Error installing Mellanox OFED on CentOS 7.9.

Infrastructure & Networking Software And Drivers
#1

I’m trying to install the Mellanox OFED on a CentOS 7.9 that I have on a Singularity container with the following command:

./mlnxofedinstall --add-kernel-support --force --skip-repo

However, I get the following error:

/lib/modules/4.17.0-rc2/build/scripts is required to build mlnx-ofa_kernel RPM.

Please install the corresponding kernel-source or kernel-devel RPM.

Error: One or more required packages for installing OFED-internal are missing.

Please install the missing packages using your Linux distribution Package Management tool.

Run:

yum install kernel-devel-4.17.0-rc2

Failed to build MLNX_OFED_LINUX for 4.17.0-rc2

When I try yum install kernel-dev-4.17.0-rc2, I get the following error:

No package kernel-devel-4.17.0-rc2 available.

Error: Nothing to do

And when I try yum install kernel-dev, an older version 3.something is installed that does not fix the issue.

This is output of uname -s -r:

Linux 4.17.0-rc2

And this is the output of ls -l /usr/src/kernels/:

3.10.0-1160.53.1.el7.x86_64

3.10.0-1160.53.1.el7.x86_64.debug

4.17.0-0.rc2.git3.1.fc29.x86_64

Could you please let me know how I can fix this?

#2

Hello Mehran,

Thank you for posting your inquiry on the NVIDIA Networking Community.

As you want to install MLNX_OFED on a non-default kernel on CentOS 7.6, please make sure you have the following RPMS installed as mentioned in the RN under section → https://docs.nvidia.com/networking/display/MLNXOFEDv494170/General+Support+in+MLNX_OFED#GeneralSupportinMLNX_OFED-HardwareandSoftwareRequirements

Be aware that in some cases, it is not possible to rebuild the driver against the installed non-default kernel, if the kernel is much more recent than the MLNX_OFED version you want to install.

Also make sure you use the same GCC version on which your kernel is build.

Another method, is to run the ‘mlnx_add_kernel_support.sh’ script manually, by pointing towards the correct kernel source. Use the ‘help’ option for the available options.

Thank you and regards,

~NVIDIA Networking Technical Support

#3

Hi Martijn,

Thanks a lot for your response. I’m wondering if this is the correct approach I’m pursuing for enabling Infiniband on a Singularity container. I have not been able to find suitable instructions. I would be so grateful if you could kindly help with that.

(I need to compile MVAPICH (MPI for Infiniband) on my Singularity container which requires enabling IB on my container.)