Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 11 Mar 2021 12:33:13 -0800
From:   Jakub Kicinski <kuba@kernel.org>
To:     <stefanc@marvell.com>
Cc:     <netdev@vger.kernel.org>, <thomas.petazzoni@bootlin.com>,
        <davem@davemloft.net>, <nadavh@marvell.com>,
        <ymarkman@marvell.com>, <linux-kernel@vger.kernel.org>,
        <linux@armlinux.org.uk>, <mw@semihalf.com>, <andrew@lunn.ch>,
        <rmk+kernel@armlinux.org.uk>, <atenart@kernel.org>,
        <rabeeh@solid-run.com>
Subject: Re: [V2 net-next] net: mvpp2: Add reserved port private flag
 configuration
Message-ID: <20210311123313.0f5e7f80@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
In-Reply-To: <1615481007-16735-1-git-send-email-stefanc@marvell.com>
References: <1615481007-16735-1-git-send-email-stefanc@marvell.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Thu, 11 Mar 2021 18:43:27 +0200 stefanc@marvell.com wrote:
> According to Armada SoC architecture and design, all the PPv2 ports
> which are populated on the same communication processor silicon die
> (CP11x) share the same Classifier and Parser engines.
> 
> Armada is an embedded platform and therefore there is a need to reserve
> some of the PPv2 ports for different use cases.
> 
> For example, a port can be reserved for a CM3 CPU running FreeRTOS for
> management purposes or by user-space data plane application.
> 
> During port reservation all common configurations are preserved and
> only RXQ, TXQ, and interrupt vectors are disabled.
> Since TXQ's are disabled, the Kernel won't transmit any packet
> from this port, and to due the closed RXQ interrupts, the Kernel won't
> receive any packet.
> The port MAC address and administrative UP/DOWN state can still
> be changed.
> The only permitted configuration in this mode is MTU change.
> The driver's .ndo_change_mtu callback has logic that switches between
> percpu_pools and shared pools buffer mode, since the buffer management
> not done by Kernel this should be permitted.

Andrew asks good questions. This looks like a strange construct.

IMO Linux should either not see the port (like it doesn't see NC-SI),
or we need representors for physical and logical ports and explicit
forwarding rules.