Hi, I'm looking into implementing tickless on nRF52 (first using systick, since it is an easy option, and then using a RTC timer, which works while CPU is asleep). My intention is to use the arch timer driver for this (drivers/timer/arch_timer.c) since it provides up_mdelay without using a busy loop and it supports both tickless and non-tickless. However, the code right now turns up to be a bit akward since the arch timer is optional so you have to support the case for when it is off. This means that tickless mode needs to be separately supported. On STM32 you can see this in stm32_timerisr.c (where there are two implementations, depending on the arch timer being ON or not) and in stm32_tickless.c (which does not uses the arch timer, but provides tickless functionality on its own).
I understand this is due to arch_timer/rtc/alarm being introduced more recently but I'm not sure if it makes sense to have them as optional instead of the standard way to provide both tickless and non-tickless modes. IMHO it would be cleaner and simpler (to implement and to configure the build for each case) to do the latter. Or is there a downside to this? If you agree, I think the change would consist of: * deprecate the "old way" (maybe open an issue which lists which arch is not yet converted), only accept the "new way" in new ports * convert each arch (by someone who is able to test it) by: - replace *_timerisr.c and *_tickless.c for each arch by one file (*_systime.c?) where the appropriate arch_timer/rtc/alarm interface is invoked, based on supported options for the arch (systick should be the basic option). - implement each possible system timer source as a arch_timer/rtc/alarm lowerhalf in its own file Best, Matias
