以下是關(guān)于volatile關(guān)鍵的外文描述：

By declaring an object volatile, the compiler is informed that the value of the object can change beyond the compiler’s control. The compiler must also assume that any accesses can have side effects—thus all accesses to the volatile object must be preserved.

There are three main reasons for declaring an object volatile:

Shared access; the object is shared between several tasks in a multitasking environment

Trigger access; as for a memory-mapped SFR where the fact that an access occurshas an effect

Modified access; where the contents of the object can change in ways not known tothe compiler.

1、Shared access

the object is shared between several tasks in a multitasking environment。

當(dāng)同一全局變量在多個(gè)線程之間被共享時(shí)，有可能會(huì)出現(xiàn)同步錯(cuò)誤，編譯器可能會(huì)將訪問(wèn)該全局變量的代碼優(yōu)化為訪問(wèn)某個(gè)寄存器，而不會(huì)再次訪問(wèn)相應(yīng)的內(nèi)存，導(dǎo)致程序運(yùn)行錯(cuò)誤。

測(cè)試代碼如下：

staticstructrt_threadv_thread1;
staticcharv_thread1_stack[8192];
staticstructrt_threadv_thread2;
staticcharv_thread2_stack[8192];

staticintflag;
staticintcount;

staticvoidrt_init_thread1_entry(void*parameter)
{
while(1)
{
rt_thread_mdelay(300);
flag=1;
rt_thread_mdelay(300);
flag=0;

if(count++>10)
{
rt_kprintf("thread1 exit.\n");
flag=1;
return;
}
}
}

staticvoidrt_init_thread2_entry(void*parameter)
{
while(1)
{
while(flag==0);
rt_kprintf("thread2 running.\n");
rt_thread_mdelay(100);

if(count++>10)
{
rt_kprintf("thread2 exit.\n");
return;
}
}
}

intvolatile_test()
{
rt_err_tresult=RT_EOK;
result=rt_thread_init(&v_thread1,"vth1",
rt_init_thread1_entry,
RT_NULL,
v_thread1_stack,sizeof(v_thread1_stack),
RT_THREAD_PRIORITY_MAX/3-1,20);
if(result==RT_EOK)
rt_thread_startup(&v_thread1);

result=rt_thread_init(&v_thread2,"vth2",
rt_init_thread2_entry,
RT_NULL,
v_thread2_stack,sizeof(v_thread2_stack),
RT_THREAD_PRIORITY_MAX/3,20);
if(result==RT_EOK)
rt_thread_startup(&v_thread2);

return0;

}
MSH_CMD_EXPORT(volatile_test,runvolatile_test);

上面的測(cè)試代碼在 O0 優(yōu)化時(shí)正常運(yùn)行，打印結(jié)果如下：

msh />volatile_test
thread2 running.
msh />thread2 running.
thread2 running.
thread2 running.
thread2 running.
thread2 running.
thread2 running.
thread2 running.
thread2 running.
thread2 exit.
thread1 exit.

但是如果開(kāi)啟 O3 優(yōu)化，則打印結(jié)果如下：

msh />volatile_test
thread1 exit.

也就是說(shuō) thread2 永遠(yuǎn)得不到運(yùn)行，那么原因是什么呢，請(qǐng)看下圖的反匯編，語(yǔ)句

while(flag==0);

被優(yōu)化成了如下匯編：

00108b4c: ldr r3, [r4, #+288] # 第一次讀取 flag 的實(shí)際值到 r3
00108b50: cmp r3,#0 # 對(duì)比 r3 的值是否為 0
00108b54: bne +0 ; # 如果不為 0 則跳轉(zhuǎn)
00108b58: b -8 ; # 再次跳轉(zhuǎn)回 cmp 語(yǔ)句繼續(xù)循環(huán)

也就是說(shuō)，整個(gè)程序被翻譯成，只讀取一次 flag 的實(shí)際值，后續(xù)一直使用 r3 寄存器中的值來(lái)進(jìn)行對(duì)比，而第一次讀取到的 r3 值為零，因此 while 的條件將永遠(yuǎn)成立，thread2 永遠(yuǎn)也得不到執(zhí)行。

2、Trigger access

as for a memory-mapped SFR（特殊功能寄存器）where the fact that an access occurs has an effect。

當(dāng)讀取類似串口設(shè)備的數(shù)據(jù)寄存器時(shí)，一定要加上 volatile，因?yàn)樵摰刂芳拇嫫髦械臄?shù)值可能會(huì)發(fā)生改變，如果不加 volatile，可能會(huì)發(fā)現(xiàn)讀取的數(shù)據(jù)是錯(cuò)誤的。

3、Modified access

where the contents of the object can change in ways not known to the compiler.

對(duì)象的內(nèi)容可能會(huì)被以編譯器不清楚的方式被修改，例如在內(nèi)核態(tài)與用戶態(tài)的程序在不同的虛擬地址訪問(wèn)同一塊物理內(nèi)存，此時(shí)如果不加上 volatile，則外部的修改無(wú)法被感知到，造成程序錯(cuò)誤。

關(guān)于優(yōu)化錯(cuò)誤

如果系統(tǒng)在低優(yōu)化等級(jí)能正常運(yùn)行，但是在高優(yōu)化的情況下的無(wú)法正常運(yùn)行，首先懷疑兩個(gè)方面：

是否是一些關(guān)鍵操作沒(méi)有添加 volatile

是否是有內(nèi)存寫穿（因?yàn)椴煌膬?yōu)化等級(jí)改變了內(nèi)存排布導(dǎo)致寫穿位置發(fā)生改變）

4、如何避免關(guān)鍵操作被優(yōu)化

情況一

如果發(fā)現(xiàn)加上了printf打印，或者調(diào)用了某個(gè)外部函數(shù)，系統(tǒng)就正常運(yùn)行了，也要懷疑是否出現(xiàn)了變量訪問(wèn)被優(yōu)化的情況，因?yàn)槿绻由狭?strong>外部函數(shù)（非本文件中的函數(shù)或其他庫(kù)中的函數(shù)）調(diào)用，則編譯器無(wú)法確定被引用的變量是否被外部函數(shù)所改變，因而會(huì)自動(dòng)從原有地址重新讀取該變量的值。

如果修改上面的測(cè)試代碼，在 while 循環(huán)中加入rt_kprintf打印如下：

while(flag==0)
{
rt_kprintf("5\n");
}

則程序仍然正常運(yùn)行，原因就是編譯器不知道rt_kprintf函數(shù)是否會(huì)修改 flag 變量，因此編譯器會(huì)嘗試每次都重新讀取flag的值。

情況二

還可以使用另外一種方式來(lái)解決這個(gè)問(wèn)題，如下：

while(flag==0)
{
asmvolatile("":::"memory");
}

If our instruction modifies memory in an unpredictable fashion, add "memory" to the list of clobbered registers. This will cause GCC to not keep memory values cached in registers across the assembler instruction. We also have to add thevolatile keywordif the memory affected is not listed in the inputs or outputs of the asm.

這將會(huì)告訴編譯器，經(jīng)過(guò)一些指令后，memory 中的數(shù)據(jù)已經(jīng)發(fā)生了變化，GCC 將不會(huì)再使用寄存器作為數(shù)據(jù)的緩存。因此再次使用這些數(shù)據(jù)時(shí)，會(huì)從內(nèi)存中重新嘗試讀取。使用關(guān)鍵字 volatile 也可以達(dá)到同樣的效果。

以下描述摘自《GCC-Inline-Assembly-HOWTO》：

Some instructions clobber some hardware registers. We have to list those registers in the clobber-list, ie the field after the third ’:’ in the asm function. This is to inform gcc that we will use and modify them ourselves. So gcc will not assume that the values it loads into these registers will be valid. We shoudn’t list the input and output registers in this list. Because, gcc knows that "asm" uses them (because they are specified explicitly as constraints). If the instructions use any other registers, implicitly or explicitly (and the registers are not present either in input or in the output constraint list), then those registers have to be specified in the clobbered list.

If our instruction can alter the condition code register, we have to add "cc" to the list of clobbered registers.

4、結(jié)論

關(guān)于 volatile 關(guān)鍵字，最重要的是要認(rèn)識(shí)到一點(diǎn)，即是否在編譯器清楚的范圍之外，所操作的變量有可能被改變，如果有這種可能性，則一定要添加上 volatile 關(guān)鍵字，以避免這種錯(cuò)誤。

歸根結(jié)底，是要確定代碼在真實(shí)運(yùn)行的狀態(tài)下，當(dāng)其訪問(wèn)某個(gè)變量時(shí)，是否真正地從這個(gè)變量所在的地址重新讀取該變量的值，而不是直接使用上次存儲(chǔ)在某個(gè)寄存器中的值。