1 /* |
|
2 * Copyright (c) 1997, 2016, 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/classLoader.hpp" |
|
27 #include "code/codeCache.hpp" |
|
28 #include "code/vtableStubs.hpp" |
|
29 #include "gc/shared/collectedHeap.inline.hpp" |
|
30 #include "interpreter/interpreter.hpp" |
|
31 #include "memory/allocation.inline.hpp" |
|
32 #include "memory/resourceArea.hpp" |
|
33 #include "memory/universe.inline.hpp" |
|
34 #include "oops/oop.inline.hpp" |
|
35 #include "oops/symbol.hpp" |
|
36 #include "runtime/deoptimization.hpp" |
|
37 #include "runtime/fprofiler.hpp" |
|
38 #include "runtime/mutexLocker.hpp" |
|
39 #include "runtime/stubCodeGenerator.hpp" |
|
40 #include "runtime/stubRoutines.hpp" |
|
41 #include "runtime/task.hpp" |
|
42 #include "runtime/thread.inline.hpp" |
|
43 #include "runtime/vframe.hpp" |
|
44 #include "utilities/macros.hpp" |
|
45 |
|
46 // Static fields of FlatProfiler |
|
47 int FlatProfiler::received_gc_ticks = 0; |
|
48 int FlatProfiler::vm_operation_ticks = 0; |
|
49 int FlatProfiler::threads_lock_ticks = 0; |
|
50 int FlatProfiler::class_loader_ticks = 0; |
|
51 int FlatProfiler::extra_ticks = 0; |
|
52 int FlatProfiler::blocked_ticks = 0; |
|
53 int FlatProfiler::deopt_ticks = 0; |
|
54 int FlatProfiler::unknown_ticks = 0; |
|
55 int FlatProfiler::interpreter_ticks = 0; |
|
56 int FlatProfiler::compiler_ticks = 0; |
|
57 int FlatProfiler::received_ticks = 0; |
|
58 int FlatProfiler::delivered_ticks = 0; |
|
59 int* FlatProfiler::bytecode_ticks = NULL; |
|
60 int* FlatProfiler::bytecode_ticks_stub = NULL; |
|
61 int FlatProfiler::all_int_ticks = 0; |
|
62 int FlatProfiler::all_comp_ticks = 0; |
|
63 int FlatProfiler::all_ticks = 0; |
|
64 bool FlatProfiler::full_profile_flag = false; |
|
65 ThreadProfiler* FlatProfiler::thread_profiler = NULL; |
|
66 ThreadProfiler* FlatProfiler::vm_thread_profiler = NULL; |
|
67 FlatProfilerTask* FlatProfiler::task = NULL; |
|
68 elapsedTimer FlatProfiler::timer; |
|
69 int FlatProfiler::interval_ticks_previous = 0; |
|
70 IntervalData* FlatProfiler::interval_data = NULL; |
|
71 |
|
72 ThreadProfiler::ThreadProfiler() { |
|
73 // Space for the ProfilerNodes |
|
74 const int area_size = 1 * ProfilerNodeSize * 1024; |
|
75 area_bottom = AllocateHeap(area_size, mtInternal); |
|
76 area_top = area_bottom; |
|
77 area_limit = area_bottom + area_size; |
|
78 |
|
79 // ProfilerNode pointer table |
|
80 table = NEW_C_HEAP_ARRAY(ProfilerNode*, table_size, mtInternal); |
|
81 initialize(); |
|
82 engaged = false; |
|
83 } |
|
84 |
|
85 ThreadProfiler::~ThreadProfiler() { |
|
86 FreeHeap(area_bottom); |
|
87 area_bottom = NULL; |
|
88 area_top = NULL; |
|
89 area_limit = NULL; |
|
90 FreeHeap(table); |
|
91 table = NULL; |
|
92 } |
|
93 |
|
94 // Statics for ThreadProfiler |
|
95 int ThreadProfiler::table_size = 1024; |
|
96 |
|
97 int ThreadProfiler::entry(int value) { |
|
98 value = (value > 0) ? value : -value; |
|
99 return value % table_size; |
|
100 } |
|
101 |
|
102 ThreadProfilerMark::ThreadProfilerMark(ThreadProfilerMark::Region r) { |
|
103 _r = r; |
|
104 _pp = NULL; |
|
105 assert(((r > ThreadProfilerMark::noRegion) && (r < ThreadProfilerMark::maxRegion)), "ThreadProfilerMark::Region out of bounds"); |
|
106 Thread* tp = Thread::current(); |
|
107 if (tp != NULL && tp->is_Java_thread()) { |
|
108 JavaThread* jtp = (JavaThread*) tp; |
|
109 ThreadProfiler* pp = jtp->get_thread_profiler(); |
|
110 _pp = pp; |
|
111 if (pp != NULL) { |
|
112 pp->region_flag[r] = true; |
|
113 } |
|
114 } |
|
115 } |
|
116 |
|
117 ThreadProfilerMark::~ThreadProfilerMark() { |
|
118 if (_pp != NULL) { |
|
119 _pp->region_flag[_r] = false; |
|
120 } |
|
121 _pp = NULL; |
|
122 } |
|
123 |
|
124 // Random other statics |
|
125 static const int col1 = 2; // position of output column 1 |
|
126 static const int col2 = 11; // position of output column 2 |
|
127 static const int col3 = 25; // position of output column 3 |
|
128 static const int col4 = 55; // position of output column 4 |
|
129 |
|
130 |
|
131 // Used for detailed profiling of nmethods. |
|
132 class PCRecorder : AllStatic { |
|
133 private: |
|
134 static int* counters; |
|
135 static address base; |
|
136 enum { |
|
137 bucket_size = 16 |
|
138 }; |
|
139 static int index_for(address pc) { return (pc - base)/bucket_size; } |
|
140 static address pc_for(int index) { return base + (index * bucket_size); } |
|
141 static int size() { |
|
142 return ((int)CodeCache::max_capacity())/bucket_size * BytesPerWord; |
|
143 } |
|
144 public: |
|
145 static address bucket_start_for(address pc) { |
|
146 if (counters == NULL) return NULL; |
|
147 return pc_for(index_for(pc)); |
|
148 } |
|
149 static int bucket_count_for(address pc) { return counters[index_for(pc)]; } |
|
150 static void init(); |
|
151 static void record(address pc); |
|
152 static void print(); |
|
153 static void print_blobs(CodeBlob* cb); |
|
154 }; |
|
155 |
|
156 int* PCRecorder::counters = NULL; |
|
157 address PCRecorder::base = NULL; |
|
158 |
|
159 void PCRecorder::init() { |
|
160 MutexLockerEx lm(CodeCache_lock, Mutex::_no_safepoint_check_flag); |
|
161 int s = size(); |
|
162 counters = NEW_C_HEAP_ARRAY(int, s, mtInternal); |
|
163 for (int index = 0; index < s; index++) { |
|
164 counters[index] = 0; |
|
165 } |
|
166 base = CodeCache::low_bound(); |
|
167 } |
|
168 |
|
169 void PCRecorder::record(address pc) { |
|
170 if (counters == NULL) return; |
|
171 assert(CodeCache::contains(pc), "must be in CodeCache"); |
|
172 counters[index_for(pc)]++; |
|
173 } |
|
174 |
|
175 |
|
176 address FlatProfiler::bucket_start_for(address pc) { |
|
177 return PCRecorder::bucket_start_for(pc); |
|
178 } |
|
179 |
|
180 int FlatProfiler::bucket_count_for(address pc) { |
|
181 return PCRecorder::bucket_count_for(pc); |
|
182 } |
|
183 |
|
184 void PCRecorder::print() { |
|
185 if (counters == NULL) return; |
|
186 |
|
187 tty->cr(); |
|
188 tty->print_cr("Printing compiled methods with PC buckets having more than " INTX_FORMAT " ticks", ProfilerPCTickThreshold); |
|
189 tty->print_cr("==================================================================="); |
|
190 tty->cr(); |
|
191 |
|
192 GrowableArray<CodeBlob*>* candidates = new GrowableArray<CodeBlob*>(20); |
|
193 |
|
194 |
|
195 int s; |
|
196 { |
|
197 MutexLockerEx lm(CodeCache_lock, Mutex::_no_safepoint_check_flag); |
|
198 s = size(); |
|
199 } |
|
200 |
|
201 for (int index = 0; index < s; index++) { |
|
202 int count = counters[index]; |
|
203 if (count > ProfilerPCTickThreshold) { |
|
204 address pc = pc_for(index); |
|
205 CodeBlob* cb = CodeCache::find_blob_unsafe(pc); |
|
206 if (cb != NULL && candidates->find(cb) < 0) { |
|
207 candidates->push(cb); |
|
208 } |
|
209 } |
|
210 } |
|
211 for (int i = 0; i < candidates->length(); i++) { |
|
212 print_blobs(candidates->at(i)); |
|
213 } |
|
214 } |
|
215 |
|
216 void PCRecorder::print_blobs(CodeBlob* cb) { |
|
217 if (cb != NULL) { |
|
218 cb->print(); |
|
219 if (cb->is_nmethod()) { |
|
220 ((nmethod*)cb)->print_code(); |
|
221 } |
|
222 tty->cr(); |
|
223 } else { |
|
224 tty->print_cr("stub code"); |
|
225 } |
|
226 } |
|
227 |
|
228 class tick_counter { // holds tick info for one node |
|
229 public: |
|
230 int ticks_in_code; |
|
231 int ticks_in_native; |
|
232 |
|
233 tick_counter() { ticks_in_code = ticks_in_native = 0; } |
|
234 tick_counter(int code, int native) { ticks_in_code = code; ticks_in_native = native; } |
|
235 |
|
236 int total() const { |
|
237 return (ticks_in_code + ticks_in_native); |
|
238 } |
|
239 |
|
240 void add(tick_counter* a) { |
|
241 ticks_in_code += a->ticks_in_code; |
|
242 ticks_in_native += a->ticks_in_native; |
|
243 } |
|
244 |
|
245 void update(TickPosition where) { |
|
246 switch(where) { |
|
247 case tp_code: ticks_in_code++; break; |
|
248 case tp_native: ticks_in_native++; break; |
|
249 } |
|
250 } |
|
251 |
|
252 void print_code(outputStream* st, int total_ticks) { |
|
253 st->print("%5.1f%% %5d ", total() * 100.0 / total_ticks, ticks_in_code); |
|
254 } |
|
255 |
|
256 void print_native(outputStream* st) { |
|
257 st->print(" + %5d ", ticks_in_native); |
|
258 } |
|
259 }; |
|
260 |
|
261 class ProfilerNode { |
|
262 private: |
|
263 ProfilerNode* _next; |
|
264 public: |
|
265 tick_counter ticks; |
|
266 |
|
267 public: |
|
268 |
|
269 void* operator new(size_t size, ThreadProfiler* tp) throw(); |
|
270 void operator delete(void* p); |
|
271 |
|
272 ProfilerNode() { |
|
273 _next = NULL; |
|
274 } |
|
275 |
|
276 virtual ~ProfilerNode() { |
|
277 if (_next) |
|
278 delete _next; |
|
279 } |
|
280 |
|
281 void set_next(ProfilerNode* n) { _next = n; } |
|
282 ProfilerNode* next() { return _next; } |
|
283 |
|
284 void update(TickPosition where) { ticks.update(where);} |
|
285 int total_ticks() { return ticks.total(); } |
|
286 |
|
287 virtual bool is_interpreted() const { return false; } |
|
288 virtual bool is_compiled() const { return false; } |
|
289 virtual bool is_stub() const { return false; } |
|
290 virtual bool is_runtime_stub() const{ return false; } |
|
291 virtual void oops_do(OopClosure* f) = 0; |
|
292 |
|
293 virtual bool interpreted_match(Method* m) const { return false; } |
|
294 virtual bool compiled_match(Method* m ) const { return false; } |
|
295 virtual bool stub_match(Method* m, const char* name) const { return false; } |
|
296 virtual bool adapter_match() const { return false; } |
|
297 virtual bool runtimeStub_match(const CodeBlob* stub, const char* name) const { return false; } |
|
298 virtual bool unknown_compiled_match(const CodeBlob* cb) const { return false; } |
|
299 |
|
300 static void print_title(outputStream* st) { |
|
301 st->print(" + native"); |
|
302 st->fill_to(col3); |
|
303 st->print("Method"); |
|
304 st->fill_to(col4); |
|
305 st->cr(); |
|
306 } |
|
307 |
|
308 static void print_total(outputStream* st, tick_counter* t, int total, const char* msg) { |
|
309 t->print_code(st, total); |
|
310 st->fill_to(col2); |
|
311 t->print_native(st); |
|
312 st->fill_to(col3); |
|
313 st->print("%s", msg); |
|
314 st->cr(); |
|
315 } |
|
316 |
|
317 virtual Method* method() = 0; |
|
318 |
|
319 virtual void print_method_on(outputStream* st) { |
|
320 int limit; |
|
321 int i; |
|
322 Method* m = method(); |
|
323 Symbol* k = m->klass_name(); |
|
324 // Print the class name with dots instead of slashes |
|
325 limit = k->utf8_length(); |
|
326 for (i = 0 ; i < limit ; i += 1) { |
|
327 char c = (char) k->byte_at(i); |
|
328 if (c == '/') { |
|
329 c = '.'; |
|
330 } |
|
331 st->print("%c", c); |
|
332 } |
|
333 if (limit > 0) { |
|
334 st->print("."); |
|
335 } |
|
336 Symbol* n = m->name(); |
|
337 limit = n->utf8_length(); |
|
338 for (i = 0 ; i < limit ; i += 1) { |
|
339 char c = (char) n->byte_at(i); |
|
340 st->print("%c", c); |
|
341 } |
|
342 if (Verbose || WizardMode) { |
|
343 // Disambiguate overloaded methods |
|
344 Symbol* sig = m->signature(); |
|
345 sig->print_symbol_on(st); |
|
346 } else if (MethodHandles::is_signature_polymorphic(m->intrinsic_id())) |
|
347 // compare with Method::print_short_name |
|
348 MethodHandles::print_as_basic_type_signature_on(st, m->signature(), true); |
|
349 } |
|
350 |
|
351 virtual void print(outputStream* st, int total_ticks) { |
|
352 ticks.print_code(st, total_ticks); |
|
353 st->fill_to(col2); |
|
354 ticks.print_native(st); |
|
355 st->fill_to(col3); |
|
356 print_method_on(st); |
|
357 st->cr(); |
|
358 } |
|
359 |
|
360 // for hashing into the table |
|
361 static int hash(Method* method) { |
|
362 // The point here is to try to make something fairly unique |
|
363 // out of the fields we can read without grabbing any locks |
|
364 // since the method may be locked when we need the hash. |
|
365 return ( |
|
366 method->code_size() ^ |
|
367 method->max_stack() ^ |
|
368 method->max_locals() ^ |
|
369 method->size_of_parameters()); |
|
370 } |
|
371 |
|
372 // for sorting |
|
373 static int compare(ProfilerNode** a, ProfilerNode** b) { |
|
374 return (*b)->total_ticks() - (*a)->total_ticks(); |
|
375 } |
|
376 }; |
|
377 |
|
378 void* ProfilerNode::operator new(size_t size, ThreadProfiler* tp) throw() { |
|
379 void* result = (void*) tp->area_top; |
|
380 tp->area_top += size; |
|
381 |
|
382 if (tp->area_top > tp->area_limit) { |
|
383 fatal("flat profiler buffer overflow"); |
|
384 } |
|
385 return result; |
|
386 } |
|
387 |
|
388 void ProfilerNode::operator delete(void* p){ |
|
389 } |
|
390 |
|
391 class interpretedNode : public ProfilerNode { |
|
392 private: |
|
393 Method* _method; |
|
394 oop _class_loader; // needed to keep metadata for the method alive |
|
395 public: |
|
396 interpretedNode(Method* method, TickPosition where) : ProfilerNode() { |
|
397 _method = method; |
|
398 _class_loader = method->method_holder()->class_loader(); |
|
399 update(where); |
|
400 } |
|
401 |
|
402 bool is_interpreted() const { return true; } |
|
403 |
|
404 bool interpreted_match(Method* m) const { |
|
405 return _method == m; |
|
406 } |
|
407 |
|
408 void oops_do(OopClosure* f) { |
|
409 f->do_oop(&_class_loader); |
|
410 } |
|
411 |
|
412 Method* method() { return _method; } |
|
413 |
|
414 static void print_title(outputStream* st) { |
|
415 st->fill_to(col1); |
|
416 st->print("%11s", "Interpreted"); |
|
417 ProfilerNode::print_title(st); |
|
418 } |
|
419 |
|
420 void print(outputStream* st, int total_ticks) { |
|
421 ProfilerNode::print(st, total_ticks); |
|
422 } |
|
423 |
|
424 void print_method_on(outputStream* st) { |
|
425 ProfilerNode::print_method_on(st); |
|
426 MethodCounters* mcs = method()->method_counters(); |
|
427 if (Verbose && mcs != NULL) mcs->invocation_counter()->print_short(); |
|
428 } |
|
429 }; |
|
430 |
|
431 class compiledNode : public ProfilerNode { |
|
432 private: |
|
433 Method* _method; |
|
434 oop _class_loader; // needed to keep metadata for the method alive |
|
435 public: |
|
436 compiledNode(Method* method, TickPosition where) : ProfilerNode() { |
|
437 _method = method; |
|
438 _class_loader = method->method_holder()->class_loader(); |
|
439 update(where); |
|
440 } |
|
441 bool is_compiled() const { return true; } |
|
442 |
|
443 bool compiled_match(Method* m) const { |
|
444 return _method == m; |
|
445 } |
|
446 |
|
447 Method* method() { return _method; } |
|
448 |
|
449 void oops_do(OopClosure* f) { |
|
450 f->do_oop(&_class_loader); |
|
451 } |
|
452 |
|
453 static void print_title(outputStream* st) { |
|
454 st->fill_to(col1); |
|
455 st->print("%11s", "Compiled"); |
|
456 ProfilerNode::print_title(st); |
|
457 } |
|
458 |
|
459 void print(outputStream* st, int total_ticks) { |
|
460 ProfilerNode::print(st, total_ticks); |
|
461 } |
|
462 |
|
463 void print_method_on(outputStream* st) { |
|
464 ProfilerNode::print_method_on(st); |
|
465 } |
|
466 }; |
|
467 |
|
468 class stubNode : public ProfilerNode { |
|
469 private: |
|
470 Method* _method; |
|
471 oop _class_loader; // needed to keep metadata for the method alive |
|
472 const char* _symbol; // The name of the nearest VM symbol (for +ProfileVM). Points to a unique string |
|
473 public: |
|
474 stubNode(Method* method, const char* name, TickPosition where) : ProfilerNode() { |
|
475 _method = method; |
|
476 _class_loader = method->method_holder()->class_loader(); |
|
477 _symbol = name; |
|
478 update(where); |
|
479 } |
|
480 |
|
481 bool is_stub() const { return true; } |
|
482 |
|
483 void oops_do(OopClosure* f) { |
|
484 f->do_oop(&_class_loader); |
|
485 } |
|
486 |
|
487 bool stub_match(Method* m, const char* name) const { |
|
488 return (_method == m) && (_symbol == name); |
|
489 } |
|
490 |
|
491 Method* method() { return _method; } |
|
492 |
|
493 static void print_title(outputStream* st) { |
|
494 st->fill_to(col1); |
|
495 st->print("%11s", "Stub"); |
|
496 ProfilerNode::print_title(st); |
|
497 } |
|
498 |
|
499 void print(outputStream* st, int total_ticks) { |
|
500 ProfilerNode::print(st, total_ticks); |
|
501 } |
|
502 |
|
503 void print_method_on(outputStream* st) { |
|
504 ProfilerNode::print_method_on(st); |
|
505 print_symbol_on(st); |
|
506 } |
|
507 |
|
508 void print_symbol_on(outputStream* st) { |
|
509 if(_symbol) { |
|
510 st->print(" (%s)", _symbol); |
|
511 } |
|
512 } |
|
513 }; |
|
514 |
|
515 class adapterNode : public ProfilerNode { |
|
516 public: |
|
517 adapterNode(TickPosition where) : ProfilerNode() { |
|
518 update(where); |
|
519 } |
|
520 bool is_compiled() const { return true; } |
|
521 |
|
522 bool adapter_match() const { return true; } |
|
523 |
|
524 Method* method() { return NULL; } |
|
525 |
|
526 void oops_do(OopClosure* f) { |
|
527 ; |
|
528 } |
|
529 |
|
530 void print(outputStream* st, int total_ticks) { |
|
531 ProfilerNode::print(st, total_ticks); |
|
532 } |
|
533 |
|
534 void print_method_on(outputStream* st) { |
|
535 st->print("%s", "adapters"); |
|
536 } |
|
537 }; |
|
538 |
|
539 class runtimeStubNode : public ProfilerNode { |
|
540 private: |
|
541 const RuntimeStub* _stub; |
|
542 const char* _symbol; // The name of the nearest VM symbol when ProfileVM is on. Points to a unique string. |
|
543 public: |
|
544 runtimeStubNode(const CodeBlob* stub, const char* name, TickPosition where) : ProfilerNode(), _stub(NULL), _symbol(name) { |
|
545 assert(stub->is_runtime_stub(), "wrong code blob"); |
|
546 _stub = (RuntimeStub*) stub; |
|
547 update(where); |
|
548 } |
|
549 |
|
550 bool is_runtime_stub() const { return true; } |
|
551 |
|
552 bool runtimeStub_match(const CodeBlob* stub, const char* name) const { |
|
553 assert(stub->is_runtime_stub(), "wrong code blob"); |
|
554 return _stub->entry_point() == ((RuntimeStub*)stub)->entry_point() && |
|
555 (_symbol == name); |
|
556 } |
|
557 |
|
558 Method* method() { return NULL; } |
|
559 |
|
560 static void print_title(outputStream* st) { |
|
561 st->fill_to(col1); |
|
562 st->print("%11s", "Runtime stub"); |
|
563 ProfilerNode::print_title(st); |
|
564 } |
|
565 |
|
566 void oops_do(OopClosure* f) { |
|
567 ; |
|
568 } |
|
569 |
|
570 void print(outputStream* st, int total_ticks) { |
|
571 ProfilerNode::print(st, total_ticks); |
|
572 } |
|
573 |
|
574 void print_method_on(outputStream* st) { |
|
575 st->print("%s", _stub->name()); |
|
576 print_symbol_on(st); |
|
577 } |
|
578 |
|
579 void print_symbol_on(outputStream* st) { |
|
580 if(_symbol) { |
|
581 st->print(" (%s)", _symbol); |
|
582 } |
|
583 } |
|
584 }; |
|
585 |
|
586 |
|
587 class unknown_compiledNode : public ProfilerNode { |
|
588 const char *_name; |
|
589 public: |
|
590 unknown_compiledNode(const CodeBlob* cb, TickPosition where) : ProfilerNode() { |
|
591 if ( cb->is_buffer_blob() ) |
|
592 _name = ((const BufferBlob*)cb)->name(); |
|
593 else |
|
594 _name = ((const SingletonBlob*)cb)->name(); |
|
595 update(where); |
|
596 } |
|
597 bool is_compiled() const { return true; } |
|
598 |
|
599 bool unknown_compiled_match(const CodeBlob* cb) const { |
|
600 if ( cb->is_buffer_blob() ) |
|
601 return !strcmp(((const BufferBlob*)cb)->name(), _name); |
|
602 else |
|
603 return !strcmp(((const SingletonBlob*)cb)->name(), _name); |
|
604 } |
|
605 |
|
606 Method* method() { return NULL; } |
|
607 |
|
608 void oops_do(OopClosure* f) { |
|
609 ; |
|
610 } |
|
611 |
|
612 void print(outputStream* st, int total_ticks) { |
|
613 ProfilerNode::print(st, total_ticks); |
|
614 } |
|
615 |
|
616 void print_method_on(outputStream* st) { |
|
617 st->print("%s", _name); |
|
618 } |
|
619 }; |
|
620 |
|
621 class vmNode : public ProfilerNode { |
|
622 private: |
|
623 const char* _name; // "optional" name obtained by os means such as dll lookup |
|
624 public: |
|
625 vmNode(const TickPosition where) : ProfilerNode() { |
|
626 _name = NULL; |
|
627 update(where); |
|
628 } |
|
629 |
|
630 vmNode(const char* name, const TickPosition where) : ProfilerNode() { |
|
631 _name = os::strdup(name); |
|
632 update(where); |
|
633 } |
|
634 |
|
635 ~vmNode() { |
|
636 if (_name != NULL) { |
|
637 os::free((void*)_name); |
|
638 } |
|
639 } |
|
640 |
|
641 const char *name() const { return _name; } |
|
642 bool is_compiled() const { return true; } |
|
643 |
|
644 bool vm_match(const char* name) const { return strcmp(name, _name) == 0; } |
|
645 |
|
646 Method* method() { return NULL; } |
|
647 |
|
648 static int hash(const char* name){ |
|
649 // Compute a simple hash |
|
650 const char* cp = name; |
|
651 int h = 0; |
|
652 |
|
653 if(name != NULL){ |
|
654 while(*cp != '\0'){ |
|
655 h = (h << 1) ^ *cp; |
|
656 cp++; |
|
657 } |
|
658 } |
|
659 return h; |
|
660 } |
|
661 |
|
662 void oops_do(OopClosure* f) { |
|
663 ; |
|
664 } |
|
665 |
|
666 void print(outputStream* st, int total_ticks) { |
|
667 ProfilerNode::print(st, total_ticks); |
|
668 } |
|
669 |
|
670 void print_method_on(outputStream* st) { |
|
671 if(_name==NULL){ |
|
672 st->print("%s", "unknown code"); |
|
673 } |
|
674 else { |
|
675 st->print("%s", _name); |
|
676 } |
|
677 } |
|
678 }; |
|
679 |
|
680 void ThreadProfiler::interpreted_update(Method* method, TickPosition where) { |
|
681 int index = entry(ProfilerNode::hash(method)); |
|
682 if (!table[index]) { |
|
683 table[index] = new (this) interpretedNode(method, where); |
|
684 } else { |
|
685 ProfilerNode* prev = table[index]; |
|
686 for(ProfilerNode* node = prev; node; node = node->next()) { |
|
687 if (node->interpreted_match(method)) { |
|
688 node->update(where); |
|
689 return; |
|
690 } |
|
691 prev = node; |
|
692 } |
|
693 prev->set_next(new (this) interpretedNode(method, where)); |
|
694 } |
|
695 } |
|
696 |
|
697 void ThreadProfiler::compiled_update(Method* method, TickPosition where) { |
|
698 int index = entry(ProfilerNode::hash(method)); |
|
699 if (!table[index]) { |
|
700 table[index] = new (this) compiledNode(method, where); |
|
701 } else { |
|
702 ProfilerNode* prev = table[index]; |
|
703 for(ProfilerNode* node = prev; node; node = node->next()) { |
|
704 if (node->compiled_match(method)) { |
|
705 node->update(where); |
|
706 return; |
|
707 } |
|
708 prev = node; |
|
709 } |
|
710 prev->set_next(new (this) compiledNode(method, where)); |
|
711 } |
|
712 } |
|
713 |
|
714 void ThreadProfiler::stub_update(Method* method, const char* name, TickPosition where) { |
|
715 int index = entry(ProfilerNode::hash(method)); |
|
716 if (!table[index]) { |
|
717 table[index] = new (this) stubNode(method, name, where); |
|
718 } else { |
|
719 ProfilerNode* prev = table[index]; |
|
720 for(ProfilerNode* node = prev; node; node = node->next()) { |
|
721 if (node->stub_match(method, name)) { |
|
722 node->update(where); |
|
723 return; |
|
724 } |
|
725 prev = node; |
|
726 } |
|
727 prev->set_next(new (this) stubNode(method, name, where)); |
|
728 } |
|
729 } |
|
730 |
|
731 void ThreadProfiler::adapter_update(TickPosition where) { |
|
732 int index = 0; |
|
733 if (!table[index]) { |
|
734 table[index] = new (this) adapterNode(where); |
|
735 } else { |
|
736 ProfilerNode* prev = table[index]; |
|
737 for(ProfilerNode* node = prev; node; node = node->next()) { |
|
738 if (node->adapter_match()) { |
|
739 node->update(where); |
|
740 return; |
|
741 } |
|
742 prev = node; |
|
743 } |
|
744 prev->set_next(new (this) adapterNode(where)); |
|
745 } |
|
746 } |
|
747 |
|
748 void ThreadProfiler::runtime_stub_update(const CodeBlob* stub, const char* name, TickPosition where) { |
|
749 int index = 0; |
|
750 if (!table[index]) { |
|
751 table[index] = new (this) runtimeStubNode(stub, name, where); |
|
752 } else { |
|
753 ProfilerNode* prev = table[index]; |
|
754 for(ProfilerNode* node = prev; node; node = node->next()) { |
|
755 if (node->runtimeStub_match(stub, name)) { |
|
756 node->update(where); |
|
757 return; |
|
758 } |
|
759 prev = node; |
|
760 } |
|
761 prev->set_next(new (this) runtimeStubNode(stub, name, where)); |
|
762 } |
|
763 } |
|
764 |
|
765 |
|
766 void ThreadProfiler::unknown_compiled_update(const CodeBlob* cb, TickPosition where) { |
|
767 int index = 0; |
|
768 if (!table[index]) { |
|
769 table[index] = new (this) unknown_compiledNode(cb, where); |
|
770 } else { |
|
771 ProfilerNode* prev = table[index]; |
|
772 for(ProfilerNode* node = prev; node; node = node->next()) { |
|
773 if (node->unknown_compiled_match(cb)) { |
|
774 node->update(where); |
|
775 return; |
|
776 } |
|
777 prev = node; |
|
778 } |
|
779 prev->set_next(new (this) unknown_compiledNode(cb, where)); |
|
780 } |
|
781 } |
|
782 |
|
783 void ThreadProfiler::vm_update(TickPosition where) { |
|
784 vm_update(NULL, where); |
|
785 } |
|
786 |
|
787 void ThreadProfiler::vm_update(const char* name, TickPosition where) { |
|
788 int index = entry(vmNode::hash(name)); |
|
789 assert(index >= 0, "Must be positive"); |
|
790 // Note that we call strdup below since the symbol may be resource allocated |
|
791 if (!table[index]) { |
|
792 table[index] = new (this) vmNode(name, where); |
|
793 } else { |
|
794 ProfilerNode* prev = table[index]; |
|
795 for(ProfilerNode* node = prev; node; node = node->next()) { |
|
796 if (((vmNode *)node)->vm_match(name)) { |
|
797 node->update(where); |
|
798 return; |
|
799 } |
|
800 prev = node; |
|
801 } |
|
802 prev->set_next(new (this) vmNode(name, where)); |
|
803 } |
|
804 } |
|
805 |
|
806 |
|
807 class FlatProfilerTask : public PeriodicTask { |
|
808 public: |
|
809 FlatProfilerTask(int interval_time) : PeriodicTask(interval_time) {} |
|
810 void task(); |
|
811 }; |
|
812 |
|
813 void FlatProfiler::record_vm_operation() { |
|
814 if (Universe::heap()->is_gc_active()) { |
|
815 FlatProfiler::received_gc_ticks += 1; |
|
816 return; |
|
817 } |
|
818 |
|
819 if (DeoptimizationMarker::is_active()) { |
|
820 FlatProfiler::deopt_ticks += 1; |
|
821 return; |
|
822 } |
|
823 |
|
824 FlatProfiler::vm_operation_ticks += 1; |
|
825 } |
|
826 |
|
827 void FlatProfiler::record_vm_tick() { |
|
828 // Profile the VM Thread itself if needed |
|
829 // This is done without getting the Threads_lock and we can go deep |
|
830 // inside Safepoint, etc. |
|
831 if( ProfileVM ) { |
|
832 ResourceMark rm; |
|
833 ExtendedPC epc; |
|
834 const char *name = NULL; |
|
835 char buf[256]; |
|
836 buf[0] = '\0'; |
|
837 |
|
838 vm_thread_profiler->inc_thread_ticks(); |
|
839 |
|
840 // Get a snapshot of a current VMThread pc (and leave it running!) |
|
841 // The call may fail in some circumstances |
|
842 epc = os::get_thread_pc(VMThread::vm_thread()); |
|
843 if(epc.pc() != NULL) { |
|
844 if (os::dll_address_to_function_name(epc.pc(), buf, sizeof(buf), NULL)) { |
|
845 name = buf; |
|
846 } |
|
847 } |
|
848 if (name != NULL) { |
|
849 vm_thread_profiler->vm_update(name, tp_native); |
|
850 } |
|
851 } |
|
852 } |
|
853 |
|
854 void FlatProfiler::record_thread_ticks() { |
|
855 |
|
856 int maxthreads, suspendedthreadcount; |
|
857 JavaThread** threadsList; |
|
858 bool interval_expired = false; |
|
859 |
|
860 if (ProfileIntervals && |
|
861 (FlatProfiler::received_ticks >= interval_ticks_previous + ProfileIntervalsTicks)) { |
|
862 interval_expired = true; |
|
863 interval_ticks_previous = FlatProfiler::received_ticks; |
|
864 } |
|
865 |
|
866 // Try not to wait for the Threads_lock |
|
867 if (Threads_lock->try_lock()) { |
|
868 { // Threads_lock scope |
|
869 maxthreads = Threads::number_of_threads(); |
|
870 threadsList = NEW_C_HEAP_ARRAY(JavaThread *, maxthreads, mtInternal); |
|
871 suspendedthreadcount = 0; |
|
872 for (JavaThread* tp = Threads::first(); tp != NULL; tp = tp->next()) { |
|
873 if (tp->is_Compiler_thread()) { |
|
874 // Only record ticks for active compiler threads |
|
875 CompilerThread* cthread = (CompilerThread*)tp; |
|
876 if (cthread->task() != NULL) { |
|
877 // The compiler is active. If we need to access any of the fields |
|
878 // of the compiler task we should suspend the CompilerThread first. |
|
879 FlatProfiler::compiler_ticks += 1; |
|
880 continue; |
|
881 } |
|
882 } |
|
883 |
|
884 // First externally suspend all threads by marking each for |
|
885 // external suspension - so it will stop at its next transition |
|
886 // Then do a safepoint |
|
887 ThreadProfiler* pp = tp->get_thread_profiler(); |
|
888 if (pp != NULL && pp->engaged) { |
|
889 MutexLockerEx ml(tp->SR_lock(), Mutex::_no_safepoint_check_flag); |
|
890 if (!tp->is_external_suspend() && !tp->is_exiting()) { |
|
891 tp->set_external_suspend(); |
|
892 threadsList[suspendedthreadcount++] = tp; |
|
893 } |
|
894 } |
|
895 } |
|
896 Threads_lock->unlock(); |
|
897 } |
|
898 // Suspend each thread. This call should just return |
|
899 // for any threads that have already self-suspended |
|
900 // Net result should be one safepoint |
|
901 for (int j = 0; j < suspendedthreadcount; j++) { |
|
902 JavaThread *tp = threadsList[j]; |
|
903 if (tp) { |
|
904 tp->java_suspend(); |
|
905 } |
|
906 } |
|
907 |
|
908 // We are responsible for resuming any thread on this list |
|
909 for (int i = 0; i < suspendedthreadcount; i++) { |
|
910 JavaThread *tp = threadsList[i]; |
|
911 if (tp) { |
|
912 ThreadProfiler* pp = tp->get_thread_profiler(); |
|
913 if (pp != NULL && pp->engaged) { |
|
914 HandleMark hm; |
|
915 FlatProfiler::delivered_ticks += 1; |
|
916 if (interval_expired) { |
|
917 FlatProfiler::interval_record_thread(pp); |
|
918 } |
|
919 // This is the place where we check to see if a user thread is |
|
920 // blocked waiting for compilation. |
|
921 if (tp->blocked_on_compilation()) { |
|
922 pp->compiler_ticks += 1; |
|
923 pp->interval_data_ref()->inc_compiling(); |
|
924 } else { |
|
925 pp->record_tick(tp); |
|
926 } |
|
927 } |
|
928 MutexLocker ml(Threads_lock); |
|
929 tp->java_resume(); |
|
930 } |
|
931 } |
|
932 if (interval_expired) { |
|
933 FlatProfiler::interval_print(); |
|
934 FlatProfiler::interval_reset(); |
|
935 } |
|
936 |
|
937 FREE_C_HEAP_ARRAY(JavaThread *, threadsList); |
|
938 } else { |
|
939 // Couldn't get the threads lock, just record that rather than blocking |
|
940 FlatProfiler::threads_lock_ticks += 1; |
|
941 } |
|
942 |
|
943 } |
|
944 |
|
945 void FlatProfilerTask::task() { |
|
946 FlatProfiler::received_ticks += 1; |
|
947 |
|
948 if (ProfileVM) { |
|
949 FlatProfiler::record_vm_tick(); |
|
950 } |
|
951 |
|
952 VM_Operation* op = VMThread::vm_operation(); |
|
953 if (op != NULL) { |
|
954 FlatProfiler::record_vm_operation(); |
|
955 if (SafepointSynchronize::is_at_safepoint()) { |
|
956 return; |
|
957 } |
|
958 } |
|
959 FlatProfiler::record_thread_ticks(); |
|
960 } |
|
961 |
|
962 void ThreadProfiler::record_interpreted_tick(JavaThread* thread, frame fr, TickPosition where, int* ticks) { |
|
963 FlatProfiler::all_int_ticks++; |
|
964 if (!FlatProfiler::full_profile()) { |
|
965 return; |
|
966 } |
|
967 |
|
968 if (!fr.is_interpreted_frame_valid(thread)) { |
|
969 // tick came at a bad time |
|
970 interpreter_ticks += 1; |
|
971 FlatProfiler::interpreter_ticks += 1; |
|
972 return; |
|
973 } |
|
974 |
|
975 // The frame has been fully validated so we can trust the method and bci |
|
976 |
|
977 Method* method = *fr.interpreter_frame_method_addr(); |
|
978 |
|
979 interpreted_update(method, where); |
|
980 |
|
981 // update byte code table |
|
982 InterpreterCodelet* desc = Interpreter::codelet_containing(fr.pc()); |
|
983 if (desc != NULL && desc->bytecode() >= 0) { |
|
984 ticks[desc->bytecode()]++; |
|
985 } |
|
986 } |
|
987 |
|
988 void ThreadProfiler::record_compiled_tick(JavaThread* thread, frame fr, TickPosition where) { |
|
989 const char *name = NULL; |
|
990 TickPosition localwhere = where; |
|
991 |
|
992 FlatProfiler::all_comp_ticks++; |
|
993 if (!FlatProfiler::full_profile()) return; |
|
994 |
|
995 CodeBlob* cb = fr.cb(); |
|
996 |
|
997 // For runtime stubs, record as native rather than as compiled |
|
998 if (cb->is_runtime_stub()) { |
|
999 RegisterMap map(thread, false); |
|
1000 fr = fr.sender(&map); |
|
1001 cb = fr.cb(); |
|
1002 localwhere = tp_native; |
|
1003 } |
|
1004 |
|
1005 Method* method = cb->is_compiled() ? cb->as_compiled_method()->method() : (Method*) NULL; |
|
1006 if (method == NULL) { |
|
1007 if (cb->is_runtime_stub()) |
|
1008 runtime_stub_update(cb, name, localwhere); |
|
1009 else |
|
1010 unknown_compiled_update(cb, localwhere); |
|
1011 } |
|
1012 else { |
|
1013 if (method->is_native()) { |
|
1014 stub_update(method, name, localwhere); |
|
1015 } else { |
|
1016 compiled_update(method, localwhere); |
|
1017 } |
|
1018 } |
|
1019 } |
|
1020 |
|
1021 extern "C" void find(int x); |
|
1022 |
|
1023 |
|
1024 void ThreadProfiler::record_tick_for_running_frame(JavaThread* thread, frame fr) { |
|
1025 // The tick happened in real code -> non VM code |
|
1026 if (fr.is_interpreted_frame()) { |
|
1027 interval_data_ref()->inc_interpreted(); |
|
1028 record_interpreted_tick(thread, fr, tp_code, FlatProfiler::bytecode_ticks); |
|
1029 return; |
|
1030 } |
|
1031 |
|
1032 if (CodeCache::contains(fr.pc())) { |
|
1033 interval_data_ref()->inc_compiled(); |
|
1034 PCRecorder::record(fr.pc()); |
|
1035 record_compiled_tick(thread, fr, tp_code); |
|
1036 return; |
|
1037 } |
|
1038 |
|
1039 if (VtableStubs::stub_containing(fr.pc()) != NULL) { |
|
1040 unknown_ticks_array[ut_vtable_stubs] += 1; |
|
1041 return; |
|
1042 } |
|
1043 |
|
1044 frame caller = fr.profile_find_Java_sender_frame(thread); |
|
1045 |
|
1046 if (caller.sp() != NULL && caller.pc() != NULL) { |
|
1047 record_tick_for_calling_frame(thread, caller); |
|
1048 return; |
|
1049 } |
|
1050 |
|
1051 unknown_ticks_array[ut_running_frame] += 1; |
|
1052 FlatProfiler::unknown_ticks += 1; |
|
1053 } |
|
1054 |
|
1055 void ThreadProfiler::record_tick_for_calling_frame(JavaThread* thread, frame fr) { |
|
1056 // The tick happened in VM code |
|
1057 interval_data_ref()->inc_native(); |
|
1058 if (fr.is_interpreted_frame()) { |
|
1059 record_interpreted_tick(thread, fr, tp_native, FlatProfiler::bytecode_ticks_stub); |
|
1060 return; |
|
1061 } |
|
1062 if (CodeCache::contains(fr.pc())) { |
|
1063 record_compiled_tick(thread, fr, tp_native); |
|
1064 return; |
|
1065 } |
|
1066 |
|
1067 frame caller = fr.profile_find_Java_sender_frame(thread); |
|
1068 |
|
1069 if (caller.sp() != NULL && caller.pc() != NULL) { |
|
1070 record_tick_for_calling_frame(thread, caller); |
|
1071 return; |
|
1072 } |
|
1073 |
|
1074 unknown_ticks_array[ut_calling_frame] += 1; |
|
1075 FlatProfiler::unknown_ticks += 1; |
|
1076 } |
|
1077 |
|
1078 void ThreadProfiler::record_tick(JavaThread* thread) { |
|
1079 FlatProfiler::all_ticks++; |
|
1080 thread_ticks += 1; |
|
1081 |
|
1082 // Here's another way to track global state changes. |
|
1083 // When the class loader starts it marks the ThreadProfiler to tell it it is in the class loader |
|
1084 // and we check that here. |
|
1085 // This is more direct, and more than one thread can be in the class loader at a time, |
|
1086 // but it does mean the class loader has to know about the profiler. |
|
1087 if (region_flag[ThreadProfilerMark::classLoaderRegion]) { |
|
1088 class_loader_ticks += 1; |
|
1089 FlatProfiler::class_loader_ticks += 1; |
|
1090 return; |
|
1091 } else if (region_flag[ThreadProfilerMark::extraRegion]) { |
|
1092 extra_ticks += 1; |
|
1093 FlatProfiler::extra_ticks += 1; |
|
1094 return; |
|
1095 } |
|
1096 // Note that the WatcherThread can now stop for safepoints |
|
1097 uint32_t debug_bits = 0; |
|
1098 if (!thread->wait_for_ext_suspend_completion(SuspendRetryCount, |
|
1099 SuspendRetryDelay, &debug_bits)) { |
|
1100 unknown_ticks_array[ut_unknown_thread_state] += 1; |
|
1101 FlatProfiler::unknown_ticks += 1; |
|
1102 return; |
|
1103 } |
|
1104 |
|
1105 frame fr; |
|
1106 |
|
1107 switch (thread->thread_state()) { |
|
1108 case _thread_in_native: |
|
1109 case _thread_in_native_trans: |
|
1110 case _thread_in_vm: |
|
1111 case _thread_in_vm_trans: |
|
1112 if (thread->profile_last_Java_frame(&fr)) { |
|
1113 if (fr.is_runtime_frame()) { |
|
1114 RegisterMap map(thread, false); |
|
1115 fr = fr.sender(&map); |
|
1116 } |
|
1117 record_tick_for_calling_frame(thread, fr); |
|
1118 } else { |
|
1119 unknown_ticks_array[ut_no_last_Java_frame] += 1; |
|
1120 FlatProfiler::unknown_ticks += 1; |
|
1121 } |
|
1122 break; |
|
1123 // handle_special_runtime_exit_condition self-suspends threads in Java |
|
1124 case _thread_in_Java: |
|
1125 case _thread_in_Java_trans: |
|
1126 if (thread->profile_last_Java_frame(&fr)) { |
|
1127 if (fr.is_safepoint_blob_frame()) { |
|
1128 RegisterMap map(thread, false); |
|
1129 fr = fr.sender(&map); |
|
1130 } |
|
1131 record_tick_for_running_frame(thread, fr); |
|
1132 } else { |
|
1133 unknown_ticks_array[ut_no_last_Java_frame] += 1; |
|
1134 FlatProfiler::unknown_ticks += 1; |
|
1135 } |
|
1136 break; |
|
1137 case _thread_blocked: |
|
1138 case _thread_blocked_trans: |
|
1139 if (thread->osthread() && thread->osthread()->get_state() == RUNNABLE) { |
|
1140 if (thread->profile_last_Java_frame(&fr)) { |
|
1141 if (fr.is_safepoint_blob_frame()) { |
|
1142 RegisterMap map(thread, false); |
|
1143 fr = fr.sender(&map); |
|
1144 record_tick_for_running_frame(thread, fr); |
|
1145 } else { |
|
1146 record_tick_for_calling_frame(thread, fr); |
|
1147 } |
|
1148 } else { |
|
1149 unknown_ticks_array[ut_no_last_Java_frame] += 1; |
|
1150 FlatProfiler::unknown_ticks += 1; |
|
1151 } |
|
1152 } else { |
|
1153 blocked_ticks += 1; |
|
1154 FlatProfiler::blocked_ticks += 1; |
|
1155 } |
|
1156 break; |
|
1157 case _thread_uninitialized: |
|
1158 case _thread_new: |
|
1159 // not used, included for completeness |
|
1160 case _thread_new_trans: |
|
1161 unknown_ticks_array[ut_no_last_Java_frame] += 1; |
|
1162 FlatProfiler::unknown_ticks += 1; |
|
1163 break; |
|
1164 default: |
|
1165 unknown_ticks_array[ut_unknown_thread_state] += 1; |
|
1166 FlatProfiler::unknown_ticks += 1; |
|
1167 break; |
|
1168 } |
|
1169 return; |
|
1170 } |
|
1171 |
|
1172 void ThreadProfiler::engage() { |
|
1173 engaged = true; |
|
1174 timer.start(); |
|
1175 } |
|
1176 |
|
1177 void ThreadProfiler::disengage() { |
|
1178 engaged = false; |
|
1179 timer.stop(); |
|
1180 } |
|
1181 |
|
1182 void ThreadProfiler::initialize() { |
|
1183 for (int index = 0; index < table_size; index++) { |
|
1184 table[index] = NULL; |
|
1185 } |
|
1186 thread_ticks = 0; |
|
1187 blocked_ticks = 0; |
|
1188 compiler_ticks = 0; |
|
1189 interpreter_ticks = 0; |
|
1190 for (int ut = 0; ut < ut_end; ut += 1) { |
|
1191 unknown_ticks_array[ut] = 0; |
|
1192 } |
|
1193 region_flag[ThreadProfilerMark::classLoaderRegion] = false; |
|
1194 class_loader_ticks = 0; |
|
1195 region_flag[ThreadProfilerMark::extraRegion] = false; |
|
1196 extra_ticks = 0; |
|
1197 timer.start(); |
|
1198 interval_data_ref()->reset(); |
|
1199 } |
|
1200 |
|
1201 void ThreadProfiler::reset() { |
|
1202 timer.stop(); |
|
1203 if (table != NULL) { |
|
1204 for (int index = 0; index < table_size; index++) { |
|
1205 ProfilerNode* n = table[index]; |
|
1206 if (n != NULL) { |
|
1207 delete n; |
|
1208 } |
|
1209 } |
|
1210 } |
|
1211 initialize(); |
|
1212 } |
|
1213 |
|
1214 void FlatProfiler::allocate_table() { |
|
1215 { // Bytecode table |
|
1216 bytecode_ticks = NEW_C_HEAP_ARRAY(int, Bytecodes::number_of_codes, mtInternal); |
|
1217 bytecode_ticks_stub = NEW_C_HEAP_ARRAY(int, Bytecodes::number_of_codes, mtInternal); |
|
1218 for(int index = 0; index < Bytecodes::number_of_codes; index++) { |
|
1219 bytecode_ticks[index] = 0; |
|
1220 bytecode_ticks_stub[index] = 0; |
|
1221 } |
|
1222 } |
|
1223 |
|
1224 if (ProfilerRecordPC) PCRecorder::init(); |
|
1225 |
|
1226 interval_data = NEW_C_HEAP_ARRAY(IntervalData, interval_print_size, mtInternal); |
|
1227 FlatProfiler::interval_reset(); |
|
1228 } |
|
1229 |
|
1230 void FlatProfiler::engage(JavaThread* mainThread, bool fullProfile) { |
|
1231 full_profile_flag = fullProfile; |
|
1232 if (bytecode_ticks == NULL) { |
|
1233 allocate_table(); |
|
1234 } |
|
1235 if(ProfileVM && (vm_thread_profiler == NULL)){ |
|
1236 vm_thread_profiler = new ThreadProfiler(); |
|
1237 } |
|
1238 if (task == NULL) { |
|
1239 task = new FlatProfilerTask(WatcherThread::delay_interval); |
|
1240 task->enroll(); |
|
1241 } |
|
1242 timer.start(); |
|
1243 if (mainThread != NULL) { |
|
1244 // When mainThread was created, it might not have a ThreadProfiler |
|
1245 ThreadProfiler* pp = mainThread->get_thread_profiler(); |
|
1246 if (pp == NULL) { |
|
1247 mainThread->set_thread_profiler(new ThreadProfiler()); |
|
1248 } else { |
|
1249 pp->reset(); |
|
1250 } |
|
1251 mainThread->get_thread_profiler()->engage(); |
|
1252 } |
|
1253 // This is where we would assign thread_profiler |
|
1254 // if we wanted only one thread_profiler for all threads. |
|
1255 thread_profiler = NULL; |
|
1256 } |
|
1257 |
|
1258 void FlatProfiler::disengage() { |
|
1259 if (!task) { |
|
1260 return; |
|
1261 } |
|
1262 timer.stop(); |
|
1263 task->disenroll(); |
|
1264 delete task; |
|
1265 task = NULL; |
|
1266 if (thread_profiler != NULL) { |
|
1267 thread_profiler->disengage(); |
|
1268 } else { |
|
1269 MutexLocker tl(Threads_lock); |
|
1270 for (JavaThread* tp = Threads::first(); tp != NULL; tp = tp->next()) { |
|
1271 ThreadProfiler* pp = tp->get_thread_profiler(); |
|
1272 if (pp != NULL) { |
|
1273 pp->disengage(); |
|
1274 } |
|
1275 } |
|
1276 } |
|
1277 } |
|
1278 |
|
1279 void FlatProfiler::reset() { |
|
1280 if (task) { |
|
1281 disengage(); |
|
1282 } |
|
1283 |
|
1284 class_loader_ticks = 0; |
|
1285 extra_ticks = 0; |
|
1286 received_gc_ticks = 0; |
|
1287 vm_operation_ticks = 0; |
|
1288 compiler_ticks = 0; |
|
1289 deopt_ticks = 0; |
|
1290 interpreter_ticks = 0; |
|
1291 blocked_ticks = 0; |
|
1292 unknown_ticks = 0; |
|
1293 received_ticks = 0; |
|
1294 delivered_ticks = 0; |
|
1295 timer.stop(); |
|
1296 } |
|
1297 |
|
1298 bool FlatProfiler::is_active() { |
|
1299 return task != NULL; |
|
1300 } |
|
1301 |
|
1302 void FlatProfiler::print_byte_code_statistics() { |
|
1303 GrowableArray <ProfilerNode*>* array = new GrowableArray<ProfilerNode*>(200); |
|
1304 |
|
1305 tty->print_cr(" Bytecode ticks:"); |
|
1306 for (int index = 0; index < Bytecodes::number_of_codes; index++) { |
|
1307 if (FlatProfiler::bytecode_ticks[index] > 0 || FlatProfiler::bytecode_ticks_stub[index] > 0) { |
|
1308 tty->print_cr(" %4d %4d = %s", |
|
1309 FlatProfiler::bytecode_ticks[index], |
|
1310 FlatProfiler::bytecode_ticks_stub[index], |
|
1311 Bytecodes::name( (Bytecodes::Code) index)); |
|
1312 } |
|
1313 } |
|
1314 tty->cr(); |
|
1315 } |
|
1316 |
|
1317 void print_ticks(const char* title, int ticks, int total) { |
|
1318 if (ticks > 0) { |
|
1319 tty->print("%5.1f%% %5d", ticks * 100.0 / total, ticks); |
|
1320 tty->fill_to(col3); |
|
1321 tty->print("%s", title); |
|
1322 tty->cr(); |
|
1323 } |
|
1324 } |
|
1325 |
|
1326 void ThreadProfiler::print(const char* thread_name) { |
|
1327 ResourceMark rm; |
|
1328 MutexLocker ppl(ProfilePrint_lock); |
|
1329 int index = 0; // Declared outside for loops for portability |
|
1330 |
|
1331 if (table == NULL) { |
|
1332 return; |
|
1333 } |
|
1334 |
|
1335 if (thread_ticks <= 0) { |
|
1336 return; |
|
1337 } |
|
1338 |
|
1339 const char* title = "too soon to tell"; |
|
1340 double secs = timer.seconds(); |
|
1341 |
|
1342 GrowableArray <ProfilerNode*>* array = new GrowableArray<ProfilerNode*>(200); |
|
1343 for(index = 0; index < table_size; index++) { |
|
1344 for(ProfilerNode* node = table[index]; node; node = node->next()) |
|
1345 array->append(node); |
|
1346 } |
|
1347 |
|
1348 array->sort(&ProfilerNode::compare); |
|
1349 |
|
1350 // compute total (sanity check) |
|
1351 int active = |
|
1352 class_loader_ticks + |
|
1353 compiler_ticks + |
|
1354 interpreter_ticks + |
|
1355 unknown_ticks(); |
|
1356 for (index = 0; index < array->length(); index++) { |
|
1357 active += array->at(index)->ticks.total(); |
|
1358 } |
|
1359 int total = active + blocked_ticks; |
|
1360 |
|
1361 tty->cr(); |
|
1362 tty->print_cr("Flat profile of %3.2f secs (%d total ticks): %s", secs, total, thread_name); |
|
1363 if (total != thread_ticks) { |
|
1364 print_ticks("Lost ticks", thread_ticks-total, thread_ticks); |
|
1365 } |
|
1366 tty->cr(); |
|
1367 |
|
1368 // print interpreted methods |
|
1369 tick_counter interpreted_ticks; |
|
1370 bool has_interpreted_ticks = false; |
|
1371 int print_count = 0; |
|
1372 for (index = 0; index < array->length(); index++) { |
|
1373 ProfilerNode* n = array->at(index); |
|
1374 if (n->is_interpreted()) { |
|
1375 interpreted_ticks.add(&n->ticks); |
|
1376 if (!has_interpreted_ticks) { |
|
1377 interpretedNode::print_title(tty); |
|
1378 has_interpreted_ticks = true; |
|
1379 } |
|
1380 if (print_count++ < ProfilerNumberOfInterpretedMethods) { |
|
1381 n->print(tty, active); |
|
1382 } |
|
1383 } |
|
1384 } |
|
1385 if (has_interpreted_ticks) { |
|
1386 if (print_count <= ProfilerNumberOfInterpretedMethods) { |
|
1387 title = "Total interpreted"; |
|
1388 } else { |
|
1389 title = "Total interpreted (including elided)"; |
|
1390 } |
|
1391 interpretedNode::print_total(tty, &interpreted_ticks, active, title); |
|
1392 tty->cr(); |
|
1393 } |
|
1394 |
|
1395 // print compiled methods |
|
1396 tick_counter compiled_ticks; |
|
1397 bool has_compiled_ticks = false; |
|
1398 print_count = 0; |
|
1399 for (index = 0; index < array->length(); index++) { |
|
1400 ProfilerNode* n = array->at(index); |
|
1401 if (n->is_compiled()) { |
|
1402 compiled_ticks.add(&n->ticks); |
|
1403 if (!has_compiled_ticks) { |
|
1404 compiledNode::print_title(tty); |
|
1405 has_compiled_ticks = true; |
|
1406 } |
|
1407 if (print_count++ < ProfilerNumberOfCompiledMethods) { |
|
1408 n->print(tty, active); |
|
1409 } |
|
1410 } |
|
1411 } |
|
1412 if (has_compiled_ticks) { |
|
1413 if (print_count <= ProfilerNumberOfCompiledMethods) { |
|
1414 title = "Total compiled"; |
|
1415 } else { |
|
1416 title = "Total compiled (including elided)"; |
|
1417 } |
|
1418 compiledNode::print_total(tty, &compiled_ticks, active, title); |
|
1419 tty->cr(); |
|
1420 } |
|
1421 |
|
1422 // print stub methods |
|
1423 tick_counter stub_ticks; |
|
1424 bool has_stub_ticks = false; |
|
1425 print_count = 0; |
|
1426 for (index = 0; index < array->length(); index++) { |
|
1427 ProfilerNode* n = array->at(index); |
|
1428 if (n->is_stub()) { |
|
1429 stub_ticks.add(&n->ticks); |
|
1430 if (!has_stub_ticks) { |
|
1431 stubNode::print_title(tty); |
|
1432 has_stub_ticks = true; |
|
1433 } |
|
1434 if (print_count++ < ProfilerNumberOfStubMethods) { |
|
1435 n->print(tty, active); |
|
1436 } |
|
1437 } |
|
1438 } |
|
1439 if (has_stub_ticks) { |
|
1440 if (print_count <= ProfilerNumberOfStubMethods) { |
|
1441 title = "Total stub"; |
|
1442 } else { |
|
1443 title = "Total stub (including elided)"; |
|
1444 } |
|
1445 stubNode::print_total(tty, &stub_ticks, active, title); |
|
1446 tty->cr(); |
|
1447 } |
|
1448 |
|
1449 // print runtime stubs |
|
1450 tick_counter runtime_stub_ticks; |
|
1451 bool has_runtime_stub_ticks = false; |
|
1452 print_count = 0; |
|
1453 for (index = 0; index < array->length(); index++) { |
|
1454 ProfilerNode* n = array->at(index); |
|
1455 if (n->is_runtime_stub()) { |
|
1456 runtime_stub_ticks.add(&n->ticks); |
|
1457 if (!has_runtime_stub_ticks) { |
|
1458 runtimeStubNode::print_title(tty); |
|
1459 has_runtime_stub_ticks = true; |
|
1460 } |
|
1461 if (print_count++ < ProfilerNumberOfRuntimeStubNodes) { |
|
1462 n->print(tty, active); |
|
1463 } |
|
1464 } |
|
1465 } |
|
1466 if (has_runtime_stub_ticks) { |
|
1467 if (print_count <= ProfilerNumberOfRuntimeStubNodes) { |
|
1468 title = "Total runtime stubs"; |
|
1469 } else { |
|
1470 title = "Total runtime stubs (including elided)"; |
|
1471 } |
|
1472 runtimeStubNode::print_total(tty, &runtime_stub_ticks, active, title); |
|
1473 tty->cr(); |
|
1474 } |
|
1475 |
|
1476 if (blocked_ticks + class_loader_ticks + interpreter_ticks + compiler_ticks + unknown_ticks() != 0) { |
|
1477 tty->fill_to(col1); |
|
1478 tty->print_cr("Thread-local ticks:"); |
|
1479 print_ticks("Blocked (of total)", blocked_ticks, total); |
|
1480 print_ticks("Class loader", class_loader_ticks, active); |
|
1481 print_ticks("Extra", extra_ticks, active); |
|
1482 print_ticks("Interpreter", interpreter_ticks, active); |
|
1483 print_ticks("Compilation", compiler_ticks, active); |
|
1484 print_ticks("Unknown: vtable stubs", unknown_ticks_array[ut_vtable_stubs], active); |
|
1485 print_ticks("Unknown: null method", unknown_ticks_array[ut_null_method], active); |
|
1486 print_ticks("Unknown: running frame", unknown_ticks_array[ut_running_frame], active); |
|
1487 print_ticks("Unknown: calling frame", unknown_ticks_array[ut_calling_frame], active); |
|
1488 print_ticks("Unknown: no pc", unknown_ticks_array[ut_no_pc], active); |
|
1489 print_ticks("Unknown: no last frame", unknown_ticks_array[ut_no_last_Java_frame], active); |
|
1490 print_ticks("Unknown: thread_state", unknown_ticks_array[ut_unknown_thread_state], active); |
|
1491 tty->cr(); |
|
1492 } |
|
1493 |
|
1494 if (WizardMode) { |
|
1495 tty->print_cr("Node area used: " INTX_FORMAT " Kb", (area_top - area_bottom) / 1024); |
|
1496 } |
|
1497 reset(); |
|
1498 } |
|
1499 |
|
1500 /* |
|
1501 ThreadProfiler::print_unknown(){ |
|
1502 if (table == NULL) { |
|
1503 return; |
|
1504 } |
|
1505 |
|
1506 if (thread_ticks <= 0) { |
|
1507 return; |
|
1508 } |
|
1509 } */ |
|
1510 |
|
1511 void FlatProfiler::print(int unused) { |
|
1512 ResourceMark rm; |
|
1513 if (thread_profiler != NULL) { |
|
1514 thread_profiler->print("All threads"); |
|
1515 } else { |
|
1516 MutexLocker tl(Threads_lock); |
|
1517 for (JavaThread* tp = Threads::first(); tp != NULL; tp = tp->next()) { |
|
1518 ThreadProfiler* pp = tp->get_thread_profiler(); |
|
1519 if (pp != NULL) { |
|
1520 pp->print(tp->get_thread_name()); |
|
1521 } |
|
1522 } |
|
1523 } |
|
1524 |
|
1525 if (ProfilerPrintByteCodeStatistics) { |
|
1526 print_byte_code_statistics(); |
|
1527 } |
|
1528 |
|
1529 if (non_method_ticks() > 0) { |
|
1530 tty->cr(); |
|
1531 tty->print_cr("Global summary of %3.2f seconds:", timer.seconds()); |
|
1532 print_ticks("Received ticks", received_ticks, received_ticks); |
|
1533 print_ticks("Received GC ticks", received_gc_ticks, received_ticks); |
|
1534 print_ticks("Compilation", compiler_ticks, received_ticks); |
|
1535 print_ticks("Deoptimization", deopt_ticks, received_ticks); |
|
1536 print_ticks("Other VM operations", vm_operation_ticks, received_ticks); |
|
1537 #ifndef PRODUCT |
|
1538 print_ticks("Blocked ticks", blocked_ticks, received_ticks); |
|
1539 print_ticks("Threads_lock blocks", threads_lock_ticks, received_ticks); |
|
1540 print_ticks("Delivered ticks", delivered_ticks, received_ticks); |
|
1541 print_ticks("All ticks", all_ticks, received_ticks); |
|
1542 #endif |
|
1543 print_ticks("Class loader", class_loader_ticks, received_ticks); |
|
1544 print_ticks("Extra ", extra_ticks, received_ticks); |
|
1545 print_ticks("Interpreter", interpreter_ticks, received_ticks); |
|
1546 print_ticks("Unknown code", unknown_ticks, received_ticks); |
|
1547 } |
|
1548 |
|
1549 PCRecorder::print(); |
|
1550 |
|
1551 if(ProfileVM){ |
|
1552 tty->cr(); |
|
1553 vm_thread_profiler->print("VM Thread"); |
|
1554 } |
|
1555 } |
|
1556 |
|
1557 void IntervalData::print_header(outputStream* st) { |
|
1558 st->print("i/c/n/g"); |
|
1559 } |
|
1560 |
|
1561 void IntervalData::print_data(outputStream* st) { |
|
1562 st->print("%d/%d/%d/%d", interpreted(), compiled(), native(), compiling()); |
|
1563 } |
|
1564 |
|
1565 void FlatProfiler::interval_record_thread(ThreadProfiler* tp) { |
|
1566 IntervalData id = tp->interval_data(); |
|
1567 int total = id.total(); |
|
1568 tp->interval_data_ref()->reset(); |
|
1569 |
|
1570 // Insertion sort the data, if it's relevant. |
|
1571 for (int i = 0; i < interval_print_size; i += 1) { |
|
1572 if (total > interval_data[i].total()) { |
|
1573 for (int j = interval_print_size - 1; j > i; j -= 1) { |
|
1574 interval_data[j] = interval_data[j-1]; |
|
1575 } |
|
1576 interval_data[i] = id; |
|
1577 break; |
|
1578 } |
|
1579 } |
|
1580 } |
|
1581 |
|
1582 void FlatProfiler::interval_print() { |
|
1583 if ((interval_data[0].total() > 0)) { |
|
1584 tty->stamp(); |
|
1585 tty->print("\t"); |
|
1586 IntervalData::print_header(tty); |
|
1587 for (int i = 0; i < interval_print_size; i += 1) { |
|
1588 if (interval_data[i].total() > 0) { |
|
1589 tty->print("\t"); |
|
1590 interval_data[i].print_data(tty); |
|
1591 } |
|
1592 } |
|
1593 tty->cr(); |
|
1594 } |
|
1595 } |
|
1596 |
|
1597 void FlatProfiler::interval_reset() { |
|
1598 for (int i = 0; i < interval_print_size; i += 1) { |
|
1599 interval_data[i].reset(); |
|
1600 } |
|
1601 } |
|
1602 |
|
1603 void ThreadProfiler::oops_do(OopClosure* f) { |
|
1604 if (table == NULL) return; |
|
1605 |
|
1606 for(int index = 0; index < table_size; index++) { |
|
1607 for(ProfilerNode* node = table[index]; node; node = node->next()) |
|
1608 node->oops_do(f); |
|
1609 } |
|
1610 } |
|
1611 |
|
1612 void FlatProfiler::oops_do(OopClosure* f) { |
|
1613 if (thread_profiler != NULL) { |
|
1614 thread_profiler->oops_do(f); |
|
1615 } else { |
|
1616 for (JavaThread* tp = Threads::first(); tp != NULL; tp = tp->next()) { |
|
1617 ThreadProfiler* pp = tp->get_thread_profiler(); |
|
1618 if (pp != NULL) { |
|
1619 pp->oops_do(f); |
|
1620 } |
|
1621 } |
|
1622 } |
|
1623 } |
|