net/cxgbe: fix Tx queue stuck with mbuf chain coalescing

When trying to coalesce mbufs with chain on Tx side, it is possible
to get stuck during queue wrap around. When coalescing this mbuf
chain fails, the Tx path returns EBUSY and when the same packet
is retried again, it couldn't get coalesced again, and the loop
repeats. Fix by pushing the packet through the normal Tx path.
Also use FW_ETH_TX_PKTS_WR to handle mbufs with chain for FW
to optimize.

Fixes: 6c2809628cd5 ("net/cxgbe: improve latency for slow traffic")
Cc: stable@dpdk.org
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>

diff --git a/drivers/net/cxgbe/sge.c b/drivers/net/cxgbe/sge.c
index 5c176004f9d5f25058f0da3611125e20d12e1650..566cd48406dbe837ce128eaa973753c40bc7887a 100644 (file)
--- a/drivers/net/cxgbe/sge.c
+++ b/drivers/net/cxgbe/sge.c
@@ -789,9 +789,9 @@ static inline void txq_advance(struct sge_txq *q, unsigned int n)
 
 #define MAX_COALESCE_LEN 64000
 
-static inline int wraps_around(struct sge_txq *q, int ndesc)
+static inline bool wraps_around(struct sge_txq *q, int ndesc)
 {
-       return (q->pidx + ndesc) > q->size ? 1 : 0;
+       return (q->pidx + ndesc) > q->size ? true : false;
 }
 
 static void tx_timer_cb(void *data)
@@ -842,7 +842,6 @@ static inline void ship_tx_pkt_coalesce_wr(struct adapter *adap,
 
        /* fill the pkts WR header */
        wr = (void *)&q->desc[q->pidx];
-       wr->op_pkd = htonl(V_FW_WR_OP(FW_ETH_TX_PKTS2_WR));
        vmwr = (void *)&q->desc[q->pidx];
 
        wr_mid = V_FW_WR_LEN16(DIV_ROUND_UP(q->coalesce.flits, 2));
@@ -852,8 +851,11 @@ static inline void ship_tx_pkt_coalesce_wr(struct adapter *adap,
        wr->npkt = q->coalesce.idx;
        wr->r3 = 0;
        if (is_pf4(adap)) {
-               wr->op_pkd = htonl(V_FW_WR_OP(FW_ETH_TX_PKTS2_WR));
                wr->type = q->coalesce.type;
+               if (likely(wr->type != 0))
+                       wr->op_pkd = htonl(V_FW_WR_OP(FW_ETH_TX_PKTS2_WR));
+               else
+                       wr->op_pkd = htonl(V_FW_WR_OP(FW_ETH_TX_PKTS_WR));
        } else {
                wr->op_pkd = htonl(V_FW_WR_OP(FW_ETH_TX_PKTS_VM_WR));
                vmwr->r4 = 0;
@@ -932,13 +934,16 @@ static inline int should_tx_packet_coalesce(struct sge_eth_txq *txq,
                ndesc = DIV_ROUND_UP(q->coalesce.flits + flits, 8);
                credits = txq_avail(q) - ndesc;
 
+               if (unlikely(wraps_around(q, ndesc)))
+                       return 0;
+
                /* If we are wrapping or this is last mbuf then, send the
                 * already coalesced mbufs and let the non-coalesce pass
                 * handle the mbuf.
                 */
-               if (unlikely(credits < 0 || wraps_around(q, ndesc))) {
+               if (unlikely(credits < 0)) {
                        ship_tx_pkt_coalesce_wr(adap, txq);
-                       return 0;
+                       return -EBUSY;
                }
 
                /* If the max coalesce len or the max WR len is reached
@@ -962,8 +967,12 @@ new:
        ndesc = flits_to_desc(q->coalesce.flits + flits);
        credits = txq_avail(q) - ndesc;
 
-       if (unlikely(credits < 0 || wraps_around(q, ndesc)))
+       if (unlikely(wraps_around(q, ndesc)))
                return 0;
+
+       if (unlikely(credits < 0))
+               return -EBUSY;
+
        q->coalesce.flits += wr_size / sizeof(__be64);
        q->coalesce.type = type;
        q->coalesce.ptr = (unsigned char *)&q->desc[q->pidx] +
@@ -1106,7 +1115,7 @@ int t4_eth_xmit(struct sge_eth_txq *txq, struct rte_mbuf *mbuf,
        unsigned int flits, ndesc, cflits;
        int l3hdr_len, l4hdr_len, eth_xtra_len;
        int len, last_desc;
-       int credits;
+       int should_coal, credits;
        u32 wr_mid;
        u64 cntrl, *end;
        bool v6;
@@ -1138,9 +1147,9 @@ out_free:
        /* align the end of coalesce WR to a 512 byte boundary */
        txq->q.coalesce.max = (8 - (txq->q.pidx & 7)) * 8;
 
-       if (!((m->ol_flags & RTE_MBUF_F_TX_TCP_SEG) ||
-                       m->pkt_len > RTE_ETHER_MAX_LEN)) {
-               if (should_tx_packet_coalesce(txq, mbuf, &cflits, adap)) {
+       if ((m->ol_flags & RTE_MBUF_F_TX_TCP_SEG) == 0) {
+               should_coal = should_tx_packet_coalesce(txq, mbuf, &cflits, adap);
+               if (should_coal > 0) {
                        if (unlikely(map_mbuf(mbuf, addr) < 0)) {
                                dev_warn(adap, "%s: mapping err for coalesce\n",
                                         __func__);
@@ -1149,8 +1158,8 @@ out_free:
                        }
                        return tx_do_packet_coalesce(txq, mbuf, cflits, adap,
                                                     pi, addr, nb_pkts);
-               } else {
-                       return -EBUSY;
+               } else if (should_coal < 0) {
+                       return should_coal;
                }
        }
 
@@ -1197,8 +1206,7 @@ out_free:
                end = (u64 *)vmwr + flits;
        }
 
-       len = 0;
-       len += sizeof(*cpl);
+       len = sizeof(*cpl);
 
        /* Coalescing skipped and we send through normal path */
        if (!(m->ol_flags & RTE_MBUF_F_TX_TCP_SEG)) {