Taking timestamps within kernel code

It's important to be able to take an accurate timestamp as kernels open employ this facility. For example, the dmesg(1) utility shows the time since the system booted in seconds.microseconds format; Ftrace traces typically show the time a function takes to execute. When in user mode, we often employ the gettimeofday(2) system call to take a timestamp. Within the kernel, several interfaces exist; commonly, the ktime_get_*() family of routines is employed for the purpose of obtaining accurate timestamps. For our purposes, the following routine is useful:

u64 ktime_get_real_ns(void);

This routine internally queries the wall (clock) time via the ktime_get_real() API and then converts the result into a nanosecond quantity. We won't bother with the internal details here. Also, several variants of this API are available; for example, ktime_get_real_fast_ns(), ktime_get_real_ts64(), and so on. The former is both fast and NMI-safe.

...