src/share/vm/classfile/symbolTable.cpp

Sat, 01 Sep 2012 13:25:18 -0400

author
coleenp
date
Sat, 01 Sep 2012 13:25:18 -0400
changeset 4037
da91efe96a93
parent 3904
ace99a6ffc83
child 4335
2aa953165ade
permissions
-rw-r--r--

6964458: Reimplement class meta-data storage to use native memory
Summary: Remove PermGen, allocate meta-data in metaspace linked to class loaders, rewrite GC walking, rewrite and rename metadata to be C++ classes
Reviewed-by: jmasa, stefank, never, coleenp, kvn, brutisso, mgerdin, dholmes, jrose, twisti, roland
Contributed-by: jmasa <jon.masamitsu@oracle.com>, stefank <stefan.karlsson@oracle.com>, mgerdin <mikael.gerdin@oracle.com>, never <tom.rodriguez@oracle.com>

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@4037 51
coleenp@4037 52 if (c_heap) {
coleenp@3682 53 // refcount starts as 1
coleenp@4037 54 assert(!DumpSharedSpaces, "never allocate to C heap");
coleenp@3682 55 sym = new (len, THREAD) Symbol(name, len, 1);
coleenp@4037 56 assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted");
coleenp@4037 57 } else {
coleenp@4037 58 if (DumpSharedSpaces) {
coleenp@4037 59 sym = new (len, ClassLoaderData::the_null_class_loader_data(), THREAD) Symbol(name, len, -1);
coleenp@3682 60 } else {
coleenp@3682 61 sym = new (len, arena(), THREAD) Symbol(name, len, -1);
coleenp@3682 62 }
coleenp@4037 63 }
coleenp@2497 64 return sym;
coleenp@2497 65 }
coleenp@2497 66
coleenp@3682 67 void SymbolTable::initialize_symbols(int arena_alloc_size) {
coleenp@3682 68 // Initialize the arena for global symbols, size passed in depends on CDS.
coleenp@3682 69 if (arena_alloc_size == 0) {
zgu@3900 70 _arena = new (mtSymbol) Arena();
coleenp@3682 71 } else {
zgu@3900 72 _arena = new (mtSymbol) Arena(arena_alloc_size);
coleenp@2497 73 }
coleenp@2497 74 }
coleenp@2497 75
coleenp@2497 76 // Call function for all symbols in the symbol table.
coleenp@2497 77 void SymbolTable::symbols_do(SymbolClosure *cl) {
coleenp@2497 78 const int n = the_table()->table_size();
coleenp@2497 79 for (int i = 0; i < n; i++) {
zgu@3900 80 for (HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
coleenp@2497 81 p != NULL;
coleenp@2497 82 p = p->next()) {
coleenp@2497 83 cl->do_symbol(p->literal_addr());
coleenp@2497 84 }
coleenp@2497 85 }
coleenp@2497 86 }
coleenp@2497 87
coleenp@2497 88 int SymbolTable::symbols_removed = 0;
coleenp@2497 89 int SymbolTable::symbols_counted = 0;
coleenp@2497 90
coleenp@2497 91 // Remove unreferenced symbols from the symbol table
coleenp@3682 92 // This is done late during GC.
coleenp@2497 93 void SymbolTable::unlink() {
coleenp@2497 94 int removed = 0;
coleenp@2497 95 int total = 0;
coleenp@2618 96 size_t memory_total = 0;
coleenp@2497 97 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 98 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 99 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
coleenp@3875 100 while (entry != NULL) {
coleenp@3875 101 // Shared entries are normally at the end of the bucket and if we run into
coleenp@3875 102 // a shared entry, then there is nothing more to remove. However, if we
coleenp@3875 103 // have rehashed the table, then the shared entries are no longer at the
coleenp@3875 104 // end of the bucket.
coleenp@3875 105 if (entry->is_shared() && !use_alternate_hashcode()) {
coleenp@2497 106 break;
coleenp@2497 107 }
coleenp@2497 108 Symbol* s = entry->literal();
coleenp@4037 109 memory_total += s->size();
coleenp@2497 110 total++;
coleenp@2497 111 assert(s != NULL, "just checking");
coleenp@2497 112 // If reference count is zero, remove.
coleenp@2497 113 if (s->refcount() == 0) {
coleenp@3875 114 assert(!entry->is_shared(), "shared entries should be kept live");
coleenp@2497 115 delete s;
coleenp@2497 116 removed++;
coleenp@2497 117 *p = entry->next();
coleenp@2497 118 the_table()->free_entry(entry);
coleenp@2497 119 } else {
coleenp@2497 120 p = entry->next_addr();
coleenp@2497 121 }
coleenp@3875 122 // get next entry
zgu@3900 123 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
coleenp@2497 124 }
coleenp@2497 125 }
coleenp@2497 126 symbols_removed += removed;
coleenp@2497 127 symbols_counted += total;
coleenp@2618 128 // Exclude printing for normal PrintGCDetails because people parse
coleenp@2618 129 // this output.
coleenp@2618 130 if (PrintGCDetails && Verbose && WizardMode) {
coleenp@2618 131 gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", total,
coleenp@2497 132 (memory_total*HeapWordSize)/1024);
coleenp@2497 133 }
coleenp@2497 134 }
coleenp@2497 135
coleenp@3865 136 // Create a new table and using alternate hash code, populate the new table
coleenp@3865 137 // with the existing strings. Set flag to use the alternate hash code afterwards.
coleenp@3865 138 void SymbolTable::rehash_table() {
coleenp@3865 139 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@3875 140 // This should never happen with -Xshare:dump but it might in testing mode.
coleenp@3875 141 if (DumpSharedSpaces) return;
coleenp@3865 142 // Create a new symbol table
coleenp@3865 143 SymbolTable* new_table = new SymbolTable();
coleenp@3865 144
coleenp@3865 145 the_table()->move_to(new_table);
coleenp@3865 146
coleenp@3865 147 // Delete the table and buckets (entries are reused in new table).
coleenp@3865 148 delete _the_table;
coleenp@3865 149 // Don't check if we need rehashing until the table gets unbalanced again.
coleenp@3865 150 // Then rehash with a new global seed.
coleenp@3865 151 _needs_rehashing = false;
coleenp@3865 152 _the_table = new_table;
coleenp@3865 153 }
coleenp@2497 154
duke@435 155 // Lookup a symbol in a bucket.
duke@435 156
coleenp@2497 157 Symbol* SymbolTable::lookup(int index, const char* name,
duke@435 158 int len, unsigned int hash) {
coleenp@3865 159 int count = 0;
zgu@3900 160 for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) {
coleenp@3865 161 count++; // count all entries in this bucket, not just ones with same hash
duke@435 162 if (e->hash() == hash) {
coleenp@2497 163 Symbol* sym = e->literal();
duke@435 164 if (sym->equals(name, len)) {
coleenp@2497 165 // something is referencing this symbol now.
coleenp@2497 166 sym->increment_refcount();
duke@435 167 return sym;
duke@435 168 }
duke@435 169 }
duke@435 170 }
coleenp@3865 171 // If the bucket size is too deep check if this hash code is insufficient.
zgu@3900 172 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
coleenp@3865 173 _needs_rehashing = check_rehash_table(count);
coleenp@3865 174 }
duke@435 175 return NULL;
duke@435 176 }
duke@435 177
coleenp@3875 178 // Pick hashing algorithm.
coleenp@3875 179 unsigned int SymbolTable::hash_symbol(const char* s, int len) {
coleenp@3865 180 return use_alternate_hashcode() ?
coleenp@3865 181 AltHashing::murmur3_32(seed(), (const jbyte*)s, len) :
coleenp@3875 182 java_lang_String::to_hash(s, len);
coleenp@3865 183 }
coleenp@3865 184
duke@435 185
duke@435 186 // We take care not to be blocking while holding the
duke@435 187 // SymbolTable_lock. Otherwise, the system might deadlock, since the
duke@435 188 // symboltable is used during compilation (VM_thread) The lock free
duke@435 189 // synchronization is simplified by the fact that we do not delete
duke@435 190 // entries in the symbol table during normal execution (only during
duke@435 191 // safepoints).
duke@435 192
coleenp@2497 193 Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) {
duke@435 194 unsigned int hashValue = hash_symbol(name, len);
duke@435 195 int index = the_table()->hash_to_index(hashValue);
duke@435 196
coleenp@2497 197 Symbol* s = the_table()->lookup(index, name, len, hashValue);
duke@435 198
duke@435 199 // Found
duke@435 200 if (s != NULL) return s;
duke@435 201
coleenp@3875 202 // Grab SymbolTable_lock first.
coleenp@3875 203 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 204
duke@435 205 // Otherwise, add to symbol to table
coleenp@3682 206 return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL);
duke@435 207 }
duke@435 208
coleenp@2497 209 Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) {
duke@435 210 char* buffer;
duke@435 211 int index, len;
duke@435 212 unsigned int hashValue;
duke@435 213 char* name;
duke@435 214 {
duke@435 215 debug_only(No_Safepoint_Verifier nsv;)
duke@435 216
duke@435 217 name = (char*)sym->base() + begin;
duke@435 218 len = end - begin;
duke@435 219 hashValue = hash_symbol(name, len);
duke@435 220 index = the_table()->hash_to_index(hashValue);
coleenp@2497 221 Symbol* s = the_table()->lookup(index, name, len, hashValue);
duke@435 222
duke@435 223 // Found
duke@435 224 if (s != NULL) return s;
duke@435 225 }
duke@435 226
duke@435 227 // Otherwise, add to symbol to table. Copy to a C string first.
duke@435 228 char stack_buf[128];
duke@435 229 ResourceMark rm(THREAD);
duke@435 230 if (len <= 128) {
duke@435 231 buffer = stack_buf;
duke@435 232 } else {
duke@435 233 buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len);
duke@435 234 }
duke@435 235 for (int i=0; i<len; i++) {
duke@435 236 buffer[i] = name[i];
duke@435 237 }
duke@435 238 // Make sure there is no safepoint in the code above since name can't move.
duke@435 239 // We can't include the code in No_Safepoint_Verifier because of the
duke@435 240 // ResourceMark.
duke@435 241
coleenp@3875 242 // Grab SymbolTable_lock first.
coleenp@3875 243 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 244
coleenp@3682 245 return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL);
duke@435 246 }
duke@435 247
coleenp@2497 248 Symbol* SymbolTable::lookup_only(const char* name, int len,
duke@435 249 unsigned int& hash) {
duke@435 250 hash = hash_symbol(name, len);
duke@435 251 int index = the_table()->hash_to_index(hash);
duke@435 252
coleenp@2497 253 Symbol* s = the_table()->lookup(index, name, len, hash);
coleenp@2497 254 return s;
duke@435 255 }
duke@435 256
phh@3427 257 // Look up the address of the literal in the SymbolTable for this Symbol*
phh@3427 258 // Do not create any new symbols
phh@3427 259 // Do not increment the reference count to keep this alive
phh@3427 260 Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){
phh@3427 261 unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length());
phh@3427 262 int index = the_table()->hash_to_index(hash);
phh@3427 263
zgu@3900 264 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) {
phh@3427 265 if (e->hash() == hash) {
phh@3427 266 Symbol* literal_sym = e->literal();
phh@3427 267 if (sym == literal_sym) {
phh@3427 268 return e->literal_addr();
phh@3427 269 }
phh@3427 270 }
phh@3427 271 }
phh@3427 272 return NULL;
phh@3427 273 }
phh@3427 274
jrose@1100 275 // Suggestion: Push unicode-based lookup all the way into the hashing
jrose@1100 276 // and probing logic, so there is no need for convert_to_utf8 until
coleenp@2497 277 // an actual new Symbol* is created.
coleenp@2497 278 Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) {
jrose@1100 279 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
jrose@1100 280 char stack_buf[128];
jrose@1100 281 if (utf8_length < (int) sizeof(stack_buf)) {
jrose@1100 282 char* chars = stack_buf;
jrose@1100 283 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 284 return lookup(chars, utf8_length, THREAD);
jrose@1100 285 } else {
jrose@1100 286 ResourceMark rm(THREAD);
jrose@1100 287 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
jrose@1100 288 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 289 return lookup(chars, utf8_length, THREAD);
jrose@1100 290 }
jrose@1100 291 }
jrose@1100 292
coleenp@2497 293 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
jrose@1100 294 unsigned int& hash) {
jrose@1100 295 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
jrose@1100 296 char stack_buf[128];
jrose@1100 297 if (utf8_length < (int) sizeof(stack_buf)) {
jrose@1100 298 char* chars = stack_buf;
jrose@1100 299 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 300 return lookup_only(chars, utf8_length, hash);
jrose@1100 301 } else {
jrose@1100 302 ResourceMark rm;
jrose@1100 303 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
jrose@1100 304 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 305 return lookup_only(chars, utf8_length, hash);
jrose@1100 306 }
jrose@1100 307 }
jrose@1100 308
coleenp@4037 309 void SymbolTable::add(ClassLoaderData* loader_data, constantPoolHandle cp,
coleenp@3682 310 int names_count,
duke@435 311 const char** names, int* lengths, int* cp_indices,
duke@435 312 unsigned int* hashValues, TRAPS) {
coleenp@3875 313 // Grab SymbolTable_lock first.
coleenp@3875 314 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 315
duke@435 316 SymbolTable* table = the_table();
coleenp@4037 317 bool added = table->basic_add(loader_data, cp, names_count, names, lengths,
duke@435 318 cp_indices, hashValues, CHECK);
duke@435 319 if (!added) {
duke@435 320 // do it the hard way
duke@435 321 for (int i=0; i<names_count; i++) {
duke@435 322 int index = table->hash_to_index(hashValues[i]);
coleenp@4037 323 bool c_heap = !loader_data->is_the_null_class_loader_data();
coleenp@3682 324 Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK);
duke@435 325 cp->symbol_at_put(cp_indices[i], sym);
duke@435 326 }
duke@435 327 }
duke@435 328 }
duke@435 329
coleenp@3682 330 Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) {
coleenp@3682 331 unsigned int hash;
coleenp@3682 332 Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash);
coleenp@3682 333 if (result != NULL) {
coleenp@3682 334 return result;
coleenp@3682 335 }
coleenp@3875 336 // Grab SymbolTable_lock first.
coleenp@3875 337 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 338
coleenp@3682 339 SymbolTable* table = the_table();
coleenp@3682 340 int index = table->hash_to_index(hash);
coleenp@3682 341 return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD);
coleenp@3682 342 }
coleenp@3682 343
coleenp@3875 344 Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len,
coleenp@3865 345 unsigned int hashValue_arg, bool c_heap, TRAPS) {
duke@435 346 assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),
duke@435 347 "proposed name of symbol must be stable");
duke@435 348
coleenp@3875 349 // Don't allow symbols to be created which cannot fit in a Symbol*.
coleenp@3875 350 if (len > Symbol::max_length()) {
coleenp@3875 351 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
coleenp@3875 352 "name is too long to represent");
coleenp@3875 353 }
coleenp@3875 354
coleenp@3875 355 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 356 No_Safepoint_Verifier nsv;
duke@435 357
coleenp@3865 358 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3875 359 // the hash value and index.
coleenp@3875 360 unsigned int hashValue;
coleenp@3875 361 int index;
coleenp@3875 362 if (use_alternate_hashcode()) {
coleenp@3875 363 hashValue = hash_symbol((const char*)name, len);
coleenp@3875 364 index = hash_to_index(hashValue);
coleenp@3875 365 } else {
coleenp@3875 366 hashValue = hashValue_arg;
coleenp@3875 367 index = index_arg;
coleenp@3875 368 }
coleenp@3865 369
duke@435 370 // Since look-up was done lock-free, we need to check if another
duke@435 371 // thread beat us in the race to insert the symbol.
coleenp@2497 372 Symbol* test = lookup(index, (char*)name, len, hashValue);
duke@435 373 if (test != NULL) {
coleenp@3682 374 // A race occurred and another thread introduced the symbol.
coleenp@2497 375 assert(test->refcount() != 0, "lookup should have incremented the count");
duke@435 376 return test;
duke@435 377 }
duke@435 378
coleenp@3682 379 // Create a new symbol.
coleenp@3682 380 Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL);
coleenp@3682 381 assert(sym->equals((char*)name, len), "symbol must be properly initialized");
coleenp@3682 382
zgu@3900 383 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
duke@435 384 add_entry(index, entry);
coleenp@2497 385 return sym;
duke@435 386 }
duke@435 387
coleenp@3682 388 // This version of basic_add adds symbols in batch from the constant pool
coleenp@3682 389 // parsing.
coleenp@4037 390 bool SymbolTable::basic_add(ClassLoaderData* loader_data, constantPoolHandle cp,
coleenp@3682 391 int names_count,
duke@435 392 const char** names, int* lengths,
duke@435 393 int* cp_indices, unsigned int* hashValues,
duke@435 394 TRAPS) {
coleenp@3682 395
coleenp@3682 396 // Check symbol names are not too long. If any are too long, don't add any.
coleenp@3682 397 for (int i = 0; i< names_count; i++) {
coleenp@3682 398 if (lengths[i] > Symbol::max_length()) {
coleenp@3682 399 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
coleenp@3682 400 "name is too long to represent");
coleenp@3682 401 }
duke@435 402 }
duke@435 403
coleenp@3875 404 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 405 No_Safepoint_Verifier nsv;
duke@435 406
coleenp@2497 407 for (int i=0; i<names_count; i++) {
coleenp@3865 408 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3865 409 // the hash value.
coleenp@3875 410 unsigned int hashValue;
coleenp@3875 411 if (use_alternate_hashcode()) {
coleenp@3875 412 hashValue = hash_symbol(names[i], lengths[i]);
coleenp@3875 413 } else {
coleenp@3875 414 hashValue = hashValues[i];
coleenp@3875 415 }
duke@435 416 // Since look-up was done lock-free, we need to check if another
duke@435 417 // thread beat us in the race to insert the symbol.
coleenp@3865 418 int index = hash_to_index(hashValue);
coleenp@3865 419 Symbol* test = lookup(index, names[i], lengths[i], hashValue);
duke@435 420 if (test != NULL) {
twisti@1040 421 // A race occurred and another thread introduced the symbol, this one
duke@435 422 // will be dropped and collected. Use test instead.
duke@435 423 cp->symbol_at_put(cp_indices[i], test);
coleenp@2497 424 assert(test->refcount() != 0, "lookup should have incremented the count");
duke@435 425 } else {
coleenp@3682 426 // Create a new symbol. The null class loader is never unloaded so these
coleenp@3682 427 // are allocated specially in a permanent arena.
coleenp@4037 428 bool c_heap = !loader_data->is_the_null_class_loader_data();
coleenp@3682 429 Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false));
coleenp@3682 430 assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized"); // why wouldn't it be???
zgu@3900 431 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
duke@435 432 add_entry(index, entry);
duke@435 433 cp->symbol_at_put(cp_indices[i], sym);
duke@435 434 }
duke@435 435 }
duke@435 436 return true;
duke@435 437 }
duke@435 438
duke@435 439
duke@435 440 void SymbolTable::verify() {
duke@435 441 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 442 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
duke@435 443 for ( ; p != NULL; p = p->next()) {
coleenp@2497 444 Symbol* s = (Symbol*)(p->literal());
duke@435 445 guarantee(s != NULL, "symbol is NULL");
duke@435 446 unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length());
duke@435 447 guarantee(p->hash() == h, "broken hash in symbol table entry");
duke@435 448 guarantee(the_table()->hash_to_index(h) == i,
duke@435 449 "wrong index in symbol table");
duke@435 450 }
duke@435 451 }
duke@435 452 }
duke@435 453
coleenp@3865 454 void SymbolTable::dump(outputStream* st) {
coleenp@3865 455 NumberSeq summary;
coleenp@3865 456 for (int i = 0; i < the_table()->table_size(); ++i) {
coleenp@3865 457 int count = 0;
zgu@3900 458 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(i);
coleenp@3865 459 e != NULL; e = e->next()) {
coleenp@3865 460 count++;
coleenp@3865 461 }
coleenp@3865 462 summary.add((double)count);
coleenp@3865 463 }
coleenp@3865 464 st->print_cr("SymbolTable statistics:");
coleenp@3865 465 st->print_cr("Number of buckets : %7d", summary.num());
coleenp@3865 466 st->print_cr("Average bucket size : %7.0f", summary.avg());
coleenp@3865 467 st->print_cr("Variance of bucket size : %7.0f", summary.variance());
coleenp@3865 468 st->print_cr("Std. dev. of bucket size: %7.0f", summary.sd());
coleenp@3865 469 st->print_cr("Maximum bucket size : %7.0f", summary.maximum());
coleenp@3865 470 }
coleenp@3865 471
duke@435 472
duke@435 473 //---------------------------------------------------------------------------
duke@435 474 // Non-product code
duke@435 475
duke@435 476 #ifndef PRODUCT
duke@435 477
duke@435 478 void SymbolTable::print_histogram() {
duke@435 479 MutexLocker ml(SymbolTable_lock);
duke@435 480 const int results_length = 100;
duke@435 481 int results[results_length];
duke@435 482 int i,j;
duke@435 483
duke@435 484 // initialize results to zero
duke@435 485 for (j = 0; j < results_length; j++) {
duke@435 486 results[j] = 0;
duke@435 487 }
duke@435 488
duke@435 489 int total = 0;
duke@435 490 int max_symbols = 0;
duke@435 491 int out_of_range = 0;
coleenp@2497 492 int memory_total = 0;
coleenp@2497 493 int count = 0;
duke@435 494 for (i = 0; i < the_table()->table_size(); i++) {
zgu@3900 495 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
duke@435 496 for ( ; p != NULL; p = p->next()) {
coleenp@4037 497 memory_total += p->literal()->size();
coleenp@2497 498 count++;
coleenp@2497 499 int counter = p->literal()->utf8_length();
duke@435 500 total += counter;
duke@435 501 if (counter < results_length) {
duke@435 502 results[counter]++;
duke@435 503 } else {
duke@435 504 out_of_range++;
duke@435 505 }
duke@435 506 max_symbols = MAX2(max_symbols, counter);
duke@435 507 }
duke@435 508 }
duke@435 509 tty->print_cr("Symbol Table:");
coleenp@2497 510 tty->print_cr("Total number of symbols %5d", count);
coleenp@2497 511 tty->print_cr("Total size in memory %5dK",
coleenp@2497 512 (memory_total*HeapWordSize)/1024);
coleenp@2497 513 tty->print_cr("Total counted %5d", symbols_counted);
coleenp@2497 514 tty->print_cr("Total removed %5d", symbols_removed);
coleenp@2497 515 if (symbols_counted > 0) {
coleenp@2497 516 tty->print_cr("Percent removed %3.2f",
coleenp@2497 517 ((float)symbols_removed/(float)symbols_counted)* 100);
coleenp@2497 518 }
coleenp@2497 519 tty->print_cr("Reference counts %5d", Symbol::_total_count);
coleenp@3682 520 tty->print_cr("Symbol arena size %5d used %5d",
coleenp@3682 521 arena()->size_in_bytes(), arena()->used());
coleenp@2497 522 tty->print_cr("Histogram of symbol length:");
duke@435 523 tty->print_cr("%8s %5d", "Total ", total);
duke@435 524 tty->print_cr("%8s %5d", "Maximum", max_symbols);
duke@435 525 tty->print_cr("%8s %3.2f", "Average",
duke@435 526 ((float) total / (float) the_table()->table_size()));
duke@435 527 tty->print_cr("%s", "Histogram:");
duke@435 528 tty->print_cr(" %s %29s", "Length", "Number chains that length");
duke@435 529 for (i = 0; i < results_length; i++) {
duke@435 530 if (results[i] > 0) {
duke@435 531 tty->print_cr("%6d %10d", i, results[i]);
duke@435 532 }
duke@435 533 }
coleenp@2497 534 if (Verbose) {
coleenp@2497 535 int line_length = 70;
coleenp@2497 536 tty->print_cr("%s %30s", " Length", "Number chains that length");
coleenp@2497 537 for (i = 0; i < results_length; i++) {
coleenp@2497 538 if (results[i] > 0) {
coleenp@2497 539 tty->print("%4d", i);
coleenp@2497 540 for (j = 0; (j < results[i]) && (j < line_length); j++) {
coleenp@2497 541 tty->print("%1s", "*");
coleenp@2497 542 }
coleenp@2497 543 if (j == line_length) {
coleenp@2497 544 tty->print("%1s", "+");
coleenp@2497 545 }
coleenp@2497 546 tty->cr();
duke@435 547 }
coleenp@2497 548 }
coleenp@2497 549 }
coleenp@2497 550 tty->print_cr(" %s %d: %d\n", "Number chains longer than",
coleenp@2497 551 results_length, out_of_range);
coleenp@2497 552 }
coleenp@2497 553
coleenp@2497 554 void SymbolTable::print() {
coleenp@2497 555 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 556 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 557 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
coleenp@2497 558 if (entry != NULL) {
coleenp@2497 559 while (entry != NULL) {
coleenp@2497 560 tty->print(PTR_FORMAT " ", entry->literal());
coleenp@2497 561 entry->literal()->print();
coleenp@2497 562 tty->print(" %d", entry->literal()->refcount());
coleenp@2497 563 p = entry->next_addr();
zgu@3900 564 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
duke@435 565 }
duke@435 566 tty->cr();
duke@435 567 }
duke@435 568 }
duke@435 569 }
duke@435 570 #endif // PRODUCT
duke@435 571
duke@435 572 // --------------------------------------------------------------------------
duke@435 573
duke@435 574 #ifdef ASSERT
duke@435 575 class StableMemoryChecker : public StackObj {
duke@435 576 enum { _bufsize = wordSize*4 };
duke@435 577
duke@435 578 address _region;
duke@435 579 jint _size;
duke@435 580 u1 _save_buf[_bufsize];
duke@435 581
duke@435 582 int sample(u1* save_buf) {
duke@435 583 if (_size <= _bufsize) {
duke@435 584 memcpy(save_buf, _region, _size);
duke@435 585 return _size;
duke@435 586 } else {
duke@435 587 // copy head and tail
duke@435 588 memcpy(&save_buf[0], _region, _bufsize/2);
duke@435 589 memcpy(&save_buf[_bufsize/2], _region + _size - _bufsize/2, _bufsize/2);
duke@435 590 return (_bufsize/2)*2;
duke@435 591 }
duke@435 592 }
duke@435 593
duke@435 594 public:
duke@435 595 StableMemoryChecker(const void* region, jint size) {
duke@435 596 _region = (address) region;
duke@435 597 _size = size;
duke@435 598 sample(_save_buf);
duke@435 599 }
duke@435 600
duke@435 601 bool verify() {
duke@435 602 u1 check_buf[sizeof(_save_buf)];
duke@435 603 int check_size = sample(check_buf);
duke@435 604 return (0 == memcmp(_save_buf, check_buf, check_size));
duke@435 605 }
duke@435 606
duke@435 607 void set_region(const void* region) { _region = (address) region; }
duke@435 608 };
duke@435 609 #endif
duke@435 610
duke@435 611
duke@435 612 // --------------------------------------------------------------------------
duke@435 613 StringTable* StringTable::_the_table = NULL;
duke@435 614
coleenp@3865 615 bool StringTable::_needs_rehashing = false;
coleenp@3865 616
coleenp@3865 617 // Pick hashing algorithm
coleenp@3875 618 unsigned int StringTable::hash_string(const jchar* s, int len) {
coleenp@3865 619 return use_alternate_hashcode() ? AltHashing::murmur3_32(seed(), s, len) :
coleenp@3875 620 java_lang_String::to_hash(s, len);
coleenp@3865 621 }
coleenp@3865 622
duke@435 623 oop StringTable::lookup(int index, jchar* name,
duke@435 624 int len, unsigned int hash) {
coleenp@3865 625 int count = 0;
zgu@3900 626 for (HashtableEntry<oop, mtSymbol>* l = bucket(index); l != NULL; l = l->next()) {
coleenp@3865 627 count++;
duke@435 628 if (l->hash() == hash) {
duke@435 629 if (java_lang_String::equals(l->literal(), name, len)) {
duke@435 630 return l->literal();
duke@435 631 }
duke@435 632 }
duke@435 633 }
coleenp@3865 634 // If the bucket size is too deep check if this hash code is insufficient.
zgu@3900 635 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
coleenp@3865 636 _needs_rehashing = check_rehash_table(count);
coleenp@3865 637 }
duke@435 638 return NULL;
duke@435 639 }
duke@435 640
duke@435 641
coleenp@3875 642 oop StringTable::basic_add(int index_arg, Handle string, jchar* name,
coleenp@3865 643 int len, unsigned int hashValue_arg, TRAPS) {
duke@435 644
duke@435 645 assert(java_lang_String::equals(string(), name, len),
duke@435 646 "string must be properly initialized");
coleenp@3875 647 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 648 No_Safepoint_Verifier nsv;
duke@435 649
coleenp@3865 650 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3875 651 // the hash value and index before second lookup.
coleenp@3875 652 unsigned int hashValue;
coleenp@3875 653 int index;
coleenp@3875 654 if (use_alternate_hashcode()) {
coleenp@3875 655 hashValue = hash_string(name, len);
coleenp@3875 656 index = hash_to_index(hashValue);
coleenp@3875 657 } else {
coleenp@3875 658 hashValue = hashValue_arg;
coleenp@3875 659 index = index_arg;
coleenp@3875 660 }
coleenp@3865 661
duke@435 662 // Since look-up was done lock-free, we need to check if another
duke@435 663 // thread beat us in the race to insert the symbol.
duke@435 664
duke@435 665 oop test = lookup(index, name, len, hashValue); // calls lookup(u1*, int)
duke@435 666 if (test != NULL) {
duke@435 667 // Entry already added
duke@435 668 return test;
duke@435 669 }
duke@435 670
zgu@3900 671 HashtableEntry<oop, mtSymbol>* entry = new_entry(hashValue, string());
duke@435 672 add_entry(index, entry);
duke@435 673 return string();
duke@435 674 }
duke@435 675
duke@435 676
coleenp@2497 677 oop StringTable::lookup(Symbol* symbol) {
duke@435 678 ResourceMark rm;
duke@435 679 int length;
duke@435 680 jchar* chars = symbol->as_unicode(length);
coleenp@3865 681 unsigned int hashValue = hash_string(chars, length);
duke@435 682 int index = the_table()->hash_to_index(hashValue);
duke@435 683 return the_table()->lookup(index, chars, length, hashValue);
duke@435 684 }
duke@435 685
duke@435 686
duke@435 687 oop StringTable::intern(Handle string_or_null, jchar* name,
duke@435 688 int len, TRAPS) {
coleenp@3865 689 unsigned int hashValue = hash_string(name, len);
duke@435 690 int index = the_table()->hash_to_index(hashValue);
coleenp@3875 691 oop found_string = the_table()->lookup(index, name, len, hashValue);
duke@435 692
duke@435 693 // Found
coleenp@3875 694 if (found_string != NULL) return found_string;
coleenp@3875 695
coleenp@3875 696 debug_only(StableMemoryChecker smc(name, len * sizeof(name[0])));
coleenp@3875 697 assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),
coleenp@3875 698 "proposed name of symbol must be stable");
coleenp@3875 699
coleenp@3875 700 Handle string;
coleenp@3875 701 // try to reuse the string if possible
coleenp@4037 702 if (!string_or_null.is_null()) {
coleenp@3875 703 string = string_or_null;
coleenp@3875 704 } else {
coleenp@4037 705 string = java_lang_String::create_from_unicode(name, len, CHECK_NULL);
coleenp@3875 706 }
coleenp@3875 707
coleenp@3875 708 // Grab the StringTable_lock before getting the_table() because it could
coleenp@3875 709 // change at safepoint.
coleenp@3875 710 MutexLocker ml(StringTable_lock, THREAD);
duke@435 711
duke@435 712 // Otherwise, add to symbol to table
coleenp@3875 713 return the_table()->basic_add(index, string, name, len,
duke@435 714 hashValue, CHECK_NULL);
duke@435 715 }
duke@435 716
coleenp@2497 717 oop StringTable::intern(Symbol* symbol, TRAPS) {
duke@435 718 if (symbol == NULL) return NULL;
duke@435 719 ResourceMark rm(THREAD);
duke@435 720 int length;
duke@435 721 jchar* chars = symbol->as_unicode(length);
duke@435 722 Handle string;
duke@435 723 oop result = intern(string, chars, length, CHECK_NULL);
duke@435 724 return result;
duke@435 725 }
duke@435 726
duke@435 727
duke@435 728 oop StringTable::intern(oop string, TRAPS)
duke@435 729 {
duke@435 730 if (string == NULL) return NULL;
duke@435 731 ResourceMark rm(THREAD);
duke@435 732 int length;
duke@435 733 Handle h_string (THREAD, string);
duke@435 734 jchar* chars = java_lang_String::as_unicode_string(string, length);
duke@435 735 oop result = intern(h_string, chars, length, CHECK_NULL);
duke@435 736 return result;
duke@435 737 }
duke@435 738
duke@435 739
duke@435 740 oop StringTable::intern(const char* utf8_string, TRAPS) {
duke@435 741 if (utf8_string == NULL) return NULL;
duke@435 742 ResourceMark rm(THREAD);
duke@435 743 int length = UTF8::unicode_length(utf8_string);
duke@435 744 jchar* chars = NEW_RESOURCE_ARRAY(jchar, length);
duke@435 745 UTF8::convert_to_unicode(utf8_string, chars, length);
duke@435 746 Handle string;
duke@435 747 oop result = intern(string, chars, length, CHECK_NULL);
duke@435 748 return result;
duke@435 749 }
duke@435 750
coleenp@2497 751 void StringTable::unlink(BoolObjectClosure* is_alive) {
coleenp@2497 752 // Readers of the table are unlocked, so we should only be removing
coleenp@2497 753 // entries at a safepoint.
coleenp@2497 754 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@2497 755 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 756 HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 757 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
coleenp@3875 758 while (entry != NULL) {
coleenp@3875 759 // Shared entries are normally at the end of the bucket and if we run into
coleenp@3875 760 // a shared entry, then there is nothing more to remove. However, if we
coleenp@3875 761 // have rehashed the table, then the shared entries are no longer at the
coleenp@3875 762 // end of the bucket.
coleenp@3875 763 if (entry->is_shared() && !use_alternate_hashcode()) {
coleenp@2497 764 break;
coleenp@2497 765 }
coleenp@2497 766 assert(entry->literal() != NULL, "just checking");
coleenp@3875 767 if (entry->is_shared() || is_alive->do_object_b(entry->literal())) {
coleenp@2497 768 p = entry->next_addr();
coleenp@2497 769 } else {
coleenp@2497 770 *p = entry->next();
coleenp@2497 771 the_table()->free_entry(entry);
coleenp@2497 772 }
zgu@3900 773 entry = (HashtableEntry<oop, mtSymbol>*)HashtableEntry<oop, mtSymbol>::make_ptr(*p);
coleenp@2497 774 }
coleenp@2497 775 }
coleenp@2497 776 }
coleenp@2497 777
coleenp@2497 778 void StringTable::oops_do(OopClosure* f) {
coleenp@2497 779 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 780 HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 781 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
coleenp@2497 782 while (entry != NULL) {
coleenp@2497 783 f->do_oop((oop*)entry->literal_addr());
coleenp@2497 784
coleenp@2497 785 // Did the closure remove the literal from the table?
coleenp@2497 786 if (entry->literal() == NULL) {
coleenp@2497 787 assert(!entry->is_shared(), "immutable hashtable entry?");
coleenp@2497 788 *p = entry->next();
coleenp@2497 789 the_table()->free_entry(entry);
coleenp@2497 790 } else {
coleenp@2497 791 p = entry->next_addr();
coleenp@2497 792 }
zgu@3900 793 entry = (HashtableEntry<oop, mtSymbol>*)HashtableEntry<oop, mtSymbol>::make_ptr(*p);
coleenp@2497 794 }
coleenp@2497 795 }
coleenp@2497 796 }
coleenp@2497 797
duke@435 798 void StringTable::verify() {
duke@435 799 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 800 HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
duke@435 801 for ( ; p != NULL; p = p->next()) {
duke@435 802 oop s = p->literal();
duke@435 803 guarantee(s != NULL, "interned string is NULL");
never@2700 804 unsigned int h = java_lang_String::hash_string(s);
duke@435 805 guarantee(p->hash() == h, "broken hash in string table entry");
duke@435 806 guarantee(the_table()->hash_to_index(h) == i,
duke@435 807 "wrong index in string table");
duke@435 808 }
duke@435 809 }
duke@435 810 }
coleenp@3865 811
coleenp@3865 812 void StringTable::dump(outputStream* st) {
coleenp@3865 813 NumberSeq summary;
coleenp@3865 814 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 815 HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
coleenp@3865 816 int count = 0;
coleenp@3865 817 for ( ; p != NULL; p = p->next()) {
coleenp@3865 818 count++;
coleenp@3865 819 }
coleenp@3865 820 summary.add((double)count);
coleenp@3865 821 }
coleenp@3865 822 st->print_cr("StringTable statistics:");
coleenp@3865 823 st->print_cr("Number of buckets : %7d", summary.num());
coleenp@3865 824 st->print_cr("Average bucket size : %7.0f", summary.avg());
coleenp@3865 825 st->print_cr("Variance of bucket size : %7.0f", summary.variance());
coleenp@3865 826 st->print_cr("Std. dev. of bucket size: %7.0f", summary.sd());
coleenp@3865 827 st->print_cr("Maximum bucket size : %7.0f", summary.maximum());
coleenp@3865 828 }
coleenp@3865 829
coleenp@3865 830
coleenp@3865 831 // Create a new table and using alternate hash code, populate the new table
coleenp@3865 832 // with the existing strings. Set flag to use the alternate hash code afterwards.
coleenp@3865 833 void StringTable::rehash_table() {
coleenp@3865 834 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@3875 835 // This should never happen with -Xshare:dump but it might in testing mode.
coleenp@3875 836 if (DumpSharedSpaces) return;
coleenp@3865 837 StringTable* new_table = new StringTable();
coleenp@3865 838
coleenp@3865 839 // Rehash the table
coleenp@3865 840 the_table()->move_to(new_table);
coleenp@3865 841
coleenp@3865 842 // Delete the table and buckets (entries are reused in new table).
coleenp@3865 843 delete _the_table;
coleenp@3865 844 // Don't check if we need rehashing until the table gets unbalanced again.
coleenp@3865 845 // Then rehash with a new global seed.
coleenp@3865 846 _needs_rehashing = false;
coleenp@3865 847 _the_table = new_table;
coleenp@3865 848 }

mercurial