Setting CONFIG_SYS_CLOCK_TICKS_PER_SEC to 1 million in a tickless build

[pavel-kirienko]

Is it a sensible idea to set CONFIG_SYS_CLOCK_TICKS_PER_SEC = 1000000 in a tickless kernel in order to increase the time resolution, or are there side effects one needs to be aware of? I am obviously aware of the more frequent timer overflows; this is not an issue in my case. Same goes for the increased power consumption by the timer.

Thanks!

[ndrs-pst]

@pavel-kirienko I think the downside would be the maximum time the kernel can wait (MAX_TICKS), which depends on the timer being used as the system tick. Digging into kernel/timeout.c might give you more insights. 😃

---

zephyr/drivers/timer/cortex_m_systick.c

Lines 15 to 21 in 012520d

	#define COUNTER_MAX 0x00ffffff
	#define TIMER_STOPPED 0xff000000
	#define CYC_PER_TICK (sys_clock_hw_cycles_per_sec() \
	/ CONFIG_SYS_CLOCK_TICKS_PER_SEC)
	#define MAX_TICKS ((k_ticks_t)(COUNTER_MAX / CYC_PER_TICK) - 1)
	#define MAX_CYCLES (MAX_TICKS * CYC_PER_TICK)

zephyr/drivers/timer/esp32_sys_timer.c

Lines 22 to 26 in 012520d

	#define CYC_PER_TICK ((uint32_t)((uint64_t)sys_clock_hw_cycles_per_sec() \
	/ (uint64_t)CONFIG_SYS_CLOCK_TICKS_PER_SEC))
	#define MAX_CYC 0xffffffffu
	#define MAX_TICKS ((MAX_CYC - CYC_PER_TICK) / CYC_PER_TICK)
	#define MIN_DELAY 1