[PATCH] of: return NUMA_NO_NODE from fallback of_node_to_nid()

[Grant Likely]

On Mon, 13 Apr 2015 11:49:31 -0500
, Rob Herring <robherring2 at gmail.com>
 wrote:
> On Mon, Apr 13, 2015 at 8:38 AM, Konstantin Khlebnikov
> <khlebnikov at yandex-team.ru> wrote:
> > On 13.04.2015 16:22, Rob Herring wrote:
> >>
> >> On Wed, Apr 8, 2015 at 11:59 AM, Konstantin Khlebnikov
> >> <khlebnikov at yandex-team.ru> wrote:
> >>>
> >>> Node 0 might be offline as well as any other numa node,
> >>> in this case kernel cannot handle memory allocation and crashes.
> >>>
> >>> Signed-off-by: Konstantin Khlebnikov <khlebnikov at yandex-team.ru>
> >>> Fixes: 0c3f061c195c ("of: implement of_node_to_nid as a weak function")
> >>> ---
> >>>   drivers/of/base.c  |    2 +-
> >>>   include/linux/of.h |    5 ++++-
> >>>   2 files changed, 5 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/of/base.c b/drivers/of/base.c
> >>> index 8f165b112e03..51f4bd16e613 100644
> >>> --- a/drivers/of/base.c
> >>> +++ b/drivers/of/base.c
> >>> @@ -89,7 +89,7 @@ EXPORT_SYMBOL(of_n_size_cells);
> >>>   #ifdef CONFIG_NUMA
> >>>   int __weak of_node_to_nid(struct device_node *np)
> >>>   {
> >>> -       return numa_node_id();
> >>> +       return NUMA_NO_NODE;
> >>
> >>
> >> This is going to break any NUMA machine that enables OF and expects
> >> the weak function to work.
> >
> >
> > Why? NUMA_NO_NODE == -1 -- this's standard "no-affinity" signal.
> > As I see powerpc/sparc versions of of_node_to_nid returns -1 if they
> > cannot find out which node should be used.
> 
> Ah, I was thinking those platforms were relying on the default
> implementation. I guess any real NUMA support is going to need to
> override this function. The arm64 patch series does that as well. We
> need to be sure this change is correct for metag which appears to be
> the only other OF enabled platform with NUMA support.
> 
> In that case, then there is little reason to keep the inline and we
> can just always enable the weak function (with your change). It is
> slightly less optimal, but the few callers hardly appear to be hot
> paths.

Sounds like you're in agreement with this patch then? Shall I apply it?

g.