1 /*
2 * Copyright (c) 1997, 2014, 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/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
39 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
40
41 // --------------------------------------------------------------------------
42 // the number of buckets a thread claims
43 const int ClaimChunkSize = 32;
44
45 SymbolTable* SymbolTable::_the_table = NULL;
46 // Static arena for symbols that are not deallocated
47 Arena* SymbolTable::_arena = NULL;
48 bool SymbolTable::_needs_rehashing = false;
49
50 Symbol* SymbolTable::allocate_symbol(const u1* name, int len, bool c_heap, TRAPS) {
51 assert (len <= Symbol::max_length(), "should be checked by caller");
52
53 Symbol* sym;
54
55 if (DumpSharedSpaces) {
56 // Allocate all symbols to CLD shared metaspace
57 sym = new (len, ClassLoaderData::the_null_class_loader_data(), THREAD) Symbol(name, len, -1);
58 } else if (c_heap) {
59 // refcount starts as 1
60 sym = new (len, THREAD) Symbol(name, len, 1);
61 assert(sym != NULL, "new should call vm_exit_out_of_memory if C_HEAP is exhausted");
62 } else {
63 // Allocate to global arena
64 sym = new (len, arena(), THREAD) Symbol(name, len, -1);
65 }
66 return sym;
67 }
68
69 void SymbolTable::initialize_symbols(int arena_alloc_size) {
70 // Initialize the arena for global symbols, size passed in depends on CDS.
71 if (arena_alloc_size == 0) {
72 _arena = new (mtSymbol) Arena();
73 } else {
74 _arena = new (mtSymbol) Arena(arena_alloc_size);
75 }
76 }
77
78 // Call function for all symbols in the symbol table.
79 void SymbolTable::symbols_do(SymbolClosure *cl) {
80 const int n = the_table()->table_size();
81 for (int i = 0; i < n; i++) {
82 for (HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
83 p != NULL;
84 p = p->next()) {
85 cl->do_symbol(p->literal_addr());
86 }
87 }
88 }
89
90 int SymbolTable::_symbols_removed = 0;
91 int SymbolTable::_symbols_counted = 0;
92 volatile int SymbolTable::_parallel_claimed_idx = 0;
93
94 void SymbolTable::buckets_unlink(int start_idx, int end_idx, int* processed, int* removed, size_t* memory_total) {
95 for (int i = start_idx; i < end_idx; ++i) {
96 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
97 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
98 while (entry != NULL) {
99 // Shared entries are normally at the end of the bucket and if we run into
100 // a shared entry, then there is nothing more to remove. However, if we
101 // have rehashed the table, then the shared entries are no longer at the
102 // end of the bucket.
103 if (entry->is_shared() && !use_alternate_hashcode()) {
104 break;
105 }
106 Symbol* s = entry->literal();
107 (*memory_total) += s->size();
108 (*processed)++;
109 assert(s != NULL, "just checking");
110 // If reference count is zero, remove.
111 if (s->refcount() == 0) {
112 assert(!entry->is_shared(), "shared entries should be kept live");
113 delete s;
114 (*removed)++;
115 *p = entry->next();
116 the_table()->free_entry(entry);
117 } else {
118 p = entry->next_addr();
119 }
120 // get next entry
121 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
122 }
123 }
124 }
125
126 // Remove unreferenced symbols from the symbol table
127 // This is done late during GC.
128 void SymbolTable::unlink(int* processed, int* removed) {
129 size_t memory_total = 0;
130 buckets_unlink(0, the_table()->table_size(), processed, removed, &memory_total);
131 _symbols_removed += *removed;
132 _symbols_counted += *processed;
133 // Exclude printing for normal PrintGCDetails because people parse
134 // this output.
135 if (PrintGCDetails && Verbose && WizardMode) {
136 gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", *processed,
137 (memory_total*HeapWordSize)/1024);
138 }
139 }
140
141 void SymbolTable::possibly_parallel_unlink(int* processed, int* removed) {
142 const int limit = the_table()->table_size();
143
144 size_t memory_total = 0;
145
146 for (;;) {
147 // Grab next set of buckets to scan
148 int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize;
149 if (start_idx >= limit) {
150 // End of table
151 break;
152 }
153
154 int end_idx = MIN2(limit, start_idx + ClaimChunkSize);
155 buckets_unlink(start_idx, end_idx, processed, removed, &memory_total);
156 }
157 Atomic::add(*processed, &_symbols_counted);
158 Atomic::add(*removed, &_symbols_removed);
159 // Exclude printing for normal PrintGCDetails because people parse
160 // this output.
161 if (PrintGCDetails && Verbose && WizardMode) {
162 gclog_or_tty->print(" [Symbols: scanned=%d removed=%d size=" SIZE_FORMAT "K] ", *processed, *removed,
163 (memory_total*HeapWordSize)/1024);
164 }
165 }
166
167 // Create a new table and using alternate hash code, populate the new table
168 // with the existing strings. Set flag to use the alternate hash code afterwards.
169 void SymbolTable::rehash_table() {
170 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
171 // This should never happen with -Xshare:dump but it might in testing mode.
172 if (DumpSharedSpaces) return;
173 // Create a new symbol table
174 SymbolTable* new_table = new SymbolTable();
175
176 the_table()->move_to(new_table);
177
178 // Delete the table and buckets (entries are reused in new table).
179 delete _the_table;
180 // Don't check if we need rehashing until the table gets unbalanced again.
181 // Then rehash with a new global seed.
182 _needs_rehashing = false;
183 _the_table = new_table;
184 }
185
186 // Lookup a symbol in a bucket.
187
188 Symbol* SymbolTable::lookup(int index, const char* name,
189 int len, unsigned int hash) {
190 int count = 0;
191 for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) {
192 count++; // count all entries in this bucket, not just ones with same hash
193 if (e->hash() == hash) {
194 Symbol* sym = e->literal();
195 if (sym->equals(name, len)) {
196 // something is referencing this symbol now.
197 sym->increment_refcount();
198 return sym;
199 }
200 }
201 }
202 // If the bucket size is too deep check if this hash code is insufficient.
203 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
204 _needs_rehashing = check_rehash_table(count);
205 }
206 return NULL;
207 }
208
209 // Pick hashing algorithm.
210 unsigned int SymbolTable::hash_symbol(const char* s, int len) {
211 return use_alternate_hashcode() ?
212 AltHashing::murmur3_32(seed(), (const jbyte*)s, len) :
213 java_lang_String::hash_code(s, len);
214 }
215
216
217 // We take care not to be blocking while holding the
218 // SymbolTable_lock. Otherwise, the system might deadlock, since the
219 // symboltable is used during compilation (VM_thread) The lock free
220 // synchronization is simplified by the fact that we do not delete
221 // entries in the symbol table during normal execution (only during
222 // safepoints).
223
224 Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) {
225 unsigned int hashValue = hash_symbol(name, len);
226 int index = the_table()->hash_to_index(hashValue);
227
228 Symbol* s = the_table()->lookup(index, name, len, hashValue);
229
230 // Found
231 if (s != NULL) return s;
232
233 // Grab SymbolTable_lock first.
234 MutexLocker ml(SymbolTable_lock, THREAD);
235
236 // Otherwise, add to symbol to table
237 return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL);
238 }
239
240 Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) {
241 char* buffer;
242 int index, len;
243 unsigned int hashValue;
244 char* name;
245 {
246 debug_only(No_Safepoint_Verifier nsv;)
247
248 name = (char*)sym->base() + begin;
249 len = end - begin;
250 hashValue = hash_symbol(name, len);
251 index = the_table()->hash_to_index(hashValue);
252 Symbol* s = the_table()->lookup(index, name, len, hashValue);
253
254 // Found
255 if (s != NULL) return s;
256 }
257
258 // Otherwise, add to symbol to table. Copy to a C string first.
259 char stack_buf[128];
260 ResourceMark rm(THREAD);
261 if (len <= 128) {
262 buffer = stack_buf;
263 } else {
264 buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len);
265 }
266 for (int i=0; i<len; i++) {
267 buffer[i] = name[i];
268 }
269 // Make sure there is no safepoint in the code above since name can't move.
270 // We can't include the code in No_Safepoint_Verifier because of the
271 // ResourceMark.
272
273 // Grab SymbolTable_lock first.
274 MutexLocker ml(SymbolTable_lock, THREAD);
275
276 return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL);
277 }
278
279 Symbol* SymbolTable::lookup_only(const char* name, int len,
280 unsigned int& hash) {
281 hash = hash_symbol(name, len);
282 int index = the_table()->hash_to_index(hash);
283
284 Symbol* s = the_table()->lookup(index, name, len, hash);
285 return s;
286 }
287
288 // Look up the address of the literal in the SymbolTable for this Symbol*
289 // Do not create any new symbols
290 // Do not increment the reference count to keep this alive
291 Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){
292 unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length());
293 int index = the_table()->hash_to_index(hash);
294
295 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) {
296 if (e->hash() == hash) {
297 Symbol* literal_sym = e->literal();
298 if (sym == literal_sym) {
299 return e->literal_addr();
300 }
301 }
302 }
303 return NULL;
304 }
305
306 // Suggestion: Push unicode-based lookup all the way into the hashing
307 // and probing logic, so there is no need for convert_to_utf8 until
308 // an actual new Symbol* is created.
309 Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) {
310 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
311 char stack_buf[128];
312 if (utf8_length < (int) sizeof(stack_buf)) {
313 char* chars = stack_buf;
314 UNICODE::convert_to_utf8(name, utf16_length, chars);
315 return lookup(chars, utf8_length, THREAD);
316 } else {
317 ResourceMark rm(THREAD);
318 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
319 UNICODE::convert_to_utf8(name, utf16_length, chars);
320 return lookup(chars, utf8_length, THREAD);
321 }
322 }
323
324 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
325 unsigned int& hash) {
326 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
327 char stack_buf[128];
328 if (utf8_length < (int) sizeof(stack_buf)) {
329 char* chars = stack_buf;
330 UNICODE::convert_to_utf8(name, utf16_length, chars);
331 return lookup_only(chars, utf8_length, hash);
332 } else {
333 ResourceMark rm;
334 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
335 UNICODE::convert_to_utf8(name, utf16_length, chars);
336 return lookup_only(chars, utf8_length, hash);
337 }
338 }
339
340 void SymbolTable::add(ClassLoaderData* loader_data, constantPoolHandle cp,
341 int names_count,
342 const char** names, int* lengths, int* cp_indices,
343 unsigned int* hashValues, TRAPS) {
344 // Grab SymbolTable_lock first.
345 MutexLocker ml(SymbolTable_lock, THREAD);
346
347 SymbolTable* table = the_table();
348 bool added = table->basic_add(loader_data, cp, names_count, names, lengths,
349 cp_indices, hashValues, CHECK);
350 if (!added) {
351 // do it the hard way
352 for (int i=0; i<names_count; i++) {
353 int index = table->hash_to_index(hashValues[i]);
354 bool c_heap = !loader_data->is_the_null_class_loader_data();
355 Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK);
356 cp->symbol_at_put(cp_indices[i], sym);
357 }
358 }
359 }
360
361 Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) {
362 unsigned int hash;
363 Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash);
364 if (result != NULL) {
365 return result;
366 }
367 // Grab SymbolTable_lock first.
368 MutexLocker ml(SymbolTable_lock, THREAD);
369
370 SymbolTable* table = the_table();
371 int index = table->hash_to_index(hash);
372 return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD);
373 }
374
375 Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len,
376 unsigned int hashValue_arg, bool c_heap, TRAPS) {
377 assert(!Universe::heap()->is_in_reserved(name),
378 "proposed name of symbol must be stable");
379
380 // Don't allow symbols to be created which cannot fit in a Symbol*.
381 if (len > Symbol::max_length()) {
382 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
383 "name is too long to represent");
384 }
385
386 // Cannot hit a safepoint in this function because the "this" pointer can move.
387 No_Safepoint_Verifier nsv;
388
389 // Check if the symbol table has been rehashed, if so, need to recalculate
390 // the hash value and index.
391 unsigned int hashValue;
392 int index;
393 if (use_alternate_hashcode()) {
394 hashValue = hash_symbol((const char*)name, len);
395 index = hash_to_index(hashValue);
396 } else {
397 hashValue = hashValue_arg;
398 index = index_arg;
399 }
400
401 // Since look-up was done lock-free, we need to check if another
402 // thread beat us in the race to insert the symbol.
403 Symbol* test = lookup(index, (char*)name, len, hashValue);
404 if (test != NULL) {
405 // A race occurred and another thread introduced the symbol.
406 assert(test->refcount() != 0, "lookup should have incremented the count");
407 return test;
408 }
409
410 // Create a new symbol.
411 Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL);
412 assert(sym->equals((char*)name, len), "symbol must be properly initialized");
413
414 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
415 add_entry(index, entry);
416 return sym;
417 }
418
419 // This version of basic_add adds symbols in batch from the constant pool
420 // parsing.
421 bool SymbolTable::basic_add(ClassLoaderData* loader_data, constantPoolHandle cp,
422 int names_count,
423 const char** names, int* lengths,
424 int* cp_indices, unsigned int* hashValues,
425 TRAPS) {
426
427 // Check symbol names are not too long. If any are too long, don't add any.
428 for (int i = 0; i< names_count; i++) {
429 if (lengths[i] > Symbol::max_length()) {
430 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
431 "name is too long to represent");
432 }
433 }
434
435 // Cannot hit a safepoint in this function because the "this" pointer can move.
436 No_Safepoint_Verifier nsv;
437
438 for (int i=0; i<names_count; i++) {
439 // Check if the symbol table has been rehashed, if so, need to recalculate
440 // the hash value.
441 unsigned int hashValue;
442 if (use_alternate_hashcode()) {
443 hashValue = hash_symbol(names[i], lengths[i]);
444 } else {
445 hashValue = hashValues[i];
446 }
447 // Since look-up was done lock-free, we need to check if another
448 // thread beat us in the race to insert the symbol.
449 int index = hash_to_index(hashValue);
450 Symbol* test = lookup(index, names[i], lengths[i], hashValue);
451 if (test != NULL) {
452 // A race occurred and another thread introduced the symbol, this one
453 // will be dropped and collected. Use test instead.
454 cp->symbol_at_put(cp_indices[i], test);
455 assert(test->refcount() != 0, "lookup should have incremented the count");
456 } else {
457 // Create a new symbol. The null class loader is never unloaded so these
458 // are allocated specially in a permanent arena.
459 bool c_heap = !loader_data->is_the_null_class_loader_data();
460 Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false));
461 assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized"); // why wouldn't it be???
462 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
463 add_entry(index, entry);
464 cp->symbol_at_put(cp_indices[i], sym);
465 }
466 }
467 return true;
468 }
469
470
471 void SymbolTable::verify() {
472 for (int i = 0; i < the_table()->table_size(); ++i) {
473 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
474 for ( ; p != NULL; p = p->next()) {
475 Symbol* s = (Symbol*)(p->literal());
476 guarantee(s != NULL, "symbol is NULL");
477 unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length());
478 guarantee(p->hash() == h, "broken hash in symbol table entry");
479 guarantee(the_table()->hash_to_index(h) == i,
480 "wrong index in symbol table");
481 }
482 }
483 }
484
485 void SymbolTable::dump(outputStream* st) {
486 the_table()->dump_table(st, "SymbolTable");
487 }
488
489
490 //---------------------------------------------------------------------------
491 // Non-product code
492
493 #ifndef PRODUCT
494
495 void SymbolTable::print_histogram() {
496 MutexLocker ml(SymbolTable_lock);
497 const int results_length = 100;
498 int results[results_length];
499 int i,j;
500
501 // initialize results to zero
502 for (j = 0; j < results_length; j++) {
503 results[j] = 0;
504 }
505
506 int total = 0;
507 int max_symbols = 0;
508 int out_of_range = 0;
509 int memory_total = 0;
510 int count = 0;
511 for (i = 0; i < the_table()->table_size(); i++) {
512 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
513 for ( ; p != NULL; p = p->next()) {
514 memory_total += p->literal()->size();
515 count++;
516 int counter = p->literal()->utf8_length();
517 total += counter;
518 if (counter < results_length) {
519 results[counter]++;
520 } else {
521 out_of_range++;
522 }
523 max_symbols = MAX2(max_symbols, counter);
524 }
525 }
526 tty->print_cr("Symbol Table:");
527 tty->print_cr("Total number of symbols %5d", count);
528 tty->print_cr("Total size in memory %5dK",
529 (memory_total*HeapWordSize)/1024);
530 tty->print_cr("Total counted %5d", _symbols_counted);
531 tty->print_cr("Total removed %5d", _symbols_removed);
532 if (_symbols_counted > 0) {
533 tty->print_cr("Percent removed %3.2f",
534 ((float)_symbols_removed/(float)_symbols_counted)* 100);
535 }
536 tty->print_cr("Reference counts %5d", Symbol::_total_count);
537 tty->print_cr("Symbol arena size %5d used %5d",
538 arena()->size_in_bytes(), arena()->used());
539 tty->print_cr("Histogram of symbol length:");
540 tty->print_cr("%8s %5d", "Total ", total);
541 tty->print_cr("%8s %5d", "Maximum", max_symbols);
542 tty->print_cr("%8s %3.2f", "Average",
543 ((float) total / (float) the_table()->table_size()));
544 tty->print_cr("%s", "Histogram:");
545 tty->print_cr(" %s %29s", "Length", "Number chains that length");
546 for (i = 0; i < results_length; i++) {
547 if (results[i] > 0) {
548 tty->print_cr("%6d %10d", i, results[i]);
549 }
550 }
551 if (Verbose) {
552 int line_length = 70;
553 tty->print_cr("%s %30s", " Length", "Number chains that length");
554 for (i = 0; i < results_length; i++) {
555 if (results[i] > 0) {
556 tty->print("%4d", i);
557 for (j = 0; (j < results[i]) && (j < line_length); j++) {
558 tty->print("%1s", "*");
559 }
560 if (j == line_length) {
561 tty->print("%1s", "+");
562 }
563 tty->cr();
564 }
565 }
566 }
567 tty->print_cr(" %s %d: %d\n", "Number chains longer than",
568 results_length, out_of_range);
569 }
570
571 void SymbolTable::print() {
572 for (int i = 0; i < the_table()->table_size(); ++i) {
573 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
574 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
575 if (entry != NULL) {
576 while (entry != NULL) {
577 tty->print(PTR_FORMAT " ", entry->literal());
578 entry->literal()->print();
579 tty->print(" %d", entry->literal()->refcount());
580 p = entry->next_addr();
581 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
582 }
583 tty->cr();
584 }
585 }
586 }
587 #endif // PRODUCT