I sent this message to Lennert Buytenhek last Friday, but he must be busy
because I haven't received a response yet. I would appreciate any input
anyone has on the subject. A few google searches didn't turn up anything
like this.
-----Original Message-----
From: Eble, Dan
Sent: Friday, January 24, 2003 4:37 PM
To: '[EMAIL PROTECTED]'
Subject: Integrated Hardware/Software Bridging
Good day,
I'm using Linux 2.4.18 in an embedded system. I would like to make a
managed ethernet switch (Broadcom 5325M) work with the Linux bridge
driver, including supporting Spanning Tree Protocol. (Following my
signature is a diagram that may help you understand the software
architecture I am about to describe. You can turn it into postscript
format with "dot -Tps -o <outputfile> <sourcefile>". If you don't have
dot, I will happily send you a postscript version.)
The BCM 5325M is a 6-port managed ethernet switch. It offers a way to
set the Spanning Tree state of each port in hardware. (The STP
algorithm itself is not implemented in hardware, but the switch modifies
its forwarding behavior according to the STP state of each port.)
The switch driver presents a network device for each port (mii0 to
mii5). The current implementation does not allow sending or receiving
any packets via these devices, but it does allow their physical aspects
to be managed with "ifconfig" and "mii-tool". My plan is to allow STP
PDUs to pass through these devices, so that they may be enslaved to a
bridge device to participate in Spanning Tree. (In the diagram, you
will see that eth0 is enslaved to the BCM 5325M driver. The switch
encapsulates STP PDUs received on all ports in specially tagged frames
and sends them to the CPU via eth0. The BCM 5325M driver will
distribute the STP PDUs to the correct miiX device and send everything
else to the eth1 device.)
Given what I've said so far, do you see a better way?
In order for this design to work, the BCM 5325M driver must be informed
of the STP state. Can you suggest a good way to accomplish this within
the existing bridge driver?
I have examined it for about 15 minutes and my first idea is to add an
element to "struct net_device" that points to a "struct stp_net_iface"
(or something like that) provided by the switch driver. When the bridge
driver changes the STP state, if this pointer is not NULL, it will call
a function within "stp_net_iface" to set the STP state in the hardware.
My second idea is to add an ioctl to do the same thing. Which is a
better idea? Is there a better third way that hasn't occurred to me?
(Is this already possible by means I have overlooked?)
Thank you for your time.
--
Dan Eble <[EMAIL PROTECTED]> _____ .
| _ |/|
Applied Innovation Inc. | |_| | |
http://www.aiinet.com/ |__/|_|_|
digraph "g" {
graph [
fontsize = "10"
fontname = "Helvetica"
]
node [
fontsize = "10"
fontname = "Helvetica"
shape = "record"
]
edge [
fontsize = "10"
fontname = "Helvetica"
dir = "none"
]
"nothing" [
style = "dashed"
shape = "ellipse"
]
"routing" [
label = "Routing"
]
"BCM" [
label = "BCM 5325M\nDriver"
]
"FEC" [
label = "FEC\nDriver"
]
"br0" [
label = "Bridging"
]
"WAN" [
label = "WAN\nDriver"
]
"routing" -> "br0" [
label = "br0"
]
"nothing" -> "BCM" [
label = "mii5"
]
"br0" -> "BCM" [
label = "mii0..4"
]
"routing" -> "WAN" [
label = "wanY"
]
"br0" -> "BCM" [
label = "eth1\n(STP off)"
]
"br0" -> "WAN" [
label = "wanX"
]
"BCM" -> "FEC" [
label = "eth0"
]
}
_______________________________________________
Bridge mailing list
[EMAIL PROTECTED]
http://www.math.leidenuniv.nl/mailman/listinfo/bridge
