On Fri, 7 Jun 2024 09:12:31 -0700
Dave Jiang <dave.jiang@xxxxxxxxx> wrote:
> On 6/7/24 7:30 AM, Jonathan Cameron wrote:
> >
> >>>> + if (is_cxl_root(parent_port)) {
> >>>> + ctx->port = parent_port;
> >>>> + cxl_coordinates_combine(ctx->coord, ctx->coord,
> >>>> + dport->coord);
> >>>
> >>> I'm a bit lost in all the levels of iteration so may have missed it.
> >>>
> >>> Do we assume that GP BW (which is the root bridge) is shared across multiple root
> >>> ports on that host bridge if they are both part of the interleave set?
> >>
> >> Do we need to count the number of RPs under a HB and do min(aggregated_RPs_BW, (GP_BW / no of RPs) * affiliated_RPs_in_region)?
> >
> > I'm not 100% sure I understand the question.
> >
> > Taking this again and expanding it another level.
> >
> >
> >
> > Host CPU
> > ______________________________________
> > | |
> > | |
> > | 3 from GP/HMAT | 3 from GP/HMAT
> > _____|_____ _______|______
> > RP RP RP RP
> > 2| |2 2| |2
> > __|__ ___|__ __|___ __|____
> > |1 |1 1| |1 |1 |1 |1 |1
> > EP EP EP EP EP EP EP EP
> >
> > Then your maths
> >
> > aggregated RPs BW is 8
> > (GP_BW/no of RPS) * affliated RPS in region.
> > = (3/2 * 4)
> > = 6
>
> While the result is the same, the math would be this below right?
> min((3/2 * 2), 4) + min((3/2 * 2), 4)
That's better, but I thought your thing above was about RPs in a HB vs
RPs across the whole thing. Hence was trying to align with that.
I'm lost and it's end of Friday. Lets work this out with code.
Jonathan
>
> > Which is correct. So yes, I think that works if we assume everything is balanced.
> > I'm fine with that assumption as that should be the common case.
> >
> >
> > Jonathan
> >
> >
>
