Name: elR10090 Date: 09/10/2003



Tiger-b18, all platforms, intermittently crashes while executing
new NSK tests against the new Thread.getAllStackTraces() feature.
Or, saying exactly, there are variable crash scenarios observed
if VM runs the tests with -Xmixed or -Xcomp mode; and running the
tests with -Xint mode looks even more complicated -- instead of
crash, sometimes NPE is observed like the following:

    Exception in thread "main" java.lang.NullPointerException
        at java.util.HashMap.hash(HashMap.java:264)
        at java.util.HashMap.put(HashMap.java:382)
        at java.lang.Thread.getAllStackTraces(Thread.java:1363)
        at nsk.stress.strace.strace002.makeSnapshot(strace002.java:126)
        at nsk.stress.strace.strace002.run(strace002.java:69)
        at nsk.stress.strace.strace002.main(strace002.java:51)

I'm providing a doit.sh stuff for GammaBase to reproduce the fault;
see it under the directory:

    /net/jano.sfbay/export/disk20/GammaBase/Bugs/${THIS_BUG_ID}

Use it like:

    time sh doit.sh $JAVA_HOME [ $OPTIONS ]

E.g.:

    JAVA_HOME=/java/re/jdk/1.5.0/latest/binaries/solaris-sparcv9
    time sh doit.sh $JAVA_HOME -d64 -Xint

The script runs the test "strace002" many times; until VM crashes,
or until NPE is thrown. A number of iterations before the fault
depends on a tested platform. The following table shows how long
it took me to see a fault for different platforms (-Xint mode):


              | Client VM | Server VM
              | 1xCPU 2(4)xCPU | 1xCPU 2(4)xCPU
--------------+------------+------------+------------+-----------
Sol. SPARC v9 | n/a | n/a | 40m, crash | 33m, crash
Sol. SPARC | 7h, crash | 20m, crash | 50m, crash | 7m, crash
Sol. x86 | 30m, NPE | 20m, crash | 2h, crash | 2m, NPE
Linux x86 | 1h, crash | - | 15m, NPE | -
Windows x86 | 15h, exit | - | 10m, crash | -

Footnotes:
(1) I didn't have a chance to test against multi-CPU Linux nor
    Windows systems.
(2) Windows -client silently exited after 15 hours of iterations;
    this looks like a different bug, but I can't understand what
    is it.

I saw various kinds of VM crash; particularly, Linux reported
problematic frame is in java.util.HashMap.hash() method. Compare
this to the NPE trace displayed above. It looks like this bug is
caused by something happening inside the HashMap.hash() method.

Following is the list of computers I've used:
(a) SPARC, 1xCPU, 450MHz, 512Mb, Solaris 8
(b) SPARC, 4xCPU , 512Mb, Solaris 7 (don't know MHz)
(c) P-III, 1xCPU, 866MHz, 512Mb, Solaris 8
(d) P-III, 2xCPU, 600MHz, 512Mb, Solaris 8
(e) P-III, 1xCPU, 866MHz, 512Mb, Linux RedHat 8.0
(f) P-II , 1xCPU, 350MHz, 384Mb, Windows 2000 Pro


Well. Following is the complete list of the tests which I've found
affected by this bug. I'm enumerating the tests here to help with
evaluating further regular QA runs of these tests against Tiger:

    nsk/stress/strace/strace001
    nsk/stress/strace/strace002
    nsk/stress/strace/strace005
    nsk/stress/strace/strace006
    nsk/stress/strace/strace008

======================================================================

Name: elR10090 Date: 09/15/2003


This bug also affects the tests:

    nsk/stress/strace/strace003
    nsk/stress/strace/strace004


======================================================================