Mastering C++ Multithreading
上QQ阅读APP看书,第一时间看更新

Hardware

The simplest hardware-based implementation on a uniprocessor (single processor core), non-SMT system is to disable interrupts, and thus, prevent the task from being changed. More commonly, a so-called busy-wait principle is employed. This is the basic principle behind a mutex--due to how the processor fetches data, only one task can obtain and read/write an atomic value in the shared memory, meaning, a variable sized the same (or smaller) as the CPU's registers. This is further detailed in Chapter 8, Atomic Operations - Working with the Hardware.

When our code tries to lock a mutex, what this does is read the value of such an atomic section of memory, and try to set it to its locked value. Since this is a single operation, only one task can change the value at any given time. Other tasks will have to wait until they can gain access in this busy-wait cycle, as shown in this diagram: