Re: [dpdk-dev] [PATCH] doc: notice for changes in kni structures

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 5/8/2017 10:46 AM, Hemant Agrawal wrote:
> On 5/4/2017 10:20 PM, Ferruh Yigit wrote:
>> On 5/3/2017 12:31 PM, Hemant Agrawal wrote:
>>> Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
>>> ---
>>>  doc/guides/rel_notes/deprecation.rst | 7 +++++++
>>>  1 file changed, 7 insertions(+)
>>>
>>> diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
>>> index a3e7c72..0c1ef2c 100644
>>> --- a/doc/guides/rel_notes/deprecation.rst
>>> +++ b/doc/guides/rel_notes/deprecation.rst
>>> @@ -81,3 +81,10 @@ Deprecation Notices
>>>
>>>    - ``rte_crpytodev_scheduler_mode_get``, replaced by ``rte_cryptodev_scheduler_mode_get``
>>>    - ``rte_crpytodev_scheduler_mode_set``, replaced by ``rte_cryptodev_scheduler_mode_set``
>>> +
>>> +* kni: additional functionality is planned to be added in kni to support mtu, macaddr,
>>> +  gso_size, promiscusity configuration.
>>> +  some of the kni structure will be changed to support additional functionality
>>> +  e.g  ``rte_kni_request`` to support promiscusity`` and mac_addr,
>>
>> rte_kni_request is between KNI library and KNI kernel module, shouldn't
>> be part of API.
>>
>>> +  ``rte_kni_mbu`` to support the configured gso_size,
>>
>> Again,  rte_kni_mbuf should be only concern of KNI kernel module.
>>
>>> +  ``rte_kni_device_info`` and ``rte_kni_conf`` to also support mtu and macaddr.
>>
>> rte_kni_device_info also between KNI library and KNI kernel module.
>>
>> I think deprecation notice not required for above ones.
>>
>> But you KNI patchset updates rte_kni_conf and rte_kni_ops.
>> These are part of KNI API and changing them cause ABI breakage,
>> but if new fields appended in these structs, this will not cause an ABI
>> breakage, and I think that is better to do instead of deprecation
>> notice, what do you think?
> 
> I agree.
>>
>>
>> And apart from above ABI issues,
>> adding new fields to "rte_kni_ops" means DPDK application that use KNI
>> should implement them, right?
> 
> Well, it depend, if the application is interested in this information or 
> not?
> 
>> So this suggest everyone require to set promiscuity of KNI device should
>> implement this.
> 
> yes!
> 
>> Can't we find another way that all can benefit from a common implementation?
> 
> how you want it differently? Any ideas?

Can having default implementations in librte_kni work? Would
applications be doing something different, lets say to set MTU?

> 
> 
>>
>> Thanks,
>> ferruh
>>
> 
>