On Tue, Jan 17, 2017 at 08:14:56AM +0100, Peter Zijlstra wrote:
> On Tue, Jan 17, 2017 at 11:05:42AM +0900, Byungchul Park wrote:
> > On Mon, Jan 16, 2017 at 04:10:01PM +0100, Peter Zijlstra wrote:
> > > On Fri, Dec 09, 2016 at 02:12:03PM +0900, Byungchul Park wrote:
>
> > > > @@ -155,6 +164,9 @@ struct lockdep_map {
> > > > int cpu;
> > > > unsigned long ip;
> > > > #endif
> > > > +#ifdef CONFIG_LOCKDEP_CROSSRELEASE
> > > > + struct cross_lock *xlock;
> > > > +#endif
> > >
> > > The use of this escapes me; why does the lockdep_map need a pointer to
> > > this?
> >
> > Lockdep interfaces e.g. lock_acquire(), lock_release() and lock_commit()
> > use lockdep_map as an arg, but crossrelease need to extract cross_lock
> > instances from that.
>
> > > Why not do something like:
> > >
> > > struct lockdep_map_cross {
> > > struct lockdep_map map;
> > > struct held_lock hlock;
> > > }
>
> Using a structure like that, you can pass lockdep_map_cross around just
> fine, since the lockdep_map is the first member, so the pointers are
> interchangeable. At worst we might need to munge a few typecasts.
>
> But then the cross release code can simply cast to the bigger type and
> have access to the extra data it knows to be there.
Right. I will apply it.
