nvethernet: Add support for ioctls and HW offloads

1. ARP offload can be configured via private ioctl. Application requesting the arp offload configuration should provide the IP address to be configured for ARP offload via ioctl data struct. Refer to struct ether_ifr_data and struct arp_offload_param for details. 2. Tx/Rx Checksum offload can be configured via ethtool 3. TCP Segmentation offload can be configured via ethtool Bug 2571001 Change-Id: If639ab9049db97f200911af456ddb8cb8433fa12 Signed-off-by: Srinivas Ramachandran <srinivasra@nvidia.com> Reviewed-on: https://git-master.nvidia.com/r/2109676 Reviewed-by: Rakesh Goyal <rgoyal@nvidia.com> Reviewed-by: Narayan Reddy <narayanr@nvidia.com> GVS: Gerrit_Virtual_Submit Reviewed-by: Ashutosh Jha <ajha@nvidia.com> Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com> Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
2025-12-23 01:31:30 +03:00 · 2019-04-19 13:50:17 -07:00
parent 0426fc74e2
commit e05030a4e2
5 changed files with 483 additions and 36 deletions
--- a/drivers/net/ethernet/nvidia/nvethernet/ether_linux.c
+++ b/drivers/net/ethernet/nvidia/nvethernet/ether_linux.c
@@ -996,6 +996,59 @@ static int ether_close(struct net_device *dev)
 	return ret;
 }

+/**
+ *	ether_handle_tso - Helper func to check if TSO is used in given skb.
+ *	@tx_pkt_cx: Pointer to packet context information structure.
+ *	@skb: socket buffer.
+ *
+ *	Algorithm:
+ *	1) Check if driver received a TSO/LSO/GSO packet
+ *	2) If so, store the packet details like MSS(Maximum Segment Size),
+ *	packet header length, packet payload length, tcp/udp header length.
+ *
+ *	Dependencies: None.
+ *
+ *	Protection: None.
+ *
+ *	Return: 0 - Not a TSO packet, 1 - success, -ve value - failure.
+ */
+static int ether_handle_tso(struct osi_tx_pkt_cx *tx_pkt_cx,
+			    struct sk_buff *skb)
+{
+	int ret = 1;
+
+	if (skb_is_gso(skb) == 0) {
+		return 0;
+	}
+
+	if (skb_header_cloned(skb)) {
+		ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
+		if (ret) {
+			return ret;
+		}
+	}
+
+	/* Start filling packet details in Tx_pkt_cx */
+	if (skb_shinfo(skb)->gso_type & (SKB_GSO_UDP)) {
+		tx_pkt_cx->tcp_udp_hdrlen = sizeof(struct udphdr);
+		tx_pkt_cx->mss = skb_shinfo(skb)->gso_size -
+			sizeof(struct udphdr);
+	} else {
+		tx_pkt_cx->tcp_udp_hdrlen = tcp_hdrlen(skb);
+		tx_pkt_cx->mss = skb_shinfo(skb)->gso_size;
+	}
+	tx_pkt_cx->total_hdrlen = skb_transport_offset(skb) +
+			tx_pkt_cx->tcp_udp_hdrlen;
+	tx_pkt_cx->payload_len = (skb->len - tx_pkt_cx->total_hdrlen);
+
+	netdev_dbg(skb->dev, "mss           =%u\n", tx_pkt_cx->mss);
+	netdev_dbg(skb->dev, "payload_len   =%u\n", tx_pkt_cx->payload_len);
+	netdev_dbg(skb->dev, "tcp_udp_hdrlen=%u\n", tx_pkt_cx->tcp_udp_hdrlen);
+	netdev_dbg(skb->dev, "total_hdrlen  =%u\n", tx_pkt_cx->total_hdrlen);
+
+	return ret;
+}
+
 /**
 *	ether_tx_swcx_alloc - Tx ring software context allocation.
 *	@dev: device instance associated with driver.
@@ -1020,16 +1073,39 @@ static int ether_tx_swcx_alloc(struct device *dev,
 	struct osi_tx_pkt_cx *tx_pkt_cx = &tx_ring->tx_pkt_cx;
 	unsigned int cur_tx_idx = tx_ring->cur_tx_idx;
 	struct osi_tx_swcx *tx_swcx = NULL;
-	unsigned int len = 0;
-	int cnt = 0;
+	unsigned int len = 0, offset = 0, size = 0;
+	int cnt = 0, ret = 0, i, num_frags;
+	struct skb_frag_struct *frag;
+	unsigned int page_idx, page_offset;
+	unsigned int max_data_len_per_txd = (unsigned int)
+		ETHER_MAX_DATA_LEN_PER_TXD_BUF; // 4KB

 	memset(tx_pkt_cx, 0, sizeof(*tx_pkt_cx));

+	ret = ether_handle_tso(tx_pkt_cx, skb);
+	if (unlikely(ret < 0)) {
+		dev_err(dev, "Unable to handle TSO packet (%d)\n", ret);
+		/* Caller will take care of consuming skb */
+		return ret;
+	}
+
+	if (ret == 0) {
+		dev_dbg(dev, "Not a TSO packet\n");
+		if (skb->ip_summed == CHECKSUM_PARTIAL) {
+			tx_pkt_cx->flags |= OSI_PKT_CX_CSUM;
+		}
+	} else {
+		tx_pkt_cx->flags |= OSI_PKT_CX_TSO;
+	}
+
 	if (unlikely(skb_vlan_tag_present(skb))) {
 		tx_pkt_cx->vtag_id = skb_vlan_tag_get(skb);
 		tx_pkt_cx->vtag_id |= (skb->priority << VLAN_PRIO_SHIFT);
-		tx_pkt_cx->flags = OSI_PKT_CX_VLAN;
+		tx_pkt_cx->flags |= OSI_PKT_CX_VLAN;
+	}

+	if (((tx_pkt_cx->flags & OSI_PKT_CX_VLAN) == OSI_PKT_CX_VLAN) ||
+	    ((tx_pkt_cx->flags & OSI_PKT_CX_TSO) == OSI_PKT_CX_TSO)) {
 		tx_swcx = tx_ring->tx_swcx + cur_tx_idx;
 		if (tx_swcx->len) {
 			return 0;
@@ -1040,28 +1116,142 @@ static int ether_tx_swcx_alloc(struct device *dev,
 		INCR_TX_DESC_INDEX(cur_tx_idx, 1U);
 	}

-	len = skb_headlen(skb);
-
-	tx_swcx = tx_ring->tx_swcx + cur_tx_idx;
-	if (tx_swcx->len) {
-		return 0;
+	if ((tx_pkt_cx->flags & OSI_PKT_CX_TSO) == OSI_PKT_CX_TSO) {
+		/* For TSO map the header in separate desc. */
+		len = tx_pkt_cx->total_hdrlen;
+	} else {
+		len = skb_headlen(skb);
 	}

-	tx_swcx->buf_phy_addr = dma_map_single(dev, skb->data,
-					       len, DMA_TO_DEVICE);
-	if (unlikely(dma_mapping_error(dev, tx_swcx->buf_phy_addr))) {
-		dev_err(dev, "failed to map Tx buffer\n");
-		return -ENOMEM;
+	/* Map the linear buffers from the skb first.
+	 * For TSO only upto TCP header is filled in first desc.
+	 */
+	while (len) {
+		tx_swcx = tx_ring->tx_swcx + cur_tx_idx;
+		if (unlikely(tx_swcx->len)) {
+			goto desc_not_free;
+		}
+
+		size = min(len, max_data_len_per_txd);
+
+		tx_swcx->buf_phy_addr = dma_map_single(dev,
+						skb->data + offset,
+						size,
+						DMA_TO_DEVICE);
+		if (unlikely(dma_mapping_error(dev,
+			     tx_swcx->buf_phy_addr))) {
+			dev_err(dev, "failed to map Tx buffer\n");
+			ret = -ENOMEM;
+			goto dma_map_failed;
+		}
+		tx_swcx->is_paged_buf = 0;
+
+		tx_swcx->len = size;
+		len -= size;
+		offset += size;
+		cnt++;
+		INCR_TX_DESC_INDEX(cur_tx_idx, 1U);
+	}
+
+	/* Map remaining payload from linear buffer
+	 * to subsequent descriptors in case of TSO
+	 */
+	if ((tx_pkt_cx->flags & OSI_PKT_CX_TSO) == OSI_PKT_CX_TSO) {
+		len = skb_headlen(skb) - tx_pkt_cx->total_hdrlen;
+		while (len) {
+			tx_swcx = tx_ring->tx_swcx + cur_tx_idx;
+
+			if (unlikely(tx_swcx->len)) {
+				goto desc_not_free;
+			}
+
+			size = min(len, max_data_len_per_txd);
+			tx_swcx->buf_phy_addr = dma_map_single(dev,
+							skb->data + offset,
+							size,
+							DMA_TO_DEVICE);
+			if (unlikely(dma_mapping_error(dev,
+				     tx_swcx->buf_phy_addr))) {
+				dev_err(dev, "failed to map Tx buffer\n");
+				ret = -ENOMEM;
+				goto dma_map_failed;
+			}
+
+			tx_swcx->is_paged_buf = 0;
+			tx_swcx->len = size;
+			len -= size;
+			offset += size;
+			cnt++;
+			INCR_TX_DESC_INDEX(cur_tx_idx, 1U);
+		}
+	}
+
+	/* Process fragmented skb's */
+	num_frags = skb_shinfo(skb)->nr_frags;
+	for (i = 0; i < num_frags; i++) {
+		offset = 0;
+		frag = &skb_shinfo(skb)->frags[i];
+		len = frag->size;
+		while (len) {
+			tx_swcx = tx_ring->tx_swcx + cur_tx_idx;
+
+			if (unlikely(tx_swcx->len)) {
+				goto desc_not_free;
+			}
+
+			size = min(len, max_data_len_per_txd);
+
+			page_idx = (frag->page_offset + offset) >> PAGE_SHIFT;
+			page_offset = (frag->page_offset + offset) & ~PAGE_MASK;
+			tx_swcx->buf_phy_addr = dma_map_page(dev,
+						(frag->page.p + page_idx),
+						page_offset, size,
+						DMA_TO_DEVICE);
+			if (unlikely(dma_mapping_error(dev,
+				     tx_swcx->buf_phy_addr))) {
+				dev_err(dev, "failed to map Tx buffer\n");
+				ret = -ENOMEM;
+				goto dma_map_failed;
+			}
+			tx_swcx->is_paged_buf = 1;
+
+			tx_swcx->len = size;
+			len -= size;
+			offset += size;
+			cnt++;
+			INCR_TX_DESC_INDEX(cur_tx_idx, 1U);
+		}
 	}

-	tx_swcx->len = len;
 	tx_swcx->buf_virt_addr = skb;
-
-	cnt++;
-
 	tx_pkt_cx->desc_cnt = cnt;

 	return cnt;
+
+desc_not_free:
+	ret = 0;
+
+dma_map_failed:
+	/* Failed to fill current desc. Rollback previous desc's */
+	while (cnt) {
+		DECR_TX_DESC_INDEX(cur_tx_idx, 1U);
+		tx_swcx = tx_ring->tx_swcx + cur_tx_idx;
+		if (tx_swcx->buf_phy_addr) {
+			if (tx_swcx->is_paged_buf) {
+				dma_unmap_page(dev, tx_swcx->buf_phy_addr,
+					       tx_swcx->len, DMA_TO_DEVICE);
+			} else {
+				dma_unmap_single(dev, tx_swcx->buf_phy_addr,
+						 tx_swcx->len, DMA_TO_DEVICE);
+			}
+			tx_swcx->buf_phy_addr = 0;
+		}
+		tx_swcx->len = 0;
+
+		tx_swcx->is_paged_buf = 0;
+		cnt--;
+	}
+	return ret;
 }

 /**
@@ -1129,8 +1319,8 @@ static int ether_start_xmit(struct sk_buff *skb, struct net_device *ndev)
 	count = ether_tx_swcx_alloc(pdata->dev, tx_ring, skb);
 	if (count <= 0) {
 		if (count == 0) {
-			netif_stop_queue(ndev);
-			netdev_err(ndev, "Tx ring is full\n");
+			netif_stop_subqueue(ndev, chan);
+			netdev_err(ndev, "Tx ring[%d] is full\n", chan);
 			return NETDEV_TX_BUSY;
 		}
 		dev_kfree_skb_any(skb);
@@ -1139,6 +1329,11 @@ static int ether_start_xmit(struct sk_buff *skb, struct net_device *ndev)

 	osi_hw_transmit(osi_dma, chan);

+	if (ether_avail_txdesc_cnt(tx_ring) < TX_DESC_THRESHOLD) {
+		netif_stop_subqueue(ndev, chan);
+		netdev_dbg(ndev, "Tx ring[%d] insufficient desc.\n", chan);
+	}
+
 	return NETDEV_TX_OK;
 }

@@ -1162,15 +1357,21 @@ static int ether_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
 {
 	int ret = -EOPNOTSUPP;

-	if (!netif_running(dev))
+	if (!dev || !rq) {
 		return -EINVAL;
+	}
+
+	if (!netif_running(dev)) {
+		return -EINVAL;
+	}

 	switch (cmd) {
 	case SIOCGMIIPHY:
 	case SIOCGMIIREG:
 	case SIOCSMIIREG:
-		if (!dev->phydev)
+		if (!dev->phydev) {
 			return -EINVAL;
+		}

 		/* generic PHY MII ioctl interface */
 		ret = phy_mii_ioctl(dev->phydev, rq, cmd);
@@ -1181,6 +1382,8 @@ static int ether_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
 		break;

 	default:
+		netdev_err(dev, "%s: Unsupported ioctl %d\n",
+			   __func__, cmd);
 		break;
 	}

@@ -1255,6 +1458,55 @@ static int ether_change_mtu(struct net_device *ndev, int new_mtu)
 	return 0;
 }

+/**
+ *	ether_set_features - Change HW features for the given ndev
+ *	@ndev: Network device structure
+ *	@feat: New features to be updated
+ *
+ *	Algorithm:
+ *	1) Check if HW supports feature requested to be changed
+ *	2) If supported, check the current status of the feature and if it
+ *	needs to be toggled, do so.
+ *
+ *	Dependencies: Ethernet interface needs to be up. Stack will enforce
+ *	the check.
+ *
+ *	Protection: None.
+ *
+ *	Return: 0 - success, negative value - failure.
+ */
+static int ether_set_features(struct net_device *ndev, netdev_features_t feat)
+{
+	int ret = 0;
+	struct ether_priv_data *pdata = netdev_priv(ndev);
+	struct osi_core_priv_data *osi_core = pdata->osi_core;
+	netdev_features_t hw_feat_cur_state = pdata->hw_feat_cur_state;
+
+	if (pdata->hw_feat.rx_coe_sel == 0U) {
+		return ret;
+	}
+
+	if ((feat & NETIF_F_RXCSUM) == NETIF_F_RXCSUM) {
+		if (!(hw_feat_cur_state & NETIF_F_RXCSUM)) {
+			ret = osi_config_rxcsum_offload(osi_core,
+							OSI_ENABLE);
+			dev_info(pdata->dev, "Rx Csum offload: Enable: %s\n",
+				 ret ? "Failed" : "Success");
+			pdata->hw_feat_cur_state |= NETIF_F_RXCSUM;
+		}
+	} else {
+		if ((hw_feat_cur_state & NETIF_F_RXCSUM)) {
+			ret = osi_config_rxcsum_offload(osi_core,
+							OSI_DISABLE);
+			dev_info(pdata->dev, "Rx Csum offload: Disable: %s\n",
+				 ret ? "Failed" : "Success");
+			pdata->hw_feat_cur_state &= ~NETIF_F_RXCSUM;
+		}
+	}
+
+	return ret;
+}
+
 static const struct net_device_ops ether_netdev_ops = {
 	.ndo_open = ether_open,
 	.ndo_stop = ether_close,
@@ -1263,6 +1515,7 @@ static const struct net_device_ops ether_netdev_ops = {
 	.ndo_set_mac_address = ether_set_mac_addr,
 	.ndo_change_mtu = ether_change_mtu,
 	.ndo_select_queue = ether_select_queue,
+	.ndo_set_features = ether_set_features,
 };

 /**
@@ -2234,6 +2487,62 @@ static int ether_set_dma_mask(struct ether_priv_data *pdata)
 	return ret;
 }

+/**	ether_set_ndev_features - Set the network device feature flags
+ *	@ndev: Network device instance
+ *	@pdata: OS dependent private data structure.
+ *
+ *	Algorithm:
+ *	1) Check the HW features supported
+ *	2) Enable corresponding feature flag so that network subsystem of OS
+ *		is aware of device capabilities.
+ *	3) Update current enable/disable state of features currently enabled
+ *
+ *	Dependencies: Netdev allocated and HW features are already parsed.
+ *
+ *	Protection: None
+ *
+ *	Return: None.
+ */
+static void ether_set_ndev_features(struct net_device *ndev,
+				    struct ether_priv_data *pdata)
+{
+	netdev_features_t features = 0;
+
+	if (pdata->hw_feat.tso_en) {
+		features |= NETIF_F_TSO;
+		features |= NETIF_F_SG;
+	}
+
+	if (pdata->hw_feat.tx_coe_sel) {
+		features |= NETIF_F_IP_CSUM;
+		features |= NETIF_F_IPV6_CSUM;
+	}
+
+	if (pdata->hw_feat.rx_coe_sel) {
+		features |= NETIF_F_RXCSUM;
+	}
+
+	/* GRO is independent of HW features */
+	features |= NETIF_F_GRO;
+
+	if (pdata->hw_feat.sa_vlan_ins) {
+		features |= NETIF_F_HW_VLAN_CTAG_TX;
+	}
+
+	/* Rx VLAN tag detection enabled by default */
+	features |= NETIF_F_HW_VLAN_CTAG_RX;
+
+	/* Features available in HW */
+	ndev->hw_features = features;
+	/* Features that can be changed by user */
+	ndev->features = features;
+	/* Features that can be inherited by vlan devices */
+	ndev->vlan_features = features;
+
+	/* Set current state of features enabled by default in HW */
+	pdata->hw_feat_cur_state = features;
+}
+
 /**
 *	ether_probe - Ethernet platform driver probe.
 *	@pdev:	platform device associated with platform driver.
@@ -2333,6 +2642,9 @@ static int ether_probe(struct platform_device *pdev)
 		goto err_dma_mask;
 	}

+	/* Set netdev features based on hw features */
+	ether_set_ndev_features(ndev, pdata);
+
 	ret = osi_get_mac_version(osi_core->base, &osi_core->mac_ver);
 	if (ret < 0) {
 		dev_err(&pdev->dev, "failed to get MAC version (%u)\n",
@@ -2355,13 +2667,6 @@ static int ether_probe(struct platform_device *pdev)
 	ndev->netdev_ops = &ether_netdev_ops;
 	ether_set_ethtool_ops(ndev);

-	ndev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
-	if (pdata->hw_feat.sa_vlan_ins) {
-		ndev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
-	}
-
-	ndev->features |= ndev->hw_features;
-
 	ret = ether_alloc_napi(pdata);
 	if (ret < 0) {
 		dev_err(&pdev->dev, "failed to allocate NAPI\n");