PTE chaining, kswapd and swapin readahead

[Rik van Riel <H.H.vanRiel@phys.uu.nl>]

Hi,

In the PTE chaining discussion/patches a while ago, I saw
that kswapd was changed in a way that it scanned memory
in physical order instead of walking the pagetables.

This has the advantage of deallocating memory in physically
adjecant chunks, which will be nice while we still have the
primitive buddy allocator we're using now.

However, it will be a major performance bottleneck when we
get around to implementing the zone allocator and swapin
readahead. This is because we don't need physical deallocation
with the zone allocatore and because swapin readahead is just
an awful lot faster when the pages are contiguous in swap.

I write this to let the PTE people (Stephen and Ben) know
that they probably shouldn't remove the pagetable walking
routines from kswapd...

Rik.
+-------------------------------------------------------------------+
| Linux memory management tour guide.        H.H.vanRiel@phys.uu.nl |
| Scouting Vries cubscout leader.      http://www.phys.uu.nl/~riel/ |
+-------------------------------------------------------------------+