Re: Predictable CPU discovery

[Ryo Shimizu <ryo%nerv.org@localhost>]

I initially overlooked the timing of the vindexsize update. Thanks Jared for pointing that out.

>Alternatively, other ports use uvm_page_recolor(9) when secondary
>CPUs are attached:
>
>https://nxr.netbsd.org/source/s?refs=uvm_page_recolor&project=src
>
>For aarch64, something like the attached patch makes sense? (Sorry,
>compile-test only since I don't have big.LITTLE processors...)

I've tested it and it seems to be fine.
I tried to force recolor on BigCluster by returning a dummy (larger) vindexsize,
but it seemed to work.

One more point.

    if (uvm.page_init_done)
       uvm_page_recolor(vindexsize / PAGE_SIZE);

It may be better to use.

Thanks
-- 
ryo shimizu