Would like to point out that it is true that it would be nice to have the spin times in adaptive spinlocks dynamically or worse case, statically configurable. But the version of the infrastructure in mainline-rt currently does not have any restrictions on the wait-time. This is the area being worked on. So, at the moment, for every spinlock, the process spins till it obtains the lock. But the reason why this does not affect the latenices is that the sleep is completely preemptable. So at the cost of lower priority tasks not being able to run while the higher prio task spins, the next highest prio task will preempt the spinning task.