Currently, if you use the -t option without an argument cyclictest presumes that it should put a measurement thread on every online cpu. This is the original logic, which is to test the entire system.
The alternative is to use -t<n> and -a<cpulist> to create <n> measurement threads and affine them to <cpulist>. One complaint I have about this is that no matter where you place the measurement threads, they're labeled 0-(n-1) in the output, so it makes it hard to map a test thread back to the cpu where it ran. Wondering if we can update the output to show the core to which a test thread is bound.

I'm also wondering if we can add a third option to say "just use the cores in your current cpu affinity mask". That way you can rejigger the tuning parameters for a system and then just kick off cyclictest without having to figure out how your -a/-t options should change to match.

I'm thinking where we've diddled the isolcpus= boot parameter and want to try cyclictest over there like this:
        # presumes isolcpus=0,12
        taskset --cpus=0,12 cyclictest -p95 -mn --affinity-mask

So that we'd get two measurment threads, affine to cpus 0 and 12. I realize that we could do this with '-t2 -a0,12' but I was hoping to simplify things.