chris@paradox.demon.co.uk said:
> Must be some useful data we could load up into memory though?  Can't
> think  what though, any big kernel data tables? maybe the kernel L1
> tlb table, or  does that change too much/shouldn't be cached?  Just
> feels like such a waste  to pull 16k across that slow bus just to
> chuck it away. 

Well, if there were 16K of constant (read-only) data, we might be able to 
snarfle that instead.  But trying to pull in read/write data would be bad 
news (the comment in the example I posted said that in that case you 
needed to pull in 32K worth *and then flush the cache afterwards*.

R.