[puddly]

OpenThread in ESP-IDF implements UART writing via their POSIX-y write without checking the return code. This is pretty common in OpenThread implementations, I believe EFR32 does the same. Unfortunately, ESP-IDF also flags USB UARTs as suspended if a few cumulative 1ms stutters pile up, resulting in the very clean loss of a single Spinel HDLC frame during heavy traffic. Doubly unfortunately, Spinel does not have global sequence counters for notifications so this loss is entirely and completely silent.

This should affect stock OpenThread firmware on the ESP32 as well. I'll try to report this to Espressif.

[Copilot]

🟡 Changes recommended

A critical transport bug can silently discard HDLC data when the USB ring fills while still signaling successful completion, so failed enqueues must be retried or explicitly fail the transport.

1 open issue

• Do not acknowledge UART chunks discarded when the USB ring is full · New

What changed in this PR

This PR addresses silent Spinel frame loss on ESP32-C6 by bypassing the console VFS and sending directly through the USB-Serial-JTAG driver.

Changes:

• Route USB Spinel writes through the driver’s TX ring buffer.
• Increase USB buffer sizes to accommodate HDLC chunks.
• Replace inline component-editing scripts with patch files, preserving the Ziggurat NCP integration.

File	Description
firmwares/​esp32-c6/​patches/​0002-send-spinel-through-the-usb-serial-jtag-driver.patch	Routes USB writes directly to the driver with a bounded timeout.
firmwares/​esp32-c6/​patches/​0001-ziggurat-owns-the-ncp-vendor-hook.patch	Preserves vendor-hook ownership and glue compilation as a patch.
firmwares/​esp32-c6/​main/​esp_ot_config.h	Sets TX and RX buffers to 4096 and 2048 bytes.
firmwares/​esp32-c6/​build.sh	Applies component patches instead of inline Python edits.

_{Review effort: Balanced}

[MattWestb]

One PM if using ESP-IDF version pre some V6 beta have some very nasty bugs in the USB-UART that shall being fixed in the latest versions and its only USB-UART communication not normal UART.

[puddly]

I ran into this on v6.0.2 but I think this is still present in v6.1

[puddly]

I've reported it to Espressif. Hopefully they can fix it 😄

[MattWestb]

I was testing H2 with the S3 boar with eth an its using the normal UART so no problem but many in the community thread and it was fixed in one pre V6 beta but is the dictated Zigbee / thread repo not the general ide one an was not going there in the v6 release. Perhaps they have missing putting it in there.

As normal great . . . . done !!!!!