Re: [PATCH 1/3] tools/liblockdep: add userspace version of READ_ONCE

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Feb 11, 2016 at 12:33:30AM +0100, Alfredo Alvarez Fernandez wrote:
> From: Alfredo Alvarez Fernandez <alfredoalvarezfernandez@gmail.com>
> 
> This was added to the kernel code in 1658d35ead5d ("list: Use
> READ_ONCE() when testing for empty lists")
> There's nothing special we need to do about it in userspace.
> 
> Signed-off-by: Alfredo Alvarez Fernandez <alfredoalvarezfernandez@gmail.com>
> ---
>  tools/lib/lockdep/uinclude/linux/compiler.h | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/tools/lib/lockdep/uinclude/linux/compiler.h b/tools/lib/lockdep/uinclude/linux/compiler.h
> index 6386dc3..fd3e56a 100644
> --- a/tools/lib/lockdep/uinclude/linux/compiler.h
> +++ b/tools/lib/lockdep/uinclude/linux/compiler.h
> @@ -3,6 +3,7 @@
>  
>  #define __used		__attribute__((__unused__))
>  #define unlikely
> +#define READ_ONCE(x) (x)
>  #define WRITE_ONCE(x, val) x=(val)

I would argue we'd still very much want the volatile cast for both READ
and WRITE_ONCE().

Why do these things have different semantics between user and kernel
space?