Re: Marvell 88E609x switch?

[Gary Thomas <gary@mlbassoc.com>]

Lennert Buytenhek wrote:
> On Fri, Feb 27, 2009 at 05:25:06AM -0700, Gary Thomas wrote:
> 
>>>>>>>>>> Is there support for this device anywhere?  In particular,
>>>>>>>>>> the M88E6095 switch.
>>>>>>>>> Not at the moment, but it should be easy enough to add.  If your
>>>>>>>>> board already runs on 2.6.28+, I can whip up some patches for you
>>>>>>>>> to try from the docs I have for that part.
>>>>>>>> That would be much appreciated, thanks.
>>>>>>> I noticed that the 6095/6095F are quite similar to the 6131 as far
>>>>>>> as the register set goes.  So something along these lines (hacky
>>>>>>> patch, breaks 6131, not for mainline) might just work to detect
>>>>>>> single 6095s (cascading DSA chips is something that needs more work,
>>>>>>> let's get the single-chip case working first).
>>>>>>>
>>>>>>> The other thing you'll need to do is create dsa platform devices
>>>>>>> for your switch chips, a la how it's done in arch/arm/mach-orion5x/
>>>>>>> or arch/arm/mach-kirkwood/ for example -- you need to pass in a struct
>>>>>>> device * for your network device, a struct device * for your mii bus,
>>>>>>> the switch MII address on the MII bus, and names of the individual
>>>>>>> ports (where you'll specify "cpu" for the port on the switch chip that
>>>>>>> the CPU is connected to).
>>>>>>>
>>>>>>> Let me know if this works.
>>>>>> Thanks, I'll give it a try.  It will take a little effort
>>>>>> to get setup as I have to work within the open firmware
>>>>>> structure (that's how all the various components are
>>>>>> specified).
>>>>> Right, we don't have OF bindings yet.  I guess this would make sense
>>>>> to do generically at some point, since there are quite a few PPC
>>>>> platforms with DSA switch chips.
>>>> Here's what I tried - (patch attached) - a trulyhorrible hack,
>>>> but I've not figured out how to get the correct device pointers
>>>> from the OF world yet.  The boot log shows that it's trying, but
>>>> I don't see the DSA layer (M88E690x driver) doing the MII indirection
>>>> that's needed for this device.
>>>>
>>>> I'm probably not starting it up correctly, but I think I followed
>>>> the examples you cited.  Any ideas?
>>> "indirection needed for this device" -- does that mean that your
>>> switch chip is configured to use the multi-chip addressing mode?
>>> (It looks like it, as most of the MII addresses return ffff in
>>> their ID registers.)  If yes, you should set ->sw_addr to whatever
>>> MII address the chip has been assigned.
>> Much better, my switch seems to be found now.
>>
>>   Distributed Switch Architecture driver version 0.1
>>   gfar_mdio_read(cf9db400, 1, 0) = 1811
>>   gfar_mdio_write(cf9db400, 1, 0, 9a03) = 0
>>   gfar_mdio_read(cf9db400, 1, 0) = 1a03
>>   gfar_mdio_read(cf9db400, 1, 1) = 953
>>   mv88e6131_probe(cf9db400, 1) = 2387
> 
> That looks like the proper indirect access sequence.
> 
> 
>>   eth0: detected a Marvell 88E6095/88E6095F switch
> 
> Yay.
> 
> 
>>   root@ppc_target:~ ls /sys/bus/mdio_bus/devices/
>>   24520:01:00  24520:01:02  24520:01:04  24520:01:06
>>   24520:01:01  24520:01:03  24520:01:05  24520:01:07
> 
> That looks good too.
> 
> 
>> However, the network subsystem still can't locate it.  It may be
>> a complication of the OF stuff and how the [gianfar] network
>> device knows what PHY to look at.
>>
>>   starting network interfaces...
>>   24520:01 not found
>>   eth0: Could not attach to PHY
> 
> It's correct that eth0 should not associate with a PHY -- since the
> MAC connects to the switch chip over GMII/RGMII/SGMII, there is no
> PHY involved.

Yes, but it's expecting to be able to talk to the PHYLIB layer
and ask things like speed, duplex, link state, etc.

> Is the gianfar driver refusing to up the interface without a PHY?
> It shouldn't do that -- IMHO it's perfectly fine to not have a PHY
> on your ethernet MAC.
> 

It seems so, yes.  I tried disabling the PHY connection in the OF
tree and now eth0 doesn't even try to come up :-(

> 
>> Also, how do I specify the [implicit] route within the switch
>> that connects '24520:01:00' to the CPU port '24520:01:0A' (if
>> there was such a thing)?  My boot loader has configured the
>> switch for this path - I've not looked through the log to see
>> what the DSA layer did.
> 
> Just specify "cpu" as the port name of port 10 in your dsa platform
> data (assuming that there's where your CPU is), that'll take care of
> it.
> 
> Does it give you Linux network interfaces for the other switch chip
> ports (1-8)?

Not sure I understand this question, there are no other network
interfaces listed:

root@ppc_target:~ cat /proc/net/dev
Inter-|   Receive                                                |  Transmit
 face |bytes    packets errs drop fifo frame compressed multicast|bytes    packets errs drop fifo colls carrier compresse
d
    lo:       0       0    0    0    0     0          0         0        0       0    0    0    0     0       0
0


-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------