[dpdk-dev] [RFC] ethdev: introduce action context APIs

[Andrew Rybchenko]

On 3/17/21 7:10 PM, Bing Zhao wrote:
> Hi Thomas,
> 
>> -----Original Message-----
>> From: Thomas Monjalon <thomas at monjalon.net>
>> Sent: Wednesday, March 17, 2021 4:29 PM
>> To: Bing Zhao <bingz at nvidia.com>
>> Cc: Ori Kam <orika at nvidia.com>; ferruh.yigit at intel.com;
>> andrew.rybchenko at oktetlabs.ru; dev at dpdk.org
>> Subject: Re: [RFC] ethdev: introduce action context APIs
>>
>> External email: Use caution opening links or attachments
>>
>>
>> 17/03/2021 08:59, Bing Zhao:
>>> The new functions rte_flow_action_ctx* that were added will
>> replace
>>> the curret shared functions rte_flow_shared_action_*.
>>>   - rte_flow_shared_action_create
>>>   - rte_flow_shared_action_destroy
>>>   - rte_flow_shared_action_update
>>>   - rte_flow_shared_action_query
>>>   + rte_flow_action_ctx_create
>>>   + rte_flow_action_ctx_destroy
>>>   + rte_flow_action_ctx_update
>>>   + rte_flow_action_ctx_query
>>>
>>> When creating a action context, it could be shared among different
>>> flows or different ports. Or it could also be used by a single
>> flow.
>>> The name "shared" is improper in a sense.
>>
>> Is it the only reason for the change?
>> I better understand "shared" even if it is sometimes not shared.
> 
> Another reason is as written in the description of commit message, currently the update interface use the rte_flow_action as the input parameter. It would limit the ability of updating, since rte_flow_action only have the "conf" w/o any mask or flag bits.
> 
> If only part of the fields need to be updated, it couldn't meet the needs.

Sorry, it is too vague. Context in the RFC looks insufficient
to understand it. Please, provide more details and real-life
examples.