1.Cortex-A9 has a private timer that supports one short and continous modes. We can configure it as continous mode and it will generates periodic interrupts without any more configuration.
2.Cortex-A53 replaces the private timer with generic timer. But the generic timer does not support continous mode. The interrupt output condition is the physical timer's compare value larger
than the counter. For example we must reprogramme the CNTP_TVAL or the CNTP_CVAL to generate the next interrupt. We must first read the current ounter , add the periodic time count and then write it to the CNTP_CVAL.
3.The periodic interrupt needs repeated software configuration. Won't it effect the accuracy of the interval between interrupts ? Is the linux kernel OS tick generated this way from generic
timer ?
I've not looked at how the Linux the kernel does this, but I suspect it simply increments the value in the comparator by a set amount. In which case, as long it doesn't take more than timer interval to respond to the interrupt, then it shouldn't matter exactly when you perform the update.