Why do I experience CPU lockups when running large volumes of data through some Mellanox cards in RHEL 6.5?

Solution Verified - Updated -

Environment

  • A system using the Mellanox Ethernet driver (mlx4_en) provided by the regular kernel or by a DUP package
  • Red Hat Enterprise Linux 6.5

Issue

Why do I experience CPU lockups when running large volumes of data through some Mellanox cards in RHEL 6.5?

Resolution

This issue is fixed in the latest RHEL 6.5 errata kernels and all RHEL 6.6 kernels.

If your system needed an mlx4_en Driver Update Program (DUP) package to install RHEL 6.5, the issue can be resolved in one of three ways:

  1. Install using the DUP package, then update to the latest RHEL 6.5.z kernel. Errata kernels 2.6.32-431.37.1.el6 and later contain a fix that resolves this problem.
  2. Install using the DUP package, then update to the RHEL 6.6 GA kernel or a 6.6 errata kernel. The RHEL 6.6 GA kernel, 2.6.32-483.el6, and all errata kernels for 6.6 contain a fix that resolves the problem.
  3. Install using RHEL 6.6, avoiding the need for this DUP package.

More information on DUP packages can be found at this link: https://access.redhat.com/articles/64322

If your system does not need a DUP package for installation, update to the latest RHEL 6.5 errata kernel, or 6.6 GA or errata kernel to resolve the problem.

Note: An issue may occur with the mlx4_en driver on RHEL 6.6 that is not related to the issue described above. The issue manifests as an oops error, and is being tracked in private Red Hat Bugzilla bug 1128611. Non-Red Hat employees will not be able to view that bug, but customers or partners with questions about the status of the issue can provide the link or bug number to their support contact who can check the status on their behalf.

This solution is part of Red Hat’s fast-track publication program, providing a huge library of solutions that Red Hat engineers have created while supporting our customers. To give you the knowledge you need the instant it becomes available, these articles may be presented in a raw and unedited form.