[RFC PATCH v4 3/4] arm64:thunder: Add initial dts for Cavium's Thunder SoC in 2 Node topology.

Ganapatrao Kulkarni gpkulkarni at gmail.com
Tue Feb 24 00:52:29 PST 2015


Hi Arnd,

On Fri, Jan 23, 2015 at 6:07 PM, Ganapatrao Kulkarni
<gpkulkarni at gmail.com> wrote:
> Hi Arnd,
>
> On Thu, Jan 22, 2015 at 11:56 PM, Arnd Bergmann <arnd at arndb.de> wrote:
>> On Thursday 22 January 2015 17:47:13 Arnd Bergmann wrote:
>>>
>>> This seems wrong still: The clusters and cores do not have unique
>>> numbers. I believe the code will not work correctly, and it won't
>>> be compliant with the binding from patch 2.
>>>
>>> I think the right way here would be to use
>>>
>>>         arm,associativity = <0 2 47>;
>>>
>>> for cpu at 20f, and
>>>
>>>         arm,associativity = <1 3 48>;
>>>
>>> for cpu at 10000. Your previous version used the numbers from
>>> the reg property, which should be fine as well if that helps:
>>>
>>>
>>>         arm,associativity = <0x0 0x200 0x20f>;
>>>
>>>         arm,associativity = <0x10000 0x10000 0x10000>;
>>>
>>> which should have the same effect as above, as long as the code
>>> can handle the numbers not being consecutive.
>>>
>>>
>>
>> Upon further consideration, I think your patch is correct after
>> all, but let me check again on PowerPC machines.
> i have removed, board id which was in previous patch, to keep the
> associativity aligned to mpidr, ie. socket,cluster id and core id.
> both previous and current mappings holds good for our design.
> our topology is 2 sockets(aff2=0and 1) , each having 3
> clusters(aff1=0to2)and each cluster having 16 cores(aff0:0to15)
> in associativity property, for our case, only first id(socket id) is
> mapped to numa, others are mentioned only to describe the topology.
>>
>>         Arnd
any further comments, how we go about these patches?
please suggest.

> thanks
> Ganapat

thanks
Ganapat



More information about the linux-arm-kernel mailing list