gl-e750-oled: add new package by ServPol · Pull Request #30695 · openwrt/packages

ServPol · 2026-10-04T17:54:49Z

📦 Package Details

Maintainer: @ServPol

Description:
Reworked after review: this is now only the OLED service for the GL.iNet GL-E750 (Mudi). The MCU transport, the battery (power_supply) and the console handling moved to the kernel/board side in openwrt/openwrt#25649 (kmod-gl-e750-mcu), which this package depends on.

gl-e750-oled is a small ucode service that builds the screen content (Wi-Fi SSIDs/keys per band, client count, uplink type, cellular state, VPN state, clock) and writes it as JSON to /dev/gl-e750-mcu.

reacts to netifd interface events and hostapd notifications instead of polling; only the clock and the cellular signal (uqmi, QMI only) are refreshed once a minute
JSON comes from ucode's encoder, no hand-made escaping
uplink type from the protocol of the interface holding the default route and the Wi-Fi mode of its network; VPN state from interfaces with a VPN protocol; nothing hardcoded
no /etc/inittab changes and no direct UART access anymore

Known issue: with the ucode currently in main, stopping the service occasionally crashes ucode on exit after hostapd notifications were received. That is ucode-lang/ucode#438, fixed upstream; the running service is not affected.

🧪 Run Testing Details

OpenWrt Version: main (6.18.55) with ath79: gl-e750: add status MCU driver (OLED, battery) openwrt#25649
OpenWrt Target/Subtarget: ath79/nand
OpenWrt Device: GL.iNet GL-E750 V1 and GL-E750 V2

The screen shows the clock, Wi-Fi bands, SIM state and follows clients joining and leaving within a second; battery data comes from /sys/class/power_supply/gl-e750-battery.

✅ Formalities

I have reviewed the CONTRIBUTING.md file for detailed contributing guidelines.

openwrt-ai · 2026-10-04T18:30:21Z

+# ttyS0 is wired to the status MCU, not to a user console. A login prompt on
+# it reads the MCU replies, so turn it off as the vendor firmware does.
+# procd reads /etc/inittab at boot, so this takes effect after a reboot.
+sed -i 's|^::askconsole:|#::askconsole:|' /etc/inittab


Nothing reverts this edit: there is no postrm/prerm that restores ::askconsole:, and /etc/inittab is in base-files' keep list, so the serial console stays disabled after apk del, and after sysupgrades too. Add a Package/gl-e750-mcu/postrm that runs the reverse sed (guarded by IPKG_INSTROOT and board name).

Generated by Claude Code

fixed, thanks

Generated by Claude Code

@BKPepe thanks for the review, reworked along those lines:

kernel side: ath79: gl-e750: add status MCU driver (OLED, battery) openwrt#25649 adds kmod-gl-e750-mcu, a serdev driver bound from the DTS. It owns the MCU UART, exposes the battery as a power_supply (gl-e750-battery), turns the MCU's power-switch event into an orderly poweroff and takes screen updates through /dev/gl-e750-mcu. The board DTS sets console=ttynull, so /etc/inittab is no longer touched.

this package is now only gl-e750-oled, a small event-driven ucode service (netifd/hostapd ubus events, JSON via ucode's encoder). It depends on kmod-gl-e750-mcu, so it can only be merged after adblock-fast: Makefile bugfixes and hijack rework #25649.

Tested on GL-E750 V1 and V2 with OpenWrt main. Marking ready for review.

openwrt-ai · 2026-10-04T18:30:24Z

+	[ "$enabled" -eq 1 ] || return 0
+
+	procd_open_instance
+	procd_set_param command /usr/sbin/gl-e750-mcu


The reload trigger never restarts the daemon. With no reload_service, reload just runs start again, and procd keeps the instance because its params are unchanged. A changed interval/hide_psk/modem_every therefore only takes effect after a manual restart. Let procd track the config file:

Suggested change

procd_set_param command /usr/sbin/gl-e750-mcu

procd_set_param command /usr/sbin/gl-e750-mcu

procd_set_param file /etc/config/gl-e750-mcu

Generated by Claude Code

fixed, thanks

Generated by Claude Code

openwrt-ai · 2026-10-04T18:30:27Z

+	# Keep kernel messages off the MCU uart, like the vendor firmware does.
+	echo "1 4 1 7" > /proc/sys/kernel/printk
+	stty -F "$TTY" 115200 raw -echo -echoe -echok -echoctl -echoke \
+		-icrnl -opost -onlcr -isig -icanon -iexten min 0 time 1


With min 0 time 1, a read() that sees no data for 100 ms returns 0. The cat "$TTY" in mcu_query (line 46) treats that as EOF and exits, so a reply is only caught if the MCU answers within about 100 ms, and the sleep 2/kill never comes into play. min 1 time 0 keeps cat blocking until it is killed, which seems to be the intent:

Suggested change

-icrnl -opost -onlcr -isig -icanon -iexten min 0 time 1

-icrnl -opost -onlcr -isig -icanon -iexten min 1 time 0

Generated by Claude Code

fixed, thanks

Generated by Claude Code

ServPol · 2026-10-04T19:11:07Z

Thanks, all three addressed in b2ea3b9 and re-tested on a GL-E750 V2 (25.12.5):

postrm restores ::askconsole: on removal (board-guarded, skipped in IPKG_INSTROOT)
procd_set_param file /etc/config/gl-e750-mcu: a config change now restarts the daemon, a plain reload with no change keeps it
tty set to min 1 time 0, so cat blocks until killed; battery data is read as before

openwrt-ai

Reviewed 1 new commits; no new issues found.

Generated by Claude Code

ServPol · 2026-10-05T07:43:45Z

Relaxed DEPENDS from @TARGET_ath79_nand to @TARGET_ath79 (34a90e6): with the nand-only dependency the CI skipped the package on every arch ("unsupported architecture"), while the mips_24kc job uses the ath79/generic SDK, so it is now actually built and tested there. On other boards the service and uci-defaults still do nothing (board_name check). The earlier powerpc_464fp failure was a SDK download checksum mismatch, unrelated.

BKPepe

I don't think this should be merged in its current form.

The GL-E750 MCU support itself is useful, but this package mixes two different responsibilities:

hardware/MCU transport and battery telemetry
userspace collection of network/UI state for the OLED

The first part should not be implemented as a polling shell daemon talking directly to /dev/ttyS0. The MCU is a real hardware controller connected through UART, and Linux already provides the appropriate serial/serdev infrastructure for protocol devices. Battery telemetry should also be exposed through the standard power_supply interface rather than a private /var/run/gl-e750-mcu/battery file.

The userspace part can then remain small and only handle the OLED presentation. It can consume normal OpenWrt/kernel/network state and send the resulting JSON to the MCU.

There are also several problems with the current implementation itself: it rebuilds JSON with destructive "escaping", polls UCI/ubus/hostapd/QMI periodically instead of reacting to state changes, uses heuristics for uplink/VPN state, hardcodes vpn_status=off, and modifies /etc/inittab to disable the serial console.

In particular, changing /etc/inittab from a package is a strong indication that the UART/MCU relationship should be handled as part of board/kernel support rather than by an application installed later.

I would therefore suggest splitting this into:

an OpenWrt/kernel change for the GL-E750 MCU device and UART/serdev driver
standard power_supply exposure for the battery state
a small userspace OLED service/package on top of that.

The OLED functionality itself is useful and worth keeping, but I don't think the current shell daemon is the right abstraction for upstream OpenWrt.

openwrt-ai

Reviewed 1 new commits; no new issues found.

Generated by Claude Code

ServPol · 2026-10-05T08:46:45Z

Thanks for the thorough review, agreed. I will rework it along those lines:

a serdev driver for the GL-E750 MCU (kmod in openwrt/openwrt, DT node under the UART), exposing the battery via power_supply and a small interface for the OLED payload
moving the console away from ttyS0 as part of the GL-E750 board support instead of editing /etc/inittab
a small event-driven OLED service on top (proper JSON encoding, no uplink/VPN heuristics)

Converting this to draft until the kernel side is up and tested on V1/V2; I will link the openwrt/openwrt PR here.

Show the router state on the OLED of the GL.iNet GL-E750 (Mudi): Wi-Fi SSIDs/keys per band, client count, uplink type, cellular state, VPN state and the clock. The MCU transport and the battery are handled by kmod-gl-e750-mcu in openwrt/openwrt; this service only builds the screen content and writes it to /dev/gl-e750-mcu. It is a ucode program that reacts to netifd interface events and hostapd notifications instead of polling; only the clock (once a minute) and the cellular signal via uqmi (once a minute, QMI only) are refreshed periodically. JSON is produced with ucode's encoder. The uplink type comes from the protocol of the interface that holds the default route and the Wi-Fi mode of its network, the VPN state from interfaces with a VPN protocol. Tested on a GL-E750 V1 and V2 with a main build that includes kmod-gl-e750-mcu: the screen follows clients joining and leaving within a second, Wi-Fi changes, and shows the clock and SIM state. With the ucode currently in main, stopping the service occasionally crashes ucode on exit after hostapd notifications were received; that is ucode-lang/ucode#438 (fixed upstream), it does not affect the running service. Assisted-by: Claude:claude-opus-5-5 Signed-off-by: Serv Pol <servpol@w420.ru>

ServPol · 2026-10-05T21:36:15Z

@BKPepe reworked as you suggested, see the reply in the review thread above. The kernel side (serdev driver, power_supply, power switch, console=ttynull in the DTS) is in openwrt/openwrt#25649; this package is now only the gl-e750-oled userspace service on top of it. Ready for another look.

openwrt Bot added Add package not following guidelines labels Oct 4, 2026

ServPol force-pushed the gl-e750-mcu branch from 2a2c0f0 to d876fe8 Compare October 4, 2026 17:56

openwrt Bot removed the not following guidelines label Oct 4, 2026

openwrt-ai reviewed Oct 4, 2026

View reviewed changes

ServPol force-pushed the gl-e750-mcu branch 2 times, most recently from 83774f7 to b2ea3b9 Compare October 4, 2026 19:09

openwrt-ai reviewed Oct 4, 2026

View reviewed changes

ServPol force-pushed the gl-e750-mcu branch from b2ea3b9 to 34a90e6 Compare October 5, 2026 07:43

BKPepe requested changes Oct 5, 2026

View reviewed changes

openwrt-ai reviewed Oct 5, 2026

View reviewed changes

ServPol marked this pull request as draft October 5, 2026 08:47

ServPol mentioned this pull request Oct 5, 2026

ath79: gl-e750: add status MCU driver (OLED, battery) openwrt/openwrt#25649

Open

ServPol force-pushed the gl-e750-mcu branch from 34a90e6 to 3cefa9b Compare October 5, 2026 20:40

ServPol changed the title ~~gl-e750-mcu: add new package~~ gl-e750-oled: add new package Oct 5, 2026

ServPol marked this pull request as ready for review October 5, 2026 20:43

	procd_set_param command /usr/sbin/gl-e750-mcu
	procd_set_param command /usr/sbin/gl-e750-mcu
	procd_set_param file /etc/config/gl-e750-mcu

	-icrnl -opost -onlcr -isig -icanon -iexten min 0 time 1
	-icrnl -opost -onlcr -isig -icanon -iexten min 1 time 0

Conversation

ServPol commented Oct 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

📦 Package Details

🧪 Run Testing Details

✅ Formalities

Uh oh!

openwrt-ai Oct 4, 2026

Choose a reason for hiding this comment

Uh oh!

openwrt-ai Oct 4, 2026

Choose a reason for hiding this comment

Uh oh!

ServPol Oct 5, 2026

Choose a reason for hiding this comment

Uh oh!

openwrt-ai Oct 4, 2026

Choose a reason for hiding this comment

Uh oh!

openwrt-ai Oct 4, 2026

Choose a reason for hiding this comment

Uh oh!

openwrt-ai Oct 4, 2026

Choose a reason for hiding this comment

Uh oh!

openwrt-ai Oct 4, 2026

Choose a reason for hiding this comment

Uh oh!

ServPol commented Oct 4, 2026

Uh oh!

openwrt-ai left a comment

Choose a reason for hiding this comment

Uh oh!

ServPol commented Oct 5, 2026

Uh oh!

BKPepe left a comment

Choose a reason for hiding this comment

Uh oh!

openwrt-ai left a comment

Choose a reason for hiding this comment

Uh oh!

ServPol commented Oct 5, 2026

Uh oh!

ServPol commented Oct 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ServPol commented Oct 4, 2026 •

edited

Loading