Add TX completion handling. I'm not quite sure, but I think that we

receive interrupts for the physical TX queues, but the TX buffers
which need to be freed were taken from the aggregated (per core) TX
queue.  This means we probably should have the physical TX queues
tied to specific cores, so that the TX enqueue and TX completion
share the same per-core info for the free-handling.  For now we only
have a single physical and aggregated TX queue, so it's comparatively
easy.

1 files changed, 39 insertions, 8 deletions

diff --git a/sys/dev/fdt/if_mvpp.c b/sys/dev/fdt/if_mvpp.c
index b367a80b255..4511646c7d1 100644
--- a/sys/dev/fdt/if_mvpp.c
+++ b/sys/dev/fdt/if_mvpp.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_mvpp.c,v 1.4 2020/06/26 09:30:10 patrick Exp $	*/
+/*	$OpenBSD: if_mvpp.c,v 1.5 2020/06/26 09:40:42 patrick Exp $	*/
 /*
  * Copyright (c) 2008, 2019 Mark Kettenis <kettenis@openbsd.org>
  * Copyright (c) 2017, 2020 Patrick Wildt <patrick@blueri.se>
@@ -280,6 +280,7 @@ void	mvpp2_rxtick(void *);
 int	mvpp2_link_intr(void *);
 int	mvpp2_intr(void *);
 void	mvpp2_tx_proc(struct mvpp2_port *, uint8_t);
+void	mvpp2_txq_proc(struct mvpp2_port *, struct mvpp2_tx_queue *);
 void	mvpp2_rx_proc(struct mvpp2_port *, uint8_t);
 void	mvpp2_rxq_proc(struct mvpp2_port *, struct mvpp2_rx_queue *);
 
@@ -1967,22 +1968,52 @@ mvpp2_intr(void *arg)
 void
 mvpp2_tx_proc(struct mvpp2_port *sc, uint8_t queues)
 {
-//	struct ifnet *ifp = &sc->sc_ac.ac_if;
 	struct mvpp2_tx_queue *txq;
-	uint32_t reg;
 	int i;
 
 	for (i = 0; i < sc->sc_ntxq; i++) {
 		txq = &sc->sc_txqs[i];
 		if ((queues & (1 << i)) == 0)
 			continue;
-		reg = mvpp2_read(sc->sc, MVPP2_TXQ_SENT_REG(txq->id));
-		printf("%s: txq %u sent reg %u\n", sc->sc_dev.dv_xname,
-		    i, (reg & MVPP2_TRANSMITTED_COUNT_MASK) >>
-		    MVPP2_TRANSMITTED_COUNT_OFFSET);
+		mvpp2_txq_proc(sc, txq);
 	}
+}
 
-	/* FIXME: tx done processing */
+void
+mvpp2_txq_proc(struct mvpp2_port *sc, struct mvpp2_tx_queue *txq)
+{
+	struct ifnet *ifp = &sc->sc_ac.ac_if;
+	struct mvpp2_tx_queue *aggr_txq = &sc->sc->sc_aggr_txqs[0];
+	struct mvpp2_buf *txb;
+	int i, idx, nsent;
+
+	nsent = (mvpp2_read(sc->sc, MVPP2_TXQ_SENT_REG(txq->id)) &
+	    MVPP2_TRANSMITTED_COUNT_MASK) >>
+	    MVPP2_TRANSMITTED_COUNT_OFFSET;
+
+	for (i = 0; i < nsent; i++) {
+		idx = aggr_txq->cons;
+		KASSERT(idx < MVPP2_AGGR_TXQ_SIZE);
+
+		txb = &aggr_txq->buf[idx];
+		if (txb->mb_m) {
+			bus_dmamap_sync(sc->sc_dmat, txb->mb_map, 0,
+			    txb->mb_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
+			bus_dmamap_unload(sc->sc_dmat, txb->mb_map);
+
+			m_freem(txb->mb_m);
+			txb->mb_m = NULL;
+		}
+
+		aggr_txq->cnt--;
+		aggr_txq->cons = (aggr_txq->cons + 1) % MVPP2_AGGR_TXQ_SIZE;
+	}
+
+	if (aggr_txq->cnt == 0)
+		ifp->if_timer = 0;
+
+	if (ifq_is_oactive(&ifp->if_snd))
+		ifq_restart(&ifp->if_snd);
 }
 
 void