1 /* 2 * Copyright (c) 2003, 2012, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "code/debugInfoRec.hpp" 27 #include "code/pcDesc.hpp" 28 #include "gc_interface/collectedHeap.inline.hpp" 29 #include "memory/space.hpp" 30 #include "memory/universe.inline.hpp" 31 #include "oops/oop.inline.hpp" 32 #include "oops/oop.inline2.hpp" 33 #include "prims/forte.hpp" 34 #include "runtime/thread.hpp" 35 #include "runtime/vframe.hpp" 36 #include "runtime/vframeArray.hpp" 37 38 // call frame copied from old .h file and renamed 39 typedef struct { 40 jint lineno; // line number in the source file 41 jmethodID method_id; // method executed in this frame 42 } ASGCT_CallFrame; 43 44 // call trace copied from old .h file and renamed 45 typedef struct { 46 JNIEnv *env_id; // Env where trace was recorded 47 jint num_frames; // number of frames in this trace 48 ASGCT_CallFrame *frames; // frames 49 } ASGCT_CallTrace; 50 51 // These name match the names reported by the forte quality kit 52 enum { 53 ticks_no_Java_frame = 0, 54 ticks_no_class_load = -1, 55 ticks_GC_active = -2, 56 ticks_unknown_not_Java = -3, 57 ticks_not_walkable_not_Java = -4, 58 ticks_unknown_Java = -5, 59 ticks_not_walkable_Java = -6, 60 ticks_unknown_state = -7, 61 ticks_thread_exit = -8, 62 ticks_deopt = -9, 63 ticks_safepoint = -10 64 }; 65 66 #if INCLUDE_JVMTI 67 68 //------------------------------------------------------- 69 70 // Native interfaces for use by Forte tools. 71 72 73 #ifndef IA64 74 75 class vframeStreamForte : public vframeStreamCommon { 76 public: 77 // constructor that starts with sender of frame fr (top_frame) 78 vframeStreamForte(JavaThread *jt, frame fr, bool stop_at_java_call_stub); 79 void forte_next(); 80 }; 81 82 83 static bool is_decipherable_compiled_frame(JavaThread* thread, frame* fr, nmethod* nm); 84 static bool is_decipherable_interpreted_frame(JavaThread* thread, 85 frame* fr, 86 Method** method_p, 87 int* bci_p); 88 89 90 91 92 vframeStreamForte::vframeStreamForte(JavaThread *jt, 93 frame fr, 94 bool stop_at_java_call_stub) : vframeStreamCommon(jt) { 95 96 _stop_at_java_call_stub = stop_at_java_call_stub; 97 _frame = fr; 98 99 // We must always have a valid frame to start filling 100 101 bool filled_in = fill_from_frame(); 102 103 assert(filled_in, "invariant"); 104 105 } 106 107 108 // Solaris SPARC Compiler1 needs an additional check on the grandparent 109 // of the top_frame when the parent of the top_frame is interpreted and 110 // the grandparent is compiled. However, in this method we do not know 111 // the relationship of the current _frame relative to the top_frame so 112 // we implement a more broad sanity check. When the previous callee is 113 // interpreted and the current sender is compiled, we verify that the 114 // current sender is also walkable. If it is not walkable, then we mark 115 // the current vframeStream as at the end. 116 void vframeStreamForte::forte_next() { 117 // handle frames with inlining 118 if (_mode == compiled_mode && 119 vframeStreamCommon::fill_in_compiled_inlined_sender()) { 120 return; 121 } 122 123 // handle general case 124 125 int loop_count = 0; 126 int loop_max = MaxJavaStackTraceDepth * 2; 127 128 129 do { 130 131 loop_count++; 132 133 // By the time we get here we should never see unsafe but better 134 // safe then segv'd 135 136 if (loop_count > loop_max || !_frame.safe_for_sender(_thread)) { 137 _mode = at_end_mode; 138 return; 139 } 140 141 _frame = _frame.sender(&_reg_map); 142 143 } while (!fill_from_frame()); 144 } 145 146 // Determine if 'fr' is a decipherable compiled frame. We are already 147 // assured that fr is for a java nmethod. 148 149 static bool is_decipherable_compiled_frame(JavaThread* thread, frame* fr, nmethod* nm) { 150 assert(nm->is_java_method(), "invariant"); 151 152 if (thread->has_last_Java_frame() && thread->last_Java_pc() == fr->pc()) { 153 // We're stopped at a call into the JVM so look for a PcDesc with 154 // the actual pc reported by the frame. 155 PcDesc* pc_desc = nm->pc_desc_at(fr->pc()); 156 157 // Did we find a useful PcDesc? 158 if (pc_desc != NULL && 159 pc_desc->scope_decode_offset() != DebugInformationRecorder::serialized_null) { 160 return true; 161 } 162 } 163 164 // We're at some random pc in the nmethod so search for the PcDesc 165 // whose pc is greater than the current PC. It's done this way 166 // because the extra PcDescs that are recorded for improved debug 167 // info record the end of the region covered by the ScopeDesc 168 // instead of the beginning. 169 PcDesc* pc_desc = nm->pc_desc_near(fr->pc() + 1); 170 171 // Now do we have a useful PcDesc? 172 if (pc_desc == NULL || 173 pc_desc->scope_decode_offset() == DebugInformationRecorder::serialized_null) { 174 // No debug information available for this pc 175 // vframeStream would explode if we try and walk the frames. 176 return false; 177 } 178 179 // This PcDesc is useful however we must adjust the frame's pc 180 // so that the vframeStream lookups will use this same pc 181 fr->set_pc(pc_desc->real_pc(nm)); 182 return true; 183 } 184 185 186 // Determine if 'fr' is a walkable interpreted frame. Returns false 187 // if it is not. *method_p, and *bci_p are not set when false is 188 // returned. *method_p is non-NULL if frame was executing a Java 189 // method. *bci_p is != -1 if a valid BCI in the Java method could 190 // be found. 191 // Note: this method returns true when a valid Java method is found 192 // even if a valid BCI cannot be found. 193 194 static bool is_decipherable_interpreted_frame(JavaThread* thread, 195 frame* fr, 196 Method** method_p, 197 int* bci_p) { 198 assert(fr->is_interpreted_frame(), "just checking"); 199 200 // top frame is an interpreted frame 201 // check if it is walkable (i.e. valid Method* and valid bci) 202 203 // Because we may be racing a gc thread the method and/or bci 204 // of a valid interpreter frame may look bad causing us to 205 // fail the is_interpreted_frame_valid test. If the thread 206 // is in any of the following states we are assured that the 207 // frame is in fact valid and we must have hit the race. 208 209 JavaThreadState state = thread->thread_state(); 210 bool known_valid = (state == _thread_in_native || 211 state == _thread_in_vm || 212 state == _thread_blocked ); 213 214 if (known_valid || fr->is_interpreted_frame_valid(thread)) { 215 216 // The frame code should completely validate the frame so that 217 // references to Method* and bci are completely safe to access 218 // If they aren't the frame code should be fixed not this 219 // code. However since gc isn't locked out the values could be 220 // stale. This is a race we can never completely win since we can't 221 // lock out gc so do one last check after retrieving their values 222 // from the frame for additional safety 223 224 Method* method = fr->interpreter_frame_method(); 225 226 // We've at least found a method. 227 // NOTE: there is something to be said for the approach that 228 // if we don't find a valid bci then the method is not likely 229 // a valid method. Then again we may have caught an interpreter 230 // frame in the middle of construction and the bci field is 231 // not yet valid. 232 233 *method_p = method; 234 if (!method->is_valid_method()) return false; 235 236 intptr_t bcx = fr->interpreter_frame_bcx(); 237 238 int bci = method->validate_bci_from_bcx(bcx); 239 240 // note: bci is set to -1 if not a valid bci 241 *bci_p = bci; 242 return true; 243 } 244 245 return false; 246 } 247 248 249 // Determine if 'fr' can be used to find an initial Java frame. 250 // Return false if it can not find a fully decipherable Java frame 251 // (in other words a frame that isn't safe to use in a vframe stream). 252 // Obviously if it can't even find a Java frame false will also be returned. 253 // 254 // If we find a Java frame decipherable or not then by definition we have 255 // identified a method and that will be returned to the caller via method_p. 256 // If we can determine a bci that is returned also. (Hmm is it possible 257 // to return a method and bci and still return false? ) 258 // 259 // The initial Java frame we find (if any) is return via initial_frame_p. 260 // 261 262 static bool find_initial_Java_frame(JavaThread* thread, 263 frame* fr, 264 frame* initial_frame_p, 265 Method** method_p, 266 int* bci_p) { 267 268 // It is possible that for a frame containing an nmethod 269 // we can capture the method but no bci. If we get no 270 // bci the frame isn't walkable but the method is usable. 271 // Therefore we init the returned Method* to NULL so the 272 // caller can make the distinction. 273 274 *method_p = NULL; 275 276 // On the initial call to this method the frame we get may not be 277 // recognizable to us. This should only happen if we are in a JRT_LEAF 278 // or something called by a JRT_LEAF method. 279 280 281 282 frame candidate = *fr; 283 284 // If the starting frame we were given has no codeBlob associated with 285 // it see if we can find such a frame because only frames with codeBlobs 286 // are possible Java frames. 287 288 if (fr->cb() == NULL) { 289 290 // See if we can find a useful frame 291 int loop_count; 292 int loop_max = MaxJavaStackTraceDepth * 2; 293 RegisterMap map(thread, false); 294 295 for (loop_count = 0; loop_count < loop_max; loop_count++) { 296 if (!candidate.safe_for_sender(thread)) return false; 297 candidate = candidate.sender(&map); 298 if (candidate.cb() != NULL) break; 299 } 300 if (candidate.cb() == NULL) return false; 301 } 302 303 // We have a frame known to be in the codeCache 304 // We will hopefully be able to figure out something to do with it. 305 int loop_count; 306 int loop_max = MaxJavaStackTraceDepth * 2; 307 RegisterMap map(thread, false); 308 309 for (loop_count = 0; loop_count < loop_max; loop_count++) { 310 311 if (candidate.is_first_frame()) { 312 // If initial frame is frame from StubGenerator and there is no 313 // previous anchor, there are no java frames associated with a method 314 return false; 315 } 316 317 if (candidate.is_interpreted_frame()) { 318 if (is_decipherable_interpreted_frame(thread, &candidate, method_p, bci_p)) { 319 *initial_frame_p = candidate; 320 return true; 321 } 322 323 // Hopefully we got some data 324 return false; 325 } 326 327 if (candidate.cb()->is_nmethod()) { 328 329 nmethod* nm = (nmethod*) candidate.cb(); 330 *method_p = nm->method(); 331 332 // If the frame isn't fully decipherable then the default 333 // value for the bci is a signal that we don't have a bci. 334 // If we have a decipherable frame this bci value will 335 // not be used. 336 337 *bci_p = -1; 338 339 *initial_frame_p = candidate; 340 341 // Native wrapper code is trivial to decode by vframeStream 342 343 if (nm->is_native_method()) return true; 344 345 // If it isn't decipherable then we have found a pc that doesn't 346 // have a PCDesc that can get us a bci however we did find 347 // a method 348 349 if (!is_decipherable_compiled_frame(thread, &candidate, nm)) { 350 return false; 351 } 352 353 // is_decipherable_compiled_frame may modify candidate's pc 354 *initial_frame_p = candidate; 355 356 assert(nm->pc_desc_at(candidate.pc()) != NULL, "if it's decipherable then pc must be valid"); 357 358 return true; 359 } 360 361 // Must be some stub frame that we don't care about 362 363 if (!candidate.safe_for_sender(thread)) return false; 364 candidate = candidate.sender(&map); 365 366 // If it isn't in the code cache something is wrong 367 // since once we find a frame in the code cache they 368 // all should be there. 369 370 if (candidate.cb() == NULL) return false; 371 372 } 373 374 return false; 375 376 } 377 378 static void forte_fill_call_trace_given_top(JavaThread* thd, 379 ASGCT_CallTrace* trace, 380 int depth, 381 frame top_frame) { 382 NoHandleMark nhm; 383 384 frame initial_Java_frame; 385 Method* method; 386 int bci; 387 int count; 388 389 count = 0; 390 assert(trace->frames != NULL, "trace->frames must be non-NULL"); 391 392 bool fully_decipherable = find_initial_Java_frame(thd, &top_frame, &initial_Java_frame, &method, &bci); 393 394 // The frame might not be walkable but still recovered a method 395 // (e.g. an nmethod with no scope info for the pc) 396 397 if (method == NULL) return; 398 399 if (!method->is_valid_method()) { 400 trace->num_frames = ticks_GC_active; // -2 401 return; 402 } 403 404 // We got a Java frame however it isn't fully decipherable 405 // so it won't necessarily be safe to use it for the 406 // initial frame in the vframe stream. 407 408 if (!fully_decipherable) { 409 // Take whatever method the top-frame decoder managed to scrape up. 410 // We look further at the top frame only if non-safepoint 411 // debugging information is available. 412 count++; 413 trace->num_frames = count; 414 trace->frames[0].method_id = method->find_jmethod_id_or_null(); 415 if (!method->is_native()) { 416 trace->frames[0].lineno = bci; 417 } else { 418 trace->frames[0].lineno = -3; 419 } 420 421 if (!initial_Java_frame.safe_for_sender(thd)) return; 422 423 RegisterMap map(thd, false); 424 initial_Java_frame = initial_Java_frame.sender(&map); 425 } 426 427 vframeStreamForte st(thd, initial_Java_frame, false); 428 429 for (; !st.at_end() && count < depth; st.forte_next(), count++) { 430 bci = st.bci(); 431 method = st.method(); 432 433 if (!method->is_valid_method()) { 434 // we throw away everything we've gathered in this sample since 435 // none of it is safe 436 trace->num_frames = ticks_GC_active; // -2 437 return; 438 } 439 440 trace->frames[count].method_id = method->find_jmethod_id_or_null(); 441 if (!method->is_native()) { 442 trace->frames[count].lineno = bci; 443 } else { 444 trace->frames[count].lineno = -3; 445 } 446 } 447 trace->num_frames = count; 448 return; 449 } 450 451 452 // Forte Analyzer AsyncGetCallTrace() entry point. Currently supported 453 // on Linux X86, Solaris SPARC and Solaris X86. 454 // 455 // Async-safe version of GetCallTrace being called from a signal handler 456 // when a LWP gets interrupted by SIGPROF but the stack traces are filled 457 // with different content (see below). 458 // 459 // This function must only be called when JVM/TI 460 // CLASS_LOAD events have been enabled since agent startup. The enabled 461 // event will cause the jmethodIDs to be allocated at class load time. 462 // The jmethodIDs cannot be allocated in a signal handler because locks 463 // cannot be grabbed in a signal handler safely. 464 // 465 // void (*AsyncGetCallTrace)(ASGCT_CallTrace *trace, jint depth, void* ucontext) 466 // 467 // Called by the profiler to obtain the current method call stack trace for 468 // a given thread. The thread is identified by the env_id field in the 469 // ASGCT_CallTrace structure. The profiler agent should allocate a ASGCT_CallTrace 470 // structure with enough memory for the requested stack depth. The VM fills in 471 // the frames buffer and the num_frames field. 472 // 473 // Arguments: 474 // 475 // trace - trace data structure to be filled by the VM. 476 // depth - depth of the call stack trace. 477 // ucontext - ucontext_t of the LWP 478 // 479 // ASGCT_CallTrace: 480 // typedef struct { 481 // JNIEnv *env_id; 482 // jint num_frames; 483 // ASGCT_CallFrame *frames; 484 // } ASGCT_CallTrace; 485 // 486 // Fields: 487 // env_id - ID of thread which executed this trace. 488 // num_frames - number of frames in the trace. 489 // (< 0 indicates the frame is not walkable). 490 // frames - the ASGCT_CallFrames that make up this trace. Callee followed by callers. 491 // 492 // ASGCT_CallFrame: 493 // typedef struct { 494 // jint lineno; 495 // jmethodID method_id; 496 // } ASGCT_CallFrame; 497 // 498 // Fields: 499 // 1) For Java frame (interpreted and compiled), 500 // lineno - bci of the method being executed or -1 if bci is not available 501 // method_id - jmethodID of the method being executed 502 // 2) For native method 503 // lineno - (-3) 504 // method_id - jmethodID of the method being executed 505 506 extern "C" { 507 JNIEXPORT 508 void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext) { 509 JavaThread* thread; 510 511 if (trace->env_id == NULL || 512 (thread = JavaThread::thread_from_jni_environment(trace->env_id)) == NULL || 513 thread->is_exiting()) { 514 515 // bad env_id, thread has exited or thread is exiting 516 trace->num_frames = ticks_thread_exit; // -8 517 return; 518 } 519 520 if (thread->in_deopt_handler()) { 521 // thread is in the deoptimization handler so return no frames 522 trace->num_frames = ticks_deopt; // -9 523 return; 524 } 525 526 assert(JavaThread::current() == thread, 527 "AsyncGetCallTrace must be called by the current interrupted thread"); 528 529 if (!JvmtiExport::should_post_class_load()) { 530 trace->num_frames = ticks_no_class_load; // -1 531 return; 532 } 533 534 if (Universe::heap()->is_gc_active()) { 535 trace->num_frames = ticks_GC_active; // -2 536 return; 537 } 538 539 switch (thread->thread_state()) { 540 case _thread_new: 541 case _thread_uninitialized: 542 case _thread_new_trans: 543 // We found the thread on the threads list above, but it is too 544 // young to be useful so return that there are no Java frames. 545 trace->num_frames = 0; 546 break; 547 case _thread_in_native: 548 case _thread_in_native_trans: 549 case _thread_blocked: 550 case _thread_blocked_trans: 551 case _thread_in_vm: 552 case _thread_in_vm_trans: 553 { 554 frame fr; 555 556 // param isInJava == false - indicate we aren't in Java code 557 if (!thread->pd_get_top_frame_for_signal_handler(&fr, ucontext, false)) { 558 trace->num_frames = ticks_unknown_not_Java; // -3 unknown frame 559 } else { 560 if (!thread->has_last_Java_frame()) { 561 trace->num_frames = 0; // No Java frames 562 } else { 563 trace->num_frames = ticks_not_walkable_not_Java; // -4 non walkable frame by default 564 forte_fill_call_trace_given_top(thread, trace, depth, fr); 565 566 // This assert would seem to be valid but it is not. 567 // It would be valid if we weren't possibly racing a gc 568 // thread. A gc thread can make a valid interpreted frame 569 // look invalid. It's a small window but it does happen. 570 // The assert is left here commented out as a reminder. 571 // assert(trace->num_frames != ticks_not_walkable_not_Java, "should always be walkable"); 572 573 } 574 } 575 } 576 break; 577 case _thread_in_Java: 578 case _thread_in_Java_trans: 579 { 580 frame fr; 581 582 // param isInJava == true - indicate we are in Java code 583 if (!thread->pd_get_top_frame_for_signal_handler(&fr, ucontext, true)) { 584 trace->num_frames = ticks_unknown_Java; // -5 unknown frame 585 } else { 586 trace->num_frames = ticks_not_walkable_Java; // -6, non walkable frame by default 587 forte_fill_call_trace_given_top(thread, trace, depth, fr); 588 } 589 } 590 break; 591 default: 592 // Unknown thread state 593 trace->num_frames = ticks_unknown_state; // -7 594 break; 595 } 596 } 597 598 599 #ifndef _WINDOWS 600 // Support for the Forte(TM) Peformance Tools collector. 601 // 602 // The method prototype is derived from libcollector.h. For more 603 // information, please see the libcollect man page. 604 605 // Method to let libcollector know about a dynamically loaded function. 606 // Because it is weakly bound, the calls become NOP's when the library 607 // isn't present. 608 #ifdef __APPLE__ 609 // XXXDARWIN: Link errors occur even when __attribute__((weak_import)) 610 // is added 611 #define collector_func_load(x0,x1,x2,x3,x4,x5,x6) (0) 612 #else 613 void collector_func_load(char* name, 614 void* null_argument_1, 615 void* null_argument_2, 616 void *vaddr, 617 int size, 618 int zero_argument, 619 void* null_argument_3); 620 #pragma weak collector_func_load 621 #define collector_func_load(x0,x1,x2,x3,x4,x5,x6) \ 622 ( collector_func_load ? collector_func_load(x0,x1,x2,x3,x4,x5,x6),(void)0 : (void)0 ) 623 #endif // __APPLE__ 624 #endif // !_WINDOWS 625 626 } // end extern "C" 627 #endif // !IA64 628 629 void Forte::register_stub(const char* name, address start, address end) { 630 #if !defined(_WINDOWS) && !defined(IA64) 631 assert(pointer_delta(end, start, sizeof(jbyte)) < INT_MAX, 632 "Code size exceeds maximum range"); 633 634 collector_func_load((char*)name, NULL, NULL, start, 635 pointer_delta(end, start, sizeof(jbyte)), 0, NULL); 636 #endif // !_WINDOWS && !IA64 637 } 638 639 #else // INCLUDE_JVMTI 640 extern "C" { 641 JNIEXPORT 642 void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext) { 643 trace->num_frames = ticks_no_class_load; // -1 644 } 645 } 646 #endif // INCLUDE_JVMTI