[coyotos-dev] Guarded page tables

[Jonathan S. Shapiro]

On Fri, 2007-06-15 at 07:59 -0700, Charles Landau wrote:
> At 9:18 AM -0400 6/15/07, Jonathan S. Shapiro wrote:
> >On Thu, 2007-06-14 at 14:09 -0700, Charles Landau wrote:
> >  > That problem is a result of allowing GPTs to have any l2v, not a
> >>  result of allowing GPTs to have a fan-out of either 8 or 16.
> >
> >Not entirely. In any case I note that you have not replied with a
> >decision procedure.
> 
> I do not have enough information.
> 
> At 4:41 PM -0400 6/14/07, Jonathan S. Shapiro wrote:
> >We need to install a new GPT in order to
> >support a newly valid address
> 
> Who needs to do this? A VCSK? A dynamic loader?

Either, but this is indeed the case whose answer is not obvious. Use
VCSK as the challenge case.

> 
> My general approach for such cases where you can't know in advance 
> what structure is optimal, would be to use only l2v values of the 
> form (log page size + n * (log of the natural arity of the GPT, 4 for 
> Coyotos)) and ignore the extra flexibility. If you need a BG or HA 
> key you might need one extra GPT; so what, we've already agreed we 
> don't know how to make it optimal.

This does indeed appear to be the right answer. The reason is that using
any other outcome causes an irregularity in one place to propagate to
all other places under the same tree.
-- 
Jonathan S. Shapiro, Ph.D.
Managing Director
The EROS Group, LLC