rdma: optimize tx wqe_init

Type: improvement
Change-Id: I7f28a3f03ab1ea8461c52743c61dc23a57965237
Signed-off-by: Damjan Marion <damarion@cisco.com>

3 files changed, 33 insertions, 13 deletions

diff --git a/src/plugins/rdma/device.c b/src/plugins/rdma/device.c
index eb13f855b1a..ba7b7de9323 100644
--- a/src/plugins/rdma/device.c
+++ b/src/plugins/rdma/device.c
@@ -617,7 +617,7 @@ rdma_txq_init (vlib_main_t * vm, rdma_device_t * rd, u16 qid, u32 n_desc)
       mlx5dv_set_ctrl_seg (&tmpl->ctrl, 0, MLX5_OPCODE_SEND, 0,
 			   txq->qp->qp_num, 0, RDMA_MLX5_WQE_DS, 0,
 			   RDMA_TXQ_DV_INVALID_ID);
-      /* FIXME: mlx5dv_set_eth_seg(&tmpl->eseg, MLX5_ETH_WQE_L3_CSUM | MLX5_ETH_WQE_L4_CSUM, 0, 0, 0); */
+      tmpl->eseg.inline_hdr_sz = htobe16 (MLX5_ETH_L2_INLINE_HEADER_SIZE);
       mlx5dv_set_data_seg (&tmpl->dseg, 0, rd->lkey, 0);
     }
 
diff --git a/src/plugins/rdma/output.c b/src/plugins/rdma/output.c
index 015208c023e..848b6fd9c96 100644
--- a/src/plugins/rdma/output.c
+++ b/src/plugins/rdma/output.c
@@ -23,10 +23,6 @@
 #include <vnet/devices/devices.h>
 #include <rdma/rdma.h>
 
-#ifndef MLX5_ETH_L2_INLINE_HEADER_SIZE
-#define MLX5_ETH_L2_INLINE_HEADER_SIZE  18
-#endif
-
 #define RDMA_TX_RETRIES 5
 
 #define RDMA_TXQ_DV_DSEG_SZ(txq)        (RDMA_MLX5_WQE_DS * RDMA_TXQ_DV_SQ_SZ(txq))
@@ -113,18 +109,28 @@ rdma_mlx5_wqe_init (rdma_mlx5_wqe_t * wqe, const void *tmpl,
 		    vlib_buffer_t * b, const u16 tail)
 {
   u16 sz = b->current_length;
-  u16 inline_sz = clib_min (sz, MLX5_ETH_L2_INLINE_HEADER_SIZE);
+  const void *cur = vlib_buffer_get_current (b);
+  uword addr = pointer_to_uword (cur);
 
   clib_memcpy_fast (wqe, tmpl, RDMA_MLX5_WQE_SZ);
-
-  wqe->ctrl.opmod_idx_opcode |= ((u32) htobe16 (tail)) << 8;
   /* speculatively copy at least MLX5_ETH_L2_INLINE_HEADER_SIZE (18-bytes) */
-  const void *cur = vlib_buffer_get_current (b);
   clib_memcpy_fast (wqe->eseg.inline_hdr_start,
 		    cur, MLX5_ETH_L2_INLINE_HEADER_SIZE);
-  wqe->eseg.inline_hdr_sz = htobe16 (inline_sz);
-  wqe->dseg.byte_count = htobe32 (sz - inline_sz);
-  wqe->dseg.addr = htobe64 (pointer_to_uword (cur) + inline_sz);
+
+  wqe->wqe_index_lo = tail;
+  wqe->wqe_index_hi = tail >> 8;
+  if (PREDICT_TRUE (sz >= MLX5_ETH_L2_INLINE_HEADER_SIZE))
+    {
+      /* inline_hdr_sz is set to MLX5_ETH_L2_INLINE_HEADER_SIZE
+         in the template */
+      wqe->dseg.byte_count = htobe32 (sz - MLX5_ETH_L2_INLINE_HEADER_SIZE);
+      wqe->dseg.addr = htobe64 (addr + MLX5_ETH_L2_INLINE_HEADER_SIZE);
+    }
+  else
+    {
+      /* dseg.byte_count and desg.addr are set to 0 in the template */
+      wqe->eseg.inline_hdr_sz = htobe16 (sz);
+    }
 }
 
 /*
diff --git a/src/plugins/rdma/rdma.h b/src/plugins/rdma/rdma.h
index 82f32ec9d01..016956ee0f2 100644
--- a/src/plugins/rdma/rdma.h
+++ b/src/plugins/rdma/rdma.h
@@ -39,10 +39,24 @@ enum
 #undef _
 };
 
+#ifndef MLX5_ETH_L2_INLINE_HEADER_SIZE
+#define MLX5_ETH_L2_INLINE_HEADER_SIZE  18
+#endif
+
 typedef struct
 {
   CLIB_ALIGN_MARK (align0, MLX5_SEND_WQE_BB);
-  struct mlx5_wqe_ctrl_seg ctrl;
+  union
+  {
+    struct mlx5_wqe_ctrl_seg ctrl;
+    struct
+    {
+      u8 opc_mod;
+      u8 wqe_index_hi;
+      u8 wqe_index_lo;
+      u8 opcode;
+    };
+  };
   struct mlx5_wqe_eth_seg eseg;
   struct mlx5_wqe_data_seg dseg;
 } rdma_mlx5_wqe_t;