On Tue, 22 Mar 2005 at 20:58 +0000, John Levon wrote:
> On Tue, Feb 22, 2005 at 10:20:34AM -0600, Rob Fowler wrote:
> > Regardless of the ultimate scheme for reporting time, transitions
> > in performance state will be valuable performance metrics in themselves
> > and should be profileable.
> This is true, but somewhat of a side issue. There are lots of
> interesting events we could and maybe should be providing support for.
> The big issue is "how do I get reliable profiles on my laptop", and I
> agree with Will we need to look at providing a synthetic event keyed to
> clock frequency for this problem.
> It will also give us the nice side effect of having an immediately
> understandable event that can be programmed as "interrupt every x
> microseconds" or whatever.
We can't do that through something based on HW events except a timer
interrupt clocked independantly. This already exists : it's called timer
int, rate is not variable but who care. Trying to key to clock frequency
something already keyed to another HW event look like a non-sense.