[SOLV]Kernel Crash Very Early in Boot Sequence (GPD MicroPC)

[tempuser]

Context

I've been trying to create a very small (~16GB) console only Gentoo installation on a GPD MicroPC in order to have rapid access to certain tools. I've been using the usual Gentoo documentation plus this page as a guide. The trouble is, it won't boot. The kernel doesn't even draw to the screen. In spite of the title of this topic, I don't even know that the handover to the kernel is taking place correctly, and I certainly don't know how to debug it, hence why I'm humbly asking for a little assistance.

What I have done

I have successfully installed the Gentoo system to a newly created ext4 partition using a live USB. I can arch-chroot into it and it appears to work just fine. I have also used efibootmgr to create the appropriate boot entry and set it as the first active one. I created the kernel by emerging sys-kernel/gentoo-sources (6.6.38), downloading the 5.2.1 kernel config from the handheld.computer page I linked above, and then running make menuconfig, and verifying everything looked sane (to my untrained eye anyway).

This might be the cause of the problem if make menuconfig didn't gracefully handle the jump from 5.2.1 to 6.6.38, or if the manufacturer changed some of the hardware during the intervening almost five years without making it obvious, but even then I would have expected the kernel to be able to write something to the screen using legacy BIOS operations (please correct me if I'm wrong).

[UPDATE] I have updated the config using the correct procedure of executing make oldconfig and answering the questions, the issue persists.

Things I know

I can confirm that CONFIG_FB_EFI=y is set in the kernel config. The kernel's file name also has the correct .efi extension.

I can be confident that the UEFI firmware can see the kernel in the ESP's filesystem as earlier during installation I accidentally put the kernel in the wrong directory (oops) and the boot failed but in a different way.

There is no initrd or initramfs in use.

The Symptoms

If I try to boot the kernel the system will just hang indefinitely. I have tried booting the kernel by:

• Turning the machine on and waiting.
  
• Going into the BIOS menu and selecting the appropriate boot entry.
  
• Manually booting the kernel via GRUB2 on the live USB.
  
• Going into the "built-in EFI shell", navigating to the kernel through the ESP's filesystem, and instructing to execute.
  

In all four cases the state of the screen prior to booting is different, and in all four cases whatever was previously on the screen persists indefinitely. Whenever this happens journalctl --list-boots will fail to see that any boot attempt was made at all, suggesting that the kernel really is failing early, and its not just a screen drawing problem (although as I mentioned above the kernel really should be able to draw to the screen using legacy BIOS operations).

Then again, I can only run journalctl when arch-chrooted into the Gentoo system from the live USB, so I don't know whether this test is valid or not. This may just be telling me when the live USB was booted.

[UPDATE] I now think that my boots attempts were showing up in journalctl, and not the live USB boot attempts, and that I just confused myself.

Other information

The built-in kernel command string is currently set to "root=/dev/sda2 fbcon=rotate:1 i915.lvds_downclock=1 i915.i915_enable_fb=1 i915.i915_enable_rc6=1 i915.enable_psr=1".

Running fdisk -l /dev/sda yields:

[pietinger]

[tempuser]

Thank you for the feedback pietinger. Sadly, no luck. I checked, and all the framebuffer options that should be enabled are. Recreating the kernel config from the 5.2.1 config using make oldconfig, answering the myriad questions (goodness the Linux kernel team have been busy these last five years), then adjusting with make menuconfig did not help either (after recompiling and installing the new kernel, naturally).

[kimchi_sg]

[tempuser]

Of course, here you go:

• .config
  
• lspci -nnk

Fingers crossed you can spot whatever blunder I've made.

[pietinger]

You really have a very clean kernel configuration ... a monolithic kernel without initramfs (like me):

[tempuser]

[pietinger]

[pietinger]

P.S.:

[tempuser]

I've enabled the NUMA settings and CONFIG_ACPI_PROCESSOR_AGGREGATOR, I've disabled APEI and PSTORE, and I've updated the built-in kernel command line to include fbcon=rotate:1. Sadly, no change (the screen hasn't even rotated). The relevant boot log just reads:

[pietinger]

Perhaps I have misunderstood something: I thought the system log (dmesg) was from a boot of your kernel - was it from the GentooCD boot ?

What exactly is the status when you boot your kernel with the options:

root=... ro earlycon=efifb efi=debug initcall_debug ignore_loglevel fbcon=rotate:1

?

What happens if you boot an UbuntuLiveCD ?
_________________
https://wiki.gentoo.org/wiki/User:Pietinger

[tempuser]

[pietinger]

[tempuser]

So here's an odd one. If I try to view the most recent boot log (#0) using journalctl -b I get the no entries message, if I try to view the second most recent boot log using journalctl -b -1 (#-1) then it does display, even if it wouldn't display before when it was the most recent boot log (#0). In other words, if I want to view the most recent boot log I need to boot the machine one more time in order to make it the second most recent boot log.

Anyway, here's the most recent boot log I can access (with the new kernel arguments you specified).

[pietinger]

Okay, I can imagine why you tried to switch it off:

[tempuser]

Sorry for taking so long to reply; I've had a much busier few weeks than anticipated.

Enabling CONFIG_X86_PAT seems to have made the system bootable. Thank you so much!

In truth, I had to navigate one last hurdle; for some reason I was getting a kernel panic related to not being able to open the root device. The listed "available partitions" were all "/dev/sdb*", i.e. there was nothing assigned to "/dev/sda". I changed the built-in command line to be of the form root=PARTUUID=<partition UUID> which fixed the problem.

The only remaining issues now are some strange error messages. This is an up to date boot log. My problem with only being able to see the penultimate boot log seems to have disappeared now that I am booting the machine and logging in successfully.

There is an "[drm] *ERROR* GPIO index 0 request failed (-2)" error message that I briefly see during boot (repeated several times). [UPDATE: apparently this is a "known issue"]

There is an "ucsi_acpi USBC000:00: possible UCSI driver bug 4" that echoes over the login prompt at the Linux terminal (also repeated several times).

Naturally, there may be other errors that I have missed.

[pietinger]

[tempuser]

Of course, this is my current .config file.

[pietinger]

Yes, I thought so: USB3 (CONFIG_TYPEC) and thunderbolt require PCI hotplug ... You have: # CONFIG_HOTPLUG_PCI_ACPI is not set

(See also: https://wiki.gentoo.org/wiki/USB/Guide#USB_Type-C_and_Thunderbolt )
_________________
https://wiki.gentoo.org/wiki/User:Pietinger

[tempuser]

I made the proposed changes, but sadly the error messages persist

This is my .config file as of now.

[pietinger]

Does you USB3 work ? (with an USB3 device) If yes, then maybe this error is not important. But you should in every case enable: CONFIG_PCI_QUIRKS is not set

(Sorry, I always think this is a configuration build from a default configuration where some options are enabled by default ... and then I forget to check it)

Do you have a Synaptics Designware PCI controler in this system ? (If yes: CONFIG_PCIE_DW_PLAT_HOST is not set)
_________________
https://wiki.gentoo.org/wiki/User:Pietinger

[tempuser]

[pietinger]