net/ice: use relaxed and remove duplicate barrier
authorGavin Hu <gavin.hu@arm.com>
Mon, 16 Sep 2019 11:27:15 +0000 (19:27 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 8 Oct 2019 10:14:30 +0000 (12:14 +0200)
To guarantee the orderings of successive stores to CIO and MMIO memory,
a lighter weight rte_io_wmb [1] can be used instead of rte_wmb, and since
the ICE_PCI_REG_WRITE API already has an inclusive rte_io_wmb, this
explicit call can even be saved.

[1] http://git.dpdk.org/dpdk/tree/lib/librte_eal/common/include/generic/
rte_atomic.h#n98

Signed-off-by: Gavin Hu <gavin.hu@arm.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
drivers/net/ice/ice_rxtx.c

index af96c0f..8785cf8 100644 (file)
@@ -402,8 +402,6 @@ ice_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
                return -ENOMEM;
        }
 
-       rte_wmb();
-
        /* Init the RX tail register. */
        ICE_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
 
@@ -1275,7 +1273,6 @@ ice_rx_alloc_bufs(struct ice_rx_queue *rxq)
        }
 
        /* Update rx tail regsiter */
-       rte_wmb();
        ICE_PCI_REG_WRITE(rxq->qrx_tail, rxq->rx_free_trigger);
 
        rxq->rx_free_trigger =
@@ -2177,8 +2174,6 @@ ice_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                                         ICE_TXD_QW1_CMD_S);
        }
 end_of_tx:
-       rte_wmb();
-
        /* update Tail register */
        ICE_PCI_REG_WRITE(txq->qtx_tail, tx_id);
        txq->tx_tail = tx_id;
@@ -2334,7 +2329,6 @@ tx_xmit_pkts(struct ice_tx_queue *txq,
                txq->tx_tail = 0;
 
        /* Update the tx tail register */
-       rte_wmb();
        ICE_PCI_REG_WRITE(txq->qtx_tail, txq->tx_tail);
 
        return nb_pkts;