Changes/EnableDrmPanic

Enable Drm Panic

This is a proposed Change for Fedora Linux.
This document represents a proposed Change. As part of the Changes process, proposals are publicly announced in order to receive community feedback. This proposal will only be implemented if approved by the Fedora Engineering Steering Committee.

Summary

Drm_panic is a new feature in the Linux kernel that displays a panic screen when a kernel panic occurs. This proposal is to enable DRM_PANIC in the Fedora kernel, to improve the kernel panic user experience.

Owner

• Name: Jocelyn Falempe, Javier Martinez Canillas
• Email: <jfalempe@redhat.com>, <javierm@redhat.com>

Current status

• Targeted release: <VERSION>/ Fedora Linux <VERSION>
• Last updated: 2024-07-11
• [Announced]
• [<will be assigned by the Wrangler> Discussion thread]
• FESCo issue: <will be assigned by the Wrangler>
• Tracker bug: <will be assigned by the Wrangler>
• Release notes tracker: <will be assigned by the Wrangler>

Detailed Description

When the linux kernel panics in Fedora 40, in most cases, the screen just freezes. If you're in a VT console, you'll be able to see the kernel debug information, but that is pretty hard to understand for users that are not kernel developers. With this feature, they will see a message saying the computer has crashed, and they need to reboot the computer. Drm_panic has been introduced in kernel v6.10, but is still under active development.

In order to enable DRM_PANIC, you need to disable VT_CONSOLE in the kernel, this is to prevent a race condition, that if you are in a VT console when the panic occurs, both fbcon and drm_panic will write to the framebuffer at the same time, leading to corrupted output. https://patchwork.freedesktop.org/series/134831/ The drawback is that tty0 won't show the kernel kmsg, and it can be harder to debug boot issues. But plymouth already takes care of this, and can display the boot kmsg when no VT console is present. https://gitlab.freedesktop.org/plymouth/plymouth/-/merge_requests/224 And the user experience would be better, because plymouth has better font and color support than fbcon.

Supported drivers are simpledrm, mgag200, ast, (and imx, tidss, on aarch64). I'm working on nouveau support, and I hope i915 and amdgpu will add support too. If the driver is not supported, you won't see the panic screen, but it won't be worse than what you have today.

Drm panic provides different panic screens. The default is "user" which will display a simple friendly message telling the user to reboot the computer. But for kernel developers, you can also set it to "kmsg", to see the last kmsg lines (so this is equivalent to the current fbcon). You can select the panic screen in Kconfig, or as a module parameter (drm.panic_screen=user) or at runtime with "echo -n kmsg > /sys/module/drm/parameters/panic_screen"

I've also made a proof of concept to add a panic screen with a QR code with debugging information, which will make it easier for users to report kernel panic in Fedora. An example can be seen here: https://github.com/kdj0c/panic_report/issues/1

Benefit to Fedora

This change will improve the user experience when a kernel panic occurs.

It's also a first step to switch to userspace console, and being able to disable CONFIG_VT in the kernel. VT and fbcon are legacy part of the kernel, that would reduce maintenance burden if we can disable them, and It will also reduce CVE impact, as userspace vulnerabilities are usually less critical.

Scope

• Proposal owners:

Write documentation on how-to debug boot issues without VT_CONSOLE. Maybe also change the systemd log configuration, so that it default to writing the log to the console.

• Other developers:

• Release engineering: #Releng issue number

I'm unsure if it has impact on the installer.

• Policies and guidelines: N/A (not needed for this Change)

• Trademark approval: N/A (not needed for this Change)

• Alignment with the Fedora Strategy:

I think it perfectly fit the "Fedora is for everyone" goal, as the current kernel panic (either UI freeze or kmsg output in VT) is not user-friendly.

Upgrade/compatibility impact

Enabling DRM_PANIC should be transparent to user, but disabling VT_CONSOLE may have a visible impact. Fortunately since Fedora 40, plymouth is able to display the kmsg messages.

For non-graphical boot, you can use systemd.log_target=console systemd.log_level=info and remove rhgb and quiet to see the kernel boot message.

But this needs to be documented, and communicated, so that users that debug boot issues, know about this change.

Early Testing (Optional)

Do you require 'QA Blueprint' support? Y/N

How To Test

Currently the easiest way to test, is to use the simpledrm driver, as it can run on all hardware. So first blacklist your driver (i915, amdgpu or nouveau), and then boot and check that you're using simpledrm. then you can trigger a kernel panic with: echo c > /proc/sysrq-trigger

As it will crash your machine, it's also possible to do this in a VM (so disabling virtio-gpu, or vmwgfx)

Also to check that you can still see the kernel messages at boot, in the grub menu, remove the "quiet" kernel command argument, and you should still see the kernel boot messages on the plymouth screen.

Dependencies

The main dependency, is to have a kernel v6.10 or later. To still see the kernel boot messages, there is also a dependency on plymouth and systemd, but the versions in F40 are already good.

Contingency Plan

• Contingency mechanism: Revert the kernel configuration changes.
• Contingency deadline: N/A (not a System Wide Change)
• Blocks release? N/A (not a System Wide Change), Yes/No

Documentation

Kernel Kconfig for DRM_PANIC: https://elixir.bootlin.com/linux/v6.10-rc7/source/drivers/gpu/drm/Kconfig#L107