sched_clock()
I just wanted to measure small intervals in kernel, which reminded me the way PRINTK_TIME does and it's enough accurate and easy for my purpose though I have to keep in mind that its accuracy(granularity) depends on architectures or platforms indeed because it comes from what kind of timer they have. And also the overflow has to be dealt with by myself. Basically its architecture default sched_clock is defined as weak symbol, mostly jiffies accuracy, and the platform version can overwrite it by just defining this function.
I took a look at omap and it uses 32k timer, but it seems that gp timer can have much smaller granularity, though...right?