Kernel synchronization mechanism

Some basic concepts about synchronization theory ：
A critical region （critical area）: Code snippet for accessing or manipulating shared data , It can be simply understood as synchronized The part in braces （ Atomicity ）

Competitive conditions （race conditions）： Two threads have the execution right of the critical section at the same time

Data inconsistency ：data unconsistency Data corruption caused by competitive conditions

Sync （synchronization）： avoid race conditions

lock ： The means to synchronize （ Door lock , Behind the door is the critical area , Only one thread is allowed to exist ） Locking and unlocking must be atomic

Atomicity ： An indivisible operation like an atom

Orderliness ： Prohibit command rearrangement

visibility ： Modifications within a thread , Another thread is visible

Common methods of kernel synchronization ：

• Atomic manipulation ： The kernel is similar to AtomicXXX, be located <linux/types.h>
• spinlocks ： Supported by assembly in the kernel cas, be located <asm/spinlock.h>
• Read write spin ： Be similar to ReadWriteLock, Can read at the same time , Only one can write , When I read it, it's a shared lock , When I write, it's exclusive
• Semaphore ： Be similar to Semaphore（PV operation down up operation , Possession and release ）  Heavyweight lock , The thread will enter wait, Suitable for long-term holding of locks
• Read and write semaphores ：downread upread downwrite upwrite （ Write more than one , You can write in sections , Less use ）( Section lock ）
• mutex （mutex）： Special semaphores （ Binary semaphore ）
• Completion variables ： Special semaphores （A Signal to B,B Waiting on the completion variable ） vfork() Wake up the parent process by completing the variable at the end of the child process Be similar to (Latch)
• BKL： Big kernel lock
• Sequential lock （2.6）： Read and write spinlocks that threads can suspend , Sequence counter （ from 0 Start , Add... As you write (+1), After writing, release (+1), I found the singular before reading , There is a write thread , wait for , The sequence before and after reading is the same , Note that there is no write thread break ）
• No preemption ：preempt_disable()
• Memory barrier ： see volatile