops/oop.inline2.hpp"
stefank@2314 36 #include "runtime/mutexLocker.hpp"
stefank@2314 37 #include "utilities/hashtable.inline.hpp"
coleenp@3865 38 #include "utilities/numberSeq.hpp"
duke@435 39
duke@435 40 // --------------------------------------------------------------------------
duke@435 41
duke@435 42 SymbolTable* SymbolTable::_the_table = NULL;
coleenp@3682 43 // Static arena for symbols that are not deallocated
coleenp@3682 44 Arena* SymbolTable::_arena = NULL;
coleenp@3865 45 bool SymbolTable::_needs_rehashing = false;
duke@435 46
coleenp@3682 47 Symbol* SymbolTable::allocate_symbol(const u1* name, int len, bool c_heap, TRAPS) {
coleenp@3875 48 assert (len <= Symbol::max_length(), "should be checked by caller");
coleenp@3875 49
coleenp@3682 50 Symbol* sym;
coleenp@3682 51 // Allocate symbols in the C heap when dumping shared spaces in case there
coleenp@3682 52 // are temporary symbols we can remove.
coleenp@3682 53 if (c_heap || DumpSharedSpaces) {
coleenp@3682 54 // refcount starts as 1
coleenp@3682 55 sym = new (len, THREAD) Symbol(name, len, 1);
coleenp@3682 56 } else {
coleenp@3682 57 sym = new (len, arena(), THREAD) Symbol(name, len, -1);
coleenp@3682 58 }
coleenp@2497 59 assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted");
coleenp@2497 60 return sym;
coleenp@2497 61 }
coleenp@2497 62
coleenp@3682 63 void SymbolTable::initialize_symbols(int arena_alloc_size) {
coleenp@3682 64 // Initialize the arena for global symbols, size passed in depends on CDS.
coleenp@3682 65 if (arena_alloc_size == 0) {
zgu@3900 66 _arena = new (mtSymbol) Arena();
coleenp@3682 67 } else {
zgu@3900 68 _arena = new (mtSymbol) Arena(arena_alloc_size);
coleenp@2497 69 }
coleenp@2497 70 }
coleenp@2497 71
coleenp@2497 72 // Call function for all symbols in the symbol table.
coleenp@2497 73 void SymbolTable::symbols_do(SymbolClosure *cl) {
coleenp@2497 74 const int n = the_table()->table_size();
coleenp@2497 75 for (int i = 0; i < n; i++) {
zgu@3900 76 for (HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
coleenp@2497 77 p != NULL;
coleenp@2497 78 p = p->next()) {
coleenp@2497 79 cl->do_symbol(p->literal_addr());
coleenp@2497 80 }
coleenp@2497 81 }
coleenp@2497 82 }
coleenp@2497 83
coleenp@2497 84 int SymbolTable::symbols_removed = 0;
coleenp@2497 85 int SymbolTable::symbols_counted = 0;
coleenp@2497 86
coleenp@2497 87 // Remove unreferenced symbols from the symbol table
coleenp@3682 88 // This is done late during GC.
coleenp@2497 89 void SymbolTable::unlink() {
coleenp@2497 90 int removed = 0;
coleenp@2497 91 int total = 0;
coleenp@2618 92 size_t memory_total = 0;
coleenp@2497 93 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 94 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 95 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
coleenp@3875 96 while (entry != NULL) {
coleenp@3875 97 // Shared entries are normally at the end of the bucket and if we run into
coleenp@3875 98 // a shared entry, then there is nothing more to remove. However, if we
coleenp@3875 99 // have rehashed the table, then the shared entries are no longer at the
coleenp@3875 100 // end of the bucket.
coleenp@3875 101 if (entry->is_shared() && !use_alternate_hashcode()) {
coleenp@2497 102 break;
coleenp@2497 103 }
coleenp@2497 104 Symbol* s = entry->literal();
coleenp@2497 105 memory_total += s->object_size();
coleenp@2497 106 total++;
coleenp@2497 107 assert(s != NULL, "just checking");
coleenp@2497 108 // If reference count is zero, remove.
coleenp@2497 109 if (s->refcount() == 0) {
coleenp@3875 110 assert(!entry->is_shared(), "shared entries should be kept live");
coleenp@2497 111 delete s;
coleenp@2497 112 removed++;
coleenp@2497 113 *p = entry->next();
coleenp@2497 114 the_table()->free_entry(entry);
coleenp@2497 115 } else {
coleenp@2497 116 p = entry->next_addr();
coleenp@2497 117 }
coleenp@3875 118 // get next entry
zgu@3900 119 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
coleenp@2497 120 }
coleenp@2497 121 }
coleenp@2497 122 symbols_removed += removed;
coleenp@2497 123 symbols_counted += total;
coleenp@2618 124 // Exclude printing for normal PrintGCDetails because people parse
coleenp@2618 125 // this output.
coleenp@2618 126 if (PrintGCDetails && Verbose && WizardMode) {
coleenp@2618 127 gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", total,
coleenp@2497 128 (memory_total*HeapWordSize)/1024);
coleenp@2497 129 }
coleenp@2497 130 }
coleenp@2497 131
coleenp@3865 132 // Create a new table and using alternate hash code, populate the new table
coleenp@3865 133 // with the existing strings. Set flag to use the alternate hash code afterwards.
coleenp@3865 134 void SymbolTable::rehash_table() {
coleenp@3865 135 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@3875 136 // This should never happen with -Xshare:dump but it might in testing mode.
coleenp@3875 137 if (DumpSharedSpaces) return;
coleenp@3865 138 // Create a new symbol table
coleenp@3865 139 SymbolTable* new_table = new SymbolTable();
coleenp@3865 140
coleenp@3865 141 the_table()->move_to(new_table);
coleenp@3865 142
coleenp@3865 143 // Delete the table and buckets (entries are reused in new table).
coleenp@3865 144 delete _the_table;
coleenp@3865 145 // Don't check if we need rehashing until the table gets unbalanced again.
coleenp@3865 146 // Then rehash with a new global seed.
coleenp@3865 147 _needs_rehashing = false;
coleenp@3865 148 _the_table = new_table;
coleenp@3865 149 }
coleenp@2497 150
duke@435 151 // Lookup a symbol in a bucket.
duke@435 152
coleenp@2497 153 Symbol* SymbolTable::lookup(int index, const char* name,
duke@435 154 int len, unsigned int hash) {
coleenp@3865 155 int count = 0;
zgu@3900 156 for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) {
coleenp@3865 157 count++; // count all entries in this bucket, not just ones with same hash
duke@435 158 if (e->hash() == hash) {
coleenp@2497 159 Symbol* sym = e->literal();
duke@435 160 if (sym->equals(name, len)) {
coleenp@2497 161 // something is referencing this symbol now.
coleenp@2497 162 sym->increment_refcount();
duke@435 163 return sym;
duke@435 164 }
duke@435 165 }
duke@435 166 }
coleenp@3865 167 // If the bucket size is too deep check if this hash code is insufficient.
zgu@3900 168 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
coleenp@3865 169 _needs_rehashing = check_rehash_table(count);
coleenp@3865 170 }
duke@435 171 return NULL;
duke@435 172 }
duke@435 173
coleenp@3875 174 // Pick hashing algorithm.
coleenp@3875 175 unsigned int SymbolTable::hash_symbol(const char* s, int len) {
coleenp@3865 176 return use_alternate_hashcode() ?
coleenp@3865 177 AltHashing::murmur3_32(seed(), (const jbyte*)s, len) :
coleenp@3875 178 java_lang_String::to_hash(s, len);
coleenp@3865 179 }
coleenp@3865 180
duke@435 181
duke@435 182 // We take care not to be blocking while holding the
duke@435 183 // SymbolTable_lock. Otherwise, the system might deadlock, since the
duke@435 184 // symboltable is used during compilation (VM_thread) The lock free
duke@435 185 // synchronization is simplified by the fact that we do not delete
duke@435 186 // entries in the symbol table during normal execution (only during
duke@435 187 // safepoints).
duke@435 188
coleenp@2497 189 Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) {
duke@435 190 unsigned int hashValue = hash_symbol(name, len);
duke@435 191 int index = the_table()->hash_to_index(hashValue);
duke@435 192
coleenp@2497 193 Symbol* s = the_table()->lookup(index, name, len, hashValue);
duke@435 194
duke@435 195 // Found
duke@435 196 if (s != NULL) return s;
duke@435 197
coleenp@3875 198 // Grab SymbolTable_lock first.
coleenp@3875 199 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 200
duke@435 201 // Otherwise, add to symbol to table
coleenp@3682 202 return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL);
duke@435 203 }
duke@435 204
coleenp@2497 205 Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) {
duke@435 206 char* buffer;
duke@435 207 int index, len;
duke@435 208 unsigned int hashValue;
duke@435 209 char* name;
duke@435 210 {
duke@435 211 debug_only(No_Safepoint_Verifier nsv;)
duke@435 212
duke@435 213 name = (char*)sym->base() + begin;
duke@435 214 len = end - begin;
duke@435 215 hashValue = hash_symbol(name, len);
duke@435 216 index = the_table()->hash_to_index(hashValue);
coleenp@2497 217 Symbol* s = the_table()->lookup(index, name, len, hashValue);
duke@435 218
duke@435 219 // Found
duke@435 220 if (s != NULL) return s;
duke@435 221 }
duke@435 222
duke@435 223 // Otherwise, add to symbol to table. Copy to a C string first.
duke@435 224 char stack_buf[128];
duke@435 225 ResourceMark rm(THREAD);
duke@435 226 if (len <= 128) {
duke@435 227 buffer = stack_buf;
duke@435 228 } else {
duke@435 229 buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len);
duke@435 230 }
duke@435 231 for (int i=0; i<len; i++) {
duke@435 232 buffer[i] = name[i];
duke@435 233 }
duke@435 234 // Make sure there is no safepoint in the code above since name can't move.
duke@435 235 // We can't include the code in No_Safepoint_Verifier because of the
duke@435 236 // ResourceMark.
duke@435 237
coleenp@3875 238 // Grab SymbolTable_lock first.
coleenp@3875 239 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 240
coleenp@3682 241 return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL);
duke@435 242 }
duke@435 243
coleenp@2497 244 Symbol* SymbolTable::lookup_only(const char* name, int len,
duke@435 245 unsigned int& hash) {
duke@435 246 hash = hash_symbol(name, len);
duke@435 247 int index = the_table()->hash_to_index(hash);
duke@435 248
coleenp@2497 249 Symbol* s = the_table()->lookup(index, name, len, hash);
coleenp@2497 250 return s;
duke@435 251 }
duke@435 252
phh@3427 253 // Look up the address of the literal in the SymbolTable for this Symbol*
phh@3427 254 // Do not create any new symbols
phh@3427 255 // Do not increment the reference count to keep this alive
phh@3427 256 Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){
phh@3427 257 unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length());
phh@3427 258 int index = the_table()->hash_to_index(hash);
phh@3427 259
zgu@3900 260 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) {
phh@3427 261 if (e->hash() == hash) {
phh@3427 262 Symbol* literal_sym = e->literal();
phh@3427 263 if (sym == literal_sym) {
phh@3427 264 return e->literal_addr();
phh@3427 265 }
phh@3427 266 }
phh@3427 267 }
phh@3427 268 return NULL;
phh@3427 269 }
phh@3427 270
jrose@1100 271 // Suggestion: Push unicode-based lookup all the way into the hashing
jrose@1100 272 // and probing logic, so there is no need for convert_to_utf8 until
coleenp@2497 273 // an actual new Symbol* is created.
coleenp@2497 274 Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) {
jrose@1100 275 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
jrose@1100 276 char stack_buf[128];
jrose@1100 277 if (utf8_length < (int) sizeof(stack_buf)) {
jrose@1100 278 char* chars = stack_buf;
jrose@1100 279 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 280 return lookup(chars, utf8_length, THREAD);
jrose@1100 281 } else {
jrose@1100 282 ResourceMark rm(THREAD);
jrose@1100 283 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
jrose@1100 284 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 285 return lookup(chars, utf8_length, THREAD);
jrose@1100 286 }
jrose@1100 287 }
jrose@1100 288
coleenp@2497 289 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
jrose@1100 290 unsigned int& hash) {
jrose@1100 291 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
jrose@1100 292 char stack_buf[128];
jrose@1100 293 if (utf8_length < (int) sizeof(stack_buf)) {
jrose@1100 294 char* chars = stack_buf;
jrose@1100 295 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 296 return lookup_only(chars, utf8_length, hash);
jrose@1100 297 } else {
jrose@1100 298 ResourceMark rm;
jrose@1100 299 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
jrose@1100 300 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 301 return lookup_only(chars, utf8_length, hash);
jrose@1100 302 }
jrose@1100 303 }
jrose@1100 304
coleenp@3682 305 void SymbolTable::add(Handle class_loader, constantPoolHandle cp,
coleenp@3682 306 int names_count,
duke@435 307 const char** names, int* lengths, int* cp_indices,
duke@435 308 unsigned int* hashValues, TRAPS) {
coleenp@3875 309 // Grab SymbolTable_lock first.
coleenp@3875 310 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 311
duke@435 312 SymbolTable* table = the_table();
coleenp@3682 313 bool added = table->basic_add(class_loader, cp, names_count, names, lengths,
duke@435 314 cp_indices, hashValues, CHECK);
duke@435 315 if (!added) {
duke@435 316 // do it the hard way
duke@435 317 for (int i=0; i<names_count; i++) {
duke@435 318 int index = table->hash_to_index(hashValues[i]);
coleenp@3682 319 bool c_heap = class_loader() != NULL;
coleenp@3682 320 Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK);
duke@435 321 cp->symbol_at_put(cp_indices[i], sym);
duke@435 322 }
duke@435 323 }
duke@435 324 }
duke@435 325
coleenp@3682 326 Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) {
coleenp@3682 327 unsigned int hash;
coleenp@3682 328 Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash);
coleenp@3682 329 if (result != NULL) {
coleenp@3682 330 return result;
coleenp@3682 331 }
coleenp@3875 332 // Grab SymbolTable_lock first.
coleenp@3875 333 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 334
coleenp@3682 335 SymbolTable* table = the_table();
coleenp@3682 336 int index = table->hash_to_index(hash);
coleenp@3682 337 return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD);
coleenp@3682 338 }
coleenp@3682 339
coleenp@3875 340 Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len,
coleenp@3865 341 unsigned int hashValue_arg, bool c_heap, TRAPS) {
duke@435 342 assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),
duke@435 343 "proposed name of symbol must be stable");
duke@435 344
coleenp@3875 345 // Don't allow symbols to be created which cannot fit in a Symbol*.
coleenp@3875 346 if (len > Symbol::max_length()) {
coleenp@3875 347 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
coleenp@3875 348 "name is too long to represent");
coleenp@3875 349 }
coleenp@3875 350
coleenp@3875 351 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 352 No_Safepoint_Verifier nsv;
duke@435 353
coleenp@3865 354 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3875 355 // the hash value and index.
coleenp@3875 356 unsigned int hashValue;
coleenp@3875 357 int index;
coleenp@3875 358 if (use_alternate_hashcode()) {
coleenp@3875 359 hashValue = hash_symbol((const char*)name, len);
coleenp@3875 360 index = hash_to_index(hashValue);
coleenp@3875 361 } else {
coleenp@3875 362 hashValue = hashValue_arg;
coleenp@3875 363 index = index_arg;
coleenp@3875 364 }
coleenp@3865 365
duke@435 366 // Since look-up was done lock-free, we need to check if another
duke@435 367 // thread beat us in the race to insert the symbol.
coleenp@2497 368 Symbol* test = lookup(index, (char*)name, len, hashValue);
duke@435 369 if (test != NULL) {
coleenp@3682 370 // A race occurred and another thread introduced the symbol.
coleenp@2497 371 assert(test->refcount() != 0, "lookup should have incremented the count");
duke@435 372 return test;
duke@435 373 }
duke@435 374
coleenp@3682 375 // Create a new symbol.
coleenp@3682 376 Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL);
coleenp@3682 377 assert(sym->equals((char*)name, len), "symbol must be properly initialized");
coleenp@3682 378
zgu@3900 379 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
duke@435 380 add_entry(index, entry);
coleenp@2497 381 return sym;
duke@435 382 }
duke@435 383
coleenp@3682 384 // This version of basic_add adds symbols in batch from the constant pool
coleenp@3682 385 // parsing.
coleenp@3682 386 bool SymbolTable::basic_add(Handle class_loader, constantPoolHandle cp,
coleenp@3682 387 int names_count,
duke@435 388 const char** names, int* lengths,
duke@435 389 int* cp_indices, unsigned int* hashValues,
duke@435 390 TRAPS) {
coleenp@3682 391
coleenp@3682 392 // Check symbol names are not too long. If any are too long, don't add any.
coleenp@3682 393 for (int i = 0; i< names_count; i++) {
coleenp@3682 394 if (lengths[i] > Symbol::max_length()) {
coleenp@3682 395 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
coleenp@3682 396 "name is too long to represent");
coleenp@3682 397 }
duke@435 398 }
duke@435 399
coleenp@3875 400 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 401 No_Safepoint_Verifier nsv;
duke@435 402
coleenp@2497 403 for (int i=0; i<names_count; i++) {
coleenp@3865 404 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3865 405 // the hash value.
coleenp@3875 406 unsigned int hashValue;
coleenp@3875 407 if (use_alternate_hashcode()) {
coleenp@3875 408 hashValue = hash_symbol(names[i], lengths[i]);
coleenp@3875 409 } else {
coleenp@3875 410 hashValue = hashValues[i];
coleenp@3875 411 }
duke@435 412 // Since look-up was done lock-free, we need to check if another
duke@435 413 // thread beat us in the race to insert the symbol.
coleenp@3865 414 int index = hash_to_index(hashValue);
coleenp@3865 415 Symbol* test = lookup(index, names[i], lengths[i], hashValue);
duke@435 416 if (test != NULL) {
twisti@1040 417 // A race occurred and another thread introduced the symbol, this one
duke@435 418 // will be dropped and collected. Use test instead.
duke@435 419 cp->symbol_at_put(cp_indices[i], test);
coleenp@2497 420 assert(test->refcount() != 0, "lookup should have incremented the count");
duke@435 421 } else {
coleenp@3682 422 // Create a new symbol. The null class loader is never unloaded so these
coleenp@3682 423 // are allocated specially in a permanent arena.
coleenp@3682 424 bool c_heap = class_loader() != NULL;
coleenp@3682 425 Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false));
coleenp@3682 426 assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized"); // why wouldn't it be???
zgu@3900 427 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
duke@435 428 add_entry(index, entry);
duke@435 429 cp->symbol_at_put(cp_indices[i], sym);
duke@435 430 }
duke@435 431 }
duke@435 432 return true;
duke@435 433 }
duke@435 434
duke@435 435
duke@435 436 void SymbolTable::verify() {
duke@435 437 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 438 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
duke@435 439 for ( ; p != NULL; p = p->next()) {
coleenp@2497 440 Symbol* s = (Symbol*)(p->literal());
duke@435 441 guarantee(s != NULL, "symbol is NULL");
duke@435 442 unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length());
duke@435 443 guarantee(p->hash() == h, "broken hash in symbol table entry");
duke@435 444 guarantee(the_table()->hash_to_index(h) == i,
duke@435 445 "wrong index in symbol table");
duke@435 446 }
duke@435 447 }
duke@435 448 }
duke@435 449
coleenp@3865 450 void SymbolTable::dump(outputStream* st) {
coleenp@3865 451 NumberSeq summary;
coleenp@3865 452 for (int i = 0; i < the_table()->table_size(); ++i) {
coleenp@3865 453 int count = 0;
zgu@3900 454 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(i);
coleenp@3865 455 e != NULL; e = e->next()) {
coleenp@3865 456 count++;
coleenp@3865 457 }
coleenp@3865 458 summary.add((double)count);
coleenp@3865 459 }
coleenp@3865 460 st->print_cr("SymbolTable statistics:");
coleenp@3865 461 st->print_cr("Number of buckets : %7d", summary.num());
coleenp@3865 462 st->print_cr("Average bucket size : %7.0f", summary.avg());
coleenp@3865 463 st->print_cr("Variance of bucket size : %7.0f", summary.variance());
coleenp@3865 464 st->print_cr("Std. dev. of bucket size: %7.0f", summary.sd());
coleenp@3865 465 st->print_cr("Maximum bucket size : %7.0f", summary.maximum());
coleenp@3865 466 }
coleenp@3865 467
duke@435 468
duke@435 469 //---------------------------------------------------------------------------
duke@435 470 // Non-product code
duke@435 471
duke@435 472 #ifndef PRODUCT
duke@435 473
duke@435 474 void SymbolTable::print_histogram() {
duke@435 475 MutexLocker ml(SymbolTable_lock);
duke@435 476 const int results_length = 100;
duke@435 477 int results[results_length];
duke@435 478 int i,j;
duke@435 479
duke@435 480 // initialize results to zero
duke@435 481 for (j = 0; j < results_length; j++) {
duke@435 482 results[j] = 0;
duke@435 483 }
duke@435 484
duke@435 485 int total = 0;
duke@435 486 int max_symbols = 0;
duke@435 487 int out_of_range = 0;
coleenp@2497 488 int memory_total = 0;
coleenp@2497 489 int count = 0;
duke@435 490 for (i = 0; i < the_table()->table_size(); i++) {
zgu@3900 491 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
duke@435 492 for ( ; p != NULL; p = p->next()) {
coleenp@2497 493 memory_total += p->literal()->object_size();
coleenp@2497 494 count++;
coleenp@2497 495 int counter = p->literal()->utf8_length();
duke@435 496 total += counter;
duke@435 497 if (counter < results_length) {
duke@435 498 results[counter]++;
duke@435 499 } else {
duke@435 500 out_of_range++;
duke@435 501 }
duke@435 502 max_symbols = MAX2(max_symbols, counter);
duke@435 503 }
duke@435 504 }
duke@435 505 tty->print_cr("Symbol Table:");
coleenp@2497 506 tty->print_cr("Total number of symbols %5d", count);
coleenp@2497 507 tty->print_cr("Total size in memory %5dK",
coleenp@2497 508 (memory_total*HeapWordSize)/1024);
coleenp@2497 509 tty->print_cr("Total counted %5d", symbols_counted);
coleenp@2497 510 tty->print_cr("Total removed %5d", symbols_removed);
coleenp@2497 511 if (symbols_counted > 0) {
coleenp@2497 512 tty->print_cr("Percent removed %3.2f",
coleenp@2497 513 ((float)symbols_removed/(float)symbols_counted)* 100);
coleenp@2497 514 }
coleenp@2497 515 tty->print_cr("Reference counts %5d", Symbol::_total_count);
coleenp@3682 516 tty->print_cr("Symbol arena size %5d used %5d",
coleenp@3682 517 arena()->size_in_bytes(), arena()->used());
coleenp@2497 518 tty->print_cr("Histogram of symbol length:");
duke@435 519 tty->print_cr("%8s %5d", "Total ", total);
duke@435 520 tty->print_cr("%8s %5d", "Maximum", max_symbols);
duke@435 521 tty->print_cr("%8s %3.2f", "Average",
duke@435 522 ((float) total / (float) the_table()->table_size()));
duke@435 523 tty->print_cr("%s", "Histogram:");
duke@435 524 tty->print_cr(" %s %29s", "Length", "Number chains that length");
duke@435 525 for (i = 0; i < results_length; i++) {
duke@435 526 if (results[i] > 0) {
duke@435 527 tty->print_cr("%6d %10d", i, results[i]);
duke@435 528 }
duke@435 529 }
coleenp@2497 530 if (Verbose) {
coleenp@2497 531 int line_length = 70;
coleenp@2497 532 tty->print_cr("%s %30s", " Length", "Number chains that length");
coleenp@2497 533 for (i = 0; i < results_length; i++) {
coleenp@2497 534 if (results[i] > 0) {
coleenp@2497 535 tty->print("%4d", i);
coleenp@2497 536 for (j = 0; (j < results[i]) && (j < line_length); j++) {
coleenp@2497 537 tty->print("%1s", "*");
coleenp@2497 538 }
coleenp@2497 539 if (j == line_length) {
coleenp@2497 540 tty->print("%1s", "+");
coleenp@2497 541 }
coleenp@2497 542 tty->cr();
duke@435 543 }
coleenp@2497 544 }
coleenp@2497 545 }
coleenp@2497 546 tty->print_cr(" %s %d: %d\n", "Number chains longer than",
coleenp@2497 547 results_length, out_of_range);
coleenp@2497 548 }
coleenp@2497 549
coleenp@2497 550 void SymbolTable::print() {
coleenp@2497 551 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 552 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 553 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
coleenp@2497 554 if (entry != NULL) {
coleenp@2497 555 while (entry != NULL) {
coleenp@2497 556 tty->print(PTR_FORMAT " ", entry->literal());
coleenp@2497 557 entry->literal()->print();
coleenp@2497 558 tty->print(" %d", entry->literal()->refcount());
coleenp@2497 559 p = entry->next_addr();
zgu@3900 560 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
duke@435 561 }
duke@435 562 tty->cr();
duke@435 563 }
duke@435 564 }
duke@435 565 }
duke@435 566 #endif // PRODUCT
duke@435 567
duke@435 568 // --------------------------------------------------------------------------
duke@435 569
duke@435 570 #ifdef ASSERT
duke@435 571 class StableMemoryChecker : public StackObj {
duke@435 572 enum { _bufsize = wordSize*4 };
duke@435 573
duke@435 574 address _region;
duke@435 575 jint _size;
duke@435 576 u1 _save_buf[_bufsize];
duke@435 577
duke@435 578 int sample(u1* save_buf) {
duke@435 579 if (_size <= _bufsize) {
duke@435 580 memcpy(save_buf, _region, _size);
duke@435 581 return _size;
duke@435 582 } else {
duke@435 583 // copy head and tail
duke@435 584 memcpy(&save_buf[0], _region, _bufsize/2);
duke@435 585 memcpy(&save_buf[_bufsize/2], _region + _size - _bufsize/2, _bufsize/2);
duke@435 586 return (_bufsize/2)*2;
duke@435 587 }
duke@435 588 }
duke@435 589
duke@435 590 public:
duke@435 591 StableMemoryChecker(const void* region, jint size) {
duke@435 592 _region = (address) region;
duke@435 593 _size = size;
duke@435 594 sample(_save_buf);
duke@435 595 }
duke@435 596
duke@435 597 bool verify() {
duke@435 598 u1 check_buf[sizeof(_save_buf)];
duke@435 599 int check_size = sample(check_buf);
duke@435 600 return (0 == memcmp(_save_buf, check_buf, check_size));
duke@435 601 }
duke@435 602
duke@435 603 void set_region(const void* region) { _region = (address) region; }
duke@435 604 };
duke@435 605 #endif
duke@435 606
duke@435 607
duke@435 608 // --------------------------------------------------------------------------
duke@435 609 StringTable* StringTable::_the_table = NULL;
duke@435 610
coleenp@3865 611 bool StringTable::_needs_rehashing = false;
coleenp@3865 612
coleenp@3865 613 // Pick hashing algorithm
coleenp@3875 614 unsigned int StringTable::hash_string(const jchar* s, int len) {
coleenp@3865 615 return use_alternate_hashcode() ? AltHashing::murmur3_32(seed(), s, len) :
coleenp@3875 616 java_lang_String::to_hash(s, len);
coleenp@3865 617 }
coleenp@3865 618
duke@435 619 oop StringTable::lookup(int index, jchar* name,
duke@435 620 int len, unsigned int hash) {
coleenp@3865 621 int count = 0;
zgu@3900 622 for (HashtableEntry<oop, mtSymbol>* l = bucket(index); l != NULL; l = l->next()) {
coleenp@3865 623 count++;
duke@435 624 if (l->hash() == hash) {
duke@435 625 if (java_lang_String::equals(l->literal(), name, len)) {
duke@435 626 return l->literal();
duke@435 627 }
duke@435 628 }
duke@435 629 }
coleenp@3865 630 // If the bucket size is too deep check if this hash code is insufficient.
zgu@3900 631 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
coleenp@3865 632 _needs_rehashing = check_rehash_table(count);
coleenp@3865 633 }
duke@435 634 return NULL;
duke@435 635 }
duke@435 636
duke@435 637
coleenp@3875 638 oop StringTable::basic_add(int index_arg, Handle string, jchar* name,
coleenp@3865 639 int len, unsigned int hashValue_arg, TRAPS) {
duke@435 640
duke@435 641 assert(java_lang_String::equals(string(), name, len),
duke@435 642 "string must be properly initialized");
coleenp@3875 643 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 644 No_Safepoint_Verifier nsv;
duke@435 645
coleenp@3865 646 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3875 647 // the hash value and index before second lookup.
coleenp@3875 648 unsigned int hashValue;
coleenp@3875 649 int index;
coleenp@3875 650 if (use_alternate_hashcode()) {
coleenp@3875 651 hashValue = hash_string(name, len);
coleenp@3875 652 index = hash_to_index(hashValue);
coleenp@3875 653 } else {
coleenp@3875 654 hashValue = hashValue_arg;
coleenp@3875 655 index = index_arg;
coleenp@3875 656 }
coleenp@3865 657
duke@435 658 // Since look-up was done lock-free, we need to check if another
duke@435 659 // thread beat us in the race to insert the symbol.
duke@435 660
duke@435 661 oop test = lookup(index, name, len, hashValue); // calls lookup(u1*, int)
duke@435 662 if (test != NULL) {
duke@435 663 // Entry already added
duke@435 664 return test;
duke@435 665 }
duke@435 666
zgu@3900 667 HashtableEntry<oop, mtSymbol>* entry = new_entry(hashValue, string());
duke@435 668 add_entry(index, entry);
duke@435 669 return string();
duke@435 670 }
duke@435 671
duke@435 672
coleenp@2497 673 oop StringTable::lookup(Symbol* symbol) {
duke@435 674 ResourceMark rm;
duke@435 675 int length;
duke@435 676 jchar* chars = symbol->as_unicode(length);
coleenp@3865 677 unsigned int hashValue = hash_string(chars, length);
duke@435 678 int index = the_table()->hash_to_index(hashValue);
duke@435 679 return the_table()->lookup(index, chars, length, hashValue);
duke@435 680 }
duke@435 681
duke@435 682
duke@435 683 oop StringTable::intern(Handle string_or_null, jchar* name,
duke@435 684 int len, TRAPS) {
coleenp@3865 685 unsigned int hashValue = hash_string(name, len);
duke@435 686 int index = the_table()->hash_to_index(hashValue);
coleenp@3875 687 oop found_string = the_table()->lookup(index, name, len, hashValue);
duke@435 688
duke@435 689 // Found
coleenp@3875 690 if (found_string != NULL) return found_string;
coleenp@3875 691
coleenp@3875 692 debug_only(StableMemoryChecker smc(name, len * sizeof(name[0])));
coleenp@3875 693 assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),
coleenp@3875 694 "proposed name of symbol must be stable");
coleenp@3875 695
coleenp@3875 696 Handle string;
coleenp@3875 697 // try to reuse the string if possible
coleenp@3875 698 if (!string_or_null.is_null() && (!JavaObjectsInPerm || string_or_null()->is_perm())) {
coleenp@3875 699 string = string_or_null;
coleenp@3875 700 } else {
coleenp@3875 701 string = java_lang_String::create_tenured_from_unicode(name, len, CHECK_NULL);
coleenp@3875 702 }
coleenp@3875 703
coleenp@3875 704 // Grab the StringTable_lock before getting the_table() because it could
coleenp@3875 705 // change at safepoint.
coleenp@3875 706 MutexLocker ml(StringTable_lock, THREAD);
duke@435 707
duke@435 708 // Otherwise, add to symbol to table
coleenp@3875 709 return the_table()->basic_add(index, string, name, len,
duke@435 710 hashValue, CHECK_NULL);
duke@435 711 }
duke@435 712
coleenp@2497 713 oop StringTable::intern(Symbol* symbol, TRAPS) {
duke@435 714 if (symbol == NULL) return NULL;
duke@435 715 ResourceMark rm(THREAD);
duke@435 716 int length;
duke@435 717 jchar* chars = symbol->as_unicode(length);
duke@435 718 Handle string;
duke@435 719 oop result = intern(string, chars, length, CHECK_NULL);
duke@435 720 return result;
duke@435 721 }
duke@435 722
duke@435 723
duke@435 724 oop StringTable::intern(oop string, TRAPS)
duke@435 725 {
duke@435 726 if (string == NULL) return NULL;
duke@435 727 ResourceMark rm(THREAD);
duke@435 728 int length;
duke@435 729 Handle h_string (THREAD, string);
duke@435 730 jchar* chars = java_lang_String::as_unicode_string(string, length);
duke@435 731 oop result = intern(h_string, chars, length, CHECK_NULL);
duke@435 732 return result;
duke@435 733 }
duke@435 734
duke@435 735
duke@435 736 oop StringTable::intern(const char* utf8_string, TRAPS) {
duke@435 737 if (utf8_string == NULL) return NULL;
duke@435 738 ResourceMark rm(THREAD);
duke@435 739 int length = UTF8::unicode_length(utf8_string);
duke@435 740 jchar* chars = NEW_RESOURCE_ARRAY(jchar, length);
duke@435 741 UTF8::convert_to_unicode(utf8_string, chars, length);
duke@435 742 Handle string;
duke@435 743 oop result = intern(string, chars, length, CHECK_NULL);
duke@435 744 return result;
duke@435 745 }
duke@435 746
coleenp@2497 747 void StringTable::unlink(BoolObjectClosure* is_alive) {
coleenp@2497 748 // Readers of the table are unlocked, so we should only be removing
coleenp@2497 749 // entries at a safepoint.
coleenp@2497 750 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@2497 751 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 752 HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 753 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
coleenp@3875 754 while (entry != NULL) {
coleenp@3875 755 // Shared entries are normally at the end of the bucket and if we run into
coleenp@3875 756 // a shared entry, then there is nothing more to remove. However, if we
coleenp@3875 757 // have rehashed the table, then the shared entries are no longer at the
coleenp@3875 758 // end of the bucket.
coleenp@3875 759 if (entry->is_shared() && !use_alternate_hashcode()) {
coleenp@2497 760 break;
coleenp@2497 761 }
coleenp@2497 762 assert(entry->literal() != NULL, "just checking");
coleenp@3875 763 if (entry->is_shared() || is_alive->do_object_b(entry->literal())) {
coleenp@2497 764 p = entry->next_addr();
coleenp@2497 765 } else {
coleenp@2497 766 *p = entry->next();
coleenp@2497 767 the_table()->free_entry(entry);
coleenp@2497 768 }
zgu@3900 769 entry = (HashtableEntry<oop, mtSymbol>*)HashtableEntry<oop, mtSymbol>::make_ptr(*p);
coleenp@2497 770 }
coleenp@2497 771 }
coleenp@2497 772 }
coleenp@2497 773
coleenp@2497 774 void StringTable::oops_do(OopClosure* f) {
coleenp@2497 775 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 776 HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 777 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
coleenp@2497 778 while (entry != NULL) {
coleenp@2497 779 f->do_oop((oop*)entry->literal_addr());
coleenp@2497 780
coleenp@2497 781 // Did the closure remove the literal from the table?
coleenp@2497 782 if (entry->literal() == NULL) {
coleenp@2497 783 assert(!entry->is_shared(), "immutable hashtable entry?");
coleenp@2497 784 *p = entry->next();
coleenp@2497 785 the_table()->free_entry(entry);
coleenp@2497 786 } else {
coleenp@2497 787 p = entry->next_addr();
coleenp@2497 788 }
zgu@3900 789 entry = (HashtableEntry<oop, mtSymbol>*)HashtableEntry<oop, mtSymbol>::make_ptr(*p);
coleenp@2497 790 }
coleenp@2497 791 }
coleenp@2497 792 }
coleenp@2497 793
duke@435 794 void StringTable::verify() {
duke@435 795 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 796 HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
duke@435 797 for ( ; p != NULL; p = p->next()) {
duke@435 798 oop s = p->literal();
duke@435 799 guarantee(s != NULL, "interned string is NULL");
jcoomes@2661 800 guarantee(s->is_perm() || !JavaObjectsInPerm, "interned string not in permspace");
never@2700 801 unsigned int h = java_lang_String::hash_string(s);
duke@435 802 guarantee(p->hash() == h, "broken hash in string table entry");
duke@435 803 guarantee(the_table()->hash_to_index(h) == i,
duke@435 804 "wrong index in string table");
duke@435 805 }
duke@435 806 }
duke@435 807 }
coleenp@3865 808
coleenp@3865 809 void StringTable::dump(outputStream* st) {
coleenp@3865 810 NumberSeq summary;
coleenp@3865 811 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 812 HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
coleenp@3865 813 int count = 0;
coleenp@3865 814 for ( ; p != NULL; p = p->next()) {
coleenp@3865 815 count++;
coleenp@3865 816 }
coleenp@3865 817 summary.add((double)count);
coleenp@3865 818 }
coleenp@3865 819 st->print_cr("StringTable statistics:");
coleenp@3865 820 st->print_cr("Number of buckets : %7d", summary.num());
coleenp@3865 821 st->print_cr("Average bucket size : %7.0f", summary.avg());
coleenp@3865 822 st->print_cr("Variance of bucket size : %7.0f", summary.variance());
coleenp@3865 823 st->print_cr("Std. dev. of bucket size: %7.0f", summary.sd());
coleenp@3865 824 st->print_cr("Maximum bucket size : %7.0f", summary.maximum());
coleenp@3865 825 }
coleenp@3865 826
coleenp@3865 827
coleenp@3865 828 // Create a new table and using alternate hash code, populate the new table
coleenp@3865 829 // with the existing strings. Set flag to use the alternate hash code afterwards.
coleenp@3865 830 void StringTable::rehash_table() {
coleenp@3865 831 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@3875 832 // This should never happen with -Xshare:dump but it might in testing mode.
coleenp@3875 833 if (DumpSharedSpaces) return;
coleenp@3865 834 StringTable* new_table = new StringTable();
coleenp@3865 835
coleenp@3865 836 // Rehash the table
coleenp@3865 837 the_table()->move_to(new_table);
coleenp@3865 838
coleenp@3865 839 // Delete the table and buckets (entries are reused in new table).
coleenp@3865 840 delete _the_table;
coleenp@3865 841 // Don't check if we need rehashing until the table gets unbalanced again.
coleenp@3865 842 // Then rehash with a new global seed.
coleenp@3865 843 _needs_rehashing = false;
coleenp@3865 844 _the_table = new_table;
coleenp@3865 845 }

mercurial