Re: Non power-of-two cache sizes and page colouring

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

> I've got a machine here with 6MB of cache (an Intel E5420), and this
> cache size confuses the page colouring code..  I end up with 96 colours,
> and then we form a mask based on this which is where the problems
> start...  The failure symtoms are UVM saying that the machine is out of
> RAM while vmstat(1) says there's still half the memory available.

In general, ncolors (and its implementation) should be deteremined
by a strategy used on looking up a cache index from a physicall address
on the CPU, otherwise it won't make sense.

BTW, is it really 2*N way set associative?
Isn't it 3*N way set associative?
i.e. isn't it 24 way rather than 16 way?
---
Izumi Tsutsui