net/i40e: fix out of order Rx read

In vPMD, when we load Rx desc with _mm_loadu_si128, the volatile
modifier will be cast away, allowing the compiler to reorder the
load instructions.

The Rx recv function's correctness is relying on these load
instructions following a strict sequence, reading the descriptors
in reverse order, so we add compiler barrier to prevent compiler
reorder.

Fixes: 9ed94e5bb04e ("i40e: add vector Rx")

Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>

diff --git a/drivers/net/i40e/i40e_rxtx_vec_sse.c b/drivers/net/i40e/i40e_rxtx_vec_sse.c
index 3607312..7c84a41 100644 (file)
--- a/drivers/net/i40e/i40e_rxtx_vec_sse.c
+++ b/drivers/net/i40e/i40e_rxtx_vec_sse.c
@@ -306,6 +306,7 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
                /* Read desc statuses backwards to avoid race condition */
                /* A.1 load 4 pkts desc */
                descs[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
+               rte_compiler_barrier();
 
                /* B.2 copy 2 mbuf point into rx_pkts  */
                _mm_storeu_si128((__m128i *)&rx_pkts[pos], mbp1);
@@ -314,8 +315,10 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
                mbp2 = _mm_loadu_si128((__m128i *)&sw_ring[pos+2]);
 
                descs[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
+               rte_compiler_barrier();
                /* B.1 load 2 mbuf point */
                descs[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
+               rte_compiler_barrier();
                descs[0] = _mm_loadu_si128((__m128i *)(rxdp));
 
                /* B.2 copy 2 mbuf point into rx_pkts  */