Re: rfc: high-resolution timer framework

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Wed, Jun 25, 2008 at 03:25:40PM +0000, Andrew Doran wrote:
>       +#ifdef LAPIC_HRTIMER
>       +void
>       +lapic_clockintr(void *arg, struct intrframe *frame)
>       +{
> 
> It breaks if the machine has more than one CPU, because clock interrupts
> will only occur on the boot CPU. Presumably it also prevents the lapic from
> being used as a timecounter source, but that's not a big deal. It's also
> unfortunate that it's optional.

This is actually one of the fundamental issues I have with the current
callout and time handling. It is inherently bound to a single CPU
system, without many good reasons.

> One solution it is to always use the i8254 for the clock interrupt, and
> trigger a broadcast IPI to the other CPUs from the i8254 clock ISR before
> calling the MI hardclock().

I think it is much better to decouple the system time from hardclock and
derive it from time counters. This is also a lot more work though.

Joerg