Ethernet on 8260

Ethernet on 8260

[Brad Bonkoski]

Hello,

I am working with a custom 8260 board which has 3 Ethernet devices that
are on using FCC to communicate.  When I try to activate all three I get
some funny routing.  Essentailly, I have them all set up on the same
network: 192.168.0.X.  When I try to ping one Ethernet device, the
system tries to respond on another.  What could lead to this problem?
The CPM appears to be placing the sk_buffs in the right queues in DPRAM,
but the sk_buffs seem to be getting mangled, many times the kernel tries
to shove an additional ARP header on them.  Any ideas as to what could
be causing this?  Is this a known problem and fixed in later versions of
fcc_enet.c?  Do I need to post additional diagonostic information?
Thanks for any help.
-Brad


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Ethernet on 8260

[Murray Jensen]

On 16 Jul 2001 17:19:36 -0700, Brad Bonkoski <bradb59@home.com> writes:
>Essentailly, I have them all set up on the same
>network: 192.168.0.X.

Use different IP network numbers for each ethernet interface. e.g.

FCC1 - eth0 - 192.168.1.x
FCC2 - eth1 - 192.168.2.x
FCC3 - eth2 - 192.168.3.x

(assuming a netmask of 255.255.255.0). If they all have the same IP *network*
number (i.e. the number is the same after applying the netmask), then Linux
will only use one of them. In other words, you can only have one route for
each subnet.

By the way, if you want Linux to forward packets between the interfaces, then
ensure you enable forwarding (echo 1 > /proc/sys/net/ipv4/ip_forward).

Also, you could by-pass all this IP stuff and do things at the ethernet level
(using a RAW socket?). Cheers!
								Murray...
--
Murray Jensen, CSIRO Manufacturing Sci & Tech,         Phone: +61 3 9662 7763
Locked Bag No. 9, Preston, Vic, 3072, Australia.         Fax: +61 3 9662 7853
Internet: Murray.Jensen@cmst.csiro.au  (old address was mjj@mlb.dmt.csiro.au)


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Ethernet on 8260

[Brad Bonkoski]

On 17 Jul 2001 14:22:21 +1000, Murray Jensen wrote:
>
> On 16 Jul 2001 17:19:36 -0700, Brad Bonkoski <bradb59@home.com> writes:
> >Essentailly, I have them all set up on the same
> >network: 192.168.0.X.
>
> Use different IP network numbers for each ethernet interface. e.g.
>
> FCC1 - eth0 - 192.168.1.x
> FCC2 - eth1 - 192.168.2.x
> FCC3 - eth2 - 192.168.3.x
>
> (assuming a netmask of 255.255.255.0). If they all have the same IP *network*
> number (i.e. the number is the same after applying the netmask), then Linux
> will only use one of them. In other words, you can only have one route for
> each subnet.
>

Does Linux _really_ do this?  Shouldn't it route at the host layer
before the network layer?  Like it I ping 192.168.0.1 it should return
the ping on that interface regardless of if another network device lives
on this network!  Or, do I have to explicitly set up host routes in the
routing table?  I guess what is the point of having multiple interfaces
on one board live on the same network, but even so, I would still think
it should choose the host route before the network route.  Any thoughts
on this?


> By the way, if you want Linux to forward packets between the interfaces, then
> ensure you enable forwarding (echo 1 > /proc/sys/net/ipv4/ip_forward).
>
> Also, you could by-pass all this IP stuff and do things at the ethernet level
> (using a RAW socket?). Cheers!
>

This is a good idea, but I would like to keep it simple, as they say
simplicity is the 'key' to High availability.

Regards,
Brad

                                                          Murray...
> --
> Murray Jensen, CSIRO Manufacturing Sci & Tech,         Phone: +61 3 9662 7763
> Locked Bag No. 9, Preston, Vic, 3072, Australia.         Fax: +61 3 9662 7853
> Internet: Murray.Jensen@cmst.csiro.au  (old address was mjj@mlb.dmt.csiro.au)
>
>
>


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Ethernet on 8260

[Murray Jensen]

On 16 Jul 2001 22:30:24 -0700, Brad Bonkoski <bradb59@home.com> writes:
>Does Linux _really_ do this?  Shouldn't it route at the host layer
>before the network layer?

I think it does look for host routes before matching network routes, but
there needs to be one in your routing table. I believe when you configure
the interface the driver installs the network route only. i.e. by default
they will clash.

>Like it I ping 192.168.0.1 it should return
>the ping on that interface regardless of if another network device lives
>on this network!  Or, do I have to explicitly set up host routes in the
>routing table?

Yes, I believe you need to explicitly add them to the routing tables.

>I guess what is the point of having multiple interfaces
>on one board live on the same network, but even so, I would still think
>it should choose the host route before the network route.  Any thoughts
>on this?

You are correct, but you have to set it up manually. Or maybe it depends on
what net-tools package you use - the one I use doesn't add the host routes.

>This is a good idea, but I would like to keep it simple, as they say
>simplicity is the 'key' to High availability.

What could be more simple than removing all that really complicated IP
protocol overhead? :-) Cheers!
								Murray...
--
Murray Jensen, CSIRO Manufacturing Sci & Tech,         Phone: +61 3 9662 7763
Locked Bag No. 9, Preston, Vic, 3072, Australia.         Fax: +61 3 9662 7853
Internet: Murray.Jensen@cmst.csiro.au  (old address was mjj@mlb.dmt.csiro.au)


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Ethernet on 8260

[Brad Bonkoski]

Thanks for the recoomendations, look like things are on thier way!

One further question, is it possible for a detection of a loss of the
actual Ethernet cable at the kernel level?  Are there any pointers to
docs I could read on this?
Thanks again!
Regards,
Brad


On 17 Jul 2001 19:09:34 +1000, Murray Jensen wrote:
>
> On 16 Jul 2001 22:30:24 -0700, Brad Bonkoski <bradb59@home.com> writes:
> >Does Linux _really_ do this?  Shouldn't it route at the host layer
> >before the network layer?
>
> I think it does look for host routes before matching network routes, but
> there needs to be one in your routing table. I believe when you configure
> the interface the driver installs the network route only. i.e. by default
> they will clash.
>
> >Like it I ping 192.168.0.1 it should return
> >the ping on that interface regardless of if another network device lives
> >on this network!  Or, do I have to explicitly set up host routes in the
> >routing table?
>
> Yes, I believe you need to explicitly add them to the routing tables.
>
> >I guess what is the point of having multiple interfaces
> >on one board live on the same network, but even so, I would still think
> >it should choose the host route before the network route.  Any thoughts
> >on this?
>
> You are correct, but you have to set it up manually. Or maybe it depends on
> what net-tools package you use - the one I use doesn't add the host routes.
>
> >This is a good idea, but I would like to keep it simple, as they say
> >simplicity is the 'key' to High availability.
>
> What could be more simple than removing all that really complicated IP
> protocol overhead? :-) Cheers!
>                                                               Murray...
> --
> Murray Jensen, CSIRO Manufacturing Sci & Tech,         Phone: +61 3 9662 7763
> Locked Bag No. 9, Preston, Vic, 3072, Australia.         Fax: +61 3 9662 7853
> Internet: Murray.Jensen@cmst.csiro.au  (old address was mjj@mlb.dmt.csiro.au)
>
>
>


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Ethernet on 8260

[Murray Jensen]

On 17 Jul 2001 12:07:19 -0700, Brad Bonkoski <bradb59@home.com> writes:
>One further question, is it possible for a detection of a loss of the
>actual Ethernet cable at the kernel level?

Depends on your board design - only the PHY knows this info and you
communicate with the PHY via the MII. But the 8260 has no specific
support for the management communication as defined by the MII spec
(MDC/MDIO - very similar to an i2c interface) - you implement this
using I/O port pins, twiddling bits to send clocked serial data to
the PHY.

The first board design dependent thing is which port pins you choose
to connect to the PHY MDC/MDIO pins. The second is which interrupt
input(s) (port C pins?) you connect the PHY interrupt output(s) to.
There will be others - probably dependent on the PHY you choose, or
even the RJ45 connector itself! e.g. on our board, there is a bi-colour
LED on the RJ45 socket which we connect to our FPGA so that with the
appropriate FPGA code loaded we can interpret the PHY state outputs and
display the appropriate colours - rather than dedicate I/O port pins
on the 8260 to this function i.e. input the PHY state, output the LED
state - wastes a lot of pins. Other board designs might use I/O port
pins. LEDs aren't so critical, but you get the idea.

Based on previous postings, I believe Dan Malek will (eventually) be
patching the FCC driver to provide interrupt driven MII/PHY support,
but you will have to hack it for your PHY etc. although most PHYs are
pretty similar so that all you might need to do is recognise your vendor
ID number.

>Are there any pointers to
>docs I could read on this?

Your PHY manual, your board design manual, and section 30.3 of the 8260 UM.
Cheers!
								Murray...
--
Murray Jensen, CSIRO Manufacturing Sci & Tech,         Phone: +61 3 9662 7763
Locked Bag No. 9, Preston, Vic, 3072, Australia.         Fax: +61 3 9662 7853
Internet: Murray.Jensen@cmst.csiro.au  (old address was mjj@mlb.dmt.csiro.au)


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Ethernet on 8260

[August Hoerandl]

On Tuesday 17 July 2001 02:19 am, Brad Bonkoski wrote:
> Hello,
>
> I am working with a custom 8260 board which has 3 Ethernet devices that
> are on using FCC to communicate.  When I try to activate all three I get
> some funny routing.  Essentailly, I have them all set up on the same
> network: 192.168.0.X.  When I try to ping one Ethernet device, the
> system tries to respond on another.  What could lead to this problem?

you should not put more then one "real" device into one subnet - the
kernel will always check with the netmask and use the first one that
matches, there is no idea of "the request was on eth2, the answer has
to go out on eth2"

hth
Gustl

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/