forte.cpp revision 1879
* This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any // These name match the names reported by the forte quality kit //------------------------------------------------------- // Native interfaces for use by Forte tools. // constructor that starts with sender of frame fr (top_frame) // We must always have a valid frame to start filling // Solaris SPARC Compiler1 needs an additional check on the grandparent // of the top_frame when the parent of the top_frame is interpreted and // the grandparent is compiled. However, in this method we do not know // the relationship of the current _frame relative to the top_frame so // we implement a more broad sanity check. When the previous callee is // interpreted and the current sender is compiled, we verify that the // current sender is also walkable. If it is not walkable, then we mark // the current vframeStream as at the end. // handle frames with inlining // By the time we get here we should never see unsafe but better // Determine if 'fr' is a decipherable compiled frame. We are already // assured that fr is for a java nmethod. // We're stopped at a call into the JVM so look for a PcDesc with // the actual pc reported by the frame. // Did we find a useful PcDesc? // We're at some random pc in the nmethod so search for the PcDesc // whose pc is greater than the current PC. It's done this way // because the extra PcDescs that are recorded for improved debug // info record the end of the region covered by the ScopeDesc // instead of the beginning. // Now do we have a useful PcDesc? // No debug information available for this pc // vframeStream would explode if we try and walk the frames. // This PcDesc is useful however we must adjust the frame's pc // so that the vframeStream lookups will use this same pc // Determine if 'fr' is a walkable interpreted frame. Returns false // if it is not. *method_p, and *bci_p are not set when false is // returned. *method_p is non-NULL if frame was executing a Java // method. *bci_p is != -1 if a valid BCI in the Java method could // Note: this method returns true when a valid Java method is found // even if a valid BCI cannot be found. // top frame is an interpreted frame // check if it is walkable (i.e. valid methodOop and valid bci) // Because we may be racing a gc thread the method and/or bci // of a valid interpreter frame may look bad causing us to // fail the is_interpreted_frame_valid test. If the thread // is in any of the following states we are assured that the // frame is in fact valid and we must have hit the race. // The frame code should completely validate the frame so that // references to methodOop and bci are completely safe to access // If they aren't the frame code should be fixed not this // code. However since gc isn't locked out the values could be // stale. This is a race we can never completely win since we can't // lock out gc so do one last check after retrieving their values // from the frame for additional safety // We've at least found a method. // NOTE: there is something to be said for the approach that // if we don't find a valid bci then the method is not likely // a valid method. Then again we may have caught an interpreter // frame in the middle of construction and the bci field is // See if gc may have invalidated method since we validated frame // note: bci is set to -1 if not a valid bci // Determine if 'fr' can be used to find an initial Java frame. // Return false if it can not find a fully decipherable Java frame // (in other words a frame that isn't safe to use in a vframe stream). // Obviously if it can't even find a Java frame false will also be returned. // If we find a Java frame decipherable or not then by definition we have // identified a method and that will be returned to the caller via method_p. // If we can determine a bci that is returned also. (Hmm is it possible // to return a method and bci and still return false? ) // The initial Java frame we find (if any) is return via initial_frame_p. // It is possible that for a frame containing an nmethod // we can capture the method but no bci. If we get no // bci the frame isn't walkable but the method is usable. // Therefore we init the returned methodOop to NULL so the // caller can make the distinction. // On the initial call to this method the frame we get may not be // recognizable to us. This should only happen if we are in a JRT_LEAF // or something called by a JRT_LEAF method. // If the starting frame we were given has no codeBlob associated with // it see if we can find such a frame because only frames with codeBlobs // are possible Java frames. // See if we can find a useful frame // We have a frame known to be in the codeCache // We will hopefully be able to figure out something to do with it. // If initial frame is frame from StubGenerator and there is no // previous anchor, there are no java frames associated with a method // Hopefully we got some data // If the frame isn't fully decipherable then the default // value for the bci is a signal that we don't have a bci. // If we have a decipherable frame this bci value will // Native wrapper code is trivial to decode by vframeStream // If it isn't decipherable then we have found a pc that doesn't // have a PCDesc that can get us a bci however we did find // is_decipherable_compiled_frame may modify candidate's pc // Must be some stub frame that we don't care about // If it isn't in the code cache something is wrong // since once we find a frame in the code cache they // call frame copied from old .h file and renamed // call trace copied from old .h file and renamed // The frame might not be walkable but still recovered a method // (e.g. an nmethod with no scope info for the pc // The method is not stored GC safe so see if GC became active // after we entered AsyncGetCallTrace() and before we try to // Yes, there is still a window after this check and before // we use methodOop below, but we can't lock out GC so that // has to be an acceptable risk. // We got a Java frame however it isn't fully decipherable // so it won't necessarily be safe to use it for the // initial frame in the vframe stream. // Take whatever method the top-frame decoder managed to scrape up. // We look further at the top frame only if non-safepoint // debugging information is available. // The method is not stored GC safe so see if GC became active // after we entered AsyncGetCallTrace() and before we try to // Yes, there is still a window after this check and before // we use methodOop below, but we can't lock out GC so that // has to be an acceptable risk. // we throw away everything we've gathered in this sample since // Forte Analyzer AsyncGetCallTrace() entry point. Currently supported // on Linux X86, Solaris SPARC and Solaris X86. // Async-safe version of GetCallTrace being called from a signal handler // when a LWP gets interrupted by SIGPROF but the stack traces are filled // with different content (see below). // This function must only be called when JVM/TI // CLASS_LOAD events have been enabled since agent startup. The enabled // event will cause the jmethodIDs to be allocated at class load time. // The jmethodIDs cannot be allocated in a signal handler because locks // cannot be grabbed in a signal handler safely. // void (*AsyncGetCallTrace)(ASGCT_CallTrace *trace, jint depth, void* ucontext) // Called by the profiler to obtain the current method call stack trace for // a given thread. The thread is identified by the env_id field in the // ASGCT_CallTrace structure. The profiler agent should allocate a ASGCT_CallTrace // structure with enough memory for the requested stack depth. The VM fills in // the frames buffer and the num_frames field. // trace - trace data structure to be filled by the VM. // depth - depth of the call stack trace. // ucontext - ucontext_t of the LWP // ASGCT_CallFrame *frames; // env_id - ID of thread which executed this trace. // num_frames - number of frames in the trace. // (< 0 indicates the frame is not walkable). // frames - the ASGCT_CallFrames that make up this trace. Callee followed by callers. // 1) For Java frame (interpreted and compiled), // lineno - bci of the method being executed or -1 if bci is not available // method_id - jmethodID of the method being executed // method_id - jmethodID of the method being executed // This is if'd out because we no longer use thread suspension. // However if someone wanted to backport this to a 5.0 jvm then this // code would be important. // The safepoint mechanism is trying to synchronize all the threads. // Since this can involve thread suspension, it is not safe for us // to be here. We can reduce the deadlock risk window by quickly // returning to the SIGPROF handler. However, it is still possible // for VMThread to catch us here or in the SIGPROF handler. If we // are suspended while holding a resource and another thread blocks // on that resource in the SIGPROF handler, then we will have a // three-thread deadlock (VMThread, this thread, the other thread). // bad env_id, thread has exited or thread is exiting // thread is in the deoptimization handler so return no frames "AsyncGetCallTrace must be called by the current interrupted thread");
// We found the thread on the threads list above, but it is too // young to be useful so return that there are no Java frames. // param isInJava == false - indicate we aren't in Java code // This assert would seem to be valid but it is not. // It would be valid if we weren't possibly racing a gc // thread. A gc thread can make a valid interpreted frame // look invalid. It's a small window but it does happen. // The assert is left here commented out as a reminder. // assert(trace->num_frames != ticks_not_walkable_not_Java, "should always be walkable"); // param isInJava == true - indicate we are in Java code // Support for the Forte(TM) Peformance Tools collector. // information, please see the libcollect man page. // Method to let libcollector know about a dynamically loaded function. // Because it is weakly bound, the calls become NOP's when the library "Code size exceeds maximum range");
#
endif // !_WINDOWS && !IA64