RDMA/rxe: Fix an error handling path in rxe_get_mcg()

The commit in the Fixes tag has shuffled some code.
Now 'mcg_num' is incremented before the kzalloc(). So if the memory
allocation fails, this increment must be undone.

Fixes: a926a90 ("RDMA/rxe: Do not call dev_mc_add/del() under a spinlock")
Link: https://lore.kernel.org/r/fe137cd8b1f17593243aa73d59c18ea71ab9ee36.1653225896.git.christophe.jaillet@wanadoo.fr
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

drivers/infiniband/sw/rxe/rxe_mcast.c

@@ -206,8 +206,10 @@ static struct rxe_mcg *rxe_get_mcg(struct rxe_dev *rxe, union ib_gid *mgid)
 
 	/* speculative alloc of new mcg */
 	mcg = kzalloc(sizeof(*mcg), GFP_KERNEL);
-	if (!mcg)
-		return ERR_PTR(-ENOMEM);
+	if (!mcg) {
+		err = -ENOMEM;
+		goto err_dec;
+	}
 
 	spin_lock_bh(&rxe->mcg_lock);
 	/* re-check to see if someone else just added it */