* Michael Ellerman <m...@ellerman.id.au> [2019-07-29 22:41:55]:
> >
> > + chip_id = of_get_ibm_chip_id(np);
> > + if (chip_id == -1)
> > + chip_id = of_node_to_nid(np);
> > +
> > of_node_put(np);
> > - return of_get_ibm_chip_id(np);
> > + return chip_id;
> > }
>
> A nid is not a chip-id.
>
Agree that nid is not a chip-id.
> This obviously happens to work for the case you've identified above but
> it's not something I'm happy to merge in general.
>
Okay.
> We could do a similar change in the topology code, but I'd probably like
> it to be restricted to when we're running under PowerVM and there are no
> chip-ids found at all.
>
So for PowerNV case and KVM guest, of_get_ibm_chip_id() always seems to
returns a valid chip-id. Its *only* in the PowerVM case that we are
returning nid as the fallback chip-id.
Do you think checking for OPAL firmware would help?
chip_id = of_get_ibm_chip_id(np);
if (chip_id == -1 && !firmware_has_feature(FW_FEATURE_OPAL))
chip_id = of_node_to_nid(np);
of_node_put(np);
or should we do
int topology_physical_package_id(int cpu)
{
int chip_id = cpu_to_chip_id(cpu)
if (chip_id == -1 && !firmware_has_feature(FW_FEATURE_OPAL))
//Fallback to nid instead of chip-id.
....
return chip_id;
}
> I'm also not clear how it will interact with migration.
>
On migration, this function would be triggered when the cpumasks are getting
updated. So I would expect this to continue working.
Or Am I missing someother migration related quirk?
> cheers
>
The other alternative that I see is
--
Thanks and Regards
Srikar Dronamraju
