Re: [RFC net-next v4 0/9] Add support for per-NAPI config via netlink
From: Stanislav Fomichev
Date: Fri Oct 04 2024 - 12:25:14 EST
On 10/03, Joe Damato wrote:
> On Thu, Oct 03, 2024 at 04:53:13PM -0700, Joe Damato wrote:
> > On Thu, Oct 03, 2024 at 04:29:37PM -0700, Stanislav Fomichev wrote:
> > > On 10/01, Joe Damato wrote:
> >
> > [...]
> >
> > > > 2. This revision seems to work (see below for a full walk through). Is
> > > > this the behavior we want? Am I missing some use case or some
> > > > behavioral thing other folks need?
> > >
> > > The walk through looks good!
> >
> > Thanks for taking a look.
> >
> > > > 3. Re a previous point made by Stanislav regarding "taking over a NAPI
> > > > ID" when the channel count changes: mlx5 seems to call napi_disable
> > > > followed by netif_napi_del for the old queues and then calls
> > > > napi_enable for the new ones. In this RFC, the NAPI ID generation
> > > > is deferred to napi_enable. This means we won't end up with two of
> > > > the same NAPI IDs added to the hash at the same time (I am pretty
> > > > sure).
> > >
> > >
> > > [..]
> > >
> > > > Can we assume all drivers will napi_disable the old queues before
> > > > napi_enable the new ones? If yes, we might not need to worry about
> > > > a NAPI ID takeover function.
> > >
> > > With the explicit driver opt-in via netif_napi_add_config, this
> > > shouldn't matter? When somebody gets to converting the drivers that
> > > don't follow this common pattern they'll have to solve the takeover
> > > part :-)
> >
> > That is true; that's a good point.
>
> Actually, sorry, that isn't strictly true. NAPI ID generation is
> moved for everything to napi_enable; they just are (or are not)
> persisted depending on whether the driver opted in to add_config or
> not.
>
> So, the change does affect all drivers. NAPI IDs won't be generated
> and added to the hash until napi_enable and they will be removed
> from the hash in napi_disable... even if you didn't opt-in to having
> storage.
>
> Opt-ing in to storage via netif_napi_add_config just means that your
> NAPI IDs (and other settings) will be persistent.
>
> Sorry about my confusion when replying earlier.
AFAIA, all control operations (ethtool or similar ones via netlink),
should grab rtnl lock. So as long as both enable/disable happen
under rtnl (and in my mind they should), I don't think there is gonna
be any user-visible side-effects of your change. But I might be wrong,
let's see if others can come up with some corner cases..