[ntp:hackers] odd behaviour (one more time)

clemens at dwf.com clemens at dwf.com
Tue Sep 9 23:59:57 UTC 2008


> clemens at dwf.com wrote:
> >> Can you tell us what has to change to accommodate different ticks? 
> >> Solaris uses both 100Hz and 1000Hz, settable at boot time. I would 
> >> like to verify that the contestants are correctly scaled in these two 
> >> cases.
> >>
> >>     
> >
> > Actually its a one-liner when configuring the Kernel.
> > Under XConfig, I choose
> >
> > 	Processor type and Features -> Timer Frequency -> 1000 Kz
> >
> > and then rebuild the kernel.
> >
> > The other options were/are 100Hz, 250Hz, and 300Hz.
> >
> >   
> This just says that for the kernel you are talking about (Linux?) it 
> changes the constants by calculating them in relation to the tick 
> frequency. OR maybe not, who knows? But since I am talking about Solaris 
> and I am trying to verify those recalculations, I am afraid that your 
> comment isn't helpful to me.

In fact, the actual config variables set are
        CONFIG_HZ_100
        thru
        CONFIG_HZ_1000
which are set to 0/1 and are unused.
and the variable
        CONFIG_HZ
which is set to 100, ... 1000
this variable is unused except to set the variable
        HZ
but trying to scan on HZ is useless as that combination is used hundreds
(thousands) of places in the kernel as parts of other variables, and trying
to sort that out (when I really want to get some dinner) isnt going to 
happen...

> So the question still stands. What needs to change in the microkernel 
> code to accommodate different tick frequencies?  Is it those things that 
> are calculated in relation to the HZ define in the microkernel reference 
> implementation?

There are several variables in timex.h that appear to be related.
Im sure that Dave could point at the right ones, and even tell us how to
compute them for a given kernel frequency.

-- 
                                        Reg.Clemens
                                        reg at dwf.com




More information about the hackers mailing list