Re: [PATCH]: Linux for Baget/MIPS (r3k & VME box)

[Dominic Sweetman <dom@algor.co.uk>]

Gleb O. Rajko (rajko@mech.math.msu.su) writes:

> Consider the following scenario. A region (physical page) is swapped (on a 
> backing store), immediately filled by new data, new virtual page is 
> mapped to this region. There is a non-zero probability that old data 
> still lives in I-cache...  

Cache management is needed when you do DMA or the CPU writes
instructions.  But in this case the need for the invalidation is quite
unconnected with the re-mapping.