Re: [net-next PATCH v3 1/3] act_skbedit: skbedit queue mapping for receive queue

[Saeed Mahameed <saeed@kernel.org>]

On 27 Oct 10:12, Roi Dayan wrote:
>
>
>On 26/10/2022 19:17, Jakub Kicinski wrote:
>> On Wed, 26 Oct 2022 14:40:39 +0300 Roi Dayan wrote:
>>> This patch broke mlx5_core TC offloads.
>>> We have a generic code part going over the enum values and have a list
>>> of action pointers to handle parsing each action without knowing the action.
>>> The list of actions depends on being aligned with the values order of
>>> the enum which I think usually new values should go to the end of the list.
>>> I'm not sure if other code parts are broken from this change but at
>>> least one part is.
>>> New values were always inserted at the end.
>>>
>>> Can you make a fixup patch to move FLOW_ACTION_RX_QUEUE_MAPPING to
>>> the end of the enum list?
>>> i.e. right before NUM_FLOW_ACTIONS.
>>
>> Odd, can you point us to the exact code that got broken?
>> There are no guarantees on ordering of kernel-internal enum
>> and I think it's a bad idea to make such precedent.
>
>
>ok. I were in the thought order is kept.
>
>You can see our usage in drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
>in function parse_tc_actions().
>we loop over the actions and get a struct with function pointers
>that represent the flow action and we use those function pointers
>to parse whats needed without parse_tc_actions() knowing the action.
>
>the function pointers are in drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/act.c
>see static struct mlx5e_tc_act *tc_acts_fdb[NUM_FLOW_ACTIONS].
>each function handling code is in a different file under that sub folder.
>
>if order is not important i guess i'll do a function to return the ops i need
>per enum value.
>please let me know if to continue this road.
>thanks

Order is not guaranteed, let's have a robust solution.
You can define an explicit static mapping in the driver, not in a for loop.