Repository navigation
Fix goodix_ts DKMS auto-rebuild, thermald status check, doc gaps - #19
pirsqr-rnd wants to merge 10 commits into
Conversation
… OSK setup iio-sensor-proxy silently exits with "No sensors or missing kernel drivers" if i2c-dev isn't loaded (find_accels() can't open /dev/i2c-* without it) -- not obvious from the service's own error message. acpi_call is separately required for the EC-level keyboard/touchpad toggle in tablet mode; without it, screen rotation still works but SW_TABLET_MODE-based toggling at the EC level is skipped. Also documents wiring up an on-screen keyboard (wvkbd-deskintl) for tablet mode on Niri, since there's no popup-on-focus mechanism like GNOME/KDE have, plus how to reach Polish diacritics via the Compose key.
DKMS_VERSION was "$(KVER)-minibook1" (e.g. "7.2.3-minibook1"), making
the /usr/src/goodix_ts-7.2.3-minibook1 source directory contain two
hyphens. Pacman's dkms hook (/usr/share/libalpm/scripts/dkms) derives
module name/version by regex-splitting "name-version" directory names
on the last hyphen, so it parsed this as name="goodix_ts-7.2.3",
version="minibook1" instead of the correct name="goodix_ts",
version="7.2.3-minibook1" -- then called dkms install with that wrong
identifier, which always failed ("Bad return status for module
build"). Confirmed reproducing on two separate kernel updates (7.2.2
and 7.2.3) in /var/log/pacman.log.
Every other module here uses a static, hyphen-free PACKAGE_VERSION
(1.0, 1.2.2, ...) so only goodix_ts hit this. Switching to
"$(KVER).minibook1" leaves exactly one hyphen in the directory name,
in the right place, so the pacman hook's automatic rebuild-on-kernel-
update should now work without manual intervention.
The services check called bare "thermald --version", which fails whenever /usr/local/sbin (or /usr/sbin, for the install-arch path) isn't on the invoking user's PATH -- reporting a healthy, correctly patched thermald install as "NOT FOUND (WRONG)". Added find_thermald() to fall back to the known install locations before giving up.
|
Thank you for the PR! Regarding the new "Login screen" section:
Are you sure? I'm finding it hard to believe neither the bootloader nor the VBT patch would reach the login screen. |
…ironment warning Fixed via a pacman hook that patches niri-session in place on every package update, per the fix shared in niri-wm/niri#254 (maintainer endorsed local patching over changing the default). Also note a separate, intermittent Niri panic on session quit (libseat backend race during teardown) - unrelated to the flash, functionally harmless so far.
# Conflicts: # GUIDE.md
The merge of origin/main silently dropped the header for the desktop-rotation/on-screen-keyboard content (left it orphaned under '## Optional: higher refresh rate') and duplicated the vbt_patch build step. Restructured: - desktop rotation + on-screen keyboard back under their own '## Optional: screen auto-rotation on other desktops' header - refresh rate content under its own '## Optional: higher refresh rate' header, keeping both the numbered explanation and upstream's added suspend/resume caution note - removed the orphaned 'On Fedora, sudo dnf remove iio-sensor-proxy' fragment left over from the old numbered install steps - fixed two internal anchor links (fstanis#7-iio-sensor-proxy, fstanis#8-vbt-patcher-display-refresh-rate) that pointed at section headers removed by the merge
check-status.sh silently skips the VBT line without intel_vbt_decode; document which package provides it.
The i2c-dev/acpi_call requirements paragraph (from c06ef5d) got stranded between 'Reboot, then run:' and the check-status.sh command during the origin/main merge, and its content was already fully duplicated (in more detail) in docs/iio-sensor-proxy.md's Runtime requirements section. - Removed the stranded paragraph; check-status.sh now follows directly after 'Reboot, then run:' - Added a troubleshooting note in its place that reflects current reality: bootstrap-arch.sh loads acpi_call automatically now, but not i2c-dev, so that's the one still worth calling out - Fixed docs/iio-sensor-proxy.md's cross-link, which pointed at the removed 'fstanis#7-iio-sensor-proxy' anchor
install_modules() called 'make -C modules/$m install' directly, bypassing the Arch PKGBUILDs added in modules/*/packaging/arch/ - unlike install_sensor_proxy(), which already builds and installs iio-sensor-proxy through makepkg + pacman -U. Consequence: re-running this script (which GUIDE.md documents as safe to do any time, e.g. after a kernel upgrade) would silently overwrite pacman-managed module installs with unpackaged files under /usr/src, desyncing the pacman database. Rewrote install_modules() to mirror install_sensor_proxy()'s makepkg + pacman -U pattern for all four modules. Added clean_unpackaged_dkms_source()/clean_unpackaged_file() so a prior 'make install' (raw DKMS, not a package) is migrated in place on first run instead of conflicting with pacman -U - verified this is a no-op once a module is already pacman-owned, and that the actual migration path works end-to-end (manually walked all four modules through dkms remove + rm -rf /usr/src/... + makepkg -si on this machine, including the goodix_9110_cfg.bin firmware file conflict this same helper now handles automatically). Syntax-checked with bash -n; not run end-to-end as root since the manual walkthrough already exercised the equivalent commands.
…creen fstanis questioned this on PR fstanis#19 (comment on 2026-09-12), correctly -- the kernel cmdline and VBT patch both set the DRM panel-orientation property, which any DRM client honors regardless of what greeter/session is running, contradicting the 'none of the methods above' claim two sections up on the very same page. Confirmed live on this machine: commented out the manual 'transforms' line in noctalia-greeter's greeter.toml and logged out. The login screen stayed correctly rotated with no config at all. What actually does not reach the login screen is the iio-sensor-proxy *dynamic* rotation - it is a userspace daemon that only starts inside the desktop session, after greetd already rendered its first frame. Clarified that distinction instead of just deleting the wrong claim, since the manual greeter.toml rotation config below is still the right answer for anyone relying on the proxy alone with no static rotation.
|
You were right to push back on that — sorry, that line was wrong. I tested it live just now: commented out the manual The kernel cmdline and VBT patch both set the DRM What doesn't reach the login screen is the |
|
Closing this one — the branch behind it (my fork's I'll re-open the individual fixes here as focused PRs against current |
Summary
Three independent fixes/additions found while running this on a CHUWI MiniBook X after a kernel update broke touchscreen resume behavior:
DKMS_VERSIONused a hyphen ($(KVER)-minibook1), which pacman'sdkmshook mis-parses when deriving module name/version from the/usr/src/<name>-<version>directory name (it splits on the last hyphen). This made the automatic rebuild-on-kernel-update silently fail every time (Error! Bad return status for module build), confirmed in/var/log/pacman.logacross two separate kernel bumps. Switching to a dot fixes the parse without any functional change to the module itself.thermaldservice check called barethermald --version, which fails whenever/usr/local/sbin(or/usr/sbinfor theinstall-archpath) isn't on the invoking user'sPATH— reporting a correctly installed, correctly patchedthermaldasNOT FOUND (WRONG). Added a small fallback that checks the known install locations.iio-sensor-proxyrequiresi2c-dev(silently exits with "No sensors or missing kernel drivers" otherwise) andacpi_call(needed for EC-level tablet-mode keyboard toggle), plus how to wire up an on-screen keyboard for tablet mode on Niri (no popup-on-focus mechanism like GNOME/KDE).Test plan
goodix_ts: reproduced the pacman-hook failure on two real kernel updates (7.2.2, 7.2.3), confirmed the fix installs cleanly and the hook's own directory-name regex now parses correctly (verified against/usr/share/libalpm/scripts/dkms'sall_nv_from_kver)check-status.sh: confirmedthermaldnow reports correctly (2.5.11.minibook1, enabled, running) instead of(WRONG)