mjhjmtu

Let's say I have a CPU with lots of cores. Is a given process, once started, bound to one of the cores?

Does that mean that a process can become sluggish even if there are several unused cores? For example, core0 is being used by process A and B, process A is using 100% of the core, and process B can not run, even though core1 is completely free.

Is this correct?

Processes are not intrinsically bound to cores. Each time a process is scheduled for execution it may be executed by any core in its affinity list. If the affinity list has not been set explicitly, a process can be run on any core.

You can check or set the affinity set for a process with the taskset command (see man taskset for its documentation). In particular, it writes that,

The Linux scheduler will honour the given CPU affinity and the process will not run on any other CPUs. Note that the Linux scheduler also supports natural CPU affinity: the scheduler attempts to keep processes on the same CPU as long as practical for performance reasons. Therefore, forcing a specific CPU affinity is useful only in certain applications.

Example

sleep 1000 & slpid=$!
echo "PID of sleep is $slpid" # 5221

taskset --pid $slpid
pid 5221's current affinity mask: 3 # bitmask

taskset --cpu-list --pid $slpid
pid 5221's current affinity list: 0,1 # list of potential cores

taskset --cpu-list --pid 0 $slpid
pid 5221's current affinity list: 0,1
pid 5221's new affinity list: 0 # bound now only to core 0

kill $slpid # all done, tidy up