Re: [Qemu-devel] [RFC PATCH v3] numa: enable sparse node numbering

[Eduardo Habkost]

On Mon, Jun 30, 2014 at 11:26:10AM -0700, Nishanth Aravamudan wrote:
[...]
> > > -        if (i == nb_numa_nodes) {
> > > +        if (i == max_numa_node) {
> > >              for (i = 0; i < max_cpus; i++) {
> > > -                set_bit(i, numa_info[i % nb_numa_nodes].node_cpu);
> > > +                do {
> > > +                    j = (j + 1) % max_numa_node;
> > > +                } while (!numa_info[j].present);
> > 
> > If you change it from "do { } while" to "while { }", you don't need to
> > initialize j to -1.
> 
> I don't think that's quite as simple as you make it out to be. If you
> use a while() loop, we won't always increment j, which means once we've
> found a present node, we'll always use that node? j here basically
> represents the *last* used nodeid, which we don't want to use again when
> we re-enter the for-loop, we want to use the next present nodeid. It
> seems like the do {} while() does this fine?  I could use a while() if I
> added another increment outside the loop, as follows:
> 
>         if (i == max_numa_nodeid) {
>             for (i = 0, j = 0; i < max_cpus; i++) {
>                 while (!numa_info[j].present) {
>                     j = (j + 1) % (max_numa_nodeid);
>                 }
>                 set_bit(i, numa_info[j].node_cpu);
>                 j = (j + 1) % (max_numa_nodeid);
>             }
>         }
> 
> If you think that is cleaner, I'll use that version.

You are right, and your "do { } while" version looks cleaner than
duplicating the "j = (j + 1) % (max_numa_nodeid)" line.

-- 
Eduardo