nvidia-kmod-common is wiping out custom blacklist in /etc/default/grub

[jnm27]

We have ahci blacklisted in /etc/default/grub. But when we install nvidia-kmod-common, it wipes out that change.

https://github.com/NVIDIA/yum-packaging-nvidia-kmod-common/blob/main/nvidia-kmod-common.spec#L13

By default this is our command line:

GRUB_CMDLINE_LINUX="blacklist=nouveau,ahci rd.driver.blacklist=nouveau,ahci initcall_blacklist=efifb_driver_init crashkernel=256M rd.lvm.lv=rootvg/rootlv rd.lvm.lv=rootvg/swaplv rd.lvm.lv=rootvg/usrlv biosdevname=0 net.ifnames=1 audit=1 audit_backlog_limit=8192 rhgb quiet"

This is what happens when we install:

D: %post(nvidia-kmod-common-3:525.78.01-1.el8.noarch): scriptlet start
D: %post(nvidia-kmod-common-3:525.78.01-1.el8.noarch): execv(/bin/sh) pid 47301
D: Plugin: calling hook scriptlet_fork_post in selinux plugin
+ type -p grubby
/sbin/grubby
+ grubby --help
+ checkGrubby=0
+ '[' 0 -eq 0 ']'
+ /usr/sbin/grubby --update-kernel=ALL --args=rd.driver.blacklist=nouveau '--remove-args=nomodeset gfxpayload=vga=normal nouveau.modeset=0 nvidia-drm.modeset=1'
+ '[' '!' -f /run/ostree-booted ']'
+ '[' -f /etc/default/grub ']'
+ . /etc/default/grub
++ GRUB_TIMEOUT=5
+++ sed 's, release .*$,,g' /etc/system-release
++ GRUB_DISTRIBUTOR='Red Hat Enterprise Linux'
++ GRUB_DEFAULT=saved
++ GRUB_DISABLE_SUBMENU=true
++ GRUB_TERMINAL_OUTPUT=console
++ GRUB_CMDLINE_LINUX='blacklist=nouveau,ahci initcall_blacklist=efifb_driver_init crashkernel=256M rd.lvm.lv=rootvg/rootlv rd.lvm.lv=rootvg/swaplv rd.lvm.lv=rootvg/usrlv biosdevname=0 net.ifnames=1 audit=1 audit_backlog_limit=8192 rhgb quiet rd.driver.blacklist=nouveau'
++ GRUB_DISABLE_RECOVERY=true
++ GRUB_ENABLE_BLSCFG=true

[kmittman]

Hmm, I had thought we had solved this with #1 (merged as https://github.com/NVIDIA/yum-packaging-nvidia-kmod-common/commit/896d3dce59b4c09be31558ae3fee115fecf8891e) by first sourcing the existing GRUB config.

Actually, I wonder if this is due to the lack of double-quotes, which was added in #2 (merged in 530 as https://github.com/NVIDIA/yum-packaging-nvidia-kmod-common/commit/e7941a2c729ce0ffaf9a3236a7005e3f4f48b321).

@jnm27 could you please check if this reproduces with 530 driver packages?

Also might be helpful if you could post cat /proc/cmdline from reboot after installation, to see if it is preserving the previous config partially or not at all.

[jnm27]

yeah, we actually can't reboot after this happens, because we need the ahci blacklist in order to boot. The cmdline in the output above does get set.

I don't think #2 would fix it because the code in that commit only applies if GRUB_CMDLINE_LINUX doesn't already exist in /etc/default/grub

[jnm27]

I'd say do something like, get the existing blacklist param, and if nouveau is already in the param, don't do anything, else append ,nouveau to it.

[jnm27]

It might be nice to have an option to not modify grub at all, for users that know that nouveau needs to be blacklisted and already have that in their config.

[jnm27]

It's actually the grubby --args which is overwriting it, after which point it's too late to check the existing GRUB_CMDLINE_LINUX, because it would be overwritten already

[jnm27]

To workaround this, I'm just going to overwrite checkGrubby=1 in the spec file to skip the check.

[scaronni-nvidia]

This is being overhauled for a newer release of the drivers like the following:

• https://github.com/negativo17/nvidia-kmod-common/blob/master/nvidia-kmod-common.spec#L74-L80
• https://github.com/negativo17/nvidia-kmod-common/blob/master/nvidia-boot-update