Unexpected behaviour with SuspendThread()

Question

Hi,
I'm building a JVMTI agent in C++ that controls the threads that are running in the JVM. My agent uses bytecode injection and JNI to intercept and suspend a thread ( using SuspendThread(jvmtiEnv* env, jthread thread) ), just before it executes particular bytecode instructions.
When I call jvmti->SuspendThread(NULL) (i.e. to suspend the current thread), I have found that although the current thread does suspend before returning to the Java environment, the lines of C++ code that follow the call to SuspendThread() are executed before the thread is suspended.
The JVMTI documentation says that "If the calling thread is specified, this function will not return until some other thread calls ResumeThread". Am I missing something here?
A simplified version of my agent code looks like this:
#include "SuspendTest.h"
#include <jvmti.h>
#include <string.h>
static jvmtiCapabilities capa;
static jvmtiEnv* jvmti_env;
int counter = 1;
* Called via JNI
JNIEXPORT void JNICALL Java_SuspendTest__1suspendMe
(JNIEnv *, jclass) {
     printf("%d: About to suspend thread\n", counter);     
     jvmti_env->SuspendThread(NULL);
     counter++;
     printf("%d: This should not be reached\n", counter);
     counter++;
     printf("%d: This should not be reached either\n", counter);
* Called when the agent is first loaded...
JNIEXPORT jint JNICALL Agent_OnLoad(JavaVM *jvm, char *options, void *reserved)
     jint res = jvm->GetEnv((void **) &jvmti_env, JVMTI_VERSION_1_0);
     if (res != JNI_OK || jvmti_env == NULL)
          printf("Error getting environment: %d", res);
     (void)memset(&capa, 0, sizeof(jvmtiCapabilities));
     capa.can_suspend = 1;
     jvmtiError err = jvmti_env->AddCapabilities(&capa);
     if (err != JVMTI_ERROR_NONE)
          printf("Error setting capabilities: %d", err);
     return JNI_OK;
JNIEXPORT void JNICALL
Agent_OnUnload(JavaVM *vm) {}When the Java_SuspendTest__1suspendMe function is called via JNI, I see on the console:
1: About to suspend thread
2: This should not be reached
3: This should not be reached eitherAny suggestions?
I should probably add that this is:
java version "1.6.0_06"
Java(TM) SE Runtime Environment (build 1.6.0_06-b02)
Java HotSpot(TM) Client VM (build 10.0-b22, mixed mode, sharing)
Linux 2.6.24-16-generic #1 SMP Thu Apr 10 13:23:42 UTC 2008 i686 GNU/Linux
Edited by: RichardBounds on Jun 19, 2008 2:40 AM

Answer 1

HotSpot doesn't suspend threads in native code as this can cause deadlocks. For this example, I believe it will "self-suspends" if you try to re-enter the VM or the native method exits.