On 17 July 2013 17:01, Rafael J. Wysocki [off-list ref] wrote:

First off, I'm not sure how many applications actually use it and I think,
if any, they should be able cope with the attribute not being present.

Of course, if it turns out that yes, there are applications using it and no,
they cannot cope with the missing attribute, we'll need to address this.
That said such applications wouldn't work with earlier kernels in which that
attribute wasn't present at all, so I suppose this is really unlikely.

So, do whichever makes more sense to you: Design things to preserve the old
behavior (which is sightly confusing) or design them to expose the attribute
if the feature is actually supported and be prepared to address the (unlikely)
case when some hypothetical applications break because of that.

Okay. Its better to keep it the way Lukasz designed it in his last patchset.