Re: [PATCH -tip tracing/kprobes 0/9] tracing/kprobes, perf: perf probe and kprobe-tracer bugfixes

[Masami Hiramatsu <mhiramat@redhat.com>]

Hi Ashwin,

Ashwin Chaugule wrote:
> Masami,
>            I really like your idea with kprobes ! I havent yet run
> this stuff on my machine, but I've been planning to utilize kprobes
> with perfevents too. Please let me know if the following functionality
> is already possible with your patches.

Thanks! I'm happy to here that you are interested in.

> The basic idea I had was to be able to profile any function in the
> kernel using kprobes and use the perfevents framework to monitor
> things like incorrect branch predictions, cache misses etc. for the
> scope of that function. That way, we can fine tune kernel functions
> that are in the hot path of kernel control flow, using gcc tricks,
> inline assembly, or even architecture specific tricks.

Actually, perf-probe can add tracepoint-like events. It will allow you
to probe function inside by both C-source-line level and address level.
perf-probe already supports inline function too.

Currently, even its syntax is unstable, but you can add an event inside
some function which is at cfile.c line 100, as below;

 perf probe -P 'p:probe1 cfile.c:100'

and you can trace it by using perf record as same as other tracepoints

 perf record -f -e kprobes:myprobe:record -F 1 -a ls


> Alternately, I think even dynamic trace could provide similar insight
> with perfevents ? If none of this is already done, I plan to work on
> this in my spare time and would like to team up with anyone else
> interested.

Yes, it's done as I said above.
But it still has long TODO list, including support type of arguments,
arrays, fields of structures and so on (of course, defining useful
syntax too). So any comments and contributes are welcome :-)

Thank you,

-- 
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America), Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com