Description of problem: On our Rackable Opteron servers, the i686 kernel detects the left-most tg3 10/100/1000 interface is eth0, the identical NIC on the right is eth1. However, the AMD64 version of RHEL3 reverses this order. This is true during the installation of RHEL3 and afterwards. Is this a known problem in the kernel? What is the fix? Version-Release number of selected component (if applicable): RHEL3
The x86_64 kernel and the i386 kernel probe for PCI controllers and busses in a different order. An x86_64 platform export will have to determine whether 1) it is even possible to make x86_64 match i386 behavior 2) whether that is even desirable I am not such an export so I'm removing myself from the CC:
Ok, reassigning to JimP.
The RHEL3 i686 and x86_64 kernels use different mechanisms to probe the PCI bus (i686 probes the bus directly, x86_64 uses ACPI tables). As a result the default enumeration of network devices may be different. If you are going to be going back and forth between the two OSs and want to force a particular enumeration, you can have the devices renamed at system startup just prior to being configured. One way to do this is to use the GUI net configurator (Main Menu | System Settings | Network) and click on the "Hardware" tab. You can then edit the entries and set their names to your liking. Alternatively, you can edit the /etc/sysconfig/network-scripts/ifcfg-eth* files. Add to each a line of the form "HWADDR=xx:xx:xx:xx:xx:xx" specifying the MAC address of the interface you want to correspond to that particular name. Closing as WONTFIX since a workaround is available.
This isn't a workaround when kickstarting. There isn't /etc/sysconfig/anything on the box if linux isn't yet installed. Think about it from an enterprises perspective. You have 300+ Opteron servers. You have many system admins over the globe. You have a standard wiring policy, where you "plug the cable into the leftmost eth port". eth0 is always the one on the left, except when you try to load kickstart with AMD64 RHEL3. Then kickstarting doesn't work. I know that changing the way the kernel does things shouldn't be taken lightly, but if both probing the ACPI tables and probing directly work, why not just use one method? Is the difference in methods really needed?
What is the status of this. If it can't be changed in either of kernels, perhaps we should just close this bug.
???
User jparadis's account has been closed
hello?
The development cycle for RHEL3 is long past. Closing this issue accordingly.