Re: [PATCH 02/16] perf: Add ability to attach registers dump to sample

[Frederic Weisbecker <fweisbec@gmail.com>]

On Thu, May 24, 2012 at 01:52:06PM +0200, Jiri Olsa wrote:
> On Thu, May 24, 2012 at 12:42:22PM +0200, Peter Zijlstra wrote:
> > On Thu, 2012-05-24 at 12:06 +0200, Stephane Eranian wrote:
> > 
> > > > What are we doing here and why?
> > > >
> > > I think this is related to a discusion we had earlier about which
> > > machine state you want
> > > to sample.
> > > 
> > > There are 3 possible machine states:
> > >   1- user level (even when sample is in kernel AND assuming you did
> > > not hit a kernel only thread)
> > >   2- interrupted state (@ PMU interrupt)
> > >   3- precise state (state captured by PEBS on Intel, for instance)
> > > 
> > > Jiri is only interested in 1/. I am interested in the other two as well.
> > > 
> > > Question: is there a situation where we could need more than one machine
> > > state per sample?
> > 
> > Well, IIRC you always wanted both 2 and 3 at the same time to compute
> > skid, thus:
> > 
> > > If not, then a single bitmask is enough.
> > 
> > Indeed, so then we get to multiple bitmasks and unless you want to be
> > restricted to the same bitmap for all these types this setup won't
> > actually work.
> 
> My intention was to make this general. I could just add
> bitmask for each type (user regs mask for now) but I wanted
> to be consistent with other SAMPLE_* stuff..
> 
> So current patch adds PERF_SAMPLE_REGS sample_type bit.
> Once it is set, the 'sample_regs' value is checked for what
> type of registers you want for sample.
> 
> Each type then has separate bitmask in case you want different
> registers for each type. Allowing whatever combination of regs dump
> being added to the sample, since it seems there's no firm
> decision on what combination might be needed.
> 
> Sure we can make the same with bitmasks for each regs type,
> and check the presence in sample by bitmask being not empty.

I believe we should allow the record of user regs and precise/irq
as well in the same time. Unless we get some real proof that both
will never be used at the same time. I really don't want us to be
limited in the future for ABI reasons.