usleep accuracy

I am using a simple usleep(100); and I noticed that the actual delay is 140 us. My specific configuration is a tickless scheduler where USEC_PER_TICK is 1.

For comparison I have setup a thread that waits on sigwaitinfo(&mask, &info);, which is signaled by a hardware timer with a timeout of 100 us. Using clock_gettime() for timing I see that this approach almost gets precisely done what I am expecting.

From what I am understanding usleep employs signals as well under the hood, however using clock_systime_ticks() for analysis it seems that it performs significantly worse. Is this expected due to the overhead?