9484 cxgbe should clean TX descriptors in timely manner

Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
Reviewed by: Ryan Zezeski <rpz@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Toomas Soome <tsoome@me.com>
Approved by: Dan McDonald <danmcd@joyent.com>

3 files changed, 39 insertions, 2 deletions

diff --git a/usr/src/uts/common/io/cxgbe/t4nex/adapter.h b/usr/src/uts/common/io/cxgbe/t4nex/adapter.h
index cf6fd497fd..7aedd30c6a 100644
--- a/usr/src/uts/common/io/cxgbe/t4nex/adapter.h
+++ b/usr/src/uts/common/io/cxgbe/t4nex/adapter.h
@@ -507,6 +507,7 @@ struct adapter {
 	struct sge sge;
 
 	struct port_info *port[MAX_NPORTS];
+	ddi_taskq_t *tq[NCHAN];
 	uint8_t chan_map[NCHAN];
 	uint32_t filter_mode;
 
diff --git a/usr/src/uts/common/io/cxgbe/t4nex/t4_nexus.c b/usr/src/uts/common/io/cxgbe/t4nex/t4_nexus.c
index 5607385283..6de5af0e30 100644
--- a/usr/src/uts/common/io/cxgbe/t4nex/t4_nexus.c
+++ b/usr/src/uts/common/io/cxgbe/t4nex/t4_nexus.c
@@ -369,6 +369,20 @@ t4_devo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
 		sc->fw_msg_handler[i] = fw_msg_not_handled;
 	}
  
+	for (i = 0; i < NCHAN; i++) {
+		(void) snprintf(name, sizeof (name), "%s-%d",
+				"reclaim", i);
+		sc->tq[i] = ddi_taskq_create(sc->dip,
+		   name, 1, TASKQ_DEFAULTPRI, 0);
+
+		if (sc->tq[i] == NULL) {
+			cxgb_printf(dip, CE_WARN,
+				   "failed to create task queues");
+			rc = DDI_FAILURE;
+			goto done;
+		}
+	}
+
 	/*
 	 * Prepare the adapter for operation.
 	 */
@@ -836,6 +850,13 @@ t4_devo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
 	ddi_prop_remove_all(dip);
 	ddi_remove_minor_node(dip, NULL);
 
+	for (i = 0; i < NCHAN; i++) {
+		if (sc->tq[i]) {
+			ddi_taskq_wait(sc->tq[i]);
+			ddi_taskq_destroy(sc->tq[i]);
+		}
+	}
+
 	if (sc->ksp != NULL)
 		kstat_delete(sc->ksp);
 	if (sc->ksp_stat != NULL)
diff --git a/usr/src/uts/common/io/cxgbe/t4nex/t4_sge.c b/usr/src/uts/common/io/cxgbe/t4nex/t4_sge.c
index 6f68670bc9..2109c6fcf2 100644
--- a/usr/src/uts/common/io/cxgbe/t4nex/t4_sge.c
+++ b/usr/src/uts/common/io/cxgbe/t4nex/t4_sge.c
@@ -693,7 +693,6 @@ t4_intr_rx_work(struct sge_iq *iq)
 uint_t
 t4_intr(caddr_t arg1, caddr_t arg2)
 {
-	/* LINTED: E_BAD_PTR_CAST_ALIGN */
 	struct sge_iq *iq = (struct sge_iq *)arg2;
 	int state;
 
@@ -1687,7 +1686,8 @@ eth_eq_alloc(struct adapter *sc, struct port_info *pi, struct sge_eq *eq)
 	    V_FW_EQ_ETH_CMD_VFN(0));
 	c.alloc_to_len16 = BE_32(F_FW_EQ_ETH_CMD_ALLOC |
 	    F_FW_EQ_ETH_CMD_EQSTART | FW_LEN16(c));
-	c.autoequiqe_to_viid = BE_32(V_FW_EQ_ETH_CMD_VIID(pi->viid));
+	c.autoequiqe_to_viid = BE_32(F_FW_EQ_ETH_CMD_AUTOEQUIQE |
+	    F_FW_EQ_ETH_CMD_AUTOEQUEQE | V_FW_EQ_ETH_CMD_VIID(pi->viid));
 	c.fetchszm_to_iqid =
 	    BE_32(V_FW_EQ_ETH_CMD_HOSTFCMODE(X_HOSTFCMODE_STATUS_PAGE) |
 	    V_FW_EQ_ETH_CMD_PCIECHN(eq->tx_chan) | F_FW_EQ_ETH_CMD_FETCHRO |
@@ -3346,6 +3346,16 @@ ring_fl_db(struct adapter *sc, struct sge_fl *fl)
 	fl->pending -= ndesc * 8;
 }
 
+static void
+tx_reclaim_task(void *arg)
+{
+	struct sge_txq *txq = arg;
+
+	TXQ_LOCK(txq);
+	reclaim_tx_descs(txq, txq->eq.qsize);
+	TXQ_UNLOCK(txq);
+}
+
 /* ARGSUSED */
 static int
 handle_sge_egr_update(struct sge_iq *iq, const struct rss_header *rss,
@@ -3355,12 +3365,17 @@ handle_sge_egr_update(struct sge_iq *iq, const struct rss_header *rss,
 	unsigned int qid = G_EGR_QID(ntohl(cpl->opcode_qid));
 	struct adapter *sc = iq->adapter;
 	struct sge *s = &sc->sge;
+	struct sge_eq *eq;
 	struct sge_txq *txq;
 
 	txq = (void *)s->eqmap[qid - s->eq_start];
+	eq = &txq->eq;
 	txq->qflush++;
 	t4_mac_tx_update(txq->port, txq);
 
+	ddi_taskq_dispatch(sc->tq[eq->tx_chan], tx_reclaim_task,
+		(void *)txq, DDI_NOSLEEP);
+
 	return (0);
 }