Re: [Q] READ_ONCE(x)++

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

On Thu, Dec 29, 2016 at 11:54:02AM +0900, Akira Yokosawa wrote:
> Hi Paul,
> 
> As I'm not so familiar with Linux kernel programming,
> I found the following hunk in commit 7945ae1a06c5 difficult to grasp at first glance.
> 
> @@ -564,7 +584,7 @@ the other will wait until the first thread releases the lock.
>   38     exit(-1);
>   39   }
>   40   for (i = 0; i < 3; i++) {
> - 41     ACCESS_ONCE(x)++;
> + 41     READ_ONCE(x)++;
>   42     poll(NULL, 0, 5);
>   43   }
>   44   if (pthread_mutex_unlock(pmlp) != 0) {
> 
> "The name READ_ONCE() seems to imply it is used for read access, but what
> happens when it is used with a "++" operator?" was what I thought.
> 
> "ACCESS_ONCE(x)++" was already somewhat confusing for me.
> 
> Once you know its definition, you can see there is no problem. But it still
> looks strange...
> 
> Don't kernel programmers feel strangeness in "READ_ONCE(x)++"?
> 
> This is just a random question. But if you could add some explanation of the
> usage, that would be of help for novice programmers.

Heh!  Good catch, thank you!

This would give a compiler error in the Linux kernel (I think, anyway),
but worked by accident given the definitions perfbook uses.

I have fixed this with your Reported-by.

							Thanx, Paul