When discussing semaphores first, we tried to avoid the spin$-$lock in the acquire $($sem wait$)$ function by blocking the calling process. But then when implementing acquire$/$sem wait and release$/$sem post, the operations on the shared variable within the these functions become the critical section and we said that they can now be protected using a spin$-$lock. Why is it OK to use the spin$-$lock when implementing semaphore operations while we were trying to avoid in the first place when discussing in the context of applications?