1 files changed, 80 insertions, 18 deletions

diff --git a/drivers/net/ethernet/broadcom/tg3.c b/drivers/net/ethernet/broadcom/tg3.c
index 5665d0c3668f..48b6191efa56 100644
--- a/drivers/net/ethernet/broadcom/tg3.c
+++ b/drivers/net/ethernet/broadcom/tg3.c
@@ -6647,9 +6647,9 @@ static void tg3_tx(struct tg3_napi *tnapi)
 
 	tnapi->tx_cons = sw_idx;
 
-	/* Need to make the tx_cons update visible to tg3_start_xmit()
+	/* Need to make the tx_cons update visible to __tg3_start_xmit()
 	 * before checking for netif_queue_stopped().  Without the
-	 * memory barrier, there is a small possibility that tg3_start_xmit()
+	 * memory barrier, there is a small possibility that __tg3_start_xmit()
 	 * will miss it and cause the queue to be stopped forever.
 	 */
 	smp_mb();
@@ -6889,7 +6889,7 @@ static int tg3_rx(struct tg3_napi *tnapi, int budget)
 				       desc_idx, *post_ptr);
 		drop_it_no_recycle:
 			/* Other statistics kept track of by card. */
-			tp->rx_dropped++;
+			tnapi->rx_dropped++;
 			goto next_pkt;
 		}
 
@@ -7889,7 +7889,7 @@ static bool tg3_tso_bug_gso_check(struct tg3_napi *tnapi, struct sk_buff *skb)
 	return skb_shinfo(skb)->gso_segs < tnapi->tx_pending / 3;
 }
 
-static netdev_tx_t tg3_start_xmit(struct sk_buff *, struct net_device *);
+static netdev_tx_t __tg3_start_xmit(struct sk_buff *, struct net_device *);
 
 /* Use GSO to workaround all TSO packets that meet HW bug conditions
  * indicated in tg3_tx_frag_set()
@@ -7918,12 +7918,14 @@ static int tg3_tso_bug(struct tg3 *tp, struct tg3_napi *tnapi,
 
 	segs = skb_gso_segment(skb, tp->dev->features &
 				    ~(NETIF_F_TSO | NETIF_F_TSO6));
-	if (IS_ERR(segs) || !segs)
+	if (IS_ERR(segs) || !segs) {
+		tnapi->tx_dropped++;
 		goto tg3_tso_bug_end;
+	}
 
 	skb_list_walk_safe(segs, seg, next) {
 		skb_mark_not_on_list(seg);
-		tg3_start_xmit(seg, tp->dev);
+		__tg3_start_xmit(seg, tp->dev);
 	}
 
 tg3_tso_bug_end:
@@ -7933,7 +7935,7 @@ tg3_tso_bug_end:
 }
 
 /* hard_start_xmit for all devices */
-static netdev_tx_t tg3_start_xmit(struct sk_buff *skb, struct net_device *dev)
+static netdev_tx_t __tg3_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
 	struct tg3 *tp = netdev_priv(dev);
 	u32 len, entry, base_flags, mss, vlan = 0;
@@ -8182,11 +8184,6 @@ static netdev_tx_t tg3_start_xmit(struct sk_buff *skb, struct net_device *dev)
 			netif_tx_wake_queue(txq);
 	}
 
-	if (!netdev_xmit_more() || netif_xmit_stopped(txq)) {
-		/* Packets are ready, update Tx producer idx on card. */
-		tw32_tx_mbox(tnapi->prodmbox, entry);
-	}
-
 	return NETDEV_TX_OK;
 
 dma_error:
@@ -8195,10 +8192,46 @@ dma_error:
 drop:
 	dev_kfree_skb_any(skb);
 drop_nofree:
-	tp->tx_dropped++;
+	tnapi->tx_dropped++;
 	return NETDEV_TX_OK;
 }
 
+static netdev_tx_t tg3_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+	struct netdev_queue *txq;
+	u16 skb_queue_mapping;
+	netdev_tx_t ret;
+
+	skb_queue_mapping = skb_get_queue_mapping(skb);
+	txq = netdev_get_tx_queue(dev, skb_queue_mapping);
+
+	ret = __tg3_start_xmit(skb, dev);
+
+	/* Notify the hardware that packets are ready by updating the TX ring
+	 * tail pointer. We respect netdev_xmit_more() thus avoiding poking
+	 * the hardware for every packet. To guarantee forward progress the TX
+	 * ring must be drained when it is full as indicated by
+	 * netif_xmit_stopped(). This needs to happen even when the current
+	 * skb was dropped or rejected with NETDEV_TX_BUSY. Otherwise packets
+	 * queued by previous __tg3_start_xmit() calls might get stuck in
+	 * the queue forever.
+	 */
+	if (!netdev_xmit_more() || netif_xmit_stopped(txq)) {
+		struct tg3_napi *tnapi;
+		struct tg3 *tp;
+
+		tp = netdev_priv(dev);
+		tnapi = &tp->napi[skb_queue_mapping];
+
+		if (tg3_flag(tp, ENABLE_TSS))
+			tnapi++;
+
+		tw32_tx_mbox(tnapi->prodmbox, tnapi->tx_prod);
+	}
+
+	return ret;
+}
+
 static void tg3_mac_loopback(struct tg3 *tp, bool enable)
 {
 	if (enable) {
@@ -9374,7 +9407,7 @@ static void __tg3_set_rx_mode(struct net_device *);
 /* tp->lock is held. */
 static int tg3_halt(struct tg3 *tp, int kind, bool silent)
 {
-	int err;
+	int err, i;
 
 	tg3_stop_fw(tp);
 
@@ -9395,6 +9428,13 @@ static int tg3_halt(struct tg3 *tp, int kind, bool silent)
 
 		/* And make sure the next sample is new data */
 		memset(tp->hw_stats, 0, sizeof(struct tg3_hw_stats));
+
+		for (i = 0; i < TG3_IRQ_MAX_VECS; ++i) {
+			struct tg3_napi *tnapi = &tp->napi[i];
+
+			tnapi->rx_dropped = 0;
+			tnapi->tx_dropped = 0;
+		}
 	}
 
 	return err;
@@ -11944,6 +11984,9 @@ static void tg3_get_nstats(struct tg3 *tp, struct rtnl_link_stats64 *stats)
 {
 	struct rtnl_link_stats64 *old_stats = &tp->net_stats_prev;
 	struct tg3_hw_stats *hw_stats = tp->hw_stats;
+	unsigned long rx_dropped;
+	unsigned long tx_dropped;
+	int i;
 
 	stats->rx_packets = old_stats->rx_packets +
 		get_stat64(&hw_stats->rx_ucast_packets) +
@@ -11990,8 +12033,26 @@ static void tg3_get_nstats(struct tg3 *tp, struct rtnl_link_stats64 *stats)
 	stats->rx_missed_errors = old_stats->rx_missed_errors +
 		get_stat64(&hw_stats->rx_discards);
 
-	stats->rx_dropped = tp->rx_dropped;
-	stats->tx_dropped = tp->tx_dropped;
+	/* Aggregate per-queue counters. The per-queue counters are updated
+	 * by a single writer, race-free. The result computed by this loop
+	 * might not be 100% accurate (counters can be updated in the middle of
+	 * the loop) but the next tg3_get_nstats() will recompute the current
+	 * value so it is acceptable.
+	 *
+	 * Note that these counters wrap around at 4G on 32bit machines.
+	 */
+	rx_dropped = (unsigned long)(old_stats->rx_dropped);
+	tx_dropped = (unsigned long)(old_stats->tx_dropped);
+
+	for (i = 0; i < tp->irq_cnt; i++) {
+		struct tg3_napi *tnapi = &tp->napi[i];
+
+		rx_dropped += tnapi->rx_dropped;
+		tx_dropped += tnapi->tx_dropped;
+	}
+
+	stats->rx_dropped = rx_dropped;
+	stats->tx_dropped = tx_dropped;
 }
 
 static int tg3_get_regs_len(struct net_device *dev)
@@ -17729,7 +17790,7 @@ static int tg3_init_one(struct pci_dev *pdev,
 	 * device behind the EPB cannot support DMA addresses > 40-bit.
 	 * On 64-bit systems with IOMMU, use 40-bit dma_mask.
 	 * On 64-bit systems without IOMMU, use 64-bit dma_mask and
-	 * do DMA address check in tg3_start_xmit().
+	 * do DMA address check in __tg3_start_xmit().
 	 */
 	if (tg3_flag(tp, IS_5788))
 		persist_dma_mask = dma_mask = DMA_BIT_MASK(32);
@@ -18127,7 +18188,8 @@ static void tg3_shutdown(struct pci_dev *pdev)
 	if (netif_running(dev))
 		dev_close(dev);
 
-	tg3_power_down(tp);
+	if (system_state == SYSTEM_POWER_OFF)
+		tg3_power_down(tp);
 
 	rtnl_unlock();