学步园

atomic_inc(&v)对变量v用锁定总线的单指令进行不可分解的"原子"级增量操作，避免v的值由于中断或多处理器同时操作造成不确定状态。

X86 用LOCK 指令实现，  不懂

ARM代码如下，  ldrex :

 ARMv6_Architecture.pdf    解释如下   还是不大明白

• LDREX{<cond>} <Rd>, [<Rn>]
This performs a load, then sets a monitor to “watch” the address
• STREX {<cond>} <Rd>, <Rm>, [<Rn>]
This performs a store and returns “success” in Rd if no intervening access
detected by the monitor.

• LDREX{<cond>} <Rd>, [<Rn>]
This performs a load, then sets a monitor to “watch” the address
• STREX {<cond>} <Rd>, <Rm>, [<Rn>]
This performs a store and returns “success” in Rd if no intervening access
detected by the monitor.

• LDREX{<cond>} <Rd>, [<Rn>]
This performs a load, then sets a monitor to “watch” the address
• STREX {<cond>} <Rd>, <Rm>, [<Rn>]
This performs a store and returns “success” in Rd if no intervening access
detected by the monitor.

static inline int atomic_add_return(int i, atomic_t *v)
{
        unsigned long tmp;
        int result;

        __asm__ __volatile__("@ atomic_add_return/n"
"1:     ldrex   %0, [%2]/n"
"       add     %0, %0, %3/n"
"       strex   %1, %0, [%2]/n"
"       teq     %1, #0/n"
"       bne     1b"
        : "=&r" (result), "=&r" (tmp)
        : "r" (&v->counter), "Ir" (i)
        : "cc");

        return result;