> > This patch gets rid of a major annoyance:  it stops the "use NO_HZ!"
> > suggestions on x86_64 and other platforms which can't support it.
>
> Hi,
>
> I both agree and disagree with this. x86_64 does support NO_HZ, you 
> just need a 2.6.24 kernel or the -hrt patchkit. So I agree the current 
> message isn't correct.... but I don't think just making it go away is 
> right either; PowerTOP should suggest getting a tickless kernel in 
> that case.

Are you desupporting non-x86 hardware then?  Because even ARM doesn't
have very strong tickless support yet.  And they've had access to that
platform level support for several years now.  (Using the original
code, NO_IDLE_HZ, which is still available on platforms which haven't
switched to GENERIC_CLOCKEVENTS.  I'm not clear if there are still any
in-kernel users for that code.)

Suggesting getting a tickless kernel would be a fine thing to put
in the "else" branch.  I figured if anyone really cared, they could
tweak this patch... and meanwhile, powertop stops spewing garbage
on my x86_64 box!

Of course, not many end-users are going to go install those HRT patches;
or, on other platforms, replace the existing timer code with new stuff
they write.  This area is a lot trickier than the casual hacker will
want to touch; it's not just x86 that has timer-related flakiness.  ;)


> > This lets the other messages -- the ones which could actually do some
> > good on this system!! -- be seen.  Else, it seems like the impossible
> > NO_HZ suggestion is the only one ever visible.
>
> hmm that would be a bug; in principle the messages should rotate (but 
> with a priority bias); it could be the priorities are just off...

In practice, I think about 90% of the messages are the pointless NO_HZ
ones.  Certainly I've had to wait several minutes just to see *one* of
the actually-useful messages appear.

- Dave

_______________________________________________
Power mailing list
[email protected]
http://www.bughost.org/mailman/listinfo/power

Reply via email to