-----Original Message-----
From: Johannes Berg <johannes@sipsolutions.net>
Sent: Friday, January 20, 2023 1:11 AM
To: Keller, Jacob E <jacob.e.keller@intel.com>; Jakub Kicinski <kuba@kernel.org>;
davem@davemloft.net
Cc: netdev@vger.kernel.org; edumazet@google.com; pabeni@redhat.com;
robh@kernel.org; stephen@networkplumber.org; ecree.xilinx@gmail.com;
sdf@google.com; f.fainelli@gmail.com; fw@strlen.de; linux-
doc@vger.kernel.org; razor@blackwall.org; nicolas.dichtel@6wind.com; Bagas
Sanjaya [off-list ref]
Subject: Re: [PATCH net-next v3 1/8] docs: add more netlink docs (incl. spec docs)

On Thu, 2023-01-19 at 16:23 -0800, Jacob Keller wrote:

quoted

Per op policy is important because otherwise it can become impossible to
safely extend a new attribute to commands over multiple kernel releases.

Yeah. I think I just realised that my issues is more with the fact that
per-op policy implies per-op attribute (identifier/number/name)space,
and if you don't have that you have attribute duplication etc.

Anyway, it just feels superfluous, not really dangerous I guess :)

Johannes

I think we want per-family attribute and op IDs, but still per-op policy that indicates which attributes are supported for a command.

This way you can re-use attributes for multiple commands, and they should behave the same semantics for those commands which support them, but the kernel can properly express which attributes are valid for a given command using the per-op policy.

I know some commands do use separate attributes for each command, or separate attribute space for sub-attributes in nests, but I am not sure how widespread that is. Devlink uses a single space for all its attributes.

Thanks,
Jake