Re: [PATCH net-next RFC 01/13] devlink: Add reload level option to devlink reload command

[Jacob Keller]

On 8/10/2020 9:53 AM, Jakub Kicinski wrote:
> On Sun, 9 Aug 2020 16:21:29 +0300 Moshe Shemesh wrote:
>> Okay, so devlink reload default for mlx5 will include also fw-activate 
>> to align with mlxsw default.
>>
>> Meaning drivers that supports fw-activate will add it to the default.
> 
> No per-driver default.
> 
> Maybe the difference between mlxsw and mlx5 can be simply explained by
> the fact that mlxsw loads firmware from /lib/firmware on every probe
> (more or less).
> 
> It's only natural for a driver which loads FW from disk to load it on
> driver reload.
> 

This seems reasonable to me as long as the drivers document this
behavior in their devlink/<driver>.rst. We shouldn't change existing
behavior. One could argue that this difference in behavior amounts to a
"driver default"... but I agree that we shouldn't enshrine that in the
interface.


>> The flow of devlink reload default on mlx5 will be:
>>
>> If there is FW image pending and live patch is suitable to apply, do 
>> live patch and driver re-initialization.
>>
>> If there is FW image pending but live patch doesn't fit do fw-reset and 
>> driver-initialization.
>>
>> If no FW image pending just do driver-initialization.
> 
> This sounds too complicated. Don't try to guess what the user wants.
> >> I still think I should on top of that add the level option to be
>> selected by the user if he prefers a specific action, so the uAPI would be:
>>
>> devlink dev reload [ netns { PID | NAME | ID } ] [ level { fw-live-patch 
>> | driver-reinit |fw-activate } ]
> 
> I'm all for the level/action.
> 

Yep, same here.

>> But I am still missing something: fw-activate implies that it will 
>> activate a new FW image stored on flash, pending activation. What if the 
>> user wants to reset and reload the FW if no new FW pending ? Should we 
>> add --force option to fw-activate level ?
> 
> Since reload does not check today if anything changed - i.e. if reload
> is actually needed, neither should fw-activate, IMO. I'd expect the
> "--force behavior" to be the default.
> 

Yep. What about if there is HW/FW that can't initiate the fw-activate
reset unless there is a pending update? I think ice firmware might
respond to the "please reset/activate" command with a specific status
code indicating that no update was pending.

I think the simplest solution is to just interpret this as a success.
Alternatively we could report a specific error to inform user that no
activation took place?