- Make note of interface MTU change if the rx queues exist, and not just

  when the interface is up.
- Add a tunable to control the TOE's rx coalesce feature (enabled by
  default as it always has been).  Consider the interface MTU or the
  coalesce size when deciding which cluster zone to use to fill the
  offload rx queue's free list.  The tunable is:
  dev.{t4nex,t5nex}.<N>.toe.rx_coalesce

MFC after:	1 day

6 files changed, 44 insertions, 8 deletions

diff --git a/sys/dev/cxgbe/adapter.h b/sys/dev/cxgbe/adapter.h
index df0976a..38af49f 100644
--- a/sys/dev/cxgbe/adapter.h
+++ b/sys/dev/cxgbe/adapter.h
@@ -132,7 +132,6 @@ enum {
 #else
 	FL_BUF_SIZES = 3,	/* cluster, jumbo9k, jumbo16k */
 #endif
-	OFLD_BUF_SIZE = MJUM16BYTES,	/* size of fl buffer for TOE rxq */
 
 	CTRL_EQ_QSIZE = 128,
 
diff --git a/sys/dev/cxgbe/offload.h b/sys/dev/cxgbe/offload.h
index 6090775..7f4b38e 100644
--- a/sys/dev/cxgbe/offload.h
+++ b/sys/dev/cxgbe/offload.h
@@ -140,6 +140,7 @@ struct tom_tunables {
 	int ddp;
 	int indsz;
 	int ddp_thres;
+	int rx_coalesce;
 };
 
 int t4_register_uld(struct uld_info *);
diff --git a/sys/dev/cxgbe/t4_main.c b/sys/dev/cxgbe/t4_main.c
index 6f96390..fde79c4 100644
--- a/sys/dev/cxgbe/t4_main.c
+++ b/sys/dev/cxgbe/t4_main.c
@@ -1072,9 +1072,10 @@ cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
 		if (rc)
 			return (rc);
 		ifp->if_mtu = mtu;
-		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
+		if (pi->flags & PORT_INIT_DONE) {
 			t4_update_fl_bufsize(ifp);
-			rc = update_mac_settings(pi, XGMAC_MTU);
+			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
+				rc = update_mac_settings(pi, XGMAC_MTU);
 		}
 		end_synchronized_op(sc, 0);
 		break;
@@ -4381,6 +4382,10 @@ t4_sysctls(struct adapter *sc)
 		    G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2));
 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp_thres", CTLFLAG_RW,
 		    &sc->tt.ddp_thres, 0, "DDP threshold");
+
+		sc->tt.rx_coalesce = 1;
+		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
+		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
 	}
 #endif
 
diff --git a/sys/dev/cxgbe/t4_sge.c b/sys/dev/cxgbe/t4_sge.c
index 89eee2d..586137a 100644
--- a/sys/dev/cxgbe/t4_sge.c
+++ b/sys/dev/cxgbe/t4_sge.c
@@ -659,6 +659,18 @@ mtu_to_bufsize(int mtu)
 	return (bufsize);
 }
 
+#ifdef TCP_OFFLOAD
+static inline int
+mtu_to_bufsize_toe(struct adapter *sc, int mtu)
+{
+
+	if (sc->tt.rx_coalesce)
+		return (G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2)));
+
+	return (mtu);
+}
+#endif
+
 int
 t4_setup_port_queues(struct port_info *pi)
 {
@@ -673,9 +685,10 @@ t4_setup_port_queues(struct port_info *pi)
 #endif
 	char name[16];
 	struct adapter *sc = pi->adapter;
+	struct ifnet *ifp = pi->ifp;
 	struct sysctl_oid *oid = device_get_sysctl_tree(pi->dev);
 	struct sysctl_oid_list *children = SYSCTL_CHILDREN(oid);
-	int bufsize = mtu_to_bufsize(pi->ifp->if_mtu);
+	int bufsize;
 
 	oid = SYSCTL_ADD_NODE(&pi->ctx, children, OID_AUTO, "rxq", CTLFLAG_RD,
 	    NULL, "rx queues");
@@ -696,6 +709,7 @@ t4_setup_port_queues(struct port_info *pi)
 	 * a) initialize iq and fl
 	 * b) allocate queue iff it will take direct interrupts.
 	 */
+	bufsize = mtu_to_bufsize(ifp->if_mtu);
 	for_each_rxq(pi, i, rxq) {
 
 		init_iq(&rxq->iq, sc, pi->tmr_idx, pi->pktc_idx, pi->qsize_rxq,
@@ -719,6 +733,7 @@ t4_setup_port_queues(struct port_info *pi)
 	}
 
 #ifdef TCP_OFFLOAD
+	bufsize = mtu_to_bufsize_toe(sc, ifp->if_mtu);
 	for_each_ofld_rxq(pi, i, ofld_rxq) {
 
 		init_iq(&ofld_rxq->iq, sc, pi->tmr_idx, pi->pktc_idx,
@@ -726,7 +741,7 @@ t4_setup_port_queues(struct port_info *pi)
 
 		snprintf(name, sizeof(name), "%s ofld_rxq%d-fl",
 		    device_get_nameunit(pi->dev), i);
-		init_fl(&ofld_rxq->fl, pi->qsize_rxq / 8, OFLD_BUF_SIZE, name);
+		init_fl(&ofld_rxq->fl, pi->qsize_rxq / 8, bufsize, name);
 
 		if (sc->flags & INTR_DIRECT ||
 		    (sc->intr_count > 1 && pi->nofldrxq > pi->nrxq)) {
@@ -1557,9 +1572,13 @@ t4_update_fl_bufsize(struct ifnet *ifp)
 {
 	struct port_info *pi = ifp->if_softc;
 	struct sge_rxq *rxq;
+#ifdef TCP_OFFLOAD
+	struct sge_ofld_rxq *ofld_rxq;
+#endif
 	struct sge_fl *fl;
-	int i, bufsize = mtu_to_bufsize(ifp->if_mtu);
+	int i, bufsize;
 
+	bufsize = mtu_to_bufsize(ifp->if_mtu);
 	for_each_rxq(pi, i, rxq) {
 		fl = &rxq->fl;
 
@@ -1567,6 +1586,16 @@ t4_update_fl_bufsize(struct ifnet *ifp)
 		set_fl_tag_idx(fl, bufsize);
 		FL_UNLOCK(fl);
 	}
+#ifdef TCP_OFFLOAD
+	bufsize = mtu_to_bufsize_toe(pi->adapter, ifp->if_mtu);
+	for_each_ofld_rxq(pi, i, ofld_rxq) {
+		fl = &ofld_rxq->fl;
+
+		FL_LOCK(fl);
+		set_fl_tag_idx(fl, bufsize);
+		FL_UNLOCK(fl);
+	}
+#endif
 }
 
 int
diff --git a/sys/dev/cxgbe/tom/t4_connect.c b/sys/dev/cxgbe/tom/t4_connect.c
index 11cb4fc..08267cd 100644
--- a/sys/dev/cxgbe/tom/t4_connect.c
+++ b/sys/dev/cxgbe/tom/t4_connect.c
@@ -251,7 +251,8 @@ calc_opt2a(struct socket *so, struct toepcb *toep)
 		opt2 |= F_T5_OPT_2_VALID;
 		opt2 |= F_CONG_CNTRL_VALID; /* OPT_2_ISS really, for T5 */
 	}
-	opt2 |= V_RX_COALESCE(M_RX_COALESCE);
+	if (sc->tt.rx_coalesce)
+		opt2 |= V_RX_COALESCE(M_RX_COALESCE);
 
 #ifdef USE_DDP_RX_FLOW_CONTROL
 	if (toep->ulp_mode == ULP_MODE_TCPDDP)
diff --git a/sys/dev/cxgbe/tom/t4_listen.c b/sys/dev/cxgbe/tom/t4_listen.c
index 7259a29..9e1dc80 100644
--- a/sys/dev/cxgbe/tom/t4_listen.c
+++ b/sys/dev/cxgbe/tom/t4_listen.c
@@ -1021,7 +1021,8 @@ calc_opt2p(struct adapter *sc, struct port_info *pi, int rxqid,
 		opt2 |= F_T5_OPT_2_VALID;
 		opt2 |= F_CONG_CNTRL_VALID; /* OPT_2_ISS really, for T5 */
 	}
-	opt2 |= V_RX_COALESCE(M_RX_COALESCE);
+	if (sc->tt.rx_coalesce)
+		opt2 |= V_RX_COALESCE(M_RX_COALESCE);
 
 #ifdef USE_DDP_RX_FLOW_CONTROL
 	if (ulp_mode == ULP_MODE_TCPDDP)