[PATCH v3] examples/l3fwd: dynamic NUMA-aware alloc lcore_conf
fengchengwen
fengchengwen at huawei.com
Mon Nov 2 05:13:06 CET 2026
On 11/1/2026 10:26 PM, Thomas Monjalon wrote:
> 31/10/2026 02:37, fengchengwen:
>> On 10/30/2026 8:59 PM, Thomas Monjalon wrote:
>>> 30/10/2026 11:52, Konstantin Ananyev:
>>>>> BTW: SOCKET_ID_ANY seemed need redefined as RTE_SOCKET_ID_ANY, If there is no
>>>>> objection, I will make the changes in the next independent patch.
>>>>
>>>> Yep, that's probably a good thing to do, and probably 26.11 is a good
>>>> time to do that.
>>>> BTW, I think it was discussed before, but I can't remember why we didn't
>>>> fix it so far.
>>>
>>> 1/ We should not break this API easily as it is used a lot.
>>> If we do the change we need to keep the compat alias for some time.
>>>
>>> 2/ The term SOCKET is used in many places instead of NUMA.
>>> There is a confusion between CPU and memory which deserves to be fixed.
>>>
>>
>> My plan is a 3-patch series:
>> 1. Add RTE_SOCKET_ID_ANY and deprecation documentation.
>> 2. Replace all in-tree SOCKET_ID_ANY with RTE_SOCKET_ID_ANY.
>> 3. Mark SOCKET_ID_ANY with RTE_DEPRECATED after in-tree migration completes,
>> so warnings only hit out-of-tree users while DPDK itself builds cleanly.
>>
>> I will avoid the bigger "socket -> NUMA node" terminology refactor in this series.
>> That would be a massive semantic change with large churn, better handled via separate
>> RFC later.
>>
>> If this plan looks acceptable, I will start the implementation.
>
> My doubt is that while introducing a new API,
> it may be better to apply it in correct places with correct definition,
> avoiding mixing CPU and memory.
OK, I will try to implement it (avoiding mixing CPU and memory)
>
>
More information about the dev
mailing list