Switch up eth0 and eth1 to BMC shared and dedicated resp. ( Zaius )

Milton Miller II miltonm at us.ibm.com
Sun Sep 3 10:12:24 AEST 2017


On 06/26/2017 about 06:06PM, Patrick Williams wrote:
>On Thu, Jun 15, 2017 at 03:00:30PM +0930, Joel Stanley wrote:
>> On Sat, Jun 10, 2017 at 2:33 AM, Adriana Kobylak <anoo at us.ibm.com>
>wrote:
>> > From: Adi Gangidi
>> > Sent: Friday, June 9, 2017 10:06 AM
>> > To: anoo at us.ibm.com; Kenneth Wilke; Brad DeBauche; Chad Somerlot;
>> > abarrera at us.ibm.com; ryan.dc.yu at foxconn.com; Rob Lippert
>> > Subject: Switch up eth0 and eth1 to BMC shared and dedicated
>resp. ( Zaius )
>> >
>> > Hi Adriana /.  Ryan ( Foxconn )
>> >
>> > Right now on openBMC
>> > BMC Shared      = eth0
>> > BMC Dedicated = eth1
>> >
>> >
>> > Instead, Is there a way we can make sure the BMC dedicated comes
>up as eth0
>> > and BMC shared link comes up as eth1, in Zaius / Barreleye G2 ?
>> >
>> >
>> > In how it works currently . you could have BMC dedicated be eth0
>or eth1
>> > depending upon the hardware config ( Whether or not Broadcom
>5719A is popped
>> > OR de-popped ) . That could confusing to support operators who
>don't
>> > understand the nuances .
>> >
>> > If you guys are busy with other things to make the change  , can
>I have your
>> > thoughts on if this can be changed ? I'm guessing this is a
>u-boot change .
>> > Any thoughts on how we could make this change ourselves ?
>> 
>> You can use a udev rule to ensure the devices are named as you
>expect.
>
>Most Linux distros have moved away from 'ethN' style naming and to
>"something else".  My laptop's network interfaces are named
>'enp0s31f6'
>and 'wlp4s0' for the ethernet and wireless interfaces respectively.
>
>https://www.freedesktop.org/wiki/Software/systemd/PredictableNetworkI
>nterfaceNames/
>
>According to that, the default order of naming we should be getting
>is something
>like this:
>
>1. Names incorporating Firmware/BIOS provided index numbers for
>on-board devices (example: eno1)
>2. Names incorporating Firmware/BIOS provided PCI Express hotplug
>slot index numbers (example: ens1)
>3. Names incorporating physical/geographical location of the
>connector of the hardware (example: enp2s0)
>4. Classic, unpredictable kernel-native ethX naming (example: eth0)
>
>Any idea why we are ending up with #4?  Is this due to device tree
>entries / naming?  It seems like using something like #3 would get
>well
>defined and consistent names no matter which network interface is
>enabled.  It should also, in theory, make all AST2400 chips arrive at
>the same names for NC-SI vs direct PHY.
>

Upstream has udev rules to name PCI devices.  Our NICs are not PCI
devices.  We don't have any BIOS or interface to firmware and I am
not aware of any standard for device tree labeling of connectors[1].

I don't think it makes sense to name the pins on the chip, I would
want to follow the phy(s) for the mac(s).


[1] other than something like ibm,location-code in PAPR.

milton



More information about the openbmc mailing list