patches for DPDK stable branches
 help / color / mirror / Atom feed
* [dpdk-stable] [PATCH] net/mlx5: fix metadata calculation in legacy mode
@ 2021-06-30  2:24 rongwei liu
  0 siblings, 0 replies; only message in thread
From: rongwei liu @ 2021-06-30  2:24 UTC (permalink / raw)
  To: matan, viacheslavo, orika, thomas, Shahaf Shuler,
	David Christensen, Jerin Jacob, Ruifeng Wang, Bruce Richardson,
	Konstantin Ananyev
  Cc: dev, rasland, stable

Currently, we store metadata in BE and need to change
it to CPU endian when attaching to mbuf.

For dv_xmeta_en = 2, no bit shift.
For dv_xmeta_en = 1, 16 bits shift.

For the non-vector mode, endian change should be first,
then bit shift.

For the vector mode, the behavior is right. In this update,
we use __builtin_clz instead of __builtin_popcount to
avoid confusing.

Fixes: 743ac28ddf0b ("net/mlx5: fix metadata calculation in legacy mode")
Cc: stable@dpdk.org

Signed-off-by: rongwei liu <rongweil@nvidia.com>
---
 drivers/net/mlx5/mlx5_rx.c               | 4 ++--
 drivers/net/mlx5/mlx5_rxtx_vec_altivec.h | 2 +-
 drivers/net/mlx5/mlx5_rxtx_vec_neon.h    | 2 +-
 drivers/net/mlx5/mlx5_rxtx_vec_sse.h     | 2 +-
 4 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rx.c b/drivers/net/mlx5/mlx5_rx.c
index 777a1d6e45..012745dc26 100644
--- a/drivers/net/mlx5/mlx5_rx.c
+++ b/drivers/net/mlx5/mlx5_rx.c
@@ -740,8 +740,8 @@ rxq_cq_to_mbuf(struct mlx5_rxq_data *rxq, struct rte_mbuf *pkt,
 		}
 	}
 	if (rxq->dynf_meta) {
-		uint32_t meta = rte_be_to_cpu_32(cqe->flow_table_metadata >>
-			__builtin_popcount(rxq->flow_meta_port_mask)) &
+		uint32_t meta = (rte_be_to_cpu_32(cqe->flow_table_metadata) >>
+			__builtin_clz(rxq->flow_meta_port_mask)) &
 			rxq->flow_meta_port_mask;
 
 		if (meta) {
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec_altivec.h b/drivers/net/mlx5/mlx5_rxtx_vec_altivec.h
index 648c59e2c2..4a0da1bd3b 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec_altivec.h
+++ b/drivers/net/mlx5/mlx5_rxtx_vec_altivec.h
@@ -1223,7 +1223,7 @@ rxq_cq_process_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,
 			int32_t offs = rxq->flow_meta_offset;
 			uint32_t mask = rxq->flow_meta_port_mask;
 			uint32_t shift =
-				__builtin_popcount(rxq->flow_meta_port_mask);
+				__builtin_clz(rxq->flow_meta_port_mask);
 			uint32_t metadata;
 
 			/* This code is subject for futher optimization. */
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec_neon.h b/drivers/net/mlx5/mlx5_rxtx_vec_neon.h
index 5c569ee199..38cbcb6fdc 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec_neon.h
+++ b/drivers/net/mlx5/mlx5_rxtx_vec_neon.h
@@ -834,7 +834,7 @@ rxq_cq_process_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,
 			int32_t offs = rxq->flow_meta_offset;
 			uint32_t mask = rxq->flow_meta_port_mask;
 			uint32_t shift =
-				__builtin_popcount(rxq->flow_meta_port_mask);
+				__builtin_clz(rxq->flow_meta_port_mask);
 
 			*RTE_MBUF_DYNFIELD(pkts[pos], offs, uint32_t *) =
 				(rte_be_to_cpu_32(container_of
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec_sse.h b/drivers/net/mlx5/mlx5_rxtx_vec_sse.h
index 661fa7273c..480583acbb 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec_sse.h
+++ b/drivers/net/mlx5/mlx5_rxtx_vec_sse.h
@@ -770,7 +770,7 @@ rxq_cq_process_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,
 			int32_t offs = rxq->flow_meta_offset;
 			uint32_t mask = rxq->flow_meta_port_mask;
 			uint32_t shift =
-				__builtin_popcount(rxq->flow_meta_port_mask);
+				__builtin_clz(rxq->flow_meta_port_mask);
 
 			*RTE_MBUF_DYNFIELD(pkts[pos], offs, uint32_t *) =
 				(rte_be_to_cpu_32
-- 
2.27.0


^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2021-06-30  2:25 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-06-30  2:24 [dpdk-stable] [PATCH] net/mlx5: fix metadata calculation in legacy mode rongwei liu

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).