REPO DEPRECATION

This repository is deprecated in favor of msm8998-mainline/linux on GitLab where development has continued where this tree left off :)

linux-mainline-oneplus5

Linux mainline kernel fork focused on OnePlus 5 (cheeseburger) / 5T (dumpling) devices.

linux-mainline-oneplus5

Building

Dependencies

Ubuntu 20.04 LTS

# apt install build-essential flex bison libssl-dev bc kmod gcc-aarch64-linux-gnu mkbootimg

Kernel and modules

export ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu-
make oneplus5_defconfig # see arch/arm64/configs/
make -j$(nproc)
make INSTALL_MOD_PATH=rootfs modules_install -j$(nproc)
tar -czf modules.tar.gz -C rootfs/lib/modules . --owner=0 --group=0 --numeric-owner

NOTE: In case you want to boot without a ramdisk you need to make menuconfig after generating a .config and tweak the following:

# change sde21 (system) below to match your desired rootfs partition
CONFIG_CMDLINE="clk_ignore_unused loglevel=3 quiet root=/dev/sde21"
CONFIG_CMDLINE_FORCE=y

Otherwise see e.g. my initramfs-tools after building; the cmdline below will contain rd.* option examples that are specific to my ramdisk, such as booting a loopback-based rootfs.img on the userdata partition root :)

Boot image

Next install osm0sis/mkbootimg to create an Android bootloader compatible boot image:

DEVICE="cheeseburger" # change depending on which device you're targeting
cat arch/arm64/boot/{Image.gz,dts/qcom/msm8998-oneplus-$DEVICE.dtb} > Image.gz-dtb
mkbootimg --kernel Image.gz-dtb --ramdisk /path/to/initramfs.cpio* \
    --cmdline "clk_ignore_unused loglevel=3 quiet rd.rootfs=/dev/sda13/rootfs.img rd.extra_hooks=configfs,rndis" \
    --base 0x0 --pagesize 4096 -o boot.img

Updating an existing boot.img

Same steps as above apply, except instead of mkbootimg run:

abootimg -u /path/to/boot.img -k Image.gz-dtb

NOTE: Kernel cmdline can also be easily tweaked using abootimg -u ... -c "cmdline=..." (assuming you’re not using CMDLINE_FORCE=y).

Manual deployment

Now you can use e.g. fastboot boot boot.img or fastboot flash boot boot.img && fastboot reboot to boot the newly compiled kernel.

Firmware

Currently probing ipa (built as a module) without firmware in place causes a kernel panic (as of v5.12). If you don’t wish to deal with this you can:

Set CONFIG_QCOM_IPA=n in make menuconfig and rebuild the kernel
Make a new boot image with modprobe.blacklist=ipa added to cmdline

Otherwise follow these steps to enable features such as Bluetooth and Wi-Fi:

wget https://github.com/JamiKettunen/firmware-mainline-oneplus5/archive/10.0.1.tar.gz
scp 10.0.1.tar.gz user@172.16.42.1:~
ssh -t user@172.16.42.1 "sudo sh -c 'mkdir -p /lib/firmware && tar -xmf 10.0.1.tar.gz --strip-components=1 -C /lib/firmware && find /lib/firmware/* -maxdepth 0 ! -name qca ! -name ath10k ! -name qcom -delete && rm 10.0.1.tar.gz && sync; echo Firmware deployed.'"

Kernel modules

Once booted you should have an SSH connection via USB RNDIS to the device, which means you can just scp over the previously made modules.tar.gz and extract it:

scp modules.tar.gz user@172.16.42.1:~
ssh -t user@172.16.42.1 "sudo sh -c 'mkdir -p /lib/modules && tar -xmf modules.tar.gz -C /lib/modules && rm modules.tar.gz && sync && reboot'"

NOTE: 172.16.42.1 is the default IP address my ramdisk assigns to the device’s USB interface from the RNDIS hook.

What works?

UFS internal storage
Samsung S6E3FA5/S6E3FC1 display panels
- There are some issues with refreshing, but the drivers are mostly usable
Synaptics touchscreen (& back/recents capacitive buttons on OP5) via RMI4
Capacitive button backlight on OP5 (see /sys/class/leds/white:kbd_backlight)
Adreno 540 GPU
- At least partially; there are various visual glitches currently on more complicated scenes
Bluetooth (requires firmware blobs)
Wi-Fi via ath10k SNOC (requires firmware blobs & 5 userspace daemons#WLAN))
- Somewhat unstable; FW keeps crashing without diag-router and on any type of network disconnect
- With ath10k_snoc driver probed the shutdown is additionally hung by some amount due to the remoteproc refusing to shutdown cleanly
Real-time clock (RTC)
- Reported timestamp is always somewhere around 1972
- Stored timestamp cannot be updated, but alarms do fire when written relative to the incorrectly reported timestamp#Real-timeclock.28RTC.29)
USB in peripheral mode
- No power delivery from the device’s USB port -> host mode practically useless for now
GPS via GPSD with Qualcomm PDS patch
Power and volume buttons
Haptics/vibration motor
Battery details query via fuel gauge
Hall effect sensor (for magnetic flip cases and whatnot)
Some cellular modem functions (tested via ModemManager/libqmi)
- PIN unlock
- Carrier info query
- Sending SMS
Notification light
- Patches for Qualcomm Light Pulse Generator from bamse’s WIP tree

What doesn’t work?

Any kind of audio input/output (except via Bluetooth)
Plenty of cellular modem functions (tested via ModemManager/libqmi)
- Receiving SMS
- Calls
- Data connection
Fingerprint sensors (& home button on OP5)
Any cameras (sensors: IMX{398,350,376K,371})
Camera flashlight
- Downstream driver here
Hardware video acceleration via Venus
- angelogioacchino.delregno@somainline.org/">Pending SDM660 patches exist on LKML; could be adapted somewhat easily for MSM8998?
Practically all sensors
- Needs “low power island” bringup + drivers as most of the sensors are controlled by this DSP
Fast charging
- This could perhaps be implemented when the proprietary dashd Android blob is reverse engineered
NFC
- While the PN544 I2C chip has a driver and can be hooked up in device tree, there appears to be an SPI “secure element” chip handling some of the tasks (and potentially even an Android blob? check on downstream!)
CPUfreq scaling
- This worked back on v5.11 but causes a kernel panic when probes since v5.12, waiting for updated patchset from Angelo that can hopefully get merged upstream
Probably everything else I forgot to mention

Other issues

Performance
- A stock postmarketOS install for example takes 40 seconds to boot; about half of the time is spent decompressing a 40 MB archive which should take only a few seconds when compared with downstream and SDM845 mainline
Battery life
- Primary sources of power consumption should be measured (how?)
Might be a nitpick but there’s some amount of ugly spam about clocks getting stuck and such in dmesg

OnePlus 5T specific quirks

Android 10 bootloader (OxygenOS 10.0.1 FW) is the only one where I’ve experienced fastboot boot working so far
- Particularly it doesn’t work with OxygenOS 9.0.11 FW flashed
If appending more than one DTB (e.g. for cheeseburger + dumpling) the device won’t boot anymore
- Perhaps downstream (DTBs) can be investigated as downstream kernel boots fine with many unordered DTBs appended?