cxgb4: implement udp tunnel callbacks

Implement ndo_udp_tunnel_add and ndo_udp_tunnel_del
to support vxlan tunnelling.

Original work by: Santosh Rastapur <santosh@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/chelsio/cxgb4/cxgb4.h

@@ -825,6 +825,10 @@ struct mbox_list {
 	struct list_head list;
 };
 
+struct mps_encap_entry {
+	atomic_t refcnt;
+};
+
 struct adapter {
 	void __iomem *regs;
 	void __iomem *bar2;
@@ -839,6 +843,8 @@ struct adapter {
 	enum chip_type chip;
 
 	int msg_enable;
+	__be16 vxlan_port;
+	u8 vxlan_port_cnt;
 
 	struct adapter_params params;
 	struct cxgb4_virt_res vres;
@@ -868,7 +874,10 @@ struct adapter {
 	unsigned int clipt_start;
 	unsigned int clipt_end;
 	struct clip_tbl *clipt;
+	unsigned int rawf_start;
+	unsigned int rawf_cnt;
 	struct smt_data *smt;
+	struct mps_encap_entry *mps_encap;
 	struct cxgb4_uld_info *uld;
 	void *uld_handle[CXGB4_ULD_MAX];
 	unsigned int num_uld;
@@ -1637,6 +1646,12 @@ int t4_free_vi(struct adapter *adap, unsigned int mbox,
 int t4_set_rxmode(struct adapter *adap, unsigned int mbox, unsigned int viid,
 		int mtu, int promisc, int all_multi, int bcast, int vlanex,
 		bool sleep_ok);
+int t4_free_raw_mac_filt(struct adapter *adap, unsigned int viid,
+			 const u8 *addr, const u8 *mask, unsigned int idx,
+			 u8 lookup_type, u8 port_id, bool sleep_ok);
+int t4_alloc_raw_mac_filt(struct adapter *adap, unsigned int viid,
+			  const u8 *addr, const u8 *mask, unsigned int idx,
+			  u8 lookup_type, u8 port_id, bool sleep_ok);
 int t4_alloc_mac_filt(struct adapter *adap, unsigned int mbox,
 		      unsigned int viid, bool free, unsigned int naddr,
 		      const u8 **addr, u16 *idx, u64 *hash, bool sleep_ok);

drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c

@@ -65,6 +65,7 @@
 #include <net/addrconf.h>
 #include <linux/uaccess.h>
 #include <linux/crash_dump.h>
+#include <net/udp_tunnel.h>
 
 #include "cxgb4.h"
 #include "cxgb4_filter.h"
@@ -2987,6 +2988,133 @@ static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
 	}
 }
 
+static void cxgb_del_udp_tunnel(struct net_device *netdev,
+				struct udp_tunnel_info *ti)
+{
+	struct port_info *pi = netdev_priv(netdev);
+	struct adapter *adapter = pi->adapter;
+	unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
+	u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
+	int ret = 0, i;
+
+	if (chip_ver < CHELSIO_T6)
+		return;
+
+	switch (ti->type) {
+	case UDP_TUNNEL_TYPE_VXLAN:
+		if (!adapter->vxlan_port_cnt ||
+		    adapter->vxlan_port != ti->port)
+			return; /* Invalid VxLAN destination port */
+
+		adapter->vxlan_port_cnt--;
+		if (adapter->vxlan_port_cnt)
+			return;
+
+		adapter->vxlan_port = 0;
+		t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A, 0);
+		break;
+	default:
+		return;
+	}
+
+	/* Matchall mac entries can be deleted only after all tunnel ports
+	 * are brought down or removed.
+	 */
+	if (!adapter->rawf_cnt)
+		return;
+	for_each_port(adapter, i) {
+		pi = adap2pinfo(adapter, i);
+		ret = t4_free_raw_mac_filt(adapter, pi->viid,
+					   match_all_mac, match_all_mac,
+					   adapter->rawf_start +
+					    pi->port_id,
+					   1, pi->port_id, true);
+		if (ret < 0) {
+			netdev_info(netdev, "Failed to free mac filter entry, for port %d\n",
+				    i);
+			return;
+		}
+		atomic_dec(&adapter->mps_encap[adapter->rawf_start +
+			   pi->port_id].refcnt);
+	}
+}
+
+static void cxgb_add_udp_tunnel(struct net_device *netdev,
+				struct udp_tunnel_info *ti)
+{
+	struct port_info *pi = netdev_priv(netdev);
+	struct adapter *adapter = pi->adapter;
+	unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
+	u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
+	int i, ret;
+
+	if (chip_ver < CHELSIO_T6)
+		return;
+
+	switch (ti->type) {
+	case UDP_TUNNEL_TYPE_VXLAN:
+		/* For T6 fw reserves last 2 entries for
+		 * storing match all mac filter (config file entry).
+		 */
+		if (!adapter->rawf_cnt)
+			return;
+
+		/* Callback for adding vxlan port can be called with the same
+		 * port for both IPv4 and IPv6. We should not disable the
+		 * offloading when the same port for both protocols is added
+		 * and later one of them is removed.
+		 */
+		if (adapter->vxlan_port_cnt &&
+		    adapter->vxlan_port == ti->port) {
+			adapter->vxlan_port_cnt++;
+			return;
+		}
+
+		/* We will support only one VxLAN port */
+		if (adapter->vxlan_port_cnt) {
+			netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
+				    be16_to_cpu(adapter->vxlan_port),
+				    be16_to_cpu(ti->port));
+			return;
+		}
+
+		adapter->vxlan_port = ti->port;
+		adapter->vxlan_port_cnt = 1;
+
+		t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A,
+			     VXLAN_V(be16_to_cpu(ti->port)) | VXLAN_EN_F);
+		break;
+	default:
+		return;
+	}
+
+	/* Create a 'match all' mac filter entry for inner mac,
+	 * if raw mac interface is supported. Once the linux kernel provides
+	 * driver entry points for adding/deleting the inner mac addresses,
+	 * we will remove this 'match all' entry and fallback to adding
+	 * exact match filters.
+	 */
+	if (adapter->rawf_cnt) {
+		for_each_port(adapter, i) {
+			pi = adap2pinfo(adapter, i);
+
+			ret = t4_alloc_raw_mac_filt(adapter, pi->viid,
+						    match_all_mac,
+						    match_all_mac,
+						    adapter->rawf_start +
+						    pi->port_id,
+						    1, pi->port_id, true);
+			if (ret < 0) {
+				netdev_info(netdev, "Failed to allocate a mac filter entry, not adding port %d\n",
+					    be16_to_cpu(ti->port));
+				cxgb_del_udp_tunnel(netdev, ti);
+				return;
+			}
+			atomic_inc(&adapter->mps_encap[ret].refcnt);
+		}
+	}
+}
+
 static netdev_features_t cxgb_fix_features(struct net_device *dev,
 					   netdev_features_t features)
 {
@@ -3018,6 +3146,8 @@ static const struct net_device_ops cxgb4_netdev_ops = {
 #endif /* CONFIG_CHELSIO_T4_FCOE */
 	.ndo_set_tx_maxrate   = cxgb_set_tx_maxrate,
 	.ndo_setup_tc         = cxgb_setup_tc,
+	.ndo_udp_tunnel_add   = cxgb_add_udp_tunnel,
+	.ndo_udp_tunnel_del   = cxgb_del_udp_tunnel,
 	.ndo_fix_features     = cxgb_fix_features,
 };
 

drivers/net/ethernet/chelsio/cxgb4/t4_hw.c

@@ -7466,6 +7466,112 @@ int t4_set_rxmode(struct adapter *adap, unsigned int mbox, unsigned int viid,
 	return t4_wr_mbox_meat(adap, mbox, &c, sizeof(c), NULL, sleep_ok);
 }
 
+/**
+ *	t4_free_raw_mac_filt - Frees a raw mac entry in mps tcam
+ *	@adap: the adapter
+ *	@viid: the VI id
+ *	@addr: the MAC address
+ *	@mask: the mask
+ *	@idx: index of the entry in mps tcam
+ *	@lookup_type: MAC address for inner (1) or outer (0) header
+ *	@port_id: the port index
+ *	@sleep_ok: call is allowed to sleep
+ *
+ *	Removes the mac entry at the specified index using raw mac interface.
+ *
+ *	Returns a negative error number on failure.
+ */
+int t4_free_raw_mac_filt(struct adapter *adap, unsigned int viid,
+			 const u8 *addr, const u8 *mask, unsigned int idx,
+			 u8 lookup_type, u8 port_id, bool sleep_ok)
+{
+	struct fw_vi_mac_cmd c;
+	struct fw_vi_mac_raw *p = &c.u.raw;
+	u32 val;
+
+	memset(&c, 0, sizeof(c));
+	c.op_to_viid = cpu_to_be32(FW_CMD_OP_V(FW_VI_MAC_CMD) |
+				   FW_CMD_REQUEST_F | FW_CMD_WRITE_F |
+				   FW_CMD_EXEC_V(0) |
+				   FW_VI_MAC_CMD_VIID_V(viid));
+	val = FW_CMD_LEN16_V(1) |
+	      FW_VI_MAC_CMD_ENTRY_TYPE_V(FW_VI_MAC_TYPE_RAW);
+	c.freemacs_to_len16 = cpu_to_be32(FW_VI_MAC_CMD_FREEMACS_V(0) |
+					  FW_CMD_LEN16_V(val));
+
+	p->raw_idx_pkd = cpu_to_be32(FW_VI_MAC_CMD_RAW_IDX_V(idx) |
+				     FW_VI_MAC_ID_BASED_FREE);
+
+	/* Lookup Type. Outer header: 0, Inner header: 1 */
+	p->data0_pkd = cpu_to_be32(DATALKPTYPE_V(lookup_type) |
+				   DATAPORTNUM_V(port_id));
+	/* Lookup mask and port mask */
+	p->data0m_pkd = cpu_to_be64(DATALKPTYPE_V(DATALKPTYPE_M) |
+				    DATAPORTNUM_V(DATAPORTNUM_M));
+
+	/* Copy the address and the mask */
+	memcpy((u8 *)&p->data1[0] + 2, addr, ETH_ALEN);
+	memcpy((u8 *)&p->data1m[0] + 2, mask, ETH_ALEN);
+
+	return t4_wr_mbox_meat(adap, adap->mbox, &c, sizeof(c), &c, sleep_ok);
+}
+
+/**
+ *	t4_alloc_raw_mac_filt - Adds a mac entry in mps tcam
+ *	@adap: the adapter
+ *	@viid: the VI id
+ *	@mac: the MAC address
+ *	@mask: the mask
+ *	@idx: index at which to add this entry
+ *	@port_id: the port index
+ *	@lookup_type: MAC address for inner (1) or outer (0) header
+ *	@sleep_ok: call is allowed to sleep
+ *
+ *	Adds the mac entry at the specified index using raw mac interface.
+ *
+ *	Returns a negative error number or the allocated index for this mac.
+ */
+int t4_alloc_raw_mac_filt(struct adapter *adap, unsigned int viid,
+			  const u8 *addr, const u8 *mask, unsigned int idx,
+			  u8 lookup_type, u8 port_id, bool sleep_ok)
+{
+	int ret = 0;
+	struct fw_vi_mac_cmd c;
+	struct fw_vi_mac_raw *p = &c.u.raw;
+	u32 val;
+
+	memset(&c, 0, sizeof(c));
+	c.op_to_viid = cpu_to_be32(FW_CMD_OP_V(FW_VI_MAC_CMD) |
+				   FW_CMD_REQUEST_F | FW_CMD_WRITE_F |
+				   FW_VI_MAC_CMD_VIID_V(viid));
+	val = FW_CMD_LEN16_V(1) |
+	      FW_VI_MAC_CMD_ENTRY_TYPE_V(FW_VI_MAC_TYPE_RAW);
+	c.freemacs_to_len16 = cpu_to_be32(val);
+
+	/* Specify that this is an inner mac address */
+	p->raw_idx_pkd = cpu_to_be32(FW_VI_MAC_CMD_RAW_IDX_V(idx));
+
+	/* Lookup Type. Outer header: 0, Inner header: 1 */
+	p->data0_pkd = cpu_to_be32(DATALKPTYPE_V(lookup_type) |
+				   DATAPORTNUM_V(port_id));
+	/* Lookup mask and port mask */
+	p->data0m_pkd = cpu_to_be64(DATALKPTYPE_V(DATALKPTYPE_M) |
+				    DATAPORTNUM_V(DATAPORTNUM_M));
+
+	/* Copy the address and the mask */
+	memcpy((u8 *)&p->data1[0] + 2, addr, ETH_ALEN);
+	memcpy((u8 *)&p->data1m[0] + 2, mask, ETH_ALEN);
+
+	ret = t4_wr_mbox_meat(adap, adap->mbox, &c, sizeof(c), &c, sleep_ok);
+	if (ret == 0) {
+		ret = FW_VI_MAC_CMD_RAW_IDX_G(be32_to_cpu(p->raw_idx_pkd));
+		if (ret != idx)
+			ret = -ENOMEM;
+	}
+
+	return ret;
+}
+
 /**
  *	t4_alloc_mac_filt - allocates exact-match filters for MAC addresses
  *	@adap: the adapter