Name: cl74495 Date: 03/19/2001


Using JVMPI with the 1.2.2 Classic VM or the 1.3 Classic VM on
an AMD K6-2 processor causes the VM to crash. Note this is very
specific to the processor; the problem does NOT occur on Intel
chips.

  To reproduce, on an AMD K6-2 running NT4.0, execute:

java -Xrunhprof:cpu=times NoSuchClass

where NoSuchClass doesn't exit. It's that easy.

This results in NT popping up a "java.exe Application Error:

The instruction at "0x504fa616" referenced memory at "0x52dca048". The memory could not be "read".

The problem was first reported against JProbe, but can be seen with hprof as well. I don't expect you to fix this, since the Classic VM is basically dead, but I would like a bug number we can refer our customers to when they report it.

Notes:
- Also crashes the Linux version on an AMD K6-2
- Customers report crash occurs on Win2K with K6-2 but not with AMD K7
- I suspect it's related to METHOD_ENTRY/METHOD_EXIT events, because hprof's cpu=samples works, and JProbe works if we use our own instrumentation in place of these events.
- It's not resolved by turning off the JIT
- 1.3.0 Client Hotspot and 1.3.1b17 Client or Server Hotspot do not show the problem.
(Review ID: 118881)
======================================================================