Kernel panic when installing RHEL4 U3 64bit with 4 GB memory - Servers

Resolving The Problem

RETAIN tip: H187528

During installing the RHEL4 U3 x86-64 OS on AMD Dual-core Platform with more than 4 GB memory, when user inserts the CD 1 in the system then press enter at the boot prompt, the installation will fail and show "Kernel panic - not syncing : PCI-DMA : high address but no IOMMU " and system hangs.

The system may be any of the following IBM servers:

• System x3105, Type 4347, any model

The system is configured with at least one of the following:

• Red Hat Linux RHEL 4, any model

The Symptoms appear to be common to all systems using AMD Opteron Dual-Core CPU with more than 4 GB memory installed. A FIX is targeted in RHEL4 U4 X86-64.

To achieve the successful installation of RHEL4 U3 X86-64, user has to disable PCI config space access via MMCONFIG by appending the appropriate parameters at the boot prompt before hand. This can be done by the following step

1. Insert the Red Hat Enterprise Linux Version 4 Update 3 CD 1 into the CD drive and boot the system.
2. In the boot screen, type linux pci=nommconf and then press Enter.
3. Follow the standard instructions that guide you through the installation.

1. When the blue splash screen is displayed, press any key to display the kernel menu.
2. Press e to edit the default kernel configuration.
3. Use the Down Arrow key to scroll to the kernel line, and press e again.
4. Type 'pci=nommconf 3' at the end of the line, then press Enter and then .b. to start in text mode.
5. Log in as root and using vi (or any other editor), edit the /boot/grub/menu.lst file. Add 'pci=nommconf' to the end of the kernel line.
6. Type init 6 to restart the server.

Basically, this is a known issue that Memory Mapped Configuration (mmconfig) of PCI Express is not fully supported for AMD CPU by RHEL4 Update 3 x86-64. All AMD dual-core Opteron boxes will be affected by this issue when network operating system enables the generation of the MCFG table in ACPI. A work around here is to disable 'mmconfig' function that processes MCFG by adding 'pci=nommconf' in the kernel command line. The network operating system kernel will then try to address the devices via the BIOS or directly but not by mmconfig function. This work around is agreed officially by Red Hat.