update: ppc IRQ blocking.

[Iain Sandoe]

Hi,

Well, much of the info presented in my message yesterday was bogus...

The instrumentation needs improvement... it's missing too many cases where
IRQs are switched on/off behind the back of the measurement (which is making
it very difficult to narrow down the rogue...)

I'm working on it.

The latency measurements *do* indicate a problem... I don't think that is
bogus.. but...

So unless:
(a) you have an idea where there might be a 8 to 10ms hold off in IRQ or
(b) you already have an IRQ blocking measurement set that includes all the
cases (in which case I'd love a copy ;-)

Ignore the post of yesterday...
Iain.

P.S. Jun - do you want a copy of what I've got so far? or wait 'till it's
done... I covered a lot more cases but still need to deal with all the stuff
going on in head.S :-(


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/