src/share/vm/classfile/symbolTable.cpp

Wed, 13 Mar 2013 15:15:56 -0400

author
coleenp
date
Wed, 13 Mar 2013 15:15:56 -0400
changeset 4718
0ede345ec7c9
parent 4335
2aa953165ade
child 4850
ede380e13960
permissions
-rw-r--r--

8009829: CDS: JDK JPRT test fails crash in Symbol::equals()
Summary: -Xshare:dump was creating a Symbol in C_heap. There's an assert there that jdk jprt wasn't hitting because it was only done in product
Reviewed-by: dholmes, hseigel, iklam

duke@435 1 /*
coleenp@4718 2 * Copyright (c) 1997, 2013, 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@4718 52 if (DumpSharedSpaces) {
coleenp@4718 53 // Allocate all symbols to CLD shared metaspace
coleenp@4718 54 sym = new (len, ClassLoaderData::the_null_class_loader_data(), THREAD) Symbol(name, len, -1);
coleenp@4718 55 } else if (c_heap) {
coleenp@3682 56 // refcount starts as 1
coleenp@3682 57 sym = new (len, THREAD) Symbol(name, len, 1);
coleenp@4037 58 assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted");
coleenp@4037 59 } else {
coleenp@4718 60 // Allocate to global arena
coleenp@3682 61 sym = new (len, arena(), THREAD) Symbol(name, len, -1);
coleenp@3682 62 }
coleenp@2497 63 return sym;
coleenp@2497 64 }
coleenp@2497 65
coleenp@3682 66 void SymbolTable::initialize_symbols(int arena_alloc_size) {
coleenp@3682 67 // Initialize the arena for global symbols, size passed in depends on CDS.
coleenp@3682 68 if (arena_alloc_size == 0) {
zgu@3900 69 _arena = new (mtSymbol) Arena();
coleenp@3682 70 } else {
zgu@3900 71 _arena = new (mtSymbol) Arena(arena_alloc_size);
coleenp@2497 72 }
coleenp@2497 73 }
coleenp@2497 74
coleenp@2497 75 // Call function for all symbols in the symbol table.
coleenp@2497 76 void SymbolTable::symbols_do(SymbolClosure *cl) {
coleenp@2497 77 const int n = the_table()->table_size();
coleenp@2497 78 for (int i = 0; i < n; i++) {
zgu@3900 79 for (HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
coleenp@2497 80 p != NULL;
coleenp@2497 81 p = p->next()) {
coleenp@2497 82 cl->do_symbol(p->literal_addr());
coleenp@2497 83 }
coleenp@2497 84 }
coleenp@2497 85 }
coleenp@2497 86
coleenp@2497 87 int SymbolTable::symbols_removed = 0;
coleenp@2497 88 int SymbolTable::symbols_counted = 0;
coleenp@2497 89
coleenp@2497 90 // Remove unreferenced symbols from the symbol table
coleenp@3682 91 // This is done late during GC.
coleenp@2497 92 void SymbolTable::unlink() {
coleenp@2497 93 int removed = 0;
coleenp@2497 94 int total = 0;
coleenp@2618 95 size_t memory_total = 0;
coleenp@2497 96 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 97 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 98 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
coleenp@3875 99 while (entry != NULL) {
coleenp@3875 100 // Shared entries are normally at the end of the bucket and if we run into
coleenp@3875 101 // a shared entry, then there is nothing more to remove. However, if we
coleenp@3875 102 // have rehashed the table, then the shared entries are no longer at the
coleenp@3875 103 // end of the bucket.
coleenp@3875 104 if (entry->is_shared() && !use_alternate_hashcode()) {
coleenp@2497 105 break;
coleenp@2497 106 }
coleenp@2497 107 Symbol* s = entry->literal();
coleenp@4037 108 memory_total += s->size();
coleenp@2497 109 total++;
coleenp@2497 110 assert(s != NULL, "just checking");
coleenp@2497 111 // If reference count is zero, remove.
coleenp@2497 112 if (s->refcount() == 0) {
coleenp@3875 113 assert(!entry->is_shared(), "shared entries should be kept live");
coleenp@2497 114 delete s;
coleenp@2497 115 removed++;
coleenp@2497 116 *p = entry->next();
coleenp@2497 117 the_table()->free_entry(entry);
coleenp@2497 118 } else {
coleenp@2497 119 p = entry->next_addr();
coleenp@2497 120 }
coleenp@3875 121 // get next entry
zgu@3900 122 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
coleenp@2497 123 }
coleenp@2497 124 }
coleenp@2497 125 symbols_removed += removed;
coleenp@2497 126 symbols_counted += total;
coleenp@2618 127 // Exclude printing for normal PrintGCDetails because people parse
coleenp@2618 128 // this output.
coleenp@2618 129 if (PrintGCDetails && Verbose && WizardMode) {
coleenp@2618 130 gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", total,
coleenp@2497 131 (memory_total*HeapWordSize)/1024);
coleenp@2497 132 }
coleenp@2497 133 }
coleenp@2497 134
coleenp@3865 135 // Create a new table and using alternate hash code, populate the new table
coleenp@3865 136 // with the existing strings. Set flag to use the alternate hash code afterwards.
coleenp@3865 137 void SymbolTable::rehash_table() {
coleenp@3865 138 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@3875 139 // This should never happen with -Xshare:dump but it might in testing mode.
coleenp@3875 140 if (DumpSharedSpaces) return;
coleenp@3865 141 // Create a new symbol table
coleenp@3865 142 SymbolTable* new_table = new SymbolTable();
coleenp@3865 143
coleenp@3865 144 the_table()->move_to(new_table);
coleenp@3865 145
coleenp@3865 146 // Delete the table and buckets (entries are reused in new table).
coleenp@3865 147 delete _the_table;
coleenp@3865 148 // Don't check if we need rehashing until the table gets unbalanced again.
coleenp@3865 149 // Then rehash with a new global seed.
coleenp@3865 150 _needs_rehashing = false;
coleenp@3865 151 _the_table = new_table;
coleenp@3865 152 }
coleenp@2497 153
duke@435 154 // Lookup a symbol in a bucket.
duke@435 155
coleenp@2497 156 Symbol* SymbolTable::lookup(int index, const char* name,
duke@435 157 int len, unsigned int hash) {
coleenp@3865 158 int count = 0;
zgu@3900 159 for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) {
coleenp@3865 160 count++; // count all entries in this bucket, not just ones with same hash
duke@435 161 if (e->hash() == hash) {
coleenp@2497 162 Symbol* sym = e->literal();
duke@435 163 if (sym->equals(name, len)) {
coleenp@2497 164 // something is referencing this symbol now.
coleenp@2497 165 sym->increment_refcount();
duke@435 166 return sym;
duke@435 167 }
duke@435 168 }
duke@435 169 }
coleenp@3865 170 // If the bucket size is too deep check if this hash code is insufficient.
zgu@3900 171 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
coleenp@3865 172 _needs_rehashing = check_rehash_table(count);
coleenp@3865 173 }
duke@435 174 return NULL;
duke@435 175 }
duke@435 176
coleenp@3875 177 // Pick hashing algorithm.
coleenp@3875 178 unsigned int SymbolTable::hash_symbol(const char* s, int len) {
coleenp@3865 179 return use_alternate_hashcode() ?
coleenp@3865 180 AltHashing::murmur3_32(seed(), (const jbyte*)s, len) :
brutisso@4335 181 java_lang_String::hash_code(s, len);
coleenp@3865 182 }
coleenp@3865 183
duke@435 184
duke@435 185 // We take care not to be blocking while holding the
duke@435 186 // SymbolTable_lock. Otherwise, the system might deadlock, since the
duke@435 187 // symboltable is used during compilation (VM_thread) The lock free
duke@435 188 // synchronization is simplified by the fact that we do not delete
duke@435 189 // entries in the symbol table during normal execution (only during
duke@435 190 // safepoints).
duke@435 191
coleenp@2497 192 Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) {
duke@435 193 unsigned int hashValue = hash_symbol(name, len);
duke@435 194 int index = the_table()->hash_to_index(hashValue);
duke@435 195
coleenp@2497 196 Symbol* s = the_table()->lookup(index, name, len, hashValue);
duke@435 197
duke@435 198 // Found
duke@435 199 if (s != NULL) return s;
duke@435 200
coleenp@3875 201 // Grab SymbolTable_lock first.
coleenp@3875 202 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 203
duke@435 204 // Otherwise, add to symbol to table
coleenp@3682 205 return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL);
duke@435 206 }
duke@435 207
coleenp@2497 208 Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) {
duke@435 209 char* buffer;
duke@435 210 int index, len;
duke@435 211 unsigned int hashValue;
duke@435 212 char* name;
duke@435 213 {
duke@435 214 debug_only(No_Safepoint_Verifier nsv;)
duke@435 215
duke@435 216 name = (char*)sym->base() + begin;
duke@435 217 len = end - begin;
duke@435 218 hashValue = hash_symbol(name, len);
duke@435 219 index = the_table()->hash_to_index(hashValue);
coleenp@2497 220 Symbol* s = the_table()->lookup(index, name, len, hashValue);
duke@435 221
duke@435 222 // Found
duke@435 223 if (s != NULL) return s;
duke@435 224 }
duke@435 225
duke@435 226 // Otherwise, add to symbol to table. Copy to a C string first.
duke@435 227 char stack_buf[128];
duke@435 228 ResourceMark rm(THREAD);
duke@435 229 if (len <= 128) {
duke@435 230 buffer = stack_buf;
duke@435 231 } else {
duke@435 232 buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len);
duke@435 233 }
duke@435 234 for (int i=0; i<len; i++) {
duke@435 235 buffer[i] = name[i];
duke@435 236 }
duke@435 237 // Make sure there is no safepoint in the code above since name can't move.
duke@435 238 // We can't include the code in No_Safepoint_Verifier because of the
duke@435 239 // ResourceMark.
duke@435 240
coleenp@3875 241 // Grab SymbolTable_lock first.
coleenp@3875 242 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 243
coleenp@3682 244 return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL);
duke@435 245 }
duke@435 246
coleenp@2497 247 Symbol* SymbolTable::lookup_only(const char* name, int len,
duke@435 248 unsigned int& hash) {
duke@435 249 hash = hash_symbol(name, len);
duke@435 250 int index = the_table()->hash_to_index(hash);
duke@435 251
coleenp@2497 252 Symbol* s = the_table()->lookup(index, name, len, hash);
coleenp@2497 253 return s;
duke@435 254 }
duke@435 255
phh@3427 256 // Look up the address of the literal in the SymbolTable for this Symbol*
phh@3427 257 // Do not create any new symbols
phh@3427 258 // Do not increment the reference count to keep this alive
phh@3427 259 Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){
phh@3427 260 unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length());
phh@3427 261 int index = the_table()->hash_to_index(hash);
phh@3427 262
zgu@3900 263 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) {
phh@3427 264 if (e->hash() == hash) {
phh@3427 265 Symbol* literal_sym = e->literal();
phh@3427 266 if (sym == literal_sym) {
phh@3427 267 return e->literal_addr();
phh@3427 268 }
phh@3427 269 }
phh@3427 270 }
phh@3427 271 return NULL;
phh@3427 272 }
phh@3427 273
jrose@1100 274 // Suggestion: Push unicode-based lookup all the way into the hashing
jrose@1100 275 // and probing logic, so there is no need for convert_to_utf8 until
coleenp@2497 276 // an actual new Symbol* is created.
coleenp@2497 277 Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) {
jrose@1100 278 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
jrose@1100 279 char stack_buf[128];
jrose@1100 280 if (utf8_length < (int) sizeof(stack_buf)) {
jrose@1100 281 char* chars = stack_buf;
jrose@1100 282 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 283 return lookup(chars, utf8_length, THREAD);
jrose@1100 284 } else {
jrose@1100 285 ResourceMark rm(THREAD);
jrose@1100 286 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
jrose@1100 287 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 288 return lookup(chars, utf8_length, THREAD);
jrose@1100 289 }
jrose@1100 290 }
jrose@1100 291
coleenp@2497 292 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
jrose@1100 293 unsigned int& hash) {
jrose@1100 294 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
jrose@1100 295 char stack_buf[128];
jrose@1100 296 if (utf8_length < (int) sizeof(stack_buf)) {
jrose@1100 297 char* chars = stack_buf;
jrose@1100 298 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 299 return lookup_only(chars, utf8_length, hash);
jrose@1100 300 } else {
jrose@1100 301 ResourceMark rm;
jrose@1100 302 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
jrose@1100 303 UNICODE::convert_to_utf8(name, utf16_length, chars);
jrose@1100 304 return lookup_only(chars, utf8_length, hash);
jrose@1100 305 }
jrose@1100 306 }
jrose@1100 307
coleenp@4037 308 void SymbolTable::add(ClassLoaderData* loader_data, constantPoolHandle cp,
coleenp@3682 309 int names_count,
duke@435 310 const char** names, int* lengths, int* cp_indices,
duke@435 311 unsigned int* hashValues, TRAPS) {
coleenp@3875 312 // Grab SymbolTable_lock first.
coleenp@3875 313 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 314
duke@435 315 SymbolTable* table = the_table();
coleenp@4037 316 bool added = table->basic_add(loader_data, cp, names_count, names, lengths,
duke@435 317 cp_indices, hashValues, CHECK);
duke@435 318 if (!added) {
duke@435 319 // do it the hard way
duke@435 320 for (int i=0; i<names_count; i++) {
duke@435 321 int index = table->hash_to_index(hashValues[i]);
coleenp@4037 322 bool c_heap = !loader_data->is_the_null_class_loader_data();
coleenp@3682 323 Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK);
duke@435 324 cp->symbol_at_put(cp_indices[i], sym);
duke@435 325 }
duke@435 326 }
duke@435 327 }
duke@435 328
coleenp@3682 329 Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) {
coleenp@3682 330 unsigned int hash;
coleenp@3682 331 Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash);
coleenp@3682 332 if (result != NULL) {
coleenp@3682 333 return result;
coleenp@3682 334 }
coleenp@3875 335 // Grab SymbolTable_lock first.
coleenp@3875 336 MutexLocker ml(SymbolTable_lock, THREAD);
coleenp@3875 337
coleenp@3682 338 SymbolTable* table = the_table();
coleenp@3682 339 int index = table->hash_to_index(hash);
coleenp@3682 340 return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD);
coleenp@3682 341 }
coleenp@3682 342
coleenp@3875 343 Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len,
coleenp@3865 344 unsigned int hashValue_arg, bool c_heap, TRAPS) {
duke@435 345 assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),
duke@435 346 "proposed name of symbol must be stable");
duke@435 347
coleenp@3875 348 // Don't allow symbols to be created which cannot fit in a Symbol*.
coleenp@3875 349 if (len > Symbol::max_length()) {
coleenp@3875 350 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
coleenp@3875 351 "name is too long to represent");
coleenp@3875 352 }
coleenp@3875 353
coleenp@3875 354 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 355 No_Safepoint_Verifier nsv;
duke@435 356
coleenp@3865 357 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3875 358 // the hash value and index.
coleenp@3875 359 unsigned int hashValue;
coleenp@3875 360 int index;
coleenp@3875 361 if (use_alternate_hashcode()) {
coleenp@3875 362 hashValue = hash_symbol((const char*)name, len);
coleenp@3875 363 index = hash_to_index(hashValue);
coleenp@3875 364 } else {
coleenp@3875 365 hashValue = hashValue_arg;
coleenp@3875 366 index = index_arg;
coleenp@3875 367 }
coleenp@3865 368
duke@435 369 // Since look-up was done lock-free, we need to check if another
duke@435 370 // thread beat us in the race to insert the symbol.
coleenp@2497 371 Symbol* test = lookup(index, (char*)name, len, hashValue);
duke@435 372 if (test != NULL) {
coleenp@3682 373 // A race occurred and another thread introduced the symbol.
coleenp@2497 374 assert(test->refcount() != 0, "lookup should have incremented the count");
duke@435 375 return test;
duke@435 376 }
duke@435 377
coleenp@3682 378 // Create a new symbol.
coleenp@3682 379 Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL);
coleenp@3682 380 assert(sym->equals((char*)name, len), "symbol must be properly initialized");
coleenp@3682 381
zgu@3900 382 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
duke@435 383 add_entry(index, entry);
coleenp@2497 384 return sym;
duke@435 385 }
duke@435 386
coleenp@3682 387 // This version of basic_add adds symbols in batch from the constant pool
coleenp@3682 388 // parsing.
coleenp@4037 389 bool SymbolTable::basic_add(ClassLoaderData* loader_data, constantPoolHandle cp,
coleenp@3682 390 int names_count,
duke@435 391 const char** names, int* lengths,
duke@435 392 int* cp_indices, unsigned int* hashValues,
duke@435 393 TRAPS) {
coleenp@3682 394
coleenp@3682 395 // Check symbol names are not too long. If any are too long, don't add any.
coleenp@3682 396 for (int i = 0; i< names_count; i++) {
coleenp@3682 397 if (lengths[i] > Symbol::max_length()) {
coleenp@3682 398 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
coleenp@3682 399 "name is too long to represent");
coleenp@3682 400 }
duke@435 401 }
duke@435 402
coleenp@3875 403 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 404 No_Safepoint_Verifier nsv;
duke@435 405
coleenp@2497 406 for (int i=0; i<names_count; i++) {
coleenp@3865 407 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3865 408 // the hash value.
coleenp@3875 409 unsigned int hashValue;
coleenp@3875 410 if (use_alternate_hashcode()) {
coleenp@3875 411 hashValue = hash_symbol(names[i], lengths[i]);
coleenp@3875 412 } else {
coleenp@3875 413 hashValue = hashValues[i];
coleenp@3875 414 }
duke@435 415 // Since look-up was done lock-free, we need to check if another
duke@435 416 // thread beat us in the race to insert the symbol.
coleenp@3865 417 int index = hash_to_index(hashValue);
coleenp@3865 418 Symbol* test = lookup(index, names[i], lengths[i], hashValue);
duke@435 419 if (test != NULL) {
twisti@1040 420 // A race occurred and another thread introduced the symbol, this one
duke@435 421 // will be dropped and collected. Use test instead.
duke@435 422 cp->symbol_at_put(cp_indices[i], test);
coleenp@2497 423 assert(test->refcount() != 0, "lookup should have incremented the count");
duke@435 424 } else {
coleenp@3682 425 // Create a new symbol. The null class loader is never unloaded so these
coleenp@3682 426 // are allocated specially in a permanent arena.
coleenp@4037 427 bool c_heap = !loader_data->is_the_null_class_loader_data();
coleenp@3682 428 Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false));
coleenp@3682 429 assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized"); // why wouldn't it be???
zgu@3900 430 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
duke@435 431 add_entry(index, entry);
duke@435 432 cp->symbol_at_put(cp_indices[i], sym);
duke@435 433 }
duke@435 434 }
duke@435 435 return true;
duke@435 436 }
duke@435 437
duke@435 438
duke@435 439 void SymbolTable::verify() {
duke@435 440 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 441 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
duke@435 442 for ( ; p != NULL; p = p->next()) {
coleenp@2497 443 Symbol* s = (Symbol*)(p->literal());
duke@435 444 guarantee(s != NULL, "symbol is NULL");
duke@435 445 unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length());
duke@435 446 guarantee(p->hash() == h, "broken hash in symbol table entry");
duke@435 447 guarantee(the_table()->hash_to_index(h) == i,
duke@435 448 "wrong index in symbol table");
duke@435 449 }
duke@435 450 }
duke@435 451 }
duke@435 452
coleenp@3865 453 void SymbolTable::dump(outputStream* st) {
coleenp@3865 454 NumberSeq summary;
coleenp@3865 455 for (int i = 0; i < the_table()->table_size(); ++i) {
coleenp@3865 456 int count = 0;
zgu@3900 457 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(i);
coleenp@3865 458 e != NULL; e = e->next()) {
coleenp@3865 459 count++;
coleenp@3865 460 }
coleenp@3865 461 summary.add((double)count);
coleenp@3865 462 }
coleenp@3865 463 st->print_cr("SymbolTable statistics:");
coleenp@3865 464 st->print_cr("Number of buckets : %7d", summary.num());
coleenp@3865 465 st->print_cr("Average bucket size : %7.0f", summary.avg());
coleenp@3865 466 st->print_cr("Variance of bucket size : %7.0f", summary.variance());
coleenp@3865 467 st->print_cr("Std. dev. of bucket size: %7.0f", summary.sd());
coleenp@3865 468 st->print_cr("Maximum bucket size : %7.0f", summary.maximum());
coleenp@3865 469 }
coleenp@3865 470
duke@435 471
duke@435 472 //---------------------------------------------------------------------------
duke@435 473 // Non-product code
duke@435 474
duke@435 475 #ifndef PRODUCT
duke@435 476
duke@435 477 void SymbolTable::print_histogram() {
duke@435 478 MutexLocker ml(SymbolTable_lock);
duke@435 479 const int results_length = 100;
duke@435 480 int results[results_length];
duke@435 481 int i,j;
duke@435 482
duke@435 483 // initialize results to zero
duke@435 484 for (j = 0; j < results_length; j++) {
duke@435 485 results[j] = 0;
duke@435 486 }
duke@435 487
duke@435 488 int total = 0;
duke@435 489 int max_symbols = 0;
duke@435 490 int out_of_range = 0;
coleenp@2497 491 int memory_total = 0;
coleenp@2497 492 int count = 0;
duke@435 493 for (i = 0; i < the_table()->table_size(); i++) {
zgu@3900 494 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
duke@435 495 for ( ; p != NULL; p = p->next()) {
coleenp@4037 496 memory_total += p->literal()->size();
coleenp@2497 497 count++;
coleenp@2497 498 int counter = p->literal()->utf8_length();
duke@435 499 total += counter;
duke@435 500 if (counter < results_length) {
duke@435 501 results[counter]++;
duke@435 502 } else {
duke@435 503 out_of_range++;
duke@435 504 }
duke@435 505 max_symbols = MAX2(max_symbols, counter);
duke@435 506 }
duke@435 507 }
duke@435 508 tty->print_cr("Symbol Table:");
coleenp@2497 509 tty->print_cr("Total number of symbols %5d", count);
coleenp@2497 510 tty->print_cr("Total size in memory %5dK",
coleenp@2497 511 (memory_total*HeapWordSize)/1024);
coleenp@2497 512 tty->print_cr("Total counted %5d", symbols_counted);
coleenp@2497 513 tty->print_cr("Total removed %5d", symbols_removed);
coleenp@2497 514 if (symbols_counted > 0) {
coleenp@2497 515 tty->print_cr("Percent removed %3.2f",
coleenp@2497 516 ((float)symbols_removed/(float)symbols_counted)* 100);
coleenp@2497 517 }
coleenp@2497 518 tty->print_cr("Reference counts %5d", Symbol::_total_count);
coleenp@3682 519 tty->print_cr("Symbol arena size %5d used %5d",
coleenp@3682 520 arena()->size_in_bytes(), arena()->used());
coleenp@2497 521 tty->print_cr("Histogram of symbol length:");
duke@435 522 tty->print_cr("%8s %5d", "Total ", total);
duke@435 523 tty->print_cr("%8s %5d", "Maximum", max_symbols);
duke@435 524 tty->print_cr("%8s %3.2f", "Average",
duke@435 525 ((float) total / (float) the_table()->table_size()));
duke@435 526 tty->print_cr("%s", "Histogram:");
duke@435 527 tty->print_cr(" %s %29s", "Length", "Number chains that length");
duke@435 528 for (i = 0; i < results_length; i++) {
duke@435 529 if (results[i] > 0) {
duke@435 530 tty->print_cr("%6d %10d", i, results[i]);
duke@435 531 }
duke@435 532 }
coleenp@2497 533 if (Verbose) {
coleenp@2497 534 int line_length = 70;
coleenp@2497 535 tty->print_cr("%s %30s", " Length", "Number chains that length");
coleenp@2497 536 for (i = 0; i < results_length; i++) {
coleenp@2497 537 if (results[i] > 0) {
coleenp@2497 538 tty->print("%4d", i);
coleenp@2497 539 for (j = 0; (j < results[i]) && (j < line_length); j++) {
coleenp@2497 540 tty->print("%1s", "*");
coleenp@2497 541 }
coleenp@2497 542 if (j == line_length) {
coleenp@2497 543 tty->print("%1s", "+");
coleenp@2497 544 }
coleenp@2497 545 tty->cr();
duke@435 546 }
coleenp@2497 547 }
coleenp@2497 548 }
coleenp@2497 549 tty->print_cr(" %s %d: %d\n", "Number chains longer than",
coleenp@2497 550 results_length, out_of_range);
coleenp@2497 551 }
coleenp@2497 552
coleenp@2497 553 void SymbolTable::print() {
coleenp@2497 554 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 555 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 556 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
coleenp@2497 557 if (entry != NULL) {
coleenp@2497 558 while (entry != NULL) {
coleenp@2497 559 tty->print(PTR_FORMAT " ", entry->literal());
coleenp@2497 560 entry->literal()->print();
coleenp@2497 561 tty->print(" %d", entry->literal()->refcount());
coleenp@2497 562 p = entry->next_addr();
zgu@3900 563 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
duke@435 564 }
duke@435 565 tty->cr();
duke@435 566 }
duke@435 567 }
duke@435 568 }
duke@435 569 #endif // PRODUCT
duke@435 570
duke@435 571 // --------------------------------------------------------------------------
duke@435 572
duke@435 573 #ifdef ASSERT
duke@435 574 class StableMemoryChecker : public StackObj {
duke@435 575 enum { _bufsize = wordSize*4 };
duke@435 576
duke@435 577 address _region;
duke@435 578 jint _size;
duke@435 579 u1 _save_buf[_bufsize];
duke@435 580
duke@435 581 int sample(u1* save_buf) {
duke@435 582 if (_size <= _bufsize) {
duke@435 583 memcpy(save_buf, _region, _size);
duke@435 584 return _size;
duke@435 585 } else {
duke@435 586 // copy head and tail
duke@435 587 memcpy(&save_buf[0], _region, _bufsize/2);
duke@435 588 memcpy(&save_buf[_bufsize/2], _region + _size - _bufsize/2, _bufsize/2);
duke@435 589 return (_bufsize/2)*2;
duke@435 590 }
duke@435 591 }
duke@435 592
duke@435 593 public:
duke@435 594 StableMemoryChecker(const void* region, jint size) {
duke@435 595 _region = (address) region;
duke@435 596 _size = size;
duke@435 597 sample(_save_buf);
duke@435 598 }
duke@435 599
duke@435 600 bool verify() {
duke@435 601 u1 check_buf[sizeof(_save_buf)];
duke@435 602 int check_size = sample(check_buf);
duke@435 603 return (0 == memcmp(_save_buf, check_buf, check_size));
duke@435 604 }
duke@435 605
duke@435 606 void set_region(const void* region) { _region = (address) region; }
duke@435 607 };
duke@435 608 #endif
duke@435 609
duke@435 610
duke@435 611 // --------------------------------------------------------------------------
duke@435 612 StringTable* StringTable::_the_table = NULL;
duke@435 613
coleenp@3865 614 bool StringTable::_needs_rehashing = false;
coleenp@3865 615
coleenp@3865 616 // Pick hashing algorithm
coleenp@3875 617 unsigned int StringTable::hash_string(const jchar* s, int len) {
coleenp@3865 618 return use_alternate_hashcode() ? AltHashing::murmur3_32(seed(), s, len) :
brutisso@4335 619 java_lang_String::hash_code(s, len);
coleenp@3865 620 }
coleenp@3865 621
duke@435 622 oop StringTable::lookup(int index, jchar* name,
duke@435 623 int len, unsigned int hash) {
coleenp@3865 624 int count = 0;
zgu@3900 625 for (HashtableEntry<oop, mtSymbol>* l = bucket(index); l != NULL; l = l->next()) {
coleenp@3865 626 count++;
duke@435 627 if (l->hash() == hash) {
duke@435 628 if (java_lang_String::equals(l->literal(), name, len)) {
duke@435 629 return l->literal();
duke@435 630 }
duke@435 631 }
duke@435 632 }
coleenp@3865 633 // If the bucket size is too deep check if this hash code is insufficient.
zgu@3900 634 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
coleenp@3865 635 _needs_rehashing = check_rehash_table(count);
coleenp@3865 636 }
duke@435 637 return NULL;
duke@435 638 }
duke@435 639
duke@435 640
coleenp@3875 641 oop StringTable::basic_add(int index_arg, Handle string, jchar* name,
coleenp@3865 642 int len, unsigned int hashValue_arg, TRAPS) {
duke@435 643
duke@435 644 assert(java_lang_String::equals(string(), name, len),
duke@435 645 "string must be properly initialized");
coleenp@3875 646 // Cannot hit a safepoint in this function because the "this" pointer can move.
coleenp@3875 647 No_Safepoint_Verifier nsv;
duke@435 648
coleenp@3865 649 // Check if the symbol table has been rehashed, if so, need to recalculate
coleenp@3875 650 // the hash value and index before second lookup.
coleenp@3875 651 unsigned int hashValue;
coleenp@3875 652 int index;
coleenp@3875 653 if (use_alternate_hashcode()) {
coleenp@3875 654 hashValue = hash_string(name, len);
coleenp@3875 655 index = hash_to_index(hashValue);
coleenp@3875 656 } else {
coleenp@3875 657 hashValue = hashValue_arg;
coleenp@3875 658 index = index_arg;
coleenp@3875 659 }
coleenp@3865 660
duke@435 661 // Since look-up was done lock-free, we need to check if another
duke@435 662 // thread beat us in the race to insert the symbol.
duke@435 663
duke@435 664 oop test = lookup(index, name, len, hashValue); // calls lookup(u1*, int)
duke@435 665 if (test != NULL) {
duke@435 666 // Entry already added
duke@435 667 return test;
duke@435 668 }
duke@435 669
zgu@3900 670 HashtableEntry<oop, mtSymbol>* entry = new_entry(hashValue, string());
duke@435 671 add_entry(index, entry);
duke@435 672 return string();
duke@435 673 }
duke@435 674
duke@435 675
coleenp@2497 676 oop StringTable::lookup(Symbol* symbol) {
duke@435 677 ResourceMark rm;
duke@435 678 int length;
duke@435 679 jchar* chars = symbol->as_unicode(length);
coleenp@3865 680 unsigned int hashValue = hash_string(chars, length);
duke@435 681 int index = the_table()->hash_to_index(hashValue);
duke@435 682 return the_table()->lookup(index, chars, length, hashValue);
duke@435 683 }
duke@435 684
duke@435 685
duke@435 686 oop StringTable::intern(Handle string_or_null, jchar* name,
duke@435 687 int len, TRAPS) {
coleenp@3865 688 unsigned int hashValue = hash_string(name, len);
duke@435 689 int index = the_table()->hash_to_index(hashValue);
coleenp@3875 690 oop found_string = the_table()->lookup(index, name, len, hashValue);
duke@435 691
duke@435 692 // Found
coleenp@3875 693 if (found_string != NULL) return found_string;
coleenp@3875 694
coleenp@3875 695 debug_only(StableMemoryChecker smc(name, len * sizeof(name[0])));
coleenp@3875 696 assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),
coleenp@3875 697 "proposed name of symbol must be stable");
coleenp@3875 698
coleenp@3875 699 Handle string;
coleenp@3875 700 // try to reuse the string if possible
coleenp@4037 701 if (!string_or_null.is_null()) {
coleenp@3875 702 string = string_or_null;
coleenp@3875 703 } else {
coleenp@4037 704 string = java_lang_String::create_from_unicode(name, len, CHECK_NULL);
coleenp@3875 705 }
coleenp@3875 706
coleenp@3875 707 // Grab the StringTable_lock before getting the_table() because it could
coleenp@3875 708 // change at safepoint.
coleenp@3875 709 MutexLocker ml(StringTable_lock, THREAD);
duke@435 710
duke@435 711 // Otherwise, add to symbol to table
coleenp@3875 712 return the_table()->basic_add(index, string, name, len,
duke@435 713 hashValue, CHECK_NULL);
duke@435 714 }
duke@435 715
coleenp@2497 716 oop StringTable::intern(Symbol* symbol, TRAPS) {
duke@435 717 if (symbol == NULL) return NULL;
duke@435 718 ResourceMark rm(THREAD);
duke@435 719 int length;
duke@435 720 jchar* chars = symbol->as_unicode(length);
duke@435 721 Handle string;
duke@435 722 oop result = intern(string, chars, length, CHECK_NULL);
duke@435 723 return result;
duke@435 724 }
duke@435 725
duke@435 726
duke@435 727 oop StringTable::intern(oop string, TRAPS)
duke@435 728 {
duke@435 729 if (string == NULL) return NULL;
duke@435 730 ResourceMark rm(THREAD);
duke@435 731 int length;
duke@435 732 Handle h_string (THREAD, string);
duke@435 733 jchar* chars = java_lang_String::as_unicode_string(string, length);
duke@435 734 oop result = intern(h_string, chars, length, CHECK_NULL);
duke@435 735 return result;
duke@435 736 }
duke@435 737
duke@435 738
duke@435 739 oop StringTable::intern(const char* utf8_string, TRAPS) {
duke@435 740 if (utf8_string == NULL) return NULL;
duke@435 741 ResourceMark rm(THREAD);
duke@435 742 int length = UTF8::unicode_length(utf8_string);
duke@435 743 jchar* chars = NEW_RESOURCE_ARRAY(jchar, length);
duke@435 744 UTF8::convert_to_unicode(utf8_string, chars, length);
duke@435 745 Handle string;
duke@435 746 oop result = intern(string, chars, length, CHECK_NULL);
duke@435 747 return result;
duke@435 748 }
duke@435 749
coleenp@2497 750 void StringTable::unlink(BoolObjectClosure* is_alive) {
coleenp@2497 751 // Readers of the table are unlocked, so we should only be removing
coleenp@2497 752 // entries at a safepoint.
coleenp@2497 753 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@2497 754 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 755 HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 756 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
coleenp@3875 757 while (entry != NULL) {
coleenp@3875 758 // Shared entries are normally at the end of the bucket and if we run into
coleenp@3875 759 // a shared entry, then there is nothing more to remove. However, if we
coleenp@3875 760 // have rehashed the table, then the shared entries are no longer at the
coleenp@3875 761 // end of the bucket.
coleenp@3875 762 if (entry->is_shared() && !use_alternate_hashcode()) {
coleenp@2497 763 break;
coleenp@2497 764 }
coleenp@2497 765 assert(entry->literal() != NULL, "just checking");
coleenp@3875 766 if (entry->is_shared() || is_alive->do_object_b(entry->literal())) {
coleenp@2497 767 p = entry->next_addr();
coleenp@2497 768 } else {
coleenp@2497 769 *p = entry->next();
coleenp@2497 770 the_table()->free_entry(entry);
coleenp@2497 771 }
zgu@3900 772 entry = (HashtableEntry<oop, mtSymbol>*)HashtableEntry<oop, mtSymbol>::make_ptr(*p);
coleenp@2497 773 }
coleenp@2497 774 }
coleenp@2497 775 }
coleenp@2497 776
coleenp@2497 777 void StringTable::oops_do(OopClosure* f) {
coleenp@2497 778 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 779 HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
zgu@3900 780 HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
coleenp@2497 781 while (entry != NULL) {
coleenp@2497 782 f->do_oop((oop*)entry->literal_addr());
coleenp@2497 783
coleenp@2497 784 // Did the closure remove the literal from the table?
coleenp@2497 785 if (entry->literal() == NULL) {
coleenp@2497 786 assert(!entry->is_shared(), "immutable hashtable entry?");
coleenp@2497 787 *p = entry->next();
coleenp@2497 788 the_table()->free_entry(entry);
coleenp@2497 789 } else {
coleenp@2497 790 p = entry->next_addr();
coleenp@2497 791 }
zgu@3900 792 entry = (HashtableEntry<oop, mtSymbol>*)HashtableEntry<oop, mtSymbol>::make_ptr(*p);
coleenp@2497 793 }
coleenp@2497 794 }
coleenp@2497 795 }
coleenp@2497 796
duke@435 797 void StringTable::verify() {
duke@435 798 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 799 HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
duke@435 800 for ( ; p != NULL; p = p->next()) {
duke@435 801 oop s = p->literal();
duke@435 802 guarantee(s != NULL, "interned string is NULL");
never@2700 803 unsigned int h = java_lang_String::hash_string(s);
duke@435 804 guarantee(p->hash() == h, "broken hash in string table entry");
duke@435 805 guarantee(the_table()->hash_to_index(h) == i,
duke@435 806 "wrong index in string table");
duke@435 807 }
duke@435 808 }
duke@435 809 }
coleenp@3865 810
coleenp@3865 811 void StringTable::dump(outputStream* st) {
coleenp@3865 812 NumberSeq summary;
coleenp@3865 813 for (int i = 0; i < the_table()->table_size(); ++i) {
zgu@3900 814 HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
coleenp@3865 815 int count = 0;
coleenp@3865 816 for ( ; p != NULL; p = p->next()) {
coleenp@3865 817 count++;
coleenp@3865 818 }
coleenp@3865 819 summary.add((double)count);
coleenp@3865 820 }
coleenp@3865 821 st->print_cr("StringTable statistics:");
coleenp@3865 822 st->print_cr("Number of buckets : %7d", summary.num());
coleenp@3865 823 st->print_cr("Average bucket size : %7.0f", summary.avg());
coleenp@3865 824 st->print_cr("Variance of bucket size : %7.0f", summary.variance());
coleenp@3865 825 st->print_cr("Std. dev. of bucket size: %7.0f", summary.sd());
coleenp@3865 826 st->print_cr("Maximum bucket size : %7.0f", summary.maximum());
coleenp@3865 827 }
coleenp@3865 828
coleenp@3865 829
coleenp@3865 830 // Create a new table and using alternate hash code, populate the new table
coleenp@3865 831 // with the existing strings. Set flag to use the alternate hash code afterwards.
coleenp@3865 832 void StringTable::rehash_table() {
coleenp@3865 833 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
coleenp@3875 834 // This should never happen with -Xshare:dump but it might in testing mode.
coleenp@3875 835 if (DumpSharedSpaces) return;
coleenp@3865 836 StringTable* new_table = new StringTable();
coleenp@3865 837
coleenp@3865 838 // Rehash the table
coleenp@3865 839 the_table()->move_to(new_table);
coleenp@3865 840
coleenp@3865 841 // Delete the table and buckets (entries are reused in new table).
coleenp@3865 842 delete _the_table;
coleenp@3865 843 // Don't check if we need rehashing until the table gets unbalanced again.
coleenp@3865 844 // Then rehash with a new global seed.
coleenp@3865 845 _needs_rehashing = false;
coleenp@3865 846 _the_table = new_table;
coleenp@3865 847 }

mercurial