Subject: Re: largepage success
To: None <firstname.lastname@example.org>
From: Alasdair Baird <email@example.com>
Date: 09/12/2000 20:07:08
> > Buffer cache pages are problematic -- pagemove().
Talking of pagemove (on the i386 at least), does it really have to
invalidate both the source and the destination page addresses? In
theory it should only need to invalidate the TLB for the destination,
allowing the source one to LRU (or whatever) out of the processor.
TLB shooting isn't cheap. I don't think both were invalidated
prior to the uvm changes, along with which came the current pagemove.