oprofile kernel driver for perf-event backend

[Robert R. <rob...@am...>]

On 01.06.11 09:45:52, Maynard Johnson wrote:

> Robert, the oprofile kernel driver now uses the perf backend?  I
> must have missed that patch series.  Was it posted to the
> oprofile-list?  If so, on what date?  At a high level, what was the
> nature of this change?  I remember there being a discussion of this
> at one time and there being some opposition to the idea of porting
> the oprofile kernel driver to perf_events -- where those opposed
> felt it would be best to simply re-write the frontend to use
> perf_events directly.  Was there some consensus reached that I
> didn't hear about?

It depends on the architecture, for arm and sh it is. It was
originally implemented for arm (v2.6.35):

 http://git.kernel.org/?p=linux/kernel/git/rric/oprofile.git;a=commitdiff;h=8c1fc96f6fd1f361428ba805103af0d0eee65179

and was abstracted (v2.6.37):

 http://git.kernel.org/?p=linux/kernel/git/rric/oprofile.git;a=commitdiff;h=3d90a00763b51e1db344a7430c966be723b67a29

to switch sh to it too (v2.6.37):

 http://git.kernel.org/?p=linux/kernel/git/rric/oprofile.git;a=commitdiff;h=86c8c04792f152c5469023885510140dd34817bc

Re-writing the frontend would be still the best, but so far there
isn't one. The changes were basically driven by the aim to merge
in-kernel profiling code which reduces maintenance efforts and
provides oprofile support (almost) out of the box for systems
supporting perf. Unforunately this went only through the architectural
kernel mailing lists and lkml.

-Robert

-- 
Advanced Micro Devices, Inc.
Operating System Research Center