nfp: offload flower vxlan endpoint MAC addresses
authorJohn Hurley <john.hurley@netronome.com>
Mon, 25 Sep 2017 10:23:38 +0000 (12:23 +0200)
committerDavid S. Miller <davem@davemloft.net>
Wed, 27 Sep 2017 04:27:36 +0000 (21:27 -0700)
Generate a list of MAC addresses of netdevs that could be used as VXLAN
tunnel end points. Give offloaded MACs an index for storage on the NFP in
the ranges:
0x100-0x1ff physical port representors
0x200-0x2ff VF port representors
0x300-0x3ff other offloads (e.g. vxlan netdevs, ovs bridges)

Assign phys and vf indexes based on unique 8 bit values in the port num.
Maintain list of other netdevs to ensure same netdev is not offloaded
twice and each gets a unique ID without exhausting the entries. Because
the IDs are unique but constant for a netdev, any changes are implemented
by overwriting the index on NFP.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/netronome/nfp/Makefile
drivers/net/ethernet/netronome/nfp/flower/cmsg.c
drivers/net/ethernet/netronome/nfp/flower/cmsg.h
drivers/net/ethernet/netronome/nfp/flower/main.c
drivers/net/ethernet/netronome/nfp/flower/main.h
drivers/net/ethernet/netronome/nfp/flower/match.c
drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c [new file with mode: 0644]

index 96e579a15cbe12ba150d94b309e3b7beb1ef19ac..becaacf1554d8f6aa93dbfa012f9a0be9625c53d 100644 (file)
@@ -37,7 +37,8 @@ nfp-objs += \
            flower/main.o \
            flower/match.o \
            flower/metadata.o \
-           flower/offload.o
+           flower/offload.o \
+           flower/tunnel_conf.o
 endif
 
 ifeq ($(CONFIG_BPF_SYSCALL),y)
index c3ca05d10fe10602652f30cb67fd7ef454e3e278..b756006dba6fd8f8dde334920c55c44d3fb22f41 100644 (file)
 #include <net/dst_metadata.h>
 
 #include "main.h"
-#include "../nfpcore/nfp_cpp.h"
 #include "../nfp_net.h"
 #include "../nfp_net_repr.h"
 #include "./cmsg.h"
 
-#define nfp_flower_cmsg_warn(app, fmt, args...)                                \
-       do {                                                            \
-               if (net_ratelimit())                                    \
-                       nfp_warn((app)->cpp, fmt, ## args);             \
-       } while (0)
-
 static struct nfp_flower_cmsg_hdr *
 nfp_flower_cmsg_get_hdr(struct sk_buff *skb)
 {
index ff42ce8a1e9c037e369e00b0e8b0515f37b17e1d..dc248193c9962393f68d2de84ef22dae9eae377d 100644 (file)
@@ -39,6 +39,7 @@
 #include <linux/types.h>
 
 #include "../nfp_app.h"
+#include "../nfpcore/nfp_cpp.h"
 
 #define NFP_FLOWER_LAYER_META          BIT(0)
 #define NFP_FLOWER_LAYER_PORT          BIT(1)
 #define NFP_FL_IPV4_TUNNEL_TYPE                GENMASK(7, 4)
 #define NFP_FL_IPV4_PRE_TUN_INDEX      GENMASK(2, 0)
 
+#define nfp_flower_cmsg_warn(app, fmt, args...)                         \
+       do {                                                            \
+               if (net_ratelimit())                                    \
+                       nfp_warn((app)->cpp, fmt, ## args);             \
+       } while (0)
+
 enum nfp_flower_tun_type {
        NFP_FL_TUNNEL_NONE =    0,
        NFP_FL_TUNNEL_VXLAN =   2,
@@ -310,6 +317,7 @@ enum nfp_flower_cmsg_type_port {
        NFP_FLOWER_CMSG_TYPE_FLOW_DEL =         2,
        NFP_FLOWER_CMSG_TYPE_MAC_REPR =         7,
        NFP_FLOWER_CMSG_TYPE_PORT_MOD =         8,
+       NFP_FLOWER_CMSG_TYPE_TUN_MAC =          11,
        NFP_FLOWER_CMSG_TYPE_FLOW_STATS =       15,
        NFP_FLOWER_CMSG_TYPE_PORT_ECHO =        16,
        NFP_FLOWER_CMSG_TYPE_MAX =              32,
@@ -343,6 +351,7 @@ enum nfp_flower_cmsg_port_type {
        NFP_FLOWER_CMSG_PORT_TYPE_UNSPEC =      0x0,
        NFP_FLOWER_CMSG_PORT_TYPE_PHYS_PORT =   0x1,
        NFP_FLOWER_CMSG_PORT_TYPE_PCIE_PORT =   0x2,
+       NFP_FLOWER_CMSG_PORT_TYPE_OTHER_PORT =  0x3,
 };
 
 enum nfp_flower_cmsg_port_vnic_type {
index 91fe0361710657ecbd6fba6dcc54b00894019480..e46e7c60d49165415e8cadd2cb432fe0182e8700 100644 (file)
@@ -436,6 +436,16 @@ static void nfp_flower_clean(struct nfp_app *app)
        app->priv = NULL;
 }
 
+static int nfp_flower_start(struct nfp_app *app)
+{
+       return nfp_tunnel_config_start(app);
+}
+
+static void nfp_flower_stop(struct nfp_app *app)
+{
+       nfp_tunnel_config_stop(app);
+}
+
 const struct nfp_app_type app_flower = {
        .id             = NFP_APP_FLOWER_NIC,
        .name           = "flower",
@@ -453,6 +463,9 @@ const struct nfp_app_type app_flower = {
        .repr_open      = nfp_flower_repr_netdev_open,
        .repr_stop      = nfp_flower_repr_netdev_stop,
 
+       .start          = nfp_flower_start,
+       .stop           = nfp_flower_stop,
+
        .ctrl_msg_rx    = nfp_flower_cmsg_rx,
 
        .sriov_enable   = nfp_flower_sriov_enable,
index cd695eabce02cf5332a29a7529bfd9ab4e04b02a..9de375acc254c68f7cc9104e8aa301b86d5d8886 100644 (file)
@@ -84,6 +84,13 @@ struct nfp_fl_stats_id {
  * @flow_table:                Hash table used to store flower rules
  * @cmsg_work:         Workqueue for control messages processing
  * @cmsg_skbs:         List of skbs for control message processing
+ * @nfp_mac_off_list:  List of MAC addresses to offload
+ * @nfp_mac_index_list:        List of unique 8-bit indexes for non NFP netdevs
+ * @nfp_mac_off_lock:  Lock for the MAC address list
+ * @nfp_mac_index_lock:        Lock for the MAC index list
+ * @nfp_mac_off_ids:   IDA to manage id assignment for offloaded macs
+ * @nfp_mac_off_count: Number of MACs in address list
+ * @nfp_tun_mac_nb:    Notifier to monitor link state
  */
 struct nfp_flower_priv {
        struct nfp_app *app;
@@ -96,6 +103,13 @@ struct nfp_flower_priv {
        DECLARE_HASHTABLE(flow_table, NFP_FLOWER_HASH_BITS);
        struct work_struct cmsg_work;
        struct sk_buff_head cmsg_skbs;
+       struct list_head nfp_mac_off_list;
+       struct list_head nfp_mac_index_list;
+       struct mutex nfp_mac_off_lock;
+       struct mutex nfp_mac_index_lock;
+       struct ida nfp_mac_off_ids;
+       int nfp_mac_off_count;
+       struct notifier_block nfp_tun_mac_nb;
 };
 
 struct nfp_fl_key_ls {
@@ -165,4 +179,8 @@ nfp_flower_remove_fl_table(struct nfp_app *app, unsigned long tc_flower_cookie);
 
 void nfp_flower_rx_flow_stats(struct nfp_app *app, struct sk_buff *skb);
 
+int nfp_tunnel_config_start(struct nfp_app *app);
+void nfp_tunnel_config_stop(struct nfp_app *app);
+void nfp_tunnel_write_macs(struct nfp_app *app);
+
 #endif
index 1fd1bab0611f09b3c2c903c5a6267b6bc05c6bc4..cb3ff6c126e8a77dd4628e404549a446f6f6668e 100644 (file)
@@ -232,6 +232,7 @@ int nfp_flower_compile_flow_match(struct tc_cls_flower_offload *flow,
                                  struct nfp_fl_payload *nfp_flow)
 {
        enum nfp_flower_tun_type tun_type = NFP_FL_TUNNEL_NONE;
+       struct nfp_repr *netdev_repr;
        int err;
        u8 *ext;
        u8 *msk;
@@ -341,6 +342,12 @@ int nfp_flower_compile_flow_match(struct tc_cls_flower_offload *flow,
                                         flow, true);
                ext += sizeof(struct nfp_flower_vxlan);
                msk += sizeof(struct nfp_flower_vxlan);
+
+               /* Configure tunnel end point MAC. */
+               if (nfp_netdev_is_nfp_repr(netdev)) {
+                       netdev_repr = netdev_priv(netdev);
+                       nfp_tunnel_write_macs(netdev_repr->app);
+               }
        }
 
        return 0;
diff --git a/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c b/drivers/net/ethernet/netronome/nfp/flower/tunnel_conf.c
new file mode 100644 (file)
index 0000000..34be858
--- /dev/null
@@ -0,0 +1,374 @@
+/*
+ * Copyright (C) 2017 Netronome Systems, Inc.
+ *
+ * This software is dual licensed under the GNU General License Version 2,
+ * June 1991 as shown in the file COPYING in the top-level directory of this
+ * source tree or the BSD 2-Clause License provided below.  You have the
+ * option to license this software under the complete terms of either license.
+ *
+ * The BSD 2-Clause License:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      1. Redistributions of source code must retain the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer.
+ *
+ *      2. Redistributions in binary form must reproduce the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer in the documentation and/or other materials
+ *         provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/etherdevice.h>
+#include <linux/idr.h>
+#include <net/dst_metadata.h>
+
+#include "cmsg.h"
+#include "main.h"
+#include "../nfp_net_repr.h"
+#include "../nfp_net.h"
+
+/**
+ * struct nfp_tun_mac_addr - configure MAC address of tunnel EP on NFP
+ * @reserved:  reserved for future use
+ * @count:     number of MAC addresses in the message
+ * @index:     index of MAC address in the lookup table
+ * @addr:      interface MAC address
+ * @addresses: series of MACs to offload
+ */
+struct nfp_tun_mac_addr {
+       __be16 reserved;
+       __be16 count;
+       struct index_mac_addr {
+               __be16 index;
+               u8 addr[ETH_ALEN];
+       } addresses[];
+};
+
+/**
+ * struct nfp_tun_mac_offload_entry - list of MACs to offload
+ * @index:     index of MAC address for offloading
+ * @addr:      interface MAC address
+ * @list:      list pointer
+ */
+struct nfp_tun_mac_offload_entry {
+       __be16 index;
+       u8 addr[ETH_ALEN];
+       struct list_head list;
+};
+
+#define NFP_MAX_MAC_INDEX       0xff
+
+/**
+ * struct nfp_tun_mac_non_nfp_idx - converts non NFP netdev ifindex to 8-bit id
+ * @ifindex:   netdev ifindex of the device
+ * @index:     index of netdevs mac on NFP
+ * @list:      list pointer
+ */
+struct nfp_tun_mac_non_nfp_idx {
+       int ifindex;
+       u8 index;
+       struct list_head list;
+};
+
+static bool nfp_tun_is_netdev_to_offload(struct net_device *netdev)
+{
+       if (!netdev->rtnl_link_ops)
+               return false;
+       if (!strcmp(netdev->rtnl_link_ops->kind, "openvswitch"))
+               return true;
+       if (!strcmp(netdev->rtnl_link_ops->kind, "vxlan"))
+               return true;
+
+       return false;
+}
+
+static int
+nfp_flower_xmit_tun_conf(struct nfp_app *app, u8 mtype, u16 plen, void *pdata)
+{
+       struct sk_buff *skb;
+       unsigned char *msg;
+
+       skb = nfp_flower_cmsg_alloc(app, plen, mtype);
+       if (!skb)
+               return -ENOMEM;
+
+       msg = nfp_flower_cmsg_get_data(skb);
+       memcpy(msg, pdata, nfp_flower_cmsg_get_data_len(skb));
+
+       nfp_ctrl_tx(app->ctrl, skb);
+       return 0;
+}
+
+void nfp_tunnel_write_macs(struct nfp_app *app)
+{
+       struct nfp_flower_priv *priv = app->priv;
+       struct nfp_tun_mac_offload_entry *entry;
+       struct nfp_tun_mac_addr *payload;
+       struct list_head *ptr, *storage;
+       int mac_count, err, pay_size;
+
+       mutex_lock(&priv->nfp_mac_off_lock);
+       if (!priv->nfp_mac_off_count) {
+               mutex_unlock(&priv->nfp_mac_off_lock);
+               return;
+       }
+
+       pay_size = sizeof(struct nfp_tun_mac_addr) +
+                  sizeof(struct index_mac_addr) * priv->nfp_mac_off_count;
+
+       payload = kzalloc(pay_size, GFP_KERNEL);
+       if (!payload) {
+               mutex_unlock(&priv->nfp_mac_off_lock);
+               return;
+       }
+
+       payload->count = cpu_to_be16(priv->nfp_mac_off_count);
+
+       mac_count = 0;
+       list_for_each_safe(ptr, storage, &priv->nfp_mac_off_list) {
+               entry = list_entry(ptr, struct nfp_tun_mac_offload_entry,
+                                  list);
+               payload->addresses[mac_count].index = entry->index;
+               ether_addr_copy(payload->addresses[mac_count].addr,
+                               entry->addr);
+               mac_count++;
+       }
+
+       err = nfp_flower_xmit_tun_conf(app, NFP_FLOWER_CMSG_TYPE_TUN_MAC,
+                                      pay_size, payload);
+
+       kfree(payload);
+
+       if (err) {
+               mutex_unlock(&priv->nfp_mac_off_lock);
+               /* Write failed so retain list for future retry. */
+               return;
+       }
+
+       /* If list was successfully offloaded, flush it. */
+       list_for_each_safe(ptr, storage, &priv->nfp_mac_off_list) {
+               entry = list_entry(ptr, struct nfp_tun_mac_offload_entry,
+                                  list);
+               list_del(&entry->list);
+               kfree(entry);
+       }
+
+       priv->nfp_mac_off_count = 0;
+       mutex_unlock(&priv->nfp_mac_off_lock);
+}
+
+static int nfp_tun_get_mac_idx(struct nfp_app *app, int ifindex)
+{
+       struct nfp_flower_priv *priv = app->priv;
+       struct nfp_tun_mac_non_nfp_idx *entry;
+       struct list_head *ptr, *storage;
+       int idx;
+
+       mutex_lock(&priv->nfp_mac_index_lock);
+       list_for_each_safe(ptr, storage, &priv->nfp_mac_index_list) {
+               entry = list_entry(ptr, struct nfp_tun_mac_non_nfp_idx, list);
+               if (entry->ifindex == ifindex) {
+                       idx = entry->index;
+                       mutex_unlock(&priv->nfp_mac_index_lock);
+                       return idx;
+               }
+       }
+
+       idx = ida_simple_get(&priv->nfp_mac_off_ids, 0,
+                            NFP_MAX_MAC_INDEX, GFP_KERNEL);
+       if (idx < 0) {
+               mutex_unlock(&priv->nfp_mac_index_lock);
+               return idx;
+       }
+
+       entry = kmalloc(sizeof(*entry), GFP_KERNEL);
+       if (!entry) {
+               mutex_unlock(&priv->nfp_mac_index_lock);
+               return -ENOMEM;
+       }
+       entry->ifindex = ifindex;
+       entry->index = idx;
+       list_add_tail(&entry->list, &priv->nfp_mac_index_list);
+       mutex_unlock(&priv->nfp_mac_index_lock);
+
+       return idx;
+}
+
+static void nfp_tun_del_mac_idx(struct nfp_app *app, int ifindex)
+{
+       struct nfp_flower_priv *priv = app->priv;
+       struct nfp_tun_mac_non_nfp_idx *entry;
+       struct list_head *ptr, *storage;
+
+       mutex_lock(&priv->nfp_mac_index_lock);
+       list_for_each_safe(ptr, storage, &priv->nfp_mac_index_list) {
+               entry = list_entry(ptr, struct nfp_tun_mac_non_nfp_idx, list);
+               if (entry->ifindex == ifindex) {
+                       ida_simple_remove(&priv->nfp_mac_off_ids,
+                                         entry->index);
+                       list_del(&entry->list);
+                       kfree(entry);
+                       break;
+               }
+       }
+       mutex_unlock(&priv->nfp_mac_index_lock);
+}
+
+static void nfp_tun_add_to_mac_offload_list(struct net_device *netdev,
+                                           struct nfp_app *app)
+{
+       struct nfp_flower_priv *priv = app->priv;
+       struct nfp_tun_mac_offload_entry *entry;
+       u16 nfp_mac_idx;
+       int port = 0;
+
+       /* Check if MAC should be offloaded. */
+       if (!is_valid_ether_addr(netdev->dev_addr))
+               return;
+
+       if (nfp_netdev_is_nfp_repr(netdev))
+               port = nfp_repr_get_port_id(netdev);
+       else if (!nfp_tun_is_netdev_to_offload(netdev))
+               return;
+
+       entry = kmalloc(sizeof(*entry), GFP_KERNEL);
+       if (!entry) {
+               nfp_flower_cmsg_warn(app, "Mem fail when offloading MAC.\n");
+               return;
+       }
+
+       if (FIELD_GET(NFP_FLOWER_CMSG_PORT_TYPE, port) ==
+           NFP_FLOWER_CMSG_PORT_TYPE_PHYS_PORT) {
+               nfp_mac_idx = port << 8 | NFP_FLOWER_CMSG_PORT_TYPE_PHYS_PORT;
+       } else if (FIELD_GET(NFP_FLOWER_CMSG_PORT_TYPE, port) ==
+                  NFP_FLOWER_CMSG_PORT_TYPE_PCIE_PORT) {
+               port = FIELD_GET(NFP_FLOWER_CMSG_PORT_VNIC, port);
+               nfp_mac_idx = port << 8 | NFP_FLOWER_CMSG_PORT_TYPE_PCIE_PORT;
+       } else {
+               /* Must assign our own unique 8-bit index. */
+               int idx = nfp_tun_get_mac_idx(app, netdev->ifindex);
+
+               if (idx < 0) {
+                       nfp_flower_cmsg_warn(app, "Can't assign non-repr MAC index.\n");
+                       kfree(entry);
+                       return;
+               }
+               nfp_mac_idx = idx << 8 | NFP_FLOWER_CMSG_PORT_TYPE_OTHER_PORT;
+       }
+
+       entry->index = cpu_to_be16(nfp_mac_idx);
+       ether_addr_copy(entry->addr, netdev->dev_addr);
+
+       mutex_lock(&priv->nfp_mac_off_lock);
+       priv->nfp_mac_off_count++;
+       list_add_tail(&entry->list, &priv->nfp_mac_off_list);
+       mutex_unlock(&priv->nfp_mac_off_lock);
+}
+
+static int nfp_tun_mac_event_handler(struct notifier_block *nb,
+                                    unsigned long event, void *ptr)
+{
+       struct nfp_flower_priv *app_priv;
+       struct net_device *netdev;
+       struct nfp_app *app;
+
+       if (event == NETDEV_DOWN || event == NETDEV_UNREGISTER) {
+               app_priv = container_of(nb, struct nfp_flower_priv,
+                                       nfp_tun_mac_nb);
+               app = app_priv->app;
+               netdev = netdev_notifier_info_to_dev(ptr);
+
+               /* If non-nfp netdev then free its offload index. */
+               if (nfp_tun_is_netdev_to_offload(netdev))
+                       nfp_tun_del_mac_idx(app, netdev->ifindex);
+       } else if (event == NETDEV_UP || event == NETDEV_CHANGEADDR ||
+                  event == NETDEV_REGISTER) {
+               app_priv = container_of(nb, struct nfp_flower_priv,
+                                       nfp_tun_mac_nb);
+               app = app_priv->app;
+               netdev = netdev_notifier_info_to_dev(ptr);
+
+               nfp_tun_add_to_mac_offload_list(netdev, app);
+
+               /* Force a list write to keep NFP up to date. */
+               nfp_tunnel_write_macs(app);
+       }
+       return NOTIFY_OK;
+}
+
+int nfp_tunnel_config_start(struct nfp_app *app)
+{
+       struct nfp_flower_priv *priv = app->priv;
+       struct net_device *netdev;
+       int err;
+
+       /* Initialise priv data for MAC offloading. */
+       priv->nfp_mac_off_count = 0;
+       mutex_init(&priv->nfp_mac_off_lock);
+       INIT_LIST_HEAD(&priv->nfp_mac_off_list);
+       priv->nfp_tun_mac_nb.notifier_call = nfp_tun_mac_event_handler;
+       mutex_init(&priv->nfp_mac_index_lock);
+       INIT_LIST_HEAD(&priv->nfp_mac_index_list);
+       ida_init(&priv->nfp_mac_off_ids);
+
+       err = register_netdevice_notifier(&priv->nfp_tun_mac_nb);
+       if (err)
+               goto err_free_mac_ida;
+
+       /* Parse netdevs already registered for MACs that need offloaded. */
+       rtnl_lock();
+       for_each_netdev(&init_net, netdev)
+               nfp_tun_add_to_mac_offload_list(netdev, app);
+       rtnl_unlock();
+
+       return 0;
+
+err_free_mac_ida:
+       ida_destroy(&priv->nfp_mac_off_ids);
+       return err;
+}
+
+void nfp_tunnel_config_stop(struct nfp_app *app)
+{
+       struct nfp_tun_mac_offload_entry *mac_entry;
+       struct nfp_flower_priv *priv = app->priv;
+       struct nfp_tun_mac_non_nfp_idx *mac_idx;
+       struct list_head *ptr, *storage;
+
+       unregister_netdevice_notifier(&priv->nfp_tun_mac_nb);
+
+       /* Free any memory that may be occupied by MAC list. */
+       mutex_lock(&priv->nfp_mac_off_lock);
+       list_for_each_safe(ptr, storage, &priv->nfp_mac_off_list) {
+               mac_entry = list_entry(ptr, struct nfp_tun_mac_offload_entry,
+                                      list);
+               list_del(&mac_entry->list);
+               kfree(mac_entry);
+       }
+       mutex_unlock(&priv->nfp_mac_off_lock);
+
+       /* Free any memory that may be occupied by MAC index list. */
+       mutex_lock(&priv->nfp_mac_index_lock);
+       list_for_each_safe(ptr, storage, &priv->nfp_mac_index_list) {
+               mac_idx = list_entry(ptr, struct nfp_tun_mac_non_nfp_idx,
+                                    list);
+               list_del(&mac_idx->list);
+               kfree(mac_idx);
+       }
+       mutex_unlock(&priv->nfp_mac_index_lock);
+
+       ida_destroy(&priv->nfp_mac_off_ids);
+}