1 /*
2  * Copyright (c) 1997, 2020, 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  */
24 
25 #include "precompiled.hpp"
26 #include "classfile/altHashing.hpp"
27 #include "classfile/compactHashtable.hpp"
28 #include "classfile/javaClasses.hpp"
29 #include "classfile/symbolTable.hpp"
30 #include "memory/allocation.inline.hpp"
31 #include "memory/archiveBuilder.hpp"
32 #include "memory/dynamicArchive.hpp"
33 #include "memory/metaspaceClosure.hpp"
34 #include "memory/metaspaceShared.hpp"
35 #include "memory/resourceArea.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "runtime/atomic.hpp"
38 #include "runtime/interfaceSupport.inline.hpp"
39 #include "runtime/timerTrace.hpp"
40 #include "services/diagnosticCommand.hpp"
41 #include "utilities/concurrentHashTable.inline.hpp"
42 #include "utilities/concurrentHashTableTasks.inline.hpp"
43 #include "utilities/utf8.hpp"
44 
45 // We used to not resize at all, so let's be conservative
46 // and not set it too short before we decide to resize,
47 // to match previous startup behavior
48 const double PREF_AVG_LIST_LEN = 8.0;
49 // 2^24 is max size, like StringTable.
50 const size_t END_SIZE = 24;
51 // If a chain gets to 100 something might be wrong
52 const size_t REHASH_LEN = 100;
53 
54 const size_t ON_STACK_BUFFER_LENGTH = 128;
55 
56 // --------------------------------------------------------------------------
57 
symbol_equals_compact_hashtable_entry(Symbol * value,const char * key,int len)58 inline bool symbol_equals_compact_hashtable_entry(Symbol* value, const char* key, int len) {
59   if (value->equals(key, len)) {
60     return true;
61   } else {
62     return false;
63   }
64 }
65 
66 static OffsetCompactHashtable<
67   const char*, Symbol*,
68   symbol_equals_compact_hashtable_entry
69 > _shared_table;
70 
71 static OffsetCompactHashtable<
72   const char*, Symbol*,
73   symbol_equals_compact_hashtable_entry
74 > _dynamic_shared_table;
75 
76 // --------------------------------------------------------------------------
77 
78 typedef ConcurrentHashTable<SymbolTableConfig, mtSymbol> SymbolTableHash;
79 static SymbolTableHash* _local_table = NULL;
80 
81 volatile bool SymbolTable::_has_work = 0;
82 volatile bool SymbolTable::_needs_rehashing = false;
83 
84 // For statistics
85 static size_t _symbols_removed = 0;
86 static size_t _symbols_counted = 0;
87 static size_t _current_size = 0;
88 
89 static volatile size_t _items_count = 0;
90 static volatile bool   _has_items_to_clean = false;
91 
92 
93 static volatile bool _alt_hash = false;
94 static volatile bool _lookup_shared_first = false;
95 
96 // Static arena for symbols that are not deallocated
97 Arena* SymbolTable::_arena = NULL;
98 
99 static uint64_t _alt_hash_seed = 0;
100 
log_trace_symboltable_helper(Symbol * sym,const char * msg)101 static inline void log_trace_symboltable_helper(Symbol* sym, const char* msg) {
102 #ifndef PRODUCT
103   ResourceMark rm;
104   log_trace(symboltable)("%s [%s]", msg, sym->as_quoted_ascii());
105 #endif // PRODUCT
106 }
107 
108 // Pick hashing algorithm.
hash_symbol(const char * s,int len,bool useAlt)109 static uintx hash_symbol(const char* s, int len, bool useAlt) {
110   return useAlt ?
111   AltHashing::halfsiphash_32(_alt_hash_seed, (const uint8_t*)s, len) :
112   java_lang_String::hash_code((const jbyte*)s, len);
113 }
114 
115 #if INCLUDE_CDS
hash_shared_symbol(const char * s,int len)116 static uintx hash_shared_symbol(const char* s, int len) {
117   return java_lang_String::hash_code((const jbyte*)s, len);
118 }
119 #endif
120 
121 class SymbolTableConfig : public AllStatic {
122 private:
123 public:
124   typedef Symbol* Value;  // value of the Node in the hashtable
125 
get_hash(Value const & value,bool * is_dead)126   static uintx get_hash(Value const& value, bool* is_dead) {
127     *is_dead = (value->refcount() == 0);
128     if (*is_dead) {
129       return 0;
130     } else {
131       return hash_symbol((const char*)value->bytes(), value->utf8_length(), _alt_hash);
132     }
133   }
134   // We use default allocation/deallocation but counted
allocate_node(size_t size,Value const & value)135   static void* allocate_node(size_t size, Value const& value) {
136     SymbolTable::item_added();
137     return AllocateHeap(size, mtSymbol);
138   }
free_node(void * memory,Value const & value)139   static void free_node(void* memory, Value const& value) {
140     // We get here because #1 some threads lost a race to insert a newly created Symbol
141     // or #2 we're cleaning up unused symbol.
142     // If #1, then the symbol can be either permanent,
143     // or regular newly created one (refcount==1)
144     // If #2, then the symbol is dead (refcount==0)
145     assert(value->is_permanent() || (value->refcount() == 1) || (value->refcount() == 0),
146            "refcount %d", value->refcount());
147     if (value->refcount() == 1) {
148       value->decrement_refcount();
149       assert(value->refcount() == 0, "expected dead symbol");
150     }
151     SymbolTable::delete_symbol(value);
152     FreeHeap(memory);
153     SymbolTable::item_removed();
154   }
155 };
156 
ceil_log2(size_t value)157 static size_t ceil_log2(size_t value) {
158   size_t ret;
159   for (ret = 1; ((size_t)1 << ret) < value; ++ret);
160   return ret;
161 }
162 
create_table()163 void SymbolTable::create_table ()  {
164   size_t start_size_log_2 = ceil_log2(SymbolTableSize);
165   _current_size = ((size_t)1) << start_size_log_2;
166   log_trace(symboltable)("Start size: " SIZE_FORMAT " (" SIZE_FORMAT ")",
167                          _current_size, start_size_log_2);
168   _local_table = new SymbolTableHash(start_size_log_2, END_SIZE, REHASH_LEN);
169 
170   // Initialize the arena for global symbols, size passed in depends on CDS.
171   if (symbol_alloc_arena_size == 0) {
172     _arena = new (mtSymbol) Arena(mtSymbol);
173   } else {
174     _arena = new (mtSymbol) Arena(mtSymbol, symbol_alloc_arena_size);
175   }
176 }
177 
delete_symbol(Symbol * sym)178 void SymbolTable::delete_symbol(Symbol* sym) {
179   if (sym->is_permanent()) {
180     MutexLocker ml(SymbolArena_lock, Mutex::_no_safepoint_check_flag); // Protect arena
181     // Deleting permanent symbol should not occur very often (insert race condition),
182     // so log it.
183     log_trace_symboltable_helper(sym, "Freeing permanent symbol");
184     if (!arena()->Afree(sym, sym->size())) {
185       log_trace_symboltable_helper(sym, "Leaked permanent symbol");
186     }
187   } else {
188     delete sym;
189   }
190 }
191 
reset_has_items_to_clean()192 void SymbolTable::reset_has_items_to_clean() { Atomic::store(&_has_items_to_clean, false); }
mark_has_items_to_clean()193 void SymbolTable::mark_has_items_to_clean()  { Atomic::store(&_has_items_to_clean, true); }
has_items_to_clean()194 bool SymbolTable::has_items_to_clean()       { return Atomic::load(&_has_items_to_clean); }
195 
item_added()196 void SymbolTable::item_added() {
197   Atomic::inc(&_items_count);
198 }
199 
item_removed()200 void SymbolTable::item_removed() {
201   Atomic::inc(&(_symbols_removed));
202   Atomic::dec(&_items_count);
203 }
204 
get_load_factor()205 double SymbolTable::get_load_factor() {
206   return (double)_items_count/_current_size;
207 }
208 
table_size()209 size_t SymbolTable::table_size() {
210   return ((size_t)1) << _local_table->get_size_log2(Thread::current());
211 }
212 
trigger_cleanup()213 void SymbolTable::trigger_cleanup() {
214   MutexLocker ml(Service_lock, Mutex::_no_safepoint_check_flag);
215   _has_work = true;
216   Service_lock->notify_all();
217 }
218 
allocate_symbol(const char * name,int len,bool c_heap)219 Symbol* SymbolTable::allocate_symbol(const char* name, int len, bool c_heap) {
220   assert (len <= Symbol::max_length(), "should be checked by caller");
221 
222   Symbol* sym;
223   if (DumpSharedSpaces) {
224     // TODO: Special handling of Symbol allocation for DumpSharedSpaces will be removed
225     // in JDK-8250989
226     c_heap = false;
227   }
228   if (c_heap) {
229     // refcount starts as 1
230     sym = new (len) Symbol((const u1*)name, len, 1);
231     assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted");
232   } else if (DumpSharedSpaces) {
233     // See comments inside Symbol::operator new(size_t, int)
234     sym = new (len) Symbol((const u1*)name, len, PERM_REFCOUNT);
235     assert(sym != NULL, "new should call vm_exit_out_of_memory if failed to allocate symbol during DumpSharedSpaces");
236   } else {
237     // Allocate to global arena
238     MutexLocker ml(SymbolArena_lock, Mutex::_no_safepoint_check_flag); // Protect arena
239     sym = new (len, arena()) Symbol((const u1*)name, len, PERM_REFCOUNT);
240   }
241   return sym;
242 }
243 
244 class SymbolsDo : StackObj {
245   SymbolClosure *_cl;
246 public:
SymbolsDo(SymbolClosure * cl)247   SymbolsDo(SymbolClosure *cl) : _cl(cl) {}
operator ()(Symbol ** value)248   bool operator()(Symbol** value) {
249     assert(value != NULL, "expected valid value");
250     assert(*value != NULL, "value should point to a symbol");
251     _cl->do_symbol(value);
252     return true;
253   };
254 };
255 
256 class SharedSymbolIterator {
257   SymbolClosure* _symbol_closure;
258 public:
SharedSymbolIterator(SymbolClosure * f)259   SharedSymbolIterator(SymbolClosure* f) : _symbol_closure(f) {}
do_value(Symbol * symbol)260   void do_value(Symbol* symbol) {
261     _symbol_closure->do_symbol(&symbol);
262   }
263 };
264 
265 // Call function for all symbols in the symbol table.
symbols_do(SymbolClosure * cl)266 void SymbolTable::symbols_do(SymbolClosure *cl) {
267   assert(SafepointSynchronize::is_at_safepoint(), "Must be at safepoint");
268   // all symbols from shared table
269   SharedSymbolIterator iter(cl);
270   _shared_table.iterate(&iter);
271   _dynamic_shared_table.iterate(&iter);
272 
273   // all symbols from the dynamic table
274   SymbolsDo sd(cl);
275   _local_table->do_safepoint_scan(sd);
276 }
277 
lookup_dynamic(const char * name,int len,unsigned int hash)278 Symbol* SymbolTable::lookup_dynamic(const char* name,
279                                     int len, unsigned int hash) {
280   Symbol* sym = do_lookup(name, len, hash);
281   assert((sym == NULL) || sym->refcount() != 0, "refcount must not be zero");
282   return sym;
283 }
284 
285 #if INCLUDE_CDS
lookup_shared(const char * name,int len,unsigned int hash)286 Symbol* SymbolTable::lookup_shared(const char* name,
287                                    int len, unsigned int hash) {
288   Symbol* sym = NULL;
289   if (!_shared_table.empty()) {
290     if (_alt_hash) {
291       // hash_code parameter may use alternate hashing algorithm but the shared table
292       // always uses the same original hash code.
293       hash = hash_shared_symbol(name, len);
294     }
295     sym = _shared_table.lookup(name, hash, len);
296     if (sym == NULL && DynamicArchive::is_mapped()) {
297       sym = _dynamic_shared_table.lookup(name, hash, len);
298     }
299   }
300   return sym;
301 }
302 #endif
303 
lookup_common(const char * name,int len,unsigned int hash)304 Symbol* SymbolTable::lookup_common(const char* name,
305                             int len, unsigned int hash) {
306   Symbol* sym;
307   if (_lookup_shared_first) {
308     sym = lookup_shared(name, len, hash);
309     if (sym == NULL) {
310       _lookup_shared_first = false;
311       sym = lookup_dynamic(name, len, hash);
312     }
313   } else {
314     sym = lookup_dynamic(name, len, hash);
315     if (sym == NULL) {
316       sym = lookup_shared(name, len, hash);
317       if (sym != NULL) {
318         _lookup_shared_first = true;
319       }
320     }
321   }
322   return sym;
323 }
324 
new_symbol(const char * name,int len)325 Symbol* SymbolTable::new_symbol(const char* name, int len) {
326   unsigned int hash = hash_symbol(name, len, _alt_hash);
327   Symbol* sym = lookup_common(name, len, hash);
328   if (sym == NULL) {
329     sym = do_add_if_needed(name, len, hash, true);
330   }
331   assert(sym->refcount() != 0, "lookup should have incremented the count");
332   assert(sym->equals(name, len), "symbol must be properly initialized");
333   return sym;
334 }
335 
new_symbol(const Symbol * sym,int begin,int end)336 Symbol* SymbolTable::new_symbol(const Symbol* sym, int begin, int end) {
337   assert(begin <= end && end <= sym->utf8_length(), "just checking");
338   assert(sym->refcount() != 0, "require a valid symbol");
339   const char* name = (const char*)sym->base() + begin;
340   int len = end - begin;
341   unsigned int hash = hash_symbol(name, len, _alt_hash);
342   Symbol* found = lookup_common(name, len, hash);
343   if (found == NULL) {
344     found = do_add_if_needed(name, len, hash, true);
345   }
346   return found;
347 }
348 
349 class SymbolTableLookup : StackObj {
350 private:
351   Thread* _thread;
352   uintx _hash;
353   int _len;
354   const char* _str;
355 public:
SymbolTableLookup(const char * key,int len,uintx hash)356   SymbolTableLookup(const char* key, int len, uintx hash)
357   : _hash(hash), _len(len), _str(key) {}
get_hash() const358   uintx get_hash() const {
359     return _hash;
360   }
equals(Symbol ** value,bool * is_dead)361   bool equals(Symbol** value, bool* is_dead) {
362     assert(value != NULL, "expected valid value");
363     assert(*value != NULL, "value should point to a symbol");
364     Symbol *sym = *value;
365     if (sym->equals(_str, _len)) {
366       if (sym->try_increment_refcount()) {
367         // something is referencing this symbol now.
368         return true;
369       } else {
370         assert(sym->refcount() == 0, "expected dead symbol");
371         *is_dead = true;
372         return false;
373       }
374     } else {
375       *is_dead = (sym->refcount() == 0);
376       return false;
377     }
378   }
379 };
380 
381 class SymbolTableGet : public StackObj {
382   Symbol* _return;
383 public:
SymbolTableGet()384   SymbolTableGet() : _return(NULL) {}
operator ()(Symbol ** value)385   void operator()(Symbol** value) {
386     assert(value != NULL, "expected valid value");
387     assert(*value != NULL, "value should point to a symbol");
388     _return = *value;
389   }
get_res_sym() const390   Symbol* get_res_sym() const {
391     return _return;
392   }
393 };
394 
do_lookup(const char * name,int len,uintx hash)395 Symbol* SymbolTable::do_lookup(const char* name, int len, uintx hash) {
396   Thread* thread = Thread::current();
397   SymbolTableLookup lookup(name, len, hash);
398   SymbolTableGet stg;
399   bool rehash_warning = false;
400   _local_table->get(thread, lookup, stg, &rehash_warning);
401   update_needs_rehash(rehash_warning);
402   Symbol* sym = stg.get_res_sym();
403   assert((sym == NULL) || sym->refcount() != 0, "found dead symbol");
404   return sym;
405 }
406 
lookup_only(const char * name,int len,unsigned int & hash)407 Symbol* SymbolTable::lookup_only(const char* name, int len, unsigned int& hash) {
408   hash = hash_symbol(name, len, _alt_hash);
409   return lookup_common(name, len, hash);
410 }
411 
412 // Suggestion: Push unicode-based lookup all the way into the hashing
413 // and probing logic, so there is no need for convert_to_utf8 until
414 // an actual new Symbol* is created.
new_symbol(const jchar * name,int utf16_length)415 Symbol* SymbolTable::new_symbol(const jchar* name, int utf16_length) {
416   int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
417   char stack_buf[ON_STACK_BUFFER_LENGTH];
418   if (utf8_length < (int) sizeof(stack_buf)) {
419     char* chars = stack_buf;
420     UNICODE::convert_to_utf8(name, utf16_length, chars);
421     return new_symbol(chars, utf8_length);
422   } else {
423     ResourceMark rm;
424     char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);
425     UNICODE::convert_to_utf8(name, utf16_length, chars);
426     return new_symbol(chars, utf8_length);
427   }
428 }
429 
lookup_only_unicode(const jchar * name,int utf16_length,unsigned int & hash)430 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
431                                          unsigned int& hash) {
432   int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
433   char stack_buf[ON_STACK_BUFFER_LENGTH];
434   if (utf8_length < (int) sizeof(stack_buf)) {
435     char* chars = stack_buf;
436     UNICODE::convert_to_utf8(name, utf16_length, chars);
437     return lookup_only(chars, utf8_length, hash);
438   } else {
439     ResourceMark rm;
440     char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);
441     UNICODE::convert_to_utf8(name, utf16_length, chars);
442     return lookup_only(chars, utf8_length, hash);
443   }
444 }
445 
new_symbols(ClassLoaderData * loader_data,const constantPoolHandle & cp,int names_count,const char ** names,int * lengths,int * cp_indices,unsigned int * hashValues)446 void SymbolTable::new_symbols(ClassLoaderData* loader_data, const constantPoolHandle& cp,
447                               int names_count, const char** names, int* lengths,
448                               int* cp_indices, unsigned int* hashValues) {
449   // Note that c_heap will be true for non-strong hidden classes and unsafe anonymous classes
450   // even if their loader is the boot loader because they will have a different cld.
451   bool c_heap = !loader_data->is_the_null_class_loader_data();
452   for (int i = 0; i < names_count; i++) {
453     const char *name = names[i];
454     int len = lengths[i];
455     unsigned int hash = hashValues[i];
456     assert(lookup_shared(name, len, hash) == NULL, "must have checked already");
457     Symbol* sym = do_add_if_needed(name, len, hash, c_heap);
458     assert(sym->refcount() != 0, "lookup should have incremented the count");
459     cp->symbol_at_put(cp_indices[i], sym);
460   }
461 }
462 
do_add_if_needed(const char * name,int len,uintx hash,bool heap)463 Symbol* SymbolTable::do_add_if_needed(const char* name, int len, uintx hash, bool heap) {
464   SymbolTableLookup lookup(name, len, hash);
465   SymbolTableGet stg;
466   bool clean_hint = false;
467   bool rehash_warning = false;
468   Symbol* sym = NULL;
469   Thread* THREAD = Thread::current();
470 
471   do {
472     // Callers have looked up the symbol once, insert the symbol.
473     sym = allocate_symbol(name, len, heap);
474     if (_local_table->insert(THREAD, lookup, sym, &rehash_warning, &clean_hint)) {
475       break;
476     }
477     // In case another thread did a concurrent add, return value already in the table.
478     // This could fail if the symbol got deleted concurrently, so loop back until success.
479     if (_local_table->get(THREAD, lookup, stg, &rehash_warning)) {
480       sym = stg.get_res_sym();
481       break;
482     }
483   } while(true);
484 
485   update_needs_rehash(rehash_warning);
486 
487   if (clean_hint) {
488     mark_has_items_to_clean();
489     check_concurrent_work();
490   }
491 
492   assert((sym == NULL) || sym->refcount() != 0, "found dead symbol");
493   return sym;
494 }
495 
new_permanent_symbol(const char * name)496 Symbol* SymbolTable::new_permanent_symbol(const char* name) {
497   unsigned int hash = 0;
498   int len = (int)strlen(name);
499   Symbol* sym = SymbolTable::lookup_only(name, len, hash);
500   if (sym == NULL) {
501     sym = do_add_if_needed(name, len, hash, false);
502   }
503   if (!sym->is_permanent()) {
504     sym->make_permanent();
505     log_trace_symboltable_helper(sym, "Asked for a permanent symbol, but got a regular one");
506   }
507   return sym;
508 }
509 
510 struct SizeFunc : StackObj {
operator ()SizeFunc511   size_t operator()(Symbol** value) {
512     assert(value != NULL, "expected valid value");
513     assert(*value != NULL, "value should point to a symbol");
514     return (*value)->size() * HeapWordSize;
515   };
516 };
517 
get_table_statistics()518 TableStatistics SymbolTable::get_table_statistics() {
519   static TableStatistics ts;
520   SizeFunc sz;
521   ts = _local_table->statistics_get(Thread::current(), sz, ts);
522   return ts;
523 }
524 
print_table_statistics(outputStream * st,const char * table_name)525 void SymbolTable::print_table_statistics(outputStream* st,
526                                          const char* table_name) {
527   SizeFunc sz;
528   _local_table->statistics_to(Thread::current(), sz, st, table_name);
529 }
530 
531 // Verification
532 class VerifySymbols : StackObj {
533 public:
operator ()(Symbol ** value)534   bool operator()(Symbol** value) {
535     guarantee(value != NULL, "expected valid value");
536     guarantee(*value != NULL, "value should point to a symbol");
537     Symbol* sym = *value;
538     guarantee(sym->equals((const char*)sym->bytes(), sym->utf8_length()),
539               "symbol must be internally consistent");
540     return true;
541   };
542 };
543 
verify()544 void SymbolTable::verify() {
545   Thread* thr = Thread::current();
546   VerifySymbols vs;
547   if (!_local_table->try_scan(thr, vs)) {
548     log_info(symboltable)("verify unavailable at this moment");
549   }
550 }
551 
552 // Dumping
553 class DumpSymbol : StackObj {
554   Thread* _thr;
555   outputStream* _st;
556 public:
DumpSymbol(Thread * thr,outputStream * st)557   DumpSymbol(Thread* thr, outputStream* st) : _thr(thr), _st(st) {}
operator ()(Symbol ** value)558   bool operator()(Symbol** value) {
559     assert(value != NULL, "expected valid value");
560     assert(*value != NULL, "value should point to a symbol");
561     Symbol* sym = *value;
562     const char* utf8_string = (const char*)sym->bytes();
563     int utf8_length = sym->utf8_length();
564     _st->print("%d %d: ", utf8_length, sym->refcount());
565     HashtableTextDump::put_utf8(_st, utf8_string, utf8_length);
566     _st->cr();
567     return true;
568   };
569 };
570 
dump(outputStream * st,bool verbose)571 void SymbolTable::dump(outputStream* st, bool verbose) {
572   if (!verbose) {
573     print_table_statistics(st, "SymbolTable");
574   } else {
575     Thread* thr = Thread::current();
576     ResourceMark rm(thr);
577     st->print_cr("VERSION: 1.1");
578     DumpSymbol ds(thr, st);
579     if (!_local_table->try_scan(thr, ds)) {
580       log_info(symboltable)("dump unavailable at this moment");
581     }
582   }
583 }
584 
585 #if INCLUDE_CDS
copy_shared_symbol_table(GrowableArray<Symbol * > * symbols,CompactHashtableWriter * writer)586 void SymbolTable::copy_shared_symbol_table(GrowableArray<Symbol*>* symbols,
587                                            CompactHashtableWriter* writer) {
588   int len = symbols->length();
589   for (int i = 0; i < len; i++) {
590     Symbol* sym = ArchiveBuilder::get_relocated_symbol(symbols->at(i));
591     unsigned int fixed_hash = hash_shared_symbol((const char*)sym->bytes(), sym->utf8_length());
592     assert(fixed_hash == hash_symbol((const char*)sym->bytes(), sym->utf8_length(), false),
593            "must not rehash during dumping");
594     sym->set_permanent();
595     if (DynamicDumpSharedSpaces) {
596       sym = DynamicArchive::buffer_to_target(sym);
597     }
598     writer->add(fixed_hash, MetaspaceShared::object_delta_u4(sym));
599   }
600 }
601 
estimate_size_for_archive()602 size_t SymbolTable::estimate_size_for_archive() {
603   return CompactHashtableWriter::estimate_size(int(_items_count));
604 }
605 
write_to_archive(GrowableArray<Symbol * > * symbols)606 void SymbolTable::write_to_archive(GrowableArray<Symbol*>* symbols) {
607   CompactHashtableWriter writer(int(_items_count),
608                                 &MetaspaceShared::stats()->symbol);
609   copy_shared_symbol_table(symbols, &writer);
610   if (!DynamicDumpSharedSpaces) {
611     _shared_table.reset();
612     writer.dump(&_shared_table, "symbol");
613 
614     // Verify the written shared table is correct -- at this point,
615     // vmSymbols has already been relocated to point to the archived
616     // version of the Symbols.
617     Symbol* sym = vmSymbols::java_lang_Object();
618     const char* name = (const char*)sym->bytes();
619     int len = sym->utf8_length();
620     unsigned int hash = hash_symbol(name, len, _alt_hash);
621     assert(sym == _shared_table.lookup(name, hash, len), "sanity");
622   } else {
623     _dynamic_shared_table.reset();
624     writer.dump(&_dynamic_shared_table, "symbol");
625   }
626 }
627 
serialize_shared_table_header(SerializeClosure * soc,bool is_static_archive)628 void SymbolTable::serialize_shared_table_header(SerializeClosure* soc,
629                                                 bool is_static_archive) {
630   OffsetCompactHashtable<const char*, Symbol*, symbol_equals_compact_hashtable_entry> * table;
631   if (is_static_archive) {
632     table = &_shared_table;
633   } else {
634     table = &_dynamic_shared_table;
635   }
636   table->serialize_header(soc);
637   if (soc->writing()) {
638     // Sanity. Make sure we don't use the shared table at dump time
639     table->reset();
640   }
641 }
642 #endif //INCLUDE_CDS
643 
644 // Concurrent work
grow(JavaThread * jt)645 void SymbolTable::grow(JavaThread* jt) {
646   SymbolTableHash::GrowTask gt(_local_table);
647   if (!gt.prepare(jt)) {
648     return;
649   }
650   log_trace(symboltable)("Started to grow");
651   {
652     TraceTime timer("Grow", TRACETIME_LOG(Debug, symboltable, perf));
653     while (gt.do_task(jt)) {
654       gt.pause(jt);
655       {
656         ThreadBlockInVM tbivm(jt);
657       }
658       gt.cont(jt);
659     }
660   }
661   gt.done(jt);
662   _current_size = table_size();
663   log_debug(symboltable)("Grown to size:" SIZE_FORMAT, _current_size);
664 }
665 
666 struct SymbolTableDoDelete : StackObj {
667   size_t _deleted;
SymbolTableDoDeleteSymbolTableDoDelete668   SymbolTableDoDelete() : _deleted(0) {}
operator ()SymbolTableDoDelete669   void operator()(Symbol** value) {
670     assert(value != NULL, "expected valid value");
671     assert(*value != NULL, "value should point to a symbol");
672     Symbol *sym = *value;
673     assert(sym->refcount() == 0, "refcount");
674     _deleted++;
675   }
676 };
677 
678 struct SymbolTableDeleteCheck : StackObj {
679   size_t _processed;
SymbolTableDeleteCheckSymbolTableDeleteCheck680   SymbolTableDeleteCheck() : _processed(0) {}
operator ()SymbolTableDeleteCheck681   bool operator()(Symbol** value) {
682     assert(value != NULL, "expected valid value");
683     assert(*value != NULL, "value should point to a symbol");
684     _processed++;
685     Symbol *sym = *value;
686     return (sym->refcount() == 0);
687   }
688 };
689 
clean_dead_entries(JavaThread * jt)690 void SymbolTable::clean_dead_entries(JavaThread* jt) {
691   SymbolTableHash::BulkDeleteTask bdt(_local_table);
692   if (!bdt.prepare(jt)) {
693     return;
694   }
695 
696   SymbolTableDeleteCheck stdc;
697   SymbolTableDoDelete stdd;
698   {
699     TraceTime timer("Clean", TRACETIME_LOG(Debug, symboltable, perf));
700     while (bdt.do_task(jt, stdc, stdd)) {
701       bdt.pause(jt);
702       {
703         ThreadBlockInVM tbivm(jt);
704       }
705       bdt.cont(jt);
706     }
707     reset_has_items_to_clean();
708     bdt.done(jt);
709   }
710 
711   Atomic::add(&_symbols_counted, stdc._processed);
712 
713   log_debug(symboltable)("Cleaned " SIZE_FORMAT " of " SIZE_FORMAT,
714                          stdd._deleted, stdc._processed);
715 }
716 
check_concurrent_work()717 void SymbolTable::check_concurrent_work() {
718   if (_has_work) {
719     return;
720   }
721   // We should clean/resize if we have
722   // more items than preferred load factor or
723   // more dead items than water mark.
724   if (has_items_to_clean() || (get_load_factor() > PREF_AVG_LIST_LEN)) {
725     log_debug(symboltable)("Concurrent work triggered, load factor: %f, items to clean: %s",
726                            get_load_factor(), has_items_to_clean() ? "true" : "false");
727     trigger_cleanup();
728   }
729 }
730 
do_concurrent_work(JavaThread * jt)731 void SymbolTable::do_concurrent_work(JavaThread* jt) {
732   double load_factor = get_load_factor();
733   log_debug(symboltable, perf)("Concurrent work, live factor: %g", load_factor);
734   // We prefer growing, since that also removes dead items
735   if (load_factor > PREF_AVG_LIST_LEN && !_local_table->is_max_size_reached()) {
736     grow(jt);
737   } else {
738     clean_dead_entries(jt);
739   }
740   _has_work = false;
741 }
742 
743 // Rehash
do_rehash()744 bool SymbolTable::do_rehash() {
745   if (!_local_table->is_safepoint_safe()) {
746     return false;
747   }
748 
749   // We use current size
750   size_t new_size = _local_table->get_size_log2(Thread::current());
751   SymbolTableHash* new_table = new SymbolTableHash(new_size, END_SIZE, REHASH_LEN);
752   // Use alt hash from now on
753   _alt_hash = true;
754   if (!_local_table->try_move_nodes_to(Thread::current(), new_table)) {
755     _alt_hash = false;
756     delete new_table;
757     return false;
758   }
759 
760   // free old table
761   delete _local_table;
762   _local_table = new_table;
763 
764   return true;
765 }
766 
rehash_table()767 void SymbolTable::rehash_table() {
768   static bool rehashed = false;
769   log_debug(symboltable)("Table imbalanced, rehashing called.");
770 
771   // Grow instead of rehash.
772   if (get_load_factor() > PREF_AVG_LIST_LEN &&
773       !_local_table->is_max_size_reached()) {
774     log_debug(symboltable)("Choosing growing over rehashing.");
775     trigger_cleanup();
776     _needs_rehashing = false;
777     return;
778   }
779 
780   // Already rehashed.
781   if (rehashed) {
782     log_warning(symboltable)("Rehashing already done, still long lists.");
783     trigger_cleanup();
784     _needs_rehashing = false;
785     return;
786   }
787 
788   _alt_hash_seed = AltHashing::compute_seed();
789 
790   if (do_rehash()) {
791     rehashed = true;
792   } else {
793     log_info(symboltable)("Resizes in progress rehashing skipped.");
794   }
795 
796   _needs_rehashing = false;
797 }
798 
799 //---------------------------------------------------------------------------
800 // Non-product code
801 
802 #ifndef PRODUCT
803 
804 class HistogramIterator : StackObj {
805 public:
806   static const size_t results_length = 100;
807   size_t counts[results_length];
808   size_t sizes[results_length];
809   size_t total_size;
810   size_t total_count;
811   size_t total_length;
812   size_t max_length;
813   size_t out_of_range_count;
814   size_t out_of_range_size;
HistogramIterator()815   HistogramIterator() : total_size(0), total_count(0), total_length(0),
816                         max_length(0), out_of_range_count(0), out_of_range_size(0) {
817     // initialize results to zero
818     for (size_t i = 0; i < results_length; i++) {
819       counts[i] = 0;
820       sizes[i] = 0;
821     }
822   }
operator ()(Symbol ** value)823   bool operator()(Symbol** value) {
824     assert(value != NULL, "expected valid value");
825     assert(*value != NULL, "value should point to a symbol");
826     Symbol* sym = *value;
827     size_t size = sym->size();
828     size_t len = sym->utf8_length();
829     if (len < results_length) {
830       counts[len]++;
831       sizes[len] += size;
832     } else {
833       out_of_range_count++;
834       out_of_range_size += size;
835     }
836     total_count++;
837     total_size += size;
838     total_length += len;
839     max_length = MAX2(max_length, len);
840 
841     return true;
842   };
843 };
844 
print_histogram()845 void SymbolTable::print_histogram() {
846   HistogramIterator hi;
847   _local_table->do_scan(Thread::current(), hi);
848   tty->print_cr("Symbol Table Histogram:");
849   tty->print_cr("  Total number of symbols  " SIZE_FORMAT_W(7), hi.total_count);
850   tty->print_cr("  Total size in memory     " SIZE_FORMAT_W(7) "K",
851           (hi.total_size * wordSize) / 1024);
852   tty->print_cr("  Total counted            " SIZE_FORMAT_W(7), _symbols_counted);
853   tty->print_cr("  Total removed            " SIZE_FORMAT_W(7), _symbols_removed);
854   if (_symbols_counted > 0) {
855     tty->print_cr("  Percent removed          %3.2f",
856           ((float)_symbols_removed / _symbols_counted) * 100);
857   }
858   tty->print_cr("  Reference counts         " SIZE_FORMAT_W(7), Symbol::_total_count);
859   tty->print_cr("  Symbol arena used        " SIZE_FORMAT_W(7) "K", arena()->used() / 1024);
860   tty->print_cr("  Symbol arena size        " SIZE_FORMAT_W(7) "K", arena()->size_in_bytes() / 1024);
861   tty->print_cr("  Total symbol length      " SIZE_FORMAT_W(7), hi.total_length);
862   tty->print_cr("  Maximum symbol length    " SIZE_FORMAT_W(7), hi.max_length);
863   tty->print_cr("  Average symbol length    %7.2f", ((float)hi.total_length / hi.total_count));
864   tty->print_cr("  Symbol length histogram:");
865   tty->print_cr("    %6s %10s %10s", "Length", "#Symbols", "Size");
866   for (size_t i = 0; i < hi.results_length; i++) {
867     if (hi.counts[i] > 0) {
868       tty->print_cr("    " SIZE_FORMAT_W(6) " " SIZE_FORMAT_W(10) " " SIZE_FORMAT_W(10) "K",
869                     i, hi.counts[i], (hi.sizes[i] * wordSize) / 1024);
870     }
871   }
872   tty->print_cr("  >=" SIZE_FORMAT_W(6) " " SIZE_FORMAT_W(10) " " SIZE_FORMAT_W(10) "K\n",
873                 hi.results_length, hi.out_of_range_count, (hi.out_of_range_size*wordSize) / 1024);
874 }
875 #endif // PRODUCT
876 
877 // Utility for dumping symbols
SymboltableDCmd(outputStream * output,bool heap)878 SymboltableDCmd::SymboltableDCmd(outputStream* output, bool heap) :
879                                  DCmdWithParser(output, heap),
880   _verbose("-verbose", "Dump the content of each symbol in the table",
881            "BOOLEAN", false, "false") {
882   _dcmdparser.add_dcmd_option(&_verbose);
883 }
884 
execute(DCmdSource source,TRAPS)885 void SymboltableDCmd::execute(DCmdSource source, TRAPS) {
886   VM_DumpHashtable dumper(output(), VM_DumpHashtable::DumpSymbols,
887                          _verbose.value());
888   VMThread::execute(&dumper);
889 }
890 
num_arguments()891 int SymboltableDCmd::num_arguments() {
892   ResourceMark rm;
893   SymboltableDCmd* dcmd = new SymboltableDCmd(NULL, false);
894   if (dcmd != NULL) {
895     DCmdMark mark(dcmd);
896     return dcmd->_dcmdparser.num_arguments();
897   } else {
898     return 0;
899   }
900 }
901