On Mon, Aug 19, 2013 at 05:31:36AM -0400, Ed Maste wrote:
> Commit 1514b275 introduces the use of atomic_read_explicit for
> once-only initializers. When using C11 atomics this becomes
> atomic_load_explicit, and the first argument needs to be non-const --
> with FreeBSD HEAD's in-tree clang 3.3 the build fails for me with:
>
> In file included from lib/bfd.c:34:
> ./lib/ovs-thread.h:482:5: error: first argument to atomic operation must be a
> pointer to non-const _Atomic type ('const atomic_bool *' (aka
> 'const _Atomic(bool) *') invalid)
> atomic_read_explicit(&once->done, &done, memory_order_relaxed);
> ^ ~~~~~~~~~~~
>
> One workaround is to sprinkle CONST_CASTs around, in ovs-thread.h and
> lib/bfd. - e.g.:
>
> --- a/lib/ovs-thread.h
> +++ b/lib/ovs-thread.h
> @@ -479,7 +479,8 @@ ovsthread_once_is_done__(const struct ovsthread_once
> *once)
> {
> bool done;
>
> - atomic_read_explicit(&once->done, &done, memory_order_relaxed);
> + atomic_read_explicit(CONST_CAST(atomic_bool *, &once->done), &done,
> + memory_order_relaxed);
> return done;
> }
>
> Ben, what do you think?
Oops. Thanks for the bug report.
I sent out a series that should fix the problem and make it more
difficult for it to recur:
http://openvswitch.org/pipermail/dev/2013-August/030888.html
http://openvswitch.org/pipermail/dev/2013-August/030889.html
http://openvswitch.org/pipermail/dev/2013-August/030890.html
The last patch is really curious and makes me wonder if I really
misunderstand something:
http://openvswitch.org/pipermail/dev/2013-August/030891.html
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
