Isolated by experiment (2026-08-09, journal/modem): flash boot alone
zeroes modem NV; erase dtbo and idle fastboot visits are safe; on-device
boot-deploy (incl. apk kernel upgrades) never trips it. Likely ABL
anti-tamper on boot-image overwrite under an unlocked bootloader.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Full-system audit findings: without an explicit provider selection the
image installed real PulseAudio (Plasma Mobile hard-path), which claims
the card and ships a wireplumber fragment disabling hardware.audio -
zero sinks forever. Mirror the dev phone's world: pipewire backend meta
+ pipewire-pulse + echo-cancel in extra_packages, pulse meta banned via
apk conflict in fp6-device-tweaks. The two hand-made /etc audio
fragments (wireplumber no-ACP rule, AEC pair for VoLTE speakerphone)
move into soc-fairphone-fp6-audio.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Full-system audit on the flashed phone found the IMU dead: the journal's
documented reinstall-loss trio (inv-icm42600-spi modules-load - fallback-only
SPI node never coldplugs; the accel poll udev rule for auto-rotate; the
iio-sensor-proxy libssc-crash restart drop-in). All device-critical, none
personal - now packaged.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two device-critical pieces previously applied only by the dev restore
script: masked sleep targets (idle-suspend resets the WCN combo firmware
and kills wifi ~40min - a device defect, not personalization) and imsd
enabled at boot, gated on /etc/imsd.env existing so unconfigured systems
don't boot a failing unit. A stranger now needs exactly: flash, wifi via
UI, write imsd.env, restart imsd.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
3 of 38 glib tests SIGABRT under qemu-user emulation; they pass on
native builders and do not touch the patched GNSS paths.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
pmbootstrap's install-time build plan is not topologically sorted; the
versioned makedep made that a hard error instead of a stale-dep build.
Explicit bottom-up build restores order; no-op when current.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
pmbootstrap's build plan is not dependency-ordered: MM built first and
grabbed the cached 1.39.0 libqmi from the volume's package repo. The
versioned makedepend states the real requirement and forces our bumped
libqmi to build first. Kernel pkgver gains the source commit date so
registry-subscribed phones receive kernel updates.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CI publishes every locally built apk to the instance's Alpine registry
(catbot credentials via the PACKAGE_TOKEN secret; step skips until it
exists). catcrafts-fp6-repo ships the registry signing key and appends
the repo URL - NOT yet in extra_packages: the registry must have its
first content before image builds may reference it (an empty repo's
missing APKINDEX would break apk in the build chroots).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The forgejo VM occasionally 500s under crawler load - a clone blip
should cost 10 seconds, not the run. dist/ keeps its inode so a shell
cd'd into it no longer breaks the next podman start.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Our MM patches require qmi-glib >= 1.39.1 (the version bumped right after
!470 merged); Alpine's April snapshot reports 1.39.0. Upstream 30f3e998
carries both the TLV work and the version bump, so the local patch goes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
apk cannot satisfy libmm-glib=<ver>-r100 before modemmanager r100 itself
is built; Alpine's reuse of depends_dev inside makedepends only works
when the binary repo already has the same pkgrel.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
git -C <repo> format-patch -o . writes relative to the -C directory, not
the caller's cwd - the files landed in work/ModemManager and the aport
shipped an APKBUILD referencing patches that did not exist.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
First flash boot reached systemd and died into emergency mode on
'FAT-fs (loop0p1): IO charset ascii not found': the tested phone config
never needed FAT NLS tables because the old install used an ext2 /boot;
the current device package uses a vfat ESP. Single flip, olddefconfig
roundtripped.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
libqmi: the upstream-merged (!470, in no release yet) LOC Register Events
client-identification TLVs, taken verbatim from upstream main (532de37e).
modemmanager: the AFW engine unlock + Position Report derivation (!1463,
still draft upstream, review-hardened and verified on device). Both as
patches on Alpine's git-snapshot aports, pkgrel=100.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
imsd turns out to be fully public already (0.2.7 snapshot + emergency
stage 1 + the make build path) - no tag needed, same git-archive dance as
the kernel. callaudioshim was a load-bearing unpackaged home-dir script on
the phone (callaudiod replacement, working in-call mic mute); now an aport
autostarted at phase 1. kde-telephony's modem daemon autostart is hidden
via /etc/skel from the imsd package, mirroring the phone's user override.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Files copied byte-identical from the tested dev phone; installed to the
exact deployed paths (the spaced tplg/UCM filenames come from the machine
driver's card name). Wired into the image via extra_packages.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Partition nodes (loopNpM) only appear via the host devtmpfs; a privileged
container's own /dev never gains them. Drop the static mknod dance, warn
when /dev is not the host's.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Alpine's packaged pmbootstrap predates pmaports' master->main rename and
fails reading channels.cfg. Pinning also matches the version the rest of
this script was written against.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
'pmbootstrap config' demands an existing config file, so write
pmbootstrap_v3.cfg ourselves (INI, aports/device/ui/systemd). Explicit
set -eu because 'sh script' ignores shebang flags - run 3 shipped a
3.5KB 'image' because failures didn't propagate. dist/ is now the only
build-user-owned path, so the tag step's git works again.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Learned from the pmbootstrap 3.11 source: aports overlays must replace
upstream dirs in one tree (duplicate pkgnames are a hard error), pmbootstrap
refuses root (drop to a sudo-capable build user), and export symlinks are
boot.img + fairphone-fp6.img. Stage 1 builds without imsd until the v0.3.0
tag exists. Temporary push trigger for bring-up.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Kernel aport forked from pmaports 7.1.2-r0, repointed at milos-linux
combined-stable with the tested FP6 config (+EFI_ZBOOT for packaging);
imsd + nftables aports from the pending pmaports submission. build.sh
carries TODO(validate) markers - not yet run end-to-end.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
