8247533: SA stack walking sometimes fails with sun.jvm.hotspot.debugger.DebuggerException: get_thread_regs failed for a lwp

Reviewed-by: sspitsyn, ysuenaga, dtitov

src/jdk.hotspot.agent/linux/native/libsaproc/LinuxDebuggerLocal.cpp

@@ -414,7 +414,13 @@ JNIEXPORT jlongArray JNICALL Java_sun_jvm_hotspot_debugger_linux_LinuxDebuggerLo
 
   struct ps_prochandle* ph = get_proc_handle(env, this_obj);
   if (get_lwp_regs(ph, lwp_id, &gregs) != true) {
-     THROW_NEW_DEBUGGER_EXCEPTION_("get_thread_regs failed for a lwp", 0);
+    // This is not considered fatal and does happen on occassion, usually with an
+    // ESRCH error. The root cause is not fully understood, but by ignoring this error
+    // and returning NULL, stacking walking code will get null registers and fallback
+    // to using the "last java frame" if setup.
+    fprintf(stdout, "WARNING: getThreadIntegerRegisterSet0: get_lwp_regs failed for lwp (%d)\n", lwp_id);
+    fflush(stdout);
+    return NULL;
   }
 
 #undef NPRGREG

src/jdk.hotspot.agent/linux/native/libsaproc/ps_proc.c

@@ -140,7 +140,8 @@ static bool process_get_lwp_regs(struct ps_prochandle* ph, pid_t pid, struct use
 
 #ifdef PTRACE_GETREGS_REQ
  if (ptrace_getregs(PTRACE_GETREGS_REQ, pid, user, NULL) < 0) {
-   print_debug("ptrace(PTRACE_GETREGS, ...) failed for lwp %d\n", pid);
+   print_debug("ptrace(PTRACE_GETREGS, ...) failed for lwp(%d) errno(%d) \"%s\"\n", pid,
+               errno, strerror(errno));
    return false;
  }
  return true;

src/jdk.hotspot.agent/macosx/native/libsaproc/MacosxDebuggerLocal.m

@@ -600,7 +600,7 @@ Java_sun_jvm_hotspot_debugger_bsd_BsdDebuggerLocal_getThreadIntegerRegisterSet0(
   JNIEnv *env, jobject this_obj,
   jlong thread_id)
 {
-  print_debug("getThreadRegisterSet0 called\n");
+  print_debug("getThreadIntegerRegisterSet0 called\n");
 
   struct ps_prochandle* ph = get_proc_handle(env, this_obj);
   if (ph != NULL && ph->core != NULL) {
@@ -620,7 +620,13 @@ Java_sun_jvm_hotspot_debugger_bsd_BsdDebuggerLocal_getThreadIntegerRegisterSet0(
   result = thread_get_state(tid, HSDB_THREAD_STATE, (thread_state_t)&state, &count);
 
   if (result != KERN_SUCCESS) {
-    print_error("getregs: thread_get_state(%d) failed (%d)\n", tid, result);
+    // This is not considered fatal. Unlike on Linux and Windows, we haven't seen a
+    // failure to get thread registers, but if it were to fail the response should
+    // be the same. By ignoring this error and returning NULL, stacking walking code
+    // will get null registers and fallback to using the "last java frame" if setup.
+    fprintf(stdout, "WARNING: getThreadIntegerRegisterSet0: thread_get_state failed (%d) for thread (%d)\n",
+            result, tid);
+    fflush(stdout);
     return NULL;
   }
 

src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/debugger/bsd/BsdDebuggerLocal.java

@@ -167,7 +167,7 @@ public class BsdDebuggerLocal extends DebuggerBase implements BsdDebugger {
                 } catch (InterruptedException x) {}
              }
              if (lastException != null) {
-                throw new DebuggerException(lastException);
+                throw new DebuggerException(lastException.getMessage(), lastException);
              } else {
                 return task;
              }

src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/debugger/bsd/BsdThread.java

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2002, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2002, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -67,9 +67,13 @@ class BsdThread implements ThreadProxy {
     public ThreadContext getContext() throws IllegalThreadStateException {
         long[] data = debugger.getThreadIntegerRegisterSet(unique_thread_id);
         ThreadContext context = BsdThreadContextFactory.createThreadContext(debugger);
+        // null means we failed to get the register set for some reason. The caller
+        // is responsible for dealing with the set of null registers in that case.
+        if (data != null) {
             for (int i = 0; i < data.length; i++) {
                 context.setRegister(i, data[i]);
             }
+        }
         return context;
     }
 

src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/debugger/linux/LinuxDebuggerLocal.java

@@ -186,7 +186,7 @@ public class LinuxDebuggerLocal extends DebuggerBase implements LinuxDebugger {
                 } catch (InterruptedException x) {}
              }
              if (lastException != null) {
-                throw new DebuggerException(lastException);
+                throw new DebuggerException(lastException.getMessage(), lastException);
              } else {
                 return task;
              }

src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/debugger/linux/LinuxThread.java

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2002, 2018, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2002, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -73,9 +73,13 @@ class LinuxThread implements ThreadProxy {
     public ThreadContext getContext() throws IllegalThreadStateException {
         long[] data = debugger.getThreadIntegerRegisterSet(lwp_id);
         ThreadContext context = LinuxThreadContextFactory.createThreadContext(debugger);
+        // null means we failed to get the register set for some reason. The caller
+        // is responsible for dealing with the set of null registers in that case.
+        if (data != null) {
             for (int i = 0; i < data.length; i++) {
                 context.setRegister(i, data[i]);
             }
+        }
         return context;
     }
 

src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/debugger/windbg/amd64/WindbgAMD64Thread.java

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2005, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -30,9 +30,9 @@ import sun.jvm.hotspot.debugger.windbg.*;
 
 class WindbgAMD64Thread implements ThreadProxy {
   private WindbgDebugger debugger;
-  private long           sysId;
+  private long           sysId; // SystemID for Windows thread, stored in OSThread::_thread_id
   private boolean        gotID;
-  private long           id;
+  private long           id;    // ThreadID for Windows thread,  returned by GetThreadIdBySystemId
 
   // The address argument must be the address of the OSThread::_thread_id
   WindbgAMD64Thread(WindbgDebugger debugger, Address addr) {
@@ -50,9 +50,13 @@ class WindbgAMD64Thread implements ThreadProxy {
   public ThreadContext getContext() throws IllegalThreadStateException {
     long[] data = debugger.getThreadIntegerRegisterSet(getThreadID());
     WindbgAMD64ThreadContext context = new WindbgAMD64ThreadContext(debugger);
+    // null means we failed to get the register set for some reason. The caller
+    // is responsible for dealing with the set of null registers in that case.
+    if (data != null) {
         for (int i = 0; i < data.length; i++) {
             context.setRegister(i, data[i]);
         }
+    }
     return context;
   }
 
@@ -86,6 +90,7 @@ class WindbgAMD64Thread implements ThreadProxy {
   private long getThreadID() {
     if (!gotID) {
        id = debugger.getThreadIdFromSysId(sysId);
+       gotID = true;
     }
 
     return id;

src/jdk.hotspot.agent/windows/native/libsaproc/sawindbg.cpp

@@ -42,6 +42,7 @@
 
 #include <limits.h>
 #include <windows.h>
+#include <inttypes.h>
 
 #define DEBUG_NO_IMPLEMENTATION
 #include <dbgeng.h>
@@ -764,9 +765,16 @@ JNIEXPORT jlong JNICALL Java_sun_jvm_hotspot_debugger_windbg_WindbgDebuggerLocal
   CHECK_EXCEPTION_(0);
 
   ULONG id = 0;
-  COM_VERIFY_OK_(ptrIDebugSystemObjects->GetThreadIdBySystemId((ULONG)sysId, &id),
+  HRESULT hr = ptrIDebugSystemObjects->GetThreadIdBySystemId((ULONG)sysId, &id);
-                 "Windbg Error: GetThreadIdBySystemId failed!", 0);
+  if (hr != S_OK) {
-
+    // This is not considered fatal and does happen on occassion, usually with an
+    // 0x80004002 "No such interface supported". The root cause is not fully understood,
+    // but by ignoring this error and returning NULL, stacking walking code will get
+    // null registers and fallback to using the "last java frame" if setup.
+   printf("WARNING: GetThreadIdBySystemId failed with 0x%x for sysId (%" PRIu64 ")\n",
+           hr, sysId);
+    return -1;
+  }
   return (jlong) id;
 }