rpma: gpspm: introduce the busy_wait_polling toggle

The performance of the librpma_gpspm engine depends heavily
on how much CPU power it can use to its work.
One can want either to take all available CPU power
and see what the maximum possible performance is
or configure it less aggressively and collect the results
when the CPU is not solely dedicated to doing this one task.

The librpma_gpspm engine allows toggling between one and another
by either waiting for incoming requests in the kernel
using rpma_conn_completion_wait() (busy_wait_polling=0)
or trying to collect the completion as soon as it appears
by polling all the time using rpma_conn_completion_get()
(busy_wait_polling=1).

Signed-off-by: Oksana Salyk <oksana.salyk@intel.com>
6 files changed