From: Leon Romanovsky <leonro@nvidia.com>
Date: Fri, 23 Jul 2021 14:39:49 +0300
Subject: RDMA/rdmavt: Decouple QP and SGE lists allocations
Patch-mainline: v5.15-rc1
Git-commit: 44da3730e046a784d088157175d9418ba60661fc
References: jsc#SLE-19249

The rdmavt QP has fields that are both needed for the control and data
path. Such mixed declaration caused to the very specific allocation flow
with kzalloc_node and SGE list embedded into the struct rvt_qp.

This patch separates QP creation to two: regular memory allocation for the
control path and specific code for the SGE list, while the access to the
later is performed through derefenced pointer.

Such pointer and its context are expected to be in the cache, so
performance difference is expected to be negligible, if any exists.

Link: https://lore.kernel.org/r/f66c1e20ccefba0db3c69c58ca9c897f062b4d1c.1627040189.git.leonro@nvidia.com
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
 drivers/infiniband/sw/rdmavt/qp.c |   13 ++++++++-----
 include/rdma/rdmavt_qp.h          |    2 +-
 2 files changed, 9 insertions(+), 6 deletions(-)

--- a/drivers/infiniband/sw/rdmavt/qp.c
+++ b/drivers/infiniband/sw/rdmavt/qp.c
@@ -1078,7 +1078,7 @@ struct ib_qp *rvt_create_qp(struct ib_pd
 	int err;
 	struct rvt_swqe *swq = NULL;
 	size_t sz;
-	size_t sg_list_sz;
+	size_t sg_list_sz = 0;
 	struct ib_qp *ret = ERR_PTR(-ENOMEM);
 	struct rvt_dev_info *rdi = ib_to_rvt(ibpd->device);
 	void *priv = NULL;
@@ -1126,8 +1126,6 @@ struct ib_qp *rvt_create_qp(struct ib_pd
 		if (!swq)
 			return ERR_PTR(-ENOMEM);
 
-		sz = sizeof(*qp);
-		sg_list_sz = 0;
 		if (init_attr->srq) {
 			struct rvt_srq *srq = ibsrq_to_rvtsrq(init_attr->srq);
 
@@ -1137,10 +1135,13 @@ struct ib_qp *rvt_create_qp(struct ib_pd
 		} else if (init_attr->cap.max_recv_sge > 1)
 			sg_list_sz = sizeof(*qp->r_sg_list) *
 				(init_attr->cap.max_recv_sge - 1);
-		qp = kzalloc_node(sz + sg_list_sz, GFP_KERNEL,
-				  rdi->dparms.node);
+		qp = kzalloc_node(sizeof(*qp), GFP_KERNEL, rdi->dparms.node);
 		if (!qp)
 			goto bail_swq;
+		qp->r_sg_list =
+			kzalloc_node(sg_list_sz, GFP_KERNEL, rdi->dparms.node);
+		if (!qp->r_sg_list)
+			goto bail_qp;
 		qp->allowed_ops = get_allowed_ops(init_attr->qp_type);
 
 		RCU_INIT_POINTER(qp->next, NULL);
@@ -1328,6 +1329,7 @@ bail_driver_priv:
 
 bail_qp:
 	kfree(qp->s_ack_queue);
+	kfree(qp->r_sg_list);
 	kfree(qp);
 
 bail_swq:
@@ -1762,6 +1764,7 @@ int rvt_destroy_qp(struct ib_qp *ibqp, s
 	kvfree(qp->r_rq.kwq);
 	rdi->driver_f.qp_priv_free(rdi, qp);
 	kfree(qp->s_ack_queue);
+	kfree(qp->r_sg_list);
 	rdma_destroy_ah_attr(&qp->remote_ah_attr);
 	rdma_destroy_ah_attr(&qp->alt_ah_attr);
 	free_ud_wq_attr(qp);
--- a/include/rdma/rdmavt_qp.h
+++ b/include/rdma/rdmavt_qp.h
@@ -444,7 +444,7 @@ struct rvt_qp {
 	/*
 	 * This sge list MUST be last. Do not add anything below here.
 	 */
-	struct rvt_sge r_sg_list[] /* verified SGEs */
+	struct rvt_sge *r_sg_list /* verified SGEs */
 		____cacheline_aligned_in_smp;
 };