Re: [PATCH 4/4][RFC] perf_counter: Allow sharing of output channels

[stephane eranian <eranian@googlemail.com>]

On Wed, Aug 19, 2009 at 2:56 PM, Peter Zijlstra<a.p.zijlstra@chello.nl> wrote:
> On Wed, 2009-08-19 at 22:36 +1000, Paul Mackerras wrote:
>> Peter Zijlstra writes:
>>
>> > Provide the ability to configure a counter to send its output to
>> > another (already existing) counter's output stream.
>>
>> What sort of thing might this be useful for?
>
> Some people complained that its tedious to mmap() for every counter and
> would like to share the mmap() output buffer between counters.
>

> This saves on address space and mlock budget and I guess fd management
> logic.
>
Interesting to see, you seem to have changed your mind on this.
I recall pointing this out in my early comments.

But anyway, here are some more comments:

   - how does this work with the remapped counts?
     Probably only see the count for the target, i.e., output,  event

   - if samples from multiple events end up in the same buffer, how do
I tell them apart,
     i.e., how do I know sample X came from event A, sample X from event B?
     This may be useful to detect patterns.


> As long as you're not mixing counters for different tasks/cpus there
> should be no performance penalty, but even if you do that it might work
> well enough on slow samples/small systems..
>
>> Does this only apply to sampling counters?
>
> Yeah, everything that would otherwise go through the mmap() buffer.
>
> I'm not sure there's anything to be done about the read(2) thing.
>
>