> [...] The 'requires rdtsc' for 586+ isnt needed
> except for SMP systems - which are all intel for x86 so the problem doesn't
> arise for that case.
it's needed in latest kernels for the timer interrupt. Note that earlier
kernels basically did dynamic linking of the timer interrupt depending on
the existence of the TSC. We had lots of (maintainance) problems due to
that complexity. The scheduling bits are SMP specific for now, because the
cost of getting cache issues wrong is magnified on SMP ... but if you look
at them, the concept is more or less generic and can be applied to UP too.
it's true that currently we group these features along the lines Intel has
cast into silicon, after all they have invented all these features. We
could provide a generic CONFIG_ framework for all x86 CPU features, and
the 'CPU selection' switch just picks one CPU-specific set of features
applicable to that CPU type, plus compilation options. I dont think that
going back to the messy (and imperfect) time.c solution is the way to go.
Especially as things are not getting simpler, the x86 space is getting
splittered even more these days. I dont think we have a choice, x86 CPUs
_are_ getting more and more different, there is no reason to pretend
otherwise.
-- mingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/