commaai / agnos-builder Goto Github PK

current rw partitions

• /persist - we should probably make this ro
  • we put the SSH host keys on there now. these should move to userdata
  • should only need to remount rw in provisioning to write the keys
• /cache - just handle corruption in agnos by reformatting
• /dsp - probably doesn't need to be rw?
• /firmware - also probably doesn't need to be rw?
• /data - we should try to unmount cleanly
• /data/media - ditch the last segment and unmount cleanly once we can make sure the rlog ends up on userdata

Useful for the updater, since we can't use the hash (takes too long) or agnos version (not unqiue).

Remove unused files and services in userspace/debs. Also break them up into smaller packages with individual functions/goals.

• PyTorch (the xx version?)
• pyside2, with shiboken generator

All these sleeps add up to at least 7 seconds of the startup time:

• initial four second sleep before weston init
  

  agnos-builder/userspace/files/weston.service

  Line 12 in de9a165

  ExecStart=/bin/bash -c "sleep 4 && \

• blocking on weston startup in comma service
  

  agnos-builder/userspace/usr/comma/comma.sh

  Line 14 in de9a165

  sleep 0.5 # weston's still starting after the socket's created

• sleep in openpilot launch script for release builds, since they show a UI too quickly
  

  agnos-builder/userspace/usr/comma/comma.sh

  Line 14 in de9a165

  sleep 0.5 # weston's still starting after the socket's created

For recent UI and networking improvements

Restarting the modem (lte + ModemManager services) from a cold boot while AGNOS is already booted seems way faster than normal startup. Is the modem service just started late in the boot process?

df2433ef9f6dcaa5|2022-02-25--11-39-54--boot

commaai/openpilot#24302

Currently, ModemManager only seems to try the first APN for a provider. It will keep trying to connect and fail with the first APN for the provider.

Need to make sure openpilot's updater handles this properly too

There should be a nice way to change the default directory for only the main tmux session to /data/openpilot instead of cd /data/openpilot in the bashrc. Ideally, this is done in openpilot's launch script. AGNOS doesn't need to know anything about /data/openpilot.

AGNOS onnx issue:

comma@tici:/data/openpilot$ python3 -c 'import onnx; onnx.load("/data/openpilot/selfdrive/modeld/models/supercombo.onnx")'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/usr/local/pyenv/versions/3.8.10/lib/python3.8/site-packages/onnx/__init__.py", line 11, in <module>
    from onnx.external_data_helper import load_external_data_for_model, write_external_data_tensors, convert_model_to_external_data
  File "/usr/local/pyenv/versions/3.8.10/lib/python3.8/site-packages/onnx/external_data_helper.py", line 14, in <module>
    from .onnx_pb import TensorProto, ModelProto
  File "/usr/local/pyenv/versions/3.8.10/lib/python3.8/site-packages/onnx/onnx_pb.py", line 8, in <module>
    from .onnx_ml_pb2 import *  # noqa
  File "/usr/local/pyenv/versions/3.8.10/lib/python3.8/site-packages/onnx/onnx_ml_pb2.py", line 33, in <module>
    _descriptor.EnumValueDescriptor(
  File "/usr/local/pyenv/versions/3.8.10/lib/python3.8/site-packages/google/protobuf/descriptor.py", line 755, in __new__
    _message.Message._CheckCalledFromGeneratedFile()
TypeError: Descriptors cannot not be created directly.
If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.
If you cannot immediately regenerate your protos, some other possible workarounds are:
 1. Downgrade the protobuf package to 3.20.x or lower.
 2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).

More information: https://developers.google.com/protocol-buffers/docs/news/2022-05-06#python-updates

Setting PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python does work, but we should fix if we can

When the power drop monitor triggers the ui is correctly killed, but the screen stays on.

tici -> tici-447087e4

Wi-Fi networks using only the WPA3 security protocol are showing the 🚫 symbol next to them, and cannot be joined.

• The Qualcomm 845 does seem to support WPA3 at a hardware level. (Source: XDA Developers)
• Discord user nelsonjchen said the missing Security Flag might be "NM_802_11_AP_SEC_KEY_MGMT_SAE". (Link to post)

See the network "DingoAteYourBaby" in the attached screenshot, which is only using WPA3. The IOT network current connected, has the exact same settings, on the same Access Point, except it is using WPA2.

The param written by the power drop monitor doesn't always seem to get written. Investigate this.

Hi comma teams,

I wanna to run "../launch_openpilot.sh" on C3 via ssh terminal.
But when "radard" and "controlsd" run the "config_realtime_process" inside "../common/realtime.py".
It gets "PermissionError: [Errno 1] Operation not permitted".

How should I fix this error ?

Thanks!!

Example: 075d96dd1286fdc9|2022-08-27--22-32-44--boot

When the power drop monitor triggers, append the current timestamp (seconds since boot) to a param. Openpilot can take care of making sure this gets cloudlogged and ends up in the qlogs next boot.

Will AGNOS work on onePlus3T or Letv Pro 3 phones？

Move commaai/openpilot#24175 into AGNOS

open questions:

• can we just clear the initial EPS bearer settings on each boot instead?
• where do the initial EPS bearer settings come from? SIM?

May just work with a newer version after #54

Something like limiting the max brightness 5% every hour the display is on. Also, limit the max brightness when no UI is running.

This should work:

sudo systemctl stop weston
sudo systemctl start weston

The weston service also shouldn't run as root, otherwise we'll need to keep hacking around the XDG_RUNTIME_DIR after restarts.

• ONNX
• pyopencl (build with PYOPENCL_CL_PRETEND_VERSION="2.0")
• pytest
• markdown https://github.com/commaai/openpilot/blob/599b1ab3ff5bc75603469916a8ecda644657dc34/common/tests/test_markdown.py#L2

Sometimes comma three's boot up into a "low battery state". In this state, some cores don't come online.

Repeatedly prints this:

Apr 28 18:59:35 tici kernel: FG: is_batt_empty: batt_soc_rt_sts: 32 vbatt: 4392078 uV msoc:0
Apr 28 18:59:35 tici kernel: FG: is_batt_empty: batt_soc_rt_sts: 32 vbatt: 4392078 uV msoc:0
Apr 28 18:59:36 tici kernel: FG: is_batt_empty: batt_soc_rt_sts: 32 vbatt: 4392078 uV msoc:0
Apr 28 18:59:36 tici kernel: FG: is_batt_empty: batt_soc_rt_sts: 32 vbatt: 4392078 uV msoc:0
Apr 28 18:59:36 tici kernel: FG: is_batt_empty: batt_soc_rt_sts: 32 vbatt: 4392078 uV msoc:0
Apr 28 18:59:36 tici kernel: FG: is_batt_empty: batt_soc_rt_sts: 32 vbatt: 4392078 uV msoc:0

from bootlogs/fbbb958f3157f063/2022-04-28--18-59-39.bz2

• set no cache dir
• TMPDIR needs to be big and not a non-tmpfs
• hook pip install to remount the rootfs as rw

Currently, sudo mount -o rw,remount / && sudo $(which pip) install scipy works.

We currently use the closed source Thundercomm version of Weston.

Get open source Weston working in AGNOS for a $5000 bounty.
https://github.com/wayland-project/weston

Hi comma teams,

How could I build my own customized "/usr/comma/setup" to disable the PubMasters of "gpsLocationExternal" and "sensorEvents" ?
Because I wanna to launch the "tools/sim/bridge.py" inside C3.

Thanks !!

Seeing about a 10% improvement in controlsd with a quick test. Also worth trying to build capnproto with LTO and PGO.

Part of the remaining jankiness with the supercaps is keeping the device up between unplugging and replugging. If that time is too short, then systemd will complete a proper shutdown, which can take a long time. We'd still like to properly shutdown, so let's try to just make it faster (10s?).

On ~10% of boots, the dongle needs to be unplugged to get it working. I haven't seen this happen with other aux peripherals like the red panda.

The way I have my C3 set up typically leaves the secondary usb port plugged into a computer in my car. This causes me a problem when the device reboots or is powered off from inactivity. As when it comes back on it boots into fastboot and the only way to disengage is to remove power wait for caps to discharge and then return power with it unplugged from this computer.

• Requesting that fastboot has a button i could tap on screen to exit fastboot
• a timeout so if no fastboot commands detected in X minutes it continues to boot
• or if it detects that the usb interface is disconnected it exits fastboot.

thanks