CODEWITHSUNDEEP
linuxx86kernelcpu-cache
Mike
Mike

Reputation: 1857

Why does clflush needs +m constant

I found that in Linux kernel, the clflush function is implemented as

asm volatile("clflush %0" : "+m" (*(volatile char __force *)__p));

I don't quite understand why +m is used here?

In my understanding, shouldn't it be implemented as 

asm volatile ("clflush (%0)" :: "r"(p));

Upvotes: 1

Views: 454

Answers (1)

Drew McGowen
Drew McGowen

Reputation: 11706

Either form of the instruction works since they both refer to the same address. However, by using +m as the constraint, it ensures that any optimization done with the code (since the function is inline) doesn't assume the data stored in the pointer __p is preserved. In other words, it prevents invalid optimization.

Upvotes: 3

Related Questions