[RFC PATCH] dt:numa: adding numa node mapping for memory nodes.

[Ganapatrao Kulkarni]

On Thu, Sep 18, 2014 at 4:42 AM, Zi Shen Lim <zlim.lnx at gmail.com> wrote:
> On Wed, Sep 17, 2014 at 2:48 PM, Nathan Lynch <Nathan_Lynch at mentor.com> wrote:
>> On 09/17/2014 03:56 AM, Ganapatrao Kulkarni wrote:
>>> From: Ganapatrao Kulkarni <ganapatrao.kulkarni at cavium.com>
>>>
>>> This patch adds property "nid" to memory node to provide the memory range to
>>> numa node id mapping.
>>>
>>> Signed-off-by: Ganapatrao Kulkarni <ganapatrao.kulkarni at cavium.com>
>>>
>>> ---
>>>  Documentation/devicetree/bindings/numa.txt | 58 ++++++++++++++++++++++++++++++
>>>  1 file changed, 58 insertions(+)
>>>  create mode 100644 Documentation/devicetree/bindings/numa.txt
>>>
>>> diff --git a/Documentation/devicetree/bindings/numa.txt b/Documentation/devicetree/bindings/numa.txt
>>> new file mode 100644
>>> index 0000000..c4a94f2
>>> --- /dev/null
>>> +++ b/Documentation/devicetree/bindings/numa.txt
>>> @@ -0,0 +1,58 @@
>>> +======================================================
>>> +numa id binding description
>>> +======================================================
>>> +
>>> +======================================================
>>> +1 - Introduction
>>> +======================================================
>>> +The device node  property "nid(numa node id)" can be added to memory
>>> +device node to map the range of memory addresses as defined in property "reg".
>>> +The property "nid" maps the memory range to the numa node id, which is used to
>>> +find the local and remory pages on numa aware systems.
>>
>> "Local" and "remote" memory are notions that relate to some other
>> resource -- typically a CPU, but also I/O resources on some systems.  It
>> seems to me that a useful NUMA binding would at least specify a "nid"
>> property, or something like it, for both cpu and memory nodes.  But this
>> document speaks only of memory nodes.
IMO, nid can be extended to cpu node also if the arch dont have any
other way to do mapping, in fact this can be scaled to any node.
for ARM arch, cpu mapping can be fetched from topology parsing and
more over cpu mapping is required much later when secondary cpus comes
online.
i have added for memory, since this is parsed very early in the boot
sequence and there is no support for dt based map at this moment(at
least i am not aware off).
numa distance matrix can also be defined, if the architecture has
variable node distance other than just local/remote.
>>
>> As Kumar said, the device tree on powerpc server systems already has
>> properties that express NUMA information.  If you can get hold of a copy
>> of the PAPR (not ePAPR) from power.org, refer to the description of
>> "ibm,associativity" and related properties.  I recall that it's a bit
>> more complex than this proposal, though.
>
> I'm not able to find a link to the actual PAPR (not ePAPR)
> specification. Anyone has a linky?
>
> I did find some code in tree, but couldn't find the bindings in
> Documentation/devicetree/.
yes, there is nothing in Documentation, if PPC are using since several
years, then i humbly request to publish the documentation for numa
bindings.
No point in reinventing the wheel. Also i did not find any spec named
PAPR, is it open spec, available to all like ePAPR?

>
> Seems like we'd care about form1?
> -----8<-----
> commit 41eab6f88f24124df89e38067b3766b7bef06ddb
> Author: Anton Blanchard <anton at samba.org>
> Date:   Sun May 16 20:22:31 2010 +0000
>
>     powerpc/numa: Use form 1 affinity to setup node distance
>
>     Form 1 affinity allows multiple entries in
> ibm,associativity-reference-points
>     which represent affinity domains in decreasing order of importance. The
>     Linux concept of a node is always the first entry, but using the other
>     values as an input to node_distance() allows the memory allocator to make
>     better decisions on which node to go first when local memory has been
>     exhausted.
>
>     We keep things simple and create an array indexed by NUMA node, capped at
>     4 entries. Each time we lookup an associativity property we initialise
>     the array which is overkill, but since we should only hit this path during
>     boot it didn't seem worth adding a per node valid bit.
>
>     Signed-off-by: Anton Blanchard <anton at samba.org>
>     Signed-off-by: Benjamin Herrenschmidt <benh at kernel.crashing.org>
> ----->8-----
>
>
>>
>> _______________________________________________
>> linux-arm-kernel mailing list
>> linux-arm-kernel at lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel