On Thu, Oct 20, 2016 at 3:17 AM, Nicolas Dichtel <nicolas.dich...@6wind.com> wrote: > > Now, we see that "it's needed" and that the analysis was wrong. If a race is > introduced by this patch, it will be hard to detect and fix it.
It is _not_ needed for protection, it is needed to shut up a warning, I thought this is pretty clear in the changelog. If we could invent some way to save BH (*lock_bhsave()), we don't need to bother this at all, currently we don't.