Re: Vexing Behavior: RHEL 5.2 (2.6.18-92) Random NIC Enumeration Between Reboots

On Wed, Jul 2, 2008 at 8:14 PM, John Zornig <j.zornig-6NCMjze5YvWHXe+LvDLADg@xxxxxxxxxxxxxxxx> wrote:

John Haxby wrote:

We were talking about this here the other day. You obviously know about the issues of hardware discovery changing from one kernel release to another and the whether you do a breadth-first search or a depth-first search.
The particular case we were talking about was that the hardware doesn't do PCI device detection in a deterministic fashion under some circumstances. It's not quite random, but it's variable enough to bevexing. I'm afraid you're stuck with using HWADDR in the ifcfg-eth* files or using a similar device uuid to distinguish one NIC from another. We suspect that this random behaviour will become the norm in future.
jch

I have a number of Dell systems, running RHEL5.2 suffering badly from this issue. They all have two on-board NICs and one or two 2-port or 4-port PCI cards.

I can't set HWADDR in the ifcfg-eth* files because most of the NICs are in bonded pairs and it is the ifcfg-bond* which has a HWADDR which is shared between the two NICs which are slaves to that bond*.

I've just discovered the pci=bfsort kernel parameter yesterday and begun to configure it on the systems, but it is too early to say if it works, as the issue is random at boot time.

I had no luck configuring udev/rules.d to do HWADDR matching using rules like:
KERNEL=="eth*",DRIVER=="bnx2",SYSFS{address}=="00:1d:09:1e:b0:a9", NAME="eth0"
It seemed to be ignored.

JZ

John -->

I hope someone more knowledgeable would have answered this by now.

First of all, the udev rules and parameters appears to be a a state of flux
or change.  For 5.x, I don't believe that "SYSFS{address}==..." is still
supported.

Next, I believe from my experimentation that using UDEV and locking to
the PCI slot is probably the best way to go (at least for my application).  I
haven't migrated to 5.2 yet, but here is a write-up I did on the CentOS
list for a similar question.  I hope it helps ...

==============  Forwarded Message =====================

On Sat, Apr 19, 2008 at 3:53 PM, Joseph L. Casale <jcasale-QeoNDF8sg3qKnZcHqi4ZagC/G2K4zDHf@xxxxxxxxxxxxxxxx> wrote:

>Modify /etc/sysconfig/network-scripts/ifcfg-ethX and remove the HWADDR
>line if you have one, and add a MACADDR with the mac address you want
>to use.
>
>Beware, some network cards may protest having the mac address changed,
>and using both HWADDR and MACADDR can cause issues. See
>/usr/share/doc/initscripts-*/sysconfig.txt for details.

Jim,
I appreciate the confirmation, that was the method I was going to use. I am only unsure about what *could* happen with the HWADDR in there, can eth{n} now maybe bind to a different nic under some circumstance?

How can I always force the nic in question to use this script?

Thank you!
jlc

Here is an outline of what I do to "lock-down" interfaces -- which relies
mainly on using a fairly new feature "udev":

   /etc/modprobe.conf:  make sure the lines --

              alias eth? <driver>
    
       are in the correct order, e.g.:

              alias eth0 e1000
              alias eth1 e1000
              alias eth2 tg3

    /etc/udev/rules.d/:  create network rules file (if needed) and
         add lines that associate a given NIC to its eth? interface.
         Use "udevinfo -a -p /sys/class/net/eth?" to get various
         features or attributes to find the NIC that you want to call
         eth<X>.  [Note: this seems to change from release to
         release, so this is a little general.]  You might want to put
         lines like:

           Kernel==eth?  ID==0000:03:02.0 Name=eth0
           Kernel==eth?  ID==0000:03:02.1 Name=eth1

          or

            Kernel==eth?  Sys{vendor}==0x8086 Sys{device}==0x032a Name=eth0
            Kernel==eth?  Sys{vendor}==0x8086 Sys{device}==0x1079 Name=eth1

    /etc/sysconfig/network-scripts/ifcfg-eth<X>:

         As other have suggested, now put MACADDR= into these files with the
         desired MAC address that you want the interface to be set to and
         delete the HWADDR.

Now, reboot, test and repeat as needed:-):-) ...

I hope that helps and is useful ...

  -rak-

Note:  I just checked a Fedora 8 box and some of the above has
changed -- udev is the way to go, but be advised that this feature
appears to be evolving and changing -- hopefully for the better!

_______________________________________________
rhelv5-list mailing list
rhelv5-list-H+wXaHxf7aLQT0dZR+AlfA@xxxxxxxxxxxxxxxx
https://www.redhat.com/mailman/listinfo/rhelv5-list