JMM Memory model concept

We know CPU Direct operation of cache registers , Instead of directly manipulating the main memory , If there is cache , It will cause data consistency problems , So in order to solve these problems , Put forward JMM standard .

JMM Its full name is Java Memory Model,Java Virtual machine defines a logical specification or abstract concept , Used to shield access differences between various hardware and operating system memory , Let's realize Java The program can achieve consistent memory access effect under various platforms . In other words, this set of specifications defines that the reading and writing of shared variables in the program can be mutually visible , Especially in the case of multithreading read and write operations , All around the atomicity of multithreading 、 visibility 、 Orderly unfolded .

Yes JMM You should also know something , as follows ：

• no need Java Memory area division of virtual machine , Is the existence of two kinds of logic , It is the result of the division of different levels , There are different emphases .

• The main memory can correspond to Java Heap memory in , Working memory can correspond to part of the stack .

• At a lower level , The main memory corresponds to the physical memory of the hardware , Working memory corresponds to registers and cache registers .

JMM Schematic diagram of operation memory

From the figure, we can find that the thread operates on shared variables , You need to copy a copy of the shared variables in the main memory to your own working memory , Then you can operate , After the operation, it is written back to the main memory . If considering the high concurrency scenario , Each thread first modifies the value in its own working memory , Suppose a thread assigns the value back first , At this time, the value of the main memory has been modified , But another thread is still operating with the data in its working memory , Then I wrote it back , Obviously, there is a problem with the data written back by the second thread , So how can we guarantee , Let a thread modify the value in main memory , Then notify other threads , This involves the visibility of threads , It's also JMM One of the three characteristics , Before we understand thread visibility, let's take a look JMM It's for me 8 Atomic operation to assist the interaction between threads and memory ?

JMM 8 Atomic operation ( Memory to memory interaction )

These eight atomic operation classes are also the basic steps to realize thread visibility , as follows ：

• lock： lock , Identify variables as thread exclusive , Operates on the main memory variable
• unlock： Unlock , Release locked variables , Only other threads can use , For main memory
• read： Read , Read the variable value from the main memory to the working memory ( Just read the working memory , Not assigned to a variable )
• load： load , hold read The read variables are put into the working memory Copy variable
• use： Use , Pass the value of a variable in the working memory to the execution engine
• assign： assignment , Assign the value received from the execution engine to the copy variable in the working memory
• store： Storage , Transfer variables in working memory to main memory ( Just stored in main memory , Not assigned to a variable )
• write： write in , hold store The stored value is written into the main memory Shared variables

The following figure shows that two threads simultaneously control shared variables in main memory flag Modification process ：

First, the thread should read the value of the shared variable from the main memory , First, lock the shared variable (lock operation ), And then through read operation , hold flag The value of is read from main memory into working memory ( It's actually a cache register ), And then through load operation , Put the working memory flag The value of is loaded into the variable copy , And then through use operation , Pass the value of the variable to CPU Execution engine .

CPU After execution , adopt assign operation , Store the calculation results in the variable copy , Then passed store operation , Copy the variable's flag Values are passed from working memory to main memory , And then execute write operation , hold flag The value rewritten of will be written to the shared variable in the main memory flag, Finally through unlock The operation thread releases the lock , Other threads can lock this variable , Repeat the above process .

Then there are several rules of memory interaction that need attention :

• Don't allow read and load、store and write The operation appears separately , The above two operations must be performed in sequence , But continuity is not guaranteed , in other words read and load、store and write Other instructions can be inserted between
• A thread is not allowed to discard its latest assign operation , That is, after the variable is changed in the working memory, the change must be synchronized to the main memory
• A thread without reason is not allowed ( There is nothing assign Modify the assignment operation ) Synchronizing data from the working memory will affect the main memory
• A new variable can only be born from main memory , It is not allowed to use an uninitialized variable directly in working memory , That is to say, to implement... On a variable user and store Before the operation , Must be executed first load and assign operation
• Only one thread is allowed to control a variable at the same time lock operation , but lock The operation can be repeated many times by the same thread , many times lock after , Only execute the same number of times unlock operation , Variables will unlock
• If you execute on a variable lock operation , Will empty the value of this variable in working memory , Before the execution engine uses this variable , Need to re load
• If a variable is not lock lock , It is not allowed to be executed unlock operation , Also can not unlock A variable that is locked by another thread
• Execute on a variable unlock Before , This variable must be synchronized to the main memory first ( perform store and write operation )