]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
net/mlx4_en: Manage flow steering rules with ethtool
authorHadar Hen Zion <hadarh@mellanox.co.il>
Thu, 5 Jul 2012 04:03:49 +0000 (04:03 +0000)
committerDavid S. Miller <davem@davemloft.net>
Sat, 7 Jul 2012 23:23:06 +0000 (16:23 -0700)
Implement the ethtool APIs for attaching L2/L3/L4 based flow steering
rules to the netdevice RX rings. Added set_rxnfc callback and enhanced
the existing get_rxnfc callback.

Signed-off-by: Hadar Hen Zion <hadarh@mellanox.co.il>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
drivers/net/ethernet/mellanox/mlx4/mlx4_en.h

index 72901ce2b088cc764a2b5005592108e660565214..3e72a2076fb556a44914cf0ff191e6a30d010ed4 100644 (file)
 #include "mlx4_en.h"
 #include "en_port.h"
 
+#define EN_ETHTOOL_QP_ATTACH (1ull << 63)
+#define EN_ETHTOOL_MAC_MASK 0xffffffffffffULL
+#define EN_ETHTOOL_SHORT_MASK cpu_to_be16(0xffff)
+#define EN_ETHTOOL_WORD_MASK  cpu_to_be32(0xffffffff)
 
 static void
 mlx4_en_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *drvinfo)
@@ -599,16 +603,369 @@ static int mlx4_en_set_rxfh_indir(struct net_device *dev,
        return err;
 }
 
+#define all_zeros_or_all_ones(field)           \
+       ((field) == 0 || (field) == (__force typeof(field))-1)
+
+static int mlx4_en_validate_flow(struct net_device *dev,
+                                struct ethtool_rxnfc *cmd)
+{
+       struct ethtool_usrip4_spec *l3_mask;
+       struct ethtool_tcpip4_spec *l4_mask;
+       struct ethhdr *eth_mask;
+       u64 full_mac = ~0ull;
+       u64 zero_mac = 0;
+
+       if (cmd->fs.location >= MAX_NUM_OF_FS_RULES)
+               return -EINVAL;
+
+       switch (cmd->fs.flow_type & ~FLOW_EXT) {
+       case TCP_V4_FLOW:
+       case UDP_V4_FLOW:
+               if (cmd->fs.m_u.tcp_ip4_spec.tos)
+                       return -EINVAL;
+               l4_mask = &cmd->fs.m_u.tcp_ip4_spec;
+               /* don't allow mask which isn't all 0 or 1 */
+               if (!all_zeros_or_all_ones(l4_mask->ip4src) ||
+                   !all_zeros_or_all_ones(l4_mask->ip4dst) ||
+                   !all_zeros_or_all_ones(l4_mask->psrc) ||
+                   !all_zeros_or_all_ones(l4_mask->pdst))
+                       return -EINVAL;
+               break;
+       case IP_USER_FLOW:
+               l3_mask = &cmd->fs.m_u.usr_ip4_spec;
+               if (l3_mask->l4_4_bytes || l3_mask->tos || l3_mask->proto ||
+                   cmd->fs.h_u.usr_ip4_spec.ip_ver != ETH_RX_NFC_IP4 ||
+                   (!l3_mask->ip4src && !l3_mask->ip4dst) ||
+                   !all_zeros_or_all_ones(l3_mask->ip4src) ||
+                   !all_zeros_or_all_ones(l3_mask->ip4dst))
+                       return -EINVAL;
+               break;
+       case ETHER_FLOW:
+               eth_mask = &cmd->fs.m_u.ether_spec;
+               /* source mac mask must not be set */
+               if (memcmp(eth_mask->h_source, &zero_mac, ETH_ALEN))
+                       return -EINVAL;
+
+               /* dest mac mask must be ff:ff:ff:ff:ff:ff */
+               if (memcmp(eth_mask->h_dest, &full_mac, ETH_ALEN))
+                       return -EINVAL;
+
+               if (!all_zeros_or_all_ones(eth_mask->h_proto))
+                       return -EINVAL;
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       if ((cmd->fs.flow_type & FLOW_EXT)) {
+               if (cmd->fs.m_ext.vlan_etype ||
+                   !(cmd->fs.m_ext.vlan_tci == 0 ||
+                     cmd->fs.m_ext.vlan_tci == cpu_to_be16(0xfff)))
+                       return -EINVAL;
+       }
+
+       return 0;
+}
+
+static int add_ip_rule(struct mlx4_en_priv *priv,
+                       struct ethtool_rxnfc *cmd,
+                       struct list_head *list_h)
+{
+       struct mlx4_spec_list *spec_l3;
+       struct ethtool_usrip4_spec *l3_mask = &cmd->fs.m_u.usr_ip4_spec;
+
+       spec_l3 = kzalloc(sizeof *spec_l3, GFP_KERNEL);
+       if (!spec_l3) {
+               en_err(priv, "Fail to alloc ethtool rule.\n");
+               return -ENOMEM;
+       }
+
+       spec_l3->id = MLX4_NET_TRANS_RULE_ID_IPV4;
+       spec_l3->ipv4.src_ip = cmd->fs.h_u.usr_ip4_spec.ip4src;
+       if (l3_mask->ip4src)
+               spec_l3->ipv4.src_ip_msk = EN_ETHTOOL_WORD_MASK;
+       spec_l3->ipv4.dst_ip = cmd->fs.h_u.usr_ip4_spec.ip4dst;
+       if (l3_mask->ip4dst)
+               spec_l3->ipv4.dst_ip_msk = EN_ETHTOOL_WORD_MASK;
+       list_add_tail(&spec_l3->list, list_h);
+
+       return 0;
+}
+
+static int add_tcp_udp_rule(struct mlx4_en_priv *priv,
+                            struct ethtool_rxnfc *cmd,
+                            struct list_head *list_h, int proto)
+{
+       struct mlx4_spec_list *spec_l3;
+       struct mlx4_spec_list *spec_l4;
+       struct ethtool_tcpip4_spec *l4_mask = &cmd->fs.m_u.tcp_ip4_spec;
+
+       spec_l3 = kzalloc(sizeof *spec_l3, GFP_KERNEL);
+       spec_l4 = kzalloc(sizeof *spec_l4, GFP_KERNEL);
+       if (!spec_l4 || !spec_l3) {
+               en_err(priv, "Fail to alloc ethtool rule.\n");
+               kfree(spec_l3);
+               kfree(spec_l4);
+               return -ENOMEM;
+       }
+
+       spec_l3->id = MLX4_NET_TRANS_RULE_ID_IPV4;
+
+       if (proto == TCP_V4_FLOW) {
+               spec_l4->id = MLX4_NET_TRANS_RULE_ID_TCP;
+               spec_l3->ipv4.src_ip = cmd->fs.h_u.tcp_ip4_spec.ip4src;
+               spec_l3->ipv4.dst_ip = cmd->fs.h_u.tcp_ip4_spec.ip4dst;
+               spec_l4->tcp_udp.src_port = cmd->fs.h_u.tcp_ip4_spec.psrc;
+               spec_l4->tcp_udp.dst_port = cmd->fs.h_u.tcp_ip4_spec.pdst;
+       } else {
+               spec_l4->id = MLX4_NET_TRANS_RULE_ID_UDP;
+               spec_l3->ipv4.src_ip = cmd->fs.h_u.udp_ip4_spec.ip4src;
+               spec_l3->ipv4.dst_ip = cmd->fs.h_u.udp_ip4_spec.ip4dst;
+               spec_l4->tcp_udp.src_port = cmd->fs.h_u.udp_ip4_spec.psrc;
+               spec_l4->tcp_udp.dst_port = cmd->fs.h_u.udp_ip4_spec.pdst;
+       }
+
+       if (l4_mask->ip4src)
+               spec_l3->ipv4.src_ip_msk = EN_ETHTOOL_WORD_MASK;
+       if (l4_mask->ip4dst)
+               spec_l3->ipv4.dst_ip_msk = EN_ETHTOOL_WORD_MASK;
+
+       if (l4_mask->psrc)
+               spec_l4->tcp_udp.src_port_msk = EN_ETHTOOL_SHORT_MASK;
+       if (l4_mask->pdst)
+               spec_l4->tcp_udp.dst_port_msk = EN_ETHTOOL_SHORT_MASK;
+
+       list_add_tail(&spec_l3->list, list_h);
+       list_add_tail(&spec_l4->list, list_h);
+
+       return 0;
+}
+
+static int mlx4_en_ethtool_to_net_trans_rule(struct net_device *dev,
+                                            struct ethtool_rxnfc *cmd,
+                                            struct list_head *rule_list_h)
+{
+       int err;
+       u64 mac;
+       __be64 be_mac;
+       struct ethhdr *eth_spec;
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_spec_list *spec_l2;
+       __be64 mac_msk = cpu_to_be64(EN_ETHTOOL_MAC_MASK << 16);
+
+       err = mlx4_en_validate_flow(dev, cmd);
+       if (err)
+               return err;
+
+       spec_l2 = kzalloc(sizeof *spec_l2, GFP_KERNEL);
+       if (!spec_l2)
+               return -ENOMEM;
+
+       mac = priv->mac & EN_ETHTOOL_MAC_MASK;
+       be_mac = cpu_to_be64(mac << 16);
+
+       spec_l2->id = MLX4_NET_TRANS_RULE_ID_ETH;
+       memcpy(spec_l2->eth.dst_mac_msk, &mac_msk, ETH_ALEN);
+       if ((cmd->fs.flow_type & ~FLOW_EXT) != ETHER_FLOW)
+               memcpy(spec_l2->eth.dst_mac, &be_mac, ETH_ALEN);
+
+       if ((cmd->fs.flow_type & FLOW_EXT) && cmd->fs.m_ext.vlan_tci) {
+               spec_l2->eth.vlan_id = cmd->fs.h_ext.vlan_tci;
+               spec_l2->eth.vlan_id_msk = cpu_to_be16(0xfff);
+       }
+
+       list_add_tail(&spec_l2->list, rule_list_h);
+
+       switch (cmd->fs.flow_type & ~FLOW_EXT) {
+       case ETHER_FLOW:
+               eth_spec = &cmd->fs.h_u.ether_spec;
+               memcpy(&spec_l2->eth.dst_mac, eth_spec->h_dest, ETH_ALEN);
+               spec_l2->eth.ether_type = eth_spec->h_proto;
+               if (eth_spec->h_proto)
+                       spec_l2->eth.ether_type_enable = 1;
+               break;
+       case IP_USER_FLOW:
+               err = add_ip_rule(priv, cmd, rule_list_h);
+               break;
+       case TCP_V4_FLOW:
+               err = add_tcp_udp_rule(priv, cmd, rule_list_h, TCP_V4_FLOW);
+               break;
+       case UDP_V4_FLOW:
+               err = add_tcp_udp_rule(priv, cmd, rule_list_h, UDP_V4_FLOW);
+               break;
+       }
+
+       return err;
+}
+
+static int mlx4_en_flow_replace(struct net_device *dev,
+                               struct ethtool_rxnfc *cmd)
+{
+       int err;
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct ethtool_flow_id *loc_rule;
+       struct mlx4_spec_list *spec, *tmp_spec;
+       u32 qpn;
+       u64 reg_id;
+
+       struct mlx4_net_trans_rule rule = {
+               .queue_mode = MLX4_NET_TRANS_Q_FIFO,
+               .exclusive = 0,
+               .allow_loopback = 1,
+               .promisc_mode = MLX4_FS_PROMISC_NONE,
+       };
+
+       rule.port = priv->port;
+       rule.priority = MLX4_DOMAIN_ETHTOOL | cmd->fs.location;
+       INIT_LIST_HEAD(&rule.list);
+
+       /* Allow direct QP attaches if the EN_ETHTOOL_QP_ATTACH flag is set */
+       if (cmd->fs.ring_cookie == RX_CLS_FLOW_DISC)
+               return -EINVAL;
+       else if (cmd->fs.ring_cookie & EN_ETHTOOL_QP_ATTACH) {
+               qpn = cmd->fs.ring_cookie & (EN_ETHTOOL_QP_ATTACH - 1);
+       } else {
+               if (cmd->fs.ring_cookie >= priv->rx_ring_num) {
+                       en_warn(priv, "rxnfc: RX ring (%llu) doesn't exist.\n",
+                               cmd->fs.ring_cookie);
+                       return -EINVAL;
+               }
+               qpn = priv->rss_map.qps[cmd->fs.ring_cookie].qpn;
+               if (!qpn) {
+                       en_warn(priv, "rxnfc: RX ring (%llu) is inactive.\n",
+                               cmd->fs.ring_cookie);
+                       return -EINVAL;
+               }
+       }
+       rule.qpn = qpn;
+       err = mlx4_en_ethtool_to_net_trans_rule(dev, cmd, &rule.list);
+       if (err)
+               goto out_free_list;
+
+       loc_rule = &priv->ethtool_rules[cmd->fs.location];
+       if (loc_rule->id) {
+               err = mlx4_flow_detach(priv->mdev->dev, loc_rule->id);
+               if (err) {
+                       en_err(priv, "Fail to detach network rule at location %d. registration id = %llx\n",
+                              cmd->fs.location, loc_rule->id);
+                       goto out_free_list;
+               }
+               loc_rule->id = 0;
+               memset(&loc_rule->flow_spec, 0,
+                      sizeof(struct ethtool_rx_flow_spec));
+       }
+       err = mlx4_flow_attach(priv->mdev->dev, &rule, &reg_id);
+       if (err) {
+               en_err(priv, "Fail to attach network rule at location %d.\n",
+                      cmd->fs.location);
+               goto out_free_list;
+       }
+       loc_rule->id = reg_id;
+       memcpy(&loc_rule->flow_spec, &cmd->fs,
+              sizeof(struct ethtool_rx_flow_spec));
+
+out_free_list:
+       list_for_each_entry_safe(spec, tmp_spec, &rule.list, list) {
+               list_del(&spec->list);
+               kfree(spec);
+       }
+       return err;
+}
+
+static int mlx4_en_flow_detach(struct net_device *dev,
+                              struct ethtool_rxnfc *cmd)
+{
+       int err = 0;
+       struct ethtool_flow_id *rule;
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+
+       if (cmd->fs.location >= MAX_NUM_OF_FS_RULES)
+               return -EINVAL;
+
+       rule = &priv->ethtool_rules[cmd->fs.location];
+       if (!rule->id) {
+               err =  -ENOENT;
+               goto out;
+       }
+
+       err = mlx4_flow_detach(priv->mdev->dev, rule->id);
+       if (err) {
+               en_err(priv, "Fail to detach network rule at location %d. registration id = 0x%llx\n",
+                      cmd->fs.location, rule->id);
+               goto out;
+       }
+       rule->id = 0;
+       memset(&rule->flow_spec, 0, sizeof(struct ethtool_rx_flow_spec));
+out:
+       return err;
+
+}
+
+static int mlx4_en_get_flow(struct net_device *dev, struct ethtool_rxnfc *cmd,
+                           int loc)
+{
+       int err = 0;
+       struct ethtool_flow_id *rule;
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+
+       if (loc < 0 || loc >= MAX_NUM_OF_FS_RULES)
+               return -EINVAL;
+
+       rule = &priv->ethtool_rules[loc];
+       if (rule->id)
+               memcpy(&cmd->fs, &rule->flow_spec,
+                      sizeof(struct ethtool_rx_flow_spec));
+       else
+               err = -ENOENT;
+
+       return err;
+}
+
+static int mlx4_en_get_num_flows(struct mlx4_en_priv *priv)
+{
+
+       int i, res = 0;
+       for (i = 0; i < MAX_NUM_OF_FS_RULES; i++) {
+               if (priv->ethtool_rules[i].id)
+                       res++;
+       }
+       return res;
+
+}
+
 static int mlx4_en_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *cmd,
                             u32 *rule_locs)
 {
        struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
        int err = 0;
+       int i = 0, priority = 0;
+
+       if ((cmd->cmd == ETHTOOL_GRXCLSRLCNT ||
+            cmd->cmd == ETHTOOL_GRXCLSRULE ||
+            cmd->cmd == ETHTOOL_GRXCLSRLALL) &&
+           mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED)
+               return -EINVAL;
 
        switch (cmd->cmd) {
        case ETHTOOL_GRXRINGS:
                cmd->data = priv->rx_ring_num;
                break;
+       case ETHTOOL_GRXCLSRLCNT:
+               cmd->rule_cnt = mlx4_en_get_num_flows(priv);
+               break;
+       case ETHTOOL_GRXCLSRULE:
+               err = mlx4_en_get_flow(dev, cmd, cmd->fs.location);
+               break;
+       case ETHTOOL_GRXCLSRLALL:
+               while ((!err || err == -ENOENT) && priority < cmd->rule_cnt) {
+                       err = mlx4_en_get_flow(dev, cmd, i);
+                       if (!err)
+                               rule_locs[priority++] = i;
+                       i++;
+               }
+               err = 0;
+               break;
        default:
                err = -EOPNOTSUPP;
                break;
@@ -617,6 +974,30 @@ static int mlx4_en_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *cmd,
        return err;
 }
 
+static int mlx4_en_set_rxnfc(struct net_device *dev, struct ethtool_rxnfc *cmd)
+{
+       int err = 0;
+       struct mlx4_en_priv *priv = netdev_priv(dev);
+       struct mlx4_en_dev *mdev = priv->mdev;
+
+       if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED)
+               return -EINVAL;
+
+       switch (cmd->cmd) {
+       case ETHTOOL_SRXCLSRLINS:
+               err = mlx4_en_flow_replace(dev, cmd);
+               break;
+       case ETHTOOL_SRXCLSRLDEL:
+               err = mlx4_en_flow_detach(dev, cmd);
+               break;
+       default:
+               en_warn(priv, "Unsupported ethtool command. (%d)\n", cmd->cmd);
+               return -EINVAL;
+       }
+
+       return err;
+}
+
 const struct ethtool_ops mlx4_en_ethtool_ops = {
        .get_drvinfo = mlx4_en_get_drvinfo,
        .get_settings = mlx4_en_get_settings,
@@ -637,6 +1018,7 @@ const struct ethtool_ops mlx4_en_ethtool_ops = {
        .get_ringparam = mlx4_en_get_ringparam,
        .set_ringparam = mlx4_en_set_ringparam,
        .get_rxnfc = mlx4_en_get_rxnfc,
+       .set_rxnfc = mlx4_en_set_rxnfc,
        .get_rxfh_indir_size = mlx4_en_get_rxfh_indir_size,
        .get_rxfh_indir = mlx4_en_get_rxfh_indir,
        .set_rxfh_indir = mlx4_en_set_rxfh_indir,
index 2d6dabe7f55dcd0429db905c1064956fd07623b4..8882e70493fee564226bdb4fa1ae8987ff160beb 100644 (file)
@@ -75,6 +75,7 @@
 #define STAMP_SHIFT            31
 #define STAMP_VAL              0x7fffffff
 #define STATS_DELAY            (HZ / 4)
+#define MAX_NUM_OF_FS_RULES    256
 
 /* Typical TSO descriptor with 16 gather entries is 352 bytes... */
 #define MAX_DESC_SIZE          512
@@ -435,6 +436,11 @@ struct mlx4_en_frag_info {
 
 #endif
 
+struct ethtool_flow_id {
+       struct ethtool_rx_flow_spec flow_spec;
+       u64 id;
+};
+
 struct mlx4_en_priv {
        struct mlx4_en_dev *mdev;
        struct mlx4_en_port_profile *prof;
@@ -444,6 +450,7 @@ struct mlx4_en_priv {
        struct net_device_stats ret_stats;
        struct mlx4_en_port_state port_state;
        spinlock_t stats_lock;
+       struct ethtool_flow_id ethtool_rules[MAX_NUM_OF_FS_RULES];
 
        unsigned long last_moder_packets[MAX_RX_RINGS];
        unsigned long last_moder_tx_packets;