sfc: commonise ethtool NFC and RXFH/RSS functions
authorEdward Cree <ecree@solarflare.com>
Mon, 29 Jun 2020 13:35:05 +0000 (14:35 +0100)
committerDavid S. Miller <davem@davemloft.net>
Tue, 30 Jun 2020 00:37:48 +0000 (17:37 -0700)
EF100 will share EF10's model of filtering, hashing and spreading.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/sfc/ethtool.c
drivers/net/ethernet/sfc/ethtool_common.c
drivers/net/ethernet/sfc/ethtool_common.h

index 55413d4..5e0051b 100644 (file)
@@ -267,678 +267,6 @@ static int efx_ethtool_reset(struct net_device *net_dev, u32 *flags)
        return efx_reset(efx, rc);
 }
 
-/* MAC address mask including only I/G bit */
-static const u8 mac_addr_ig_mask[ETH_ALEN] __aligned(2) = {0x01, 0, 0, 0, 0, 0};
-
-#define IP4_ADDR_FULL_MASK     ((__force __be32)~0)
-#define IP_PROTO_FULL_MASK     0xFF
-#define PORT_FULL_MASK         ((__force __be16)~0)
-#define ETHER_TYPE_FULL_MASK   ((__force __be16)~0)
-
-static inline void ip6_fill_mask(__be32 *mask)
-{
-       mask[0] = mask[1] = mask[2] = mask[3] = ~(__be32)0;
-}
-
-static int efx_ethtool_get_class_rule(struct efx_nic *efx,
-                                     struct ethtool_rx_flow_spec *rule,
-                                     u32 *rss_context)
-{
-       struct ethtool_tcpip4_spec *ip_entry = &rule->h_u.tcp_ip4_spec;
-       struct ethtool_tcpip4_spec *ip_mask = &rule->m_u.tcp_ip4_spec;
-       struct ethtool_usrip4_spec *uip_entry = &rule->h_u.usr_ip4_spec;
-       struct ethtool_usrip4_spec *uip_mask = &rule->m_u.usr_ip4_spec;
-       struct ethtool_tcpip6_spec *ip6_entry = &rule->h_u.tcp_ip6_spec;
-       struct ethtool_tcpip6_spec *ip6_mask = &rule->m_u.tcp_ip6_spec;
-       struct ethtool_usrip6_spec *uip6_entry = &rule->h_u.usr_ip6_spec;
-       struct ethtool_usrip6_spec *uip6_mask = &rule->m_u.usr_ip6_spec;
-       struct ethhdr *mac_entry = &rule->h_u.ether_spec;
-       struct ethhdr *mac_mask = &rule->m_u.ether_spec;
-       struct efx_filter_spec spec;
-       int rc;
-
-       rc = efx_filter_get_filter_safe(efx, EFX_FILTER_PRI_MANUAL,
-                                       rule->location, &spec);
-       if (rc)
-               return rc;
-
-       if (spec.dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP)
-               rule->ring_cookie = RX_CLS_FLOW_DISC;
-       else
-               rule->ring_cookie = spec.dmaq_id;
-
-       if ((spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE) &&
-           spec.ether_type == htons(ETH_P_IP) &&
-           (spec.match_flags & EFX_FILTER_MATCH_IP_PROTO) &&
-           (spec.ip_proto == IPPROTO_TCP || spec.ip_proto == IPPROTO_UDP) &&
-           !(spec.match_flags &
-             ~(EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_OUTER_VID |
-               EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_REM_HOST |
-               EFX_FILTER_MATCH_IP_PROTO |
-               EFX_FILTER_MATCH_LOC_PORT | EFX_FILTER_MATCH_REM_PORT))) {
-               rule->flow_type = ((spec.ip_proto == IPPROTO_TCP) ?
-                                  TCP_V4_FLOW : UDP_V4_FLOW);
-               if (spec.match_flags & EFX_FILTER_MATCH_LOC_HOST) {
-                       ip_entry->ip4dst = spec.loc_host[0];
-                       ip_mask->ip4dst = IP4_ADDR_FULL_MASK;
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_REM_HOST) {
-                       ip_entry->ip4src = spec.rem_host[0];
-                       ip_mask->ip4src = IP4_ADDR_FULL_MASK;
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_LOC_PORT) {
-                       ip_entry->pdst = spec.loc_port;
-                       ip_mask->pdst = PORT_FULL_MASK;
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_REM_PORT) {
-                       ip_entry->psrc = spec.rem_port;
-                       ip_mask->psrc = PORT_FULL_MASK;
-               }
-       } else if ((spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE) &&
-           spec.ether_type == htons(ETH_P_IPV6) &&
-           (spec.match_flags & EFX_FILTER_MATCH_IP_PROTO) &&
-           (spec.ip_proto == IPPROTO_TCP || spec.ip_proto == IPPROTO_UDP) &&
-           !(spec.match_flags &
-             ~(EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_OUTER_VID |
-               EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_REM_HOST |
-               EFX_FILTER_MATCH_IP_PROTO |
-               EFX_FILTER_MATCH_LOC_PORT | EFX_FILTER_MATCH_REM_PORT))) {
-               rule->flow_type = ((spec.ip_proto == IPPROTO_TCP) ?
-                                  TCP_V6_FLOW : UDP_V6_FLOW);
-               if (spec.match_flags & EFX_FILTER_MATCH_LOC_HOST) {
-                       memcpy(ip6_entry->ip6dst, spec.loc_host,
-                              sizeof(ip6_entry->ip6dst));
-                       ip6_fill_mask(ip6_mask->ip6dst);
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_REM_HOST) {
-                       memcpy(ip6_entry->ip6src, spec.rem_host,
-                              sizeof(ip6_entry->ip6src));
-                       ip6_fill_mask(ip6_mask->ip6src);
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_LOC_PORT) {
-                       ip6_entry->pdst = spec.loc_port;
-                       ip6_mask->pdst = PORT_FULL_MASK;
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_REM_PORT) {
-                       ip6_entry->psrc = spec.rem_port;
-                       ip6_mask->psrc = PORT_FULL_MASK;
-               }
-       } else if (!(spec.match_flags &
-                    ~(EFX_FILTER_MATCH_LOC_MAC | EFX_FILTER_MATCH_LOC_MAC_IG |
-                      EFX_FILTER_MATCH_REM_MAC | EFX_FILTER_MATCH_ETHER_TYPE |
-                      EFX_FILTER_MATCH_OUTER_VID))) {
-               rule->flow_type = ETHER_FLOW;
-               if (spec.match_flags &
-                   (EFX_FILTER_MATCH_LOC_MAC | EFX_FILTER_MATCH_LOC_MAC_IG)) {
-                       ether_addr_copy(mac_entry->h_dest, spec.loc_mac);
-                       if (spec.match_flags & EFX_FILTER_MATCH_LOC_MAC)
-                               eth_broadcast_addr(mac_mask->h_dest);
-                       else
-                               ether_addr_copy(mac_mask->h_dest,
-                                               mac_addr_ig_mask);
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_REM_MAC) {
-                       ether_addr_copy(mac_entry->h_source, spec.rem_mac);
-                       eth_broadcast_addr(mac_mask->h_source);
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE) {
-                       mac_entry->h_proto = spec.ether_type;
-                       mac_mask->h_proto = ETHER_TYPE_FULL_MASK;
-               }
-       } else if (spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE &&
-                  spec.ether_type == htons(ETH_P_IP) &&
-                  !(spec.match_flags &
-                    ~(EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_OUTER_VID |
-                      EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_REM_HOST |
-                      EFX_FILTER_MATCH_IP_PROTO))) {
-               rule->flow_type = IPV4_USER_FLOW;
-               uip_entry->ip_ver = ETH_RX_NFC_IP4;
-               if (spec.match_flags & EFX_FILTER_MATCH_IP_PROTO) {
-                       uip_mask->proto = IP_PROTO_FULL_MASK;
-                       uip_entry->proto = spec.ip_proto;
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_LOC_HOST) {
-                       uip_entry->ip4dst = spec.loc_host[0];
-                       uip_mask->ip4dst = IP4_ADDR_FULL_MASK;
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_REM_HOST) {
-                       uip_entry->ip4src = spec.rem_host[0];
-                       uip_mask->ip4src = IP4_ADDR_FULL_MASK;
-               }
-       } else if (spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE &&
-                  spec.ether_type == htons(ETH_P_IPV6) &&
-                  !(spec.match_flags &
-                    ~(EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_OUTER_VID |
-                      EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_REM_HOST |
-                      EFX_FILTER_MATCH_IP_PROTO))) {
-               rule->flow_type = IPV6_USER_FLOW;
-               if (spec.match_flags & EFX_FILTER_MATCH_IP_PROTO) {
-                       uip6_mask->l4_proto = IP_PROTO_FULL_MASK;
-                       uip6_entry->l4_proto = spec.ip_proto;
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_LOC_HOST) {
-                       memcpy(uip6_entry->ip6dst, spec.loc_host,
-                              sizeof(uip6_entry->ip6dst));
-                       ip6_fill_mask(uip6_mask->ip6dst);
-               }
-               if (spec.match_flags & EFX_FILTER_MATCH_REM_HOST) {
-                       memcpy(uip6_entry->ip6src, spec.rem_host,
-                              sizeof(uip6_entry->ip6src));
-                       ip6_fill_mask(uip6_mask->ip6src);
-               }
-       } else {
-               /* The above should handle all filters that we insert */
-               WARN_ON(1);
-               return -EINVAL;
-       }
-
-       if (spec.match_flags & EFX_FILTER_MATCH_OUTER_VID) {
-               rule->flow_type |= FLOW_EXT;
-               rule->h_ext.vlan_tci = spec.outer_vid;
-               rule->m_ext.vlan_tci = htons(0xfff);
-       }
-
-       if (spec.flags & EFX_FILTER_FLAG_RX_RSS) {
-               rule->flow_type |= FLOW_RSS;
-               *rss_context = spec.rss_context;
-       }
-
-       return rc;
-}
-
-static int
-efx_ethtool_get_rxnfc(struct net_device *net_dev,
-                     struct ethtool_rxnfc *info, u32 *rule_locs)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-       u32 rss_context = 0;
-       s32 rc = 0;
-
-       switch (info->cmd) {
-       case ETHTOOL_GRXRINGS:
-               info->data = efx->n_rx_channels;
-               return 0;
-
-       case ETHTOOL_GRXFH: {
-               struct efx_rss_context *ctx = &efx->rss_context;
-               __u64 data;
-
-               mutex_lock(&efx->rss_lock);
-               if (info->flow_type & FLOW_RSS && info->rss_context) {
-                       ctx = efx_find_rss_context_entry(efx, info->rss_context);
-                       if (!ctx) {
-                               rc = -ENOENT;
-                               goto out_unlock;
-                       }
-               }
-
-               data = 0;
-               if (!efx_rss_active(ctx)) /* No RSS */
-                       goto out_setdata_unlock;
-
-               switch (info->flow_type & ~FLOW_RSS) {
-               case UDP_V4_FLOW:
-               case UDP_V6_FLOW:
-                       if (ctx->rx_hash_udp_4tuple)
-                               data = (RXH_L4_B_0_1 | RXH_L4_B_2_3 |
-                                       RXH_IP_SRC | RXH_IP_DST);
-                       else
-                               data = RXH_IP_SRC | RXH_IP_DST;
-                       break;
-               case TCP_V4_FLOW:
-               case TCP_V6_FLOW:
-                       data = (RXH_L4_B_0_1 | RXH_L4_B_2_3 |
-                               RXH_IP_SRC | RXH_IP_DST);
-                       break;
-               case SCTP_V4_FLOW:
-               case SCTP_V6_FLOW:
-               case AH_ESP_V4_FLOW:
-               case AH_ESP_V6_FLOW:
-               case IPV4_FLOW:
-               case IPV6_FLOW:
-                       data = RXH_IP_SRC | RXH_IP_DST;
-                       break;
-               default:
-                       break;
-               }
-out_setdata_unlock:
-               info->data = data;
-out_unlock:
-               mutex_unlock(&efx->rss_lock);
-               return rc;
-       }
-
-       case ETHTOOL_GRXCLSRLCNT:
-               info->data = efx_filter_get_rx_id_limit(efx);
-               if (info->data == 0)
-                       return -EOPNOTSUPP;
-               info->data |= RX_CLS_LOC_SPECIAL;
-               info->rule_cnt =
-                       efx_filter_count_rx_used(efx, EFX_FILTER_PRI_MANUAL);
-               return 0;
-
-       case ETHTOOL_GRXCLSRULE:
-               if (efx_filter_get_rx_id_limit(efx) == 0)
-                       return -EOPNOTSUPP;
-               rc = efx_ethtool_get_class_rule(efx, &info->fs, &rss_context);
-               if (rc < 0)
-                       return rc;
-               if (info->fs.flow_type & FLOW_RSS)
-                       info->rss_context = rss_context;
-               return 0;
-
-       case ETHTOOL_GRXCLSRLALL:
-               info->data = efx_filter_get_rx_id_limit(efx);
-               if (info->data == 0)
-                       return -EOPNOTSUPP;
-               rc = efx_filter_get_rx_ids(efx, EFX_FILTER_PRI_MANUAL,
-                                          rule_locs, info->rule_cnt);
-               if (rc < 0)
-                       return rc;
-               info->rule_cnt = rc;
-               return 0;
-
-       default:
-               return -EOPNOTSUPP;
-       }
-}
-
-static inline bool ip6_mask_is_full(__be32 mask[4])
-{
-       return !~(mask[0] & mask[1] & mask[2] & mask[3]);
-}
-
-static inline bool ip6_mask_is_empty(__be32 mask[4])
-{
-       return !(mask[0] | mask[1] | mask[2] | mask[3]);
-}
-
-static int efx_ethtool_set_class_rule(struct efx_nic *efx,
-                                     struct ethtool_rx_flow_spec *rule,
-                                     u32 rss_context)
-{
-       struct ethtool_tcpip4_spec *ip_entry = &rule->h_u.tcp_ip4_spec;
-       struct ethtool_tcpip4_spec *ip_mask = &rule->m_u.tcp_ip4_spec;
-       struct ethtool_usrip4_spec *uip_entry = &rule->h_u.usr_ip4_spec;
-       struct ethtool_usrip4_spec *uip_mask = &rule->m_u.usr_ip4_spec;
-       struct ethtool_tcpip6_spec *ip6_entry = &rule->h_u.tcp_ip6_spec;
-       struct ethtool_tcpip6_spec *ip6_mask = &rule->m_u.tcp_ip6_spec;
-       struct ethtool_usrip6_spec *uip6_entry = &rule->h_u.usr_ip6_spec;
-       struct ethtool_usrip6_spec *uip6_mask = &rule->m_u.usr_ip6_spec;
-       u32 flow_type = rule->flow_type & ~(FLOW_EXT | FLOW_RSS);
-       struct ethhdr *mac_entry = &rule->h_u.ether_spec;
-       struct ethhdr *mac_mask = &rule->m_u.ether_spec;
-       enum efx_filter_flags flags = 0;
-       struct efx_filter_spec spec;
-       int rc;
-
-       /* Check that user wants us to choose the location */
-       if (rule->location != RX_CLS_LOC_ANY)
-               return -EINVAL;
-
-       /* Range-check ring_cookie */
-       if (rule->ring_cookie >= efx->n_rx_channels &&
-           rule->ring_cookie != RX_CLS_FLOW_DISC)
-               return -EINVAL;
-
-       /* Check for unsupported extensions */
-       if ((rule->flow_type & FLOW_EXT) &&
-           (rule->m_ext.vlan_etype || rule->m_ext.data[0] ||
-            rule->m_ext.data[1]))
-               return -EINVAL;
-
-       if (efx->rx_scatter)
-               flags |= EFX_FILTER_FLAG_RX_SCATTER;
-       if (rule->flow_type & FLOW_RSS)
-               flags |= EFX_FILTER_FLAG_RX_RSS;
-
-       efx_filter_init_rx(&spec, EFX_FILTER_PRI_MANUAL, flags,
-                          (rule->ring_cookie == RX_CLS_FLOW_DISC) ?
-                          EFX_FILTER_RX_DMAQ_ID_DROP : rule->ring_cookie);
-
-       if (rule->flow_type & FLOW_RSS)
-               spec.rss_context = rss_context;
-
-       switch (flow_type) {
-       case TCP_V4_FLOW:
-       case UDP_V4_FLOW:
-               spec.match_flags = (EFX_FILTER_MATCH_ETHER_TYPE |
-                                   EFX_FILTER_MATCH_IP_PROTO);
-               spec.ether_type = htons(ETH_P_IP);
-               spec.ip_proto = flow_type == TCP_V4_FLOW ? IPPROTO_TCP
-                                                        : IPPROTO_UDP;
-               if (ip_mask->ip4dst) {
-                       if (ip_mask->ip4dst != IP4_ADDR_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_LOC_HOST;
-                       spec.loc_host[0] = ip_entry->ip4dst;
-               }
-               if (ip_mask->ip4src) {
-                       if (ip_mask->ip4src != IP4_ADDR_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_REM_HOST;
-                       spec.rem_host[0] = ip_entry->ip4src;
-               }
-               if (ip_mask->pdst) {
-                       if (ip_mask->pdst != PORT_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_LOC_PORT;
-                       spec.loc_port = ip_entry->pdst;
-               }
-               if (ip_mask->psrc) {
-                       if (ip_mask->psrc != PORT_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_REM_PORT;
-                       spec.rem_port = ip_entry->psrc;
-               }
-               if (ip_mask->tos)
-                       return -EINVAL;
-               break;
-
-       case TCP_V6_FLOW:
-       case UDP_V6_FLOW:
-               spec.match_flags = (EFX_FILTER_MATCH_ETHER_TYPE |
-                                   EFX_FILTER_MATCH_IP_PROTO);
-               spec.ether_type = htons(ETH_P_IPV6);
-               spec.ip_proto = flow_type == TCP_V6_FLOW ? IPPROTO_TCP
-                                                        : IPPROTO_UDP;
-               if (!ip6_mask_is_empty(ip6_mask->ip6dst)) {
-                       if (!ip6_mask_is_full(ip6_mask->ip6dst))
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_LOC_HOST;
-                       memcpy(spec.loc_host, ip6_entry->ip6dst, sizeof(spec.loc_host));
-               }
-               if (!ip6_mask_is_empty(ip6_mask->ip6src)) {
-                       if (!ip6_mask_is_full(ip6_mask->ip6src))
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_REM_HOST;
-                       memcpy(spec.rem_host, ip6_entry->ip6src, sizeof(spec.rem_host));
-               }
-               if (ip6_mask->pdst) {
-                       if (ip6_mask->pdst != PORT_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_LOC_PORT;
-                       spec.loc_port = ip6_entry->pdst;
-               }
-               if (ip6_mask->psrc) {
-                       if (ip6_mask->psrc != PORT_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_REM_PORT;
-                       spec.rem_port = ip6_entry->psrc;
-               }
-               if (ip6_mask->tclass)
-                       return -EINVAL;
-               break;
-
-       case IPV4_USER_FLOW:
-               if (uip_mask->l4_4_bytes || uip_mask->tos || uip_mask->ip_ver ||
-                   uip_entry->ip_ver != ETH_RX_NFC_IP4)
-                       return -EINVAL;
-               spec.match_flags = EFX_FILTER_MATCH_ETHER_TYPE;
-               spec.ether_type = htons(ETH_P_IP);
-               if (uip_mask->ip4dst) {
-                       if (uip_mask->ip4dst != IP4_ADDR_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_LOC_HOST;
-                       spec.loc_host[0] = uip_entry->ip4dst;
-               }
-               if (uip_mask->ip4src) {
-                       if (uip_mask->ip4src != IP4_ADDR_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_REM_HOST;
-                       spec.rem_host[0] = uip_entry->ip4src;
-               }
-               if (uip_mask->proto) {
-                       if (uip_mask->proto != IP_PROTO_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_IP_PROTO;
-                       spec.ip_proto = uip_entry->proto;
-               }
-               break;
-
-       case IPV6_USER_FLOW:
-               if (uip6_mask->l4_4_bytes || uip6_mask->tclass)
-                       return -EINVAL;
-               spec.match_flags = EFX_FILTER_MATCH_ETHER_TYPE;
-               spec.ether_type = htons(ETH_P_IPV6);
-               if (!ip6_mask_is_empty(uip6_mask->ip6dst)) {
-                       if (!ip6_mask_is_full(uip6_mask->ip6dst))
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_LOC_HOST;
-                       memcpy(spec.loc_host, uip6_entry->ip6dst, sizeof(spec.loc_host));
-               }
-               if (!ip6_mask_is_empty(uip6_mask->ip6src)) {
-                       if (!ip6_mask_is_full(uip6_mask->ip6src))
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_REM_HOST;
-                       memcpy(spec.rem_host, uip6_entry->ip6src, sizeof(spec.rem_host));
-               }
-               if (uip6_mask->l4_proto) {
-                       if (uip6_mask->l4_proto != IP_PROTO_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_IP_PROTO;
-                       spec.ip_proto = uip6_entry->l4_proto;
-               }
-               break;
-
-       case ETHER_FLOW:
-               if (!is_zero_ether_addr(mac_mask->h_dest)) {
-                       if (ether_addr_equal(mac_mask->h_dest,
-                                            mac_addr_ig_mask))
-                               spec.match_flags |= EFX_FILTER_MATCH_LOC_MAC_IG;
-                       else if (is_broadcast_ether_addr(mac_mask->h_dest))
-                               spec.match_flags |= EFX_FILTER_MATCH_LOC_MAC;
-                       else
-                               return -EINVAL;
-                       ether_addr_copy(spec.loc_mac, mac_entry->h_dest);
-               }
-               if (!is_zero_ether_addr(mac_mask->h_source)) {
-                       if (!is_broadcast_ether_addr(mac_mask->h_source))
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_REM_MAC;
-                       ether_addr_copy(spec.rem_mac, mac_entry->h_source);
-               }
-               if (mac_mask->h_proto) {
-                       if (mac_mask->h_proto != ETHER_TYPE_FULL_MASK)
-                               return -EINVAL;
-                       spec.match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
-                       spec.ether_type = mac_entry->h_proto;
-               }
-               break;
-
-       default:
-               return -EINVAL;
-       }
-
-       if ((rule->flow_type & FLOW_EXT) && rule->m_ext.vlan_tci) {
-               if (rule->m_ext.vlan_tci != htons(0xfff))
-                       return -EINVAL;
-               spec.match_flags |= EFX_FILTER_MATCH_OUTER_VID;
-               spec.outer_vid = rule->h_ext.vlan_tci;
-       }
-
-       rc = efx_filter_insert_filter(efx, &spec, true);
-       if (rc < 0)
-               return rc;
-
-       rule->location = rc;
-       return 0;
-}
-
-static int efx_ethtool_set_rxnfc(struct net_device *net_dev,
-                                struct ethtool_rxnfc *info)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-
-       if (efx_filter_get_rx_id_limit(efx) == 0)
-               return -EOPNOTSUPP;
-
-       switch (info->cmd) {
-       case ETHTOOL_SRXCLSRLINS:
-               return efx_ethtool_set_class_rule(efx, &info->fs,
-                                                 info->rss_context);
-
-       case ETHTOOL_SRXCLSRLDEL:
-               return efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_MANUAL,
-                                                info->fs.location);
-
-       default:
-               return -EOPNOTSUPP;
-       }
-}
-
-static u32 efx_ethtool_get_rxfh_indir_size(struct net_device *net_dev)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-
-       if (efx->n_rx_channels == 1)
-               return 0;
-       return ARRAY_SIZE(efx->rss_context.rx_indir_table);
-}
-
-static u32 efx_ethtool_get_rxfh_key_size(struct net_device *net_dev)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-
-       return efx->type->rx_hash_key_size;
-}
-
-static int efx_ethtool_get_rxfh(struct net_device *net_dev, u32 *indir, u8 *key,
-                               u8 *hfunc)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-       int rc;
-
-       rc = efx->type->rx_pull_rss_config(efx);
-       if (rc)
-               return rc;
-
-       if (hfunc)
-               *hfunc = ETH_RSS_HASH_TOP;
-       if (indir)
-               memcpy(indir, efx->rss_context.rx_indir_table,
-                      sizeof(efx->rss_context.rx_indir_table));
-       if (key)
-               memcpy(key, efx->rss_context.rx_hash_key,
-                      efx->type->rx_hash_key_size);
-       return 0;
-}
-
-static int efx_ethtool_set_rxfh(struct net_device *net_dev, const u32 *indir,
-                               const u8 *key, const u8 hfunc)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-
-       /* Hash function is Toeplitz, cannot be changed */
-       if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
-               return -EOPNOTSUPP;
-       if (!indir && !key)
-               return 0;
-
-       if (!key)
-               key = efx->rss_context.rx_hash_key;
-       if (!indir)
-               indir = efx->rss_context.rx_indir_table;
-
-       return efx->type->rx_push_rss_config(efx, true, indir, key);
-}
-
-static int efx_ethtool_get_rxfh_context(struct net_device *net_dev, u32 *indir,
-                                       u8 *key, u8 *hfunc, u32 rss_context)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-       struct efx_rss_context *ctx;
-       int rc = 0;
-
-       if (!efx->type->rx_pull_rss_context_config)
-               return -EOPNOTSUPP;
-
-       mutex_lock(&efx->rss_lock);
-       ctx = efx_find_rss_context_entry(efx, rss_context);
-       if (!ctx) {
-               rc = -ENOENT;
-               goto out_unlock;
-       }
-       rc = efx->type->rx_pull_rss_context_config(efx, ctx);
-       if (rc)
-               goto out_unlock;
-
-       if (hfunc)
-               *hfunc = ETH_RSS_HASH_TOP;
-       if (indir)
-               memcpy(indir, ctx->rx_indir_table, sizeof(ctx->rx_indir_table));
-       if (key)
-               memcpy(key, ctx->rx_hash_key, efx->type->rx_hash_key_size);
-out_unlock:
-       mutex_unlock(&efx->rss_lock);
-       return rc;
-}
-
-static int efx_ethtool_set_rxfh_context(struct net_device *net_dev,
-                                       const u32 *indir, const u8 *key,
-                                       const u8 hfunc, u32 *rss_context,
-                                       bool delete)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-       struct efx_rss_context *ctx;
-       bool allocated = false;
-       int rc;
-
-       if (!efx->type->rx_push_rss_context_config)
-               return -EOPNOTSUPP;
-       /* Hash function is Toeplitz, cannot be changed */
-       if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
-               return -EOPNOTSUPP;
-
-       mutex_lock(&efx->rss_lock);
-
-       if (*rss_context == ETH_RXFH_CONTEXT_ALLOC) {
-               if (delete) {
-                       /* alloc + delete == Nothing to do */
-                       rc = -EINVAL;
-                       goto out_unlock;
-               }
-               ctx = efx_alloc_rss_context_entry(efx);
-               if (!ctx) {
-                       rc = -ENOMEM;
-                       goto out_unlock;
-               }
-               ctx->context_id = EFX_MCDI_RSS_CONTEXT_INVALID;
-               /* Initialise indir table and key to defaults */
-               efx_set_default_rx_indir_table(efx, ctx);
-               netdev_rss_key_fill(ctx->rx_hash_key, sizeof(ctx->rx_hash_key));
-               allocated = true;
-       } else {
-               ctx = efx_find_rss_context_entry(efx, *rss_context);
-               if (!ctx) {
-                       rc = -ENOENT;
-                       goto out_unlock;
-               }
-       }
-
-       if (delete) {
-               /* delete this context */
-               rc = efx->type->rx_push_rss_context_config(efx, ctx, NULL, NULL);
-               if (!rc)
-                       efx_free_rss_context_entry(ctx);
-               goto out_unlock;
-       }
-
-       if (!key)
-               key = ctx->rx_hash_key;
-       if (!indir)
-               indir = ctx->rx_indir_table;
-
-       rc = efx->type->rx_push_rss_context_config(efx, ctx, indir, key);
-       if (rc && allocated)
-               efx_free_rss_context_entry(ctx);
-       else
-               *rss_context = ctx->user_id;
-out_unlock:
-       mutex_unlock(&efx->rss_lock);
-       return rc;
-}
-
 static int efx_ethtool_get_ts_info(struct net_device *net_dev,
                                   struct ethtool_ts_info *ts_info)
 {
index b919611..d7d8795 100644 (file)
@@ -13,6 +13,7 @@
 #include "mcdi.h"
 #include "nic.h"
 #include "selftest.h"
+#include "rx_common.h"
 #include "ethtool_common.h"
 
 struct efx_sw_stat_desc {
@@ -601,3 +602,674 @@ int efx_ethtool_set_fecparam(struct net_device *net_dev,
 
        return rc;
 }
+
+/* MAC address mask including only I/G bit */
+static const u8 mac_addr_ig_mask[ETH_ALEN] __aligned(2) = {0x01, 0, 0, 0, 0, 0};
+
+#define IP4_ADDR_FULL_MASK     ((__force __be32)~0)
+#define IP_PROTO_FULL_MASK     0xFF
+#define PORT_FULL_MASK         ((__force __be16)~0)
+#define ETHER_TYPE_FULL_MASK   ((__force __be16)~0)
+
+static inline void ip6_fill_mask(__be32 *mask)
+{
+       mask[0] = mask[1] = mask[2] = mask[3] = ~(__be32)0;
+}
+
+static int efx_ethtool_get_class_rule(struct efx_nic *efx,
+                                     struct ethtool_rx_flow_spec *rule,
+                                     u32 *rss_context)
+{
+       struct ethtool_tcpip4_spec *ip_entry = &rule->h_u.tcp_ip4_spec;
+       struct ethtool_tcpip4_spec *ip_mask = &rule->m_u.tcp_ip4_spec;
+       struct ethtool_usrip4_spec *uip_entry = &rule->h_u.usr_ip4_spec;
+       struct ethtool_usrip4_spec *uip_mask = &rule->m_u.usr_ip4_spec;
+       struct ethtool_tcpip6_spec *ip6_entry = &rule->h_u.tcp_ip6_spec;
+       struct ethtool_tcpip6_spec *ip6_mask = &rule->m_u.tcp_ip6_spec;
+       struct ethtool_usrip6_spec *uip6_entry = &rule->h_u.usr_ip6_spec;
+       struct ethtool_usrip6_spec *uip6_mask = &rule->m_u.usr_ip6_spec;
+       struct ethhdr *mac_entry = &rule->h_u.ether_spec;
+       struct ethhdr *mac_mask = &rule->m_u.ether_spec;
+       struct efx_filter_spec spec;
+       int rc;
+
+       rc = efx_filter_get_filter_safe(efx, EFX_FILTER_PRI_MANUAL,
+                                       rule->location, &spec);
+       if (rc)
+               return rc;
+
+       if (spec.dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP)
+               rule->ring_cookie = RX_CLS_FLOW_DISC;
+       else
+               rule->ring_cookie = spec.dmaq_id;
+
+       if ((spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE) &&
+           spec.ether_type == htons(ETH_P_IP) &&
+           (spec.match_flags & EFX_FILTER_MATCH_IP_PROTO) &&
+           (spec.ip_proto == IPPROTO_TCP || spec.ip_proto == IPPROTO_UDP) &&
+           !(spec.match_flags &
+             ~(EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_OUTER_VID |
+               EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_REM_HOST |
+               EFX_FILTER_MATCH_IP_PROTO |
+               EFX_FILTER_MATCH_LOC_PORT | EFX_FILTER_MATCH_REM_PORT))) {
+               rule->flow_type = ((spec.ip_proto == IPPROTO_TCP) ?
+                                  TCP_V4_FLOW : UDP_V4_FLOW);
+               if (spec.match_flags & EFX_FILTER_MATCH_LOC_HOST) {
+                       ip_entry->ip4dst = spec.loc_host[0];
+                       ip_mask->ip4dst = IP4_ADDR_FULL_MASK;
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_REM_HOST) {
+                       ip_entry->ip4src = spec.rem_host[0];
+                       ip_mask->ip4src = IP4_ADDR_FULL_MASK;
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_LOC_PORT) {
+                       ip_entry->pdst = spec.loc_port;
+                       ip_mask->pdst = PORT_FULL_MASK;
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_REM_PORT) {
+                       ip_entry->psrc = spec.rem_port;
+                       ip_mask->psrc = PORT_FULL_MASK;
+               }
+       } else if ((spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE) &&
+           spec.ether_type == htons(ETH_P_IPV6) &&
+           (spec.match_flags & EFX_FILTER_MATCH_IP_PROTO) &&
+           (spec.ip_proto == IPPROTO_TCP || spec.ip_proto == IPPROTO_UDP) &&
+           !(spec.match_flags &
+             ~(EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_OUTER_VID |
+               EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_REM_HOST |
+               EFX_FILTER_MATCH_IP_PROTO |
+               EFX_FILTER_MATCH_LOC_PORT | EFX_FILTER_MATCH_REM_PORT))) {
+               rule->flow_type = ((spec.ip_proto == IPPROTO_TCP) ?
+                                  TCP_V6_FLOW : UDP_V6_FLOW);
+               if (spec.match_flags & EFX_FILTER_MATCH_LOC_HOST) {
+                       memcpy(ip6_entry->ip6dst, spec.loc_host,
+                              sizeof(ip6_entry->ip6dst));
+                       ip6_fill_mask(ip6_mask->ip6dst);
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_REM_HOST) {
+                       memcpy(ip6_entry->ip6src, spec.rem_host,
+                              sizeof(ip6_entry->ip6src));
+                       ip6_fill_mask(ip6_mask->ip6src);
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_LOC_PORT) {
+                       ip6_entry->pdst = spec.loc_port;
+                       ip6_mask->pdst = PORT_FULL_MASK;
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_REM_PORT) {
+                       ip6_entry->psrc = spec.rem_port;
+                       ip6_mask->psrc = PORT_FULL_MASK;
+               }
+       } else if (!(spec.match_flags &
+                    ~(EFX_FILTER_MATCH_LOC_MAC | EFX_FILTER_MATCH_LOC_MAC_IG |
+                      EFX_FILTER_MATCH_REM_MAC | EFX_FILTER_MATCH_ETHER_TYPE |
+                      EFX_FILTER_MATCH_OUTER_VID))) {
+               rule->flow_type = ETHER_FLOW;
+               if (spec.match_flags &
+                   (EFX_FILTER_MATCH_LOC_MAC | EFX_FILTER_MATCH_LOC_MAC_IG)) {
+                       ether_addr_copy(mac_entry->h_dest, spec.loc_mac);
+                       if (spec.match_flags & EFX_FILTER_MATCH_LOC_MAC)
+                               eth_broadcast_addr(mac_mask->h_dest);
+                       else
+                               ether_addr_copy(mac_mask->h_dest,
+                                               mac_addr_ig_mask);
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_REM_MAC) {
+                       ether_addr_copy(mac_entry->h_source, spec.rem_mac);
+                       eth_broadcast_addr(mac_mask->h_source);
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE) {
+                       mac_entry->h_proto = spec.ether_type;
+                       mac_mask->h_proto = ETHER_TYPE_FULL_MASK;
+               }
+       } else if (spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE &&
+                  spec.ether_type == htons(ETH_P_IP) &&
+                  !(spec.match_flags &
+                    ~(EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_OUTER_VID |
+                      EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_REM_HOST |
+                      EFX_FILTER_MATCH_IP_PROTO))) {
+               rule->flow_type = IPV4_USER_FLOW;
+               uip_entry->ip_ver = ETH_RX_NFC_IP4;
+               if (spec.match_flags & EFX_FILTER_MATCH_IP_PROTO) {
+                       uip_mask->proto = IP_PROTO_FULL_MASK;
+                       uip_entry->proto = spec.ip_proto;
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_LOC_HOST) {
+                       uip_entry->ip4dst = spec.loc_host[0];
+                       uip_mask->ip4dst = IP4_ADDR_FULL_MASK;
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_REM_HOST) {
+                       uip_entry->ip4src = spec.rem_host[0];
+                       uip_mask->ip4src = IP4_ADDR_FULL_MASK;
+               }
+       } else if (spec.match_flags & EFX_FILTER_MATCH_ETHER_TYPE &&
+                  spec.ether_type == htons(ETH_P_IPV6) &&
+                  !(spec.match_flags &
+                    ~(EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_OUTER_VID |
+                      EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_REM_HOST |
+                      EFX_FILTER_MATCH_IP_PROTO))) {
+               rule->flow_type = IPV6_USER_FLOW;
+               if (spec.match_flags & EFX_FILTER_MATCH_IP_PROTO) {
+                       uip6_mask->l4_proto = IP_PROTO_FULL_MASK;
+                       uip6_entry->l4_proto = spec.ip_proto;
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_LOC_HOST) {
+                       memcpy(uip6_entry->ip6dst, spec.loc_host,
+                              sizeof(uip6_entry->ip6dst));
+                       ip6_fill_mask(uip6_mask->ip6dst);
+               }
+               if (spec.match_flags & EFX_FILTER_MATCH_REM_HOST) {
+                       memcpy(uip6_entry->ip6src, spec.rem_host,
+                              sizeof(uip6_entry->ip6src));
+                       ip6_fill_mask(uip6_mask->ip6src);
+               }
+       } else {
+               /* The above should handle all filters that we insert */
+               WARN_ON(1);
+               return -EINVAL;
+       }
+
+       if (spec.match_flags & EFX_FILTER_MATCH_OUTER_VID) {
+               rule->flow_type |= FLOW_EXT;
+               rule->h_ext.vlan_tci = spec.outer_vid;
+               rule->m_ext.vlan_tci = htons(0xfff);
+       }
+
+       if (spec.flags & EFX_FILTER_FLAG_RX_RSS) {
+               rule->flow_type |= FLOW_RSS;
+               *rss_context = spec.rss_context;
+       }
+
+       return rc;
+}
+
+int efx_ethtool_get_rxnfc(struct net_device *net_dev,
+                         struct ethtool_rxnfc *info, u32 *rule_locs)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+       u32 rss_context = 0;
+       s32 rc = 0;
+
+       switch (info->cmd) {
+       case ETHTOOL_GRXRINGS:
+               info->data = efx->n_rx_channels;
+               return 0;
+
+       case ETHTOOL_GRXFH: {
+               struct efx_rss_context *ctx = &efx->rss_context;
+               __u64 data;
+
+               mutex_lock(&efx->rss_lock);
+               if (info->flow_type & FLOW_RSS && info->rss_context) {
+                       ctx = efx_find_rss_context_entry(efx, info->rss_context);
+                       if (!ctx) {
+                               rc = -ENOENT;
+                               goto out_unlock;
+                       }
+               }
+
+               data = 0;
+               if (!efx_rss_active(ctx)) /* No RSS */
+                       goto out_setdata_unlock;
+
+               switch (info->flow_type & ~FLOW_RSS) {
+               case UDP_V4_FLOW:
+               case UDP_V6_FLOW:
+                       if (ctx->rx_hash_udp_4tuple)
+                               data = (RXH_L4_B_0_1 | RXH_L4_B_2_3 |
+                                       RXH_IP_SRC | RXH_IP_DST);
+                       else
+                               data = RXH_IP_SRC | RXH_IP_DST;
+                       break;
+               case TCP_V4_FLOW:
+               case TCP_V6_FLOW:
+                       data = (RXH_L4_B_0_1 | RXH_L4_B_2_3 |
+                               RXH_IP_SRC | RXH_IP_DST);
+                       break;
+               case SCTP_V4_FLOW:
+               case SCTP_V6_FLOW:
+               case AH_ESP_V4_FLOW:
+               case AH_ESP_V6_FLOW:
+               case IPV4_FLOW:
+               case IPV6_FLOW:
+                       data = RXH_IP_SRC | RXH_IP_DST;
+                       break;
+               default:
+                       break;
+               }
+out_setdata_unlock:
+               info->data = data;
+out_unlock:
+               mutex_unlock(&efx->rss_lock);
+               return rc;
+       }
+
+       case ETHTOOL_GRXCLSRLCNT:
+               info->data = efx_filter_get_rx_id_limit(efx);
+               if (info->data == 0)
+                       return -EOPNOTSUPP;
+               info->data |= RX_CLS_LOC_SPECIAL;
+               info->rule_cnt =
+                       efx_filter_count_rx_used(efx, EFX_FILTER_PRI_MANUAL);
+               return 0;
+
+       case ETHTOOL_GRXCLSRULE:
+               if (efx_filter_get_rx_id_limit(efx) == 0)
+                       return -EOPNOTSUPP;
+               rc = efx_ethtool_get_class_rule(efx, &info->fs, &rss_context);
+               if (rc < 0)
+                       return rc;
+               if (info->fs.flow_type & FLOW_RSS)
+                       info->rss_context = rss_context;
+               return 0;
+
+       case ETHTOOL_GRXCLSRLALL:
+               info->data = efx_filter_get_rx_id_limit(efx);
+               if (info->data == 0)
+                       return -EOPNOTSUPP;
+               rc = efx_filter_get_rx_ids(efx, EFX_FILTER_PRI_MANUAL,
+                                          rule_locs, info->rule_cnt);
+               if (rc < 0)
+                       return rc;
+               info->rule_cnt = rc;
+               return 0;
+
+       default:
+               return -EOPNOTSUPP;
+       }
+}
+
+static inline bool ip6_mask_is_full(__be32 mask[4])
+{
+       return !~(mask[0] & mask[1] & mask[2] & mask[3]);
+}
+
+static inline bool ip6_mask_is_empty(__be32 mask[4])
+{
+       return !(mask[0] | mask[1] | mask[2] | mask[3]);
+}
+
+static int efx_ethtool_set_class_rule(struct efx_nic *efx,
+                                     struct ethtool_rx_flow_spec *rule,
+                                     u32 rss_context)
+{
+       struct ethtool_tcpip4_spec *ip_entry = &rule->h_u.tcp_ip4_spec;
+       struct ethtool_tcpip4_spec *ip_mask = &rule->m_u.tcp_ip4_spec;
+       struct ethtool_usrip4_spec *uip_entry = &rule->h_u.usr_ip4_spec;
+       struct ethtool_usrip4_spec *uip_mask = &rule->m_u.usr_ip4_spec;
+       struct ethtool_tcpip6_spec *ip6_entry = &rule->h_u.tcp_ip6_spec;
+       struct ethtool_tcpip6_spec *ip6_mask = &rule->m_u.tcp_ip6_spec;
+       struct ethtool_usrip6_spec *uip6_entry = &rule->h_u.usr_ip6_spec;
+       struct ethtool_usrip6_spec *uip6_mask = &rule->m_u.usr_ip6_spec;
+       u32 flow_type = rule->flow_type & ~(FLOW_EXT | FLOW_RSS);
+       struct ethhdr *mac_entry = &rule->h_u.ether_spec;
+       struct ethhdr *mac_mask = &rule->m_u.ether_spec;
+       enum efx_filter_flags flags = 0;
+       struct efx_filter_spec spec;
+       int rc;
+
+       /* Check that user wants us to choose the location */
+       if (rule->location != RX_CLS_LOC_ANY)
+               return -EINVAL;
+
+       /* Range-check ring_cookie */
+       if (rule->ring_cookie >= efx->n_rx_channels &&
+           rule->ring_cookie != RX_CLS_FLOW_DISC)
+               return -EINVAL;
+
+       /* Check for unsupported extensions */
+       if ((rule->flow_type & FLOW_EXT) &&
+           (rule->m_ext.vlan_etype || rule->m_ext.data[0] ||
+            rule->m_ext.data[1]))
+               return -EINVAL;
+
+       if (efx->rx_scatter)
+               flags |= EFX_FILTER_FLAG_RX_SCATTER;
+       if (rule->flow_type & FLOW_RSS)
+               flags |= EFX_FILTER_FLAG_RX_RSS;
+
+       efx_filter_init_rx(&spec, EFX_FILTER_PRI_MANUAL, flags,
+                          (rule->ring_cookie == RX_CLS_FLOW_DISC) ?
+                          EFX_FILTER_RX_DMAQ_ID_DROP : rule->ring_cookie);
+
+       if (rule->flow_type & FLOW_RSS)
+               spec.rss_context = rss_context;
+
+       switch (flow_type) {
+       case TCP_V4_FLOW:
+       case UDP_V4_FLOW:
+               spec.match_flags = (EFX_FILTER_MATCH_ETHER_TYPE |
+                                   EFX_FILTER_MATCH_IP_PROTO);
+               spec.ether_type = htons(ETH_P_IP);
+               spec.ip_proto = flow_type == TCP_V4_FLOW ? IPPROTO_TCP
+                                                        : IPPROTO_UDP;
+               if (ip_mask->ip4dst) {
+                       if (ip_mask->ip4dst != IP4_ADDR_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_LOC_HOST;
+                       spec.loc_host[0] = ip_entry->ip4dst;
+               }
+               if (ip_mask->ip4src) {
+                       if (ip_mask->ip4src != IP4_ADDR_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_REM_HOST;
+                       spec.rem_host[0] = ip_entry->ip4src;
+               }
+               if (ip_mask->pdst) {
+                       if (ip_mask->pdst != PORT_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_LOC_PORT;
+                       spec.loc_port = ip_entry->pdst;
+               }
+               if (ip_mask->psrc) {
+                       if (ip_mask->psrc != PORT_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_REM_PORT;
+                       spec.rem_port = ip_entry->psrc;
+               }
+               if (ip_mask->tos)
+                       return -EINVAL;
+               break;
+
+       case TCP_V6_FLOW:
+       case UDP_V6_FLOW:
+               spec.match_flags = (EFX_FILTER_MATCH_ETHER_TYPE |
+                                   EFX_FILTER_MATCH_IP_PROTO);
+               spec.ether_type = htons(ETH_P_IPV6);
+               spec.ip_proto = flow_type == TCP_V6_FLOW ? IPPROTO_TCP
+                                                        : IPPROTO_UDP;
+               if (!ip6_mask_is_empty(ip6_mask->ip6dst)) {
+                       if (!ip6_mask_is_full(ip6_mask->ip6dst))
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_LOC_HOST;
+                       memcpy(spec.loc_host, ip6_entry->ip6dst, sizeof(spec.loc_host));
+               }
+               if (!ip6_mask_is_empty(ip6_mask->ip6src)) {
+                       if (!ip6_mask_is_full(ip6_mask->ip6src))
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_REM_HOST;
+                       memcpy(spec.rem_host, ip6_entry->ip6src, sizeof(spec.rem_host));
+               }
+               if (ip6_mask->pdst) {
+                       if (ip6_mask->pdst != PORT_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_LOC_PORT;
+                       spec.loc_port = ip6_entry->pdst;
+               }
+               if (ip6_mask->psrc) {
+                       if (ip6_mask->psrc != PORT_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_REM_PORT;
+                       spec.rem_port = ip6_entry->psrc;
+               }
+               if (ip6_mask->tclass)
+                       return -EINVAL;
+               break;
+
+       case IPV4_USER_FLOW:
+               if (uip_mask->l4_4_bytes || uip_mask->tos || uip_mask->ip_ver ||
+                   uip_entry->ip_ver != ETH_RX_NFC_IP4)
+                       return -EINVAL;
+               spec.match_flags = EFX_FILTER_MATCH_ETHER_TYPE;
+               spec.ether_type = htons(ETH_P_IP);
+               if (uip_mask->ip4dst) {
+                       if (uip_mask->ip4dst != IP4_ADDR_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_LOC_HOST;
+                       spec.loc_host[0] = uip_entry->ip4dst;
+               }
+               if (uip_mask->ip4src) {
+                       if (uip_mask->ip4src != IP4_ADDR_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_REM_HOST;
+                       spec.rem_host[0] = uip_entry->ip4src;
+               }
+               if (uip_mask->proto) {
+                       if (uip_mask->proto != IP_PROTO_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_IP_PROTO;
+                       spec.ip_proto = uip_entry->proto;
+               }
+               break;
+
+       case IPV6_USER_FLOW:
+               if (uip6_mask->l4_4_bytes || uip6_mask->tclass)
+                       return -EINVAL;
+               spec.match_flags = EFX_FILTER_MATCH_ETHER_TYPE;
+               spec.ether_type = htons(ETH_P_IPV6);
+               if (!ip6_mask_is_empty(uip6_mask->ip6dst)) {
+                       if (!ip6_mask_is_full(uip6_mask->ip6dst))
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_LOC_HOST;
+                       memcpy(spec.loc_host, uip6_entry->ip6dst, sizeof(spec.loc_host));
+               }
+               if (!ip6_mask_is_empty(uip6_mask->ip6src)) {
+                       if (!ip6_mask_is_full(uip6_mask->ip6src))
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_REM_HOST;
+                       memcpy(spec.rem_host, uip6_entry->ip6src, sizeof(spec.rem_host));
+               }
+               if (uip6_mask->l4_proto) {
+                       if (uip6_mask->l4_proto != IP_PROTO_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_IP_PROTO;
+                       spec.ip_proto = uip6_entry->l4_proto;
+               }
+               break;
+
+       case ETHER_FLOW:
+               if (!is_zero_ether_addr(mac_mask->h_dest)) {
+                       if (ether_addr_equal(mac_mask->h_dest,
+                                            mac_addr_ig_mask))
+                               spec.match_flags |= EFX_FILTER_MATCH_LOC_MAC_IG;
+                       else if (is_broadcast_ether_addr(mac_mask->h_dest))
+                               spec.match_flags |= EFX_FILTER_MATCH_LOC_MAC;
+                       else
+                               return -EINVAL;
+                       ether_addr_copy(spec.loc_mac, mac_entry->h_dest);
+               }
+               if (!is_zero_ether_addr(mac_mask->h_source)) {
+                       if (!is_broadcast_ether_addr(mac_mask->h_source))
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_REM_MAC;
+                       ether_addr_copy(spec.rem_mac, mac_entry->h_source);
+               }
+               if (mac_mask->h_proto) {
+                       if (mac_mask->h_proto != ETHER_TYPE_FULL_MASK)
+                               return -EINVAL;
+                       spec.match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
+                       spec.ether_type = mac_entry->h_proto;
+               }
+               break;
+
+       default:
+               return -EINVAL;
+       }
+
+       if ((rule->flow_type & FLOW_EXT) && rule->m_ext.vlan_tci) {
+               if (rule->m_ext.vlan_tci != htons(0xfff))
+                       return -EINVAL;
+               spec.match_flags |= EFX_FILTER_MATCH_OUTER_VID;
+               spec.outer_vid = rule->h_ext.vlan_tci;
+       }
+
+       rc = efx_filter_insert_filter(efx, &spec, true);
+       if (rc < 0)
+               return rc;
+
+       rule->location = rc;
+       return 0;
+}
+
+int efx_ethtool_set_rxnfc(struct net_device *net_dev,
+                         struct ethtool_rxnfc *info)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+
+       if (efx_filter_get_rx_id_limit(efx) == 0)
+               return -EOPNOTSUPP;
+
+       switch (info->cmd) {
+       case ETHTOOL_SRXCLSRLINS:
+               return efx_ethtool_set_class_rule(efx, &info->fs,
+                                                 info->rss_context);
+
+       case ETHTOOL_SRXCLSRLDEL:
+               return efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_MANUAL,
+                                                info->fs.location);
+
+       default:
+               return -EOPNOTSUPP;
+       }
+}
+
+u32 efx_ethtool_get_rxfh_indir_size(struct net_device *net_dev)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+
+       if (efx->n_rx_channels == 1)
+               return 0;
+       return ARRAY_SIZE(efx->rss_context.rx_indir_table);
+}
+
+u32 efx_ethtool_get_rxfh_key_size(struct net_device *net_dev)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+
+       return efx->type->rx_hash_key_size;
+}
+
+int efx_ethtool_get_rxfh(struct net_device *net_dev, u32 *indir, u8 *key,
+                        u8 *hfunc)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+       int rc;
+
+       rc = efx->type->rx_pull_rss_config(efx);
+       if (rc)
+               return rc;
+
+       if (hfunc)
+               *hfunc = ETH_RSS_HASH_TOP;
+       if (indir)
+               memcpy(indir, efx->rss_context.rx_indir_table,
+                      sizeof(efx->rss_context.rx_indir_table));
+       if (key)
+               memcpy(key, efx->rss_context.rx_hash_key,
+                      efx->type->rx_hash_key_size);
+       return 0;
+}
+
+int efx_ethtool_set_rxfh(struct net_device *net_dev, const u32 *indir,
+                        const u8 *key, const u8 hfunc)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+
+       /* Hash function is Toeplitz, cannot be changed */
+       if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
+               return -EOPNOTSUPP;
+       if (!indir && !key)
+               return 0;
+
+       if (!key)
+               key = efx->rss_context.rx_hash_key;
+       if (!indir)
+               indir = efx->rss_context.rx_indir_table;
+
+       return efx->type->rx_push_rss_config(efx, true, indir, key);
+}
+
+int efx_ethtool_get_rxfh_context(struct net_device *net_dev, u32 *indir,
+                                u8 *key, u8 *hfunc, u32 rss_context)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+       struct efx_rss_context *ctx;
+       int rc = 0;
+
+       if (!efx->type->rx_pull_rss_context_config)
+               return -EOPNOTSUPP;
+
+       mutex_lock(&efx->rss_lock);
+       ctx = efx_find_rss_context_entry(efx, rss_context);
+       if (!ctx) {
+               rc = -ENOENT;
+               goto out_unlock;
+       }
+       rc = efx->type->rx_pull_rss_context_config(efx, ctx);
+       if (rc)
+               goto out_unlock;
+
+       if (hfunc)
+               *hfunc = ETH_RSS_HASH_TOP;
+       if (indir)
+               memcpy(indir, ctx->rx_indir_table, sizeof(ctx->rx_indir_table));
+       if (key)
+               memcpy(key, ctx->rx_hash_key, efx->type->rx_hash_key_size);
+out_unlock:
+       mutex_unlock(&efx->rss_lock);
+       return rc;
+}
+
+int efx_ethtool_set_rxfh_context(struct net_device *net_dev,
+                                const u32 *indir, const u8 *key,
+                                const u8 hfunc, u32 *rss_context,
+                                bool delete)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+       struct efx_rss_context *ctx;
+       bool allocated = false;
+       int rc;
+
+       if (!efx->type->rx_push_rss_context_config)
+               return -EOPNOTSUPP;
+       /* Hash function is Toeplitz, cannot be changed */
+       if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
+               return -EOPNOTSUPP;
+
+       mutex_lock(&efx->rss_lock);
+
+       if (*rss_context == ETH_RXFH_CONTEXT_ALLOC) {
+               if (delete) {
+                       /* alloc + delete == Nothing to do */
+                       rc = -EINVAL;
+                       goto out_unlock;
+               }
+               ctx = efx_alloc_rss_context_entry(efx);
+               if (!ctx) {
+                       rc = -ENOMEM;
+                       goto out_unlock;
+               }
+               ctx->context_id = EFX_MCDI_RSS_CONTEXT_INVALID;
+               /* Initialise indir table and key to defaults */
+               efx_set_default_rx_indir_table(efx, ctx);
+               netdev_rss_key_fill(ctx->rx_hash_key, sizeof(ctx->rx_hash_key));
+               allocated = true;
+       } else {
+               ctx = efx_find_rss_context_entry(efx, *rss_context);
+               if (!ctx) {
+                       rc = -ENOENT;
+                       goto out_unlock;
+               }
+       }
+
+       if (delete) {
+               /* delete this context */
+               rc = efx->type->rx_push_rss_context_config(efx, ctx, NULL, NULL);
+               if (!rc)
+                       efx_free_rss_context_entry(ctx);
+               goto out_unlock;
+       }
+
+       if (!key)
+               key = ctx->rx_hash_key;
+       if (!indir)
+               indir = ctx->rx_indir_table;
+
+       rc = efx->type->rx_push_rss_context_config(efx, ctx, indir, key);
+       if (rc && allocated)
+               efx_free_rss_context_entry(ctx);
+       else
+               *rss_context = ctx->user_id;
+out_unlock:
+       mutex_unlock(&efx->rss_lock);
+       return rc;
+}
index eaa1fd9..024a78c 100644 (file)
@@ -37,4 +37,20 @@ int efx_ethtool_get_fecparam(struct net_device *net_dev,
                             struct ethtool_fecparam *fecparam);
 int efx_ethtool_set_fecparam(struct net_device *net_dev,
                             struct ethtool_fecparam *fecparam);
+int efx_ethtool_get_rxnfc(struct net_device *net_dev,
+                         struct ethtool_rxnfc *info, u32 *rule_locs);
+int efx_ethtool_set_rxnfc(struct net_device *net_dev,
+                         struct ethtool_rxnfc *info);
+u32 efx_ethtool_get_rxfh_indir_size(struct net_device *net_dev);
+u32 efx_ethtool_get_rxfh_key_size(struct net_device *net_dev);
+int efx_ethtool_get_rxfh(struct net_device *net_dev, u32 *indir, u8 *key,
+                        u8 *hfunc);
+int efx_ethtool_set_rxfh(struct net_device *net_dev,
+                        const u32 *indir, const u8 *key, const u8 hfunc);
+int efx_ethtool_get_rxfh_context(struct net_device *net_dev, u32 *indir,
+                                u8 *key, u8 *hfunc, u32 rss_context);
+int efx_ethtool_set_rxfh_context(struct net_device *net_dev,
+                                const u32 *indir, const u8 *key,
+                                const u8 hfunc, u32 *rss_context,
+                                bool delete);
 #endif