]> git.karo-electronics.de Git - linux-beck.git/commitdiff
sfc: Make most filter operations NIC-type-specific
authorBen Hutchings <bhutchings@solarflare.com>
Thu, 8 Nov 2012 01:46:53 +0000 (01:46 +0000)
committerBen Hutchings <bhutchings@solarflare.com>
Thu, 22 Aug 2013 18:25:57 +0000 (19:25 +0100)
Aside from accelerated RFS, there is almost nothing that can be shared
between the filter table implementations for the Falcon architecture
and EF10.

Move the few shared functions into efx.c and rx.c and the rest into
farch.c.  Introduce efx_nic_type operations for the implementation and
inline wrapper functions that call these.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
drivers/net/ethernet/sfc/Makefile
drivers/net/ethernet/sfc/efx.c
drivers/net/ethernet/sfc/efx.h
drivers/net/ethernet/sfc/falcon.c
drivers/net/ethernet/sfc/farch.c
drivers/net/ethernet/sfc/filter.c [deleted file]
drivers/net/ethernet/sfc/net_driver.h
drivers/net/ethernet/sfc/nic.h
drivers/net/ethernet/sfc/rx.c
drivers/net/ethernet/sfc/siena.c

index ef7410f014d64220735020f01a75873c05d388cd..a61272661a732ec04864a1f2cc13598bb6c308f9 100644 (file)
@@ -1,5 +1,4 @@
 sfc-y                  += efx.o nic.o farch.o falcon.o siena.o tx.o rx.o \
-                          filter.o \
                           selftest.o ethtool.o qt202x_phy.o mdio_10g.o \
                           tenxpress.o txc43128_phy.o falcon_boards.o \
                           mcdi.o mcdi_port.o mcdi_mon.o ptp.o
index 49d06ca79d7d1e69b8738549541900b7ad53978c..a2daaae266d72b368e97ce823d67e58832d99961 100644 (file)
@@ -611,7 +611,7 @@ static void efx_start_datapath(struct efx_nic *efx)
 
        /* RX filters also have scatter-enabled flags */
        if (efx->rx_scatter != old_rx_scatter)
-               efx_filter_update_rx_scatter(efx);
+               efx->type->filter_update_rx_scatter(efx);
 
        /* We must keep at least one descriptor in a TX ring empty.
         * We could avoid this when the queue size does not exactly
@@ -1499,6 +1499,44 @@ static void efx_remove_nic(struct efx_nic *efx)
        efx->type->remove(efx);
 }
 
+static int efx_probe_filters(struct efx_nic *efx)
+{
+       int rc;
+
+       spin_lock_init(&efx->filter_lock);
+
+       rc = efx->type->filter_table_probe(efx);
+       if (rc)
+               return rc;
+
+#ifdef CONFIG_RFS_ACCEL
+       if (efx->type->offload_features & NETIF_F_NTUPLE) {
+               efx->rps_flow_id = kcalloc(efx->type->max_rx_ip_filters,
+                                          sizeof(*efx->rps_flow_id),
+                                          GFP_KERNEL);
+               if (!efx->rps_flow_id) {
+                       efx->type->filter_table_remove(efx);
+                       return -ENOMEM;
+               }
+       }
+#endif
+
+       return 0;
+}
+
+static void efx_remove_filters(struct efx_nic *efx)
+{
+#ifdef CONFIG_RFS_ACCEL
+       kfree(efx->rps_flow_id);
+#endif
+       efx->type->filter_table_remove(efx);
+}
+
+static void efx_restore_filters(struct efx_nic *efx)
+{
+       efx->type->filter_table_restore(efx);
+}
+
 /**************************************************************************
  *
  * NIC startup/shutdown
index 45de5b9fedbee153764631170d7c3f5343f3b768..9e3573872e57c9902b71d43f48961c89f3225f04 100644 (file)
@@ -68,27 +68,92 @@ extern void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue);
 #define EFX_TXQ_MIN_ENT(efx)   (2 * efx_tx_max_skb_descs(efx))
 
 /* Filters */
-extern int efx_probe_filters(struct efx_nic *efx);
-extern void efx_restore_filters(struct efx_nic *efx);
-extern void efx_remove_filters(struct efx_nic *efx);
-extern void efx_filter_update_rx_scatter(struct efx_nic *efx);
-extern s32 efx_filter_insert_filter(struct efx_nic *efx,
-                                   struct efx_filter_spec *spec,
-                                   bool replace);
-extern int efx_filter_remove_id_safe(struct efx_nic *efx,
-                                    enum efx_filter_priority priority,
-                                    u32 filter_id);
-extern int efx_filter_get_filter_safe(struct efx_nic *efx,
-                                     enum efx_filter_priority priority,
-                                     u32 filter_id, struct efx_filter_spec *);
-extern void efx_filter_clear_rx(struct efx_nic *efx,
-                               enum efx_filter_priority priority);
-extern u32 efx_filter_count_rx_used(struct efx_nic *efx,
-                                   enum efx_filter_priority priority);
-extern u32 efx_filter_get_rx_id_limit(struct efx_nic *efx);
-extern s32 efx_filter_get_rx_ids(struct efx_nic *efx,
-                                enum efx_filter_priority priority,
-                                u32 *buf, u32 size);
+
+/**
+ * efx_filter_insert_filter - add or replace a filter
+ * @efx: NIC in which to insert the filter
+ * @spec: Specification for the filter
+ * @replace_equal: Flag for whether the specified filter may replace an
+ *     existing filter with equal priority
+ *
+ * On success, return the filter ID.
+ * On failure, return a negative error code.
+ *
+ * If an existing filter has equal match values to the new filter
+ * spec, then the new filter might replace it, depending on the
+ * relative priorities.  If the existing filter has lower priority, or
+ * if @replace_equal is set and it has equal priority, then it is
+ * replaced.  Otherwise the function fails, returning -%EPERM if
+ * the existing filter has higher priority or -%EEXIST if it has
+ * equal priority.
+ */
+static inline s32 efx_filter_insert_filter(struct efx_nic *efx,
+                                          struct efx_filter_spec *spec,
+                                          bool replace_equal)
+{
+       return efx->type->filter_insert(efx, spec, replace_equal);
+}
+
+/**
+ * efx_filter_remove_id_safe - remove a filter by ID, carefully
+ * @efx: NIC from which to remove the filter
+ * @priority: Priority of filter, as passed to @efx_filter_insert_filter
+ * @filter_id: ID of filter, as returned by @efx_filter_insert_filter
+ *
+ * This function will range-check @filter_id, so it is safe to call
+ * with a value passed from userland.
+ */
+static inline int efx_filter_remove_id_safe(struct efx_nic *efx,
+                                           enum efx_filter_priority priority,
+                                           u32 filter_id)
+{
+       return efx->type->filter_remove_safe(efx, priority, filter_id);
+}
+
+/**
+ * efx_filter_get_filter_safe - retrieve a filter by ID, carefully
+ * @efx: NIC from which to remove the filter
+ * @priority: Priority of filter, as passed to @efx_filter_insert_filter
+ * @filter_id: ID of filter, as returned by @efx_filter_insert_filter
+ * @spec: Buffer in which to store filter specification
+ *
+ * This function will range-check @filter_id, so it is safe to call
+ * with a value passed from userland.
+ */
+static inline int
+efx_filter_get_filter_safe(struct efx_nic *efx,
+                          enum efx_filter_priority priority,
+                          u32 filter_id, struct efx_filter_spec *spec)
+{
+       return efx->type->filter_get_safe(efx, priority, filter_id, spec);
+}
+
+/**
+ * efx_farch_filter_clear_rx - remove RX filters by priority
+ * @efx: NIC from which to remove the filters
+ * @priority: Maximum priority to remove
+ */
+static inline void efx_filter_clear_rx(struct efx_nic *efx,
+                                      enum efx_filter_priority priority)
+{
+       return efx->type->filter_clear_rx(efx, priority);
+}
+
+static inline u32 efx_filter_count_rx_used(struct efx_nic *efx,
+                                          enum efx_filter_priority priority)
+{
+       return efx->type->filter_count_rx_used(efx, priority);
+}
+static inline u32 efx_filter_get_rx_id_limit(struct efx_nic *efx)
+{
+       return efx->type->filter_get_rx_id_limit(efx);
+}
+static inline s32 efx_filter_get_rx_ids(struct efx_nic *efx,
+                                       enum efx_filter_priority priority,
+                                       u32 *buf, u32 size)
+{
+       return efx->type->filter_get_rx_ids(efx, priority, buf, size);
+}
 #ifdef CONFIG_RFS_ACCEL
 extern int efx_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
                          u16 rxq_index, u32 flow_id);
index 0fd8a88905a607d73c9d05aebe94bc4b85506891..6ea28f8e879294d6199e4294c116d1f189dfc41a 100644 (file)
@@ -2400,6 +2400,21 @@ const struct efx_nic_type falcon_a1_nic_type = {
        .ev_read_ack = efx_farch_ev_read_ack,
        .ev_test_generate = efx_farch_ev_test_generate,
 
+       /* We don't expose the filter table on Falcon A1 as it is not
+        * mapped into function 0, but these implementations still
+        * work with a degenerate case of all tables set to size 0.
+        */
+       .filter_table_probe = efx_farch_filter_table_probe,
+       .filter_table_restore = efx_farch_filter_table_restore,
+       .filter_table_remove = efx_farch_filter_table_remove,
+       .filter_insert = efx_farch_filter_insert,
+       .filter_remove_safe = efx_farch_filter_remove_safe,
+       .filter_get_safe = efx_farch_filter_get_safe,
+       .filter_clear_rx = efx_farch_filter_clear_rx,
+       .filter_count_rx_used = efx_farch_filter_count_rx_used,
+       .filter_get_rx_id_limit = efx_farch_filter_get_rx_id_limit,
+       .filter_get_rx_ids = efx_farch_filter_get_rx_ids,
+
        .revision = EFX_REV_FALCON_A1,
        .txd_ptr_tbl_base = FR_AA_TX_DESC_PTR_TBL_KER,
        .rxd_ptr_tbl_base = FR_AA_RX_DESC_PTR_TBL_KER,
@@ -2468,6 +2483,21 @@ const struct efx_nic_type falcon_b0_nic_type = {
        .ev_process = efx_farch_ev_process,
        .ev_read_ack = efx_farch_ev_read_ack,
        .ev_test_generate = efx_farch_ev_test_generate,
+       .filter_table_probe = efx_farch_filter_table_probe,
+       .filter_table_restore = efx_farch_filter_table_restore,
+       .filter_table_remove = efx_farch_filter_table_remove,
+       .filter_update_rx_scatter = efx_farch_filter_update_rx_scatter,
+       .filter_insert = efx_farch_filter_insert,
+       .filter_remove_safe = efx_farch_filter_remove_safe,
+       .filter_get_safe = efx_farch_filter_get_safe,
+       .filter_clear_rx = efx_farch_filter_clear_rx,
+       .filter_count_rx_used = efx_farch_filter_count_rx_used,
+       .filter_get_rx_id_limit = efx_farch_filter_get_rx_id_limit,
+       .filter_get_rx_ids = efx_farch_filter_get_rx_ids,
+#ifdef CONFIG_RFS_ACCEL
+       .filter_rfs_insert = efx_farch_filter_rfs_insert,
+       .filter_rfs_expire_one = efx_farch_filter_rfs_expire_one,
+#endif
 
        .revision = EFX_REV_FALCON_B0,
        .txd_ptr_tbl_base = FR_BZ_TX_DESC_PTR_TBL,
@@ -2483,5 +2513,6 @@ const struct efx_nic_type falcon_b0_nic_type = {
        .timer_period_max =  1 << FRF_AB_TC_TIMER_VAL_WIDTH,
        .offload_features = NETIF_F_IP_CSUM | NETIF_F_RXHASH | NETIF_F_NTUPLE,
        .mcdi_max_ver = -1,
+       .max_rx_ip_filters = FR_BZ_RX_FILTER_TBL0_ROWS,
 };
 
index 7f50882f11ab363492a8ad0d81793f4ccdef482a..19418045f3e0db0c0a6b8325604568b95c101e01 100644 (file)
@@ -1779,3 +1779,1111 @@ void efx_farch_init_common(struct efx_nic *efx)
                efx_writeo(efx, &temp, FR_BZ_TX_PACE);
        }
 }
+
+/**************************************************************************
+ *
+ * Filter tables
+ *
+ **************************************************************************
+ */
+
+/* "Fudge factors" - difference between programmed value and actual depth.
+ * Due to pipelined implementation we need to program H/W with a value that
+ * is larger than the hop limit we want.
+ */
+#define EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD 3
+#define EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL 1
+
+/* Hard maximum search limit.  Hardware will time-out beyond 200-something.
+ * We also need to avoid infinite loops in efx_farch_filter_search() when the
+ * table is full.
+ */
+#define EFX_FARCH_FILTER_CTL_SRCH_MAX 200
+
+/* Don't try very hard to find space for performance hints, as this is
+ * counter-productive. */
+#define EFX_FARCH_FILTER_CTL_SRCH_HINT_MAX 5
+
+enum efx_farch_filter_type {
+       EFX_FARCH_FILTER_TCP_FULL = 0,
+       EFX_FARCH_FILTER_TCP_WILD,
+       EFX_FARCH_FILTER_UDP_FULL,
+       EFX_FARCH_FILTER_UDP_WILD,
+       EFX_FARCH_FILTER_MAC_FULL = 4,
+       EFX_FARCH_FILTER_MAC_WILD,
+       EFX_FARCH_FILTER_UC_DEF = 8,
+       EFX_FARCH_FILTER_MC_DEF,
+       EFX_FARCH_FILTER_TYPE_COUNT,            /* number of specific types */
+};
+
+enum efx_farch_filter_table_id {
+       EFX_FARCH_FILTER_TABLE_RX_IP = 0,
+       EFX_FARCH_FILTER_TABLE_RX_MAC,
+       EFX_FARCH_FILTER_TABLE_RX_DEF,
+       EFX_FARCH_FILTER_TABLE_TX_MAC,
+       EFX_FARCH_FILTER_TABLE_COUNT,
+};
+
+enum efx_farch_filter_index {
+       EFX_FARCH_FILTER_INDEX_UC_DEF,
+       EFX_FARCH_FILTER_INDEX_MC_DEF,
+       EFX_FARCH_FILTER_SIZE_RX_DEF,
+};
+
+struct efx_farch_filter_spec {
+       u8      type:4;
+       u8      priority:4;
+       u8      flags;
+       u16     dmaq_id;
+       u32     data[3];
+};
+
+struct efx_farch_filter_table {
+       enum efx_farch_filter_table_id id;
+       u32             offset;         /* address of table relative to BAR */
+       unsigned        size;           /* number of entries */
+       unsigned        step;           /* step between entries */
+       unsigned        used;           /* number currently used */
+       unsigned long   *used_bitmap;
+       struct efx_farch_filter_spec *spec;
+       unsigned        search_limit[EFX_FARCH_FILTER_TYPE_COUNT];
+};
+
+struct efx_farch_filter_state {
+       struct efx_farch_filter_table table[EFX_FARCH_FILTER_TABLE_COUNT];
+};
+
+static void
+efx_farch_filter_table_clear_entry(struct efx_nic *efx,
+                                  struct efx_farch_filter_table *table,
+                                  unsigned int filter_idx);
+
+/* The filter hash function is LFSR polynomial x^16 + x^3 + 1 of a 32-bit
+ * key derived from the n-tuple.  The initial LFSR state is 0xffff. */
+static u16 efx_farch_filter_hash(u32 key)
+{
+       u16 tmp;
+
+       /* First 16 rounds */
+       tmp = 0x1fff ^ key >> 16;
+       tmp = tmp ^ tmp >> 3 ^ tmp >> 6;
+       tmp = tmp ^ tmp >> 9;
+       /* Last 16 rounds */
+       tmp = tmp ^ tmp << 13 ^ key;
+       tmp = tmp ^ tmp >> 3 ^ tmp >> 6;
+       return tmp ^ tmp >> 9;
+}
+
+/* To allow for hash collisions, filter search continues at these
+ * increments from the first possible entry selected by the hash. */
+static u16 efx_farch_filter_increment(u32 key)
+{
+       return key * 2 - 1;
+}
+
+static enum efx_farch_filter_table_id
+efx_farch_filter_spec_table_id(const struct efx_farch_filter_spec *spec)
+{
+       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
+                    (EFX_FARCH_FILTER_TCP_FULL >> 2));
+       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
+                    (EFX_FARCH_FILTER_TCP_WILD >> 2));
+       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
+                    (EFX_FARCH_FILTER_UDP_FULL >> 2));
+       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
+                    (EFX_FARCH_FILTER_UDP_WILD >> 2));
+       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_MAC !=
+                    (EFX_FARCH_FILTER_MAC_FULL >> 2));
+       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_MAC !=
+                    (EFX_FARCH_FILTER_MAC_WILD >> 2));
+       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_TX_MAC !=
+                    EFX_FARCH_FILTER_TABLE_RX_MAC + 2);
+       return (spec->type >> 2) + ((spec->flags & EFX_FILTER_FLAG_TX) ? 2 : 0);
+}
+
+static void efx_farch_filter_push_rx_config(struct efx_nic *efx)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       struct efx_farch_filter_table *table;
+       efx_oword_t filter_ctl;
+
+       efx_reado(efx, &filter_ctl, FR_BZ_RX_FILTER_CTL);
+
+       table = &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
+       EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_TCP_FULL_SRCH_LIMIT,
+                           table->search_limit[EFX_FARCH_FILTER_TCP_FULL] +
+                           EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
+       EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_TCP_WILD_SRCH_LIMIT,
+                           table->search_limit[EFX_FARCH_FILTER_TCP_WILD] +
+                           EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
+       EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_UDP_FULL_SRCH_LIMIT,
+                           table->search_limit[EFX_FARCH_FILTER_UDP_FULL] +
+                           EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
+       EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_UDP_WILD_SRCH_LIMIT,
+                           table->search_limit[EFX_FARCH_FILTER_UDP_WILD] +
+                           EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
+
+       table = &state->table[EFX_FARCH_FILTER_TABLE_RX_MAC];
+       if (table->size) {
+               EFX_SET_OWORD_FIELD(
+                       filter_ctl, FRF_CZ_ETHERNET_FULL_SEARCH_LIMIT,
+                       table->search_limit[EFX_FARCH_FILTER_MAC_FULL] +
+                       EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
+               EFX_SET_OWORD_FIELD(
+                       filter_ctl, FRF_CZ_ETHERNET_WILDCARD_SEARCH_LIMIT,
+                       table->search_limit[EFX_FARCH_FILTER_MAC_WILD] +
+                       EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
+       }
+
+       table = &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
+       if (table->size) {
+               EFX_SET_OWORD_FIELD(
+                       filter_ctl, FRF_CZ_UNICAST_NOMATCH_Q_ID,
+                       table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].dmaq_id);
+               EFX_SET_OWORD_FIELD(
+                       filter_ctl, FRF_CZ_UNICAST_NOMATCH_RSS_ENABLED,
+                       !!(table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].flags &
+                          EFX_FILTER_FLAG_RX_RSS));
+               EFX_SET_OWORD_FIELD(
+                       filter_ctl, FRF_CZ_MULTICAST_NOMATCH_Q_ID,
+                       table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].dmaq_id);
+               EFX_SET_OWORD_FIELD(
+                       filter_ctl, FRF_CZ_MULTICAST_NOMATCH_RSS_ENABLED,
+                       !!(table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].flags &
+                          EFX_FILTER_FLAG_RX_RSS));
+
+               /* There is a single bit to enable RX scatter for all
+                * unmatched packets.  Only set it if scatter is
+                * enabled in both filter specs.
+                */
+               EFX_SET_OWORD_FIELD(
+                       filter_ctl, FRF_BZ_SCATTER_ENBL_NO_MATCH_Q,
+                       !!(table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].flags &
+                          table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].flags &
+                          EFX_FILTER_FLAG_RX_SCATTER));
+       } else if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
+               /* We don't expose 'default' filters because unmatched
+                * packets always go to the queue number found in the
+                * RSS table.  But we still need to set the RX scatter
+                * bit here.
+                */
+               EFX_SET_OWORD_FIELD(
+                       filter_ctl, FRF_BZ_SCATTER_ENBL_NO_MATCH_Q,
+                       efx->rx_scatter);
+       }
+
+       efx_writeo(efx, &filter_ctl, FR_BZ_RX_FILTER_CTL);
+}
+
+static void efx_farch_filter_push_tx_limits(struct efx_nic *efx)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       struct efx_farch_filter_table *table;
+       efx_oword_t tx_cfg;
+
+       efx_reado(efx, &tx_cfg, FR_AZ_TX_CFG);
+
+       table = &state->table[EFX_FARCH_FILTER_TABLE_TX_MAC];
+       if (table->size) {
+               EFX_SET_OWORD_FIELD(
+                       tx_cfg, FRF_CZ_TX_ETH_FILTER_FULL_SEARCH_RANGE,
+                       table->search_limit[EFX_FARCH_FILTER_MAC_FULL] +
+                       EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
+               EFX_SET_OWORD_FIELD(
+                       tx_cfg, FRF_CZ_TX_ETH_FILTER_WILD_SEARCH_RANGE,
+                       table->search_limit[EFX_FARCH_FILTER_MAC_WILD] +
+                       EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
+       }
+
+       efx_writeo(efx, &tx_cfg, FR_AZ_TX_CFG);
+}
+
+static int
+efx_farch_filter_from_gen_spec(struct efx_farch_filter_spec *spec,
+                              const struct efx_filter_spec *gen_spec)
+{
+       bool is_full = false;
+
+       if ((gen_spec->flags & EFX_FILTER_FLAG_RX_RSS) &&
+           gen_spec->rss_context != EFX_FILTER_RSS_CONTEXT_DEFAULT)
+               return -EINVAL;
+
+       spec->priority = gen_spec->priority;
+       spec->flags = gen_spec->flags;
+       spec->dmaq_id = gen_spec->dmaq_id;
+
+       switch (gen_spec->match_flags) {
+       case (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_IP_PROTO |
+             EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT |
+             EFX_FILTER_MATCH_REM_HOST | EFX_FILTER_MATCH_REM_PORT):
+               is_full = true;
+               /* fall through */
+       case (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_IP_PROTO |
+             EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT): {
+               __be32 rhost, host1, host2;
+               __be16 rport, port1, port2;
+
+               EFX_BUG_ON_PARANOID(!(gen_spec->flags & EFX_FILTER_FLAG_RX));
+
+               if (gen_spec->ether_type != htons(ETH_P_IP))
+                       return -EPROTONOSUPPORT;
+               if (gen_spec->loc_port == 0 ||
+                   (is_full && gen_spec->rem_port == 0))
+                       return -EADDRNOTAVAIL;
+               switch (gen_spec->ip_proto) {
+               case IPPROTO_TCP:
+                       spec->type = (is_full ? EFX_FARCH_FILTER_TCP_FULL :
+                                     EFX_FARCH_FILTER_TCP_WILD);
+                       break;
+               case IPPROTO_UDP:
+                       spec->type = (is_full ? EFX_FARCH_FILTER_UDP_FULL :
+                                     EFX_FARCH_FILTER_UDP_WILD);
+                       break;
+               default:
+                       return -EPROTONOSUPPORT;
+               }
+
+               /* Filter is constructed in terms of source and destination,
+                * with the odd wrinkle that the ports are swapped in a UDP
+                * wildcard filter.  We need to convert from local and remote
+                * (= zero for wildcard) addresses.
+                */
+               rhost = is_full ? gen_spec->rem_host[0] : 0;
+               rport = is_full ? gen_spec->rem_port : 0;
+               host1 = rhost;
+               host2 = gen_spec->loc_host[0];
+               if (!is_full && gen_spec->ip_proto == IPPROTO_UDP) {
+                       port1 = gen_spec->loc_port;
+                       port2 = rport;
+               } else {
+                       port1 = rport;
+                       port2 = gen_spec->loc_port;
+               }
+               spec->data[0] = ntohl(host1) << 16 | ntohs(port1);
+               spec->data[1] = ntohs(port2) << 16 | ntohl(host1) >> 16;
+               spec->data[2] = ntohl(host2);
+
+               break;
+       }
+
+       case EFX_FILTER_MATCH_LOC_MAC | EFX_FILTER_MATCH_OUTER_VID:
+               is_full = true;
+               /* fall through */
+       case EFX_FILTER_MATCH_LOC_MAC:
+               spec->type = (is_full ? EFX_FARCH_FILTER_MAC_FULL :
+                             EFX_FARCH_FILTER_MAC_WILD);
+               spec->data[0] = is_full ? ntohs(gen_spec->outer_vid) : 0;
+               spec->data[1] = (gen_spec->loc_mac[2] << 24 |
+                                gen_spec->loc_mac[3] << 16 |
+                                gen_spec->loc_mac[4] << 8 |
+                                gen_spec->loc_mac[5]);
+               spec->data[2] = (gen_spec->loc_mac[0] << 8 |
+                                gen_spec->loc_mac[1]);
+               break;
+
+       case EFX_FILTER_MATCH_LOC_MAC_IG:
+               spec->type = (is_multicast_ether_addr(gen_spec->loc_mac) ?
+                             EFX_FARCH_FILTER_MC_DEF :
+                             EFX_FARCH_FILTER_UC_DEF);
+               memset(spec->data, 0, sizeof(spec->data)); /* ensure equality */
+               break;
+
+       default:
+               return -EPROTONOSUPPORT;
+       }
+
+       return 0;
+}
+
+static void
+efx_farch_filter_to_gen_spec(struct efx_filter_spec *gen_spec,
+                            const struct efx_farch_filter_spec *spec)
+{
+       bool is_full = false;
+
+       /* *gen_spec should be completely initialised, to be consistent
+        * with efx_filter_init_{rx,tx}() and in case we want to copy
+        * it back to userland.
+        */
+       memset(gen_spec, 0, sizeof(*gen_spec));
+
+       gen_spec->priority = spec->priority;
+       gen_spec->flags = spec->flags;
+       gen_spec->dmaq_id = spec->dmaq_id;
+
+       switch (spec->type) {
+       case EFX_FARCH_FILTER_TCP_FULL:
+       case EFX_FARCH_FILTER_UDP_FULL:
+               is_full = true;
+               /* fall through */
+       case EFX_FARCH_FILTER_TCP_WILD:
+       case EFX_FARCH_FILTER_UDP_WILD: {
+               __be32 host1, host2;
+               __be16 port1, port2;
+
+               gen_spec->match_flags =
+                       EFX_FILTER_MATCH_ETHER_TYPE |
+                       EFX_FILTER_MATCH_IP_PROTO |
+                       EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT;
+               if (is_full)
+                       gen_spec->match_flags |= (EFX_FILTER_MATCH_REM_HOST |
+                                                 EFX_FILTER_MATCH_REM_PORT);
+               gen_spec->ether_type = htons(ETH_P_IP);
+               gen_spec->ip_proto =
+                       (spec->type == EFX_FARCH_FILTER_TCP_FULL ||
+                        spec->type == EFX_FARCH_FILTER_TCP_WILD) ?
+                       IPPROTO_TCP : IPPROTO_UDP;
+
+               host1 = htonl(spec->data[0] >> 16 | spec->data[1] << 16);
+               port1 = htons(spec->data[0]);
+               host2 = htonl(spec->data[2]);
+               port2 = htons(spec->data[1] >> 16);
+               if (spec->flags & EFX_FILTER_FLAG_TX) {
+                       gen_spec->loc_host[0] = host1;
+                       gen_spec->rem_host[0] = host2;
+               } else {
+                       gen_spec->loc_host[0] = host2;
+                       gen_spec->rem_host[0] = host1;
+               }
+               if (!!(gen_spec->flags & EFX_FILTER_FLAG_TX) ^
+                   (!is_full && gen_spec->ip_proto == IPPROTO_UDP)) {
+                       gen_spec->loc_port = port1;
+                       gen_spec->rem_port = port2;
+               } else {
+                       gen_spec->loc_port = port2;
+                       gen_spec->rem_port = port1;
+               }
+
+               break;
+       }
+
+       case EFX_FARCH_FILTER_MAC_FULL:
+               is_full = true;
+               /* fall through */
+       case EFX_FARCH_FILTER_MAC_WILD:
+               gen_spec->match_flags = EFX_FILTER_MATCH_LOC_MAC;
+               if (is_full)
+                       gen_spec->match_flags |= EFX_FILTER_MATCH_OUTER_VID;
+               gen_spec->loc_mac[0] = spec->data[2] >> 8;
+               gen_spec->loc_mac[1] = spec->data[2];
+               gen_spec->loc_mac[2] = spec->data[1] >> 24;
+               gen_spec->loc_mac[3] = spec->data[1] >> 16;
+               gen_spec->loc_mac[4] = spec->data[1] >> 8;
+               gen_spec->loc_mac[5] = spec->data[1];
+               gen_spec->outer_vid = htons(spec->data[0]);
+               break;
+
+       case EFX_FARCH_FILTER_UC_DEF:
+       case EFX_FARCH_FILTER_MC_DEF:
+               gen_spec->match_flags = EFX_FILTER_MATCH_LOC_MAC_IG;
+               gen_spec->loc_mac[0] = spec->type == EFX_FARCH_FILTER_MC_DEF;
+               break;
+
+       default:
+               WARN_ON(1);
+               break;
+       }
+}
+
+static void
+efx_farch_filter_reset_rx_def(struct efx_nic *efx, unsigned filter_idx)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       struct efx_farch_filter_table *table =
+               &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
+       struct efx_farch_filter_spec *spec = &table->spec[filter_idx];
+
+       /* If there's only one channel then disable RSS for non VF
+        * traffic, thereby allowing VFs to use RSS when the PF can't.
+        */
+       spec->type = EFX_FARCH_FILTER_UC_DEF + filter_idx;
+       spec->priority = EFX_FILTER_PRI_MANUAL;
+       spec->flags = (EFX_FILTER_FLAG_RX |
+                      (efx->n_rx_channels > 1 ? EFX_FILTER_FLAG_RX_RSS : 0) |
+                      (efx->rx_scatter ? EFX_FILTER_FLAG_RX_SCATTER : 0));
+       spec->dmaq_id = 0;
+       table->used_bitmap[0] |= 1 << filter_idx;
+}
+
+/* Build a filter entry and return its n-tuple key. */
+static u32 efx_farch_filter_build(efx_oword_t *filter,
+                                 struct efx_farch_filter_spec *spec)
+{
+       u32 data3;
+
+       switch (efx_farch_filter_spec_table_id(spec)) {
+       case EFX_FARCH_FILTER_TABLE_RX_IP: {
+               bool is_udp = (spec->type == EFX_FARCH_FILTER_UDP_FULL ||
+                              spec->type == EFX_FARCH_FILTER_UDP_WILD);
+               EFX_POPULATE_OWORD_7(
+                       *filter,
+                       FRF_BZ_RSS_EN,
+                       !!(spec->flags & EFX_FILTER_FLAG_RX_RSS),
+                       FRF_BZ_SCATTER_EN,
+                       !!(spec->flags & EFX_FILTER_FLAG_RX_SCATTER),
+                       FRF_BZ_TCP_UDP, is_udp,
+                       FRF_BZ_RXQ_ID, spec->dmaq_id,
+                       EFX_DWORD_2, spec->data[2],
+                       EFX_DWORD_1, spec->data[1],
+                       EFX_DWORD_0, spec->data[0]);
+               data3 = is_udp;
+               break;
+       }
+
+       case EFX_FARCH_FILTER_TABLE_RX_MAC: {
+               bool is_wild = spec->type == EFX_FARCH_FILTER_MAC_WILD;
+               EFX_POPULATE_OWORD_7(
+                       *filter,
+                       FRF_CZ_RMFT_RSS_EN,
+                       !!(spec->flags & EFX_FILTER_FLAG_RX_RSS),
+                       FRF_CZ_RMFT_SCATTER_EN,
+                       !!(spec->flags & EFX_FILTER_FLAG_RX_SCATTER),
+                       FRF_CZ_RMFT_RXQ_ID, spec->dmaq_id,
+                       FRF_CZ_RMFT_WILDCARD_MATCH, is_wild,
+                       FRF_CZ_RMFT_DEST_MAC_HI, spec->data[2],
+                       FRF_CZ_RMFT_DEST_MAC_LO, spec->data[1],
+                       FRF_CZ_RMFT_VLAN_ID, spec->data[0]);
+               data3 = is_wild;
+               break;
+       }
+
+       case EFX_FARCH_FILTER_TABLE_TX_MAC: {
+               bool is_wild = spec->type == EFX_FARCH_FILTER_MAC_WILD;
+               EFX_POPULATE_OWORD_5(*filter,
+                                    FRF_CZ_TMFT_TXQ_ID, spec->dmaq_id,
+                                    FRF_CZ_TMFT_WILDCARD_MATCH, is_wild,
+                                    FRF_CZ_TMFT_SRC_MAC_HI, spec->data[2],
+                                    FRF_CZ_TMFT_SRC_MAC_LO, spec->data[1],
+                                    FRF_CZ_TMFT_VLAN_ID, spec->data[0]);
+               data3 = is_wild | spec->dmaq_id << 1;
+               break;
+       }
+
+       default:
+               BUG();
+       }
+
+       return spec->data[0] ^ spec->data[1] ^ spec->data[2] ^ data3;
+}
+
+static bool efx_farch_filter_equal(const struct efx_farch_filter_spec *left,
+                                  const struct efx_farch_filter_spec *right)
+{
+       if (left->type != right->type ||
+           memcmp(left->data, right->data, sizeof(left->data)))
+               return false;
+
+       if (left->flags & EFX_FILTER_FLAG_TX &&
+           left->dmaq_id != right->dmaq_id)
+               return false;
+
+       return true;
+}
+
+/*
+ * Construct/deconstruct external filter IDs.  At least the RX filter
+ * IDs must be ordered by matching priority, for RX NFC semantics.
+ *
+ * Deconstruction needs to be robust against invalid IDs so that
+ * efx_filter_remove_id_safe() and efx_filter_get_filter_safe() can
+ * accept user-provided IDs.
+ */
+
+#define EFX_FARCH_FILTER_MATCH_PRI_COUNT       5
+
+static const u8 efx_farch_filter_type_match_pri[EFX_FARCH_FILTER_TYPE_COUNT] = {
+       [EFX_FARCH_FILTER_TCP_FULL]     = 0,
+       [EFX_FARCH_FILTER_UDP_FULL]     = 0,
+       [EFX_FARCH_FILTER_TCP_WILD]     = 1,
+       [EFX_FARCH_FILTER_UDP_WILD]     = 1,
+       [EFX_FARCH_FILTER_MAC_FULL]     = 2,
+       [EFX_FARCH_FILTER_MAC_WILD]     = 3,
+       [EFX_FARCH_FILTER_UC_DEF]       = 4,
+       [EFX_FARCH_FILTER_MC_DEF]       = 4,
+};
+
+static const enum efx_farch_filter_table_id efx_farch_filter_range_table[] = {
+       EFX_FARCH_FILTER_TABLE_RX_IP,   /* RX match pri 0 */
+       EFX_FARCH_FILTER_TABLE_RX_IP,
+       EFX_FARCH_FILTER_TABLE_RX_MAC,
+       EFX_FARCH_FILTER_TABLE_RX_MAC,
+       EFX_FARCH_FILTER_TABLE_RX_DEF,  /* RX match pri 4 */
+       EFX_FARCH_FILTER_TABLE_TX_MAC,  /* TX match pri 0 */
+       EFX_FARCH_FILTER_TABLE_TX_MAC,  /* TX match pri 1 */
+};
+
+#define EFX_FARCH_FILTER_INDEX_WIDTH 13
+#define EFX_FARCH_FILTER_INDEX_MASK ((1 << EFX_FARCH_FILTER_INDEX_WIDTH) - 1)
+
+static inline u32
+efx_farch_filter_make_id(const struct efx_farch_filter_spec *spec,
+                        unsigned int index)
+{
+       unsigned int range;
+
+       range = efx_farch_filter_type_match_pri[spec->type];
+       if (!(spec->flags & EFX_FILTER_FLAG_RX))
+               range += EFX_FARCH_FILTER_MATCH_PRI_COUNT;
+
+       return range << EFX_FARCH_FILTER_INDEX_WIDTH | index;
+}
+
+static inline enum efx_farch_filter_table_id
+efx_farch_filter_id_table_id(u32 id)
+{
+       unsigned int range = id >> EFX_FARCH_FILTER_INDEX_WIDTH;
+
+       if (range < ARRAY_SIZE(efx_farch_filter_range_table))
+               return efx_farch_filter_range_table[range];
+       else
+               return EFX_FARCH_FILTER_TABLE_COUNT; /* invalid */
+}
+
+static inline unsigned int efx_farch_filter_id_index(u32 id)
+{
+       return id & EFX_FARCH_FILTER_INDEX_MASK;
+}
+
+u32 efx_farch_filter_get_rx_id_limit(struct efx_nic *efx)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       unsigned int range = EFX_FARCH_FILTER_MATCH_PRI_COUNT - 1;
+       enum efx_farch_filter_table_id table_id;
+
+       do {
+               table_id = efx_farch_filter_range_table[range];
+               if (state->table[table_id].size != 0)
+                       return range << EFX_FARCH_FILTER_INDEX_WIDTH |
+                               state->table[table_id].size;
+       } while (range--);
+
+       return 0;
+}
+
+s32 efx_farch_filter_insert(struct efx_nic *efx,
+                           struct efx_filter_spec *gen_spec,
+                           bool replace_equal)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       struct efx_farch_filter_table *table;
+       struct efx_farch_filter_spec spec;
+       efx_oword_t filter;
+       int rep_index, ins_index;
+       unsigned int depth = 0;
+       int rc;
+
+       rc = efx_farch_filter_from_gen_spec(&spec, gen_spec);
+       if (rc)
+               return rc;
+
+       table = &state->table[efx_farch_filter_spec_table_id(&spec)];
+       if (table->size == 0)
+               return -EINVAL;
+
+       netif_vdbg(efx, hw, efx->net_dev,
+                  "%s: type %d search_limit=%d", __func__, spec.type,
+                  table->search_limit[spec.type]);
+
+       if (table->id == EFX_FARCH_FILTER_TABLE_RX_DEF) {
+               /* One filter spec per type */
+               BUILD_BUG_ON(EFX_FARCH_FILTER_INDEX_UC_DEF != 0);
+               BUILD_BUG_ON(EFX_FARCH_FILTER_INDEX_MC_DEF !=
+                            EFX_FARCH_FILTER_MC_DEF - EFX_FARCH_FILTER_UC_DEF);
+               rep_index = spec.type - EFX_FARCH_FILTER_UC_DEF;
+               ins_index = rep_index;
+
+               spin_lock_bh(&efx->filter_lock);
+       } else {
+               /* Search concurrently for
+                * (1) a filter to be replaced (rep_index): any filter
+                *     with the same match values, up to the current
+                *     search depth for this type, and
+                * (2) the insertion point (ins_index): (1) or any
+                *     free slot before it or up to the maximum search
+                *     depth for this priority
+                * We fail if we cannot find (2).
+                *
+                * We can stop once either
+                * (a) we find (1), in which case we have definitely
+                *     found (2) as well; or
+                * (b) we have searched exhaustively for (1), and have
+                *     either found (2) or searched exhaustively for it
+                */
+               u32 key = efx_farch_filter_build(&filter, &spec);
+               unsigned int hash = efx_farch_filter_hash(key);
+               unsigned int incr = efx_farch_filter_increment(key);
+               unsigned int max_rep_depth = table->search_limit[spec.type];
+               unsigned int max_ins_depth =
+                       spec.priority <= EFX_FILTER_PRI_HINT ?
+                       EFX_FARCH_FILTER_CTL_SRCH_HINT_MAX :
+                       EFX_FARCH_FILTER_CTL_SRCH_MAX;
+               unsigned int i = hash & (table->size - 1);
+
+               ins_index = -1;
+               depth = 1;
+
+               spin_lock_bh(&efx->filter_lock);
+
+               for (;;) {
+                       if (!test_bit(i, table->used_bitmap)) {
+                               if (ins_index < 0)
+                                       ins_index = i;
+                       } else if (efx_farch_filter_equal(&spec,
+                                                         &table->spec[i])) {
+                               /* Case (a) */
+                               if (ins_index < 0)
+                                       ins_index = i;
+                               rep_index = i;
+                               break;
+                       }
+
+                       if (depth >= max_rep_depth &&
+                           (ins_index >= 0 || depth >= max_ins_depth)) {
+                               /* Case (b) */
+                               if (ins_index < 0) {
+                                       rc = -EBUSY;
+                                       goto out;
+                               }
+                               rep_index = -1;
+                               break;
+                       }
+
+                       i = (i + incr) & (table->size - 1);
+                       ++depth;
+               }
+       }
+
+       /* If we found a filter to be replaced, check whether we
+        * should do so
+        */
+       if (rep_index >= 0) {
+               struct efx_farch_filter_spec *saved_spec =
+                       &table->spec[rep_index];
+
+               if (spec.priority == saved_spec->priority && !replace_equal) {
+                       rc = -EEXIST;
+                       goto out;
+               }
+               if (spec.priority < saved_spec->priority) {
+                       rc = -EPERM;
+                       goto out;
+               }
+       }
+
+       /* Insert the filter */
+       if (ins_index != rep_index) {
+               __set_bit(ins_index, table->used_bitmap);
+               ++table->used;
+       }
+       table->spec[ins_index] = spec;
+
+       if (table->id == EFX_FARCH_FILTER_TABLE_RX_DEF) {
+               efx_farch_filter_push_rx_config(efx);
+       } else {
+               if (table->search_limit[spec.type] < depth) {
+                       table->search_limit[spec.type] = depth;
+                       if (spec.flags & EFX_FILTER_FLAG_TX)
+                               efx_farch_filter_push_tx_limits(efx);
+                       else
+                               efx_farch_filter_push_rx_config(efx);
+               }
+
+               efx_writeo(efx, &filter,
+                          table->offset + table->step * ins_index);
+
+               /* If we were able to replace a filter by inserting
+                * at a lower depth, clear the replaced filter
+                */
+               if (ins_index != rep_index && rep_index >= 0)
+                       efx_farch_filter_table_clear_entry(efx, table,
+                                                          rep_index);
+       }
+
+       netif_vdbg(efx, hw, efx->net_dev,
+                  "%s: filter type %d index %d rxq %u set",
+                  __func__, spec.type, ins_index, spec.dmaq_id);
+       rc = efx_farch_filter_make_id(&spec, ins_index);
+
+out:
+       spin_unlock_bh(&efx->filter_lock);
+       return rc;
+}
+
+static void
+efx_farch_filter_table_clear_entry(struct efx_nic *efx,
+                                  struct efx_farch_filter_table *table,
+                                  unsigned int filter_idx)
+{
+       static efx_oword_t filter;
+
+       if (table->id == EFX_FARCH_FILTER_TABLE_RX_DEF) {
+               /* RX default filters must always exist */
+               efx_farch_filter_reset_rx_def(efx, filter_idx);
+               efx_farch_filter_push_rx_config(efx);
+       } else if (test_bit(filter_idx, table->used_bitmap)) {
+               __clear_bit(filter_idx, table->used_bitmap);
+               --table->used;
+               memset(&table->spec[filter_idx], 0, sizeof(table->spec[0]));
+
+               efx_writeo(efx, &filter,
+                          table->offset + table->step * filter_idx);
+
+               /* If this filter required a greater search depth than
+                * any other, the search limit for its type can now be
+                * decreased.  However, it is hard to determine that
+                * unless the table has become completely empty - in
+                * which case, all its search limits can be set to 0.
+                */
+               if (unlikely(table->used == 0)) {
+                       memset(table->search_limit, 0,
+                              sizeof(table->search_limit));
+                       if (table->id == EFX_FARCH_FILTER_TABLE_TX_MAC)
+                               efx_farch_filter_push_tx_limits(efx);
+                       else
+                               efx_farch_filter_push_rx_config(efx);
+               }
+       }
+}
+
+int efx_farch_filter_remove_safe(struct efx_nic *efx,
+                                enum efx_filter_priority priority,
+                                u32 filter_id)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       enum efx_farch_filter_table_id table_id;
+       struct efx_farch_filter_table *table;
+       unsigned int filter_idx;
+       struct efx_farch_filter_spec *spec;
+       int rc;
+
+       table_id = efx_farch_filter_id_table_id(filter_id);
+       if ((unsigned int)table_id >= EFX_FARCH_FILTER_TABLE_COUNT)
+               return -ENOENT;
+       table = &state->table[table_id];
+
+       filter_idx = efx_farch_filter_id_index(filter_id);
+       if (filter_idx >= table->size)
+               return -ENOENT;
+       spec = &table->spec[filter_idx];
+
+       spin_lock_bh(&efx->filter_lock);
+
+       if (test_bit(filter_idx, table->used_bitmap) &&
+           spec->priority == priority) {
+               efx_farch_filter_table_clear_entry(efx, table, filter_idx);
+               rc = 0;
+       } else {
+               rc = -ENOENT;
+       }
+
+       spin_unlock_bh(&efx->filter_lock);
+
+       return rc;
+}
+
+int efx_farch_filter_get_safe(struct efx_nic *efx,
+                             enum efx_filter_priority priority,
+                             u32 filter_id, struct efx_filter_spec *spec_buf)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       enum efx_farch_filter_table_id table_id;
+       struct efx_farch_filter_table *table;
+       struct efx_farch_filter_spec *spec;
+       unsigned int filter_idx;
+       int rc;
+
+       table_id = efx_farch_filter_id_table_id(filter_id);
+       if ((unsigned int)table_id >= EFX_FARCH_FILTER_TABLE_COUNT)
+               return -ENOENT;
+       table = &state->table[table_id];
+
+       filter_idx = efx_farch_filter_id_index(filter_id);
+       if (filter_idx >= table->size)
+               return -ENOENT;
+       spec = &table->spec[filter_idx];
+
+       spin_lock_bh(&efx->filter_lock);
+
+       if (test_bit(filter_idx, table->used_bitmap) &&
+           spec->priority == priority) {
+               efx_farch_filter_to_gen_spec(spec_buf, spec);
+               rc = 0;
+       } else {
+               rc = -ENOENT;
+       }
+
+       spin_unlock_bh(&efx->filter_lock);
+
+       return rc;
+}
+
+static void
+efx_farch_filter_table_clear(struct efx_nic *efx,
+                            enum efx_farch_filter_table_id table_id,
+                            enum efx_filter_priority priority)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       struct efx_farch_filter_table *table = &state->table[table_id];
+       unsigned int filter_idx;
+
+       spin_lock_bh(&efx->filter_lock);
+       for (filter_idx = 0; filter_idx < table->size; ++filter_idx)
+               if (table->spec[filter_idx].priority <= priority)
+                       efx_farch_filter_table_clear_entry(efx, table,
+                                                          filter_idx);
+       spin_unlock_bh(&efx->filter_lock);
+}
+
+void efx_farch_filter_clear_rx(struct efx_nic *efx,
+                              enum efx_filter_priority priority)
+{
+       efx_farch_filter_table_clear(efx, EFX_FARCH_FILTER_TABLE_RX_IP,
+                                    priority);
+       efx_farch_filter_table_clear(efx, EFX_FARCH_FILTER_TABLE_RX_MAC,
+                                    priority);
+}
+
+u32 efx_farch_filter_count_rx_used(struct efx_nic *efx,
+                                  enum efx_filter_priority priority)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       enum efx_farch_filter_table_id table_id;
+       struct efx_farch_filter_table *table;
+       unsigned int filter_idx;
+       u32 count = 0;
+
+       spin_lock_bh(&efx->filter_lock);
+
+       for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
+            table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
+            table_id++) {
+               table = &state->table[table_id];
+               for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
+                       if (test_bit(filter_idx, table->used_bitmap) &&
+                           table->spec[filter_idx].priority == priority)
+                               ++count;
+               }
+       }
+
+       spin_unlock_bh(&efx->filter_lock);
+
+       return count;
+}
+
+s32 efx_farch_filter_get_rx_ids(struct efx_nic *efx,
+                               enum efx_filter_priority priority,
+                               u32 *buf, u32 size)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       enum efx_farch_filter_table_id table_id;
+       struct efx_farch_filter_table *table;
+       unsigned int filter_idx;
+       s32 count = 0;
+
+       spin_lock_bh(&efx->filter_lock);
+
+       for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
+            table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
+            table_id++) {
+               table = &state->table[table_id];
+               for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
+                       if (test_bit(filter_idx, table->used_bitmap) &&
+                           table->spec[filter_idx].priority == priority) {
+                               if (count == size) {
+                                       count = -EMSGSIZE;
+                                       goto out;
+                               }
+                               buf[count++] = efx_farch_filter_make_id(
+                                       &table->spec[filter_idx], filter_idx);
+                       }
+               }
+       }
+out:
+       spin_unlock_bh(&efx->filter_lock);
+
+       return count;
+}
+
+/* Restore filter stater after reset */
+void efx_farch_filter_table_restore(struct efx_nic *efx)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       enum efx_farch_filter_table_id table_id;
+       struct efx_farch_filter_table *table;
+       efx_oword_t filter;
+       unsigned int filter_idx;
+
+       spin_lock_bh(&efx->filter_lock);
+
+       for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
+               table = &state->table[table_id];
+
+               /* Check whether this is a regular register table */
+               if (table->step == 0)
+                       continue;
+
+               for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
+                       if (!test_bit(filter_idx, table->used_bitmap))
+                               continue;
+                       efx_farch_filter_build(&filter, &table->spec[filter_idx]);
+                       efx_writeo(efx, &filter,
+                                  table->offset + table->step * filter_idx);
+               }
+       }
+
+       efx_farch_filter_push_rx_config(efx);
+       efx_farch_filter_push_tx_limits(efx);
+
+       spin_unlock_bh(&efx->filter_lock);
+}
+
+void efx_farch_filter_table_remove(struct efx_nic *efx)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       enum efx_farch_filter_table_id table_id;
+
+       for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
+               kfree(state->table[table_id].used_bitmap);
+               vfree(state->table[table_id].spec);
+       }
+       kfree(state);
+}
+
+int efx_farch_filter_table_probe(struct efx_nic *efx)
+{
+       struct efx_farch_filter_state *state;
+       struct efx_farch_filter_table *table;
+       unsigned table_id;
+
+       state = kzalloc(sizeof(struct efx_farch_filter_state), GFP_KERNEL);
+       if (!state)
+               return -ENOMEM;
+       efx->filter_state = state;
+
+       if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
+               table = &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
+               table->id = EFX_FARCH_FILTER_TABLE_RX_IP;
+               table->offset = FR_BZ_RX_FILTER_TBL0;
+               table->size = FR_BZ_RX_FILTER_TBL0_ROWS;
+               table->step = FR_BZ_RX_FILTER_TBL0_STEP;
+       }
+
+       if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
+               table = &state->table[EFX_FARCH_FILTER_TABLE_RX_MAC];
+               table->id = EFX_FARCH_FILTER_TABLE_RX_MAC;
+               table->offset = FR_CZ_RX_MAC_FILTER_TBL0;
+               table->size = FR_CZ_RX_MAC_FILTER_TBL0_ROWS;
+               table->step = FR_CZ_RX_MAC_FILTER_TBL0_STEP;
+
+               table = &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
+               table->id = EFX_FARCH_FILTER_TABLE_RX_DEF;
+               table->size = EFX_FARCH_FILTER_SIZE_RX_DEF;
+
+               table = &state->table[EFX_FARCH_FILTER_TABLE_TX_MAC];
+               table->id = EFX_FARCH_FILTER_TABLE_TX_MAC;
+               table->offset = FR_CZ_TX_MAC_FILTER_TBL0;
+               table->size = FR_CZ_TX_MAC_FILTER_TBL0_ROWS;
+               table->step = FR_CZ_TX_MAC_FILTER_TBL0_STEP;
+       }
+
+       for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
+               table = &state->table[table_id];
+               if (table->size == 0)
+                       continue;
+               table->used_bitmap = kcalloc(BITS_TO_LONGS(table->size),
+                                            sizeof(unsigned long),
+                                            GFP_KERNEL);
+               if (!table->used_bitmap)
+                       goto fail;
+               table->spec = vzalloc(table->size * sizeof(*table->spec));
+               if (!table->spec)
+                       goto fail;
+       }
+
+       if (state->table[EFX_FARCH_FILTER_TABLE_RX_DEF].size) {
+               /* RX default filters must always exist */
+               unsigned i;
+               for (i = 0; i < EFX_FARCH_FILTER_SIZE_RX_DEF; i++)
+                       efx_farch_filter_reset_rx_def(efx, i);
+       }
+
+       efx_farch_filter_push_rx_config(efx);
+
+       return 0;
+
+fail:
+       efx_farch_filter_table_remove(efx);
+       return -ENOMEM;
+}
+
+/* Update scatter enable flags for filters pointing to our own RX queues */
+void efx_farch_filter_update_rx_scatter(struct efx_nic *efx)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       enum efx_farch_filter_table_id table_id;
+       struct efx_farch_filter_table *table;
+       efx_oword_t filter;
+       unsigned int filter_idx;
+
+       spin_lock_bh(&efx->filter_lock);
+
+       for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
+            table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
+            table_id++) {
+               table = &state->table[table_id];
+
+               for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
+                       if (!test_bit(filter_idx, table->used_bitmap) ||
+                           table->spec[filter_idx].dmaq_id >=
+                           efx->n_rx_channels)
+                               continue;
+
+                       if (efx->rx_scatter)
+                               table->spec[filter_idx].flags |=
+                                       EFX_FILTER_FLAG_RX_SCATTER;
+                       else
+                               table->spec[filter_idx].flags &=
+                                       ~EFX_FILTER_FLAG_RX_SCATTER;
+
+                       if (table_id == EFX_FARCH_FILTER_TABLE_RX_DEF)
+                               /* Pushed by efx_farch_filter_push_rx_config() */
+                               continue;
+
+                       efx_farch_filter_build(&filter, &table->spec[filter_idx]);
+                       efx_writeo(efx, &filter,
+                                  table->offset + table->step * filter_idx);
+               }
+       }
+
+       efx_farch_filter_push_rx_config(efx);
+
+       spin_unlock_bh(&efx->filter_lock);
+}
+
+#ifdef CONFIG_RFS_ACCEL
+
+s32 efx_farch_filter_rfs_insert(struct efx_nic *efx,
+                               struct efx_filter_spec *gen_spec)
+{
+       return efx_farch_filter_insert(efx, gen_spec, true);
+}
+
+bool efx_farch_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
+                                    unsigned int index)
+{
+       struct efx_farch_filter_state *state = efx->filter_state;
+       struct efx_farch_filter_table *table =
+               &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
+
+       if (test_bit(index, table->used_bitmap) &&
+           table->spec[index].priority == EFX_FILTER_PRI_HINT &&
+           rps_may_expire_flow(efx->net_dev, table->spec[index].dmaq_id,
+                               flow_id, index)) {
+               efx_farch_filter_table_clear_entry(efx, table, index);
+               return true;
+       }
+
+       return false;
+}
+
+#endif /* CONFIG_RFS_ACCEL */
diff --git a/drivers/net/ethernet/sfc/filter.c b/drivers/net/ethernet/sfc/filter.c
deleted file mode 100644 (file)
index ad66376..0000000
+++ /dev/null
@@ -1,1244 +0,0 @@
-/****************************************************************************
- * Driver for Solarflare Solarstorm network controllers and boards
- * Copyright 2005-2010 Solarflare Communications Inc.
- *
- * This program is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License version 2 as published
- * by the Free Software Foundation, incorporated herein by reference.
- */
-
-#include <linux/in.h>
-#include <net/ip.h>
-#include "efx.h"
-#include "filter.h"
-#include "io.h"
-#include "nic.h"
-#include "farch_regs.h"
-
-/* "Fudge factors" - difference between programmed value and actual depth.
- * Due to pipelined implementation we need to program H/W with a value that
- * is larger than the hop limit we want.
- */
-#define EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD 3
-#define EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL 1
-
-/* Hard maximum search limit.  Hardware will time-out beyond 200-something.
- * We also need to avoid infinite loops in efx_farch_filter_search() when the
- * table is full.
- */
-#define EFX_FARCH_FILTER_CTL_SRCH_MAX 200
-
-/* Don't try very hard to find space for performance hints, as this is
- * counter-productive. */
-#define EFX_FARCH_FILTER_CTL_SRCH_HINT_MAX 5
-
-enum efx_farch_filter_type {
-       EFX_FARCH_FILTER_TCP_FULL = 0,
-       EFX_FARCH_FILTER_TCP_WILD,
-       EFX_FARCH_FILTER_UDP_FULL,
-       EFX_FARCH_FILTER_UDP_WILD,
-       EFX_FARCH_FILTER_MAC_FULL = 4,
-       EFX_FARCH_FILTER_MAC_WILD,
-       EFX_FARCH_FILTER_UC_DEF = 8,
-       EFX_FARCH_FILTER_MC_DEF,
-       EFX_FARCH_FILTER_TYPE_COUNT,            /* number of specific types */
-};
-
-enum efx_farch_filter_table_id {
-       EFX_FARCH_FILTER_TABLE_RX_IP = 0,
-       EFX_FARCH_FILTER_TABLE_RX_MAC,
-       EFX_FARCH_FILTER_TABLE_RX_DEF,
-       EFX_FARCH_FILTER_TABLE_TX_MAC,
-       EFX_FARCH_FILTER_TABLE_COUNT,
-};
-
-enum efx_farch_filter_index {
-       EFX_FARCH_FILTER_INDEX_UC_DEF,
-       EFX_FARCH_FILTER_INDEX_MC_DEF,
-       EFX_FARCH_FILTER_SIZE_RX_DEF,
-};
-
-struct efx_farch_filter_spec {
-       u8      type:4;
-       u8      priority:4;
-       u8      flags;
-       u16     dmaq_id;
-       u32     data[3];
-};
-
-struct efx_farch_filter_table {
-       enum efx_farch_filter_table_id id;
-       u32             offset;         /* address of table relative to BAR */
-       unsigned        size;           /* number of entries */
-       unsigned        step;           /* step between entries */
-       unsigned        used;           /* number currently used */
-       unsigned long   *used_bitmap;
-       struct efx_farch_filter_spec *spec;
-       unsigned        search_limit[EFX_FARCH_FILTER_TYPE_COUNT];
-};
-
-struct efx_farch_filter_state {
-       struct efx_farch_filter_table table[EFX_FARCH_FILTER_TABLE_COUNT];
-};
-
-static void
-efx_farch_filter_table_clear_entry(struct efx_nic *efx,
-                                  struct efx_farch_filter_table *table,
-                                  unsigned int filter_idx);
-
-/* The filter hash function is LFSR polynomial x^16 + x^3 + 1 of a 32-bit
- * key derived from the n-tuple.  The initial LFSR state is 0xffff. */
-static u16 efx_farch_filter_hash(u32 key)
-{
-       u16 tmp;
-
-       /* First 16 rounds */
-       tmp = 0x1fff ^ key >> 16;
-       tmp = tmp ^ tmp >> 3 ^ tmp >> 6;
-       tmp = tmp ^ tmp >> 9;
-       /* Last 16 rounds */
-       tmp = tmp ^ tmp << 13 ^ key;
-       tmp = tmp ^ tmp >> 3 ^ tmp >> 6;
-       return tmp ^ tmp >> 9;
-}
-
-/* To allow for hash collisions, filter search continues at these
- * increments from the first possible entry selected by the hash. */
-static u16 efx_farch_filter_increment(u32 key)
-{
-       return key * 2 - 1;
-}
-
-static enum efx_farch_filter_table_id
-efx_farch_filter_spec_table_id(const struct efx_farch_filter_spec *spec)
-{
-       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
-                    (EFX_FARCH_FILTER_TCP_FULL >> 2));
-       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
-                    (EFX_FARCH_FILTER_TCP_WILD >> 2));
-       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
-                    (EFX_FARCH_FILTER_UDP_FULL >> 2));
-       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_IP !=
-                    (EFX_FARCH_FILTER_UDP_WILD >> 2));
-       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_MAC !=
-                    (EFX_FARCH_FILTER_MAC_FULL >> 2));
-       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_RX_MAC !=
-                    (EFX_FARCH_FILTER_MAC_WILD >> 2));
-       BUILD_BUG_ON(EFX_FARCH_FILTER_TABLE_TX_MAC !=
-                    EFX_FARCH_FILTER_TABLE_RX_MAC + 2);
-       return (spec->type >> 2) + ((spec->flags & EFX_FILTER_FLAG_TX) ? 2 : 0);
-}
-
-static void efx_farch_filter_push_rx_config(struct efx_nic *efx)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       struct efx_farch_filter_table *table;
-       efx_oword_t filter_ctl;
-
-       efx_reado(efx, &filter_ctl, FR_BZ_RX_FILTER_CTL);
-
-       table = &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
-       EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_TCP_FULL_SRCH_LIMIT,
-                           table->search_limit[EFX_FARCH_FILTER_TCP_FULL] +
-                           EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
-       EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_TCP_WILD_SRCH_LIMIT,
-                           table->search_limit[EFX_FARCH_FILTER_TCP_WILD] +
-                           EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
-       EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_UDP_FULL_SRCH_LIMIT,
-                           table->search_limit[EFX_FARCH_FILTER_UDP_FULL] +
-                           EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
-       EFX_SET_OWORD_FIELD(filter_ctl, FRF_BZ_UDP_WILD_SRCH_LIMIT,
-                           table->search_limit[EFX_FARCH_FILTER_UDP_WILD] +
-                           EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
-
-       table = &state->table[EFX_FARCH_FILTER_TABLE_RX_MAC];
-       if (table->size) {
-               EFX_SET_OWORD_FIELD(
-                       filter_ctl, FRF_CZ_ETHERNET_FULL_SEARCH_LIMIT,
-                       table->search_limit[EFX_FARCH_FILTER_MAC_FULL] +
-                       EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
-               EFX_SET_OWORD_FIELD(
-                       filter_ctl, FRF_CZ_ETHERNET_WILDCARD_SEARCH_LIMIT,
-                       table->search_limit[EFX_FARCH_FILTER_MAC_WILD] +
-                       EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
-       }
-
-       table = &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
-       if (table->size) {
-               EFX_SET_OWORD_FIELD(
-                       filter_ctl, FRF_CZ_UNICAST_NOMATCH_Q_ID,
-                       table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].dmaq_id);
-               EFX_SET_OWORD_FIELD(
-                       filter_ctl, FRF_CZ_UNICAST_NOMATCH_RSS_ENABLED,
-                       !!(table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].flags &
-                          EFX_FILTER_FLAG_RX_RSS));
-               EFX_SET_OWORD_FIELD(
-                       filter_ctl, FRF_CZ_MULTICAST_NOMATCH_Q_ID,
-                       table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].dmaq_id);
-               EFX_SET_OWORD_FIELD(
-                       filter_ctl, FRF_CZ_MULTICAST_NOMATCH_RSS_ENABLED,
-                       !!(table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].flags &
-                          EFX_FILTER_FLAG_RX_RSS));
-
-               /* There is a single bit to enable RX scatter for all
-                * unmatched packets.  Only set it if scatter is
-                * enabled in both filter specs.
-                */
-               EFX_SET_OWORD_FIELD(
-                       filter_ctl, FRF_BZ_SCATTER_ENBL_NO_MATCH_Q,
-                       !!(table->spec[EFX_FARCH_FILTER_INDEX_UC_DEF].flags &
-                          table->spec[EFX_FARCH_FILTER_INDEX_MC_DEF].flags &
-                          EFX_FILTER_FLAG_RX_SCATTER));
-       } else if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
-               /* We don't expose 'default' filters because unmatched
-                * packets always go to the queue number found in the
-                * RSS table.  But we still need to set the RX scatter
-                * bit here.
-                */
-               EFX_SET_OWORD_FIELD(
-                       filter_ctl, FRF_BZ_SCATTER_ENBL_NO_MATCH_Q,
-                       efx->rx_scatter);
-       }
-
-       efx_writeo(efx, &filter_ctl, FR_BZ_RX_FILTER_CTL);
-}
-
-static void efx_farch_filter_push_tx_limits(struct efx_nic *efx)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       struct efx_farch_filter_table *table;
-       efx_oword_t tx_cfg;
-
-       efx_reado(efx, &tx_cfg, FR_AZ_TX_CFG);
-
-       table = &state->table[EFX_FARCH_FILTER_TABLE_TX_MAC];
-       if (table->size) {
-               EFX_SET_OWORD_FIELD(
-                       tx_cfg, FRF_CZ_TX_ETH_FILTER_FULL_SEARCH_RANGE,
-                       table->search_limit[EFX_FARCH_FILTER_MAC_FULL] +
-                       EFX_FARCH_FILTER_CTL_SRCH_FUDGE_FULL);
-               EFX_SET_OWORD_FIELD(
-                       tx_cfg, FRF_CZ_TX_ETH_FILTER_WILD_SEARCH_RANGE,
-                       table->search_limit[EFX_FARCH_FILTER_MAC_WILD] +
-                       EFX_FARCH_FILTER_CTL_SRCH_FUDGE_WILD);
-       }
-
-       efx_writeo(efx, &tx_cfg, FR_AZ_TX_CFG);
-}
-
-static int
-efx_farch_filter_from_gen_spec(struct efx_farch_filter_spec *spec,
-                              const struct efx_filter_spec *gen_spec)
-{
-       bool is_full = false;
-
-       if ((gen_spec->flags & EFX_FILTER_FLAG_RX_RSS) &&
-           gen_spec->rss_context != EFX_FILTER_RSS_CONTEXT_DEFAULT)
-               return -EINVAL;
-
-       spec->priority = gen_spec->priority;
-       spec->flags = gen_spec->flags;
-       spec->dmaq_id = gen_spec->dmaq_id;
-
-       switch (gen_spec->match_flags) {
-       case (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_IP_PROTO |
-             EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT |
-             EFX_FILTER_MATCH_REM_HOST | EFX_FILTER_MATCH_REM_PORT):
-               is_full = true;
-               /* fall through */
-       case (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_IP_PROTO |
-             EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT): {
-               __be32 rhost, host1, host2;
-               __be16 rport, port1, port2;
-
-               EFX_BUG_ON_PARANOID(!(gen_spec->flags & EFX_FILTER_FLAG_RX));
-
-               if (gen_spec->ether_type != htons(ETH_P_IP))
-                       return -EPROTONOSUPPORT;
-               if (gen_spec->loc_port == 0 ||
-                   (is_full && gen_spec->rem_port == 0))
-                       return -EADDRNOTAVAIL;
-               switch (gen_spec->ip_proto) {
-               case IPPROTO_TCP:
-                       spec->type = (is_full ? EFX_FARCH_FILTER_TCP_FULL :
-                                     EFX_FARCH_FILTER_TCP_WILD);
-                       break;
-               case IPPROTO_UDP:
-                       spec->type = (is_full ? EFX_FARCH_FILTER_UDP_FULL :
-                                     EFX_FARCH_FILTER_UDP_WILD);
-                       break;
-               default:
-                       return -EPROTONOSUPPORT;
-               }
-
-               /* Filter is constructed in terms of source and destination,
-                * with the odd wrinkle that the ports are swapped in a UDP
-                * wildcard filter.  We need to convert from local and remote
-                * (= zero for wildcard) addresses.
-                */
-               rhost = is_full ? gen_spec->rem_host[0] : 0;
-               rport = is_full ? gen_spec->rem_port : 0;
-               host1 = rhost;
-               host2 = gen_spec->loc_host[0];
-               if (!is_full && gen_spec->ip_proto == IPPROTO_UDP) {
-                       port1 = gen_spec->loc_port;
-                       port2 = rport;
-               } else {
-                       port1 = rport;
-                       port2 = gen_spec->loc_port;
-               }
-               spec->data[0] = ntohl(host1) << 16 | ntohs(port1);
-               spec->data[1] = ntohs(port2) << 16 | ntohl(host1) >> 16;
-               spec->data[2] = ntohl(host2);
-
-               break;
-       }
-
-       case EFX_FILTER_MATCH_LOC_MAC | EFX_FILTER_MATCH_OUTER_VID:
-               is_full = true;
-               /* fall through */
-       case EFX_FILTER_MATCH_LOC_MAC:
-               spec->type = (is_full ? EFX_FARCH_FILTER_MAC_FULL :
-                             EFX_FARCH_FILTER_MAC_WILD);
-               spec->data[0] = is_full ? ntohs(gen_spec->outer_vid) : 0;
-               spec->data[1] = (gen_spec->loc_mac[2] << 24 |
-                                gen_spec->loc_mac[3] << 16 |
-                                gen_spec->loc_mac[4] << 8 |
-                                gen_spec->loc_mac[5]);
-               spec->data[2] = (gen_spec->loc_mac[0] << 8 |
-                                gen_spec->loc_mac[1]);
-               break;
-
-       case EFX_FILTER_MATCH_LOC_MAC_IG:
-               spec->type = (is_multicast_ether_addr(gen_spec->loc_mac) ?
-                             EFX_FARCH_FILTER_MC_DEF :
-                             EFX_FARCH_FILTER_UC_DEF);
-               memset(spec->data, 0, sizeof(spec->data)); /* ensure equality */
-               break;
-
-       default:
-               return -EPROTONOSUPPORT;
-       }
-
-       return 0;
-}
-
-static void
-efx_farch_filter_to_gen_spec(struct efx_filter_spec *gen_spec,
-                            const struct efx_farch_filter_spec *spec)
-{
-       bool is_full = false;
-
-       /* *gen_spec should be completely initialised, to be consistent
-        * with efx_filter_init_{rx,tx}() and in case we want to copy
-        * it back to userland.
-        */
-       memset(gen_spec, 0, sizeof(*gen_spec));
-
-       gen_spec->priority = spec->priority;
-       gen_spec->flags = spec->flags;
-       gen_spec->dmaq_id = spec->dmaq_id;
-
-       switch (spec->type) {
-       case EFX_FARCH_FILTER_TCP_FULL:
-       case EFX_FARCH_FILTER_UDP_FULL:
-               is_full = true;
-               /* fall through */
-       case EFX_FARCH_FILTER_TCP_WILD:
-       case EFX_FARCH_FILTER_UDP_WILD: {
-               __be32 host1, host2;
-               __be16 port1, port2;
-
-               gen_spec->match_flags =
-                       EFX_FILTER_MATCH_ETHER_TYPE |
-                       EFX_FILTER_MATCH_IP_PROTO |
-                       EFX_FILTER_MATCH_LOC_HOST | EFX_FILTER_MATCH_LOC_PORT;
-               if (is_full)
-                       gen_spec->match_flags |= (EFX_FILTER_MATCH_REM_HOST |
-                                                 EFX_FILTER_MATCH_REM_PORT);
-               gen_spec->ether_type = htons(ETH_P_IP);
-               gen_spec->ip_proto =
-                       (spec->type == EFX_FARCH_FILTER_TCP_FULL ||
-                        spec->type == EFX_FARCH_FILTER_TCP_WILD) ?
-                       IPPROTO_TCP : IPPROTO_UDP;
-
-               host1 = htonl(spec->data[0] >> 16 | spec->data[1] << 16);
-               port1 = htons(spec->data[0]);
-               host2 = htonl(spec->data[2]);
-               port2 = htons(spec->data[1] >> 16);
-               if (spec->flags & EFX_FILTER_FLAG_TX) {
-                       gen_spec->loc_host[0] = host1;
-                       gen_spec->rem_host[0] = host2;
-               } else {
-                       gen_spec->loc_host[0] = host2;
-                       gen_spec->rem_host[0] = host1;
-               }
-               if (!!(gen_spec->flags & EFX_FILTER_FLAG_TX) ^
-                   (!is_full && gen_spec->ip_proto == IPPROTO_UDP)) {
-                       gen_spec->loc_port = port1;
-                       gen_spec->rem_port = port2;
-               } else {
-                       gen_spec->loc_port = port2;
-                       gen_spec->rem_port = port1;
-               }
-
-               break;
-       }
-
-       case EFX_FARCH_FILTER_MAC_FULL:
-               is_full = true;
-               /* fall through */
-       case EFX_FARCH_FILTER_MAC_WILD:
-               gen_spec->match_flags = EFX_FILTER_MATCH_LOC_MAC;
-               if (is_full)
-                       gen_spec->match_flags |= EFX_FILTER_MATCH_OUTER_VID;
-               gen_spec->loc_mac[0] = spec->data[2] >> 8;
-               gen_spec->loc_mac[1] = spec->data[2];
-               gen_spec->loc_mac[2] = spec->data[1] >> 24;
-               gen_spec->loc_mac[3] = spec->data[1] >> 16;
-               gen_spec->loc_mac[4] = spec->data[1] >> 8;
-               gen_spec->loc_mac[5] = spec->data[1];
-               gen_spec->outer_vid = htons(spec->data[0]);
-               break;
-
-       case EFX_FARCH_FILTER_UC_DEF:
-       case EFX_FARCH_FILTER_MC_DEF:
-               gen_spec->match_flags = EFX_FILTER_MATCH_LOC_MAC_IG;
-               gen_spec->loc_mac[0] = spec->type == EFX_FARCH_FILTER_MC_DEF;
-               break;
-
-       default:
-               WARN_ON(1);
-               break;
-       }
-}
-
-static void
-efx_farch_filter_reset_rx_def(struct efx_nic *efx, unsigned filter_idx)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       struct efx_farch_filter_table *table =
-               &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
-       struct efx_farch_filter_spec *spec = &table->spec[filter_idx];
-
-       /* If there's only one channel then disable RSS for non VF
-        * traffic, thereby allowing VFs to use RSS when the PF can't.
-        */
-       spec->type = EFX_FARCH_FILTER_UC_DEF + filter_idx;
-       spec->priority = EFX_FILTER_PRI_MANUAL;
-       spec->flags = (EFX_FILTER_FLAG_RX |
-                      (efx->n_rx_channels > 1 ? EFX_FILTER_FLAG_RX_RSS : 0) |
-                      (efx->rx_scatter ? EFX_FILTER_FLAG_RX_SCATTER : 0));
-       spec->dmaq_id = 0;
-       table->used_bitmap[0] |= 1 << filter_idx;
-}
-
-/* Build a filter entry and return its n-tuple key. */
-static u32 efx_farch_filter_build(efx_oword_t *filter,
-                                 struct efx_farch_filter_spec *spec)
-{
-       u32 data3;
-
-       switch (efx_farch_filter_spec_table_id(spec)) {
-       case EFX_FARCH_FILTER_TABLE_RX_IP: {
-               bool is_udp = (spec->type == EFX_FARCH_FILTER_UDP_FULL ||
-                              spec->type == EFX_FARCH_FILTER_UDP_WILD);
-               EFX_POPULATE_OWORD_7(
-                       *filter,
-                       FRF_BZ_RSS_EN,
-                       !!(spec->flags & EFX_FILTER_FLAG_RX_RSS),
-                       FRF_BZ_SCATTER_EN,
-                       !!(spec->flags & EFX_FILTER_FLAG_RX_SCATTER),
-                       FRF_BZ_TCP_UDP, is_udp,
-                       FRF_BZ_RXQ_ID, spec->dmaq_id,
-                       EFX_DWORD_2, spec->data[2],
-                       EFX_DWORD_1, spec->data[1],
-                       EFX_DWORD_0, spec->data[0]);
-               data3 = is_udp;
-               break;
-       }
-
-       case EFX_FARCH_FILTER_TABLE_RX_MAC: {
-               bool is_wild = spec->type == EFX_FARCH_FILTER_MAC_WILD;
-               EFX_POPULATE_OWORD_7(
-                       *filter,
-                       FRF_CZ_RMFT_RSS_EN,
-                       !!(spec->flags & EFX_FILTER_FLAG_RX_RSS),
-                       FRF_CZ_RMFT_SCATTER_EN,
-                       !!(spec->flags & EFX_FILTER_FLAG_RX_SCATTER),
-                       FRF_CZ_RMFT_RXQ_ID, spec->dmaq_id,
-                       FRF_CZ_RMFT_WILDCARD_MATCH, is_wild,
-                       FRF_CZ_RMFT_DEST_MAC_HI, spec->data[2],
-                       FRF_CZ_RMFT_DEST_MAC_LO, spec->data[1],
-                       FRF_CZ_RMFT_VLAN_ID, spec->data[0]);
-               data3 = is_wild;
-               break;
-       }
-
-       case EFX_FARCH_FILTER_TABLE_TX_MAC: {
-               bool is_wild = spec->type == EFX_FARCH_FILTER_MAC_WILD;
-               EFX_POPULATE_OWORD_5(*filter,
-                                    FRF_CZ_TMFT_TXQ_ID, spec->dmaq_id,
-                                    FRF_CZ_TMFT_WILDCARD_MATCH, is_wild,
-                                    FRF_CZ_TMFT_SRC_MAC_HI, spec->data[2],
-                                    FRF_CZ_TMFT_SRC_MAC_LO, spec->data[1],
-                                    FRF_CZ_TMFT_VLAN_ID, spec->data[0]);
-               data3 = is_wild | spec->dmaq_id << 1;
-               break;
-       }
-
-       default:
-               BUG();
-       }
-
-       return spec->data[0] ^ spec->data[1] ^ spec->data[2] ^ data3;
-}
-
-static bool efx_farch_filter_equal(const struct efx_farch_filter_spec *left,
-                                  const struct efx_farch_filter_spec *right)
-{
-       if (left->type != right->type ||
-           memcmp(left->data, right->data, sizeof(left->data)))
-               return false;
-
-       if (left->flags & EFX_FILTER_FLAG_TX &&
-           left->dmaq_id != right->dmaq_id)
-               return false;
-
-       return true;
-}
-
-/*
- * Construct/deconstruct external filter IDs.  At least the RX filter
- * IDs must be ordered by matching priority, for RX NFC semantics.
- *
- * Deconstruction needs to be robust against invalid IDs so that
- * efx_filter_remove_id_safe() and efx_filter_get_filter_safe() can
- * accept user-provided IDs.
- */
-
-#define EFX_FARCH_FILTER_MATCH_PRI_COUNT       5
-
-static const u8 efx_farch_filter_type_match_pri[EFX_FARCH_FILTER_TYPE_COUNT] = {
-       [EFX_FARCH_FILTER_TCP_FULL]     = 0,
-       [EFX_FARCH_FILTER_UDP_FULL]     = 0,
-       [EFX_FARCH_FILTER_TCP_WILD]     = 1,
-       [EFX_FARCH_FILTER_UDP_WILD]     = 1,
-       [EFX_FARCH_FILTER_MAC_FULL]     = 2,
-       [EFX_FARCH_FILTER_MAC_WILD]     = 3,
-       [EFX_FARCH_FILTER_UC_DEF]       = 4,
-       [EFX_FARCH_FILTER_MC_DEF]       = 4,
-};
-
-static const enum efx_farch_filter_table_id efx_farch_filter_range_table[] = {
-       EFX_FARCH_FILTER_TABLE_RX_IP,   /* RX match pri 0 */
-       EFX_FARCH_FILTER_TABLE_RX_IP,
-       EFX_FARCH_FILTER_TABLE_RX_MAC,
-       EFX_FARCH_FILTER_TABLE_RX_MAC,
-       EFX_FARCH_FILTER_TABLE_RX_DEF,  /* RX match pri 4 */
-       EFX_FARCH_FILTER_TABLE_TX_MAC,  /* TX match pri 0 */
-       EFX_FARCH_FILTER_TABLE_TX_MAC,  /* TX match pri 1 */
-};
-
-#define EFX_FARCH_FILTER_INDEX_WIDTH 13
-#define EFX_FARCH_FILTER_INDEX_MASK ((1 << EFX_FARCH_FILTER_INDEX_WIDTH) - 1)
-
-static inline u32
-efx_farch_filter_make_id(const struct efx_farch_filter_spec *spec,
-                        unsigned int index)
-{
-       unsigned int range;
-
-       range = efx_farch_filter_type_match_pri[spec->type];
-       if (!(spec->flags & EFX_FILTER_FLAG_RX))
-               range += EFX_FARCH_FILTER_MATCH_PRI_COUNT;
-
-       return range << EFX_FARCH_FILTER_INDEX_WIDTH | index;
-}
-
-static inline enum efx_farch_filter_table_id
-efx_farch_filter_id_table_id(u32 id)
-{
-       unsigned int range = id >> EFX_FARCH_FILTER_INDEX_WIDTH;
-
-       if (range < ARRAY_SIZE(efx_farch_filter_range_table))
-               return efx_farch_filter_range_table[range];
-       else
-               return EFX_FARCH_FILTER_TABLE_COUNT; /* invalid */
-}
-
-static inline unsigned int efx_farch_filter_id_index(u32 id)
-{
-       return id & EFX_FARCH_FILTER_INDEX_MASK;
-}
-
-u32 efx_filter_get_rx_id_limit(struct efx_nic *efx)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       unsigned int range = EFX_FARCH_FILTER_MATCH_PRI_COUNT - 1;
-       enum efx_farch_filter_table_id table_id;
-
-       do {
-               table_id = efx_farch_filter_range_table[range];
-               if (state->table[table_id].size != 0)
-                       return range << EFX_FARCH_FILTER_INDEX_WIDTH |
-                               state->table[table_id].size;
-       } while (range--);
-
-       return 0;
-}
-
-/**
- * efx_filter_insert_filter - add or replace a filter
- * @efx: NIC in which to insert the filter
- * @spec: Specification for the filter
- * @replace_equal: Flag for whether the specified filter may replace an
- *     existing filter with equal priority
- *
- * On success, return the filter ID.
- * On failure, return a negative error code.
- *
- * If an existing filter has equal match values to the new filter
- * spec, then the new filter might replace it, depending on the
- * relative priorities.  If the existing filter has lower priority, or
- * if @replace_equal is set and it has equal priority, then it is
- * replaced.  Otherwise the function fails, returning -%EPERM if
- * the existing filter has higher priority or -%EEXIST if it has
- * equal priority.
- */
-s32 efx_filter_insert_filter(struct efx_nic *efx,
-                            struct efx_filter_spec *gen_spec,
-                            bool replace_equal)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       struct efx_farch_filter_table *table;
-       struct efx_farch_filter_spec spec;
-       efx_oword_t filter;
-       int rep_index, ins_index;
-       unsigned int depth = 0;
-       int rc;
-
-       rc = efx_farch_filter_from_gen_spec(&spec, gen_spec);
-       if (rc)
-               return rc;
-
-       table = &state->table[efx_farch_filter_spec_table_id(&spec)];
-       if (table->size == 0)
-               return -EINVAL;
-
-       netif_vdbg(efx, hw, efx->net_dev,
-                  "%s: type %d search_limit=%d", __func__, spec.type,
-                  table->search_limit[spec.type]);
-
-       if (table->id == EFX_FARCH_FILTER_TABLE_RX_DEF) {
-               /* One filter spec per type */
-               BUILD_BUG_ON(EFX_FARCH_FILTER_INDEX_UC_DEF != 0);
-               BUILD_BUG_ON(EFX_FARCH_FILTER_INDEX_MC_DEF !=
-                            EFX_FARCH_FILTER_MC_DEF - EFX_FARCH_FILTER_UC_DEF);
-               rep_index = spec.type - EFX_FARCH_FILTER_UC_DEF;
-               ins_index = rep_index;
-
-               spin_lock_bh(&efx->filter_lock);
-       } else {
-               /* Search concurrently for
-                * (1) a filter to be replaced (rep_index): any filter
-                *     with the same match values, up to the current
-                *     search depth for this type, and
-                * (2) the insertion point (ins_index): (1) or any
-                *     free slot before it or up to the maximum search
-                *     depth for this priority
-                * We fail if we cannot find (2).
-                *
-                * We can stop once either
-                * (a) we find (1), in which case we have definitely
-                *     found (2) as well; or
-                * (b) we have searched exhaustively for (1), and have
-                *     either found (2) or searched exhaustively for it
-                */
-               u32 key = efx_farch_filter_build(&filter, &spec);
-               unsigned int hash = efx_farch_filter_hash(key);
-               unsigned int incr = efx_farch_filter_increment(key);
-               unsigned int max_rep_depth = table->search_limit[spec.type];
-               unsigned int max_ins_depth =
-                       spec.priority <= EFX_FILTER_PRI_HINT ?
-                       EFX_FARCH_FILTER_CTL_SRCH_HINT_MAX :
-                       EFX_FARCH_FILTER_CTL_SRCH_MAX;
-               unsigned int i = hash & (table->size - 1);
-
-               ins_index = -1;
-               depth = 1;
-
-               spin_lock_bh(&efx->filter_lock);
-
-               for (;;) {
-                       if (!test_bit(i, table->used_bitmap)) {
-                               if (ins_index < 0)
-                                       ins_index = i;
-                       } else if (efx_farch_filter_equal(&spec,
-                                                         &table->spec[i])) {
-                               /* Case (a) */
-                               if (ins_index < 0)
-                                       ins_index = i;
-                               rep_index = i;
-                               break;
-                       }
-
-                       if (depth >= max_rep_depth &&
-                           (ins_index >= 0 || depth >= max_ins_depth)) {
-                               /* Case (b) */
-                               if (ins_index < 0) {
-                                       rc = -EBUSY;
-                                       goto out;
-                               }
-                               rep_index = -1;
-                               break;
-                       }
-
-                       i = (i + incr) & (table->size - 1);
-                       ++depth;
-               }
-       }
-
-       /* If we found a filter to be replaced, check whether we
-        * should do so
-        */
-       if (rep_index >= 0) {
-               struct efx_farch_filter_spec *saved_spec =
-                       &table->spec[rep_index];
-
-               if (spec.priority == saved_spec->priority && !replace_equal) {
-                       rc = -EEXIST;
-                       goto out;
-               }
-               if (spec.priority < saved_spec->priority) {
-                       rc = -EPERM;
-                       goto out;
-               }
-       }
-
-       /* Insert the filter */
-       if (ins_index != rep_index) {
-               __set_bit(ins_index, table->used_bitmap);
-               ++table->used;
-       }
-       table->spec[ins_index] = spec;
-
-       if (table->id == EFX_FARCH_FILTER_TABLE_RX_DEF) {
-               efx_farch_filter_push_rx_config(efx);
-       } else {
-               if (table->search_limit[spec.type] < depth) {
-                       table->search_limit[spec.type] = depth;
-                       if (spec.flags & EFX_FILTER_FLAG_TX)
-                               efx_farch_filter_push_tx_limits(efx);
-                       else
-                               efx_farch_filter_push_rx_config(efx);
-               }
-
-               efx_writeo(efx, &filter,
-                          table->offset + table->step * ins_index);
-
-               /* If we were able to replace a filter by inserting
-                * at a lower depth, clear the replaced filter
-                */
-               if (ins_index != rep_index && rep_index >= 0)
-                       efx_farch_filter_table_clear_entry(efx, table,
-                                                          rep_index);
-       }
-
-       netif_vdbg(efx, hw, efx->net_dev,
-                  "%s: filter type %d index %d rxq %u set",
-                  __func__, spec.type, ins_index, spec.dmaq_id);
-       rc = efx_farch_filter_make_id(&spec, ins_index);
-
-out:
-       spin_unlock_bh(&efx->filter_lock);
-       return rc;
-}
-
-static void
-efx_farch_filter_table_clear_entry(struct efx_nic *efx,
-                                  struct efx_farch_filter_table *table,
-                                  unsigned int filter_idx)
-{
-       static efx_oword_t filter;
-
-       if (table->id == EFX_FARCH_FILTER_TABLE_RX_DEF) {
-               /* RX default filters must always exist */
-               efx_farch_filter_reset_rx_def(efx, filter_idx);
-               efx_farch_filter_push_rx_config(efx);
-       } else if (test_bit(filter_idx, table->used_bitmap)) {
-               __clear_bit(filter_idx, table->used_bitmap);
-               --table->used;
-               memset(&table->spec[filter_idx], 0, sizeof(table->spec[0]));
-
-               efx_writeo(efx, &filter,
-                          table->offset + table->step * filter_idx);
-
-               /* If this filter required a greater search depth than
-                * any other, the search limit for its type can now be
-                * decreased.  However, it is hard to determine that
-                * unless the table has become completely empty - in
-                * which case, all its search limits can be set to 0.
-                */
-               if (unlikely(table->used == 0)) {
-                       memset(table->search_limit, 0,
-                              sizeof(table->search_limit));
-                       if (table->id == EFX_FARCH_FILTER_TABLE_TX_MAC)
-                               efx_farch_filter_push_tx_limits(efx);
-                       else
-                               efx_farch_filter_push_rx_config(efx);
-               }
-       }
-}
-
-/**
- * efx_filter_remove_id_safe - remove a filter by ID, carefully
- * @efx: NIC from which to remove the filter
- * @priority: Priority of filter, as passed to @efx_filter_insert_filter
- * @filter_id: ID of filter, as returned by @efx_filter_insert_filter
- *
- * This function will range-check @filter_id, so it is safe to call
- * with a value passed from userland.
- */
-int efx_filter_remove_id_safe(struct efx_nic *efx,
-                             enum efx_filter_priority priority,
-                             u32 filter_id)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       enum efx_farch_filter_table_id table_id;
-       struct efx_farch_filter_table *table;
-       unsigned int filter_idx;
-       struct efx_farch_filter_spec *spec;
-       int rc;
-
-       table_id = efx_farch_filter_id_table_id(filter_id);
-       if ((unsigned int)table_id >= EFX_FARCH_FILTER_TABLE_COUNT)
-               return -ENOENT;
-       table = &state->table[table_id];
-
-       filter_idx = efx_farch_filter_id_index(filter_id);
-       if (filter_idx >= table->size)
-               return -ENOENT;
-       spec = &table->spec[filter_idx];
-
-       spin_lock_bh(&efx->filter_lock);
-
-       if (test_bit(filter_idx, table->used_bitmap) &&
-           spec->priority == priority) {
-               efx_farch_filter_table_clear_entry(efx, table, filter_idx);
-               rc = 0;
-       } else {
-               rc = -ENOENT;
-       }
-
-       spin_unlock_bh(&efx->filter_lock);
-
-       return rc;
-}
-
-/**
- * efx_filter_get_filter_safe - retrieve a filter by ID, carefully
- * @efx: NIC from which to remove the filter
- * @priority: Priority of filter, as passed to @efx_filter_insert_filter
- * @filter_id: ID of filter, as returned by @efx_filter_insert_filter
- * @spec: Buffer in which to store filter specification
- *
- * This function will range-check @filter_id, so it is safe to call
- * with a value passed from userland.
- */
-int efx_filter_get_filter_safe(struct efx_nic *efx,
-                              enum efx_filter_priority priority,
-                              u32 filter_id, struct efx_filter_spec *spec_buf)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       enum efx_farch_filter_table_id table_id;
-       struct efx_farch_filter_table *table;
-       struct efx_farch_filter_spec *spec;
-       unsigned int filter_idx;
-       int rc;
-
-       table_id = efx_farch_filter_id_table_id(filter_id);
-       if ((unsigned int)table_id >= EFX_FARCH_FILTER_TABLE_COUNT)
-               return -ENOENT;
-       table = &state->table[table_id];
-
-       filter_idx = efx_farch_filter_id_index(filter_id);
-       if (filter_idx >= table->size)
-               return -ENOENT;
-       spec = &table->spec[filter_idx];
-
-       spin_lock_bh(&efx->filter_lock);
-
-       if (test_bit(filter_idx, table->used_bitmap) &&
-           spec->priority == priority) {
-               efx_farch_filter_to_gen_spec(spec_buf, spec);
-               rc = 0;
-       } else {
-               rc = -ENOENT;
-       }
-
-       spin_unlock_bh(&efx->filter_lock);
-
-       return rc;
-}
-
-static void
-efx_farch_filter_table_clear(struct efx_nic *efx,
-                            enum efx_farch_filter_table_id table_id,
-                            enum efx_filter_priority priority)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       struct efx_farch_filter_table *table = &state->table[table_id];
-       unsigned int filter_idx;
-
-       spin_lock_bh(&efx->filter_lock);
-       for (filter_idx = 0; filter_idx < table->size; ++filter_idx)
-               if (table->spec[filter_idx].priority <= priority)
-                       efx_farch_filter_table_clear_entry(efx, table,
-                                                          filter_idx);
-       spin_unlock_bh(&efx->filter_lock);
-}
-
-/**
- * efx_filter_clear_rx - remove RX filters by priority
- * @efx: NIC from which to remove the filters
- * @priority: Maximum priority to remove
- */
-void efx_filter_clear_rx(struct efx_nic *efx, enum efx_filter_priority priority)
-{
-       efx_farch_filter_table_clear(efx, EFX_FARCH_FILTER_TABLE_RX_IP,
-                                    priority);
-       efx_farch_filter_table_clear(efx, EFX_FARCH_FILTER_TABLE_RX_MAC,
-                                    priority);
-}
-
-u32 efx_filter_count_rx_used(struct efx_nic *efx,
-                            enum efx_filter_priority priority)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       enum efx_farch_filter_table_id table_id;
-       struct efx_farch_filter_table *table;
-       unsigned int filter_idx;
-       u32 count = 0;
-
-       spin_lock_bh(&efx->filter_lock);
-
-       for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
-            table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
-            table_id++) {
-               table = &state->table[table_id];
-               for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
-                       if (test_bit(filter_idx, table->used_bitmap) &&
-                           table->spec[filter_idx].priority == priority)
-                               ++count;
-               }
-       }
-
-       spin_unlock_bh(&efx->filter_lock);
-
-       return count;
-}
-
-s32 efx_filter_get_rx_ids(struct efx_nic *efx,
-                         enum efx_filter_priority priority,
-                         u32 *buf, u32 size)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       enum efx_farch_filter_table_id table_id;
-       struct efx_farch_filter_table *table;
-       unsigned int filter_idx;
-       s32 count = 0;
-
-       spin_lock_bh(&efx->filter_lock);
-
-       for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
-            table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
-            table_id++) {
-               table = &state->table[table_id];
-               for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
-                       if (test_bit(filter_idx, table->used_bitmap) &&
-                           table->spec[filter_idx].priority == priority) {
-                               if (count == size) {
-                                       count = -EMSGSIZE;
-                                       goto out;
-                               }
-                               buf[count++] = efx_farch_filter_make_id(
-                                       &table->spec[filter_idx], filter_idx);
-                       }
-               }
-       }
-out:
-       spin_unlock_bh(&efx->filter_lock);
-
-       return count;
-}
-
-/* Restore filter stater after reset */
-void efx_restore_filters(struct efx_nic *efx)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       enum efx_farch_filter_table_id table_id;
-       struct efx_farch_filter_table *table;
-       efx_oword_t filter;
-       unsigned int filter_idx;
-
-       spin_lock_bh(&efx->filter_lock);
-
-       for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
-               table = &state->table[table_id];
-
-               /* Check whether this is a regular register table */
-               if (table->step == 0)
-                       continue;
-
-               for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
-                       if (!test_bit(filter_idx, table->used_bitmap))
-                               continue;
-                       efx_farch_filter_build(&filter, &table->spec[filter_idx]);
-                       efx_writeo(efx, &filter,
-                                  table->offset + table->step * filter_idx);
-               }
-       }
-
-       efx_farch_filter_push_rx_config(efx);
-       efx_farch_filter_push_tx_limits(efx);
-
-       spin_unlock_bh(&efx->filter_lock);
-}
-
-int efx_probe_filters(struct efx_nic *efx)
-{
-       struct efx_farch_filter_state *state;
-       struct efx_farch_filter_table *table;
-       unsigned table_id;
-
-       state = kzalloc(sizeof(struct efx_farch_filter_state), GFP_KERNEL);
-       if (!state)
-               return -ENOMEM;
-       efx->filter_state = state;
-
-       spin_lock_init(&efx->filter_lock);
-
-       if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
-#ifdef CONFIG_RFS_ACCEL
-               efx->rps_flow_id = kcalloc(FR_BZ_RX_FILTER_TBL0_ROWS,
-                                          sizeof(*efx->rps_flow_id),
-                                          GFP_KERNEL);
-               if (!efx->rps_flow_id)
-                       goto fail;
-#endif
-               table = &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
-               table->id = EFX_FARCH_FILTER_TABLE_RX_IP;
-               table->offset = FR_BZ_RX_FILTER_TBL0;
-               table->size = FR_BZ_RX_FILTER_TBL0_ROWS;
-               table->step = FR_BZ_RX_FILTER_TBL0_STEP;
-       }
-
-       if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
-               table = &state->table[EFX_FARCH_FILTER_TABLE_RX_MAC];
-               table->id = EFX_FARCH_FILTER_TABLE_RX_MAC;
-               table->offset = FR_CZ_RX_MAC_FILTER_TBL0;
-               table->size = FR_CZ_RX_MAC_FILTER_TBL0_ROWS;
-               table->step = FR_CZ_RX_MAC_FILTER_TBL0_STEP;
-
-               table = &state->table[EFX_FARCH_FILTER_TABLE_RX_DEF];
-               table->id = EFX_FARCH_FILTER_TABLE_RX_DEF;
-               table->size = EFX_FARCH_FILTER_SIZE_RX_DEF;
-
-               table = &state->table[EFX_FARCH_FILTER_TABLE_TX_MAC];
-               table->id = EFX_FARCH_FILTER_TABLE_TX_MAC;
-               table->offset = FR_CZ_TX_MAC_FILTER_TBL0;
-               table->size = FR_CZ_TX_MAC_FILTER_TBL0_ROWS;
-               table->step = FR_CZ_TX_MAC_FILTER_TBL0_STEP;
-       }
-
-       for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
-               table = &state->table[table_id];
-               if (table->size == 0)
-                       continue;
-               table->used_bitmap = kcalloc(BITS_TO_LONGS(table->size),
-                                            sizeof(unsigned long),
-                                            GFP_KERNEL);
-               if (!table->used_bitmap)
-                       goto fail;
-               table->spec = vzalloc(table->size * sizeof(*table->spec));
-               if (!table->spec)
-                       goto fail;
-       }
-
-       if (state->table[EFX_FARCH_FILTER_TABLE_RX_DEF].size) {
-               /* RX default filters must always exist */
-               unsigned i;
-               for (i = 0; i < EFX_FARCH_FILTER_SIZE_RX_DEF; i++)
-                       efx_farch_filter_reset_rx_def(efx, i);
-       }
-
-       efx_farch_filter_push_rx_config(efx);
-
-       return 0;
-
-fail:
-       efx_remove_filters(efx);
-       return -ENOMEM;
-}
-
-void efx_remove_filters(struct efx_nic *efx)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       enum efx_farch_filter_table_id table_id;
-
-       for (table_id = 0; table_id < EFX_FARCH_FILTER_TABLE_COUNT; table_id++) {
-               kfree(state->table[table_id].used_bitmap);
-               vfree(state->table[table_id].spec);
-       }
-#ifdef CONFIG_RFS_ACCEL
-       kfree(efx->rps_flow_id);
-#endif
-       kfree(state);
-}
-
-/* Update scatter enable flags for filters pointing to our own RX queues */
-void efx_filter_update_rx_scatter(struct efx_nic *efx)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       enum efx_farch_filter_table_id table_id;
-       struct efx_farch_filter_table *table;
-       efx_oword_t filter;
-       unsigned int filter_idx;
-
-       spin_lock_bh(&efx->filter_lock);
-
-       for (table_id = EFX_FARCH_FILTER_TABLE_RX_IP;
-            table_id <= EFX_FARCH_FILTER_TABLE_RX_DEF;
-            table_id++) {
-               table = &state->table[table_id];
-
-               for (filter_idx = 0; filter_idx < table->size; filter_idx++) {
-                       if (!test_bit(filter_idx, table->used_bitmap) ||
-                           table->spec[filter_idx].dmaq_id >=
-                           efx->n_rx_channels)
-                               continue;
-
-                       if (efx->rx_scatter)
-                               table->spec[filter_idx].flags |=
-                                       EFX_FILTER_FLAG_RX_SCATTER;
-                       else
-                               table->spec[filter_idx].flags &=
-                                       ~EFX_FILTER_FLAG_RX_SCATTER;
-
-                       if (table_id == EFX_FARCH_FILTER_TABLE_RX_DEF)
-                               /* Pushed by efx_farch_filter_push_rx_config() */
-                               continue;
-
-                       efx_farch_filter_build(&filter, &table->spec[filter_idx]);
-                       efx_writeo(efx, &filter,
-                                  table->offset + table->step * filter_idx);
-               }
-       }
-
-       efx_farch_filter_push_rx_config(efx);
-
-       spin_unlock_bh(&efx->filter_lock);
-}
-
-#ifdef CONFIG_RFS_ACCEL
-
-int efx_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
-                  u16 rxq_index, u32 flow_id)
-{
-       struct efx_nic *efx = netdev_priv(net_dev);
-       struct efx_channel *channel;
-       struct efx_filter_spec spec;
-       const struct iphdr *ip;
-       const __be16 *ports;
-       int nhoff;
-       int rc;
-
-       nhoff = skb_network_offset(skb);
-
-       if (skb->protocol == htons(ETH_P_8021Q)) {
-               EFX_BUG_ON_PARANOID(skb_headlen(skb) <
-                                   nhoff + sizeof(struct vlan_hdr));
-               if (((const struct vlan_hdr *)skb->data + nhoff)->
-                   h_vlan_encapsulated_proto != htons(ETH_P_IP))
-                       return -EPROTONOSUPPORT;
-
-               /* This is IP over 802.1q VLAN.  We can't filter on the
-                * IP 5-tuple and the vlan together, so just strip the
-                * vlan header and filter on the IP part.
-                */
-               nhoff += sizeof(struct vlan_hdr);
-       } else if (skb->protocol != htons(ETH_P_IP)) {
-               return -EPROTONOSUPPORT;
-       }
-
-       /* RFS must validate the IP header length before calling us */
-       EFX_BUG_ON_PARANOID(skb_headlen(skb) < nhoff + sizeof(*ip));
-       ip = (const struct iphdr *)(skb->data + nhoff);
-       if (ip_is_fragment(ip))
-               return -EPROTONOSUPPORT;
-       EFX_BUG_ON_PARANOID(skb_headlen(skb) < nhoff + 4 * ip->ihl + 4);
-       ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
-
-       efx_filter_init_rx(&spec, EFX_FILTER_PRI_HINT,
-                          efx->rx_scatter ? EFX_FILTER_FLAG_RX_SCATTER : 0,
-                          rxq_index);
-       rc = efx_filter_set_ipv4_full(&spec, ip->protocol,
-                                     ip->daddr, ports[1], ip->saddr, ports[0]);
-       if (rc)
-               return rc;
-
-       rc = efx_filter_insert_filter(efx, &spec, true);
-       if (rc < 0)
-               return rc;
-
-       /* Remember this so we can check whether to expire the filter later */
-       efx->rps_flow_id[rc] = flow_id;
-       channel = efx_get_channel(efx, skb_get_rx_queue(skb));
-       ++channel->rfs_filters_added;
-
-       netif_info(efx, rx_status, efx->net_dev,
-                  "steering %s %pI4:%u:%pI4:%u to queue %u [flow %u filter %d]\n",
-                  (ip->protocol == IPPROTO_TCP) ? "TCP" : "UDP",
-                  &ip->saddr, ntohs(ports[0]), &ip->daddr, ntohs(ports[1]),
-                  rxq_index, flow_id, rc);
-
-       return rc;
-}
-
-bool __efx_filter_rfs_expire(struct efx_nic *efx, unsigned quota)
-{
-       struct efx_farch_filter_state *state = efx->filter_state;
-       struct efx_farch_filter_table *table =
-               &state->table[EFX_FARCH_FILTER_TABLE_RX_IP];
-       unsigned mask = table->size - 1;
-       unsigned index;
-       unsigned stop;
-
-       if (!spin_trylock_bh(&efx->filter_lock))
-               return false;
-
-       index = efx->rps_expire_index;
-       stop = (index + quota) & mask;
-
-       while (index != stop) {
-               if (test_bit(index, table->used_bitmap) &&
-                   table->spec[index].priority == EFX_FILTER_PRI_HINT &&
-                   rps_may_expire_flow(efx->net_dev,
-                                       table->spec[index].dmaq_id,
-                                       efx->rps_flow_id[index], index)) {
-                       netif_info(efx, rx_status, efx->net_dev,
-                                  "expiring filter %d [flow %u]\n",
-                                  index, efx->rps_flow_id[index]);
-                       efx_farch_filter_table_clear_entry(efx, table, index);
-               }
-               index = (index + 1) & mask;
-       }
-
-       efx->rps_expire_index = stop;
-
-       spin_unlock_bh(&efx->filter_lock);
-       return true;
-}
-
-#endif /* CONFIG_RFS_ACCEL */
index 5287a3cd639c25782afd01f11a310920ccc9ad91..d35ce1410376ceaa529171d390f2acd5de288709 100644 (file)
@@ -30,6 +30,7 @@
 
 #include "enum.h"
 #include "bitfield.h"
+#include "filter.h"
 
 /**************************************************************************
  *
@@ -1025,6 +1026,24 @@ static inline unsigned int efx_port_num(struct efx_nic *efx)
  * @ev_process: Process events for a queue, up to the given NAPI quota
  * @ev_read_ack: Acknowledge read events on a queue, rearming its IRQ
  * @ev_test_generate: Generate a test event
+ * @filter_table_probe: Probe filter capabilities and set up filter software state
+ * @filter_table_restore: Restore filters removed from hardware
+ * @filter_table_remove: Remove filters from hardware and tear down software state
+ * @filter_update_rx_scatter: Update filters after change to rx scatter setting
+ * @filter_insert: add or replace a filter
+ * @filter_remove_safe: remove a filter by ID, carefully
+ * @filter_get_safe: retrieve a filter by ID, carefully
+ * @filter_clear_rx: remove RX filters by priority
+ * @filter_count_rx_used: Get the number of filters in use at a given priority
+ * @filter_get_rx_id_limit: Get maximum value of a filter id, plus 1
+ * @filter_get_rx_ids: Get list of RX filters at a given priority
+ * @filter_rfs_insert: Add or replace a filter for RFS.  This must be
+ *     atomic.  The hardware change may be asynchronous but should
+ *     not be delayed for long.  It may fail if this can't be done
+ *     atomically.
+ * @filter_rfs_expire_one: Consider expiring a filter inserted for RFS.
+ *     This must check whether the specified table entry is used by RFS
+ *     and that rps_may_expire_flow() returns true for it.
  * @revision: Hardware architecture revision
  * @txd_ptr_tbl_base: TX descriptor ring base address
  * @rxd_ptr_tbl_base: RX descriptor ring base address
@@ -1102,6 +1121,32 @@ struct efx_nic_type {
        int (*ev_process)(struct efx_channel *channel, int quota);
        void (*ev_read_ack)(struct efx_channel *channel);
        void (*ev_test_generate)(struct efx_channel *channel);
+       int (*filter_table_probe)(struct efx_nic *efx);
+       void (*filter_table_restore)(struct efx_nic *efx);
+       void (*filter_table_remove)(struct efx_nic *efx);
+       void (*filter_update_rx_scatter)(struct efx_nic *efx);
+       s32 (*filter_insert)(struct efx_nic *efx,
+                            struct efx_filter_spec *spec, bool replace);
+       int (*filter_remove_safe)(struct efx_nic *efx,
+                                 enum efx_filter_priority priority,
+                                 u32 filter_id);
+       int (*filter_get_safe)(struct efx_nic *efx,
+                              enum efx_filter_priority priority,
+                              u32 filter_id, struct efx_filter_spec *);
+       void (*filter_clear_rx)(struct efx_nic *efx,
+                               enum efx_filter_priority priority);
+       u32 (*filter_count_rx_used)(struct efx_nic *efx,
+                                   enum efx_filter_priority priority);
+       u32 (*filter_get_rx_id_limit)(struct efx_nic *efx);
+       s32 (*filter_get_rx_ids)(struct efx_nic *efx,
+                                enum efx_filter_priority priority,
+                                u32 *buf, u32 size);
+#ifdef CONFIG_RFS_ACCEL
+       s32 (*filter_rfs_insert)(struct efx_nic *efx,
+                                struct efx_filter_spec *spec);
+       bool (*filter_rfs_expire_one)(struct efx_nic *efx, u32 flow_id,
+                                     unsigned int index);
+#endif
 
        int revision;
        unsigned int txd_ptr_tbl_base;
@@ -1117,6 +1162,7 @@ struct efx_nic_type {
        unsigned int timer_period_max;
        netdev_features_t offload_features;
        int mcdi_max_ver;
+       unsigned int max_rx_ip_filters;
 };
 
 /**************************************************************************
index 25e25b6357982c7089d4ec901601e11154e0dd4d..69298c918e97c5dd92068f575d4305fe66694c2b 100644 (file)
@@ -404,6 +404,34 @@ extern int efx_farch_ev_process(struct efx_channel *channel, int quota);
 extern void efx_farch_ev_read_ack(struct efx_channel *channel);
 extern void efx_farch_ev_test_generate(struct efx_channel *channel);
 
+/* Falcon/Siena filter operations */
+extern int efx_farch_filter_table_probe(struct efx_nic *efx);
+extern void efx_farch_filter_table_restore(struct efx_nic *efx);
+extern void efx_farch_filter_table_remove(struct efx_nic *efx);
+extern void efx_farch_filter_update_rx_scatter(struct efx_nic *efx);
+extern s32 efx_farch_filter_insert(struct efx_nic *efx,
+                                  struct efx_filter_spec *spec, bool replace);
+extern int efx_farch_filter_remove_safe(struct efx_nic *efx,
+                                       enum efx_filter_priority priority,
+                                       u32 filter_id);
+extern int efx_farch_filter_get_safe(struct efx_nic *efx,
+                                    enum efx_filter_priority priority,
+                                    u32 filter_id, struct efx_filter_spec *);
+extern void efx_farch_filter_clear_rx(struct efx_nic *efx,
+                                     enum efx_filter_priority priority);
+extern u32 efx_farch_filter_count_rx_used(struct efx_nic *efx,
+                                         enum efx_filter_priority priority);
+extern u32 efx_farch_filter_get_rx_id_limit(struct efx_nic *efx);
+extern s32 efx_farch_filter_get_rx_ids(struct efx_nic *efx,
+                                      enum efx_filter_priority priority,
+                                      u32 *buf, u32 size);
+#ifdef CONFIG_RFS_ACCEL
+extern s32 efx_farch_filter_rfs_insert(struct efx_nic *efx,
+                                      struct efx_filter_spec *spec);
+extern bool efx_farch_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
+                                           unsigned int index);
+#endif
+
 extern bool efx_nic_event_present(struct efx_channel *channel);
 
 /* Some statistics are computed as A - B where A and B each increase
index f2b78cd2baf80744cc9a984224acc4e9407f2eed..12990929e2745e3fa62da7c5c57e5bd8caba1121 100644 (file)
@@ -21,6 +21,7 @@
 #include <net/checksum.h>
 #include "net_driver.h"
 #include "efx.h"
+#include "filter.h"
 #include "nic.h"
 #include "selftest.h"
 #include "workarounds.h"
@@ -802,3 +803,96 @@ module_param(rx_refill_threshold, uint, 0444);
 MODULE_PARM_DESC(rx_refill_threshold,
                 "RX descriptor ring refill threshold (%)");
 
+#ifdef CONFIG_RFS_ACCEL
+
+int efx_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
+                  u16 rxq_index, u32 flow_id)
+{
+       struct efx_nic *efx = netdev_priv(net_dev);
+       struct efx_channel *channel;
+       struct efx_filter_spec spec;
+       const struct iphdr *ip;
+       const __be16 *ports;
+       int nhoff;
+       int rc;
+
+       nhoff = skb_network_offset(skb);
+
+       if (skb->protocol == htons(ETH_P_8021Q)) {
+               EFX_BUG_ON_PARANOID(skb_headlen(skb) <
+                                   nhoff + sizeof(struct vlan_hdr));
+               if (((const struct vlan_hdr *)skb->data + nhoff)->
+                   h_vlan_encapsulated_proto != htons(ETH_P_IP))
+                       return -EPROTONOSUPPORT;
+
+               /* This is IP over 802.1q VLAN.  We can't filter on the
+                * IP 5-tuple and the vlan together, so just strip the
+                * vlan header and filter on the IP part.
+                */
+               nhoff += sizeof(struct vlan_hdr);
+       } else if (skb->protocol != htons(ETH_P_IP)) {
+               return -EPROTONOSUPPORT;
+       }
+
+       /* RFS must validate the IP header length before calling us */
+       EFX_BUG_ON_PARANOID(skb_headlen(skb) < nhoff + sizeof(*ip));
+       ip = (const struct iphdr *)(skb->data + nhoff);
+       if (ip_is_fragment(ip))
+               return -EPROTONOSUPPORT;
+       EFX_BUG_ON_PARANOID(skb_headlen(skb) < nhoff + 4 * ip->ihl + 4);
+       ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
+
+       efx_filter_init_rx(&spec, EFX_FILTER_PRI_HINT,
+                          efx->rx_scatter ? EFX_FILTER_FLAG_RX_SCATTER : 0,
+                          rxq_index);
+       rc = efx_filter_set_ipv4_full(&spec, ip->protocol,
+                                     ip->daddr, ports[1], ip->saddr, ports[0]);
+       if (rc)
+               return rc;
+
+       rc = efx->type->filter_rfs_insert(efx, &spec);
+       if (rc < 0)
+               return rc;
+
+       /* Remember this so we can check whether to expire the filter later */
+       efx->rps_flow_id[rc] = flow_id;
+       channel = efx_get_channel(efx, skb_get_rx_queue(skb));
+       ++channel->rfs_filters_added;
+
+       netif_info(efx, rx_status, efx->net_dev,
+                  "steering %s %pI4:%u:%pI4:%u to queue %u [flow %u filter %d]\n",
+                  (ip->protocol == IPPROTO_TCP) ? "TCP" : "UDP",
+                  &ip->saddr, ntohs(ports[0]), &ip->daddr, ntohs(ports[1]),
+                  rxq_index, flow_id, rc);
+
+       return rc;
+}
+
+bool __efx_filter_rfs_expire(struct efx_nic *efx, unsigned int quota)
+{
+       bool (*expire_one)(struct efx_nic *efx, u32 flow_id, unsigned int index);
+       unsigned int index, size;
+       u32 flow_id;
+
+       if (!spin_trylock_bh(&efx->filter_lock))
+               return false;
+
+       expire_one = efx->type->filter_rfs_expire_one;
+       index = efx->rps_expire_index;
+       size = efx->type->max_rx_ip_filters;
+       while (quota--) {
+               flow_id = efx->rps_flow_id[index];
+               if (expire_one(efx, flow_id, index))
+                       netif_info(efx, rx_status, efx->net_dev,
+                                  "expired filter %d [flow %u]\n",
+                                  index, flow_id);
+               if (++index == size)
+                       index = 0;
+       }
+       efx->rps_expire_index = index;
+
+       spin_unlock_bh(&efx->filter_lock);
+       return true;
+}
+
+#endif /* CONFIG_RFS_ACCEL */
index 23e573149bd66a76696e0a4967c233647fe8502d..5120cd8f706beb6295894dfac8ee7f65e430450d 100644 (file)
@@ -736,6 +736,21 @@ const struct efx_nic_type siena_a0_nic_type = {
        .ev_process = efx_farch_ev_process,
        .ev_read_ack = efx_farch_ev_read_ack,
        .ev_test_generate = efx_farch_ev_test_generate,
+       .filter_table_probe = efx_farch_filter_table_probe,
+       .filter_table_restore = efx_farch_filter_table_restore,
+       .filter_table_remove = efx_farch_filter_table_remove,
+       .filter_update_rx_scatter = efx_farch_filter_update_rx_scatter,
+       .filter_insert = efx_farch_filter_insert,
+       .filter_remove_safe = efx_farch_filter_remove_safe,
+       .filter_get_safe = efx_farch_filter_get_safe,
+       .filter_clear_rx = efx_farch_filter_clear_rx,
+       .filter_count_rx_used = efx_farch_filter_count_rx_used,
+       .filter_get_rx_id_limit = efx_farch_filter_get_rx_id_limit,
+       .filter_get_rx_ids = efx_farch_filter_get_rx_ids,
+#ifdef CONFIG_RFS_ACCEL
+       .filter_rfs_insert = efx_farch_filter_rfs_insert,
+       .filter_rfs_expire_one = efx_farch_filter_rfs_expire_one,
+#endif
 
        .revision = EFX_REV_SIENA_A0,
        .txd_ptr_tbl_base = FR_BZ_TX_DESC_PTR_TBL,
@@ -752,4 +767,5 @@ const struct efx_nic_type siena_a0_nic_type = {
        .offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
                             NETIF_F_RXHASH | NETIF_F_NTUPLE),
        .mcdi_max_ver = 1,
+       .max_rx_ip_filters = FR_BZ_RX_FILTER_TBL0_ROWS,
 };