LibreEcho Linux for MT8163

This repository is the Linux kernel source for LibreEcho MT8163 targets, kept on real upstream stable history. It is intentionally separate from LibreEcho-Platform, which owns ARM32 product tooling, initramfs, feature packaging, OTA verification, and the historical 3.18 compatibility tree.

Status: this repository replaces LibreEcho-Linux-6.1, which is a flat snapshot without upstream history (LibreEcho-Linux-6.1#53). It builds the same kernel as LibreEcho-Linux-6.1 (same sources, same configuration); the production pipeline still builds from LibreEcho-Linux-6.1 until the build is migrated here.

Provenance and layout

• Upstream: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git

• Branch point: v6.1.178 (dc5c83b7f5f83ea99aea5c771b1ec77458263a15)

• main is v6.1.178 plus the LibreEcho commits, one per subsystem, replayed from LibreEcho-Linux-6.1 ac890a53b229. The replayed tree matches that snapshot except for unreferenced stale copies under arch/arm/boot/dts/mediatek/, and upstream dotfiles (.kunitconfig, .yamllint, tools/testing/selftests/arm64/tags) that were lost when the snapshot was imported.

• The MT8163 CONSYS connectivity stack (WMT/STP/BTIF, Wi-Fi, Bluetooth; about 213k lines of vendor code) lives in LibreEcho-mt8163-connectivity and is included here as a git subtree at drivers/misc/mediatek/mt8163-connectivity, from its linux-6.1-in-tree branch. It is still built into the kernel, as in LibreEcho-Linux-6.1. Make connectivity changes on that branch, then update the subtree:

  git subtree pull --squash --prefix=drivers/misc/mediatek/mt8163-connectivity \
    https://github.com/aslater3/LibreEcho-mt8163-connectivity linux-6.1-in-tree

  The connectivity repository's main holds the out-of-tree module build, which is planned as a later, separately reviewed change.

To list the LibreEcho changes:

git log --first-parent --no-merges v6.1.178..main

Stable updates

Stable releases are merged, not re-imported:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git tag v6.1.N
git checkout -b fix/linux-6.1.N main
git merge --no-ff v6.1.N

The merge still needs conflict review, the full ARM32 build and contract checks, and hardware acceptance where relevant. Renovate's .github/renovate-upstreams.env marker remains a review signal only.

Production kernel parity

Until a functional change is intended, this tree must build exactly the production kernel. .github/workflows/parity.yml runs libreecho/parity_check.sh, which builds this tree and the LibreEcho-Linux-6.1 revision pinned in libreecho/parity.json with pinned build metadata and requires a byte-identical zImage, DTBs, System.map and .config:

git clone https://github.com/aslater3/LibreEcho-Linux-6.1 ref
git -C ref checkout "$(jq -r .ref libreecho/parity.json)"
libreecho/parity_check.sh ref

When an intended functional change lands, move or remove the pin in the same pull request.

Next LTS canary

.github/workflows/next-lts-canary.yml runs libreecho/next_lts_canary.sh every night. The script replays the LibreEcho commits onto the latest 6.18.y release and reports which ones apply cleanly. The connectivity subtree is not replayed; the connectivity repository runs its own nightly compile canary against the same release.

The tree is not a claim that the MT8163 changes are upstream-ready. A clean kernel source baseline does not by itself constitute a complete boot image or stable OTA release; those require the paired product tooling and independent image verification.

Production audio contract

Final hardware-verified contract (2026-08-07): the programme is mono, but PCM 23 must use a two-channel transport with identical left/right samples. One-channel MonoRight / DACSETUP=0x24 is superseded: live output isolation showed clean woofer audio but noise or silence from the tweeter.

The Radar-Puffin speaker path is:

• producer input: stereo S16_LE, 48 kHz;
• programme processing: mix/downmix/duck/limit once to mono;
• hardware output: duplicated mono in S16_LE, 48 kHz, 2 channels;
• hardware PCM: PCM 23 / hw:0,23;
• board channel configuration: Stereo;
• codec route: normal DACSETUP=0x14 (left slot to left DAC, right slot to right DAC);
• HP/HPL/HPR speaker routes enabled; LO/LOL/LOR line-out routes disabled;
• Audio_DacMux_Setting=Off;
• PRB_P2 resource class 12, calibrated crossover, and IFACE3 DACMOD2BCLK are retained.

The shared audio engine renders one mono sample per frame and writes it to both PCM channels. This is not stereo programme audio; it is the transport required for the left/HPL tweeter and right/HPR woofer DAC paths. The userspace engine and packaging live in the separate LibreEcho-Platform tooling repository.

Build boundary

The production pipeline must pass this repository as LIBREECHO_KERNEL_SRC and pass the separate tooling repository as LIBREECHO_TOOLING_SRC. Do not use the tooling repository as the kernel source: it contains a Linux 3.18 tree and will produce incompatible DTC/build results when used as a 6.1 source.

Use an external output directory, for example:

make O=/path/to/kernel-output ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- \
  LD=arm-linux-gnueabihf-ld.bfd mt8163_arm32_defconfig
make O=/path/to/kernel-output ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- \
  LD=arm-linux-gnueabihf-ld.bfd olddefconfig
make O=/path/to/kernel-output ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- \
  LD=arm-linux-gnueabihf-ld.bfd zImage dtbs -j8

The final LibreEcho pipeline remains the authority for the stock-v184 Android boot envelope, initramfs, feature payloads, signing, and image verification. This repository must not be used to flash a device directly.

Kernel target scaffolding

libreecho/targets.json maps radar_puffin and biscuit to their kernel defconfig and dtb. Both currently use mt8163_arm32_defconfig and libreecho-radar-puffin.dtb: this is deliberate build parity, not Biscuit hardware acceptance. Biscuit will get its own reviewed DTS and hardware validation in a later change; no new DTS or audio/config fragments are added here.

Validate the table and exercise its CI planner without hardware:

python3 -B libreecho/validate_targets.py
python3 -B libreecho/test_targets.py

CI reads this table, builds each distinct (defconfig, dtb) tuple once, then uses a publication matrix over both targets to upload libreecho-kernel-<target>-zImage, -dtb, -System.map and -config artifacts. At parity these are copies of the same build outputs. The legacy Radar bundle libreecho-linux-arm32-<sha> is also retained. Changing Biscuit's tuple to valid in-tree sources adds a build lane automatically; unknown IDs or missing sources fail validation. The build still uses olddefconfig, zImage dtbs, the existing ARMHF GCC/linker flags and no config fragments, preserving Radar's CI recipe.

Contract tests

The LibreEcho source contract tests sit next to the code they check. Run them all with:

python3 -B libreecho/run_contract_tests.py

MT8163 pstore/ramoops source boundary

Issue #6 is intentionally source-blocked on the current kernel baseline. The production DT contains platform-owned reserved-memory nodes, including the vendor mediatek,ram_console region, but it does not identify a firmware-owned region that is safe to hand to ramoops. This tree therefore does not add a new memory address, reuse an existing reserved range, or enable only CONFIG_PSTORE / CONFIG_PSTORE_RAM in the kernel defconfig.

The production initramfs and its /data archival/rotation policy are owned by the separate LibreEcho-Platform repository, not by this kernel source tree. A complete implementation must be reviewed across all three layers before any one of them is enabled here:

1. firmware/LK and the production DT must agree on a retained, non-overlapping reserved region;
2. mt8163_arm32_defconfig must enable both pstore and ramoops support; and
3. the product initramfs must mount /sys/fs/pstore, archive records, and apply the approved rotation policy.

The host-side boundary is checked with:

python3 -B tools/mt8163_pstore_contract_test.py

That check prevents a partial source integration; it does not claim warm-reset retention, deliberate-panic capture, or diagnostics-surface availability. Those remain physical/product integration gates.

MT8163 Bluetooth HCI contract

The MT8163 Bluetooth bridge is built into the production kernel. The committed mt8163_arm32_defconfig explicitly enables the complete closure:

CONFIG_BT=y
CONFIG_BT_BREDR=y
CONFIG_BT_LE=y
CONFIG_MTK_BTIF=y
CONFIG_MTK_COMBO_BT=y
CONFIG_MTK_MT8163_BLUEZ_HCI=y

The bridge registers hci0 before userspace WMT preparation is complete. It therefore cancels the automatic HCI setup work queued by hci_register_dev() and marks the controller for deferred setup; the first SET_POWERED management command is allowed through that narrow gate only after WMT has configured the BTIF transport. The normal HCI open/setup sequence then clears HCI_SETUP and populates the controller information. Without this exception, Linux rejects the first controller-indexed management command while the transport is deliberately waiting for userspace preparation.

The source contract can be checked without hardware:

python3 -B tools/mt8163_bluetooth_contract_test.py

This static check does not replace live HCI controller-information, discovery, pairing, and connection acceptance on the target device.