src/share/vm/classfile/symbolTable.cpp

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

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

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

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

mercurial