Name: kb87695 Date: 10/09/2000



JVMDI field watch bug

A small testcase is enclosed.

public class Bug {
    static int idx;

    public static void main(String args[]) {
        idx++;
        System.out.println("Access Static Field: " + idx);
    }
}

How to reporduce the bug?

1, java_g -Xdebug -Xnoagent -Xrunjdwp:transport=dt_socket,server=y,address=8000 Bug
2, jdb -attach <host>:8000
3, set the watch of access/modification of static field in jdb: watch all Bug.idx
4, continue, then you will see the assertion failure as below
#
# HotSpot Virtual Machine Error, assertion failure
# Please report this error at
# http://java.sun.com/cgi-bin/bugreport.cgi
#
# assert(has_last_Java_frame(), "must have last_Java_sp() when suspended")
#
# Error ID: /export/home/liangg/pdl/hotspot/src/os_cpu/solaris_sparc/vm/thread_solaris_sparc.hpp, 152 [ Patched ]
#
# Problematic Thread: prio=5 tid=0x114fc58 nid=0x8 runnable
#
assertion failure
assert(has_last_Java_frame(), "must have last_Java_sp() when suspended")

Note: With java but not java_g, it will crash the VM too.

Why it is broken?

According the the JVMDI specification, a field watch should catch the feild
access from both Java code and JNI calls, even the access from the JVMDI user
should trigger the watch. When watching the field modification, the DebuggerThread
(a JavaThread) makes a JNI call to GetStaticIntField. And this is actually considered
as a field access from JNI by the JVMDI, so jni_getField_probe will try to post an
event of field access. But the jvmdi::jni_GetField_probe takes the current thread
which is indeed the DebuggerThread as the first actual parameter. Note, the
DebuggerThread has never set the last_Java_frame or last_Java_sp since it never
makes a JavaCall. When jvmdi::post_field_access_event tries to retrieve the MethodOop
and bcp from the last_frame of the thread, boom, it will hit the assertion.

This bug is easily reproducible on a sparc box. The JDK used to reproduce the bug
is jdk1.3, the VM is HotSpot 2.0 Client VM.
$java -version
java version "1.3beta"
Java HotSpot(TM) Client VM (mixed mode)

###@###.### 2000-10-09

This Bug was verified on Solaris 2.7 with JDK 1.3 FCS (B28) (HotSpot Client)

NOTE: This bug is a SHOWSTOPPER for HP, They are requesting a fix by
           mid November.
-Kevin
###@###.### 2000-10-09

(Review ID: 110269)
======================================================================