Re: [GIT PULL 0/6] perf/ebpf basic integration

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Thu, Oct 29, 2015 at 01:17:32PM +0100, Ingo Molnar escreveu:
> * Arnaldo Carvalho de Melo <acme@kernel.org> wrote:
> > 	Please take a look at the changeset comments, I made notes in most of
> > them, this seems like a nice cutoff point to allow basic testing, by
> > developers, for the very basic integration of perf and ebpf, i.e. we can,
> > having a ready built object file, built with clang, be able to use it as a perf
> > event, passing it via -e/--event, etc.

> > 	Wang has lots more in the queue and I intend to work on them till I get
> > all reviewed/tested/merged, i.e. in the immediate future.

> > 	What do you think? Fair to get his smaller gulp now? Or do you want to
> > get it all the way with the code to get a .c file, build it, etc, that is
> > ready, but I haven't reviewed/tested yet?

> > 	Ah, this is on top of what I sent to you via perf/core yesterday.
 
> I'm fine with this, as long as this bit:
 
> > More work is about to be reviewed, tested and merged that will allow the whole
> > process of going from a .c file to an .o file via clang, etc to be done
> > automagically. (Wang Nan)
 
> ... is treated as the primary interface. Very few people will use object files, so 
> we need to integrate the whole life-time workflow, from instrumentation source 
> code to perf output.

Sure, I'm doing the review/test now, but at the same time, stressing
that you _can_ shortcut all that and, having a suitable .o file, use all
the other perf features with it is also cool and something worth
pointing out.

To _have to_ produce a .o to use this is, of course, bad and shortly
will be done completely automatically in as a streamlined way as we can
conceive.

- Arnaldo