Re: [PATCH rdma-next v5 3/8] RDMA/restrack: Add general infrastructure to track RDMA resources

[Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 951 bytes --]

On Tue, Jan 23, 2018 at 10:58:48AM -0700, Jason Gunthorpe wrote:
> On Tue, Jan 23, 2018 at 10:54:33AM -0700, Jason Gunthorpe wrote:
> > On Mon, Jan 22, 2018 at 02:51:14PM +0200, Leon Romanovsky wrote:
> > > +	mutex_lock(&dev->res.mutex);
> > > +	hash_del_rcu(&res->node);
> > > +	mutex_unlock(&dev->res.mutex);
> > > +
> > > +	res->valid = false;
> > > +
> > > +	if (res->task)
> > > +		put_task_struct(res->task);
> > > +	synchronize_srcu(&dev->res.srcu);
> > > +}
> >
> > This locking is wrong..
>
> Actually, just don't use RCU.
>
> We can't afford to call synchronize_srcu in the destroy() path of
> every object and the netlink user is not critical enough to bother
> optimizing the read side.

In large systems with many objects, such rwlock will block easily destroy
of the objects.

>
> Replace the rcu and the mutex with a simple rwlock.

My initial proposal was SRCU per-res to protect data with rwlock per-type to protect lists.

>
> Jason

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]