• At a minimum, modern multiprocessors (i.e., concurrent architectures with a shared memory) provide machine instructions that the systems programmer can use to implement locks and barriers (e.g., test and set, fetch and add).
• Some multiprocessors provide hardware support for processes, context switching, locks, and the spinning aspect of semaphores.
• However, in general concurrent programming mechanisms are implemented in software as a part of the OS and/or concurrent programing environment (e.g., Pthreads library, Java JVM, OS kernel of Unix, Windows NT)
• The basis for the implementation is software kernel(or nucleus) of OS, shared by all processes. Such a kernel is a small set of data structures and subroutines that provide a virtual processor to each process so that the process has the illusion that it is executing on its own processor, where:
• The data structures represent the states of processes, semaphores, and condition variables.
• The subroutines implement primitive (i.e., executed atomically)  operations on the data structures.
• OS kernel can support processes on a single processor or on multiprocessors,  starting and killing processes, providing priority scheduling, dynamic memory and processor allocation, implementing semaphores and monitors.