[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <48072595c9c344fea9c268fd81e4d06e@realtek.com>
Date: Thu, 23 May 2024 06:29:55 +0000
From: Larry Chiu <larry.chiu@...ltek.com>
To: Andrew Lunn <andrew@...n.ch>
CC: Justin Lai <justinlai0215@...ltek.com>,
"kuba@...nel.org"
<kuba@...nel.org>,
"davem@...emloft.net" <davem@...emloft.net>,
"edumazet@...gle.com" <edumazet@...gle.com>,
"pabeni@...hat.com"
<pabeni@...hat.com>,
"linux-kernel@...r.kernel.org"
<linux-kernel@...r.kernel.org>,
"netdev@...r.kernel.org"
<netdev@...r.kernel.org>,
"jiri@...nulli.us" <jiri@...nulli.us>,
"horms@...nel.org" <horms@...nel.org>,
Ping-Ke Shih <pkshih@...ltek.com>
Subject: RE: [PATCH net-next v19 01/13] rtase: Add pci table supported in this module
> > Thank you very much for your clear reply.
> >
> > As I mentioned, it works like a NIC connected to an Ethernet Switch, not a
> > Management port.
> > The packets from this GMAC are routed according to switch rules such as
> > ACL, L2, .... and it does not control packet forwarding through any special
> > header or descriptor. In this case, we have our switch tool which is used
> > for provisioning these rules in advance. Once the switch boots up, the
> > rules will be configured into the switch after the initialization. With this
> > driver and the provisioning by our switch tool, it can make switch forward
> > the frame as what you want. So it's not a DSA like device.
>
> How does spanning tree work? You need to send bridge PDUs out specific
> ports. Or do you not support STP and your network must never have
> loops otherwise it dies in a broadcast storm? That does not sound very
> reliable.
>
> There are other protocols which require sending packets out specific
> ports. Are they simply not supported?
>
This port is not a CPU port, nor a management port, and therefore does not
manage any protocols of the switch. These protocols are implemented by the
CPU inside the Ethernet switch core.
> > In another case, we do have other function which is used for controlling
> > the switch registers instead of sending packets from the switch ports.
> > At the meanwhile, we are investigating how to implement the function to
> > Integrate into switchdev.
>
> In general, we don't support configuration of hardware from user
> space, which is what your switch tool sounds like. We will want to see
> a switchdev driver of some form.
>
> It might be you need to use VLAN overlays, using
> net/dsa/tag_8021q.c. Each port of the switch is given a dedicated
> VLAN, and the switch needs to add/strip the VLAN header. Its not
> great, but it does allow 'simple' switches to have basic functionality
> if they are missing header/dma descriptor support for selecting ports.
>
> Andrew
Typically, a NIC connected to the network may go through a switch, as show
below. Our design saves the two PHYs in the middle and connects to the Host
through PCIe.
This driver just service the transmit/receive packets for one port in the RTL90xx
with PCIe interface. Other programs that the switch needs to execute are
managed by the CPU inside the switch core.
* *************************
* * *
* * PC/Host *
* * *
* * +-------------+ *
* * | NIC | *
* ***********++************
* | PHY |
* ||
* | PHY |
* +-------------++----------------+
* | | MAC | |
* | +-----+ |
* | |
* | Ethernet Switch Core |
* | |
* | +-----+ +-----+ |
* | | MAC |...........| MAC | |
* +---+-----+-----------+-----+---+
* | PHY |...........| PHY |
|
|
|
V
* *************************
* * *
* * PC/Host *
* * *
* * +-------------+ *
* * | NIC | *
* ***********++************
* ||
* +-------------++----------------+
* | | MAC | |
* | +-----+ |
* | |
* | Ethernet Switch Core |
* | |
* | +-----+ +-----+ |
* | | MAC |...........| MAC | |
* +---+-----+-----------+-----+---+
* | PHY |...........| PHY |
Powered by blists - more mailing lists