RDMA: Globally allocate and release QP memory

Convert QP object to follow IB/core general allocation scheme.  That
change allows us to make sure that restrack properly kref the memory.

Link: https://lore.kernel.org/r/48e767124758aeecc433360ddd85eaa6325b34d9.1627040189.git.leonro@nvidia.com
Reviewed-by: Gal Pressman <galpress@amazon.com> #efa
Tested-by: Gal Pressman <galpress@amazon.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com> #rdma and core
Tested-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Tested-by: Tatyana Nikolova <tatyana.e.nikolova@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

4 files changed, 45 insertions, 27 deletions

diff --git a/drivers/infiniband/core/core_priv.h b/drivers/infiniband/core/core_priv.h
index 647cca4e0240..fa2e0bbaf8c7 100644
--- a/drivers/infiniband/core/core_priv.h
+++ b/drivers/infiniband/core/core_priv.h
@@ -322,13 +322,14 @@ _ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
 	      struct ib_uqp_object *uobj, const char *caller)
 {
 	struct ib_qp *qp;
+	int ret;
 
 	if (!dev->ops.create_qp)
 		return ERR_PTR(-EOPNOTSUPP);
 
-	qp = dev->ops.create_qp(pd, attr, udata);
-	if (IS_ERR(qp))
-		return qp;
+	qp = rdma_zalloc_drv_obj_numa(dev, ib_qp);
+	if (!qp)
+		return ERR_PTR(-ENOMEM);
 
 	qp->device = dev;
 	qp->pd = pd;
@@ -337,14 +338,10 @@ _ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
 
 	qp->qp_type = attr->qp_type;
 	qp->rwq_ind_tbl = attr->rwq_ind_tbl;
-	qp->send_cq = attr->send_cq;
-	qp->recv_cq = attr->recv_cq;
 	qp->srq = attr->srq;
-	qp->rwq_ind_tbl = attr->rwq_ind_tbl;
 	qp->event_handler = attr->event_handler;
 	qp->port = attr->port_num;
 
-	atomic_set(&qp->usecnt, 0);
 	spin_lock_init(&qp->mr_lock);
 	INIT_LIST_HEAD(&qp->rdma_mrs);
 	INIT_LIST_HEAD(&qp->sig_mrs);
@@ -352,8 +349,25 @@ _ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
 	rdma_restrack_new(&qp->res, RDMA_RESTRACK_QP);
 	WARN_ONCE(!udata && !caller, "Missing kernel QP owner");
 	rdma_restrack_set_name(&qp->res, udata ? NULL : caller);
+	ret = dev->ops.create_qp(qp, attr, udata);
+	if (ret)
+		goto err_create;
+
+	/*
+	 * TODO: The mlx4 internally overwrites send_cq and recv_cq.
+	 * Unfortunately, it is not an easy task to fix that driver.
+	 */
+	qp->send_cq = attr->send_cq;
+	qp->recv_cq = attr->recv_cq;
+
 	rdma_restrack_add(&qp->res);
 	return qp;
+
+err_create:
+	rdma_restrack_put(&qp->res);
+	kfree(qp);
+	return ERR_PTR(ret);
+
 }
 
 struct rdma_dev_addr;
diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index 9056f48bdca6..f4814bb7f082 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -2654,6 +2654,7 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops)
 	SET_DEVICE_OP(dev_ops, get_hw_stats);
 	SET_DEVICE_OP(dev_ops, get_link_layer);
 	SET_DEVICE_OP(dev_ops, get_netdev);
+	SET_DEVICE_OP(dev_ops, get_numa_node);
 	SET_DEVICE_OP(dev_ops, get_port_immutable);
 	SET_DEVICE_OP(dev_ops, get_vector_affinity);
 	SET_DEVICE_OP(dev_ops, get_vf_config);
@@ -2710,6 +2711,7 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops)
 	SET_OBJ_SIZE(dev_ops, ib_cq);
 	SET_OBJ_SIZE(dev_ops, ib_mw);
 	SET_OBJ_SIZE(dev_ops, ib_pd);
+	SET_OBJ_SIZE(dev_ops, ib_qp);
 	SET_OBJ_SIZE(dev_ops, ib_rwq_ind_table);
 	SET_OBJ_SIZE(dev_ops, ib_srq);
 	SET_OBJ_SIZE(dev_ops, ib_ucontext);
diff --git a/drivers/infiniband/core/restrack.c b/drivers/infiniband/core/restrack.c
index 033207882c82..1f935d9f6178 100644
--- a/drivers/infiniband/core/restrack.c
+++ b/drivers/infiniband/core/restrack.c
@@ -343,7 +343,7 @@ void rdma_restrack_del(struct rdma_restrack_entry *res)
 	rt = &dev->res[res->type];
 
 	old = xa_erase(&rt->xa, res->id);
-	if (res->type == RDMA_RESTRACK_MR || res->type == RDMA_RESTRACK_QP)
+	if (res->type == RDMA_RESTRACK_MR)
 		return;
 	WARN_ON(old != res);
 
diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index 7036967e4c0b..a164609c2ee7 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -1963,30 +1963,32 @@ int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata)
 		rdma_rw_cleanup_mrs(qp);
 
 	rdma_counter_unbind_qp(qp, true);
-	rdma_restrack_del(&qp->res);
 	ret = qp->device->ops.destroy_qp(qp, udata);
-	if (!ret) {
-		if (alt_path_sgid_attr)
-			rdma_put_gid_attr(alt_path_sgid_attr);
-		if (av_sgid_attr)
-			rdma_put_gid_attr(av_sgid_attr);
-		if (pd)
-			atomic_dec(&pd->usecnt);
-		if (scq)
-			atomic_dec(&scq->usecnt);
-		if (rcq)
-			atomic_dec(&rcq->usecnt);
-		if (srq)
-			atomic_dec(&srq->usecnt);
-		if (ind_tbl)
-			atomic_dec(&ind_tbl->usecnt);
-		if (sec)
-			ib_destroy_qp_security_end(sec);
-	} else {
+	if (ret) {
 		if (sec)
 			ib_destroy_qp_security_abort(sec);
+		return ret;
 	}
 
+	if (alt_path_sgid_attr)
+		rdma_put_gid_attr(alt_path_sgid_attr);
+	if (av_sgid_attr)
+		rdma_put_gid_attr(av_sgid_attr);
+	if (pd)
+		atomic_dec(&pd->usecnt);
+	if (scq)
+		atomic_dec(&scq->usecnt);
+	if (rcq)
+		atomic_dec(&rcq->usecnt);
+	if (srq)
+		atomic_dec(&srq->usecnt);
+	if (ind_tbl)
+		atomic_dec(&ind_tbl->usecnt);
+	if (sec)
+		ib_destroy_qp_security_end(sec);
+
+	rdma_restrack_del(&qp->res);
+	kfree(qp);
 	return ret;
 }
 EXPORT_SYMBOL(ib_destroy_qp_user);