Subject: Re: uvm_page_physload
To: Izumi Tsutsui <firstname.lastname@example.org>
From: Jachym Holecek <email@example.com>
Date: 08/25/2006 16:52:14
# Izumi Tsutsui 2006-08-25:
> But I'm not sure whether it's worth to have struct vm_page for
> pages which aren't free. Actually many ports don't register
> memory where kernel itself is loaded (or memory for msgbuf
> which is not managed), for example.
... or memory where console is mapped.
Correct me if I'm wrong -- as long as those pages are mapped outside
VM_MIN_KERNEL_ADDRESS ... VM_MAX_KERNEL_ADDRESS
region it's fine to keep them out of uvm(9) knowledge. Otherwise,
uvm(9) needs to know so that we don't get two mappings for the same
kva by accident (on ibm4xx this is considered a "programming error"
and the behaviour is "undefined" -- I guess other ports will be
Most evbppc ports are buggy ITR, I'm currently cleaning up a fix.