Re: [PATCH net-next v3 1/3] net: rss: Require indir table size or hash key size to support RSS

[Jakub Kicinski <kuba@kernel.org>]

On Thu, 8 Oct 2026 11:04:44 +0200 Maxime Chevallier wrote:
> >> Yes it does, that's the intended behaviour here. RSS bits are all optional,
> >> but does it make sense to report hashing params when the indirection table
> >> is size 0 and key len is 0 ?  
> > 
> > Not sure if there's an implementation in the tree but 
> > it's possible to implement poor man's RSS on strict
> > power-of-2 queue set with some hashing built into the
> > Rx engine, without a key. We'd still want to control
> > which fields are hashed in this case.
> > 
> > In the Netlink API we combine the two so we'd break this?  
> 
> Hmm I wasn't aware of that :( I guess the safe choice is to check on -EOPNOTSUPP
> for .get_rxfh() + 0-len indir table / 0-size key and skip the device during the
> dump instead of returning early ?

Not sure I'm following the suggestion, isn't that what the v3
already does? I think the API was suggesting we let the device
progress into the dump, and rely on the existing "unset" checks
before the nla_put_ calls filter out the data which wasn't provided
(which may be all the data).

Not sure it's worth it, could you poke around to see if a device
which reports zero-length key + indir table actually exists?