OSDN Git Service

RDMA/mlx5: Don't call to restrack recursively
authorLeon Romanovsky <leonro@mellanox.com>
Tue, 22 Sep 2020 09:11:03 +0000 (12:11 +0300)
committerJason Gunthorpe <jgg@nvidia.com>
Tue, 22 Sep 2020 22:47:35 +0000 (19:47 -0300)
The restrack is going to manage memory of all IB objects and must be
called before object is created. GSI QP in the mlx5_ib separated between
creating dummy interface and HW object beneath. This was achieved by
double call to ib_create_qp().

In order to skip such reentry call to internal driver create_qp code.

Link: https://lore.kernel.org/r/20200922091106.2152715-3-leon@kernel.org
Reviewed-by: Mark Zhang <markz@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
drivers/infiniband/hw/mlx5/gsi.c

index 40d4181..d9f300f 100644 (file)
@@ -182,13 +182,25 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
                hw_init_attr.cap.max_send_sge = 0;
                hw_init_attr.cap.max_inline_data = 0;
        }
-       gsi->rx_qp = ib_create_qp(pd, &hw_init_attr);
+
+       gsi->rx_qp = mlx5_ib_create_qp(pd, &hw_init_attr, NULL);
        if (IS_ERR(gsi->rx_qp)) {
                mlx5_ib_warn(dev, "unable to create hardware GSI QP. error %ld\n",
                             PTR_ERR(gsi->rx_qp));
                ret = PTR_ERR(gsi->rx_qp);
                goto err_destroy_cq;
        }
+       gsi->rx_qp->device = pd->device;
+       gsi->rx_qp->pd = pd;
+       gsi->rx_qp->real_qp = gsi->rx_qp;
+
+       gsi->rx_qp->qp_type = hw_init_attr.qp_type;
+       gsi->rx_qp->send_cq = hw_init_attr.send_cq;
+       gsi->rx_qp->recv_cq = hw_init_attr.recv_cq;
+       gsi->rx_qp->event_handler = hw_init_attr.event_handler;
+       spin_lock_init(&gsi->rx_qp->mr_lock);
+       INIT_LIST_HEAD(&gsi->rx_qp->rdma_mrs);
+       INIT_LIST_HEAD(&gsi->rx_qp->sig_mrs);
 
        dev->devr.ports[init_attr->port_num - 1].gsi = gsi;
 
@@ -219,7 +231,7 @@ int mlx5_ib_gsi_destroy_qp(struct ib_qp *qp)
        mlx5_ib_dbg(dev, "destroying GSI QP\n");
 
        mutex_lock(&dev->devr.mutex);
-       ret = ib_destroy_qp(gsi->rx_qp);
+       ret = mlx5_ib_destroy_qp(gsi->rx_qp, NULL);
        if (ret) {
                mlx5_ib_warn(dev, "unable to destroy hardware GSI QP. error %d\n",
                             ret);