Re: [PATCH 0/2] Indirect phy access for mv88e6171

2015-02-15 12:25 GMT-08:00 Andrew Lunn [off-list ref]:

quoted

I'll try. My primary problem right now is that I run Lennert Buytenhek's
patch set to support bridging offload (aka hardware bridging) on top of
the upstream dsa code, and the upstream code now supports a competing /
alternate means  to support bridging/switching offload (NET_SWITCHDEV)
which doesn't work with dsa (at least not yet). So I'll have to figure
out if / how I can run your patches with my code base, or how I can add
add support for NET_SWITCHDEV into dsa.

We should be adding switchdev support to DSA for hardware
bridging. The concepts in Lennert Buytenhek's should be a good
starting point for this.

In fact, there is not much to be implemented in DSA, since we already
have everything in place in net-next now:

- net_device notifier to learn which ports are leaving/joining the bridge
- hook an abstraction for ndo_switch_port_stp_update
- hook an abstraction for ndo_fdb_{add,del,dump}

quoted

Do you know if there are any efforts going on in that direction ?

Florian has expressed an interest in getting hardware bridging
working. I've no idea if he has looked at it from the perspective of
switchdev.

I have some patches that leverage the switchdev-related patches and
bring an abstraction to DSA, they should not be fundamentally
different at the DSA driver level, since most of the abstraction is
done in net/dsa/slave.c, I plan on posting these patches tomorrow once
I am back home. I tested these with the bcm_sf2 driver, so testing
with Marvell hardware would be more than welcome.

Thanks!
-- 
Florian