[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <20191206.204617.258963174428895178.davem@davemloft.net>
Date: Fri, 06 Dec 2019 20:46:17 -0800 (PST)
From: David Miller <davem@...emloft.net>
To: john.hurley@...ronome.com
Cc: netdev@...r.kernel.org, simon.horman@...ronome.com,
jakub.kicinski@...ronome.com, oss-drivers@...ronome.com
Subject: Re: [PATCH net 0/2] Ensure egress un/bind are relayed with
indirect blocks
From: John Hurley <john.hurley@...ronome.com>
Date: Thu, 5 Dec 2019 17:03:33 +0000
> On register and unregister for indirect blocks, a command is called that
> sends a bind/unbind event to the registering driver. This command assumes
> that the bind to indirect block will be on ingress. However, drivers such
> as NFP have allowed binding to clsact qdiscs as well as ingress qdiscs
> from mainline Linux 5.2. A clsact qdisc binds to an ingress and an egress
> block.
>
> Rather than assuming that an indirect bind is always ingress, modify the
> function names to remove the ingress tag (patch 1). In cls_api, which is
> used by NFP to offload TC flower, generate bind/unbind message for both
> ingress and egress blocks on the event of indirectly
> registering/unregistering from that block. Doing so mimics the behaviour
> of both ingress and clsact qdiscs on initialise and destroy.
>
> This now ensures that drivers such as NFP receive the correct binder type
> for the indirect block registration.
Series applied and queued up for -stable.
Powered by blists - more mailing lists