> On Wed, Dec 04, 2002 at 10:38:36AM +0100, Kevin D. Kissell wrote:
> >
> > Which version of the 4Kc manual are you looking at? I'm looking
> > at a very recent version of the 4Kc Software User's Manual
> > (version 1.17, dated September 25, 2002), and it only shows
> > Hit_Writeback_D to be invalid for *secondary and teritary*
> > caches, which makes sense, since the 4KSc doesn't have any.
> >
>
> I was looking at rev 1.12, Jan 3, 2001.
>
> Good to know that 4K family does have Hit_WRiteback_D. However,
> since it is "recommanded" instead of "required". Shouldn't we
> still use "Hit_Writeback_Inv_D" just to be on the safe side?
Pardon me, but I thought that you were talking about
hit-writeback-invalidates to begin with. Indeed, I had
thought that we had organized things so that Linux always
did writeback-invalidates and never simple writebacks,
just to be on the safe side, as you say, but tampoline code
is a special case where I can see no possible multiprocessor
coherence issues with failing to invalidate the local Dcache
copy. In any case, it would be100% correct for a pure
"hit writeback" to be a no-op on a write-through cache,
since there is never anything dirty to write back.
Kevin K.
|