Under the circumstance that `rx_tail` wrap back to zero
and the advance speed of `rx_tail` is greater than `rxrearm_start`,
`rx_tail` will catch up with `rxrearm_start` and surpass it.
This may cause some mbufs be reused by applicaion.
So we need to make some restrictions to ensure that
`rx_tail` will not exceed `rxrearm_start`.
e.g.
RDH: 972 RDT: 991 rxrearm_nb: 991 rxrearm_start: 992 rx_tail: 959
RDH: 1004 RDT: 1023 rxrearm_nb: 991 rxrearm_start: 0 rx_tail: 991
RDH: 12 RDT: 31 rxrearm_nb: 991 rxrearm_start: 32 rx_tail: 1023
RDH: 31 RDT: 63 rxrearm_nb: 960 rxrearm_start: 64 rx_tail: 0
RDH: 95 RDT: 95 rxrearm_nb: 1016 rxrearm_start: 96 rx_tail: 88
RDH: 95 RDT: 127 rxrearm_nb: 991 rxrearm_start: 128 rx_tail: 95
...
RDH: 908 RDT: 927 rxrearm_nb: 991 rxrearm_start: 928 rx_tail: 895
RDH: 940 RDT: 959 rxrearm_nb: 991 rxrearm_start: 960 rx_tail: 927
RDH: 980 RDT: 991 rxrearm_nb: 991 rxrearm_start: 992 rx_tail: 959
RDH: 991 RDT: 991 rxrearm_nb: 1026 rxrearm_start: 992 rx_tail: 994
when `rx_tail` catches up with `rxrearm_start`,
2(994 - 992) mbufs be reused by applicaion !
Bugzilla ID: 882
Fixes: 5a3cca342417 ("net/ixgbe: fix vector Rx")
Cc: jia....@intel.com
Cc: sta...@dpdk.org
Signed-off-by: Bin Zheng <zhengbin.89...@bytedance.com>
---
drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c | 18 +++++++++++++++---
1 file changed, 15 insertions(+), 3 deletions(-)
diff --git a/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
b/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
index 1eed949495..33d2e96266 100644
--- a/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
+++ b/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
@@ -364,9 +364,6 @@ _recv_raw_pkts_vec(struct ixgbe_rx_queue *rxq, struct
rte_mbuf **rx_pkts,
uint8_t vlan_flags;
uint16_t udp_p_flag = 0; /* Rx Descriptor UDP header present */
- /* nb_pkts has to be floor-aligned to RTE_IXGBE_DESCS_PER_LOOP */
- nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_IXGBE_DESCS_PER_LOOP);
-
/* Just the act of getting into the function from the application is
* going to cost about 7 cycles
*/
@@ -380,6 +377,21 @@ _recv_raw_pkts_vec(struct ixgbe_rx_queue *rxq, struct
rte_mbuf **rx_pkts,
if (rxq->rxrearm_nb > RTE_IXGBE_RXQ_REARM_THRESH)
ixgbe_rxq_rearm(rxq);
+ /*
+ * Under the circumstance that `rx_tail` wrap back to zero
+ * and the advance speed of `rx_tail` is greater than `rxrearm_start`,
+ * `rx_tail` will catch up with `rxrearm_start` and surpass it.
+ * This may cause some mbufs be reused by applicaion.
+ *
+ * So we need to make some restrictions to ensure that
+ * `rx_tail` will not exceed `rxrearm_start`.
+ */
+ if (rxq->rx_tail < rxq->rxrearm_start)
+ nb_pkts = RTE_MIN(nb_pkts, rxq->rxrearm_start - rxq->rx_tail -
1);
+
+ /* nb_pkts has to be floor-aligned to RTE_IXGBE_DESCS_PER_LOOP */
+ nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_IXGBE_DESCS_PER_LOOP);
+
/* Before we start moving massive data around, check to see if
* there is actually a packet available
*/
--
2.25.1
