On Mon, Jul 18, 2022 at 10:07 PM Christoph Hellwig <hch@xxxxxx> wrote:
>
> On Mon, Jul 18, 2022 at 11:20:39AM -0700, Song Liu wrote:
> > > > static inline struct mddev *mddev_get(struct mddev *mddev)
> > > > {
> > > > + lockdep_assert_held(&all_mddevs_lock);
> > > > +
> > > > + if (mddev->deleted)
> > > > + return NULL;
> > > > atomic_inc(&mddev->active);
> > > > return mddev;
> > > > }
> > >
> > > Why can't we use 'atomic_inc_unless_zero' here and do away with the
> > > additional 'deleted' flag?
>
> > I think atomic_inc_unless_zero() should work here. Alternatively, we can
> > also grab a big from mddev->flags as MD_DELETED.
>
> s/big/bit/ ?
yeah, bit...
>
> Yes, this could use mddev->flags. I don't think atomic_inc_unless_zero
> would work, as an array can have a refcount of 0 but we still allow
> to take new references to it, the deletion only happens if the other
> conditions in mddev_put are met.
>
> Do you want me to repin the series using a flag?
Respin the series sounds good. Please also address Hannes' feedback
on 8/10.
Thanks,
Song
