Re: [RFC PATCH 1/3] ethdev: add set link state on close API

[Stephen Hemminger <stephen@networkplumber.org>]

On Wed, 3 Sep 2025 09:22:49 +0000
"Loftus, Ciara" <ciara.loftus@intel.com> wrote:

> > 
> > Hi Ciara,
> > 
> > On Fri, 29 Aug 2025, Ciara Loftus wrote:
> >   
> > > Allow the user to configure the state a link should be in after
> > > the device has been stopped/closed. Make this configurable
> > > through the new experimental rte_eth_dev_set_link_state_on_close
> > > API. Three states are allowed:
> > > 1. down: bring (or keep) the link down
> > > 2. up: bring (or keep) the link up
> > > 3. initial: restore the link to the state it was in when the device was
> > > started.
> > >  
> > 
> > Perhaps it pays to name it 'status' instead of 'state', as the latter is not
> > confined to just 'up/down' but also includes the speed, technology, FEC, etc.  
> 
> Sure, I think this would be a logical change.
> 
> > 
> > As for the three cases:
> > 
> > (3)
> > When the driver gets probed, it may discover the link being in some state that
> > the previous driver left us, (A). Then if the user does not invoke any APIs to
> > tweak the link settings and proceeds straight to 'rte_eth_dev_start', the link
> > will be in some best-effort state (B). Otherwise it will be in other state, (C).
> > 
> > Which one is implied by 'when the device was started'?  
> 
> I think what we want is (A) - the state the previous driver left us in. I will
> look to improve the descriptions if there are further revisions.
> 
> > 
> > (2)
> > Does 'keep' describe the link being up prior to 'close'? If so, then if the
> > link is not 'up' prior to 'close', this commands to bring it up, which may
> > not be possible if the link partner is gone or the cable is unplugged.  
> 
> Yes, I guess this is conditional on external factors such as link partner etc.
> The documentation can be changed to reflect this. "Try" to bring the
> link up.
> 
> > 
> > (1)
> > This is OK. If the driver is capable to force link down, in electrical idle,
> > it will do that. If it can't do this, the whole API won't be supported.
> > 
> > Thank you.  
> 
> Thank you for your feedback.
> 
> Ciara

I would prefer that:
  - the link state was the same across all drivers.
  - the link state on close was documented.
  - the link state on close followed what BSD and Linux do.

It might mean that a more complete set of link states was supported.
See Linux operstate.