Wed, Feb 25, 2015 at 08:03:48AM CET, zajec5@gmail.com wrote:

On 25 February 2015 at 01:39, Florian Fainelli [off-list ref] wrote:

quoted

On 24/02/15 16:15, Andrew Lunn wrote:

quoted

On Tue, Feb 24, 2015 at 02:55:58PM -0800, Florian Fainelli wrote:

quoted

On 24/02/15 14:50, Rafa?? Mi??ecki wrote:

quoted

As said in the commit message, these switches are really simple
devices. We can't actually send packets to the particular ports
(unless something has changed in the more recent hardware).

These switches all support Broadcom tags, so you could use your host CPU
Ethernet MAC to send/receive packets to/from specific ports of the
switch, and then this is just like DSA, but everything that you say
below is true.

If this hardware does support the concept of tags compatible to the
existing Broadcom Starfighter 2, should we not do that? Is there a
good reason these chips should use a difference abstraction than
Starfighter 2 and the Marvell devices?

No, I think that would be a reasonable thing to do, I am still a little
unclear how the older BCM5325 and friends work with respect to broadcom
tags, but regardless, DSA now understands switches that do not support
tags, so we should be good with some sort of b53 library.

DSA doesn't allow me to use switch (hardware) VLANs. I would need to
handle all forwarding in CPU. The most powerful Broadcom MIPS SoC -
BCM4706 - can handle about 130 Mb/s. Less powerful - BCM4718A1 - only
about 50 Mb/s.
This is way too bad for performance when there is a 1 Gb/s switch and
too bad for CPU usage.

Please correct me if I'm wrong Florian, but I believe that DSA does
allow setting up chips to do forwarding without packets going to CPU.

b53 looks like it fits nicely into DSA architecture. I believe that it
should be integrated there.

I guess the whole idea of switchdev was to allow what proposed version
of b53 does? To handle VLANs in hardware.

-- 
Rafał