Broken EFI Support in the Raspberry Pi Kernel

[alsldlflglhljlk]

Describe the bug

For some reason, compiling the Raspberry Pi kernel with EFI boot support breaks many features of Raspberry Pi OS. For example, I lose the following features when booting with EFI: • The ability for audio output • WiFi support • The ability to power off (I get [23634.13552] Rebooting system but it never turns off) • Bluetooth support • Crashes with overclocking • etc

However, when booting with the stock Debian kernel, have none of the above issues (except bluetooth), but I don't have /dev/gpiomem, /dev/vcio, nor /dev/vc-mem. This is why I wish to enable the EFI stub in the Raspberry Pi kernel.

Steps to reproduce the behaviour

1. Download the latest kernel source git clone --depth=1 --branch=rpi-6.2.y https://github.com/raspberrypi/linux
2. Download my configuration cd ./linux && make -j distclean && wget https://pastebin.com/raw/AVmn0AgF -O ./.config
3. Update configuration in case anything has changed make -j oldconfig
4. Compile the kernel make -j6 dtbs Image modules
5. Install devicetree and modules make -j6 dtbs_install modules_install
6. Paste the kernel in /boot sudo cp ./arch/arm64/boot/Image /boot
7. Download EFI Firmware and unzip its contents in /boot.
8. Reboot
9. Press F1 when you see the Raspberry Pi Logo to enter the EFI Shell.
10. Type fs0: in the EFI shell.
11. Type Image root=/dev/mmcblk0p2 rootwait ro (replacing /dev/mmcblk0p2 with whatever device you boot from)

All of the above steps have been performed on a Raspberry Pi, I have not cross-compiled anything

Device (s)

Raspberry Pi 4 Mod. B

System

1. cat /etc/rpi-issue Not found
2. vcgencmd version Dec 12 2022 11:56:56 Copyright (c) 2021 Broadcom version ed6f6b8fcdc6476410b9cf75d141633461d34bdd (clean) (release) (start)
3. uname -a Linux raspberrypi 6.2.16-v8+ #5 SMP PREEMT Sun Jul 16 18:48:02 CDT 2023 aarch64 GNU/Linux

Logs

No response

Additional context

No response

[pelwell]

Your config (when saved as a defconfig file) shows a lot of differences from bcm2711_defconfig:

$ diff arch/arm64/configs/bcm2711_defconfig  defconfig
2a3,4
> CONFIG_KERNEL_XZ=y
> CONFIG_DEFAULT_INIT="/sbin/init"
45c47
< CONFIG_CMDLINE="console=ttyAMA0,115200 kgdboc=ttyAMA0,115200 root=/dev/mmcblk0p2 rootfstype=ext4 rootwait"
---
> CONFIG_CMDLINE="console=ttyAMA0,115200 kgdboc=ttyAMA0,115200 root=/dev/sda2 rootfstype=ext4 rootwait"
60a63,64
> CONFIG_NVHE_EL2_DEBUG=y
> CONFIG_PROTECTED_NVHE_STACKTRACE=y
65a70,72
> CONFIG_MODULE_SIG=y
> CONFIG_MODULE_SIG_FORCE=y
> CONFIG_MODULE_SIG_SHA512=y
66a74,75
> CONFIG_MODULE_DECOMPRESS=y
> CONFIG_TRIM_UNUSED_KSYMS=y
449a459
> CONFIG_FIRMWARE_MEMMAP=y
451c461,469
< # CONFIG_EFI_VARS_PSTORE is not set
---
> CONFIG_FW_CFG_SYSFS=m
> CONFIG_FW_CFG_SYSFS_CMDLINE=y
> CONFIG_SYSFB_SIMPLEFB=y
> CONFIG_EFI_ZBOOT=y
> CONFIG_EFI_BOOTLOADER_CONTROL=y
> CONFIG_EFI_CAPSULE_LOADER=y
> CONFIG_EFI_TEST=y
> CONFIG_RESET_ATTACK_MITIGATION=y
> CONFIG_EFI_COCO_SECRET=y
459a478
> CONFIG_ZRAM_DEF_COMP_ZSTD=y
465a485
> CONFIG_BCM2835_SMI=y
707a728
> CONFIG_I3C=m

What is the minimal set of changes needed to show the problem? I can ignore DEFAULT_INIT and CMDLINE changes.

[pelwell]

Have you tried using your custom kernel without the EFI firmware?

[alsldlflglhljlk]

Have you tried using your custom kernel without the EFI firmware?

@pelwell Yes, and the only advantage of doing so is the ability to power off / reboot.

[pelwell]

If the kernel works as expected without the EFI firmware (i.e. if you have audio output, WiFi support, e.tc.), then isn't the problem in the EFI firmware?

[alsldlflglhljlk]

@pelwell I'm currently recompiling the kernel with only a minimal set of changes.

CONFIG_FW_CFG_SYSFS=m CONFIG_FW_CFG_SYSFS_CMDLINE=y CONFIG_SYSFB_SIMPLEFB=y CONFIG_EFI_ZBOOT=y CONFIG_EFI_BOOTLOADER_CONTROL=y CONFIG_EFI_CAPSULE_LOADER=y CONFIG_EFI_TEST=y CONFIG_RESET_ATTACK_MITIGATION=y CONFIG_EFI_COCO_SECRET=y

is necessary because otherwise I wouldn't be able to see the earliest kernel messages and wouldn't be able to see if my kernel panicked at the very beginning.

[alsldlflglhljlk]

If the kernel works as expected without the EFI firmware (i.e. if you have audio output, WiFi support, e.tc.), then isn't the problem in the EFI firmware?

No, if I boot without the EFI firmware, I can power off correctly. That's the only difference.

[pelwell]

Ah, you mean difference from with EFI firmware, not difference from the standard kernel without EFI.

[pelwell]

I'm keen to try with a minimal set of changes when you have it. As I said, my limited set of changes showed no problems without the EFI firmware.

[alsldlflglhljlk]

(don't mind the crazy idea) Is it possible to patch the mainline kernel to include support for /dev/gpiomem, /dev/vcio, etc?

[pelwell]

Of course it's possible - you just need to copy across the drivers, config settings and possibly some Device Tree settings.

[alsldlflglhljlk]

Alas, it was not meant to be. I reconfigured my kernel with the smallest possible amount of changes to bcm2711_defconfig, but I still observe the same behaviours I described earlier. I think the best way to do this would be to copy over the required files to the mainline kernel. I think I know the drivers I'll need to copy over, but I'm not sure how to tell the configuration how to compile those drivers.