]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
net: airoha: Add L2 hw acceleration support
authorLorenzo Bianconi <lorenzo@kernel.org>
Wed, 9 Apr 2025 09:47:15 +0000 (11:47 +0200)
committerJakub Kicinski <kuba@kernel.org>
Sat, 12 Apr 2025 03:15:36 +0000 (20:15 -0700)
Similar to mtk driver, introduce the capability to offload L2 traffic
defining flower rules in the PSE/PPE engine available on EN7581 SoC.
Since the hw always reports L2/L3/L4 flower rules, link all L2 rules
sharing the same L2 info (with different L3/L4 info) in the L2 subflows
list of a given L2 PPE entry.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Michal Kubiak <michal.kubiak@intel.com>
Link: https://patch.msgid.link/20250409-airoha-flowtable-l2b-v2-2-4a1e3935ea92@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/ethernet/airoha/airoha_eth.c
drivers/net/ethernet/airoha/airoha_eth.h
drivers/net/ethernet/airoha/airoha_ppe.c

index d748dc6de92367365db9f9548f9af52a7fdac187..723eba7cfa0cf405dc2ef9248b34bf6af897ceed 100644 (file)
@@ -694,7 +694,7 @@ static int airoha_qdma_rx_process(struct airoha_queue *q, int budget)
 
                reason = FIELD_GET(AIROHA_RXD4_PPE_CPU_REASON, msg1);
                if (reason == PPE_CPU_REASON_HIT_UNBIND_RATE_REACHED)
-                       airoha_ppe_check_skb(eth->ppe, hash);
+                       airoha_ppe_check_skb(eth->ppe, q->skb, hash);
 
                done++;
                napi_gro_receive(&q->napi, q->skb);
index 86e08832246dfdc4fc448a8aeceb7032d633e812..e82abfc1a67bda7f675342327ae07601d81f2b8f 100644 (file)
@@ -431,10 +431,14 @@ enum airoha_flow_entry_type {
 struct airoha_flow_table_entry {
        union {
                struct hlist_node list; /* PPE L3 flow entry */
-               struct rhash_head l2_node; /* L2 flow entry */
+               struct {
+                       struct rhash_head l2_node;  /* L2 flow entry */
+                       struct hlist_head l2_flows; /* PPE L2 subflows list */
+               };
        };
 
        struct airoha_foe_entry data;
+       struct hlist_node l2_subflow_node; /* PPE L2 subflow entry */
        u32 hash;
 
        enum airoha_flow_entry_type type;
@@ -548,7 +552,8 @@ u32 airoha_rmw(void __iomem *base, u32 offset, u32 mask, u32 val);
 bool airoha_is_valid_gdm_port(struct airoha_eth *eth,
                              struct airoha_gdm_port *port);
 
-void airoha_ppe_check_skb(struct airoha_ppe *ppe, u16 hash);
+void airoha_ppe_check_skb(struct airoha_ppe *ppe, struct sk_buff *skb,
+                         u16 hash);
 int airoha_ppe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
                                 void *cb_priv);
 int airoha_ppe_init(struct airoha_eth *eth);
index 7219125ade134614d12bfad3f02b48392962d88f..d4969c2a03524253ace346dd6d862981ebfb44ed 100644 (file)
@@ -204,6 +204,15 @@ static int airoha_get_dsa_port(struct net_device **dev)
 #endif
 }
 
+static void airoha_ppe_foe_set_bridge_addrs(struct airoha_foe_bridge *br,
+                                           struct ethhdr *eh)
+{
+       br->dest_mac_hi = get_unaligned_be32(eh->h_dest);
+       br->dest_mac_lo = get_unaligned_be16(eh->h_dest + 4);
+       br->src_mac_hi = get_unaligned_be16(eh->h_source);
+       br->src_mac_lo = get_unaligned_be32(eh->h_source + 2);
+}
+
 static int airoha_ppe_foe_entry_prepare(struct airoha_eth *eth,
                                        struct airoha_foe_entry *hwe,
                                        struct net_device *dev, int type,
@@ -254,13 +263,7 @@ static int airoha_ppe_foe_entry_prepare(struct airoha_eth *eth,
 
        qdata = FIELD_PREP(AIROHA_FOE_SHAPER_ID, 0x7f);
        if (type == PPE_PKT_TYPE_BRIDGE) {
-               hwe->bridge.dest_mac_hi = get_unaligned_be32(data->eth.h_dest);
-               hwe->bridge.dest_mac_lo =
-                       get_unaligned_be16(data->eth.h_dest + 4);
-               hwe->bridge.src_mac_hi =
-                       get_unaligned_be16(data->eth.h_source);
-               hwe->bridge.src_mac_lo =
-                       get_unaligned_be32(data->eth.h_source + 2);
+               airoha_ppe_foe_set_bridge_addrs(&hwe->bridge, &data->eth);
                hwe->bridge.data = qdata;
                hwe->bridge.ib2 = val;
                l2 = &hwe->bridge.l2.common;
@@ -385,6 +388,19 @@ static u32 airoha_ppe_foe_get_entry_hash(struct airoha_foe_entry *hwe)
                hv3 = hwe->ipv6.src_ip[1] ^ hwe->ipv6.dest_ip[1];
                hv3 ^= hwe->ipv6.src_ip[0];
                break;
+       case PPE_PKT_TYPE_BRIDGE: {
+               struct airoha_foe_mac_info *l2 = &hwe->bridge.l2;
+
+               hv1 = l2->common.src_mac_hi & 0xffff;
+               hv1 = hv1 << 16 | l2->src_mac_lo;
+
+               hv2 = l2->common.dest_mac_lo;
+               hv2 = hv2 << 16;
+               hv2 = hv2 | ((l2->common.src_mac_hi & 0xffff0000) >> 16);
+
+               hv3 = l2->common.dest_mac_hi;
+               break;
+       }
        case PPE_PKT_TYPE_IPV4_DSLITE:
        case PPE_PKT_TYPE_IPV6_6RD:
        default:
@@ -496,15 +512,24 @@ static void airoha_ppe_foe_remove_flow(struct airoha_ppe *ppe,
                airoha_ppe_foe_commit_entry(ppe, &e->data, e->hash);
                e->hash = 0xffff;
        }
+       if (e->type == FLOW_TYPE_L2_SUBFLOW) {
+               hlist_del_init(&e->l2_subflow_node);
+               kfree(e);
+       }
 }
 
 static void airoha_ppe_foe_remove_l2_flow(struct airoha_ppe *ppe,
                                          struct airoha_flow_table_entry *e)
 {
+       struct hlist_head *head = &e->l2_flows;
+       struct hlist_node *n;
+
        lockdep_assert_held(&ppe_lock);
 
        rhashtable_remove_fast(&ppe->l2_flows, &e->l2_node,
                               airoha_l2_flow_table_params);
+       hlist_for_each_entry_safe(e, n, head, l2_subflow_node)
+               airoha_ppe_foe_remove_flow(ppe, e);
 }
 
 static void airoha_ppe_foe_flow_remove_entry(struct airoha_ppe *ppe,
@@ -520,10 +545,56 @@ static void airoha_ppe_foe_flow_remove_entry(struct airoha_ppe *ppe,
        spin_unlock_bh(&ppe_lock);
 }
 
-static void airoha_ppe_foe_insert_entry(struct airoha_ppe *ppe, u32 hash)
+static int
+airoha_ppe_foe_commit_subflow_entry(struct airoha_ppe *ppe,
+                                   struct airoha_flow_table_entry *e,
+                                   u32 hash)
+{
+       u32 mask = AIROHA_FOE_IB1_BIND_PACKET_TYPE | AIROHA_FOE_IB1_BIND_UDP;
+       struct airoha_foe_entry *hwe_p, hwe;
+       struct airoha_flow_table_entry *f;
+       struct airoha_foe_mac_info *l2;
+       int type;
+
+       hwe_p = airoha_ppe_foe_get_entry(ppe, hash);
+       if (!hwe_p)
+               return -EINVAL;
+
+       f = kzalloc(sizeof(*f), GFP_ATOMIC);
+       if (!f)
+               return -ENOMEM;
+
+       hlist_add_head(&f->l2_subflow_node, &e->l2_flows);
+       f->type = FLOW_TYPE_L2_SUBFLOW;
+       f->hash = hash;
+
+       memcpy(&hwe, hwe_p, sizeof(*hwe_p));
+       hwe.ib1 = (hwe.ib1 & mask) | (e->data.ib1 & ~mask);
+       l2 = &hwe.bridge.l2;
+       memcpy(l2, &e->data.bridge.l2, sizeof(*l2));
+
+       type = FIELD_GET(AIROHA_FOE_IB1_BIND_PACKET_TYPE, hwe.ib1);
+       if (type == PPE_PKT_TYPE_IPV4_HNAPT)
+               memcpy(&hwe.ipv4.new_tuple, &hwe.ipv4.orig_tuple,
+                      sizeof(hwe.ipv4.new_tuple));
+       else if (type >= PPE_PKT_TYPE_IPV6_ROUTE_3T &&
+                l2->common.etype == ETH_P_IP)
+               l2->common.etype = ETH_P_IPV6;
+
+       hwe.bridge.ib2 = e->data.bridge.ib2;
+       airoha_ppe_foe_commit_entry(ppe, &hwe, hash);
+
+       return 0;
+}
+
+static void airoha_ppe_foe_insert_entry(struct airoha_ppe *ppe,
+                                       struct sk_buff *skb,
+                                       u32 hash)
 {
        struct airoha_flow_table_entry *e;
+       struct airoha_foe_bridge br = {};
        struct airoha_foe_entry *hwe;
+       bool commit_done = false;
        struct hlist_node *n;
        u32 index, state;
 
@@ -539,12 +610,33 @@ static void airoha_ppe_foe_insert_entry(struct airoha_ppe *ppe, u32 hash)
 
        index = airoha_ppe_foe_get_entry_hash(hwe);
        hlist_for_each_entry_safe(e, n, &ppe->foe_flow[index], list) {
-               if (airoha_ppe_foe_compare_entry(e, hwe)) {
-                       airoha_ppe_foe_commit_entry(ppe, &e->data, hash);
-                       e->hash = hash;
-                       break;
+               if (e->type == FLOW_TYPE_L2_SUBFLOW) {
+                       state = FIELD_GET(AIROHA_FOE_IB1_BIND_STATE, hwe->ib1);
+                       if (state != AIROHA_FOE_STATE_BIND) {
+                               e->hash = 0xffff;
+                               airoha_ppe_foe_remove_flow(ppe, e);
+                       }
+                       continue;
+               }
+
+               if (commit_done || !airoha_ppe_foe_compare_entry(e, hwe)) {
+                       e->hash = 0xffff;
+                       continue;
                }
+
+               airoha_ppe_foe_commit_entry(ppe, &e->data, hash);
+               commit_done = true;
+               e->hash = hash;
        }
+
+       if (commit_done)
+               goto unlock;
+
+       airoha_ppe_foe_set_bridge_addrs(&br, eth_hdr(skb));
+       e = rhashtable_lookup_fast(&ppe->l2_flows, &br,
+                                  airoha_l2_flow_table_params);
+       if (e)
+               airoha_ppe_foe_commit_subflow_entry(ppe, e, hash);
 unlock:
        spin_unlock_bh(&ppe_lock);
 }
@@ -899,7 +991,8 @@ int airoha_ppe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
        return err;
 }
 
-void airoha_ppe_check_skb(struct airoha_ppe *ppe, u16 hash)
+void airoha_ppe_check_skb(struct airoha_ppe *ppe, struct sk_buff *skb,
+                         u16 hash)
 {
        u16 now, diff;
 
@@ -912,7 +1005,7 @@ void airoha_ppe_check_skb(struct airoha_ppe *ppe, u16 hash)
                return;
 
        ppe->foe_check_time[hash] = now;
-       airoha_ppe_foe_insert_entry(ppe, hash);
+       airoha_ppe_foe_insert_entry(ppe, skb, hash);
 }
 
 int airoha_ppe_init(struct airoha_eth *eth)