Name: clC74495 Date: 07/19/99


This bug happens only under Solaris 2.7. I have reproduced on
a 2 processor Ultra60 machine.

java -Xrunhprof:cpu=times simple produces the following output:

CPU TIME (ms) BEGIN (total = 0) Fri Jul 16 15:44:54 1999
rank self accum count trace method
   1 NaN% NaN% 0 1 <empty trace>
   2 NaN% NaN% 0 2 java/security/AccessController.getStackAccessControlContext
   3 NaN% NaN% 0 4 java/lang/Compiler.initialize
   4 NaN% NaN% 0 3 java/lang/System.initProperties

Notice all the NaN's .
Diagnosis indicates that the Solaris gethrvtime function used by
sysThreadCPUTime (used by hprof) always returns 0, leading to
the divide by 0 in hprof_trace.c and the NaN error.
(Review ID: 85708)
======================================================================

Name: skT88420 Date: 10/06/99

=20
Ultra60, Solaris 7
jdk1.2.2 and jdk1.3beta

Creating a hprof profile with option cpu=3Dtimes produces a
profile with lots of NaN% values


TEST CASE:
Start a jvm with runhprof:cpu=3Dtimes enabled, for example

/opt/jdk1.3/bin/java -classic -Xrunhprof:cpu=3Dtimes,file=3D/tmp/profile.tx=
t sun.applet.AppletViewer

The end of /tmp/profile.txt looks like this:

CPU TIME (ms) BEGIN (total =3D 0) Sat Oct 2 17:48:29 1999
rank self accum count trace method
   1 NaN% NaN% 0 1 <empty trace>
   2 NaN% NaN% 0 2 java/security/AccessController.getStackAcc=
essCo
ntrolContext
   3 NaN% NaN% 0 3 java/lang/System.initProperties

RELATED BUGS:
4255155
(Review ID: 96063)
======================================================================