src/share/vm/prims/forte.cpp

Wed, 27 Aug 2014 08:19:12 -0400

author
zgu
date
Wed, 27 Aug 2014 08:19:12 -0400
changeset 7074
833b0f92429a
parent 6911
ce8f6bb717c9
child 7535
7ae4e26cb1e0
child 7854
e8260b6328fb
permissions
-rw-r--r--

8046598: Scalable Native memory tracking development
Summary: Enhance scalability of native memory tracking
Reviewed-by: coleenp, ctornqvi, gtriantafill

     1 /*
     2  * Copyright (c) 2003, 2013, 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  */
    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/javaCalls.hpp"
    35 #include "runtime/thread.inline.hpp"
    36 #include "runtime/vframe.hpp"
    37 #include "runtime/vframeArray.hpp"
    39 // call frame copied from old .h file and renamed
    40 typedef struct {
    41     jint lineno;                      // line number in the source file
    42     jmethodID method_id;              // method executed in this frame
    43 } ASGCT_CallFrame;
    45 // call trace copied from old .h file and renamed
    46 typedef struct {
    47     JNIEnv *env_id;                   // Env where trace was recorded
    48     jint num_frames;                  // number of frames in this trace
    49     ASGCT_CallFrame *frames;          // frames
    50 } ASGCT_CallTrace;
    52 // These name match the names reported by the forte quality kit
    53 enum {
    54   ticks_no_Java_frame         =  0,
    55   ticks_no_class_load         = -1,
    56   ticks_GC_active             = -2,
    57   ticks_unknown_not_Java      = -3,
    58   ticks_not_walkable_not_Java = -4,
    59   ticks_unknown_Java          = -5,
    60   ticks_not_walkable_Java     = -6,
    61   ticks_unknown_state         = -7,
    62   ticks_thread_exit           = -8,
    63   ticks_deopt                 = -9,
    64   ticks_safepoint             = -10
    65 };
    67 #if INCLUDE_JVMTI
    69 //-------------------------------------------------------
    71 // Native interfaces for use by Forte tools.
    74 #if !defined(IA64) && !defined(PPC64)
    76 class vframeStreamForte : public vframeStreamCommon {
    77  public:
    78   // constructor that starts with sender of frame fr (top_frame)
    79   vframeStreamForte(JavaThread *jt, frame fr, bool stop_at_java_call_stub);
    80   void forte_next();
    81 };
    84 static bool is_decipherable_compiled_frame(JavaThread* thread, frame* fr, nmethod* nm);
    85 static bool is_decipherable_interpreted_frame(JavaThread* thread,
    86                                               frame* fr,
    87                                               Method** method_p,
    88                                               int* bci_p);
    93 vframeStreamForte::vframeStreamForte(JavaThread *jt,
    94                                      frame fr,
    95                                      bool stop_at_java_call_stub) : vframeStreamCommon(jt) {
    97   _stop_at_java_call_stub = stop_at_java_call_stub;
    98   _frame = fr;
   100   // We must always have a valid frame to start filling
   102   bool filled_in = fill_from_frame();
   104   assert(filled_in, "invariant");
   106 }
   109 // Solaris SPARC Compiler1 needs an additional check on the grandparent
   110 // of the top_frame when the parent of the top_frame is interpreted and
   111 // the grandparent is compiled. However, in this method we do not know
   112 // the relationship of the current _frame relative to the top_frame so
   113 // we implement a more broad sanity check. When the previous callee is
   114 // interpreted and the current sender is compiled, we verify that the
   115 // current sender is also walkable. If it is not walkable, then we mark
   116 // the current vframeStream as at the end.
   117 void vframeStreamForte::forte_next() {
   118   // handle frames with inlining
   119   if (_mode == compiled_mode &&
   120       vframeStreamCommon::fill_in_compiled_inlined_sender()) {
   121     return;
   122   }
   124   // handle general case
   126   int loop_count = 0;
   127   int loop_max = MaxJavaStackTraceDepth * 2;
   130   do {
   132     loop_count++;
   134     // By the time we get here we should never see unsafe but better
   135     // safe then segv'd
   137     if (loop_count > loop_max || !_frame.safe_for_sender(_thread)) {
   138       _mode = at_end_mode;
   139       return;
   140     }
   142     _frame = _frame.sender(&_reg_map);
   144   } while (!fill_from_frame());
   145 }
   147 // Determine if 'fr' is a decipherable compiled frame. We are already
   148 // assured that fr is for a java nmethod.
   150 static bool is_decipherable_compiled_frame(JavaThread* thread, frame* fr, nmethod* nm) {
   151   assert(nm->is_java_method(), "invariant");
   153   if (thread->has_last_Java_frame() && thread->last_Java_pc() == fr->pc()) {
   154     // We're stopped at a call into the JVM so look for a PcDesc with
   155     // the actual pc reported by the frame.
   156     PcDesc* pc_desc = nm->pc_desc_at(fr->pc());
   158     // Did we find a useful PcDesc?
   159     if (pc_desc != NULL &&
   160         pc_desc->scope_decode_offset() != DebugInformationRecorder::serialized_null) {
   161       return true;
   162     }
   163   }
   165   // We're at some random pc in the nmethod so search for the PcDesc
   166   // whose pc is greater than the current PC.  It's done this way
   167   // because the extra PcDescs that are recorded for improved debug
   168   // info record the end of the region covered by the ScopeDesc
   169   // instead of the beginning.
   170   PcDesc* pc_desc = nm->pc_desc_near(fr->pc() + 1);
   172   // Now do we have a useful PcDesc?
   173   if (pc_desc == NULL ||
   174       pc_desc->scope_decode_offset() == DebugInformationRecorder::serialized_null) {
   175     // No debug information available for this pc
   176     // vframeStream would explode if we try and walk the frames.
   177     return false;
   178   }
   180   // This PcDesc is useful however we must adjust the frame's pc
   181   // so that the vframeStream lookups will use this same pc
   182   fr->set_pc(pc_desc->real_pc(nm));
   183   return true;
   184 }
   187 // Determine if 'fr' is a walkable interpreted frame. Returns false
   188 // if it is not. *method_p, and *bci_p are not set when false is
   189 // returned. *method_p is non-NULL if frame was executing a Java
   190 // method. *bci_p is != -1 if a valid BCI in the Java method could
   191 // be found.
   192 // Note: this method returns true when a valid Java method is found
   193 // even if a valid BCI cannot be found.
   195 static bool is_decipherable_interpreted_frame(JavaThread* thread,
   196                                               frame* fr,
   197                                               Method** method_p,
   198                                               int* bci_p) {
   199   assert(fr->is_interpreted_frame(), "just checking");
   201   // top frame is an interpreted frame
   202   // check if it is walkable (i.e. valid Method* and valid bci)
   204   // Because we may be racing a gc thread the method and/or bci
   205   // of a valid interpreter frame may look bad causing us to
   206   // fail the is_interpreted_frame_valid test. If the thread
   207   // is in any of the following states we are assured that the
   208   // frame is in fact valid and we must have hit the race.
   210   JavaThreadState state = thread->thread_state();
   211   bool known_valid = (state == _thread_in_native ||
   212                       state == _thread_in_vm ||
   213                       state == _thread_blocked );
   215   if (known_valid || fr->is_interpreted_frame_valid(thread)) {
   217     // The frame code should completely validate the frame so that
   218     // references to Method* and bci are completely safe to access
   219     // If they aren't the frame code should be fixed not this
   220     // code. However since gc isn't locked out the values could be
   221     // stale. This is a race we can never completely win since we can't
   222     // lock out gc so do one last check after retrieving their values
   223     // from the frame for additional safety
   225     Method* method = fr->interpreter_frame_method();
   227     // We've at least found a method.
   228     // NOTE: there is something to be said for the approach that
   229     // if we don't find a valid bci then the method is not likely
   230     // a valid method. Then again we may have caught an interpreter
   231     // frame in the middle of construction and the bci field is
   232     // not yet valid.
   234     *method_p = method;
   235     if (!method->is_valid_method()) return false;
   237     intptr_t bcx = fr->interpreter_frame_bcx();
   239     int      bci = method->validate_bci_from_bcx(bcx);
   241     // note: bci is set to -1 if not a valid bci
   242     *bci_p = bci;
   243     return true;
   244   }
   246   return false;
   247 }
   250 // Determine if 'fr' can be used to find an initial Java frame.
   251 // Return false if it can not find a fully decipherable Java frame
   252 // (in other words a frame that isn't safe to use in a vframe stream).
   253 // Obviously if it can't even find a Java frame false will also be returned.
   254 //
   255 // If we find a Java frame decipherable or not then by definition we have
   256 // identified a method and that will be returned to the caller via method_p.
   257 // If we can determine a bci that is returned also. (Hmm is it possible
   258 // to return a method and bci and still return false? )
   259 //
   260 // The initial Java frame we find (if any) is return via initial_frame_p.
   261 //
   263 static bool find_initial_Java_frame(JavaThread* thread,
   264                                     frame* fr,
   265                                     frame* initial_frame_p,
   266                                     Method** method_p,
   267                                     int* bci_p) {
   269   // It is possible that for a frame containing an nmethod
   270   // we can capture the method but no bci. If we get no
   271   // bci the frame isn't walkable but the method is usable.
   272   // Therefore we init the returned Method* to NULL so the
   273   // caller can make the distinction.
   275   *method_p = NULL;
   277   // On the initial call to this method the frame we get may not be
   278   // recognizable to us. This should only happen if we are in a JRT_LEAF
   279   // or something called by a JRT_LEAF method.
   283   frame candidate = *fr;
   285   // If the starting frame we were given has no codeBlob associated with
   286   // it see if we can find such a frame because only frames with codeBlobs
   287   // are possible Java frames.
   289   if (fr->cb() == NULL) {
   291     // See if we can find a useful frame
   292     int loop_count;
   293     int loop_max = MaxJavaStackTraceDepth * 2;
   294     RegisterMap map(thread, false);
   296     for (loop_count = 0; loop_count < loop_max; loop_count++) {
   297       if (!candidate.safe_for_sender(thread)) return false;
   298       candidate = candidate.sender(&map);
   299       if (candidate.cb() != NULL) break;
   300     }
   301     if (candidate.cb() == NULL) return false;
   302   }
   304   // We have a frame known to be in the codeCache
   305   // We will hopefully be able to figure out something to do with it.
   306   int loop_count;
   307   int loop_max = MaxJavaStackTraceDepth * 2;
   308   RegisterMap map(thread, false);
   310   for (loop_count = 0; loop_count < loop_max; loop_count++) {
   312     if (candidate.is_entry_frame()) {
   313       // jcw is NULL if the java call wrapper couldn't be found
   314       JavaCallWrapper *jcw = candidate.entry_frame_call_wrapper_if_safe(thread);
   315       // If initial frame is frame from StubGenerator and there is no
   316       // previous anchor, there are no java frames associated with a method
   317       if (jcw == NULL || jcw->is_first_frame()) {
   318         return false;
   319       }
   320     }
   322     if (candidate.is_interpreted_frame()) {
   323       if (is_decipherable_interpreted_frame(thread, &candidate, method_p, bci_p)) {
   324         *initial_frame_p = candidate;
   325         return true;
   326       }
   328       // Hopefully we got some data
   329       return false;
   330     }
   332     if (candidate.cb()->is_nmethod()) {
   334       nmethod* nm = (nmethod*) candidate.cb();
   335       *method_p = nm->method();
   337       // If the frame isn't fully decipherable then the default
   338       // value for the bci is a signal that we don't have a bci.
   339       // If we have a decipherable frame this bci value will
   340       // not be used.
   342       *bci_p = -1;
   344       *initial_frame_p = candidate;
   346       // Native wrapper code is trivial to decode by vframeStream
   348       if (nm->is_native_method()) return true;
   350       // If it isn't decipherable then we have found a pc that doesn't
   351       // have a PCDesc that can get us a bci however we did find
   352       // a method
   354       if (!is_decipherable_compiled_frame(thread, &candidate, nm)) {
   355         return false;
   356       }
   358       // is_decipherable_compiled_frame may modify candidate's pc
   359       *initial_frame_p = candidate;
   361       assert(nm->pc_desc_at(candidate.pc()) != NULL, "if it's decipherable then pc must be valid");
   363       return true;
   364     }
   366     // Must be some stub frame that we don't care about
   368     if (!candidate.safe_for_sender(thread)) return false;
   369     candidate = candidate.sender(&map);
   371     // If it isn't in the code cache something is wrong
   372     // since once we find a frame in the code cache they
   373     // all should be there.
   375     if (candidate.cb() == NULL) return false;
   377   }
   379   return false;
   381 }
   383 static void forte_fill_call_trace_given_top(JavaThread* thd,
   384                                             ASGCT_CallTrace* trace,
   385                                             int depth,
   386                                             frame top_frame) {
   387   NoHandleMark nhm;
   389   frame initial_Java_frame;
   390   Method* method;
   391   int bci;
   392   int count;
   394   count = 0;
   395   assert(trace->frames != NULL, "trace->frames must be non-NULL");
   397   bool fully_decipherable = find_initial_Java_frame(thd, &top_frame, &initial_Java_frame, &method, &bci);
   399   // The frame might not be walkable but still recovered a method
   400   // (e.g. an nmethod with no scope info for the pc)
   402   if (method == NULL) return;
   404   if (!method->is_valid_method()) {
   405     trace->num_frames = ticks_GC_active; // -2
   406     return;
   407   }
   409   // We got a Java frame however it isn't fully decipherable
   410   // so it won't necessarily be safe to use it for the
   411   // initial frame in the vframe stream.
   413   if (!fully_decipherable) {
   414     // Take whatever method the top-frame decoder managed to scrape up.
   415     // We look further at the top frame only if non-safepoint
   416     // debugging information is available.
   417     count++;
   418     trace->num_frames = count;
   419     trace->frames[0].method_id = method->find_jmethod_id_or_null();
   420     if (!method->is_native()) {
   421       trace->frames[0].lineno = bci;
   422     } else {
   423       trace->frames[0].lineno = -3;
   424     }
   426     if (!initial_Java_frame.safe_for_sender(thd)) return;
   428     RegisterMap map(thd, false);
   429     initial_Java_frame = initial_Java_frame.sender(&map);
   430   }
   432   vframeStreamForte st(thd, initial_Java_frame, false);
   434   for (; !st.at_end() && count < depth; st.forte_next(), count++) {
   435     bci = st.bci();
   436     method = st.method();
   438     if (!method->is_valid_method()) {
   439       // we throw away everything we've gathered in this sample since
   440       // none of it is safe
   441       trace->num_frames = ticks_GC_active; // -2
   442       return;
   443     }
   445     trace->frames[count].method_id = method->find_jmethod_id_or_null();
   446     if (!method->is_native()) {
   447       trace->frames[count].lineno = bci;
   448     } else {
   449       trace->frames[count].lineno = -3;
   450     }
   451   }
   452   trace->num_frames = count;
   453   return;
   454 }
   457 // Forte Analyzer AsyncGetCallTrace() entry point. Currently supported
   458 // on Linux X86, Solaris SPARC and Solaris X86.
   459 //
   460 // Async-safe version of GetCallTrace being called from a signal handler
   461 // when a LWP gets interrupted by SIGPROF but the stack traces are filled
   462 // with different content (see below).
   463 //
   464 // This function must only be called when JVM/TI
   465 // CLASS_LOAD events have been enabled since agent startup. The enabled
   466 // event will cause the jmethodIDs to be allocated at class load time.
   467 // The jmethodIDs cannot be allocated in a signal handler because locks
   468 // cannot be grabbed in a signal handler safely.
   469 //
   470 // void (*AsyncGetCallTrace)(ASGCT_CallTrace *trace, jint depth, void* ucontext)
   471 //
   472 // Called by the profiler to obtain the current method call stack trace for
   473 // a given thread. The thread is identified by the env_id field in the
   474 // ASGCT_CallTrace structure. The profiler agent should allocate a ASGCT_CallTrace
   475 // structure with enough memory for the requested stack depth. The VM fills in
   476 // the frames buffer and the num_frames field.
   477 //
   478 // Arguments:
   479 //
   480 //   trace    - trace data structure to be filled by the VM.
   481 //   depth    - depth of the call stack trace.
   482 //   ucontext - ucontext_t of the LWP
   483 //
   484 // ASGCT_CallTrace:
   485 //   typedef struct {
   486 //       JNIEnv *env_id;
   487 //       jint num_frames;
   488 //       ASGCT_CallFrame *frames;
   489 //   } ASGCT_CallTrace;
   490 //
   491 // Fields:
   492 //   env_id     - ID of thread which executed this trace.
   493 //   num_frames - number of frames in the trace.
   494 //                (< 0 indicates the frame is not walkable).
   495 //   frames     - the ASGCT_CallFrames that make up this trace. Callee followed by callers.
   496 //
   497 //  ASGCT_CallFrame:
   498 //    typedef struct {
   499 //        jint lineno;
   500 //        jmethodID method_id;
   501 //    } ASGCT_CallFrame;
   502 //
   503 //  Fields:
   504 //    1) For Java frame (interpreted and compiled),
   505 //       lineno    - bci of the method being executed or -1 if bci is not available
   506 //       method_id - jmethodID of the method being executed
   507 //    2) For native method
   508 //       lineno    - (-3)
   509 //       method_id - jmethodID of the method being executed
   511 extern "C" {
   512 JNIEXPORT
   513 void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext) {
   514   JavaThread* thread;
   516   if (trace->env_id == NULL ||
   517     (thread = JavaThread::thread_from_jni_environment(trace->env_id)) == NULL ||
   518     thread->is_exiting()) {
   520     // bad env_id, thread has exited or thread is exiting
   521     trace->num_frames = ticks_thread_exit; // -8
   522     return;
   523   }
   525   if (thread->in_deopt_handler()) {
   526     // thread is in the deoptimization handler so return no frames
   527     trace->num_frames = ticks_deopt; // -9
   528     return;
   529   }
   531   assert(JavaThread::current() == thread,
   532          "AsyncGetCallTrace must be called by the current interrupted thread");
   534   if (!JvmtiExport::should_post_class_load()) {
   535     trace->num_frames = ticks_no_class_load; // -1
   536     return;
   537   }
   539   if (Universe::heap()->is_gc_active()) {
   540     trace->num_frames = ticks_GC_active; // -2
   541     return;
   542   }
   544   switch (thread->thread_state()) {
   545   case _thread_new:
   546   case _thread_uninitialized:
   547   case _thread_new_trans:
   548     // We found the thread on the threads list above, but it is too
   549     // young to be useful so return that there are no Java frames.
   550     trace->num_frames = 0;
   551     break;
   552   case _thread_in_native:
   553   case _thread_in_native_trans:
   554   case _thread_blocked:
   555   case _thread_blocked_trans:
   556   case _thread_in_vm:
   557   case _thread_in_vm_trans:
   558     {
   559       frame fr;
   561       // param isInJava == false - indicate we aren't in Java code
   562       if (!thread->pd_get_top_frame_for_signal_handler(&fr, ucontext, false)) {
   563         trace->num_frames = ticks_unknown_not_Java;  // -3 unknown frame
   564       } else {
   565         if (!thread->has_last_Java_frame()) {
   566           trace->num_frames = 0; // No Java frames
   567         } else {
   568           trace->num_frames = ticks_not_walkable_not_Java;    // -4 non walkable frame by default
   569           forte_fill_call_trace_given_top(thread, trace, depth, fr);
   571           // This assert would seem to be valid but it is not.
   572           // It would be valid if we weren't possibly racing a gc
   573           // thread. A gc thread can make a valid interpreted frame
   574           // look invalid. It's a small window but it does happen.
   575           // The assert is left here commented out as a reminder.
   576           // assert(trace->num_frames != ticks_not_walkable_not_Java, "should always be walkable");
   578         }
   579       }
   580     }
   581     break;
   582   case _thread_in_Java:
   583   case _thread_in_Java_trans:
   584     {
   585       frame fr;
   587       // param isInJava == true - indicate we are in Java code
   588       if (!thread->pd_get_top_frame_for_signal_handler(&fr, ucontext, true)) {
   589         trace->num_frames = ticks_unknown_Java;  // -5 unknown frame
   590       } else {
   591         trace->num_frames = ticks_not_walkable_Java;  // -6, non walkable frame by default
   592         forte_fill_call_trace_given_top(thread, trace, depth, fr);
   593       }
   594     }
   595     break;
   596   default:
   597     // Unknown thread state
   598     trace->num_frames = ticks_unknown_state; // -7
   599     break;
   600   }
   601 }
   604 #ifndef _WINDOWS
   605 // Support for the Forte(TM) Peformance Tools collector.
   606 //
   607 // The method prototype is derived from libcollector.h. For more
   608 // information, please see the libcollect man page.
   610 // Method to let libcollector know about a dynamically loaded function.
   611 // Because it is weakly bound, the calls become NOP's when the library
   612 // isn't present.
   613 #ifdef __APPLE__
   614 // XXXDARWIN: Link errors occur even when __attribute__((weak_import))
   615 // is added
   616 #define collector_func_load(x0,x1,x2,x3,x4,x5,x6) (0)
   617 #else
   618 void    collector_func_load(char* name,
   619                             void* null_argument_1,
   620                             void* null_argument_2,
   621                             void *vaddr,
   622                             int size,
   623                             int zero_argument,
   624                             void* null_argument_3);
   625 #pragma weak collector_func_load
   626 #define collector_func_load(x0,x1,x2,x3,x4,x5,x6) \
   627         ( collector_func_load ? collector_func_load(x0,x1,x2,x3,x4,x5,x6),(void)0 : (void)0 )
   628 #endif // __APPLE__
   629 #endif // !_WINDOWS
   631 } // end extern "C"
   632 #endif // !IA64 && !PPC64
   634 void Forte::register_stub(const char* name, address start, address end) {
   635 #if !defined(_WINDOWS) && !defined(IA64) && !defined(PPC64)
   636   assert(pointer_delta(end, start, sizeof(jbyte)) < INT_MAX,
   637          "Code size exceeds maximum range");
   639   collector_func_load((char*)name, NULL, NULL, start,
   640     pointer_delta(end, start, sizeof(jbyte)), 0, NULL);
   641 #endif // !_WINDOWS && !IA64 && !PPC64
   642 }
   644 #else // INCLUDE_JVMTI
   645 extern "C" {
   646   JNIEXPORT
   647   void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext) {
   648     trace->num_frames = ticks_no_class_load; // -1
   649   }
   650 }
   651 #endif // INCLUDE_JVMTI

mercurial