Re: [Xen-devel] [RFC PATCH v2 04/25] x86: NUMA: Add accessors for acpi_numa, numa_off and numa_fake variables

[Julien Grall <julien.grall@xxxxxxx>]

On 25/04/17 15:54, Vijay Kilari wrote:
> On Tue, Apr 25, 2017 at 5:58 PM, Julien Grall <julien.grall@xxxxxxx> wrote:
> > > > > By setting 1, we are enabling acpi_numa by default. If not enabled, the
> > > > > below
> > > > > call has check srat_disabled() before proceeding fails.
> > > >
> > > >
> > > > My understanding is on x86 acpi_numa is disabled by default and will be
> > > > enabled if they are able to parse the SRAT. So why are you changing the
> > > > behavior for x86?
> > >
> > > acpi_numa = 0 means it is enabled by default on x86.
> >
> > In acpi_scan_nodes:
> >
> > if (acpi_numa <= 0)
> >   return -1;
> >
> > So it does not seem that 0 means enabled.
> IMO, In x86
>          -1 means disabled
>           0 enabled but not numa initialized
>           1 enabled and numa initialized.
>
> I clubbed 0 & 1.
From your description 0 and 1 have different meaning, so I don't see 
how you can merge them that easily without any explanation.
Anyway, I will leave x86 maintainers give their opinion here.

> I was referring to below code in x86 where in acpi_numa = 0 means
> srat_disabled() returns false. Which means acpi_numa is enabled implicitly.
>
> int srat_disabled(void)
> {
>       return numa_off || acpi_numa < 0;
> }
>
> When I changed acpi_numa to bool, it is initialized to 1 and changed
> below code.
>
> bool srat_disabled(void)
> {
>     return numa_off || acpi_numa == 0;
> }
>
> Also this srat_disabed() is used in acpi_numa_memory_affinity_init which is
> called from acpi_numa_init() before calling acpi_scan_nodes().


--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel