User:Z3ntu/Fairphone2-mainline

From postmarketOS Wiki

Component support table

Component Codename Mainline kernel Since when
Volume & camera buttons gpio-keys Working v4.15-rc1
Power key pm8941-pwrkey Working already upstream
Internal storage sdhci1 Working v4.15-rc1
SD card storage sdhci2 Working v4.16-rc1
Old Display otm1902b Kind-of working out-of-tree
Old Touchscreen Synaptics DSX? Kind-of working out-of-tree
New Display s6d6fa1 Working out-of-tree
New Touchscreen Ilitek ili2120 Working v5.19-rc1
Vibration motor TI DRV2603 Working v5.3-rc1
Notification LED qpnp / lpg Working v6.0-rc1
Magnetometer AK8963 No driver[1]
Gyroscope & Accelerometer LSM330D No driver[1]
Proximity & Light Murata LT1PA01 No driver[1]
WiFi WCN3680/prima Working v5.6-rc1
Bluetooth WCN3680 Working v5.6-rc1
FM WCN3680 No driver
GPU Adreno 330 Working out-of-tree
Modem msm8974-mss-pil Working v5.6-rc1
Old Front Camera OV2685 Working (WIP) out-of-tree
Old Rear Camera OV8865 Working (WIP) out-of-tree
New Front Camera OV5670 Working (WIP) out-of-tree
New Rear Camera OV12870 No driver
  1. 1.0 1.1 1.2 Connected to the DSP and the driver is implemented in the ADSP firmware. Would need a kernel or user space implementaion talking QMI to the ADSP to plumb the sensors to user space where it can be used for e.g. rotation.

Test notes

Power key

evtest /dev/input/event0

Volume & camera buttons

evtest /dev/input/event1

Touchscreen

evtest /dev/input/event2

Vibration motor

fftest /dev/input/event3

Internal storage

ls /dev/mmcblk0*

SD card storage

ls /dev/mmcblk1*

Display

Framebuffer console should display the kernel messages

ls -al /dev/fb0
echo 255 > /sys/class/backlight/fd922800.dsi.0/brightness

GPU

kmscube

Notification LED

echo 255 > /sys/class/leds/rgb\:status/brightness

WiFi / Bluetooth / FM

echo start > /sys/class/remoteproc/remoteprocX/state
# set mac address according to "MAC addresses" section
# wifi and bluetooth interfaces should just work

Regarding FM:

someone was doing some initial work, but i got the feeling that without lpass running it didn't want to play ball

MAC addresses

$ qmicli -d /dev/wwan0qmi0 --dms-get-mac-address wlan
$ ip link set dev wlan0 address 12:34:56:78:9A:BC

$ qmicli -d /dev/wwan0qmi0 --dms-get-mac-address bt
# Note, that the bluetooth mac address is backwards (e.g. BC:9A:78:56:34:12 is actually 12:34:56:78:9A:BC)
# You can reverse the mac address using the following script:
$ echo BC:9A:78:56:34:12 | awk 'BEGIN{FS=OFS=":"} {s=$NF; for (i=NF-1; i>=1; i--) s = s OFS $i; print s}'
$ btmgmt public-addr 12:34:56:78:9A:BC

It looks like the Bluetooth initialization (including querying the MAC address via QMI and setting it) is done with hci_qcomm_init on downstream (init.qcom.bt.sh). It looks like there is no official public source version available for that.

For WiFi this seems to be done with libwcnss_qmi.so, a source version of that can be found here.

Sensors

Notes from IRC:

the client would need to implement the sensor qmi protocols, which we haven't tried to open up

qmi is just a method of encoding structured data; exactly like protobuf

libqmi implements a set of protocols that uses qmi to encode its data

so there's another set of 10-15 protocols, that uses qmi to encode messages, related to sensors

i think these are implemented in libsensors1.so - or something like that

but iirc libsensors1 is tied to the downstream AF_MSMIPC, and won't work on AF_QIPCRTR

Further notes: https://gitlab.freedesktop.org/mobile-broadband/libqmi/issues/21

Modem

The modem gets detected with ofono but you need to do additional manual setup because it's a Dual SIM device: https://wiki.postmarketos.org/wiki/User:TravMurav/Dual-Sim_QMI_draft

The msm-modem-uim-selection package should handle this but hasn't been tested yet.

Mobile Data

Upstream kernel contains rmnet core driver (equivalent to downstream msm_rmnet.c) but not the platform specific part (or rather transport specific part). ./drivers/net/ethernet/msm/msm_rmnet_bam.c is (likely) used on downstream msm8974. Newer Qualcomm devices use the IPA driver but 8974 is older than that. The actual transport used is arch/arm/mach-msm/bam_dmux.c.

Mobile Data is working! At the moment these resources are required:

The above should all be integrated now.

Cameras

Old modules:

  • Front: OV2685
  • Back: OV8865

New modules:

  • Front: OV5670
  • Back: OV12870 (no sensor driver in Linux)

The cameras can produce a picture that can be displayed using the libcamera qcam test utility. Possibly other libcamera-based apps also might work.

Currently based on a separate kernel branch (might change), you anyways also need to rebuild your dtb with the correct camera sensors manually enabled. There's no dynamic runtime detection support. Not sure if that's even possible in Linux itself.

Replace /boot/dtbs/qcom-msm8974pro-fairphone-fp2.dtb and run sudo postmarketos-mkinitfs to update boot.img in /boot/.

Then you can try running qcam:

# Adjust environment variables to be able to launch app via SSH
eval $(cat /proc/`pidof phosh`/environ | tr '\0' '\n' | grep -v '^PS1=' | awk '{ print "export \"" $0 "\"" }')

OV2685

QT_QPA_PLATFORM=wayland qcam -v -r gles

OV8865

v4l2-ctl -d /dev/v4l-subdev19 -c exposure=2500
QT_QPA_PLATFORM=wayland qcam -v -r gles -s width=3264,height=1836

(all other sizes don't deliver any data correctly)

OV5670

QT_QPA_PLATFORM=wayland qcam -v -r gles -s width=1280,height=720

Camera Flash LED

Old camera module

The flash in the old camera module is controlled by the pm8941 and this hardware block is accessible on the address @d300. The qcom-lpg driver used for the notification LED is relatively similar but it should be a separate driver for the camera flash.

A driver was posted to LKML in 2020.

New camera module

For the new camera module (12MP one) the following driver is being used:

# cat /sys/devices/fda0c000.qcom,cci/c6.ti,flash0/leds/torch_dual/device/uevent
DRIVER=leds-lm3644
OF_NAME=ti,flash0
OF_FULLNAME=/soc/qcom,cci@fda0C000/ti,flash0@c6
OF_COMPATIBLE_0=ti,lm3644
OF_COMPATIBLE_N=1
MODALIAS=of:Nti,flash0T<NULL>Cti,lm3644

The situation seems to be more complicated than with the old module as the lm3644 driver takes over parts of the driver controlling pm8941@d300, which might be complicated to solve nicely on mainline.

The lm3644 is an i2c device sitting on the camera interface i2c bus. There's no driver in mainline for the lm3644 yet.

http://www.ti.com/lit/ds/symlink/lm3644.pdf

Audio

An initial (unsuccessful) test with out-of-tree patches from flto was done with the following config options: CONFIG_SLIMBUS, CONFIG_SLIM_QCOM_CTRL, CONFIG_SLIM_QCOM_NGD_CTRL, CONFIG_QCOM_APR, CONFIG_SND_SOC_WCD9320, CONFIG_SND_SOC_QCOM, CONFIG_SND_SOC_LPASS_*, CONFIG_SND_SOC_QDSP6_*, CONFIG_SND_SOC_MSM8996.

Headphone audio kind of works with one or both of these branches:

Others

See Brian Masney's Nexus 5 mainline to-do list.

Kernel config options

Base config: qcom_defconfig

Kernel config option Function
CONFIG_USB_ETH_RNDIS USB networking
CONFIG_INPUT_PM8941_PWRKEY Power key
CONFIG_DRM_PANEL_SIMPLE Display
CONFIG_DMA_CMA Also display
CONFIG_TOUCHSCREEN_ILI210X Touchscreen
CONFIG_INPUT_GPIO_VIBRA Vibration motor
CONFIG_LEDS_QCOM_LPG Notification leds
CONFIG_WCN36XX WiFi
CONFIG_BT_QCOMSMD Bluetooth
CONFIG_BATTERY_BMS Battery fuel gauge
CONFIG_SYSCON_REBOOT_MODE Reboot mode
CONFIG_I2C_QCOM_CCI Camera Control Interface

cmdline

earlycon=msm_serial_dm,0xf991e000 PMOS_NO_OUTPUT_REDIRECT clk_ignore_unused pd_ignore_unused cma=500m msm.vram=192m msm.allow_vram_carveout=1