Re: OProfile 0.6 released

[Hiro Y. <hyo...@mi...>]

Thanks for your quick reply.

> > Does oprofile support PEBS (precise event based sampling)
> > of Pentium 4/Intel Xeon processors?
> 
> No.
> 
> > If no, do you have any plan to support it?
> 
> I don't have any such plans ... I can't comment on what other people may
> or may not be planning :)
> 
> But there's no sign of it happening

Let me advertise my tool which supports PEBS. :-)
http://sourceforge.jp/projects/hardmeter

The nice thing of the PEBS are that you can sample
the register(s) which have address of memory which are
accessed.

example of PEBS records
#eflags  liner_ip   eax      ebx      ecx      edx      esi      edi      ebp      esp
00000206 c0142bba 00000b00 ef11ed00 00000b00 f63c8000 0000a9e1 ef11ed80 00000000 f63c9e74
00000202 c0142bba 00000b00 ed775100 00000b00 f63c8000 00009657 ed775180 00000000 f63c9e74
00000202 c0142bba 00000b00 ebef9200 00000b00 f63c8000 000082ce ebef9280 00000000 f63c9e74

the disassembled source code of c0142bba (invalidate_bdev)
is the following
                        bh_next = bh->b_next_free;

                        /* Another device? */
                        if (bh->b_dev != dev)
c0142bb3:       8b 44 24 08             mov    0x8(%esp,1),%eax
c0142bb7:       8b 7b 20                mov    0x20(%ebx),%edi
c0142bba:       66 39 43 0c             cmp    %ax,0xc(%ebx)
...

The l1_cache_miss is occured when the address 0xef11ed00 is
accessed and so on.

What do you think? Is it nice?

Regards,
  Hiro
--
Hiro Yoshioka/CTO, Miracle Linux
mailto:hyo...@mi...
http://sourceforge.jp/projects/hardmeter