[PATCH] net/mlx5: fix statistics for error packets on vectorized Rx

[PATCH] net/mlx5: fix statistics for error packets on vectorized Rx

[Yongseok Koh]

If there's a Rx completion with error (e.g, MTU mismatch), it is handled
later out of main burst loop as a slow path for performance reason.
Statistics should be corrected by subtracting counters of errored packets.
Also, the last entry of mlx5_ptype_table[] must be RTE_PTYPE_ALL_MASK to
mark error in completion.

Fixes: ea16068c0064 ("net/mlx5: fix L4 packet type support")
Fixes: 6cb559d67b83 ("net/mlx5: add vectorized Rx/Tx burst for x86")

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
 drivers/net/mlx5/mlx5_rxtx.c         |  3 ++-
 drivers/net/mlx5/mlx5_rxtx_vec_sse.c | 16 ++++++++++++++--
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 2572a1652..b07bcd118 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -94,7 +94,8 @@ mlx5_set_ptype_table(void)
 	unsigned int i;
 	uint32_t (*p)[RTE_DIM(mlx5_ptype_table)] = &mlx5_ptype_table;
 
-	for (i = 0; i < RTE_DIM(mlx5_ptype_table); ++i)
+	/* Last entry must not be overwritten, reserved for errored packet. */
+	for (i = 0; i < RTE_DIM(mlx5_ptype_table) - 1; ++i)
 		(*p)[i] = RTE_PTYPE_UNKNOWN;
 	/*
 	 * The index to the array should have:
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec_sse.c b/drivers/net/mlx5/mlx5_rxtx_vec_sse.c
index 74e595386..4202411ab 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec_sse.c
+++ b/drivers/net/mlx5/mlx5_rxtx_vec_sse.c
@@ -885,16 +885,28 @@ rxq_handle_pending_error(struct rxq *rxq, struct rte_mbuf **pkts,
 {
 	uint16_t n = 0;
 	unsigned int i;
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	uint32_t err_bytes = 0;
+#endif
 
 	for (i = 0; i < pkts_n; ++i) {
 		struct rte_mbuf *pkt = pkts[i];
 
-		if (pkt->packet_type == RTE_PTYPE_ALL_MASK)
+		if (pkt->packet_type == RTE_PTYPE_ALL_MASK) {
+#ifdef MLX5_PMD_SOFT_COUNTERS
+			err_bytes += PKT_LEN(pkt);
+#endif
 			rte_pktmbuf_free_seg(pkt);
-		else
+		} else {
 			pkts[n++] = pkt;
+		}
 	}
 	rxq->stats.idropped += (pkts_n - n);
+#ifdef MLX5_PMD_SOFT_COUNTERS
+	/* Correct counters of errored completions. */
+	rxq->stats.ipackets -= (pkts_n - n);
+	rxq->stats.ibytes -= err_bytes;
+#endif
 	rxq->pending_err = 0;
 	return n;
 }
-- 
2.11.0

Re: [PATCH] net/mlx5: fix statistics for error packets on vectorized Rx

[Thomas Monjalon]

02/08/2017 19:25, Yongseok Koh:
> If there's a Rx completion with error (e.g, MTU mismatch), it is handled
> later out of main burst loop as a slow path for performance reason.
> Statistics should be corrected by subtracting counters of errored packets.
> Also, the last entry of mlx5_ptype_table[] must be RTE_PTYPE_ALL_MASK to
> mark error in completion.
> 
> Fixes: ea16068c0064 ("net/mlx5: fix L4 packet type support")
> Fixes: 6cb559d67b83 ("net/mlx5: add vectorized Rx/Tx burst for x86")
> 
> Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
> Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>

Applied, thanks