1 files changed, 108 insertions, 28 deletions

diff --git a/drivers/net/ethernet/broadcom/tg3.c b/drivers/net/ethernet/broadcom/tg3.c
index f1c8ff5b63..f52830dfb2 100644
--- a/drivers/net/ethernet/broadcom/tg3.c
+++ b/drivers/net/ethernet/broadcom/tg3.c
@@ -6314,6 +6314,46 @@ err_out:
 	return -EOPNOTSUPP;
 }
 
+static void tg3_hwclock_to_timestamp(struct tg3 *tp, u64 hwclock,
+				     struct skb_shared_hwtstamps *timestamp)
+{
+	memset(timestamp, 0, sizeof(struct skb_shared_hwtstamps));
+	timestamp->hwtstamp  = ns_to_ktime((hwclock & TG3_TSTAMP_MASK) +
+					   tp->ptp_adjust);
+}
+
+static void tg3_read_tx_tstamp(struct tg3 *tp, u64 *hwclock)
+{
+	*hwclock = tr32(TG3_TX_TSTAMP_LSB);
+	*hwclock |= (u64)tr32(TG3_TX_TSTAMP_MSB) << 32;
+}
+
+static long tg3_ptp_ts_aux_work(struct ptp_clock_info *ptp)
+{
+	struct tg3 *tp = container_of(ptp, struct tg3, ptp_info);
+	struct skb_shared_hwtstamps timestamp;
+	u64 hwclock;
+
+	if (tp->ptp_txts_retrycnt > 2)
+		goto done;
+
+	tg3_read_tx_tstamp(tp, &hwclock);
+
+	if (hwclock != tp->pre_tx_ts) {
+		tg3_hwclock_to_timestamp(tp, hwclock, &timestamp);
+		skb_tstamp_tx(tp->tx_tstamp_skb, &timestamp);
+		goto done;
+	}
+	tp->ptp_txts_retrycnt++;
+	return HZ / 10;
+done:
+	dev_consume_skb_any(tp->tx_tstamp_skb);
+	tp->tx_tstamp_skb = NULL;
+	tp->ptp_txts_retrycnt = 0;
+	tp->pre_tx_ts = 0;
+	return -1;
+}
+
 static const struct ptp_clock_info tg3_ptp_caps = {
 	.owner		= THIS_MODULE,
 	.name		= "tg3 clock",
@@ -6325,19 +6365,12 @@ static const struct ptp_clock_info tg3_ptp_caps = {
 	.pps		= 0,
 	.adjfine	= tg3_ptp_adjfine,
 	.adjtime	= tg3_ptp_adjtime,
+	.do_aux_work	= tg3_ptp_ts_aux_work,
 	.gettimex64	= tg3_ptp_gettimex,
 	.settime64	= tg3_ptp_settime,
 	.enable		= tg3_ptp_enable,
 };
 
-static void tg3_hwclock_to_timestamp(struct tg3 *tp, u64 hwclock,
-				     struct skb_shared_hwtstamps *timestamp)
-{
-	memset(timestamp, 0, sizeof(struct skb_shared_hwtstamps));
-	timestamp->hwtstamp  = ns_to_ktime((hwclock & TG3_TSTAMP_MASK) +
-					   tp->ptp_adjust);
-}
-
 /* tp->lock must be held */
 static void tg3_ptp_init(struct tg3 *tp)
 {
@@ -6368,6 +6401,8 @@ static void tg3_ptp_fini(struct tg3 *tp)
 	ptp_clock_unregister(tp->ptp_clock);
 	tp->ptp_clock = NULL;
 	tp->ptp_adjust = 0;
+	dev_consume_skb_any(tp->tx_tstamp_skb);
+	tp->tx_tstamp_skb = NULL;
 }
 
 static inline int tg3_irq_sync(struct tg3 *tp)
@@ -6546,6 +6581,7 @@ static void tg3_tx(struct tg3_napi *tnapi)
 
 	while (sw_idx != hw_idx) {
 		struct tg3_tx_ring_info *ri = &tnapi->tx_buffers[sw_idx];
+		bool complete_skb_later = false;
 		struct sk_buff *skb = ri->skb;
 		int i, tx_bug = 0;
 
@@ -6556,12 +6592,17 @@ static void tg3_tx(struct tg3_napi *tnapi)
 
 		if (tnapi->tx_ring[sw_idx].len_flags & TXD_FLAG_HWTSTAMP) {
 			struct skb_shared_hwtstamps timestamp;
-			u64 hwclock = tr32(TG3_TX_TSTAMP_LSB);
-			hwclock |= (u64)tr32(TG3_TX_TSTAMP_MSB) << 32;
+			u64 hwclock;
 
-			tg3_hwclock_to_timestamp(tp, hwclock, &timestamp);
-
-			skb_tstamp_tx(skb, &timestamp);
+			tg3_read_tx_tstamp(tp, &hwclock);
+			if (hwclock != tp->pre_tx_ts) {
+				tg3_hwclock_to_timestamp(tp, hwclock, &timestamp);
+				skb_tstamp_tx(skb, &timestamp);
+				tp->pre_tx_ts = 0;
+			} else {
+				tp->tx_tstamp_skb = skb;
+				complete_skb_later = true;
+			}
 		}
 
 		dma_unmap_single(&tp->pdev->dev, dma_unmap_addr(ri, mapping),
@@ -6599,7 +6640,10 @@ static void tg3_tx(struct tg3_napi *tnapi)
 		pkts_compl++;
 		bytes_compl += skb->len;
 
-		dev_consume_skb_any(skb);
+		if (!complete_skb_later)
+			dev_consume_skb_any(skb);
+		else
+			ptp_schedule_worker(tp->ptp_clock, 0);
 
 		if (unlikely(tx_bug)) {
 			tg3_tx_recover(tp);
@@ -6611,9 +6655,9 @@ 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();
@@ -7853,7 +7897,7 @@ 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()
@@ -7889,7 +7933,7 @@ static int tg3_tso_bug(struct tg3 *tp, struct tg3_napi *tnapi,
 
 	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:
@@ -7899,7 +7943,7 @@ 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;
@@ -8038,8 +8082,13 @@ static netdev_tx_t tg3_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
 	if ((unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) &&
 	    tg3_flag(tp, TX_TSTAMP_EN)) {
-		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
-		base_flags |= TXD_FLAG_HWTSTAMP;
+		tg3_full_lock(tp, 0);
+		if (!tp->pre_tx_ts) {
+			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+			base_flags |= TXD_FLAG_HWTSTAMP;
+			tg3_read_tx_tstamp(tp, &tp->pre_tx_ts);
+		}
+		tg3_full_unlock(tp);
 	}
 
 	len = skb_headlen(skb);
@@ -8143,11 +8192,6 @@ 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:
@@ -8160,6 +8204,42 @@ drop_nofree:
 	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) {
@@ -17044,7 +17124,7 @@ static u32 tg3_calc_dma_bndry(struct tg3 *tp, u32 val)
 	    !tg3_flag(tp, PCI_EXPRESS))
 		goto out;
 
-#if defined(CONFIG_PPC64) || defined(CONFIG_IA64) || defined(CONFIG_PARISC)
+#if defined(CONFIG_PPC64) || defined(CONFIG_PARISC)
 	goal = BOUNDARY_MULTI_CACHELINE;
 #else
 #if defined(CONFIG_SPARC64) || defined(CONFIG_ALPHA)
@@ -17719,7 +17799,7 @@ 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);