Re: [PATCH] Radix-tree pagecache for 2.5

[David S. Miller]

   From: Momchil Velikov <[e-mail]>
   Date: 01 Feb 2002 11:01:50 +0200
   
   Does cache line bounce (shared somewhere -> exclusive elsewhere) cost
   more that a simple miss (present nowhere -> exclusive somewhere) ?

They are about equal.  For coherency purposes all cpus have to listen
to all the transactions anyways to see if they have a match in their
L2 caches (and thus must provide the data to the requestor).

Perhaps the exclusive somewhere --> exclusive somewhere else is a bit
more expensive because you eat a write port for the cache line move on
the processor providing the data.