Merge branch 'bcmgenet-Turn-on-offloads-by-default'

Doug Berger says:

====================
net: bcmgenet: Turn on offloads by default

This commit stack is based on Florian's commit 4e8aedfe78c7 ("net:
systemport: Turn on offloads by default") and enables the offloads for
the bcmgenet driver by default.

The first commit adds support for the HIGHDMA feature to the driver.

The second converts the Tx checksum implementation to use the generic
hardware logic rather than the deprecated IP centric methods.

The third modifies the Rx checksum implementation to use the hardware
offload to compute the complete checksum rather than filtering out bad
packets detected by the hardware's IP centric implementation. This may
increase processing load by passing bad packets to the network stack,
but it provides for more flexible handling of packets by the network
stack without requiring software computation of the checksum.

The remaining commits mirror the extensions Florian made to the sysport
driver to retain symmetry with that driver and to make the benefits of
the hardware offloads more ubiquitous.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 67 insertions, 50 deletions

diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.c b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
index 32f1245a69e2..3ee7917e3fc0 100644
--- a/drivers/net/ethernet/broadcom/genet/bcmgenet.c
+++ b/drivers/net/ethernet/broadcom/genet/bcmgenet.c
@@ -2,7 +2,7 @@
 /*
  * Broadcom GENET (Gigabit Ethernet) controller driver
  *
- * Copyright (c) 2014-2017 Broadcom
+ * Copyright (c) 2014-2019 Broadcom
  */
 
 #define pr_fmt(fmt)				"bcmgenet: " fmt
@@ -508,8 +508,8 @@ static int bcmgenet_set_link_ksettings(struct net_device *dev,
 	return phy_ethtool_ksettings_set(dev->phydev, cmd);
 }
 
-static int bcmgenet_set_rx_csum(struct net_device *dev,
-				netdev_features_t wanted)
+static void bcmgenet_set_rx_csum(struct net_device *dev,
+				 netdev_features_t wanted)
 {
 	struct bcmgenet_priv *priv = netdev_priv(dev);
 	u32 rbuf_chk_ctrl;
@@ -521,7 +521,7 @@ static int bcmgenet_set_rx_csum(struct net_device *dev,
 
 	/* enable rx checksumming */
 	if (rx_csum_en)
-		rbuf_chk_ctrl |= RBUF_RXCHK_EN;
+		rbuf_chk_ctrl |= RBUF_RXCHK_EN | RBUF_L3_PARSE_DIS;
 	else
 		rbuf_chk_ctrl &= ~RBUF_RXCHK_EN;
 	priv->desc_rxchk_en = rx_csum_en;
@@ -535,12 +535,10 @@ static int bcmgenet_set_rx_csum(struct net_device *dev,
 		rbuf_chk_ctrl &= ~RBUF_SKIP_FCS;
 
 	bcmgenet_rbuf_writel(priv, rbuf_chk_ctrl, RBUF_CHK_CTRL);
-
-	return 0;
 }
 
-static int bcmgenet_set_tx_csum(struct net_device *dev,
-				netdev_features_t wanted)
+static void bcmgenet_set_tx_csum(struct net_device *dev,
+				 netdev_features_t wanted)
 {
 	struct bcmgenet_priv *priv = netdev_priv(dev);
 	bool desc_64b_en;
@@ -549,7 +547,7 @@ static int bcmgenet_set_tx_csum(struct net_device *dev,
 	tbuf_ctrl = bcmgenet_tbuf_ctrl_get(priv);
 	rbuf_ctrl = bcmgenet_rbuf_readl(priv, RBUF_CTRL);
 
-	desc_64b_en = !!(wanted & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM));
+	desc_64b_en = !!(wanted & NETIF_F_HW_CSUM);
 
 	/* enable 64 bytes descriptor in both directions (RBUF and TBUF) */
 	if (desc_64b_en) {
@@ -563,21 +561,27 @@ static int bcmgenet_set_tx_csum(struct net_device *dev,
 
 	bcmgenet_tbuf_ctrl_set(priv, tbuf_ctrl);
 	bcmgenet_rbuf_writel(priv, rbuf_ctrl, RBUF_CTRL);
-
-	return 0;
 }
 
 static int bcmgenet_set_features(struct net_device *dev,
 				 netdev_features_t features)
 {
-	netdev_features_t changed = features ^ dev->features;
-	netdev_features_t wanted = dev->wanted_features;
-	int ret = 0;
+	struct bcmgenet_priv *priv = netdev_priv(dev);
+	u32 reg;
+	int ret;
 
-	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
-		ret = bcmgenet_set_tx_csum(dev, wanted);
-	if (changed & (NETIF_F_RXCSUM))
-		ret = bcmgenet_set_rx_csum(dev, wanted);
+	ret = clk_prepare_enable(priv->clk);
+	if (ret)
+		return ret;
+
+	/* Make sure we reflect the value of CRC_CMD_FWD */
+	reg = bcmgenet_umac_readl(priv, UMAC_CMD);
+	priv->crc_fwd_en = !!(reg & CMD_CRC_FWD);
+
+	bcmgenet_set_tx_csum(dev, features);
+	bcmgenet_set_rx_csum(dev, features);
+
+	clk_disable_unprepare(priv->clk);
 
 	return ret;
 }
@@ -857,6 +861,9 @@ static const struct bcmgenet_stats bcmgenet_gstrings_stats[] = {
 	STAT_GENET_SOFT_MIB("alloc_rx_buff_failed", mib.alloc_rx_buff_failed),
 	STAT_GENET_SOFT_MIB("rx_dma_failed", mib.rx_dma_failed),
 	STAT_GENET_SOFT_MIB("tx_dma_failed", mib.tx_dma_failed),
+	STAT_GENET_SOFT_MIB("tx_realloc_tsb", mib.tx_realloc_tsb),
+	STAT_GENET_SOFT_MIB("tx_realloc_tsb_failed",
+			    mib.tx_realloc_tsb_failed),
 	/* Per TX queues */
 	STAT_GENET_Q(0),
 	STAT_GENET_Q(1),
@@ -1483,6 +1490,7 @@ static void bcmgenet_tx_reclaim_all(struct net_device *dev)
 static struct sk_buff *bcmgenet_put_tx_csum(struct net_device *dev,
 					    struct sk_buff *skb)
 {
+	struct bcmgenet_priv *priv = netdev_priv(dev);
 	struct status_64 *status = NULL;
 	struct sk_buff *new_skb;
 	u16 offset;
@@ -1495,12 +1503,15 @@ static struct sk_buff *bcmgenet_put_tx_csum(struct net_device *dev,
 		 * enough headroom for us to insert 64B status block.
 		 */
 		new_skb = skb_realloc_headroom(skb, sizeof(*status));
-		dev_kfree_skb(skb);
 		if (!new_skb) {
+			dev_kfree_skb_any(skb);
+			priv->mib.tx_realloc_tsb_failed++;
 			dev->stats.tx_dropped++;
 			return NULL;
 		}
+		dev_consume_skb_any(skb);
 		skb = new_skb;
+		priv->mib.tx_realloc_tsb++;
 	}
 
 	skb_push(skb, sizeof(*status));
@@ -1516,24 +1527,19 @@ static struct sk_buff *bcmgenet_put_tx_csum(struct net_device *dev,
 			ip_proto = ipv6_hdr(skb)->nexthdr;
 			break;
 		default:
-			return skb;
+			/* don't use UDP flag */
+			ip_proto = 0;
+			break;
 		}
 
 		offset = skb_checksum_start_offset(skb) - sizeof(*status);
 		tx_csum_info = (offset << STATUS_TX_CSUM_START_SHIFT) |
-				(offset + skb->csum_offset);
+				(offset + skb->csum_offset) |
+				STATUS_TX_CSUM_LV;
 
-		/* Set the length valid bit for TCP and UDP and just set
-		 * the special UDP flag for IPv4, else just set to 0.
-		 */
-		if (ip_proto == IPPROTO_TCP || ip_proto == IPPROTO_UDP) {
-			tx_csum_info |= STATUS_TX_CSUM_LV;
-			if (ip_proto == IPPROTO_UDP &&
-			    ip_ver == htons(ETH_P_IP))
-				tx_csum_info |= STATUS_TX_CSUM_PROTO_UDP;
-		} else {
-			tx_csum_info = 0;
-		}
+		/* Set the special UDP flag for UDP */
+		if (ip_proto == IPPROTO_UDP)
+			tx_csum_info |= STATUS_TX_CSUM_PROTO_UDP;
 
 		status->tx_csum_info = tx_csum_info;
 	}
@@ -1744,7 +1750,6 @@ static unsigned int bcmgenet_desc_rx(struct bcmgenet_rx_ring *ring,
 	unsigned int bytes_processed = 0;
 	unsigned int p_index, mask;
 	unsigned int discards;
-	unsigned int chksum_ok = 0;
 
 	/* Clear status before servicing to reduce spurious interrupts */
 	if (ring->index == DESC_INDEX) {
@@ -1795,9 +1800,15 @@ static unsigned int bcmgenet_desc_rx(struct bcmgenet_rx_ring *ring,
 				dmadesc_get_length_status(priv, cb->bd_addr);
 		} else {
 			struct status_64 *status;
+			__be16 rx_csum;
 
 			status = (struct status_64 *)skb->data;
 			dma_length_status = status->length_status;
+			rx_csum = (__force __be16)(status->rx_csum & 0xffff);
+			if (priv->desc_rxchk_en) {
+				skb->csum = (__force __wsum)ntohs(rx_csum);
+				skb->ip_summed = CHECKSUM_COMPLETE;
+			}
 		}
 
 		/* DMA flags and length are still valid no matter how
@@ -1840,18 +1851,12 @@ static unsigned int bcmgenet_desc_rx(struct bcmgenet_rx_ring *ring,
 			goto next;
 		} /* error packet */
 
-		chksum_ok = (dma_flag & priv->dma_rx_chk_bit) &&
-			     priv->desc_rxchk_en;
-
 		skb_put(skb, len);
 		if (priv->desc_64b_en) {
 			skb_pull(skb, 64);
 			len -= 64;
 		}
 
-		if (likely(chksum_ok))
-			skb->ip_summed = CHECKSUM_UNNECESSARY;
-
 		/* remove hardware 2bytes added for IP alignment */
 		skb_pull(skb, 2);
 		len -= 2;
@@ -2886,9 +2891,10 @@ static int bcmgenet_open(struct net_device *dev)
 
 	init_umac(priv);
 
-	/* Make sure we reflect the value of CRC_CMD_FWD */
-	reg = bcmgenet_umac_readl(priv, UMAC_CMD);
-	priv->crc_fwd_en = !!(reg & CMD_CRC_FWD);
+	/* Apply features again in case we changed them while interface was
+	 * down
+	 */
+	bcmgenet_set_features(dev, dev->features);
 
 	bcmgenet_set_hw_addr(priv, dev->dev_addr);
 
@@ -3327,19 +3333,15 @@ static void bcmgenet_set_hw_params(struct bcmgenet_priv *priv)
 	if (GENET_IS_V5(priv) || GENET_IS_V4(priv)) {
 		bcmgenet_dma_regs = bcmgenet_dma_regs_v3plus;
 		genet_dma_ring_regs = genet_dma_ring_regs_v4;
-		priv->dma_rx_chk_bit = DMA_RX_CHK_V3PLUS;
 	} else if (GENET_IS_V3(priv)) {
 		bcmgenet_dma_regs = bcmgenet_dma_regs_v3plus;
 		genet_dma_ring_regs = genet_dma_ring_regs_v123;
-		priv->dma_rx_chk_bit = DMA_RX_CHK_V3PLUS;
 	} else if (GENET_IS_V2(priv)) {
 		bcmgenet_dma_regs = bcmgenet_dma_regs_v2;
 		genet_dma_ring_regs = genet_dma_ring_regs_v123;
-		priv->dma_rx_chk_bit = DMA_RX_CHK_V12;
 	} else if (GENET_IS_V1(priv)) {
 		bcmgenet_dma_regs = bcmgenet_dma_regs_v1;
 		genet_dma_ring_regs = genet_dma_ring_regs_v123;
-		priv->dma_rx_chk_bit = DMA_RX_CHK_V12;
 	}
 
 	/* enum genet_version starts at 1 */
@@ -3535,9 +3537,11 @@ static int bcmgenet_probe(struct platform_device *pdev)
 
 	priv->msg_enable = netif_msg_init(-1, GENET_MSG_DEFAULT);
 
-	/* Set hardware features */
-	dev->hw_features |= NETIF_F_SG | NETIF_F_IP_CSUM |
-		NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
+	/* Set default features */
+	dev->features |= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_HW_CSUM |
+			 NETIF_F_RXCSUM;
+	dev->hw_features |= dev->features;
+	dev->vlan_features |= dev->features;
 
 	/* Request the WOL interrupt and advertise suspend if available */
 	priv->wol_irq_disabled = true;
@@ -3574,6 +3578,14 @@ static int bcmgenet_probe(struct platform_device *pdev)
 
 	bcmgenet_set_hw_params(priv);
 
+	err = -EIO;
+	if (priv->hw_params->flags & GENET_HAS_40BITS)
+		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
+	if (err)
+		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
+	if (err)
+		goto err;
+
 	/* Mii wait queue */
 	init_waitqueue_head(&priv->wq);
 	/* Always use RX_BUF_LENGTH (2KB) buffer for all chips */
@@ -3689,6 +3701,9 @@ static int bcmgenet_resume(struct device *d)
 	genphy_config_aneg(dev->phydev);
 	bcmgenet_mii_config(priv->dev, false);
 
+	/* Restore enabled features */
+	bcmgenet_set_features(dev, dev->features);
+
 	bcmgenet_set_hw_addr(priv, dev->dev_addr);
 
 	if (priv->internal_phy) {
diff --git a/drivers/net/ethernet/broadcom/genet/bcmgenet.h b/drivers/net/ethernet/broadcom/genet/bcmgenet.h
index a5659197598f..61a6fe9f4cec 100644
--- a/drivers/net/ethernet/broadcom/genet/bcmgenet.h
+++ b/drivers/net/ethernet/broadcom/genet/bcmgenet.h
@@ -144,6 +144,8 @@ struct bcmgenet_mib_counters {
 	u32	alloc_rx_buff_failed;
 	u32	rx_dma_failed;
 	u32	tx_dma_failed;
+	u32	tx_realloc_tsb;
+	u32	tx_realloc_tsb_failed;
 };
 
 #define UMAC_HD_BKP_CTRL		0x004
@@ -251,6 +253,7 @@ struct bcmgenet_mib_counters {
 #define RBUF_CHK_CTRL			0x14
 #define  RBUF_RXCHK_EN			(1 << 0)
 #define  RBUF_SKIP_FCS			(1 << 4)
+#define  RBUF_L3_PARSE_DIS		(1 << 5)
 
 #define RBUF_ENERGY_CTRL		0x9c
 #define  RBUF_EEE_EN			(1 << 0)
@@ -663,7 +666,6 @@ struct bcmgenet_priv {
 	bool desc_rxchk_en;
 	bool crc_fwd_en;
 
-	unsigned int dma_rx_chk_bit;
 	u32 dma_max_burst_length;
 
 	u32 msg_enable;