[Stopkaa]

Contribution description

This PR introduces timed tranmission for the radio HAL. Timed transmission is useful in various MAC-Layers and Protocols like TSCH(openwsn), DSME(opendsme) or Thread(openthread). Right now this functionality is unused but feel free to pick this up.
Could be also used in #22539. Also this can be used for periodically send 15.4 frames like in the hal-test

Testing procedure

Run the HAL test for the support devices tests/net/ieee802154_hal and execute the txtsnd_after command:
NRF52840DK

txtsnd_after FF:FF 10 5000
2026-10-29 11:02:44,759 # txtsnd_after FF:FF 10 5000
2026-10-29 11:02:44,762 # Sending at radio time 383640229
2026-10-29 11:02:49,762 # Transmission succeeded

ESP32C6-devkit

 txtsnd_after FF:FF 10 5000   
2026-10-29 11:07:38,638 # txtsnd_after FF:FF 10 5000  
2026-10-29 11:07:38,640 # Sending at radio time 109272382  
2026-10-29 11:07:43,638 # Transmission succeeded  

FRDM-KW41Z

  
txtsnd_after FF:FF 10 5000
2026-10-29 11:12:27,596 # txtsnd_after FF:FF 10 5000
2026-10-29 11:12:27,600 # Sending at radio time 55356224
2026-10-29 11:12:32,599 # Transmission succeeded

Issues/PRs references

• none

Declaration of AI-Tools / LLMs usage:

AI-Tools / LLMs that were used are:

• Claude Opus 5.5 for documentation and research on the timer capabilitys of different radios

[riot-ci]

Murdock results

❌ FAILED

710d72e tests/net/ieee802154_hal: add timed tx test

Success	Failures	Total	Runtime
943	1	13041	02m:03s

Build failures (1)

Application	Target	Toolchain	Runtime (s)	Worker
tests/pkg/openwsn	nrf52840dk	gnu	2.80	mobi6

Artifacts

[gschorcht]

Suggested change

	* around at 2^32. It is only valid while the device is on. Time values
	* around at 2^32. It is only valid while the device is on.

[gschorcht]

Suggested change

	* @pre The device is on
	* @pre The device supports @ref IEEE802154_CAP_TIMED_TX
	* @pre The device is on.
	* @pre The device supports @ref IEEE802154_CAP_TIMED_TX.

[gschorcht]

I know, the period at the end of this short sentence is also missing in other functions, but this doesn't make it right 😉

[gschorcht]

Suggested change

	#include "esp_timer.h"
	#include "iolist.h"
	#include "log.h"
	#include "net/ieee802154/radio.h"
	#include "iolist.h"
	#include "log.h"
	#include "net/ieee802154/radio.h"

[gschorcht]

Suggested change

	return (uint32_t)esp_timer_get_time();
	return ztimer::get_now(ZTIMER_MSEC);

From the very beginning, the goal was to keep the RIOT code as independent as possible from ESP-IDF in order to limit the amount of source code that needed to be compiled in from ESP-IDF. The dependencies between the ESP-IDF modules are very complex and more and more ESP-IDF source code must be compiled in very quickly, which makes the manual maintenance of the Makefiles in cpu/esp32/esp-idf very hard. For this reason, various basic functions have been implemented in RIOT, including a system time in milliseconds that can be retrieved using system_get_time.

Since we already use ztimer_msec modul, the RIOT function should be used.

[gschorcht]

I've given it some more thought. If the intention was to use a timer that is independent of RIOT's ztimer, you should use system_get_time to be independent of ESP-IDF, even though the ESP-IDF IEEE 802.15.4 driver uses esp_timer_get_time internally.

Suggested change

	return (uint32_t)esp_timer_get_time();
	return (uint32_t)system_get_time();

[Stopkaa]

That's some useful insight I didn't have. The idea behind it was if the radio has a dedicated only for the radio it should be used instead of ztimer if this not the case for example at86rf233 don't have such a timer the functionality would be offloaded to the submac instead. Going through the esp code I notice that the transmit_at function uses esp_timer_get_time() and thought it would use the dedicated timer then. But didn't seems like it would

[gschorcht]

Okay, I've thought about it some more 🤔 As you suggested 👍, we should use esp_timer_get_time instead of system_get_time here. So please ignore my change request.

The reason for this is that system_get_time uses a different timer than esp_timer_get_time. While system_get_time uses a real timer (TIMER_0 from TIMER_GROUP_0) for the system time, esp_timer_get_time uses a special system timer peripheral. On the ESP32, an additional counter (the LACT) from TIMER_GROUP_0 is used; all other ESP32x SoC variants have a dedicated SYSTIMER peripheral that is used for the system time.

Since the implementation of esp_ieee80215_transmit_at uses the ESP-IDF system time as its time base, we should also use this for our interface. The time returned by esp_time_get_time is likely to differ slightly from the time returned by system_get_time.

Since the ESP-IDF esp_timer is compiled in anyway in cpu/esp32/esp-idf/common, using the esp_timer_get_time function here doesn't increase the dependency on ESP-IDF.

Sorry for the confusion.

[gschorcht]

This discussion inspires me to clean up the implementation of the system time in RIOT for the ESP32. Since we’re including the ESP-IDF module esp_timer anyway, we could implement the function system_get_time based on esp_timer_get_time.

The reason system_get_time hasn’t relied on esp_time_get_time until now is that, in earlier versions of ESP-IDF, it wasn’t necessary to include the esp_timer module. This requirement was added over time, but system_get_time wasn’t updated accordingly.