[SOLVED] linode, kernel, boot, grub

[pietinger]

[jankom]

efibootmgr - I'll explore that later when I have more quiet time.

I tried to copy bzImage file from linode /usr/src/linux-6.6.52-gentoo/ directory to linode /boot as vmlinuz-6.6.52-gentoo-x86_64; grub-mkconfig recognized it, but reboot ended up in kernel panic as before, probably because of missing initramfs. Rescued again, OK.

As far as not seeing a grub screen maybe explained by the linode doc here,
https://techdocs.akamai.com/cloud-computing/docs/manage-the-kernel-on-a-compute-instance
section Asynchronous SCSI scans. I tried those but so far no success.
I use the "Upstream kernel" which I interpret that the kernel they used to build my linode in 2023, and this is why it is 6.1.32 and not 6.6.52 (my latest eix-syn) with GRUB-2.

Maybe the linked linode document could shed some light on the situation for you all who have a lot more linux experience than me.

Again, thank you for your continued interest in this topic.
jankom (Janos)

[Ralphred]

[jankom]

I just ran efibootmgr on my home pc and on linode:

[pietinger]

Okay, boot your old 6.1-kernel and do a "lsmod". Give it us (if it is too long then please use wgetpaste also).
Then give us the kernel .config of your new 6.6 kernel.
(I will check if everything you need is static in there)

In the meantime do number 4 and number 5 of this chapter (if you have done it already do nothing):
https://techdocs.akamai.com/cloud-computing/docs/manage-the-kernel-on-a-compute-instance#installing-the-upstream-kernel-and-configuring-grub

In number 4 dont use the example; instead use this:

[pingtoo]

I found this thread very interesting, Especially the root is whole disk without partition and still able to use grub.

So further research lead me to grub-mkstandalone which I think explain how linode(Akamai) setup. Since grub-mkstandaline create a image kind like initramfs (but grub specific), it created a bootable image that grub load in to memory with embedded grup.conf in the memory disk which possibly scripted to load (root)/boot/grub.conf to finish boot.

So according to first document Run a Distribution-Supplied Kernel with PV-GRUB lead to second document Manage the kernel on a Compute Instance and the description in
View and modify the kernel in Cloud Manager GRUB 2 You should be able to use you manually created custom kernel to boot your linode compute instance. In the second document at bottom there is short section on how to create grub.conf can be use by GRUB 2 method.

In the first document it is said current linode use KVM for compute instance. I guess QEMU was involve, it is most likely the QEMU is setup to boot from the grub-mkstandalone created image file that is why first part of grub was invisible to Linux because it is in memory and got replaced by the time kernel start. However it is hard to guess if QEMU was setup with BIOS mode or UEFI mode to boot, I think grub-mkstandalone can generate both format.

I am not sure about your kernel configuration if it is sufficient for use with linode compute instance. But I notice that the kernel command line options for your old kernel

[jankom]

@plattinger:

[pietinger]

[pietinger]

jankom,

I guess you took the kernel .config from 6.1 and made a "make oldconfig" ... Okay, first step is correct.

Now you must go into "make menuconfig" and search with / for every module in this list and change it to static <*> (DONT try to edit the .config; do it in "make menuconfig")

[pietinger]

Moved from Portage & Programming to Kernel & Hardware.
_________________
https://wiki.gentoo.org/wiki/User:Pietinger

[jankom]

Let me first recompile kernel, then try /dev/sda and finally the grub screen tuning. - jankom

[pietinger]

[jankom]

Kernel recompiled, bzImage copied to /boot as vmlinuz... - see

[pingtoo]

jankom,

I think you can go ahead try to boot. if you follow pietinger suggested compile modules into static then you should be able go further into rootfs.

If you boot have problem in start services then you can try the recovery mode for your new kernel.

The "single' is to tell "init" only start enough for console. I think the 'lish' should allow you connect console. although I am not sure how that work. you may need to try set up kernel command line to let kernel know which tty is console.

[GDH-gentoo]

Doesn't the documentation from Akamai ask that certain variables in /etc/grub/default be set in a certain way? And a console= and other kernel parameters? I don't see that reflected in the latest grub.cfg...
_________________

[jankom]

Reboot into kernel 6.6.52 OK. grub screen is not there, although I have not mage the changes for lish.
But!
it does not seem to be connected to outside world, ping IP or domain name packets is lost.

Will rescue to previous kernel (I have no grub screen so I cannot select which kernel to boot)..

jankom

[jankom]

Rescue OK, back to "norlma", i.e. 6.1.32 kernel.
jankom
https://jgklinux.jangkom.com/

[jankom]

Rush to glory - I did not do make install-modules.

Stupid mistake.

It's OK now, Internet and 6.6.52 kernel.

The only thing left is the lish console so that I can have a grub screen for booting - not very important.

Thanks again, all of you. Great learning experience!

jankom (Janos)

[pietinger]

[jankom]

Just a final, curiosity question: how important it is to hav a /usr/src/linux symlinc pointing to the directory of the current running kernel?

[pingtoo]

[jankom]

Thanks - jankom

[jankom]

I'm copying here explanation from linode Customer Service regarding lack of grub screen:

[pietinger]

This thread could certainly help other users who use a Linode instance ... if they find this thread ... so may I ask you to add “Linode” (or Linode instance) to the title? (edit the headline in your first post)
_________________
https://wiki.gentoo.org/wiki/User:Pietinger

[jankom]

Yes