[patch 0/9][RFC] Oprofile JIT support - Second RFC post against oprofile 0.9.2

[Jens W. <jen...@de...>]

Hello,

This is the second RFC patch for oprofile JIT support. 
It is still against 0.9.2. The overall design is described
in my first post RFC post:
http://www.nabble.com/-PATCH-0-6--RFC--Oprofile-JIT-code-support-t3728729.html

As announced in my previous "status" mail I stepped on two issues:

"Method Unload":

As it turns out, the Sun JVM, as well as the IBM JVM do method unloading and
reuse memory areas (on x86).

In this patch set I added some code to opjitconv which detects the overlaps
and produces an artifical symbol that has all symbol names concatinated by
+ that are within this address range.

I also added support of the method unload JVMTI event and added timestamps
to the data that libopagent writes. There is no support in opjitconv for
this right now. The idea is to add code to opjitconv that it only selects
symbols that are "life" during the sample session.

"Adjacent VMA regions":

The idea of creating an ELF for each VMA region has a problem. 
Adjacent VMA regions with code may exist, meaning that a VMA 
starts at the end of another VMA. A single method may lie
on the border and begin in one VMA and end in another VMA.
Usually adjacent VMAs are merged to a big one by the OS, but we 
think that this is not the case when Oprofile is in action, 
because it registers unmap hooks.

I also investigated that with the SUN hot spot client JIT engine
the code is sparsely scattered over the region.

This patch set contains a new opjitconv that produces
a single ELF file from the written JIT dump file. There are multiple
.text sections in the ELF file that are created by some heuristic.

The oprofile daemon code is "experimentally" changed to
produce only a single big VMA region for the annonymous samples.
In consequence, it may make sense to remove the code that does
split annonymous samples in VMA regions at all.


The code is tested on x86/32bit (SUN JVM 1.5, IBM JVM 1.5) and
s390x/64bit (IBM JVM 1.5)

Best,

Jens

--
  Jens Wilke 
  Linux on System z - Application Development Tools 
  phone +49-(0)7031-16-3936 - tl *120-3936 - email jen...@de... 
  IBM Germany Lab, Schoenaicher Str. 220, 71032 Boeblingen