Time: chrony replaces timesyncd, with NICT/JP-pool NTP primary and gpsd
SHM as an offline-only refclock; makestep any time and fake-hwclock
cover the battery-less RTC.
Antenna: the modem cleared the supply again after the service set it,
so the service is now idempotent, configurable (ANTENNA=active|passive,
passive keeps it off) and re-checked by a 30 s timer.
gpsd: a gpsdctl udev hook spawned a stray gpsd without -n at boot that
blocked the real service; gpsd.service now binds to the NMEA device.
GNSS: CGNSSMODE=15 (GLONASS, BeiDou, Galileo, QZSS) and all supported
NMEA sentences; both persist across a modem reset. Adds sim7600-at.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
UPS input comes from the ignition-switched socket via a 15 V PD
trigger, so input loss means cranking or ignition off. The INA219
(monitor-only) gives the battery current sign; a host service would mark
cranks, then after a grace period close the recording and power off,
with a voltage floor as backstop. Records the open hardware problem:
after poweroff the UPS keeps the board powered, so it drains and
doesn't auto-start at ignition on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
15-19 V in, 3S 21700 pass-through UPS with an INA219 that the car unit
sees on I2C bus 7 at 0x41 (read-only check: 12.34 V, idle). Bench test
held the Orin >=20 s across a supply swap. Notes the 20 V PD caveat and
the intended ignition-off shutdown via battery current direction.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An active antenna on the roof received nothing until the modem's
antenna supply was switched on (AT+CVAUXV=3050, AT+CVAUXS=1); then +5.5
dB over the passive one. The settings reset on every modem restart and
AT&W doesn't keep them, so interface 03 is reserved as /dev/sim7600-at
(ModemManager ignores it) and a udev-started oneshot re-applies them.
gpsd now re-attaches via gpsdctl after re-enumeration. Verified with a
live modem reset: antenna, LTE and the GNSS fix all recovered unattended.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
In the car with ignition on, the listen-only probe found 500 kbps (2673
frames in 3 s, zero RX errors; nothing at other rates). The OBD port
isn't gatewayed: 27 periodic IDs, ~893 frames/s. A udev rule now starts
can-up@can_pcan on boot and hot-plug with 500k and listen-only, to be
dropped when obd-poller becomes the single writer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
500 kbps is the likely era standard but unconfirmed for the Colt, so no
persistent bitrate yet. Adds a listen-only probe (installed on the car
unit as can-bitrate-probe) that cycles common rates without disturbing
the bus, the fallback if the OBD port is gatewayed and silent, and what
to persist once the rate is known.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Summarises the measured cost as roughly 3 s blank per PD state change,
with keep-awake in the app handling sleep, and notes that a marginal
charger renegotiating PD repeatedly would cut the link each time.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
No USB-C detach, one gadget reconnect and a fresh DHCP exchange. With
sleep ruled out, the hub's PD edges cost ~3.5 s (unplug) and ~2.2 s
(plug-in).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Idle on battery, the tablet suspends ~60 s after screen-off and only
answers at its DHCP-renewal wakeups while the USB link stays up. Kept
awake with a video, a PD unplug costs 3.1 s and then nothing. The
earlier replug silences were the tablet waking from sleep.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Unplug causes the full USB-C detach (~3.5 s total). Replug causes no
detach but ~10 s of disruption: the tablet goes silent while the Orin's
link stays up, then reconnects its gadget once. Only an app-level
heartbeat would see the replug outage.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three PD unplug/replug cycles each caused one full USB-C detach; the
link flapped ~3 times and settled within 5-12 s with no intervention.
Treated as a service-level requirement (yukart decision 24), not a
host fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tablet as USB host, Orin's stock L4T gadget as the network side: the
tablet chose NCM and got 192.168.55.100. Android only routes app traffic
over it once it is the default network, so the app has to bind to the
Ethernet network explicitly.
Mass storage is off (enable_ums=0); the config isn't a conffile, so an
nvidia-l4t-init upgrade restores it and the report says how to check.
The Orin's USB-C port (FUSB301, no PD) can't charge the tablet; the PD
hub on hand is the practical fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The car unit has no display of its own (the tablet app is the display),
and unified memory is better spent on the voice stack. GDM/GNOME no
longer start, a text login stays on tty1, and PulseAudio/PipeWire are
disabled for all users while ALSA stays for mic capture. Saves ~200 MB
idle; every step is a reversible disable, not a removal.
Also notes the pre-existing nightly logrotate failure from the NVIDIA
jetson-logging config.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
project-yukart's map-tile sync treats NetworkManager's metered flag as
a second guard against syncing over LTE; the flag was never set. It is
now, live on the car unit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the WiFi radio off, the default route moved to wwan0 within 10 s and
all IPv4/DNS/Tailscale samples passed. Tailscale went direct to halogen
over povo's global IPv6 (no DERP), and an SSH session over the tailnet
survived both switches. WiFi took back IPv4 within 15 s and IPv6 within
~45 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The backup pin doesn't charge, so a rechargeable ML2032 would just
drain with a third of the capacity. Adds the expected life (~1.5-2 yr
typical), polarity and post-fit test notes, and marks the item low
priority since NTP covers almost every online boot.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The dev kit carrier has pads but no connector populated, so the RTC
resets whenever power is removed. Captures the connector part NVIDIA's
docs reportedly specify, the community substitute, and the R560 note;
battery type and footprint still need checking against the schematic.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
timesyncd now uses ntp.nict.jp (stratum 1), with the JP pool and Ubuntu
as fallback, over whichever uplink holds the default route. Timezone
moves from Asia/Yakutsk to Asia/Tokyo: same offset, right place.
Records that the RTC loses time at power-off (the cold boot came up at
June 4 until NTP synced), and the planned chrony + gpsd setup once
GNSS has a fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With both adapters attached at power-on, pcan registered first and took
can0, and the udev rule renamed it to can_pcan, confirming the race and
the fix. qmi_wwan auto-loaded, povo auto-connected below WiFi, and gpsd
came up on the stable symlink.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The driver source sat in /tmp and was gone after a reboot, leaving the
"after a kernel update" steps unusable. Re-fetched PEAK 9.2.0 into
~/infra/peak-linux-driver-9.2.0 per GUIDELINES §1; it is the same
release as the installed pcan.ko (Release_20260707_n).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mttcan and pcan both load at boot, so the PCAN adapter could take can0
and push the onboard controller to can1. A udev rule now names them
can_onboard (platform c310000.mttcan) and can_pcan (USB 0c72:000c),
both outside the canN pool so a transient canN can never collide.
Also records why the SIM7600 needs nothing more: ModemManager and NM
don't use ttyUSB numbers, and gpsd already uses a symlink.
can_onboard is verified live; can_pcan is checked against the driver
source only (adapter not plugged in).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The L4T kernel ships without qmi_wwan/cdc-wdm, and PPP dial-up made KDDI
detach the povo SIM (attach APN was au's; PPP re-activating the default
bearer still detached). Built both modules out of tree from v5.15.148
into ~/infra/qmi_wwan, set the attach APN to povo.jp, and added an NM
connection at metric 1000 so LTE ranks below Ethernet and WiFi.
GNSS autostarts on the modem and gpsd reads /dev/sim7600-gps (udev
symlink on interface 01); no fix yet, tested indoors.
BOM: NMEA is on ttyUSB1 not ttyUSB3, and the data path is QMI, not
RNDIS/PPP.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The original report was written on another host. On this machine the
pcan module (NETDEV_SUPPORT) is built, installed under misc/, and loaded
at boot via /etc/modules; the adapter was not yet plugged in. Adds the
verification and post-kernel-update rebuild steps.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Orin Nano Developer Kit exposes no DSI connector — the module carries DSI
on SODIMM pins but the reference carrier board does not break them out, and
DisplayPort is its only display output. The Waveshare panel is also a
Raspberry Pi accessory and mechanically incompatible. It could not have been
connected without a custom carrier board.
Moot regardless: the tablet is the display, and no dedicated display is in
scope. The entry was still marked research pending / TBD.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013kj58Jugkic783b8ZrtTVZ
Replaces `type: default` — one value across every document, read by nothing.
Types are now reference, guidelines, plan and report; plans carry status and
updated, reports carry date. Convention documented in project-yukart REPOS.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013kj58Jugkic783b8ZrtTVZ
mikkeli
mikkeli