On Thu, Sep 21, 2023 at 11:27:16AM +0300, Leon Romanovsky wrote:
> On Wed, Sep 20, 2023 at 11:01:12AM -0300, Jason Gunthorpe wrote:
> > On Wed, Sep 20, 2023 at 12:56:18PM +0300, Leon Romanovsky wrote:
> > > @@ -1796,6 +1804,10 @@ static int cache_ent_find_and_store(struct mlx5_ib_dev *dev,
> > > }
> > >
> > > mutex_lock(&cache->rb_lock);
> > > + if (cache->disable) {
> > > + mutex_unlock(&cache->rb_lock);
> > > + return 0;
> > > + }
> >
> >
> > I don't get this.
> >
> > Shouldn't we just initialize the ent->disabled to cache->disabled if
> > we happen to be creating a new ent?
> I'm convinced that new entries can't be created when we are calling to
> mlx5_mkey_cache_cleanup().
That's a better answer
Jason
