IB/mlx5: posting klm/mtt list inline in the send queue for reg_wr

As most kernel RDMA ULPs, (e.g. NVMe over Fabrics in its default
"register_always=Y" mode) registers and invalidates user buffer
upon each IO.

Today the mlx5 driver is posting the registration work
request using scatter/gather entry for the MTT/KLM list.
The fetch of the MTT/KLM list becomes the bottleneck in
number of IO operation could be done by NVMe over Fabrics
host driver on a single adapter as shown below.

This patch is adding the support for inline registration
work request upon MTT/KLM list of size <=64B.

The result for NVMe over Fabrics is increase of > x3.5 for small
IOs as shown below, I expect other ULPs (e.g iSER, SRP, NFS over RDMA)
performance to be enhanced as well.

The following results were taken against a single NVMe-oF (RoCE link layer)
subsystem with a single namespace backed by null_blk using fio benchmark
(with rw=randread, numjobs=48, iodepth={16,64}, ioengine=libaio direct=1):

ConnectX-5 (pci Width x16)
---------------------------

Block Size       s/g reg_wr            inline reg_wr
++++++++++     +++++++++++++++        ++++++++++++++++
512B            1302.8K/34.82%         4951.9K/99.02%
1KB             1284.3K/33.86%         4232.7K/98.09%
2KB             1238.6K/34.1%          2797.5K/80.04%
4KB             1169.3K/32.46%         1941.3K/61.35%
8KB             1013.4K/30.08%         1236.6K/39.47%
16KB            695.7K/20.19%          696.9K/20.59%
32KB            350.3K/9.64%           350.6K/10.3%
64KB            175.86K/5.27%          175.9K/5.28%

ConnectX-4 (pci Width x8)
---------------------------

Block Size       s/g reg_wr            inline reg_wr
++++++++++     +++++++++++++++        ++++++++++++++++
512B            1285.8K/42.66%          4242.7K/98.18%
1KB             1254.1K/41.74%          3569.2K/96.00%
2KB             1185.9K/39.83%          2173.9K/75.58%
4KB             1069.4K/36.46%          1343.3K/47.47%
8KB             755.1K/27.77%           748.7K/29.14%

Tested-by: Nitzan Carmi <nitzanc@mellanox.com>
Signed-off-by: Idan Burstein <idanb@mellanox.com>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

Signed-off-by: Doug Ledford <dledford@redhat.com>

drivers/infiniband/hw/mlx5/qp.c

@@ -54,6 +54,7 @@ enum {
 
 enum {
 	MLX5_IB_SQ_STRIDE	= 6,
+	MLX5_IB_SQ_UMR_INLINE_THRESHOLD = 64,
 };
 
 static const u32 mlx5_ib_opcode[] = {
@@ -298,7 +299,9 @@ static int sq_overhead(struct ib_qp_init_attr *attr)
 			max(sizeof(struct mlx5_wqe_atomic_seg) +
 			    sizeof(struct mlx5_wqe_raddr_seg),
 			    sizeof(struct mlx5_wqe_umr_ctrl_seg) +
-			    sizeof(struct mlx5_mkey_seg));
+			    sizeof(struct mlx5_mkey_seg) +
+			    MLX5_IB_SQ_UMR_INLINE_THRESHOLD /
+			    MLX5_IB_UMR_OCTOWORD);
 		break;
 
 	case IB_QPT_XRC_TGT:
@@ -3633,13 +3636,15 @@ static __be64 sig_mkey_mask(void)
 }
 
 static void set_reg_umr_seg(struct mlx5_wqe_umr_ctrl_seg *umr,
-			    struct mlx5_ib_mr *mr)
+			    struct mlx5_ib_mr *mr, bool umr_inline)
 {
 	int size = mr->ndescs * mr->desc_size;
 
 	memset(umr, 0, sizeof(*umr));
 
 	umr->flags = MLX5_UMR_CHECK_NOT_FREE;
+	if (umr_inline)
+		umr->flags |= MLX5_UMR_INLINE;
 	umr->xlt_octowords = cpu_to_be16(get_xlt_octo(size));
 	umr->mkey_mask = frwr_mkey_mask();
 }
@@ -3823,6 +3828,24 @@ static void set_reg_data_seg(struct mlx5_wqe_data_seg *dseg,
 	dseg->lkey = cpu_to_be32(pd->ibpd.local_dma_lkey);
 }
 
+static void set_reg_umr_inline_seg(void *seg, struct mlx5_ib_qp *qp,
+				   struct mlx5_ib_mr *mr, int mr_list_size)
+{
+	void *qend = qp->sq.qend;
+	void *addr = mr->descs;
+	int copy;
+
+	if (unlikely(seg + mr_list_size > qend)) {
+		copy = qend - seg;
+		memcpy(seg, addr, copy);
+		addr += copy;
+		mr_list_size -= copy;
+		seg = mlx5_get_send_wqe(qp, 0);
+	}
+	memcpy(seg, addr, mr_list_size);
+	seg += mr_list_size;
+}
+
 static __be32 send_ieth(struct ib_send_wr *wr)
 {
 	switch (wr->opcode) {
@@ -4217,6 +4240,8 @@ static int set_reg_wr(struct mlx5_ib_qp *qp,
 {
 	struct mlx5_ib_mr *mr = to_mmr(wr->mr);
 	struct mlx5_ib_pd *pd = to_mpd(qp->ibqp.pd);
+	int mr_list_size = mr->ndescs * mr->desc_size;
+	bool umr_inline = mr_list_size <= MLX5_IB_SQ_UMR_INLINE_THRESHOLD;
 
 	if (unlikely(wr->wr.send_flags & IB_SEND_INLINE)) {
 		mlx5_ib_warn(to_mdev(qp->ibqp.device),
@@ -4224,7 +4249,7 @@ static int set_reg_wr(struct mlx5_ib_qp *qp,
 		return -EINVAL;
 	}
 
-	set_reg_umr_seg(*seg, mr);
+	set_reg_umr_seg(*seg, mr, umr_inline);
 	*seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
 	*size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
 	if (unlikely((*seg == qp->sq.qend)))
@@ -4236,10 +4261,14 @@ static int set_reg_wr(struct mlx5_ib_qp *qp,
 	if (unlikely((*seg == qp->sq.qend)))
 		*seg = mlx5_get_send_wqe(qp, 0);
 
-	set_reg_data_seg(*seg, mr, pd);
-	*seg += sizeof(struct mlx5_wqe_data_seg);
-	*size += (sizeof(struct mlx5_wqe_data_seg) / 16);
-
+	if (umr_inline) {
+		set_reg_umr_inline_seg(*seg, qp, mr, mr_list_size);
+		*size += get_xlt_octo(mr_list_size);
+	} else {
+		set_reg_data_seg(*seg, mr, pd);
+		*seg += sizeof(struct mlx5_wqe_data_seg);
+		*size += (sizeof(struct mlx5_wqe_data_seg) / 16);
+	}
 	return 0;
 }