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 #if INCLUDE_ALL_GCS
39 #include "gc_implementation/g1/g1StringDedup.hpp"
40 #endif
41
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;
46
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");
49
50 Symbol* sym;
51
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 }
65
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 }
74
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 }
86
87 int SymbolTable::_symbols_removed = 0;
88 int SymbolTable::_symbols_counted = 0;
89 volatile int SymbolTable::_parallel_claimed_idx = 0;
90
91 void SymbolTable::buckets_unlink(int start_idx, int end_idx, int* processed, int* removed, size_t* memory_total) {
92 for (int i = start_idx; i < end_idx; ++i) {
93 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
94 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
95 while (entry != NULL) {
96 // Shared entries are normally at the end of the bucket and if we run into
97 // a shared entry, then there is nothing more to remove. However, if we
98 // have rehashed the table, then the shared entries are no longer at the
99 // end of the bucket.
100 if (entry->is_shared() && !use_alternate_hashcode()) {
101 break;
102 }
103 Symbol* s = entry->literal();
104 (*memory_total) += s->size();
105 (*processed)++;
106 assert(s != NULL, "just checking");
107 // If reference count is zero, remove.
108 if (s->refcount() == 0) {
109 assert(!entry->is_shared(), "shared entries should be kept live");
110 delete s;
111 (*removed)++;
112 *p = entry->next();
113 the_table()->free_entry(entry);
114 } else {
115 p = entry->next_addr();
116 }
117 // get next entry
118 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
119 }
120 }
121 }
122
123 // Remove unreferenced symbols from the symbol table
124 // This is done late during GC.
125 void SymbolTable::unlink(int* processed, int* removed) {
126 size_t memory_total = 0;
127 buckets_unlink(0, the_table()->table_size(), processed, removed, &memory_total);
128 _symbols_removed += *removed;
129 _symbols_counted += *processed;
130 // Exclude printing for normal PrintGCDetails because people parse
131 // this output.
132 if (PrintGCDetails && Verbose && WizardMode) {
133 gclog_or_tty->print(" [Symbols=%d size=" SIZE_FORMAT "K] ", *processed,
134 (memory_total*HeapWordSize)/1024);
135 }
136 }
137
138 void SymbolTable::possibly_parallel_unlink(int* processed, int* removed) {
139 const int limit = the_table()->table_size();
140
141 size_t memory_total = 0;
142
143 for (;;) {
144 // Grab next set of buckets to scan
145 int start_idx = Atomic::add(ClaimChunkSize, &_parallel_claimed_idx) - ClaimChunkSize;
146 if (start_idx >= limit) {
147 // End of table
148 break;
149 }
150
151 int end_idx = MIN2(limit, start_idx + ClaimChunkSize);
152 buckets_unlink(start_idx, end_idx, processed, removed, &memory_total);
153 }
154 Atomic::add(*processed, &_symbols_counted);
155 Atomic::add(*removed, &_symbols_removed);
156 // Exclude printing for normal PrintGCDetails because people parse
157 // this output.
158 if (PrintGCDetails && Verbose && WizardMode) {
159 gclog_or_tty->print(" [Symbols: scanned=%d removed=%d size=" SIZE_FORMAT "K] ", *processed, *removed,
160 (memory_total*HeapWordSize)/1024);
161 }
162 }
163
164 // Create a new table and using alternate hash code, populate the new table
165 // with the existing strings. Set flag to use the alternate hash code afterwards.
166 void SymbolTable::rehash_table() {
167 assert(SafepointSynchronize::is_at_safepoint(), "must be at safepoint");
168 // This should never happen with -Xshare:dump but it might in testing mode.
169 if (DumpSharedSpaces) return;
170 // Create a new symbol table
171 SymbolTable* new_table = new SymbolTable();
172
173 the_table()->move_to(new_table);
174
175 // Delete the table and buckets (entries are reused in new table).
176 delete _the_table;
177 // Don't check if we need rehashing until the table gets unbalanced again.
178 // Then rehash with a new global seed.
179 _needs_rehashing = false;
180 _the_table = new_table;
181 }
182
183 // Lookup a symbol in a bucket.
184
185 Symbol* SymbolTable::lookup(int index, const char* name,
186 int len, unsigned int hash) {
187 int count = 0;
188 for (HashtableEntry<Symbol*, mtSymbol>* e = bucket(index); e != NULL; e = e->next()) {
189 count++; // count all entries in this bucket, not just ones with same hash
190 if (e->hash() == hash) {
191 Symbol* sym = e->literal();
192 if (sym->equals(name, len)) {
193 // something is referencing this symbol now.
194 sym->increment_refcount();
195 return sym;
196 }
197 }
198 }
199 // If the bucket size is too deep check if this hash code is insufficient.
200 if (count >= BasicHashtable<mtSymbol>::rehash_count && !needs_rehashing()) {
201 _needs_rehashing = check_rehash_table(count);
202 }
203 return NULL;
204 }
205
206 // Pick hashing algorithm.
207 unsigned int SymbolTable::hash_symbol(const char* s, int len) {
208 return use_alternate_hashcode() ?
209 AltHashing::murmur3_32(seed(), (const jbyte*)s, len) :
210 java_lang_String::hash_code(s, len);
211 }
212
213
214 // We take care not to be blocking while holding the
215 // SymbolTable_lock. Otherwise, the system might deadlock, since the
216 // symboltable is used during compilation (VM_thread) The lock free
217 // synchronization is simplified by the fact that we do not delete
218 // entries in the symbol table during normal execution (only during
219 // safepoints).
220
221 Symbol* SymbolTable::lookup(const char* name, int len, TRAPS) {
222 unsigned int hashValue = hash_symbol(name, len);
223 int index = the_table()->hash_to_index(hashValue);
224
225 Symbol* s = the_table()->lookup(index, name, len, hashValue);
226
227 // Found
228 if (s != NULL) return s;
229
230 // Grab SymbolTable_lock first.
231 MutexLocker ml(SymbolTable_lock, THREAD);
232
233 // Otherwise, add to symbol to table
234 return the_table()->basic_add(index, (u1*)name, len, hashValue, true, CHECK_NULL);
235 }
236
237 Symbol* SymbolTable::lookup(const Symbol* sym, int begin, int end, TRAPS) {
238 char* buffer;
239 int index, len;
240 unsigned int hashValue;
241 char* name;
242 {
243 debug_only(No_Safepoint_Verifier nsv;)
244
245 name = (char*)sym->base() + begin;
246 len = end - begin;
247 hashValue = hash_symbol(name, len);
248 index = the_table()->hash_to_index(hashValue);
249 Symbol* s = the_table()->lookup(index, name, len, hashValue);
250
251 // Found
252 if (s != NULL) return s;
253 }
254
255 // Otherwise, add to symbol to table. Copy to a C string first.
256 char stack_buf[128];
257 ResourceMark rm(THREAD);
258 if (len <= 128) {
259 buffer = stack_buf;
260 } else {
261 buffer = NEW_RESOURCE_ARRAY_IN_THREAD(THREAD, char, len);
262 }
263 for (int i=0; i<len; i++) {
264 buffer[i] = name[i];
265 }
266 // Make sure there is no safepoint in the code above since name can't move.
267 // We can't include the code in No_Safepoint_Verifier because of the
268 // ResourceMark.
269
270 // Grab SymbolTable_lock first.
271 MutexLocker ml(SymbolTable_lock, THREAD);
272
273 return the_table()->basic_add(index, (u1*)buffer, len, hashValue, true, CHECK_NULL);
274 }
275
276 Symbol* SymbolTable::lookup_only(const char* name, int len,
277 unsigned int& hash) {
278 hash = hash_symbol(name, len);
279 int index = the_table()->hash_to_index(hash);
280
281 Symbol* s = the_table()->lookup(index, name, len, hash);
282 return s;
283 }
284
285 // Look up the address of the literal in the SymbolTable for this Symbol*
286 // Do not create any new symbols
287 // Do not increment the reference count to keep this alive
288 Symbol** SymbolTable::lookup_symbol_addr(Symbol* sym){
289 unsigned int hash = hash_symbol((char*)sym->bytes(), sym->utf8_length());
290 int index = the_table()->hash_to_index(hash);
291
292 for (HashtableEntry<Symbol*, mtSymbol>* e = the_table()->bucket(index); e != NULL; e = e->next()) {
293 if (e->hash() == hash) {
294 Symbol* literal_sym = e->literal();
295 if (sym == literal_sym) {
296 return e->literal_addr();
297 }
298 }
299 }
300 return NULL;
301 }
302
303 // Suggestion: Push unicode-based lookup all the way into the hashing
304 // and probing logic, so there is no need for convert_to_utf8 until
305 // an actual new Symbol* is created.
306 Symbol* SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) {
307 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
308 char stack_buf[128];
309 if (utf8_length < (int) sizeof(stack_buf)) {
310 char* chars = stack_buf;
311 UNICODE::convert_to_utf8(name, utf16_length, chars);
312 return lookup(chars, utf8_length, THREAD);
313 } else {
314 ResourceMark rm(THREAD);
315 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
316 UNICODE::convert_to_utf8(name, utf16_length, chars);
317 return lookup(chars, utf8_length, THREAD);
318 }
319 }
320
321 Symbol* SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
322 unsigned int& hash) {
323 int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
324 char stack_buf[128];
325 if (utf8_length < (int) sizeof(stack_buf)) {
326 char* chars = stack_buf;
327 UNICODE::convert_to_utf8(name, utf16_length, chars);
328 return lookup_only(chars, utf8_length, hash);
329 } else {
330 ResourceMark rm;
331 char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
332 UNICODE::convert_to_utf8(name, utf16_length, chars);
333 return lookup_only(chars, utf8_length, hash);
334 }
335 }
336
337 void SymbolTable::add(ClassLoaderData* loader_data, constantPoolHandle cp,
338 int names_count,
339 const char** names, int* lengths, int* cp_indices,
340 unsigned int* hashValues, TRAPS) {
341 // Grab SymbolTable_lock first.
342 MutexLocker ml(SymbolTable_lock, THREAD);
343
344 SymbolTable* table = the_table();
345 bool added = table->basic_add(loader_data, cp, names_count, names, lengths,
346 cp_indices, hashValues, CHECK);
347 if (!added) {
348 // do it the hard way
349 for (int i=0; i<names_count; i++) {
350 int index = table->hash_to_index(hashValues[i]);
351 bool c_heap = !loader_data->is_the_null_class_loader_data();
352 Symbol* sym = table->basic_add(index, (u1*)names[i], lengths[i], hashValues[i], c_heap, CHECK);
353 cp->symbol_at_put(cp_indices[i], sym);
354 }
355 }
356 }
357
358 Symbol* SymbolTable::new_permanent_symbol(const char* name, TRAPS) {
359 unsigned int hash;
360 Symbol* result = SymbolTable::lookup_only((char*)name, (int)strlen(name), hash);
361 if (result != NULL) {
362 return result;
363 }
364 // Grab SymbolTable_lock first.
365 MutexLocker ml(SymbolTable_lock, THREAD);
366
367 SymbolTable* table = the_table();
368 int index = table->hash_to_index(hash);
369 return table->basic_add(index, (u1*)name, (int)strlen(name), hash, false, THREAD);
370 }
371
372 Symbol* SymbolTable::basic_add(int index_arg, u1 *name, int len,
373 unsigned int hashValue_arg, bool c_heap, TRAPS) {
374 assert(!Universe::heap()->is_in_reserved(name),
375 "proposed name of symbol must be stable");
376
377 // Don't allow symbols to be created which cannot fit in a Symbol*.
378 if (len > Symbol::max_length()) {
379 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
380 "name is too long to represent");
381 }
382
383 // Cannot hit a safepoint in this function because the "this" pointer can move.
384 No_Safepoint_Verifier nsv;
385
386 // Check if the symbol table has been rehashed, if so, need to recalculate
387 // the hash value and index.
388 unsigned int hashValue;
389 int index;
390 if (use_alternate_hashcode()) {
391 hashValue = hash_symbol((const char*)name, len);
392 index = hash_to_index(hashValue);
393 } else {
394 hashValue = hashValue_arg;
395 index = index_arg;
396 }
397
398 // Since look-up was done lock-free, we need to check if another
399 // thread beat us in the race to insert the symbol.
400 Symbol* test = lookup(index, (char*)name, len, hashValue);
401 if (test != NULL) {
402 // A race occurred and another thread introduced the symbol.
403 assert(test->refcount() != 0, "lookup should have incremented the count");
404 return test;
405 }
406
407 // Create a new symbol.
408 Symbol* sym = allocate_symbol(name, len, c_heap, CHECK_NULL);
409 assert(sym->equals((char*)name, len), "symbol must be properly initialized");
410
411 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
412 add_entry(index, entry);
413 return sym;
414 }
415
416 // This version of basic_add adds symbols in batch from the constant pool
417 // parsing.
418 bool SymbolTable::basic_add(ClassLoaderData* loader_data, constantPoolHandle cp,
419 int names_count,
420 const char** names, int* lengths,
421 int* cp_indices, unsigned int* hashValues,
422 TRAPS) {
423
424 // Check symbol names are not too long. If any are too long, don't add any.
425 for (int i = 0; i< names_count; i++) {
426 if (lengths[i] > Symbol::max_length()) {
427 THROW_MSG_0(vmSymbols::java_lang_InternalError(),
428 "name is too long to represent");
429 }
430 }
431
432 // Cannot hit a safepoint in this function because the "this" pointer can move.
433 No_Safepoint_Verifier nsv;
434
435 for (int i=0; i<names_count; i++) {
436 // Check if the symbol table has been rehashed, if so, need to recalculate
437 // the hash value.
438 unsigned int hashValue;
439 if (use_alternate_hashcode()) {
440 hashValue = hash_symbol(names[i], lengths[i]);
441 } else {
442 hashValue = hashValues[i];
443 }
444 // Since look-up was done lock-free, we need to check if another
445 // thread beat us in the race to insert the symbol.
446 int index = hash_to_index(hashValue);
447 Symbol* test = lookup(index, names[i], lengths[i], hashValue);
448 if (test != NULL) {
449 // A race occurred and another thread introduced the symbol, this one
450 // will be dropped and collected. Use test instead.
451 cp->symbol_at_put(cp_indices[i], test);
452 assert(test->refcount() != 0, "lookup should have incremented the count");
453 } else {
454 // Create a new symbol. The null class loader is never unloaded so these
455 // are allocated specially in a permanent arena.
456 bool c_heap = !loader_data->is_the_null_class_loader_data();
457 Symbol* sym = allocate_symbol((const u1*)names[i], lengths[i], c_heap, CHECK_(false));
458 assert(sym->equals(names[i], lengths[i]), "symbol must be properly initialized"); // why wouldn't it be???
459 HashtableEntry<Symbol*, mtSymbol>* entry = new_entry(hashValue, sym);
460 add_entry(index, entry);
461 cp->symbol_at_put(cp_indices[i], sym);
462 }
463 }
464 return true;
465 }
466
467
468 void SymbolTable::verify() {
469 for (int i = 0; i < the_table()->table_size(); ++i) {
470 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
471 for ( ; p != NULL; p = p->next()) {
472 Symbol* s = (Symbol*)(p->literal());
473 guarantee(s != NULL, "symbol is NULL");
474 unsigned int h = hash_symbol((char*)s->bytes(), s->utf8_length());
475 guarantee(p->hash() == h, "broken hash in symbol table entry");
476 guarantee(the_table()->hash_to_index(h) == i,
477 "wrong index in symbol table");
478 }
479 }
480 }
481
482 void SymbolTable::dump(outputStream* st) {
483 the_table()->dump_table(st, "SymbolTable");
484 }
485
486
487 //---------------------------------------------------------------------------
488 // Non-product code
489
490 #ifndef PRODUCT
491
492 void SymbolTable::print_histogram() {
493 MutexLocker ml(SymbolTable_lock);
494 const int results_length = 100;
495 int results[results_length];
496 int i,j;
497
498 // initialize results to zero
499 for (j = 0; j < results_length; j++) {
500 results[j] = 0;
501 }
502
503 int total = 0;
504 int max_symbols = 0;
505 int out_of_range = 0;
506 int memory_total = 0;
507 int count = 0;
508 for (i = 0; i < the_table()->table_size(); i++) {
509 HashtableEntry<Symbol*, mtSymbol>* p = the_table()->bucket(i);
510 for ( ; p != NULL; p = p->next()) {
511 memory_total += p->literal()->size();
512 count++;
513 int counter = p->literal()->utf8_length();
514 total += counter;
515 if (counter < results_length) {
516 results[counter]++;
517 } else {
518 out_of_range++;
519 }
520 max_symbols = MAX2(max_symbols, counter);
521 }
522 }
523 tty->print_cr("Symbol Table:");
524 tty->print_cr("Total number of symbols %5d", count);
525 tty->print_cr("Total size in memory %5dK",
526 (memory_total*HeapWordSize)/1024);
527 tty->print_cr("Total counted %5d", _symbols_counted);
528 tty->print_cr("Total removed %5d", _symbols_removed);
529 if (_symbols_counted > 0) {
530 tty->print_cr("Percent removed %3.2f",
531 ((float)_symbols_removed/(float)_symbols_counted)* 100);
532 }
533 tty->print_cr("Reference counts %5d", Symbol::_total_count);
534 tty->print_cr("Symbol arena size %5d used %5d",
535 arena()->size_in_bytes(), arena()->used());
536 tty->print_cr("Histogram of symbol length:");
537 tty->print_cr("%8s %5d", "Total ", total);
538 tty->print_cr("%8s %5d", "Maximum", max_symbols);
539 tty->print_cr("%8s %3.2f", "Average",
540 ((float) total / (float) the_table()->table_size()));
541 tty->print_cr("%s", "Histogram:");
542 tty->print_cr(" %s %29s", "Length", "Number chains that length");
543 for (i = 0; i < results_length; i++) {
544 if (results[i] > 0) {
545 tty->print_cr("%6d %10d", i, results[i]);
546 }
547 }
548 if (Verbose) {
549 int line_length = 70;
550 tty->print_cr("%s %30s", " Length", "Number chains that length");
551 for (i = 0; i < results_length; i++) {
552 if (results[i] > 0) {
553 tty->print("%4d", i);
554 for (j = 0; (j < results[i]) && (j < line_length); j++) {
555 tty->print("%1s", "*");
556 }
557 if (j == line_length) {
558 tty->print("%1s", "+");
559 }
560 tty->cr();
561 }
562 }
563 }
564 tty->print_cr(" %s %d: %d\n", "Number chains longer than",
565 results_length, out_of_range);
566 }
567
568 void SymbolTable::print() {
569 for (int i = 0; i < the_table()->table_size(); ++i) {
570 HashtableEntry<Symbol*, mtSymbol>** p = the_table()->bucket_addr(i);
571 HashtableEntry<Symbol*, mtSymbol>* entry = the_table()->bucket(i);
572 if (entry != NULL) {
573 while (entry != NULL) {
574 tty->print(PTR_FORMAT " ", entry->literal());
575 entry->literal()->print();
576 tty->print(" %d", entry->literal()->refcount());
577 p = entry->next_addr();
578 entry = (HashtableEntry<Symbol*, mtSymbol>*)HashtableEntry<Symbol*, mtSymbol>::make_ptr(*p);
579 }
580 tty->cr();
581 }
582 }
583 }
584 #endif // PRODUCT