On Fri, 2019-06-14 at 08:14 -0700, Ben Greear wrote: > > > So, maybe I return instead the elapsed time in the netlink API instead of a > timestamp. I think that will give me the value that I am looking for, > and I can still print out the 'real' time in iw so any tools reading that > output and do some simple math and figure out the 'real' associated-at time.
I don't think that's good. There's a delay between filling the message and then processing it in userspace. We had this in the scan code and learned that was full of races. The better thing is to use CLOCK_BOOTTIME nanoseconds, and then userspace can use clock_gettime() to get the current time etc. and then do whatever calculations it needs. If it wants to print the realtime timestamp, it could even calculate that (with some jitter) by doing clock_gettime() on both realtime and boottime clocks... johannes
