Jonathan S. Shapiro
shap at eros-os.org
Sun Jan 15 13:27:58 EST 2006
On Sun, 2006-01-15 at 19:06 +0100, Matthieu Lemerre wrote:
> What would be the form of the constraint? Would it be a simple couple
> (address_min, adress_max), or a more complicated scheme?
> Neal originally wanted to express constraints on physical memory
> location by using two words :
> - bits in the first word will indicate bits which have to be on in the
> address of the physical memory eventually choosen
> - bits on in the second word will indicate bits which have to be off in
> the address of the physical memory eventually choosen
Not yet decided. I haven't thought Neal's proposal through, but my
initial thought is that matters are a little more complicated than this.
There are really two constraints to consider:
1. The constraint imposed by the pinset. This is a constraint specified
to the pinset admission control agent so that it can ensure that the
requested limit can be satisfied.
The problem here is that computing a simultaneous satisfaction for many
pinsets is probably easier if the constraint is specified as a (base,
bound, rounding alignment) tuple. The problem that needs to be solved
here is similar to admission control in real-time scheduling.
2. The constraint provided at runtime by the requesting process. This is
a placement preference. Neal's solution would be okay here, but it may
not be ideal given the pre-existing constraint. Once you already know
that you are in a given range, it might be that a modular offset is
I'll talk to Neal to learn what he was trying to accomplish. I'm not
committed to the particular constraint approach that I am proposing. I'm
only committed to the idea that the admission control agent needs to be
able to make sure that all outstanding pinset constraints can be
More information about the coyotos-dev