Re: Marvell DSA switch 88E6165 - CPU port in PHY mode

[mdsmith@xxxxxxxxxxxxxx]

Le 2012-03-05 18:12, Lennert Buytenhek a écrit :
> On Sun, Mar 04, 2012 at 03:11:53PM +0100, Matthew D. Smith wrote:
>
> > Hello - have a question about CPU port configuration in the net/dsa
> > driver code. Can't seem to work out how to bring up the CPU 
> > connected
> > port on the switch (in our case this is Port 4 hardware strapped 
> > with
> > its embedded PHY enabled).
>
> Hello!  I know of at least a couple of platforms that have a DSA chip
> connected to the CPU via an MDI type link, but I'm not sure why it's
> not working for you.  I'd suggest double-checking all the port 5 PHY
> registers against the chip documentation..

Hi - perhaps I misunderstood something about the built-in PHY on Port 
4. We expected to see the PHY enabled in hardware by setting logic on 
the P4_MODE pins. But this doesn't seem to be sufficient as I guess one 
still has to enable the PHY in the driver code (like how you do it for 
the LAN interfaces in net/dsa/slave.c).

So, after digging through the net/dsa source I found that there is 
nothing explicit in the code to enable the CPU port if it needs a PHY. I 
would like to say that way it is written assumes that it is always RGMII 
so I guess the other platforms you refer to above use customized 
(non-mainline) driver code?

I have managed to get the DSA switch working by making the following 
changes:

--- ./kernel/linux-source-3.2-csb1724/net/dsa/dsa.c
+++ ./kernel/linux-source-3.2-csb1724/net/dsa/dsa.c (unsaved)
@@ -124,7 +124,6 @@
 			}
 			dst->cpu_switch = index;
 			dst->cpu_port = i;
+			ds->phys_port_mask |= 1 << i;
 		} else if (!strcmp(name, "dsa")) {
 			ds->dsa_port_mask |= 1 << i;
 		} else {

--- ./kernel/linux-source-3.2-csb1724/net/dsa/mv88e6123_61_65.c
+++ ./kernel/linux-source-3.2-csb1724/net/dsa/mv88e6123_61_65.c 
(unsaved)
@@ -193,7 +193,12 @@
 	 * full duplex.
 	 */
 	if (dsa_is_cpu_port(ds, p) || ds->dsa_port_mask & (1 << p)) {
-		REG_WRITE(addr, 0x01, 0x003e);
+		REG_WRITE(addr, 0x01, 0x0003);
 	}
 	else
 		REG_WRITE(addr, 0x01, 0x0003);

Basically this tells the DSA driver to create an interface for the CPU 
port. Perhaps overkill but it does bring up the built-in PHY in the same 
way it brings up the LAN slave interfaces. As slave.c sets autoneg on I 
needed to change the register value from 1000FD to autoneg. Likewise for 
the CPU GE01 port in my platform init code.

I don't really like this, but it works. Is there a better way?

regards
Matthew
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html