Re: [RFC 0/3] ethdev: introduce device reinitialisation API

[David Marchand <david.marchand@redhat.com>]

Hello,

On Tue, 23 Dec 2025 at 02:16, Stephen Hemminger
<stephen@networkplumber.org> wrote:
>
> On Fri, 17 Oct 2025 11:11:06 +0000
> Ciara Loftus <ciara.loftus@intel.com> wrote:
>
> > This RFC arose from a discussion [1] around whether a generic API that
> > performs device reset and reinitialisation would be useful. While the
> > net effect of a call to such an API can be achieved with a sequence of
> > calls to existing APIs (reset, configure, rx/tx queue_setup), the
> > benefit of encapsulating this all under one API is twofold:
> > 1. the burden is taken off the user for supplying the device
> > configuration in the call, we use the existing device conf under the
> > hood and only need the port_id as input.
> > 2. it can prevent potential unnecessary/repeat reconfiguration steps.
> > The PMD dev_reinit ops will only perform the necessary reconfiguration
> > steps after the device reset.
> >
> > Patch 1 implements the API, Patch 2 implements one example of the reinit
> > op for the iavf driver and Patch 3 integrates the API into testpmd.
> >
> > [1] https://mails.dpdk.org/archives/dev/2025-October/325386.html
> >
> > Ciara Loftus (3):
> >   ethdev: introduce device reinitialisation API
> >   net/iavf: implement device reinitialisation callback
> >   app/testpmd: support port reinitialisation
>
>
> It is reasonable idea, but unfortunately eth_dev_ops is likely exposed enough
> to applications that it could be an ABI change.

eth_dev_ops is opaque to applications.
We don't guarantee driver ABI.


-- 
David Marchand