src/share/vm/classfile/symbolTable.cpp

Mon, 29 Apr 2013 16:13:57 -0400

author
hseigel
date
Mon, 29 Apr 2013 16:13:57 -0400
changeset 4987
f258c5828eb8
parent 4850
ede380e13960
child 5144
a5d6f0c3585f
permissions
-rw-r--r--

8011773: Some tests on Interned String crashed JVM with OOM
Summary: Instead of terminating the VM, throw OutOfMemoryError exceptions.
Reviewed-by: coleenp, dholmes

     1 /*
     2  * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
     3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4  *
     5  * This code is free software; you can redistribute it and/or modify it
     6  * under the terms of the GNU General Public License version 2 only, as
     7  * published by the Free Software Foundation.
     8  *
     9  * This code is distributed in the hope that it will be useful, but WITHOUT
    10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    12  * version 2 for more details (a copy is included in the LICENSE file that
    13  * accompanied this code).
    14  *
    15  * You should have received a copy of the GNU General Public License version
    16  * 2 along with this work; if not, write to the Free Software Foundation,
    17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    18  *
    19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    20  * or visit www.oracle.com if you need additional information or have any
    21  * questions.
    22  *
    23  */
    25 #include "precompiled.hpp"
    26 #include "classfile/altHashing.hpp"
    27 #include "classfile/javaClasses.hpp"
    28 #include "classfile/symbolTable.hpp"
    29 #include "classfile/systemDictionary.hpp"
    30 #include "gc_interface/collectedHeap.inline.hpp"
    31 #include "memory/allocation.inline.hpp"
    32 #include "memory/filemap.hpp"
    33 #include "memory/gcLocker.inline.hpp"
    34 #include "oops/oop.inline.hpp"
    35 #include "oops/oop.inline2.hpp"
    36 #include "runtime/mutexLocker.hpp"
    37 #include "utilities/hashtable.inline.hpp"
    38 #include "utilities/numberSeq.hpp"
    40 // --------------------------------------------------------------------------
    42 SymbolTable* SymbolTable::_the_table = NULL;
    43 // Static arena for symbols that are not deallocated
    44 Arena* SymbolTable::_arena = NULL;
    45 bool SymbolTable::_needs_rehashing = false;
    47 Symbol* SymbolTable::allocate_symbol(const u1* name, int len, bool c_heap, TRAPS) {
    48   assert (len <= Symbol::max_length(), "should be checked by caller");
    50   Symbol* sym;
    52   if (DumpSharedSpaces) {
    53     // Allocate all symbols to CLD shared metaspace
    54     sym = new (len, ClassLoaderData::the_null_class_loader_data(), THREAD) Symbol(name, len, -1);
    55   } else if (c_heap) {
    56     // refcount starts as 1
    57     sym = new (len, THREAD) Symbol(name, len, 1);
    58     assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted");
    59   } else {
    60     // Allocate to global arena
    61     sym = new (len, arena(), THREAD) Symbol(name, len, -1);
    62   }
    63   return sym;
    64 }
    66 void SymbolTable::initialize_symbols(int arena_alloc_size) {
    67   // Initialize the arena for global symbols, size passed in depends on CDS.
    68   if (arena_alloc_size == 0) {
    69     _arena = new (mtSymbol) Arena();
    70   } else {
    71     _arena = new (mtSymbol) Arena(arena_alloc_size);
    72   }
    73 }
    75 // Call function for all symbols in the symbol table.
    76 void SymbolTable::symbols_do(SymbolClosure *cl) {
    77   const int n = the_table()->table_size();
    78   for (int i = 0; i < n; i++) {
    79     for (HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
    80          p != NULL;
    81          p = p->next()) {
    82       cl->do_symbol(p->literal_addr());
    83     }
    84   }
    85 }
    87 int SymbolTable::symbols_removed = 0;
    88 int SymbolTable::symbols_counted = 0;
    90 // Remove unreferenced symbols from the symbol table
    91 // This is done late during GC.
    92 void SymbolTable::unlink() {
    93   int removed = 0;
    94   int total = 0;
    95   size_t memory_total = 0;
    96   for (int i = 0; i < the_table()->table_size(); ++i) {
    97     HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
    98     HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
    99     while (entry != NULL) {
   100       // Shared entries are normally at the end of the bucket and if we run into
   101       // a shared entry, then there is nothing more to remove. However, if we
   102       // have rehashed the table, then the shared entries are no longer at the
   103       // end of the bucket.
   104       if (entry->is_shared() && !use_alternate_hashcode()) {
   105         break;
   106       }
   107       Symbol* s = entry->literal();
   108       memory_total += s->size();
   109       total++;
   110       assert(s != NULL, "just checking");
   111       // If reference count is zero, remove.
   112       if (s->refcount() == 0) {
   113         assert(!entry->is_shared(), "shared entries should be kept live");
   114         delete s;
   115         removed++;
   116         *p = entry->next();
   117         the_table()->free_entry(entry);
   118       } else {
   119         p = entry->next_addr();
   120       }
   121       // get next entry
   122       entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
   123     }
   124   }
   125   symbols_removed += removed;
   126   symbols_counted += total;
   127   // Exclude printing for normal PrintGCDetails because people parse
   128   // this output.
   129   if (PrintGCDetails && Verbose && WizardMode) {
   130     gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", total,
   131                         (memory_total*HeapWordSize)/1024);
   132   }
   133 }
   135 // Create a new table and using alternate hash code, populate the new table
   136 // with the existing strings.   Set flag to use the alternate hash code afterwards.
   137 void SymbolTable::rehash_table() {
   138   assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
   139   // This should never happen with -Xshare:dump but it might in testing mode.
   140   if (DumpSharedSpaces) return;
   141   // Create a new symbol table
   142   SymbolTable* new_table = new SymbolTable();
   144   the_table()->move_to(new_table);
   146   // Delete the table and buckets (entries are reused in new table).
   147   delete _the_table;
   148   // Don't check if we need rehashing until the table gets unbalanced again.
   149   // Then rehash with a new global seed.
   150   _needs_rehashing = false;
   151   _the_table = new_table;
   152 }
   154 // Lookup a symbol in a bucket.
   156 Symbol* SymbolTable::lookup(int index, const char* name,
   157                               int len, unsigned int hash) {
   158   int count = 0;
   159   for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) {
   160     count++;  // count all entries in this bucket, not just ones with same hash
   161     if (e->hash() == hash) {
   162       Symbol* sym = e->literal();
   163       if (sym->equals(name, len)) {
   164         // something is referencing this symbol now.
   165         sym->increment_refcount();
   166         return sym;
   167       }
   168     }
   169   }
   170   // If the bucket size is too deep check if this hash code is insufficient.
   171   if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
   172     _needs_rehashing = check_rehash_table(count);
   173   }
   174   return NULL;
   175 }
   177 // Pick hashing algorithm.
   178 unsigned int SymbolTable::hash_symbol(const char* s, int len) {
   179   return use_alternate_hashcode() ?
   180            AltHashing::murmur3_32(seed(), (const jbyte*)s, len) :
   181            java_lang_String::hash_code(s, len);
   182 }
   185 // We take care not to be blocking while holding the
   186 // SymbolTable_lock. Otherwise, the system might deadlock, since the
   187 // symboltable is used during compilation (VM_thread) The lock free
   188 // synchronization is simplified by the fact that we do not delete
   189 // entries in the symbol table during normal execution (only during
   190 // safepoints).
   192 Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) {
   193   unsigned int hashValue = hash_symbol(name, len);
   194   int index = the_table()->hash_to_index(hashValue);
   196   Symbol* s = the_table()->lookup(index, name, len, hashValue);
   198   // Found
   199   if (s != NULL) return s;
   201   // Grab SymbolTable_lock first.
   202   MutexLocker ml(SymbolTable_lock, THREAD);
   204   // Otherwise, add to symbol to table
   205   return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL);
   206 }
   208 Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) {
   209   char* buffer;
   210   int index, len;
   211   unsigned int hashValue;
   212   char* name;
   213   {
   214     debug_only(No_Safepoint_Verifier nsv;)
   216     name = (char*)sym->base() + begin;
   217     len = end - begin;
   218     hashValue = hash_symbol(name, len);
   219     index = the_table()->hash_to_index(hashValue);
   220     Symbol* s = the_table()->lookup(index, name, len, hashValue);
   222     // Found
   223     if (s != NULL) return s;
   224   }
   226   // Otherwise, add to symbol to table. Copy to a C string first.
   227   char stack_buf[128];
   228   ResourceMark rm(THREAD);
   229   if (len <= 128) {
   230     buffer = stack_buf;
   231   } else {
   232     buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len);
   233   }
   234   for (int i=0; i<len; i++) {
   235     buffer[i] = name[i];
   236   }
   237   // Make sure there is no safepoint in the code above since name can't move.
   238   // We can't include the code in No_Safepoint_Verifier because of the
   239   // ResourceMark.
   241   // Grab SymbolTable_lock first.
   242   MutexLocker ml(SymbolTable_lock, THREAD);
   244   return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL);
   245 }
   247 Symbol* SymbolTable::lookup_only(const char* name, int len,
   248                                    unsigned int& hash) {
   249   hash = hash_symbol(name, len);
   250   int index = the_table()->hash_to_index(hash);
   252   Symbol* s = the_table()->lookup(index, name, len, hash);
   253   return s;
   254 }
   256 // Look up the address of the literal in the SymbolTable for this Symbol*
   257 // Do not create any new symbols
   258 // Do not increment the reference count to keep this alive
   259 Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){
   260   unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length());
   261   int index = the_table()->hash_to_index(hash);
   263   for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) {
   264     if (e->hash() == hash) {
   265       Symbol* literal_sym = e->literal();
   266       if (sym == literal_sym) {
   267         return e->literal_addr();
   268       }
   269     }
   270   }
   271   return NULL;
   272 }
   274 // Suggestion: Push unicode-based lookup all the way into the hashing
   275 // and probing logic, so there is no need for convert_to_utf8 until
   276 // an actual new Symbol* is created.
   277 Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) {
   278   int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
   279   char stack_buf[128];
   280   if (utf8_length < (int) sizeof(stack_buf)) {
   281     char* chars = stack_buf;
   282     UNICODE::convert_to_utf8(name, utf16_length, chars);
   283     return lookup(chars, utf8_length, THREAD);
   284   } else {
   285     ResourceMark rm(THREAD);
   286     char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
   287     UNICODE::convert_to_utf8(name, utf16_length, chars);
   288     return lookup(chars, utf8_length, THREAD);
   289   }
   290 }
   292 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
   293                                            unsigned int& hash) {
   294   int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
   295   char stack_buf[128];
   296   if (utf8_length < (int) sizeof(stack_buf)) {
   297     char* chars = stack_buf;
   298     UNICODE::convert_to_utf8(name, utf16_length, chars);
   299     return lookup_only(chars, utf8_length, hash);
   300   } else {
   301     ResourceMark rm;
   302     char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
   303     UNICODE::convert_to_utf8(name, utf16_length, chars);
   304     return lookup_only(chars, utf8_length, hash);
   305   }
   306 }
   308 void SymbolTable::add(ClassLoaderData* loader_data, constantPoolHandle cp,
   309                       int names_count,
   310                       const char** names, int* lengths, int* cp_indices,
   311                       unsigned int* hashValues, TRAPS) {
   312   // Grab SymbolTable_lock first.
   313   MutexLocker ml(SymbolTable_lock, THREAD);
   315   SymbolTable* table = the_table();
   316   bool added = table->basic_add(loader_data, cp, names_count, names, lengths,
   317                                 cp_indices, hashValues, CHECK);
   318   if (!added) {
   319     // do it the hard way
   320     for (int i=0; i<names_count; i++) {
   321       int index = table->hash_to_index(hashValues[i]);
   322       bool c_heap = !loader_data->is_the_null_class_loader_data();
   323       Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK);
   324       cp->symbol_at_put(cp_indices[i], sym);
   325     }
   326   }
   327 }
   329 Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) {
   330   unsigned int hash;
   331   Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash);
   332   if (result != NULL) {
   333     return result;
   334   }
   335   // Grab SymbolTable_lock first.
   336   MutexLocker ml(SymbolTable_lock, THREAD);
   338   SymbolTable* table = the_table();
   339   int index = table->hash_to_index(hash);
   340   return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD);
   341 }
   343 Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len,
   344                                unsigned int hashValue_arg, bool c_heap, TRAPS) {
   345   assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),
   346          "proposed name of symbol must be stable");
   348   // Don't allow symbols to be created which cannot fit in a Symbol*.
   349   if (len > Symbol::max_length()) {
   350     THROW_MSG_0(vmSymbols::java_lang_InternalError(),
   351                 "name is too long to represent");
   352   }
   354   // Cannot hit a safepoint in this function because the "this" pointer can move.
   355   No_Safepoint_Verifier nsv;
   357   // Check if the symbol table has been rehashed, if so, need to recalculate
   358   // the hash value and index.
   359   unsigned int hashValue;
   360   int index;
   361   if (use_alternate_hashcode()) {
   362     hashValue = hash_symbol((const char*)name, len);
   363     index = hash_to_index(hashValue);
   364   } else {
   365     hashValue = hashValue_arg;
   366     index = index_arg;
   367   }
   369   // Since look-up was done lock-free, we need to check if another
   370   // thread beat us in the race to insert the symbol.
   371   Symbol* test = lookup(index, (char*)name, len, hashValue);
   372   if (test != NULL) {
   373     // A race occurred and another thread introduced the symbol.
   374     assert(test->refcount() != 0, "lookup should have incremented the count");
   375     return test;
   376   }
   378   // Create a new symbol.
   379   Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL);
   380   assert(sym->equals((char*)name, len), "symbol must be properly initialized");
   382   HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
   383   add_entry(index, entry);
   384   return sym;
   385 }
   387 // This version of basic_add adds symbols in batch from the constant pool
   388 // parsing.
   389 bool SymbolTable::basic_add(ClassLoaderData* loader_data, constantPoolHandle cp,
   390                             int names_count,
   391                             const char** names, int* lengths,
   392                             int* cp_indices, unsigned int* hashValues,
   393                             TRAPS) {
   395   // Check symbol names are not too long.  If any are too long, don't add any.
   396   for (int i = 0; i< names_count; i++) {
   397     if (lengths[i] > Symbol::max_length()) {
   398       THROW_MSG_0(vmSymbols::java_lang_InternalError(),
   399                   "name is too long to represent");
   400     }
   401   }
   403   // Cannot hit a safepoint in this function because the "this" pointer can move.
   404   No_Safepoint_Verifier nsv;
   406   for (int i=0; i<names_count; i++) {
   407     // Check if the symbol table has been rehashed, if so, need to recalculate
   408     // the hash value.
   409     unsigned int hashValue;
   410     if (use_alternate_hashcode()) {
   411       hashValue = hash_symbol(names[i], lengths[i]);
   412     } else {
   413       hashValue = hashValues[i];
   414     }
   415     // Since look-up was done lock-free, we need to check if another
   416     // thread beat us in the race to insert the symbol.
   417     int index = hash_to_index(hashValue);
   418     Symbol* test = lookup(index, names[i], lengths[i], hashValue);
   419     if (test != NULL) {
   420       // A race occurred and another thread introduced the symbol, this one
   421       // will be dropped and collected. Use test instead.
   422       cp->symbol_at_put(cp_indices[i], test);
   423       assert(test->refcount() != 0, "lookup should have incremented the count");
   424     } else {
   425       // Create a new symbol.  The null class loader is never unloaded so these
   426       // are allocated specially in a permanent arena.
   427       bool c_heap = !loader_data->is_the_null_class_loader_data();
   428       Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false));
   429       assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized");  // why wouldn't it be???
   430       HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
   431       add_entry(index, entry);
   432       cp->symbol_at_put(cp_indices[i], sym);
   433     }
   434   }
   435   return true;
   436 }
   439 void SymbolTable::verify() {
   440   for (int i = 0; i < the_table()->table_size(); ++i) {
   441     HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
   442     for ( ; p != NULL; p = p->next()) {
   443       Symbol* s = (Symbol*)(p->literal());
   444       guarantee(s != NULL, "symbol is NULL");
   445       unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length());
   446       guarantee(p->hash() == h, "broken hash in symbol table entry");
   447       guarantee(the_table()->hash_to_index(h) == i,
   448                 "wrong index in symbol table");
   449     }
   450   }
   451 }
   453 void SymbolTable::dump(outputStream* st) {
   454   NumberSeq summary;
   455   for (int i = 0; i < the_table()->table_size(); ++i) {
   456     int count = 0;
   457     for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(i);
   458        e != NULL; e = e->next()) {
   459       count++;
   460     }
   461     summary.add((double)count);
   462   }
   463   st->print_cr("SymbolTable statistics:");
   464   st->print_cr("Number of buckets       : %7d", summary.num());
   465   st->print_cr("Average bucket size     : %7.0f", summary.avg());
   466   st->print_cr("Variance of bucket size : %7.0f", summary.variance());
   467   st->print_cr("Std. dev. of bucket size: %7.0f", summary.sd());
   468   st->print_cr("Maximum bucket size     : %7.0f", summary.maximum());
   469 }
   472 //---------------------------------------------------------------------------
   473 // Non-product code
   475 #ifndef PRODUCT
   477 void SymbolTable::print_histogram() {
   478   MutexLocker ml(SymbolTable_lock);
   479   const int results_length = 100;
   480   int results[results_length];
   481   int i,j;
   483   // initialize results to zero
   484   for (j = 0; j < results_length; j++) {
   485     results[j] = 0;
   486   }
   488   int total = 0;
   489   int max_symbols = 0;
   490   int out_of_range = 0;
   491   int memory_total = 0;
   492   int count = 0;
   493   for (i = 0; i < the_table()->table_size(); i++) {
   494     HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
   495     for ( ; p != NULL; p = p->next()) {
   496       memory_total += p->literal()->size();
   497       count++;
   498       int counter = p->literal()->utf8_length();
   499       total += counter;
   500       if (counter < results_length) {
   501         results[counter]++;
   502       } else {
   503         out_of_range++;
   504       }
   505       max_symbols = MAX2(max_symbols, counter);
   506     }
   507   }
   508   tty->print_cr("Symbol Table:");
   509   tty->print_cr("Total number of symbols  %5d", count);
   510   tty->print_cr("Total size in memory     %5dK",
   511           (memory_total*HeapWordSize)/1024);
   512   tty->print_cr("Total counted            %5d", symbols_counted);
   513   tty->print_cr("Total removed            %5d", symbols_removed);
   514   if (symbols_counted > 0) {
   515     tty->print_cr("Percent removed          %3.2f",
   516           ((float)symbols_removed/(float)symbols_counted)* 100);
   517   }
   518   tty->print_cr("Reference counts         %5d", Symbol::_total_count);
   519   tty->print_cr("Symbol arena size        %5d used %5d",
   520                  arena()->size_in_bytes(), arena()->used());
   521   tty->print_cr("Histogram of symbol length:");
   522   tty->print_cr("%8s %5d", "Total  ", total);
   523   tty->print_cr("%8s %5d", "Maximum", max_symbols);
   524   tty->print_cr("%8s %3.2f", "Average",
   525           ((float) total / (float) the_table()->table_size()));
   526   tty->print_cr("%s", "Histogram:");
   527   tty->print_cr(" %s %29s", "Length", "Number chains that length");
   528   for (i = 0; i < results_length; i++) {
   529     if (results[i] > 0) {
   530       tty->print_cr("%6d %10d", i, results[i]);
   531     }
   532   }
   533   if (Verbose) {
   534     int line_length = 70;
   535     tty->print_cr("%s %30s", " Length", "Number chains that length");
   536     for (i = 0; i < results_length; i++) {
   537       if (results[i] > 0) {
   538         tty->print("%4d", i);
   539         for (j = 0; (j < results[i]) && (j < line_length);  j++) {
   540           tty->print("%1s", "*");
   541         }
   542         if (j == line_length) {
   543           tty->print("%1s", "+");
   544         }
   545         tty->cr();
   546       }
   547     }
   548   }
   549   tty->print_cr(" %s %d: %d\n", "Number chains longer than",
   550                     results_length, out_of_range);
   551 }
   553 void SymbolTable::print() {
   554   for (int i = 0; i < the_table()->table_size(); ++i) {
   555     HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
   556     HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
   557     if (entry != NULL) {
   558       while (entry != NULL) {
   559         tty->print(PTR_FORMAT " ", entry->literal());
   560         entry->literal()->print();
   561         tty->print(" %d", entry->literal()->refcount());
   562         p = entry->next_addr();
   563         entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
   564       }
   565       tty->cr();
   566     }
   567   }
   568 }
   569 #endif // PRODUCT
   571 // --------------------------------------------------------------------------
   573 #ifdef ASSERT
   574 class StableMemoryChecker : public StackObj {
   575   enum { _bufsize = wordSize*4 };
   577   address _region;
   578   jint    _size;
   579   u1      _save_buf[_bufsize];
   581   int sample(u1* save_buf) {
   582     if (_size <= _bufsize) {
   583       memcpy(save_buf, _region, _size);
   584       return _size;
   585     } else {
   586       // copy head and tail
   587       memcpy(&save_buf[0],          _region,                      _bufsize/2);
   588       memcpy(&save_buf[_bufsize/2], _region + _size - _bufsize/2, _bufsize/2);
   589       return (_bufsize/2)*2;
   590     }
   591   }
   593  public:
   594   StableMemoryChecker(const void* region, jint size) {
   595     _region = (address) region;
   596     _size   = size;
   597     sample(_save_buf);
   598   }
   600   bool verify() {
   601     u1 check_buf[sizeof(_save_buf)];
   602     int check_size = sample(check_buf);
   603     return (0 == memcmp(_save_buf, check_buf, check_size));
   604   }
   606   void set_region(const void* region) { _region = (address) region; }
   607 };
   608 #endif
   611 // --------------------------------------------------------------------------
   612 StringTable* StringTable::_the_table = NULL;
   614 bool StringTable::_needs_rehashing = false;
   616 // Pick hashing algorithm
   617 unsigned int StringTable::hash_string(const jchar* s, int len) {
   618   return use_alternate_hashcode() ? AltHashing::murmur3_32(seed(), s, len) :
   619                                     java_lang_String::hash_code(s, len);
   620 }
   622 oop StringTable::lookup(int index, jchar* name,
   623                         int len, unsigned int hash) {
   624   int count = 0;
   625   for (HashtableEntry<oop, mtSymbol>* l = bucket(index); l != NULL; l = l->next()) {
   626     count++;
   627     if (l->hash() == hash) {
   628       if (java_lang_String::equals(l->literal(), name, len)) {
   629         return l->literal();
   630       }
   631     }
   632   }
   633   // If the bucket size is too deep check if this hash code is insufficient.
   634   if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
   635     _needs_rehashing = check_rehash_table(count);
   636   }
   637   return NULL;
   638 }
   641 oop StringTable::basic_add(int index_arg, Handle string, jchar* name,
   642                            int len, unsigned int hashValue_arg, TRAPS) {
   644   assert(java_lang_String::equals(string(), name, len),
   645          "string must be properly initialized");
   646   // Cannot hit a safepoint in this function because the "this" pointer can move.
   647   No_Safepoint_Verifier nsv;
   649   // Check if the symbol table has been rehashed, if so, need to recalculate
   650   // the hash value and index before second lookup.
   651   unsigned int hashValue;
   652   int index;
   653   if (use_alternate_hashcode()) {
   654     hashValue = hash_string(name, len);
   655     index = hash_to_index(hashValue);
   656   } else {
   657     hashValue = hashValue_arg;
   658     index = index_arg;
   659   }
   661   // Since look-up was done lock-free, we need to check if another
   662   // thread beat us in the race to insert the symbol.
   664   oop test = lookup(index, name, len, hashValue); // calls lookup(u1*, int)
   665   if (test != NULL) {
   666     // Entry already added
   667     return test;
   668   }
   670   HashtableEntry<oop, mtSymbol>* entry = new_entry(hashValue, string());
   671   add_entry(index, entry);
   672   return string();
   673 }
   676 oop StringTable::lookup(Symbol* symbol) {
   677   ResourceMark rm;
   678   int length;
   679   jchar* chars = symbol->as_unicode(length);
   680   return lookup(chars, length);
   681 }
   684 oop StringTable::lookup(jchar* name, int len) {
   685   unsigned int hash = hash_string(name, len);
   686   int index = the_table()->hash_to_index(hash);
   687   return the_table()->lookup(index, name, len, hash);
   688 }
   691 oop StringTable::intern(Handle string_or_null, jchar* name,
   692                         int len, TRAPS) {
   693   unsigned int hashValue = hash_string(name, len);
   694   int index = the_table()->hash_to_index(hashValue);
   695   oop found_string = the_table()->lookup(index, name, len, hashValue);
   697   // Found
   698   if (found_string != NULL) return found_string;
   700   debug_only(StableMemoryChecker smc(name, len * sizeof(name[0])));
   701   assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),
   702          "proposed name of symbol must be stable");
   704   Handle string;
   705   // try to reuse the string if possible
   706   if (!string_or_null.is_null()) {
   707     string = string_or_null;
   708   } else {
   709     string = java_lang_String::create_from_unicode(name, len, CHECK_NULL);
   710   }
   712   // Grab the StringTable_lock before getting the_table() because it could
   713   // change at safepoint.
   714   MutexLocker ml(StringTable_lock, THREAD);
   716   // Otherwise, add to symbol to table
   717   return the_table()->basic_add(index, string, name, len,
   718                                 hashValue, CHECK_NULL);
   719 }
   721 oop StringTable::intern(Symbol* symbol, TRAPS) {
   722   if (symbol == NULL) return NULL;
   723   ResourceMark rm(THREAD);
   724   int length;
   725   jchar* chars = symbol->as_unicode(length);
   726   Handle string;
   727   oop result = intern(string, chars, length, CHECK_NULL);
   728   return result;
   729 }
   732 oop StringTable::intern(oop string, TRAPS)
   733 {
   734   if (string == NULL) return NULL;
   735   ResourceMark rm(THREAD);
   736   int length;
   737   Handle h_string (THREAD, string);
   738   jchar* chars = java_lang_String::as_unicode_string(string, length, CHECK_NULL);
   739   oop result = intern(h_string, chars, length, CHECK_NULL);
   740   return result;
   741 }
   744 oop StringTable::intern(const char* utf8_string, TRAPS) {
   745   if (utf8_string == NULL) return NULL;
   746   ResourceMark rm(THREAD);
   747   int length = UTF8::unicode_length(utf8_string);
   748   jchar* chars = NEW_RESOURCE_ARRAY(jchar, length);
   749   UTF8::convert_to_unicode(utf8_string, chars, length);
   750   Handle string;
   751   oop result = intern(string, chars, length, CHECK_NULL);
   752   return result;
   753 }
   755 void StringTable::unlink(BoolObjectClosure* is_alive) {
   756   // Readers of the table are unlocked, so we should only be removing
   757   // entries at a safepoint.
   758   assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
   759   for (int i = 0; i < the_table()->table_size(); ++i) {
   760     HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
   761     HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
   762     while (entry != NULL) {
   763       // Shared entries are normally at the end of the bucket and if we run into
   764       // a shared entry, then there is nothing more to remove. However, if we
   765       // have rehashed the table, then the shared entries are no longer at the
   766       // end of the bucket.
   767       if (entry->is_shared() && !use_alternate_hashcode()) {
   768         break;
   769       }
   770       assert(entry->literal() != NULL, "just checking");
   771       if (entry->is_shared() || is_alive->do_object_b(entry->literal())) {
   772         p = entry->next_addr();
   773       } else {
   774         *p = entry->next();
   775         the_table()->free_entry(entry);
   776       }
   777       entry = (HashtableEntry<oop, mtSymbol>*)HashtableEntry<oop, mtSymbol>::make_ptr(*p);
   778     }
   779   }
   780 }
   782 void StringTable::oops_do(OopClosure* f) {
   783   for (int i = 0; i < the_table()->table_size(); ++i) {
   784     HashtableEntry<oop, mtSymbol>** p = the_table()->bucket_addr(i);
   785     HashtableEntry<oop, mtSymbol>* entry = the_table()->bucket(i);
   786     while (entry != NULL) {
   787       f->do_oop((oop*)entry->literal_addr());
   789       // Did the closure remove the literal from the table?
   790       if (entry->literal() == NULL) {
   791         assert(!entry->is_shared(), "immutable hashtable entry?");
   792         *p = entry->next();
   793         the_table()->free_entry(entry);
   794       } else {
   795         p = entry->next_addr();
   796       }
   797       entry = (HashtableEntry<oop, mtSymbol>*)HashtableEntry<oop, mtSymbol>::make_ptr(*p);
   798     }
   799   }
   800 }
   802 void StringTable::verify() {
   803   for (int i = 0; i < the_table()->table_size(); ++i) {
   804     HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
   805     for ( ; p != NULL; p = p->next()) {
   806       oop s = p->literal();
   807       guarantee(s != NULL, "interned string is NULL");
   808       unsigned int h = java_lang_String::hash_string(s);
   809       guarantee(p->hash() == h, "broken hash in string table entry");
   810       guarantee(the_table()->hash_to_index(h) == i,
   811                 "wrong index in string table");
   812     }
   813   }
   814 }
   816 void StringTable::dump(outputStream* st) {
   817   NumberSeq summary;
   818   for (int i = 0; i < the_table()->table_size(); ++i) {
   819     HashtableEntry<oop, mtSymbol>* p = the_table()->bucket(i);
   820     int count = 0;
   821     for ( ; p != NULL; p = p->next()) {
   822       count++;
   823     }
   824     summary.add((double)count);
   825   }
   826   st->print_cr("StringTable statistics:");
   827   st->print_cr("Number of buckets       : %7d", summary.num());
   828   st->print_cr("Average bucket size     : %7.0f", summary.avg());
   829   st->print_cr("Variance of bucket size : %7.0f", summary.variance());
   830   st->print_cr("Std. dev. of bucket size: %7.0f", summary.sd());
   831   st->print_cr("Maximum bucket size     : %7.0f", summary.maximum());
   832 }
   835 // Create a new table and using alternate hash code, populate the new table
   836 // with the existing strings.   Set flag to use the alternate hash code afterwards.
   837 void StringTable::rehash_table() {
   838   assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
   839   // This should never happen with -Xshare:dump but it might in testing mode.
   840   if (DumpSharedSpaces) return;
   841   StringTable* new_table = new StringTable();
   843   // Rehash the table
   844   the_table()->move_to(new_table);
   846   // Delete the table and buckets (entries are reused in new table).
   847   delete _the_table;
   848   // Don't check if we need rehashing until the table gets unbalanced again.
   849   // Then rehash with a new global seed.
   850   _needs_rehashing = false;
   851   _the_table = new_table;
   852 }

mercurial