[parisc-linux] Progress - Update

Frank Rowand frank_rowand@hp.com
Wed, 24 Nov 1999 14:38:49 -0800


Philipp Rumpf wrote:
> 
> >| are you up-to-date with the cvs tree (I have no idea which caches the 735
> >| has, but cache flushes won't harm (well, they do harm performance)) ?
> >
> > The cache flush instructions are architected to cause a processor to memory
> > transfer iff the cache line is dirty, while an instruction is executed,
> > unless a dirty cache line is referenced, memory bandwidth is not consumed.
> 
> At the current rate of cache flushes (way way too high), the main factor wrt
> performance is indeed the execution of the instructions.
> 
> As soon as we've implemented page colouring (which turns out to be really very
> close to large page support), I expect us to get along with very very few
> cache flushes - basically I hope we can avoid them completely.
> 
>         Philipp Rumpf

No such luck.  You'll need cache flushing in drivers for non-coherent IO.

-Frank