Re: [PATCH rdma-next 2/2] RDMA/core: Optimize XRC target lookup

[Jason Gunthorpe <jgg@mellanox.com>]

On Mon, Jun 22, 2020 at 03:57:29PM +0300, Maor Gottlieb wrote:
> 
> On 6/22/2020 3:29 PM, Jason Gunthorpe wrote:
> > On Sun, Jun 21, 2020 at 01:41:10PM +0300, Leon Romanovsky wrote:
> > > @@ -2318,19 +2313,18 @@ EXPORT_SYMBOL(ib_alloc_xrcd_user);
> > >   int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata)
> > >   {
> > > +	unsigned long index;
> > >   	struct ib_qp *qp;
> > >   	int ret;
> > >   	if (atomic_read(&xrcd->usecnt))
> > >   		return -EBUSY;
> > > -	while (!list_empty(&xrcd->tgt_qp_list)) {
> > > -		qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
> > > +	xa_for_each(&xrcd->tgt_qps, index, qp) {
> > >   		ret = ib_destroy_qp(qp);
> > >   		if (ret)
> > >   			return ret;
> > >   	}
> > Why doesn't this need to hold the tgt_qps_rwsem?
> > 
> > Jason
> 
> Actually, we don't need this part of code. if usecnt is zero so we don't
> have any tgt qp in the list. I guess it is leftovers of ib_release_qp which
> was already deleted.

Then have a WARN_ON that the xarray is empty

Jason