tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: A simple cpufreq(9)

On Sat, Sep 24, 2011 at 08:35:11AM +0000, Michael van Elst wrote:
> What is wrong with the "abstraction" of having a number of ordered
> performance states? Hiding the states behind something pretending to
> be a continuum (wether this is a MHz value or a percentage doesn't matter)
> causes confusion. You never know what performance state you selected
> and you start to assume that 100% is twice as fast as 50% even when
> there is no correlation.

Nothing. As I wrote, (an integer) percentage is an ordered scale. I agree:
perhaps not the best one. But you have to satisfy:

        1. Boolean scale (already: ichlpcib(4), piixpcib(4), PowerPC).

        2. Interval scale (usually x86).

        3. Interval scale with nonuniform intervals (can be on x86/ARM).

If I'd have to pick, I would take the first one, as you originally wrote.

- Jukka.


And this assertion holds: a machine-independent implementation that can not
satisfy the six machine-dependent implementations currently in the tree is
inherently and completely broken. Thus, mixing latencies and whatnot to the
soup can be done later, preferably in the MD implementations.

Home | Main Index | Thread Index | Old Index