nix-files/hosts/by-name/moby/kernel.nix
Colin 6ec3126321 moby: fix display driver reload check to run before unl0kr
this should fix the no-graphics-on-boot bug i'm seeing. it was previously fixed for lightdm and greetd: just not unl0kr
2024-03-02 19:50:50 +00:00

92 lines
4.0 KiB
Nix

{ pkgs, ... }:
let
dmesg = "${pkgs.util-linux}/bin/dmesg";
grep = "${pkgs.gnugrep}/bin/grep";
modprobe = "${pkgs.kmod}/bin/modprobe";
ensureHWReady = ''
# common boot failure:
# blank screen (no backlight even), with the following log:
# ```syslog
# sun8i-dw-hdmi 1ee0000.hdmi: Couldn't get the HDMI PHY
# ...
# sun4i-drm display-engine: Couldn't bind all pipelines components
# ...
# sun8i-dw-hdmi: probe of 1ee0000.hdmi failed with error -17
# ```
#
# in particular, that `probe ... failed` occurs *only* on failed boots
# (the other messages might sometimes occur even on successful runs?)
#
# reloading the sun8i hdmi driver usually gets the screen on, showing boot text.
# then restarting display-manager.service gets us to the login.
#
# NB: the above log is default level. though less specific, there's a `err` level message that also signals this:
# sun4i-drm display-engine: failed to bind 1ee0000.hdmi (ops sun8i_dw_hdmi_ops [sun8i_drm_hdmi]): -17
# NB: this is the most common, but not the only, failure mode for `display-manager`.
# another error seems characterized by these dmesg logs, in which reprobing sun8i_drm_hdmi does not fix:
# ```syslog
# sun6i-mipi-dsi 1ca0000.dsi: Couldn't get the MIPI D-PHY
# sun4i-drm display-engine: Couldn't bind all pipelines components
# sun6i-mipi-dsi 1ca0000.dsi: Couldn't register our component
# ```
if (${dmesg} --kernel --level err --color=never --notime | ${grep} -q 'sun4i-drm display-engine: failed to bind 1ee0000.hdmi')
then
echo "reprobing sun8i_drm_hdmi"
# if a command here fails it errors the whole service, so prefer to log instead
${modprobe} -r sun8i_drm_hdmi || echo "failed to unload sun8i_drm_hdmi"
${modprobe} sun8i_drm_hdmi || echo "failed to load sub8i_drm_hdmi"
fi
'';
in
{
boot.kernelPackages = pkgs.linuxPackagesFor pkgs.linux-megous;
# boot.kernelPackages = pkgs.linuxPackagesFor pkgs.linux-manjaro;
# boot.kernelPackages = pkgs.linuxPackagesFor pkgs.linux_latest;
# alternatively, apply patches directly to stock nixos kernel:
# boot.kernelPatches = manjaroPatches ++ [
# (patchDefconfig kernelConfig)
# ];
# configure nixos to build a compressed kernel image, since it doesn't usually do that for aarch64 target.
# without this i run out of /boot space in < 10 generations
nixpkgs.hostPlatform.linux-kernel = {
# defaults:
name = "aarch64-multiplatform";
baseConfig = "defconfig";
DTB = true;
autoModules = true;
preferBuiltin = true;
# extraConfig = ...
# ^-- raspberry pi stuff: we don't need it.
# target = "Image"; # <-- default
target = "Image.gz"; # <-- compress the kernel image
# target = "zImage"; # <-- confuses other parts of nixos :-(
};
# disable proximity sensor.
# the filtering/calibration is bad that it causes the screen to go fully dark at times.
boot.blacklistedKernelModules = [ "stk3310" ];
boot.kernelParams = [
# without this some GUI apps fail: `DRM_IOCTL_MODE_CREATE_DUMB failed: Cannot allocate memory`
# this is because they can't allocate enough video ram.
# see related nixpkgs issue: <https://github.com/NixOS/nixpkgs/issues/260222>
# TODO(2023/12/03): remove once mesa 23.3.1 lands: <https://github.com/NixOS/nixpkgs/pull/265740>
#
# the default CMA seems to be 32M.
# i was running fine with 256MB from 2022/07-ish through 2022/12-ish, but then the phone quit reliably coming back from sleep (phosh): maybe a memory leak?
# `cat /proc/meminfo` to see CmaTotal/CmaFree if interested in tuning this.
"cma=512M"
# 2023/10/20: potential fix for the lima (GPU) timeout bugs:
# - <https://gitlab.com/postmarketOS/pmaports/-/issues/805#note_890467824>
"lima.sched_timeout_ms=2000"
];
# services.xserver.displayManager.job.preStart = ensureHWReady;
# systemd.services.greetd.preStart = ensureHWReady;
systemd.services.unl0kr.preStart = ensureHWReady;
}