[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <ZHb2GkallaC/4wyO@nanopsycho>
Date: Wed, 31 May 2023 09:24:10 +0200
From: Jiri Pirko <jiri@...nulli.us>
To: Jakub Kicinski <kuba@...nel.org>
Cc: netdev@...r.kernel.org, pabeni@...hat.com, davem@...emloft.net,
edumazet@...gle.com, saeedm@...dia.com, moshe@...dia.com,
simon.horman@...igine.com, leon@...nel.org
Subject: Re: [patch net-next] devlink: bring port new reply back
Wed, May 31, 2023 at 08:53:39AM CEST, kuba@...nel.org wrote:
>On Wed, 31 May 2023 08:36:25 +0200 Jiri Pirko wrote:
>> >> FWIW it should be fairly trivial to write tests for notifications and
>> >> replies now that YNL exists and describes devlink..
>> >
>> >Actually, I'm not 100% sure notifications work for devlink, with its
>> >rtnl-inspired command ID sharing.
>>
>> Could you elaborate more where could be a problem?
>
>right here
>
>https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/tree/tools/net/ynl/lib/ynl.py#n518
>
>;) If we treat Netlink as more of an RPC than.. state replication(?)
>mechanism having responses and notifications with the same ID is a bit
>awkward. I felt like I had to make a recommendation in YNL either to
>ask users not to enable notifications and issue commands on the same
>socket, or for family authors to use different IDs. I went with the
>latter. And made YNL be a bit conservative as to what it will consider
>to be a notification.
I see. I don't think we can change this devlink kernel behaviour though.
Anyway, as the command issuer does not enable notifications, he should
be okay.
Powered by blists - more mailing lists