Re: [SeaBIOS] SeaBIOS PAE experiment

[Kevin O'Connor]

On Tue, Sep 29, 2015 at 10:19:19PM +0200, Rudolf Marek wrote:
> Hi Kevin,
> > The reason why PAE is interesting (instead of standard i386 paging)
> > is that it allows for 64bit mappings and because one can set it up
> > with just a single level page directory of 2MB pages.  The single
> > level page directory makes maintaining it much easier.
>
> There is a performance penalty if you do it for first 2MB of phys memory on some
> P6/XeonP4 CPUs.  Others could trigger undefined behaviour.
>
> The reason for this are the special regions in the 640KB-1MB covered by various
> MTRRs.  See 11.11.9 Large Page Size Considerations in the Intel manual for the
> details.

Thanks - good catch.  It looks like that would require using 4KB pages
for identity mapping the first 2MB or ram.

-Kevin