On 09.10.23 15:27, Ilya Maximets wrote:
> On 10/8/23 10:00, Jakob Meng wrote:
>>
>> On 06.10.23 20:00, Ilya Maximets wrote:
>>> On 10/6/23 09:49, Jakob Meng wrote:
>>>> On 05.10.23 21:08, Ilya Maximets wrote:
>>>>> On 10/4/23 14:21, jm...@redhat.com wrote:
>>>>>> diff --git a/tests/pmd.at b/tests/pmd.at
>>>>>> index 7bdaca9e7..fb838286b 100644
>>>>>> --- a/tests/pmd.at
>>>>>> +++ b/tests/pmd.at
>>>>>> @@ -93,11 +93,11 @@ pmd thread numa_id <cleared> core_id <cleared>:
>>>>>> overhead: NOT AVAIL
>>>>>> ])
>>>>>>
>>>>>> -AT_CHECK([ovs-appctl dpif/show | sed
>>>>>> 's/\(tx_queues=\)[[0-9]]*/\1<cleared>/g'], [0], [dnl
>>>>>> +AT_CHECK([ovs-appctl dpif/show | sed
>>>>>> 's/\(numa_id=\)[[0-9]]*/\1<cleared>/g'], [0], [dnl
>>>>> Hi, Jakub. Could you explain what is going on here?
>>>>>
>>>>> I'm a bit confused. We cear out the number of tx_queues in some
>>>>> tests, because it is determined dynamically in the datapath. But
>>>>> why clearing the numa id instead? The numa id is not changing if
>>>>> not explicitly set in the port configuration, IIRC.
>>>>>
>>>>> Best regards, Ilya Maximets.
>>>> True, in pmd.at we use "--dummy-numa" to set the numa sockets and cores.
>>>> For most cases it is simply 0,0,0,0 or else numa_id is configured so
>>>> numa_id should be known to us (esp. because netdev-dummy simply stores the
>>>> numa id internally). I cleared numa_id because the surrounding code does
>>>> as well and for the sake of testing that the option is printed, it is
>>>> enough. But if you prefer I will replace <cleared> with specific numa ids
>>>> in the next patch revision.
>>> We only clear numa_id in the PMD thread header but not in the port
>>> configuration. There is no point clearing it.
>> Will change it in next patchset but I am waiting for your full review first
>> ;)
>>> We do clear number of tx queues however, and I'm not sure why you
>>> switched to not doing that anymore. Looks like an unrelated change.
>> configured_tx_queues is no longer returned from get_config() and
>> requested_tx_queues has been renamed to n_txq.
>>
>> pmd.at uses netdev dummy. By default n_txq is initialized to 1 for netdevs
>> [0] and netdev-dummy copies this to requested_n_txq [1] which is then
>> returned as n_txq in netdev dummy's get_config().
>>
>> pmd.at does not change n_txq so we know that n_txq is always 1 and thus we
>> do not need the regex clearing?!
> True. I forgot I removed the implementation of .set_tx_multiq()
> callback from netdev-dummy 7 years ago [1]. :D
>
> So, yes, no clearing required in the current code. It will be
> needed only if we decide to implement this callback again in the
> future.
>
> [1] d66e4a5e7e10 ("netdev-dummy: Add n_txq option.")
7 years ago... 😱 At that time i was still coding in C++ and chasing the dream
of a phd😂
Send patch v5 without clearing *tx_queues nor numa_id:
https://patchwork.ozlabs.org/project/openvswitch/patch/20231011101105.21803-1-jm...@redhat.com/
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
