On Wed, Jan 30, 2019 at 11:30:41AM -0800, Alexei Starovoitov wrote:
> On Wed, Jan 30, 2019 at 11:15:30AM +0100, Peter Zijlstra wrote:
> > On Tue, Jan 29, 2019 at 08:04:56PM -0800, Alexei Starovoitov wrote:
> > > Lockdep warns about false positive:
> >
> > This is not a false positive, and you probably also need to use
> > down_read_non_owner() to match this up_read_non_owner().
> >
> > {up,down}_read() and {up,down}_read_non_owner() are not only different
> > in the lockdep annotation; there is also optimistic spin stuff that
> > relies on 'owner' tracking.
>
> Can you point out in the code the spin bit?
Hurmph, looks like you're right. I got lost in that stuff again. I hate
that rwsem code :/
Rewriting that all is on the todo list somewhere, but it's far down :/
> As far as I can see sem->owner is debug only feature.
> All owner checks are done under CONFIG_DEBUG_RWSEMS.
>
> Also there is no down_read_trylock_non_owner() at the moment.
> We can argue about it for -next, but I'd rather silence lockdep
> with this patch today.
I don't agree with calling is silencing; it fixes an mis-use of the API.
But yes, keep the patch as is for now.
