Thu, 24 Mar 2016 21:38:15 -0700
8150752: Share Class Data
Reviewed-by: acorn, hseigel, mschoene
duke@435 | 1 | /* |
pliden@6413 | 2 | * Copyright (c) 1997, 2014, 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" |
pliden@6413 | 38 | #if INCLUDE_ALL_GCS |
stefank@6992 | 39 | #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp" |
pliden@6413 | 40 | #include "gc_implementation/g1/g1StringDedup.hpp" |
pliden@6413 | 41 | #endif |
duke@435 | 42 | |
drchase@6680 | 43 | PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC |
drchase@6680 | 44 | |
duke@435 | 45 | // -------------------------------------------------------------------------- |
duke@435 | 46 | |
tschatzl@6229 | 47 | // the number of buckets a thread claims |
tschatzl@6229 | 48 | const int ClaimChunkSize = 32; |
tschatzl@6229 | 49 | |
duke@435 | 50 | SymbolTable* SymbolTable::_the_table = NULL; |
coleenp@3682 | 51 | // Static arena for symbols that are not deallocated |
coleenp@3682 | 52 | Arena* SymbolTable::_arena = NULL; |
coleenp@3865 | 53 | bool SymbolTable::_needs_rehashing = false; |
duke@435 | 54 | |
coleenp@3682 | 55 | Symbol* SymbolTable::allocate_symbol(const u1* name, int len, bool c_heap, TRAPS) { |
coleenp@3875 | 56 | assert (len <= Symbol::max_length(), "should be checked by caller"); |
coleenp@3875 | 57 | |
coleenp@3682 | 58 | Symbol* sym; |
coleenp@4037 | 59 | |
coleenp@4718 | 60 | if (DumpSharedSpaces) { |
coleenp@4718 | 61 | // Allocate all symbols to CLD shared metaspace |
coleenp@4718 | 62 | sym = new (len, ClassLoaderData::the_null_class_loader_data(), THREAD) Symbol(name, len, -1); |
coleenp@4718 | 63 | } else if (c_heap) { |
coleenp@3682 | 64 | // refcount starts as 1 |
coleenp@3682 | 65 | sym = new (len, THREAD) Symbol(name, len, 1); |
coleenp@4037 | 66 | assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted"); |
coleenp@4037 | 67 | } else { |
coleenp@4718 | 68 | // Allocate to global arena |
coleenp@3682 | 69 | sym = new (len, arena(), THREAD) Symbol(name, len, -1); |
coleenp@3682 | 70 | } |
coleenp@2497 | 71 | return sym; |
coleenp@2497 | 72 | } |
coleenp@2497 | 73 | |
coleenp@3682 | 74 | void SymbolTable::initialize_symbols(int arena_alloc_size) { |
coleenp@3682 | 75 | // Initialize the arena for global symbols, size passed in depends on CDS. |
coleenp@3682 | 76 | if (arena_alloc_size == 0) { |
zgu@7074 | 77 | _arena = new (mtSymbol) Arena(mtSymbol); |
coleenp@3682 | 78 | } else { |
zgu@7074 | 79 | _arena = new (mtSymbol) Arena(mtSymbol, arena_alloc_size); |
coleenp@2497 | 80 | } |
coleenp@2497 | 81 | } |
coleenp@2497 | 82 | |
coleenp@2497 | 83 | // Call function for all symbols in the symbol table. |
coleenp@2497 | 84 | void SymbolTable::symbols_do(SymbolClosure *cl) { |
coleenp@2497 | 85 | const int n = the_table()->table_size(); |
coleenp@2497 | 86 | for (int i = 0; i < n; i++) { |
zgu@3900 | 87 | for (HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i); |
coleenp@2497 | 88 | p != NULL; |
coleenp@2497 | 89 | p = p->next()) { |
coleenp@2497 | 90 | cl->do_symbol(p->literal_addr()); |
coleenp@2497 | 91 | } |
coleenp@2497 | 92 | } |
coleenp@2497 | 93 | } |
coleenp@2497 | 94 | |
tschatzl@6229 | 95 | int SymbolTable::_symbols_removed = 0; |
tschatzl@6229 | 96 | int SymbolTable::_symbols_counted = 0; |
tschatzl@6229 | 97 | volatile int SymbolTable::_parallel_claimed_idx = 0; |
coleenp@2497 | 98 | |
tschatzl@6229 | 99 | void SymbolTable::buckets_unlink(int start_idx, int end_idx, int* processed, int* removed, size_t* memory_total) { |
tschatzl@6229 | 100 | for (int i = start_idx; i < end_idx; ++i) { |
zgu@3900 | 101 | HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i); |
zgu@3900 | 102 | HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i); |
coleenp@3875 | 103 | while (entry != NULL) { |
coleenp@3875 | 104 | // Shared entries are normally at the end of the bucket and if we run into |
coleenp@3875 | 105 | // a shared entry, then there is nothing more to remove. However, if we |
coleenp@3875 | 106 | // have rehashed the table, then the shared entries are no longer at the |
coleenp@3875 | 107 | // end of the bucket. |
coleenp@3875 | 108 | if (entry->is_shared() && !use_alternate_hashcode()) { |
coleenp@2497 | 109 | break; |
coleenp@2497 | 110 | } |
coleenp@2497 | 111 | Symbol* s = entry->literal(); |
tschatzl@6229 | 112 | (*memory_total) += s->size(); |
tschatzl@6229 | 113 | (*processed)++; |
coleenp@2497 | 114 | assert(s != NULL, "just checking"); |
coleenp@2497 | 115 | // If reference count is zero, remove. |
coleenp@2497 | 116 | if (s->refcount() == 0) { |
coleenp@3875 | 117 | assert(!entry->is_shared(), "shared entries should be kept live"); |
coleenp@2497 | 118 | delete s; |
tschatzl@6229 | 119 | (*removed)++; |
coleenp@2497 | 120 | *p = entry->next(); |
coleenp@2497 | 121 | the_table()->free_entry(entry); |
coleenp@2497 | 122 | } else { |
coleenp@2497 | 123 | p = entry->next_addr(); |
coleenp@2497 | 124 | } |
coleenp@3875 | 125 | // get next entry |
zgu@3900 | 126 | entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p); |
coleenp@2497 | 127 | } |
coleenp@2497 | 128 | } |
tschatzl@6229 | 129 | } |
tschatzl@6229 | 130 | |
tschatzl@6229 | 131 | // Remove unreferenced symbols from the symbol table |
tschatzl@6229 | 132 | // This is done late during GC. |
tschatzl@6229 | 133 | void SymbolTable::unlink(int* processed, int* removed) { |
tschatzl@6229 | 134 | size_t memory_total = 0; |
tschatzl@6229 | 135 | buckets_unlink(0, the_table()->table_size(), processed, removed, &memory_total); |
tschatzl@6229 | 136 | _symbols_removed += *removed; |
tschatzl@6229 | 137 | _symbols_counted += *processed; |
coleenp@2618 | 138 | // Exclude printing for normal PrintGCDetails because people parse |
coleenp@2618 | 139 | // this output. |
coleenp@2618 | 140 | if (PrintGCDetails && Verbose && WizardMode) { |
tschatzl@6229 | 141 | gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", *processed, |
tschatzl@6229 | 142 | (memory_total*HeapWordSize)/1024); |
tschatzl@6229 | 143 | } |
tschatzl@6229 | 144 | } |
tschatzl@6229 | 145 | |
tschatzl@6229 | 146 | void SymbolTable::possibly_parallel_unlink(int* processed, int* removed) { |
tschatzl@6229 | 147 | const int limit = the_table()->table_size(); |
tschatzl@6229 | 148 | |
tschatzl@6229 | 149 | size_t memory_total = 0; |
tschatzl@6229 | 150 | |
tschatzl@6229 | 151 | for (;;) { |
tschatzl@6229 | 152 | // Grab next set of buckets to scan |
tschatzl@6229 | 153 | int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize; |
tschatzl@6229 | 154 | if (start_idx >= limit) { |
tschatzl@6229 | 155 | // End of table |
tschatzl@6229 | 156 | break; |
tschatzl@6229 | 157 | } |
tschatzl@6229 | 158 | |
tschatzl@6229 | 159 | int end_idx = MIN2(limit, start_idx + ClaimChunkSize); |
tschatzl@6229 | 160 | buckets_unlink(start_idx, end_idx, processed, removed, &memory_total); |
tschatzl@6229 | 161 | } |
tschatzl@6229 | 162 | Atomic::add(*processed, &_symbols_counted); |
tschatzl@6229 | 163 | Atomic::add(*removed, &_symbols_removed); |
tschatzl@6229 | 164 | // Exclude printing for normal PrintGCDetails because people parse |
tschatzl@6229 | 165 | // this output. |
tschatzl@6229 | 166 | if (PrintGCDetails && Verbose && WizardMode) { |
tschatzl@6229 | 167 | gclog_or_tty->print(" [Symbols: scanned=%d removed=%d size=" SIZE_FORMAT "K] ", *processed, *removed, |
coleenp@2497 | 168 | (memory_total*HeapWordSize)/1024); |
coleenp@2497 | 169 | } |
coleenp@2497 | 170 | } |
coleenp@2497 | 171 | |
coleenp@3865 | 172 | // Create a new table and using alternate hash code, populate the new table |
coleenp@3865 | 173 | // with the existing strings. Set flag to use the alternate hash code afterwards. |
coleenp@3865 | 174 | void SymbolTable::rehash_table() { |
coleenp@3865 | 175 | assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint"); |
coleenp@3875 | 176 | // This should never happen with -Xshare:dump but it might in testing mode. |
coleenp@3875 | 177 | if (DumpSharedSpaces) return; |
coleenp@3865 | 178 | // Create a new symbol table |
coleenp@3865 | 179 | SymbolTable* new_table = new SymbolTable(); |
coleenp@3865 | 180 | |
coleenp@3865 | 181 | the_table()->move_to(new_table); |
coleenp@3865 | 182 | |
coleenp@3865 | 183 | // Delete the table and buckets (entries are reused in new table). |
coleenp@3865 | 184 | delete _the_table; |
coleenp@3865 | 185 | // Don't check if we need rehashing until the table gets unbalanced again. |
coleenp@3865 | 186 | // Then rehash with a new global seed. |
coleenp@3865 | 187 | _needs_rehashing = false; |
coleenp@3865 | 188 | _the_table = new_table; |
coleenp@3865 | 189 | } |
coleenp@2497 | 190 | |
duke@435 | 191 | // Lookup a symbol in a bucket. |
duke@435 | 192 | |
coleenp@2497 | 193 | Symbol* SymbolTable::lookup(int index, const char* name, |
duke@435 | 194 | int len, unsigned int hash) { |
coleenp@3865 | 195 | int count = 0; |
zgu@3900 | 196 | for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) { |
coleenp@3865 | 197 | count++; // count all entries in this bucket, not just ones with same hash |
duke@435 | 198 | if (e->hash() == hash) { |
coleenp@2497 | 199 | Symbol* sym = e->literal(); |
duke@435 | 200 | if (sym->equals(name, len)) { |
coleenp@2497 | 201 | // something is referencing this symbol now. |
coleenp@2497 | 202 | sym->increment_refcount(); |
duke@435 | 203 | return sym; |
duke@435 | 204 | } |
duke@435 | 205 | } |
duke@435 | 206 | } |
coleenp@3865 | 207 | // If the bucket size is too deep check if this hash code is insufficient. |
mgerdin@7207 | 208 | if (count >= rehash_count && !needs_rehashing()) { |
coleenp@3865 | 209 | _needs_rehashing = check_rehash_table(count); |
coleenp@3865 | 210 | } |
duke@435 | 211 | return NULL; |
duke@435 | 212 | } |
duke@435 | 213 | |
coleenp@3875 | 214 | // Pick hashing algorithm. |
coleenp@3875 | 215 | unsigned int SymbolTable::hash_symbol(const char* s, int len) { |
coleenp@3865 | 216 | return use_alternate_hashcode() ? |
coleenp@3865 | 217 | AltHashing::murmur3_32(seed(), (const jbyte*)s, len) : |
brutisso@4335 | 218 | java_lang_String::hash_code(s, len); |
coleenp@3865 | 219 | } |
coleenp@3865 | 220 | |
duke@435 | 221 | |
duke@435 | 222 | // We take care not to be blocking while holding the |
duke@435 | 223 | // SymbolTable_lock. Otherwise, the system might deadlock, since the |
duke@435 | 224 | // symboltable is used during compilation (VM_thread) The lock free |
duke@435 | 225 | // synchronization is simplified by the fact that we do not delete |
duke@435 | 226 | // entries in the symbol table during normal execution (only during |
duke@435 | 227 | // safepoints). |
duke@435 | 228 | |
coleenp@2497 | 229 | Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) { |
duke@435 | 230 | unsigned int hashValue = hash_symbol(name, len); |
duke@435 | 231 | int index = the_table()->hash_to_index(hashValue); |
duke@435 | 232 | |
coleenp@2497 | 233 | Symbol* s = the_table()->lookup(index, name, len, hashValue); |
duke@435 | 234 | |
duke@435 | 235 | // Found |
duke@435 | 236 | if (s != NULL) return s; |
duke@435 | 237 | |
coleenp@3875 | 238 | // Grab SymbolTable_lock first. |
coleenp@3875 | 239 | MutexLocker ml(SymbolTable_lock, THREAD); |
coleenp@3875 | 240 | |
duke@435 | 241 | // Otherwise, add to symbol to table |
coleenp@3682 | 242 | return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL); |
duke@435 | 243 | } |
duke@435 | 244 | |
coleenp@2497 | 245 | Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) { |
duke@435 | 246 | char* buffer; |
duke@435 | 247 | int index, len; |
duke@435 | 248 | unsigned int hashValue; |
duke@435 | 249 | char* name; |
duke@435 | 250 | { |
duke@435 | 251 | debug_only(No_Safepoint_Verifier nsv;) |
duke@435 | 252 | |
duke@435 | 253 | name = (char*)sym->base() + begin; |
duke@435 | 254 | len = end - begin; |
duke@435 | 255 | hashValue = hash_symbol(name, len); |
duke@435 | 256 | index = the_table()->hash_to_index(hashValue); |
coleenp@2497 | 257 | Symbol* s = the_table()->lookup(index, name, len, hashValue); |
duke@435 | 258 | |
duke@435 | 259 | // Found |
duke@435 | 260 | if (s != NULL) return s; |
duke@435 | 261 | } |
duke@435 | 262 | |
duke@435 | 263 | // Otherwise, add to symbol to table. Copy to a C string first. |
duke@435 | 264 | char stack_buf[128]; |
duke@435 | 265 | ResourceMark rm(THREAD); |
duke@435 | 266 | if (len <= 128) { |
duke@435 | 267 | buffer = stack_buf; |
duke@435 | 268 | } else { |
duke@435 | 269 | buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len); |
duke@435 | 270 | } |
duke@435 | 271 | for (int i=0; i<len; i++) { |
duke@435 | 272 | buffer[i] = name[i]; |
duke@435 | 273 | } |
duke@435 | 274 | // Make sure there is no safepoint in the code above since name can't move. |
duke@435 | 275 | // We can't include the code in No_Safepoint_Verifier because of the |
duke@435 | 276 | // ResourceMark. |
duke@435 | 277 | |
coleenp@3875 | 278 | // Grab SymbolTable_lock first. |
coleenp@3875 | 279 | MutexLocker ml(SymbolTable_lock, THREAD); |
coleenp@3875 | 280 | |
coleenp@3682 | 281 | return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL); |
duke@435 | 282 | } |
duke@435 | 283 | |
coleenp@2497 | 284 | Symbol* SymbolTable::lookup_only(const char* name, int len, |
duke@435 | 285 | unsigned int& hash) { |
duke@435 | 286 | hash = hash_symbol(name, len); |
duke@435 | 287 | int index = the_table()->hash_to_index(hash); |
duke@435 | 288 | |
coleenp@2497 | 289 | Symbol* s = the_table()->lookup(index, name, len, hash); |
coleenp@2497 | 290 | return s; |
duke@435 | 291 | } |
duke@435 | 292 | |
phh@3427 | 293 | // Look up the address of the literal in the SymbolTable for this Symbol* |
phh@3427 | 294 | // Do not create any new symbols |
phh@3427 | 295 | // Do not increment the reference count to keep this alive |
phh@3427 | 296 | Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){ |
phh@3427 | 297 | unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length()); |
phh@3427 | 298 | int index = the_table()->hash_to_index(hash); |
phh@3427 | 299 | |
zgu@3900 | 300 | for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) { |
phh@3427 | 301 | if (e->hash() == hash) { |
phh@3427 | 302 | Symbol* literal_sym = e->literal(); |
phh@3427 | 303 | if (sym == literal_sym) { |
phh@3427 | 304 | return e->literal_addr(); |
phh@3427 | 305 | } |
phh@3427 | 306 | } |
phh@3427 | 307 | } |
phh@3427 | 308 | return NULL; |
phh@3427 | 309 | } |
phh@3427 | 310 | |
jrose@1100 | 311 | // Suggestion: Push unicode-based lookup all the way into the hashing |
jrose@1100 | 312 | // and probing logic, so there is no need for convert_to_utf8 until |
coleenp@2497 | 313 | // an actual new Symbol* is created. |
coleenp@2497 | 314 | Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) { |
jrose@1100 | 315 | int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length); |
jrose@1100 | 316 | char stack_buf[128]; |
jrose@1100 | 317 | if (utf8_length < (int) sizeof(stack_buf)) { |
jrose@1100 | 318 | char* chars = stack_buf; |
jrose@1100 | 319 | UNICODE::convert_to_utf8(name, utf16_length, chars); |
jrose@1100 | 320 | return lookup(chars, utf8_length, THREAD); |
jrose@1100 | 321 | } else { |
jrose@1100 | 322 | ResourceMark rm(THREAD); |
jrose@1100 | 323 | char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);; |
jrose@1100 | 324 | UNICODE::convert_to_utf8(name, utf16_length, chars); |
jrose@1100 | 325 | return lookup(chars, utf8_length, THREAD); |
jrose@1100 | 326 | } |
jrose@1100 | 327 | } |
jrose@1100 | 328 | |
coleenp@2497 | 329 | Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length, |
jrose@1100 | 330 | unsigned int& hash) { |
jrose@1100 | 331 | int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length); |
jrose@1100 | 332 | char stack_buf[128]; |
jrose@1100 | 333 | if (utf8_length < (int) sizeof(stack_buf)) { |
jrose@1100 | 334 | char* chars = stack_buf; |
jrose@1100 | 335 | UNICODE::convert_to_utf8(name, utf16_length, chars); |
jrose@1100 | 336 | return lookup_only(chars, utf8_length, hash); |
jrose@1100 | 337 | } else { |
jrose@1100 | 338 | ResourceMark rm; |
jrose@1100 | 339 | char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);; |
jrose@1100 | 340 | UNICODE::convert_to_utf8(name, utf16_length, chars); |
jrose@1100 | 341 | return lookup_only(chars, utf8_length, hash); |
jrose@1100 | 342 | } |
jrose@1100 | 343 | } |
jrose@1100 | 344 | |
coleenp@4037 | 345 | void SymbolTable::add(ClassLoaderData* loader_data, constantPoolHandle cp, |
coleenp@3682 | 346 | int names_count, |
duke@435 | 347 | const char** names, int* lengths, int* cp_indices, |
duke@435 | 348 | unsigned int* hashValues, TRAPS) { |
coleenp@3875 | 349 | // Grab SymbolTable_lock first. |
coleenp@3875 | 350 | MutexLocker ml(SymbolTable_lock, THREAD); |
coleenp@3875 | 351 | |
duke@435 | 352 | SymbolTable* table = the_table(); |
coleenp@4037 | 353 | bool added = table->basic_add(loader_data, cp, names_count, names, lengths, |
duke@435 | 354 | cp_indices, hashValues, CHECK); |
duke@435 | 355 | if (!added) { |
duke@435 | 356 | // do it the hard way |
duke@435 | 357 | for (int i=0; i<names_count; i++) { |
duke@435 | 358 | int index = table->hash_to_index(hashValues[i]); |
coleenp@4037 | 359 | bool c_heap = !loader_data->is_the_null_class_loader_data(); |
coleenp@3682 | 360 | Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK); |
duke@435 | 361 | cp->symbol_at_put(cp_indices[i], sym); |
duke@435 | 362 | } |
duke@435 | 363 | } |
duke@435 | 364 | } |
duke@435 | 365 | |
coleenp@3682 | 366 | Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) { |
coleenp@3682 | 367 | unsigned int hash; |
coleenp@3682 | 368 | Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash); |
coleenp@3682 | 369 | if (result != NULL) { |
coleenp@3682 | 370 | return result; |
coleenp@3682 | 371 | } |
coleenp@3875 | 372 | // Grab SymbolTable_lock first. |
coleenp@3875 | 373 | MutexLocker ml(SymbolTable_lock, THREAD); |
coleenp@3875 | 374 | |
coleenp@3682 | 375 | SymbolTable* table = the_table(); |
coleenp@3682 | 376 | int index = table->hash_to_index(hash); |
coleenp@3682 | 377 | return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD); |
coleenp@3682 | 378 | } |
coleenp@3682 | 379 | |
coleenp@3875 | 380 | Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len, |
coleenp@3865 | 381 | unsigned int hashValue_arg, bool c_heap, TRAPS) { |
stefank@5769 | 382 | assert(!Universe::heap()->is_in_reserved(name), |
duke@435 | 383 | "proposed name of symbol must be stable"); |
duke@435 | 384 | |
coleenp@3875 | 385 | // Don't allow symbols to be created which cannot fit in a Symbol*. |
coleenp@3875 | 386 | if (len > Symbol::max_length()) { |
coleenp@3875 | 387 | THROW_MSG_0(vmSymbols::java_lang_InternalError(), |
coleenp@3875 | 388 | "name is too long to represent"); |
coleenp@3875 | 389 | } |
coleenp@3875 | 390 | |
coleenp@3875 | 391 | // Cannot hit a safepoint in this function because the "this" pointer can move. |
coleenp@3875 | 392 | No_Safepoint_Verifier nsv; |
duke@435 | 393 | |
coleenp@3865 | 394 | // Check if the symbol table has been rehashed, if so, need to recalculate |
coleenp@3875 | 395 | // the hash value and index. |
coleenp@3875 | 396 | unsigned int hashValue; |
coleenp@3875 | 397 | int index; |
coleenp@3875 | 398 | if (use_alternate_hashcode()) { |
coleenp@3875 | 399 | hashValue = hash_symbol((const char*)name, len); |
coleenp@3875 | 400 | index = hash_to_index(hashValue); |
coleenp@3875 | 401 | } else { |
coleenp@3875 | 402 | hashValue = hashValue_arg; |
coleenp@3875 | 403 | index = index_arg; |
coleenp@3875 | 404 | } |
coleenp@3865 | 405 | |
duke@435 | 406 | // Since look-up was done lock-free, we need to check if another |
duke@435 | 407 | // thread beat us in the race to insert the symbol. |
coleenp@2497 | 408 | Symbol* test = lookup(index, (char*)name, len, hashValue); |
duke@435 | 409 | if (test != NULL) { |
coleenp@3682 | 410 | // A race occurred and another thread introduced the symbol. |
coleenp@2497 | 411 | assert(test->refcount() != 0, "lookup should have incremented the count"); |
duke@435 | 412 | return test; |
duke@435 | 413 | } |
duke@435 | 414 | |
coleenp@3682 | 415 | // Create a new symbol. |
coleenp@3682 | 416 | Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL); |
coleenp@3682 | 417 | assert(sym->equals((char*)name, len), "symbol must be properly initialized"); |
coleenp@3682 | 418 | |
zgu@3900 | 419 | HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym); |
duke@435 | 420 | add_entry(index, entry); |
coleenp@2497 | 421 | return sym; |
duke@435 | 422 | } |
duke@435 | 423 | |
coleenp@3682 | 424 | // This version of basic_add adds symbols in batch from the constant pool |
coleenp@3682 | 425 | // parsing. |
coleenp@4037 | 426 | bool SymbolTable::basic_add(ClassLoaderData* loader_data, constantPoolHandle cp, |
coleenp@3682 | 427 | int names_count, |
duke@435 | 428 | const char** names, int* lengths, |
duke@435 | 429 | int* cp_indices, unsigned int* hashValues, |
duke@435 | 430 | TRAPS) { |
coleenp@3682 | 431 | |
coleenp@3682 | 432 | // Check symbol names are not too long. If any are too long, don't add any. |
coleenp@3682 | 433 | for (int i = 0; i< names_count; i++) { |
coleenp@3682 | 434 | if (lengths[i] > Symbol::max_length()) { |
coleenp@3682 | 435 | THROW_MSG_0(vmSymbols::java_lang_InternalError(), |
coleenp@3682 | 436 | "name is too long to represent"); |
coleenp@3682 | 437 | } |
duke@435 | 438 | } |
duke@435 | 439 | |
coleenp@3875 | 440 | // Cannot hit a safepoint in this function because the "this" pointer can move. |
coleenp@3875 | 441 | No_Safepoint_Verifier nsv; |
duke@435 | 442 | |
coleenp@2497 | 443 | for (int i=0; i<names_count; i++) { |
coleenp@3865 | 444 | // Check if the symbol table has been rehashed, if so, need to recalculate |
coleenp@3865 | 445 | // the hash value. |
coleenp@3875 | 446 | unsigned int hashValue; |
coleenp@3875 | 447 | if (use_alternate_hashcode()) { |
coleenp@3875 | 448 | hashValue = hash_symbol(names[i], lengths[i]); |
coleenp@3875 | 449 | } else { |
coleenp@3875 | 450 | hashValue = hashValues[i]; |
coleenp@3875 | 451 | } |
duke@435 | 452 | // Since look-up was done lock-free, we need to check if another |
duke@435 | 453 | // thread beat us in the race to insert the symbol. |
coleenp@3865 | 454 | int index = hash_to_index(hashValue); |
coleenp@3865 | 455 | Symbol* test = lookup(index, names[i], lengths[i], hashValue); |
duke@435 | 456 | if (test != NULL) { |
twisti@1040 | 457 | // A race occurred and another thread introduced the symbol, this one |
duke@435 | 458 | // will be dropped and collected. Use test instead. |
duke@435 | 459 | cp->symbol_at_put(cp_indices[i], test); |
coleenp@2497 | 460 | assert(test->refcount() != 0, "lookup should have incremented the count"); |
duke@435 | 461 | } else { |
coleenp@3682 | 462 | // Create a new symbol. The null class loader is never unloaded so these |
coleenp@3682 | 463 | // are allocated specially in a permanent arena. |
coleenp@4037 | 464 | bool c_heap = !loader_data->is_the_null_class_loader_data(); |
coleenp@3682 | 465 | Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false)); |
coleenp@3682 | 466 | assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized"); // why wouldn't it be??? |
zgu@3900 | 467 | HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym); |
duke@435 | 468 | add_entry(index, entry); |
duke@435 | 469 | cp->symbol_at_put(cp_indices[i], sym); |
duke@435 | 470 | } |
duke@435 | 471 | } |
duke@435 | 472 | return true; |
duke@435 | 473 | } |
duke@435 | 474 | |
duke@435 | 475 | |
duke@435 | 476 | void SymbolTable::verify() { |
duke@435 | 477 | for (int i = 0; i < the_table()->table_size(); ++i) { |
zgu@3900 | 478 | HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i); |
duke@435 | 479 | for ( ; p != NULL; p = p->next()) { |
coleenp@2497 | 480 | Symbol* s = (Symbol*)(p->literal()); |
duke@435 | 481 | guarantee(s != NULL, "symbol is NULL"); |
duke@435 | 482 | unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length()); |
duke@435 | 483 | guarantee(p->hash() == h, "broken hash in symbol table entry"); |
duke@435 | 484 | guarantee(the_table()->hash_to_index(h) == i, |
duke@435 | 485 | "wrong index in symbol table"); |
duke@435 | 486 | } |
duke@435 | 487 | } |
duke@435 | 488 | } |
duke@435 | 489 | |
coleenp@3865 | 490 | void SymbolTable::dump(outputStream* st) { |
iklam@5144 | 491 | the_table()->dump_table(st, "SymbolTable"); |
coleenp@3865 | 492 | } |
coleenp@3865 | 493 | |
duke@435 | 494 | |
duke@435 | 495 | //--------------------------------------------------------------------------- |
duke@435 | 496 | // Non-product code |
duke@435 | 497 | |
duke@435 | 498 | #ifndef PRODUCT |
duke@435 | 499 | |
duke@435 | 500 | void SymbolTable::print_histogram() { |
duke@435 | 501 | MutexLocker ml(SymbolTable_lock); |
duke@435 | 502 | const int results_length = 100; |
duke@435 | 503 | int results[results_length]; |
duke@435 | 504 | int i,j; |
duke@435 | 505 | |
duke@435 | 506 | // initialize results to zero |
duke@435 | 507 | for (j = 0; j < results_length; j++) { |
duke@435 | 508 | results[j] = 0; |
duke@435 | 509 | } |
duke@435 | 510 | |
duke@435 | 511 | int total = 0; |
duke@435 | 512 | int max_symbols = 0; |
duke@435 | 513 | int out_of_range = 0; |
coleenp@2497 | 514 | int memory_total = 0; |
coleenp@2497 | 515 | int count = 0; |
duke@435 | 516 | for (i = 0; i < the_table()->table_size(); i++) { |
zgu@3900 | 517 | HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i); |
duke@435 | 518 | for ( ; p != NULL; p = p->next()) { |
coleenp@4037 | 519 | memory_total += p->literal()->size(); |
coleenp@2497 | 520 | count++; |
coleenp@2497 | 521 | int counter = p->literal()->utf8_length(); |
duke@435 | 522 | total += counter; |
duke@435 | 523 | if (counter < results_length) { |
duke@435 | 524 | results[counter]++; |
duke@435 | 525 | } else { |
duke@435 | 526 | out_of_range++; |
duke@435 | 527 | } |
duke@435 | 528 | max_symbols = MAX2(max_symbols, counter); |
duke@435 | 529 | } |
duke@435 | 530 | } |
duke@435 | 531 | tty->print_cr("Symbol Table:"); |
coleenp@2497 | 532 | tty->print_cr("Total number of symbols %5d", count); |
coleenp@2497 | 533 | tty->print_cr("Total size in memory %5dK", |
coleenp@2497 | 534 | (memory_total*HeapWordSize)/1024); |
tschatzl@6229 | 535 | tty->print_cr("Total counted %5d", _symbols_counted); |
tschatzl@6229 | 536 | tty->print_cr("Total removed %5d", _symbols_removed); |
tschatzl@6229 | 537 | if (_symbols_counted > 0) { |
coleenp@2497 | 538 | tty->print_cr("Percent removed %3.2f", |
tschatzl@6229 | 539 | ((float)_symbols_removed/(float)_symbols_counted)* 100); |
coleenp@2497 | 540 | } |
coleenp@2497 | 541 | tty->print_cr("Reference counts %5d", Symbol::_total_count); |
coleenp@3682 | 542 | tty->print_cr("Symbol arena size %5d used %5d", |
coleenp@3682 | 543 | arena()->size_in_bytes(), arena()->used()); |
coleenp@2497 | 544 | tty->print_cr("Histogram of symbol length:"); |
duke@435 | 545 | tty->print_cr("%8s %5d", "Total ", total); |
duke@435 | 546 | tty->print_cr("%8s %5d", "Maximum", max_symbols); |
duke@435 | 547 | tty->print_cr("%8s %3.2f", "Average", |
duke@435 | 548 | ((float) total / (float) the_table()->table_size())); |
duke@435 | 549 | tty->print_cr("%s", "Histogram:"); |
duke@435 | 550 | tty->print_cr(" %s %29s", "Length", "Number chains that length"); |
duke@435 | 551 | for (i = 0; i < results_length; i++) { |
duke@435 | 552 | if (results[i] > 0) { |
duke@435 | 553 | tty->print_cr("%6d %10d", i, results[i]); |
duke@435 | 554 | } |
duke@435 | 555 | } |
coleenp@2497 | 556 | if (Verbose) { |
coleenp@2497 | 557 | int line_length = 70; |
coleenp@2497 | 558 | tty->print_cr("%s %30s", " Length", "Number chains that length"); |
coleenp@2497 | 559 | for (i = 0; i < results_length; i++) { |
coleenp@2497 | 560 | if (results[i] > 0) { |
coleenp@2497 | 561 | tty->print("%4d", i); |
coleenp@2497 | 562 | for (j = 0; (j < results[i]) && (j < line_length); j++) { |
coleenp@2497 | 563 | tty->print("%1s", "*"); |
coleenp@2497 | 564 | } |
coleenp@2497 | 565 | if (j == line_length) { |
coleenp@2497 | 566 | tty->print("%1s", "+"); |
coleenp@2497 | 567 | } |
coleenp@2497 | 568 | tty->cr(); |
duke@435 | 569 | } |
coleenp@2497 | 570 | } |
coleenp@2497 | 571 | } |
coleenp@2497 | 572 | tty->print_cr(" %s %d: %d\n", "Number chains longer than", |
coleenp@2497 | 573 | results_length, out_of_range); |
coleenp@2497 | 574 | } |
coleenp@2497 | 575 | |
coleenp@2497 | 576 | void SymbolTable::print() { |
coleenp@2497 | 577 | for (int i = 0; i < the_table()->table_size(); ++i) { |
zgu@3900 | 578 | HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i); |
zgu@3900 | 579 | HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i); |
coleenp@2497 | 580 | if (entry != NULL) { |
coleenp@2497 | 581 | while (entry != NULL) { |
coleenp@2497 | 582 | tty->print(PTR_FORMAT " ", entry->literal()); |
coleenp@2497 | 583 | entry->literal()->print(); |
coleenp@2497 | 584 | tty->print(" %d", entry->literal()->refcount()); |
coleenp@2497 | 585 | p = entry->next_addr(); |
zgu@3900 | 586 | entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p); |
duke@435 | 587 | } |
duke@435 | 588 | tty->cr(); |
duke@435 | 589 | } |
duke@435 | 590 | } |
duke@435 | 591 | } |
duke@435 | 592 | #endif // PRODUCT |
duke@435 | 593 | |
duke@435 | 594 | // -------------------------------------------------------------------------- |
duke@435 | 595 | |
duke@435 | 596 | #ifdef ASSERT |
duke@435 | 597 | class StableMemoryChecker : public StackObj { |
duke@435 | 598 | enum { _bufsize = wordSize*4 }; |
duke@435 | 599 | |
duke@435 | 600 | address _region; |
duke@435 | 601 | jint _size; |
duke@435 | 602 | u1 _save_buf[_bufsize]; |
duke@435 | 603 | |
duke@435 | 604 | int sample(u1* save_buf) { |
duke@435 | 605 | if (_size <= _bufsize) { |
duke@435 | 606 | memcpy(save_buf, _region, _size); |
duke@435 | 607 | return _size; |
duke@435 | 608 | } else { |
duke@435 | 609 | // copy head and tail |
duke@435 | 610 | memcpy(&save_buf[0], _region, _bufsize/2); |
duke@435 | 611 | memcpy(&save_buf[_bufsize/2], _region + _size - _bufsize/2, _bufsize/2); |
duke@435 | 612 | return (_bufsize/2)*2; |
duke@435 | 613 | } |
duke@435 | 614 | } |
duke@435 | 615 | |
duke@435 | 616 | public: |
duke@435 | 617 | StableMemoryChecker(const void* region, jint size) { |
duke@435 | 618 | _region = (address) region; |
duke@435 | 619 | _size = size; |
duke@435 | 620 | sample(_save_buf); |
duke@435 | 621 | } |
duke@435 | 622 | |
duke@435 | 623 | bool verify() { |
duke@435 | 624 | u1 check_buf[sizeof(_save_buf)]; |
duke@435 | 625 | int check_size = sample(check_buf); |
duke@435 | 626 | return (0 == memcmp(_save_buf, check_buf, check_size)); |
duke@435 | 627 | } |
duke@435 | 628 | |
duke@435 | 629 | void set_region(const void* region) { _region = (address) region; } |
duke@435 | 630 | }; |
duke@435 | 631 | #endif |
duke@435 | 632 | |
duke@435 | 633 | |
duke@435 | 634 | // -------------------------------------------------------------------------- |
duke@435 | 635 | StringTable* StringTable::_the_table = NULL; |
duke@435 | 636 | |
coleenp@3865 | 637 | bool StringTable::_needs_rehashing = false; |
coleenp@3865 | 638 | |
johnc@5277 | 639 | volatile int StringTable::_parallel_claimed_idx = 0; |
johnc@5277 | 640 | |
coleenp@3865 | 641 | // Pick hashing algorithm |
coleenp@3875 | 642 | unsigned int StringTable::hash_string(const jchar* s, int len) { |
coleenp@3865 | 643 | return use_alternate_hashcode() ? AltHashing::murmur3_32(seed(), s, len) : |
brutisso@4335 | 644 | java_lang_String::hash_code(s, len); |
coleenp@3865 | 645 | } |
coleenp@3865 | 646 | |
duke@435 | 647 | oop StringTable::lookup(int index, jchar* name, |
duke@435 | 648 | int len, unsigned int hash) { |
coleenp@3865 | 649 | int count = 0; |
zgu@3900 | 650 | for (HashtableEntry<oop, mtSymbol>* l = bucket(index); l != NULL; l = l->next()) { |
coleenp@3865 | 651 | count++; |
duke@435 | 652 | if (l->hash() == hash) { |
duke@435 | 653 | if (java_lang_String::equals(l->literal(), name, len)) { |
duke@435 | 654 | return l->literal(); |
duke@435 | 655 | } |
duke@435 | 656 | } |
duke@435 | 657 | } |
coleenp@3865 | 658 | // If the bucket size is too deep check if this hash code is insufficient. |
mgerdin@7207 | 659 | if (count >= rehash_count && !needs_rehashing()) { |
coleenp@3865 | 660 | _needs_rehashing = check_rehash_table(count); |
coleenp@3865 | 661 | } |
duke@435 | 662 | return NULL; |
duke@435 | 663 | } |
duke@435 | 664 | |
duke@435 | 665 | |
coleenp@3875 | 666 | oop StringTable::basic_add(int index_arg, Handle string, jchar* name, |
coleenp@3865 | 667 | int len, unsigned int hashValue_arg, TRAPS) { |
duke@435 | 668 | |
duke@435 | 669 | assert(java_lang_String::equals(string(), name, len), |
duke@435 | 670 | "string must be properly initialized"); |
coleenp@3875 | 671 | // Cannot hit a safepoint in this function because the "this" pointer can move. |
coleenp@3875 | 672 | No_Safepoint_Verifier nsv; |
duke@435 | 673 | |
coleenp@3865 | 674 | // Check if the symbol table has been rehashed, if so, need to recalculate |
coleenp@3875 | 675 | // the hash value and index before second lookup. |
coleenp@3875 | 676 | unsigned int hashValue; |
coleenp@3875 | 677 | int index; |
coleenp@3875 | 678 | if (use_alternate_hashcode()) { |
coleenp@3875 | 679 | hashValue = hash_string(name, len); |
coleenp@3875 | 680 | index = hash_to_index(hashValue); |
coleenp@3875 | 681 | } else { |
coleenp@3875 | 682 | hashValue = hashValue_arg; |
coleenp@3875 | 683 | index = index_arg; |
coleenp@3875 | 684 | } |
coleenp@3865 | 685 | |
duke@435 | 686 | // Since look-up was done lock-free, we need to check if another |
duke@435 | 687 | // thread beat us in the race to insert the symbol. |
duke@435 | 688 | |
duke@435 | 689 | oop test = lookup(index, name, len, hashValue); // calls lookup(u1*, int) |
duke@435 | 690 | if (test != NULL) { |
duke@435 | 691 | // Entry already added |
duke@435 | 692 | return test; |
duke@435 | 693 | } |
duke@435 | 694 | |
zgu@3900 | 695 | HashtableEntry<oop, mtSymbol>* entry = new_entry(hashValue, string()); |
duke@435 | 696 | add_entry(index, entry); |
duke@435 | 697 | return string(); |
duke@435 | 698 | } |
duke@435 | 699 | |
duke@435 | 700 | |
coleenp@2497 | 701 | oop StringTable::lookup(Symbol* symbol) { |
duke@435 | 702 | ResourceMark rm; |
duke@435 | 703 | int length; |
duke@435 | 704 | jchar* chars = symbol->as_unicode(length); |
mgerdin@4850 | 705 | return lookup(chars, length); |
mgerdin@4850 | 706 | } |
mgerdin@4850 | 707 | |
stefank@6992 | 708 | // Tell the GC that this string was looked up in the StringTable. |
stefank@6992 | 709 | static void ensure_string_alive(oop string) { |
stefank@6992 | 710 | // A lookup in the StringTable could return an object that was previously |
stefank@6992 | 711 | // considered dead. The SATB part of G1 needs to get notified about this |
stefank@6992 | 712 | // potential resurrection, otherwise the marking might not find the object. |
stefank@6992 | 713 | #if INCLUDE_ALL_GCS |
stefank@6992 | 714 | if (UseG1GC && string != NULL) { |
stefank@6992 | 715 | G1SATBCardTableModRefBS::enqueue(string); |
stefank@6992 | 716 | } |
stefank@6992 | 717 | #endif |
stefank@6992 | 718 | } |
mgerdin@4850 | 719 | |
mgerdin@4850 | 720 | oop StringTable::lookup(jchar* name, int len) { |
mgerdin@4850 | 721 | unsigned int hash = hash_string(name, len); |
mgerdin@4850 | 722 | int index = the_table()->hash_to_index(hash); |
stefank@6992 | 723 | oop string = the_table()->lookup(index, name, len, hash); |
stefank@6992 | 724 | |
stefank@6992 | 725 | ensure_string_alive(string); |
stefank@6992 | 726 | |
stefank@6992 | 727 | return string; |
duke@435 | 728 | } |
duke@435 | 729 | |
duke@435 | 730 | |
duke@435 | 731 | oop StringTable::intern(Handle string_or_null, jchar* name, |
duke@435 | 732 | int len, TRAPS) { |
coleenp@3865 | 733 | unsigned int hashValue = hash_string(name, len); |
duke@435 | 734 | int index = the_table()->hash_to_index(hashValue); |
coleenp@3875 | 735 | oop found_string = the_table()->lookup(index, name, len, hashValue); |
duke@435 | 736 | |
duke@435 | 737 | // Found |
stefank@6992 | 738 | if (found_string != NULL) { |
stefank@6992 | 739 | ensure_string_alive(found_string); |
stefank@6992 | 740 | return found_string; |
stefank@6992 | 741 | } |
coleenp@3875 | 742 | |
coleenp@3875 | 743 | debug_only(StableMemoryChecker smc(name, len * sizeof(name[0]))); |
stefank@5769 | 744 | assert(!Universe::heap()->is_in_reserved(name), |
coleenp@3875 | 745 | "proposed name of symbol must be stable"); |
coleenp@3875 | 746 | |
coleenp@3875 | 747 | Handle string; |
coleenp@3875 | 748 | // try to reuse the string if possible |
coleenp@4037 | 749 | if (!string_or_null.is_null()) { |
coleenp@3875 | 750 | string = string_or_null; |
coleenp@3875 | 751 | } else { |
coleenp@4037 | 752 | string = java_lang_String::create_from_unicode(name, len, CHECK_NULL); |
coleenp@3875 | 753 | } |
coleenp@3875 | 754 | |
pliden@6413 | 755 | #if INCLUDE_ALL_GCS |
pliden@6413 | 756 | if (G1StringDedup::is_enabled()) { |
pliden@6413 | 757 | // Deduplicate the string before it is interned. Note that we should never |
pliden@6413 | 758 | // deduplicate a string after it has been interned. Doing so will counteract |
pliden@6413 | 759 | // compiler optimizations done on e.g. interned string literals. |
pliden@6413 | 760 | G1StringDedup::deduplicate(string()); |
pliden@6413 | 761 | } |
pliden@6413 | 762 | #endif |
pliden@6413 | 763 | |
coleenp@3875 | 764 | // Grab the StringTable_lock before getting the_table() because it could |
coleenp@3875 | 765 | // change at safepoint. |
stefank@6992 | 766 | oop added_or_found; |
stefank@6992 | 767 | { |
stefank@6992 | 768 | MutexLocker ml(StringTable_lock, THREAD); |
stefank@6992 | 769 | // Otherwise, add to symbol to table |
stefank@6992 | 770 | added_or_found = the_table()->basic_add(index, string, name, len, |
stefank@6992 | 771 | hashValue, CHECK_NULL); |
stefank@6992 | 772 | } |
duke@435 | 773 | |
stefank@6992 | 774 | ensure_string_alive(added_or_found); |
stefank@6992 | 775 | |
stefank@6992 | 776 | return added_or_found; |
duke@435 | 777 | } |
duke@435 | 778 | |
coleenp@2497 | 779 | oop StringTable::intern(Symbol* symbol, TRAPS) { |
duke@435 | 780 | if (symbol == NULL) return NULL; |
duke@435 | 781 | ResourceMark rm(THREAD); |
duke@435 | 782 | int length; |
duke@435 | 783 | jchar* chars = symbol->as_unicode(length); |
duke@435 | 784 | Handle string; |
duke@435 | 785 | oop result = intern(string, chars, length, CHECK_NULL); |
duke@435 | 786 | return result; |
duke@435 | 787 | } |
duke@435 | 788 | |
duke@435 | 789 | |
duke@435 | 790 | oop StringTable::intern(oop string, TRAPS) |
duke@435 | 791 | { |
duke@435 | 792 | if (string == NULL) return NULL; |
duke@435 | 793 | ResourceMark rm(THREAD); |
duke@435 | 794 | int length; |
duke@435 | 795 | Handle h_string (THREAD, string); |
hseigel@4987 | 796 | jchar* chars = java_lang_String::as_unicode_string(string, length, CHECK_NULL); |
duke@435 | 797 | oop result = intern(h_string, chars, length, CHECK_NULL); |
duke@435 | 798 | return result; |
duke@435 | 799 | } |
duke@435 | 800 | |
duke@435 | 801 | |
duke@435 | 802 | oop StringTable::intern(const char* utf8_string, TRAPS) { |
duke@435 | 803 | if (utf8_string == NULL) return NULL; |
duke@435 | 804 | ResourceMark rm(THREAD); |
duke@435 | 805 | int length = UTF8::unicode_length(utf8_string); |
duke@435 | 806 | jchar* chars = NEW_RESOURCE_ARRAY(jchar, length); |
duke@435 | 807 | UTF8::convert_to_unicode(utf8_string, chars, length); |
duke@435 | 808 | Handle string; |
duke@435 | 809 | oop result = intern(string, chars, length, CHECK_NULL); |
duke@435 | 810 | return result; |
duke@435 | 811 | } |
duke@435 | 812 | |
tschatzl@6229 | 813 | void StringTable::unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int* processed, int* removed) { |
tschatzl@6229 | 814 | buckets_unlink_or_oops_do(is_alive, f, 0, the_table()->table_size(), processed, removed); |
tschatzl@6229 | 815 | } |
tschatzl@6229 | 816 | |
tschatzl@6229 | 817 | void StringTable::possibly_parallel_unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int* processed, int* removed) { |
coleenp@2497 | 818 | // Readers of the table are unlocked, so we should only be removing |
coleenp@2497 | 819 | // entries at a safepoint. |
coleenp@2497 | 820 | assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint"); |
tschatzl@6229 | 821 | const int limit = the_table()->table_size(); |
stefank@5195 | 822 | |
tschatzl@6229 | 823 | for (;;) { |
tschatzl@6229 | 824 | // Grab next set of buckets to scan |
tschatzl@6229 | 825 | int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize; |
tschatzl@6229 | 826 | if (start_idx >= limit) { |
tschatzl@6229 | 827 | // End of table |
tschatzl@6229 | 828 | break; |
coleenp@2497 | 829 | } |
tschatzl@6229 | 830 | |
tschatzl@6229 | 831 | int end_idx = MIN2(limit, start_idx + ClaimChunkSize); |
tschatzl@6229 | 832 | buckets_unlink_or_oops_do(is_alive, f, start_idx, end_idx, processed, removed); |
coleenp@2497 | 833 | } |
coleenp@2497 | 834 | } |
coleenp@2497 | 835 | |
tschatzl@6229 | 836 | void StringTable::buckets_oops_do(OopClosure* f, int start_idx, int end_idx) { |
johnc@5277 | 837 | const int limit = the_table()->table_size(); |
johnc@5277 | 838 | |
johnc@5277 | 839 | assert(0 <= start_idx && start_idx <= limit, |
tschatzl@6229 | 840 | err_msg("start_idx (" INT32_FORMAT ") is out of bounds", start_idx)); |
johnc@5277 | 841 | assert(0 <= end_idx && end_idx <= limit, |
tschatzl@6229 | 842 | err_msg("end_idx (" INT32_FORMAT ") is out of bounds", end_idx)); |
johnc@5277 | 843 | assert(start_idx <= end_idx, |
tschatzl@6229 | 844 | err_msg("Index ordering: start_idx=" INT32_FORMAT", end_idx=" INT32_FORMAT, |
johnc@5277 | 845 | start_idx, end_idx)); |
johnc@5277 | 846 | |
johnc@5277 | 847 | for (int i = start_idx; i < end_idx; i += 1) { |
zgu@3900 | 848 | HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i); |
coleenp@2497 | 849 | while (entry != NULL) { |
stefank@5195 | 850 | assert(!entry->is_shared(), "CDS not used for the StringTable"); |
stefank@5195 | 851 | |
coleenp@2497 | 852 | f->do_oop((oop*)entry->literal_addr()); |
coleenp@2497 | 853 | |
stefank@5195 | 854 | entry = entry->next(); |
coleenp@2497 | 855 | } |
coleenp@2497 | 856 | } |
coleenp@2497 | 857 | } |
coleenp@2497 | 858 | |
tschatzl@6229 | 859 | void StringTable::buckets_unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* f, int start_idx, int end_idx, int* processed, int* removed) { |
tschatzl@6229 | 860 | const int limit = the_table()->table_size(); |
tschatzl@6229 | 861 | |
tschatzl@6229 | 862 | assert(0 <= start_idx && start_idx <= limit, |
tschatzl@6229 | 863 | err_msg("start_idx (" INT32_FORMAT ") is out of bounds", start_idx)); |
tschatzl@6229 | 864 | assert(0 <= end_idx && end_idx <= limit, |
tschatzl@6229 | 865 | err_msg("end_idx (" INT32_FORMAT ") is out of bounds", end_idx)); |
tschatzl@6229 | 866 | assert(start_idx <= end_idx, |
tschatzl@6229 | 867 | err_msg("Index ordering: start_idx=" INT32_FORMAT", end_idx=" INT32_FORMAT, |
tschatzl@6229 | 868 | start_idx, end_idx)); |
tschatzl@6229 | 869 | |
tschatzl@6229 | 870 | for (int i = start_idx; i < end_idx; ++i) { |
tschatzl@6229 | 871 | HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i); |
tschatzl@6229 | 872 | HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i); |
tschatzl@6229 | 873 | while (entry != NULL) { |
tschatzl@6229 | 874 | assert(!entry->is_shared(), "CDS not used for the StringTable"); |
tschatzl@6229 | 875 | |
tschatzl@6229 | 876 | if (is_alive->do_object_b(entry->literal())) { |
tschatzl@6229 | 877 | if (f != NULL) { |
tschatzl@6229 | 878 | f->do_oop((oop*)entry->literal_addr()); |
tschatzl@6229 | 879 | } |
tschatzl@6229 | 880 | p = entry->next_addr(); |
tschatzl@6229 | 881 | } else { |
tschatzl@6229 | 882 | *p = entry->next(); |
tschatzl@6229 | 883 | the_table()->free_entry(entry); |
tschatzl@6229 | 884 | (*removed)++; |
tschatzl@6229 | 885 | } |
tschatzl@6229 | 886 | (*processed)++; |
tschatzl@6229 | 887 | entry = *p; |
tschatzl@6229 | 888 | } |
tschatzl@6229 | 889 | } |
tschatzl@6229 | 890 | } |
tschatzl@6229 | 891 | |
johnc@5277 | 892 | void StringTable::oops_do(OopClosure* f) { |
tschatzl@6229 | 893 | buckets_oops_do(f, 0, the_table()->table_size()); |
johnc@5277 | 894 | } |
johnc@5277 | 895 | |
johnc@5277 | 896 | void StringTable::possibly_parallel_oops_do(OopClosure* f) { |
johnc@5277 | 897 | const int limit = the_table()->table_size(); |
johnc@5277 | 898 | |
johnc@5277 | 899 | for (;;) { |
johnc@5277 | 900 | // Grab next set of buckets to scan |
johnc@5277 | 901 | int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize; |
johnc@5277 | 902 | if (start_idx >= limit) { |
johnc@5277 | 903 | // End of table |
johnc@5277 | 904 | break; |
johnc@5277 | 905 | } |
johnc@5277 | 906 | |
johnc@5277 | 907 | int end_idx = MIN2(limit, start_idx + ClaimChunkSize); |
tschatzl@6229 | 908 | buckets_oops_do(f, start_idx, end_idx); |
johnc@5277 | 909 | } |
johnc@5277 | 910 | } |
johnc@5277 | 911 | |
dcubed@5743 | 912 | // This verification is part of Universe::verify() and needs to be quick. |
dcubed@5743 | 913 | // See StringTable::verify_and_compare() below for exhaustive verification. |
duke@435 | 914 | void StringTable::verify() { |
duke@435 | 915 | for (int i = 0; i < the_table()->table_size(); ++i) { |
zgu@3900 | 916 | HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i); |
duke@435 | 917 | for ( ; p != NULL; p = p->next()) { |
duke@435 | 918 | oop s = p->literal(); |
duke@435 | 919 | guarantee(s != NULL, "interned string is NULL"); |
never@2700 | 920 | unsigned int h = java_lang_String::hash_string(s); |
duke@435 | 921 | guarantee(p->hash() == h, "broken hash in string table entry"); |
duke@435 | 922 | guarantee(the_table()->hash_to_index(h) == i, |
duke@435 | 923 | "wrong index in string table"); |
duke@435 | 924 | } |
duke@435 | 925 | } |
duke@435 | 926 | } |
coleenp@3865 | 927 | |
coleenp@3865 | 928 | void StringTable::dump(outputStream* st) { |
iklam@5144 | 929 | the_table()->dump_table(st, "StringTable"); |
coleenp@3865 | 930 | } |
coleenp@3865 | 931 | |
dcubed@5743 | 932 | StringTable::VerifyRetTypes StringTable::compare_entries( |
dcubed@5743 | 933 | int bkt1, int e_cnt1, |
dcubed@5743 | 934 | HashtableEntry<oop, mtSymbol>* e_ptr1, |
dcubed@5743 | 935 | int bkt2, int e_cnt2, |
dcubed@5743 | 936 | HashtableEntry<oop, mtSymbol>* e_ptr2) { |
dcubed@5743 | 937 | // These entries are sanity checked by verify_and_compare_entries() |
dcubed@5743 | 938 | // before this function is called. |
dcubed@5743 | 939 | oop str1 = e_ptr1->literal(); |
dcubed@5743 | 940 | oop str2 = e_ptr2->literal(); |
dcubed@5743 | 941 | |
dcubed@5743 | 942 | if (str1 == str2) { |
dcubed@5743 | 943 | tty->print_cr("ERROR: identical oop values (0x" PTR_FORMAT ") " |
dcubed@5743 | 944 | "in entry @ bucket[%d][%d] and entry @ bucket[%d][%d]", |
hseigel@5784 | 945 | (void *)str1, bkt1, e_cnt1, bkt2, e_cnt2); |
dcubed@5743 | 946 | return _verify_fail_continue; |
dcubed@5743 | 947 | } |
dcubed@5743 | 948 | |
dcubed@5743 | 949 | if (java_lang_String::equals(str1, str2)) { |
dcubed@5743 | 950 | tty->print_cr("ERROR: identical String values in entry @ " |
dcubed@5743 | 951 | "bucket[%d][%d] and entry @ bucket[%d][%d]", |
dcubed@5743 | 952 | bkt1, e_cnt1, bkt2, e_cnt2); |
dcubed@5743 | 953 | return _verify_fail_continue; |
dcubed@5743 | 954 | } |
dcubed@5743 | 955 | |
dcubed@5743 | 956 | return _verify_pass; |
dcubed@5743 | 957 | } |
dcubed@5743 | 958 | |
dcubed@5743 | 959 | StringTable::VerifyRetTypes StringTable::verify_entry(int bkt, int e_cnt, |
dcubed@5743 | 960 | HashtableEntry<oop, mtSymbol>* e_ptr, |
dcubed@5743 | 961 | StringTable::VerifyMesgModes mesg_mode) { |
dcubed@5743 | 962 | |
dcubed@5743 | 963 | VerifyRetTypes ret = _verify_pass; // be optimistic |
dcubed@5743 | 964 | |
dcubed@5743 | 965 | oop str = e_ptr->literal(); |
dcubed@5743 | 966 | if (str == NULL) { |
dcubed@5743 | 967 | if (mesg_mode == _verify_with_mesgs) { |
dcubed@5743 | 968 | tty->print_cr("ERROR: NULL oop value in entry @ bucket[%d][%d]", bkt, |
dcubed@5743 | 969 | e_cnt); |
dcubed@5743 | 970 | } |
dcubed@5743 | 971 | // NULL oop means no more verifications are possible |
dcubed@5743 | 972 | return _verify_fail_done; |
dcubed@5743 | 973 | } |
dcubed@5743 | 974 | |
dcubed@5743 | 975 | if (str->klass() != SystemDictionary::String_klass()) { |
dcubed@5743 | 976 | if (mesg_mode == _verify_with_mesgs) { |
dcubed@5743 | 977 | tty->print_cr("ERROR: oop is not a String in entry @ bucket[%d][%d]", |
dcubed@5743 | 978 | bkt, e_cnt); |
dcubed@5743 | 979 | } |
dcubed@5743 | 980 | // not a String means no more verifications are possible |
dcubed@5743 | 981 | return _verify_fail_done; |
dcubed@5743 | 982 | } |
dcubed@5743 | 983 | |
dcubed@5743 | 984 | unsigned int h = java_lang_String::hash_string(str); |
dcubed@5743 | 985 | if (e_ptr->hash() != h) { |
dcubed@5743 | 986 | if (mesg_mode == _verify_with_mesgs) { |
dcubed@5743 | 987 | tty->print_cr("ERROR: broken hash value in entry @ bucket[%d][%d], " |
dcubed@5743 | 988 | "bkt_hash=%d, str_hash=%d", bkt, e_cnt, e_ptr->hash(), h); |
dcubed@5743 | 989 | } |
dcubed@5743 | 990 | ret = _verify_fail_continue; |
dcubed@5743 | 991 | } |
dcubed@5743 | 992 | |
dcubed@5743 | 993 | if (the_table()->hash_to_index(h) != bkt) { |
dcubed@5743 | 994 | if (mesg_mode == _verify_with_mesgs) { |
dcubed@5743 | 995 | tty->print_cr("ERROR: wrong index value for entry @ bucket[%d][%d], " |
dcubed@5743 | 996 | "str_hash=%d, hash_to_index=%d", bkt, e_cnt, h, |
dcubed@5743 | 997 | the_table()->hash_to_index(h)); |
dcubed@5743 | 998 | } |
dcubed@5743 | 999 | ret = _verify_fail_continue; |
dcubed@5743 | 1000 | } |
dcubed@5743 | 1001 | |
dcubed@5743 | 1002 | return ret; |
dcubed@5743 | 1003 | } |
dcubed@5743 | 1004 | |
dcubed@5743 | 1005 | // See StringTable::verify() above for the quick verification that is |
dcubed@5743 | 1006 | // part of Universe::verify(). This verification is exhaustive and |
dcubed@5743 | 1007 | // reports on every issue that is found. StringTable::verify() only |
dcubed@5743 | 1008 | // reports on the first issue that is found. |
dcubed@5743 | 1009 | // |
dcubed@5743 | 1010 | // StringTable::verify_entry() checks: |
dcubed@5743 | 1011 | // - oop value != NULL (same as verify()) |
dcubed@5743 | 1012 | // - oop value is a String |
dcubed@5743 | 1013 | // - hash(String) == hash in entry (same as verify()) |
dcubed@5743 | 1014 | // - index for hash == index of entry (same as verify()) |
dcubed@5743 | 1015 | // |
dcubed@5743 | 1016 | // StringTable::compare_entries() checks: |
dcubed@5743 | 1017 | // - oops are unique across all entries |
dcubed@5743 | 1018 | // - String values are unique across all entries |
dcubed@5743 | 1019 | // |
dcubed@5743 | 1020 | int StringTable::verify_and_compare_entries() { |
dcubed@5743 | 1021 | assert(StringTable_lock->is_locked(), "sanity check"); |
dcubed@5743 | 1022 | |
dcubed@5743 | 1023 | int fail_cnt = 0; |
dcubed@5743 | 1024 | |
dcubed@5743 | 1025 | // first, verify all the entries individually: |
dcubed@5743 | 1026 | for (int bkt = 0; bkt < the_table()->table_size(); bkt++) { |
dcubed@5743 | 1027 | HashtableEntry<oop, mtSymbol>* e_ptr = the_table()->bucket(bkt); |
dcubed@5743 | 1028 | for (int e_cnt = 0; e_ptr != NULL; e_ptr = e_ptr->next(), e_cnt++) { |
dcubed@5743 | 1029 | VerifyRetTypes ret = verify_entry(bkt, e_cnt, e_ptr, _verify_with_mesgs); |
dcubed@5743 | 1030 | if (ret != _verify_pass) { |
dcubed@5743 | 1031 | fail_cnt++; |
dcubed@5743 | 1032 | } |
dcubed@5743 | 1033 | } |
dcubed@5743 | 1034 | } |
dcubed@5743 | 1035 | |
dcubed@5743 | 1036 | // Optimization: if the above check did not find any failures, then |
dcubed@5743 | 1037 | // the comparison loop below does not need to call verify_entry() |
dcubed@5743 | 1038 | // before calling compare_entries(). If there were failures, then we |
dcubed@5743 | 1039 | // have to call verify_entry() to see if the entry can be passed to |
dcubed@5743 | 1040 | // compare_entries() safely. When we call verify_entry() in the loop |
dcubed@5743 | 1041 | // below, we do so quietly to void duplicate messages and we don't |
dcubed@5743 | 1042 | // increment fail_cnt because the failures have already been counted. |
dcubed@5743 | 1043 | bool need_entry_verify = (fail_cnt != 0); |
dcubed@5743 | 1044 | |
dcubed@5743 | 1045 | // second, verify all entries relative to each other: |
dcubed@5743 | 1046 | for (int bkt1 = 0; bkt1 < the_table()->table_size(); bkt1++) { |
dcubed@5743 | 1047 | HashtableEntry<oop, mtSymbol>* e_ptr1 = the_table()->bucket(bkt1); |
dcubed@5743 | 1048 | for (int e_cnt1 = 0; e_ptr1 != NULL; e_ptr1 = e_ptr1->next(), e_cnt1++) { |
dcubed@5743 | 1049 | if (need_entry_verify) { |
dcubed@5743 | 1050 | VerifyRetTypes ret = verify_entry(bkt1, e_cnt1, e_ptr1, |
dcubed@5743 | 1051 | _verify_quietly); |
dcubed@5743 | 1052 | if (ret == _verify_fail_done) { |
dcubed@5743 | 1053 | // cannot use the current entry to compare against other entries |
dcubed@5743 | 1054 | continue; |
dcubed@5743 | 1055 | } |
dcubed@5743 | 1056 | } |
dcubed@5743 | 1057 | |
dcubed@5743 | 1058 | for (int bkt2 = bkt1; bkt2 < the_table()->table_size(); bkt2++) { |
dcubed@5743 | 1059 | HashtableEntry<oop, mtSymbol>* e_ptr2 = the_table()->bucket(bkt2); |
dcubed@5743 | 1060 | int e_cnt2; |
dcubed@5743 | 1061 | for (e_cnt2 = 0; e_ptr2 != NULL; e_ptr2 = e_ptr2->next(), e_cnt2++) { |
dcubed@5743 | 1062 | if (bkt1 == bkt2 && e_cnt2 <= e_cnt1) { |
dcubed@5743 | 1063 | // skip the entries up to and including the one that |
dcubed@5743 | 1064 | // we're comparing against |
dcubed@5743 | 1065 | continue; |
dcubed@5743 | 1066 | } |
dcubed@5743 | 1067 | |
dcubed@5743 | 1068 | if (need_entry_verify) { |
dcubed@5743 | 1069 | VerifyRetTypes ret = verify_entry(bkt2, e_cnt2, e_ptr2, |
dcubed@5743 | 1070 | _verify_quietly); |
dcubed@5743 | 1071 | if (ret == _verify_fail_done) { |
dcubed@5743 | 1072 | // cannot compare against this entry |
dcubed@5743 | 1073 | continue; |
dcubed@5743 | 1074 | } |
dcubed@5743 | 1075 | } |
dcubed@5743 | 1076 | |
dcubed@5743 | 1077 | // compare two entries, report and count any failures: |
dcubed@5743 | 1078 | if (compare_entries(bkt1, e_cnt1, e_ptr1, bkt2, e_cnt2, e_ptr2) |
dcubed@5743 | 1079 | != _verify_pass) { |
dcubed@5743 | 1080 | fail_cnt++; |
dcubed@5743 | 1081 | } |
dcubed@5743 | 1082 | } |
dcubed@5743 | 1083 | } |
dcubed@5743 | 1084 | } |
dcubed@5743 | 1085 | } |
dcubed@5743 | 1086 | return fail_cnt; |
dcubed@5743 | 1087 | } |
coleenp@3865 | 1088 | |
coleenp@3865 | 1089 | // Create a new table and using alternate hash code, populate the new table |
coleenp@3865 | 1090 | // with the existing strings. Set flag to use the alternate hash code afterwards. |
coleenp@3865 | 1091 | void StringTable::rehash_table() { |
coleenp@3865 | 1092 | assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint"); |
coleenp@3875 | 1093 | // This should never happen with -Xshare:dump but it might in testing mode. |
coleenp@3875 | 1094 | if (DumpSharedSpaces) return; |
coleenp@3865 | 1095 | StringTable* new_table = new StringTable(); |
coleenp@3865 | 1096 | |
coleenp@3865 | 1097 | // Rehash the table |
coleenp@3865 | 1098 | the_table()->move_to(new_table); |
coleenp@3865 | 1099 | |
coleenp@3865 | 1100 | // Delete the table and buckets (entries are reused in new table). |
coleenp@3865 | 1101 | delete _the_table; |
coleenp@3865 | 1102 | // Don't check if we need rehashing until the table gets unbalanced again. |
coleenp@3865 | 1103 | // Then rehash with a new global seed. |
coleenp@3865 | 1104 | _needs_rehashing = false; |
coleenp@3865 | 1105 | _the_table = new_table; |
coleenp@3865 | 1106 | } |