Error installing bootloader on SuperMicro with NVMe SSD disk

[ffuj]

I'm trying to install CentOS 6.10 (yes, I know it's almost EOL) on a SuperMicro 5019A-12TN4 SuperServer with an NVMe SSD disk. Regardless of whether I use kickstart-based text mode or interactive mode, Anaconda fails very late in the installation process with the message "There was an error installing the bootloader. The system may not be bootable." In text mode, at the point of the failure, I can use <alt-F2> to switch to an alternate console and read an addition error message from efibootmgr when it is invoked: "Fatal: Couldn't open either sysfs or procfs directories for accessing EFI variables. Try 'modprobe efivars' as root." The modprobe doesn't work.

The server is configured to use UEFI to boot. The kickstart file contains the following partition definitions:

Code: Select all

        part /boot/efi --fstype=efi --grow --size=20 --maxsize=200
        part /boot --fstype=ext4 --size=512
        part / --fstype=ext4 --size=30720
        part /altroot --fstype=ext4 --size=30720
        part /var --fstype=ext4 --size=10240
        part /opt --fstype=ext4 --size=10240
        part /diag --fstype=ext4 --size=15360
        part /home --fstype=ext4 --size=10240 --grow

Other relevant kickstart parameters:

Code: Select all

bootloader --location=partition --driveorder=nvme0n1 --append="LANG=en_US.UTF-8 SYSFONT=latarcyrheb-sun16 KEYBOARDTYPE=pc KEYTABLE=us quiet biosdevname=eth0"
zerombr
clearpart --all --drives=nvme0n1

Mounted filesystems at time of error (as seen from Anaconda):


Mounted filesystems on NVMe target disk:


What is causing this failure? Do I need to specify something else in my kickstart?

Thank you for any suggestions or tips to help solve this.

Jim

[TrevorH]

CentOS 6 predates NVMe M.2 drives by about 8 years. Red Hat stopped adding new hardware support to it about 3 years ago so I am unsurprised that it does not work. CentOS 6 uses grub legacy and that probably knows nothing about nvme at all.

[ffuj]

Fair enough... so I'll try replacing the NVMe with a SATA SSD and see what happens.

Thank you!

[ffuj]

Update: I replaced the NVMe disk with a SATA SSD and switched from UEFI boot to legacy boot mode, after which I was able to successfully install and boot CentOS 6.10.

[firaksd]

Is this really a disk issue? Why would disk restrict yours from installing centOS?
Infogeekers.

[TrevorH]

This is an nvme disk so it does not appear as /dev/sda or /dev/hda but rather as e.g. /dev/nvmen0 so lots of utilities within linux do not know that it is a disk. CentOS 6 also uses grub legacy that's been around since the early 1990s and hasn't been updated since grub2 was written in about 2002. If your bootloader doesn't understand that an NVMe disk is a disk then it stands very little chance of being able to boot from it.

[MartinR]

I've just found on this diagram on WikiCommons which may explain things. If you have a large monitor, click on the 1052x1380 display for more quality. Look down at the bottom and you'll see the SCSI mid-level box which contains the familiar /dev/sdan. That feeds down to the low-level drivers that understand different physical devices. Now cast your eye across to the right and you'll see "nvme device", follow the lines up and you can see that it uses it's own low-level and high-level drivers which are totally unrelated to SCSI. The different driver allows 64 data transfers over 4 paths and removes the bus bottleneck from SSDs. Note however the version number at the top of the page: kernel version 4.0. As always, Trevor's analysis is spot on: your software is just too old to see newer hardware.

[ffuj]

MartinR: Great find!! Thank you (and Trevor) for pointing that out.

Jim

[ffuj]

A little more research... NVMe has been supported by the Linux kernel since 3.3.

Support for NVMe was also back-ported into RHEL/CentOS 6. See these articles for more information:

https://nvmexpress.org/resources/driver ... formation/
https://itpeernetwork.intel.com/nvm-exp ... #gs.w43oe6