[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230110125915.62d428fb@kernel.org>
Date: Tue, 10 Jan 2023 12:59:15 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: Jiri Pirko <jiri@...nulli.us>
Cc: netdev@...r.kernel.org, davem@...emloft.net, pabeni@...hat.com,
edumazet@...gle.com, michael.chan@...adcom.com,
yisen.zhuang@...wei.com, salil.mehta@...wei.com,
jesse.brandeburg@...el.com, anthony.l.nguyen@...el.com,
tariqt@...dia.com, saeedm@...dia.com, leon@...nel.org,
idosch@...dia.com, petrm@...dia.com, mailhol.vincent@...adoo.fr,
jacob.e.keller@...el.com, gal@...dia.com
Subject: Re: [patch net-next v3 01/11] devlink: remove devlink features
On Tue, 10 Jan 2023 08:12:10 +0100 Jiri Pirko wrote:
>> Right, but this is not 100% equivalent because we generate the
>> notifications _before_ we try to reload_down:
>>
>> devlink_ns_change_notify(devlink, dest_net, curr_net, false);
>> err = devlink->ops->reload_down(devlink, !!dest_net, action, limit, extack);
>> if (err)
>> return err;
>>
>> IDK why, I haven't investigated.
>
> Right, but that is done even in other cases where down can't be done. I
> I think there's a bug here, down DEL notification is sent before calling
> down which can potentially fail. I think the notification call should be
> moved after reload_down() call. Then the bahaviour would stay the same
> for the features case and will get fixed for the reload_down() reject
> cases. What do you think?
I was gonna say that it sounds reasonable, and that maybe we should
be in fact using devlink_notify_register() instead of the custom
instance-and-params-only devlink_ns_change_notify().
But then I looked at who added this counter-intuitive code
and found out it's for a reason - see 05a7f4a8dff19.
So you gotta check if mlx5 still has this problem...
Powered by blists - more mailing lists