Re: [PATCH/RFC 7/7] kernel: Force ACCESS_ONCE to work only on scalar types

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Nov 24, 2014 at 12:29:07PM -0800, Alexei Starovoitov wrote:
> On Mon, Nov 24, 2014 at 11:07 AM, Christian Borntraeger
> <borntraeger@de.ibm.com> wrote:
> >
> > Anyone with a new propopal? ;-)                                        ^
> 
> one more proposal :)
> #define __ACCESS_ONCE(x) ({ typeof(x) __var = 0; (volatile typeof(x) *)&(x); })
> #define ACCESS_ONCE(x) (*__ACCESS_ONCE(x))
> 
> works as lvalue...
> the basic idea is the same:
> constant zero can be used to initialize any scalar
> (including pointers), but unions and structs will fail to compile as:
> "error: invalid initializer"
> 
> If I'm reading pr58145 gcc bug report correctly, it
> miscompiles only structs, so we can let ACCESS_ONCE
> to work on unions. Then the following will rejects structs only:
> #define __ACCESS_ONCE(x) ({ (typeof(x))0; (volatile typeof(x) *)&(x); })
> #define ACCESS_ONCE(x) (*__ACCESS_ONCE(x))

You beat me to it.  ;-)

However, your approach would allow an ACCESS_ONCE() of a long long
to compile on 32-bit systems where it has to be broken up into a
pair of 32-bit accesses.

							Thanx, Paul